nhs r&d hta programme hta · nihr health technology assessment programme. t. he health...

Health Technology Assessm

ent 2007;Vol. 11: No. 17

Screening for type 2 diabetes: literature review and econom

ic modelling

Screening for type 2 diabetes: literature review and economic modelling

N Waugh, G Scotland, P McNamee, M Gillett, A Brennan, E Goyder, R Williamsand A John

Health Technology Assessment 2007; Vol. 11: No. 17

HTAHealth Technology AssessmentNHS R&D HTA Programmewww.hta.ac.uk

The National Coordinating Centre for Health Technology Assessment,Mailpoint 728, Boldrewood,University of Southampton,Southampton, SO16 7PX, UK.Fax: +44 (0) 23 8059 5639 Email: [email protected]://www.hta.ac.uk ISSN 1366-5278

FeedbackThe HTA Programme and the authors would like to know

your views about this report.

The Correspondence Page on the HTA website(http://www.hta.ac.uk) is a convenient way to publish

your comments. If you prefer, you can send your comments to the address below, telling us whether you would like

us to transfer them to the website.

We look forward to hearing from you.

May 2007

Copyright notice

© Queen's Printer and Controller of HMSO 2007 HTA reports may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising Violations should be reported to [email protected] Applications for commercial reproduction should be addressed to HMSO, The Copyright Unit, St Clements House, 2–16 Colegate, Norwich NR3 1BQ

How to obtain copies of this and other HTA Programme reports.An electronic version of this publication, in Adobe Acrobat format, is available for downloading free ofcharge for personal use from the HTA website (http://www.hta.ac.uk). A fully searchable CD-ROM isalso available (see below).

Printed copies of HTA monographs cost £20 each (post and packing free in the UK) to both public andprivate sector purchasers from our Despatch Agents.

Non-UK purchasers will have to pay a small fee for post and packing. For European countries the cost is£2 per monograph and for the rest of the world £3 per monograph.

You can order HTA monographs from our Despatch Agents:

– fax (with credit card or official purchase order) – post (with credit card or official purchase order or cheque)– phone during office hours (credit card only).

Additionally the HTA website allows you either to pay securely by credit card or to print out yourorder and then post or fax it.

Contact details are as follows:HTA Despatch Email: [email protected]/o Direct Mail Works Ltd Tel: 02392 492 0004 Oakwood Business Centre Fax: 02392 478 555Downley, HAVANT PO9 2NP, UK Fax from outside the UK: +44 2392 478 555

NHS libraries can subscribe free of charge. Public libraries can subscribe at a very reduced cost of £100 for each volume (normally comprising 30–40 titles). The commercial subscription rate is £300 per volume. Please see our website for details. Subscriptions can only be purchased for the current orforthcoming volume.

Payment methods

Paying by chequeIf you pay by cheque, the cheque must be in pounds sterling, made payable to Direct Mail Works Ltdand drawn on a bank with a UK address.

Paying by credit cardThe following cards are accepted by phone, fax, post or via the website ordering pages: Delta, Eurocard,Mastercard, Solo, Switch and Visa. We advise against sending credit card details in a plain email.

Paying by official purchase orderYou can post or fax these, but they must be from public bodies (i.e. NHS or universities) within the UK.We cannot at present accept purchase orders from commercial companies or from outside the UK.

How do I get a copy of HTA on CD?

Please use the form on the HTA website (www.hta.ac.uk/htacd.htm). Or contact Direct Mail Works (seecontact details above) by email, post, fax or phone. HTA on CD is currently free of charge worldwide.

The website also provides information about the HTA Programme and lists the membership of the variouscommittees.

HTA

Screening for type 2 diabetes:literature review and economicmodelling

N Waugh,1* G Scotland,2 P McNamee,2

M Gillett,3 A Brennan,3 E Goyder,4 R Williams5

and A John5

1 Department of Public Health, University of Aberdeen, UK2 Health Economics Research Unit, University of Aberdeen, UK3 Department of Health Economics and Decision Science, ScHARR,

University of Sheffield, UK4 Department of Public Health, University of Sheffield, UK5 Department of Public Health, University of Swansea, UK

* Corresponding author

Declared competing interests of authors: none

Published May 2007

This report should be referenced as follows:

Waugh N, Scotland G, McNamee P, Gillett M, Brennan A, Goyder E, et al. Screening fortype 2 diabetes: literature review and economic modelling. Health Technol Assess2007;11(17).

Health Technology Assessment is indexed and abstracted in Index Medicus/MEDLINE,Excerpta Medica/EMBASE and Science Citation Index Expanded (SciSearch®) and Current Contents®/Clinical Medicine.

NIHR Health Technology Assessment Programme

The Health Technology Assessment (HTA) programme, now part of the National Institute for HealthResearch (NIHR), was set up in 1993. It produces high-quality research information on the costs,

effectiveness and broader impact of health technologies for those who use, manage and provide care inthe NHS. ‘Health technologies’ are broadly defined to include all interventions used to promote health,prevent and treat disease, and improve rehabilitation and long-term care, rather than settings of care.The research findings from the HTA Programme directly influence decision-making bodies such as theNational Institute for Health and Clinical Excellence (NICE) and the National Screening Committee(NSC). HTA findings also help to improve the quality of clinical practice in the NHS indirectly in thatthey form a key component of the ‘National Knowledge Service’.The HTA Programme is needs-led in that it fills gaps in the evidence needed by the NHS. There arethree routes to the start of projects. First is the commissioned route. Suggestions for research are actively sought from people working in theNHS, the public and consumer groups and professional bodies such as royal colleges and NHS trusts.These suggestions are carefully prioritised by panels of independent experts (including NHS serviceusers). The HTA Programme then commissions the research by competitive tender. Secondly, the HTA Programme provides grants for clinical trials for researchers who identify researchquestions. These are assessed for importance to patients and the NHS, and scientific rigour.Thirdly, through its Technology Assessment Report (TAR) call-off contract, the HTA Programmecommissions bespoke reports, principally for NICE, but also for other policy-makers. TARs bring together evidence on the value of specific technologies.Some HTA research projects, including TARs, may take only months, others need several years. They cancost from as little as £40,000 to over £1 million, and may involve synthesising existing evidence,undertaking a trial, or other research collecting new data to answer a research problem.The final reports from HTA projects are peer-reviewed by a number of independent expert refereesbefore publication in the widely read monograph series Health Technology Assessment.

Criteria for inclusion in the HTA monograph seriesReports are published in the HTA monograph series if (1) they have resulted from work for the HTAProgramme, and (2) they are of a sufficiently high scientific quality as assessed by the referees and editors.Reviews in Health Technology Assessment are termed ‘systematic’ when the account of the search,appraisal and synthesis methods (to minimise biases and random errors) would, in theory, permit thereplication of the review by others.

The research reported in this monograph was commissioned by the HTA Programme as project number05/02/01. The contractual start date was in March 2005. The draft report began editorial review inOctober 2005 and was accepted for publication in October 2006. As the funder, by devising acommissioning brief, the HTA Programme specified the research question and study design. The authorshave been wholly responsible for all data collection, analysis and interpretation, and for writing up theirwork. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and wouldlike to thank the referees for their constructive comments on the draft document. However, they do notaccept liability for damages or losses arising from material published in this report.The views expressed in this publication are those of the authors and not necessarily those of the HTA Programme or the Department of Health.

Editor-in-Chief: Professor Tom WalleySeries Editors: Dr Aileen Clarke, Dr Peter Davidson, Dr Chris Hyde,

Dr John Powell, Dr Rob Riemsma and Dr Ken SteinManaging Editors: Sally Bailey and Sarah Llewellyn Lloyd

ISSN 1366-5278

© Queen’s Printer and Controller of HMSO 2007This monograph may be freely reproduced for the purposes of private research and study and may be included in professional journals providedthat suitable acknowledgement is made and the reproduction is not associated with any form of advertising.

Applications for commercial reproduction should be addressed to: NCCHTA, Mailpoint 728, Boldrewood, University of Southampton,Southampton, SO16 7PX, UK.

Published by Gray Publishing, Tunbridge Wells, Kent, on behalf of NCCHTA.Printed on acid-free paper in the UK by St Edmundsbury Press Ltd, Bury St Edmunds, Suffolk. G

Objectives: To reconsider the aims of screening forundiagnosed diabetes, and whether screening should befor other abnormalities of glucose metabolism such asimpaired glucose tolerance (IGT), or the ‘metabolicsyndrome’. Also to update the previous review for theNational Screening Committee (NSC) on screening fordiabetes, including reviewing choice of screening test;to consider what measures would be taken if IGT andimpaired fasting glucose (IFG) were identified byscreening, and in particular to examine evidence ontreatment to prevent progression to diabetes in thesegroups; to examine the cost-effectiveness of screening;and to consider groups at higher risk at whichscreening might be targeted.Data sources: Electronic databases were searched upto the end of June 2005.Review methods: Literature searches and reviewconcentrated on evidence published since the lastreview of screening, both reviews and primary studies.The review of economic studies included only thosemodels that covered screening. The new modellingextended an existing diabetes treatment model bydeveloping a screening module. The NSC has a set ofcriteria, which it applies to new screening proposals.These criteria cover the condition, the screening testor tests, treatment and the screening programme.Screening for diabetes was considered using thesecriteria.Results: Detection of lesser degrees of glucoseintolerance such as IGT is worthwhile, partly becausethe risk of cardiovascular disease (CVD) can bereduced by treatment aimed at reducing cholesterollevel and blood pressure, and partly because somediabetes can be prevented. Several trials have shown

that both lifestyle measures and pharmacologicaltreatment can reduce the proportion of people withIGT who would otherwise develop diabetes. Screeningcould be two-stage, starting with the selection ofpeople at higher risk. The second-stage choice of testfor blood glucose remains a problem, as in the lastreview for NSC. The best test is the oral glucosetolerance test (OGTT), but it is the most expensive, isinconvenient and has weak reproducibility. Fastingplasma glucose would miss people with IGT. Glycatedhaemoglobin does not require fasting, and may be thebest compromise. It may be that more people wouldbe tested and diagnosed if the more convenient testwas used, rather than the OGTT. Five economicstudies assessed the costs and short-term outcomes ofusing different screening tests. None examined thelong-term impact of different proportions of falsenegatives. All considered the costs that would beincurred and the numbers identified by different tests,or different cut-offs. Results differed depending ondifferent assumptions. They did not give a clear guideas to which test would be the best in any UK screeningprogramme, but all recognised that the choice of cut-off would be a compromise between sensitivity andspecificity; there is no perfect test. The modellingexercise concluded that screening for diabetes appearsto be cost-effective for the 40–70-year age band, moreso for the older age bands, but even in the 40–49-yearage group, the incremental cost-effectiveness ratio forscreening versus no screening is only £10,216 perquality-adjusted life-year. Screening is more cost-effective for people in the hypertensive and obesesubgroups and the costs of screening are offset in many groups by lower future treatment costs. The


iii

© Queen’s Printer and Controller of HMSO 2007. All rights reserved.

Abstract

Screening for type 2 diabetes: literature review and economicmodelling

N Waugh,1* G Scotland,2 P McNamee,2 M Gillett,3 A Brennan,3 E Goyder,4

R Williams5 and A John5

1 Department of Public Health, University of Aberdeen, UK2 Health Economics Research Unit, University of Aberdeen, UK3 Department of Health Economics and Decision Science, ScHARR, University of Sheffield, UK4 Department of Public Health, University of Sheffield, UK5 Department of Public Health, University of Swansea, UK* Corresponding author

iv

cost-effectiveness of screening is determined as muchby, if not more than, assumptions about the degree ofcontrol of blood glucose and future treatmentprotocols than by assumptions relating to the screeningprogramme. The very low cost now of statins is also animportant factor. Although the prevalence of diabetesincreases with age, the relative risk of CVD falls,reducing the benefits of screening. Screening fordiabetes meets most of the NSC criteria, but probablyfails on three: criterion 12, on optimisation of existingmanagement of the condition; criterion 13, whichrequires that there should be evidence from high-quality randomised controlled trials (RCTs) showingthat a screening programme would reduce mortality ormorbidity; and criterion 18, that there should beadequate staffing and facilities for all aspects of theprogramme. It is uncertain whether criterion 19, thatall other options, including prevention, should havebeen considered, is met. The issue here is whether allmethods of improving lifestyles in order to reduceobesity and increase exercise have been sufficientlytried. The rise in overweight and obesity suggests that

health promotion interventions have not so far beeneffective.Conclusions: The case for screening for undiagnoseddiabetes is probably somewhat stronger than it was atthe last review, because of the greater options forreduction of CVD, principally through the use ofstatins, and because of the rising prevalence of obesityand hence type 2 diabetes. However, there is also agood case for screening for IGT, with the aim ofpreventing some future diabetes and reducing CVD.Further research is needed into the duration ofundiagnosed diabetes, and whether the rise in bloodglucose levels is linear throughout or whether theremay be a slower initial phase followed by anacceleration around the time of clinical diagnosis. This has implications for the interval after whichscreening would be repeated. Further research is also needed into the natural history of IGT, and in particular what determines progression todiabetes. An RCT of the type required by NSCcriterion 13 is under way but will not report for about7 years.

Abstract


v


Contents

List of abbreviations .................................. vii

Executive summary .................................... xi

1 Background ................................................ 1Screening for type 2 diabetes – the issues .................................................... 1The condition ............................................. 2The test ...................................................... 10The treatment ............................................ 10Trends in overweight and obesity .............. 10The prevalence of diabetes – diagnosed and undiagnosed ........................................ 11The cost of diabetes ................................... 12

2 Previous reviews ........................................ 13Wareham and Griffin (2001) ...................... 13Engelgau and colleagues (2000) ................ 13Harris and colleagues (2003) ..................... 14The CDC Working Group .......................... 15Other reviews ............................................. 16Diabetes UK ............................................... 17The American Diabetes Association .......... 17Exercise ...................................................... 17

3 Screening tests .......................................... 19Stage 1 – selection by risk factors .............. 19Stage 2 – glucose testing ............................ 22Conclusions ................................................ 25

4 Review of economic models and evaluations ................................................. 27Overview ..................................................... 27Economic models assessing long-term costs and/or consequences of screening for type 2 diabetes ...................................... 28Economic models assessing the cost-effectiveness of identifying and treating

people with impaired glucose tolerance ..................................................... 42Studies assessing the costs and short-term outcomes of diabetes screening tests ............................................ 56

5 Modelling the cost-effectiveness of screening for type 2 diabetes .................... 65Introduction ............................................... 65Methods ...................................................... 65Results ........................................................ 73Discussion ................................................... 80Background to methods ............................. 83

6 Discussion ................................................... 85The aims of screening ................................ 85Screening interval ...................................... 85Does screening for diabetes and IGT meet the NSC criteria? ............................... 86

Acknowledgements .................................... 93

References .................................................. 95

Appendix 1 Search strategies .................... 107

Appendix 2 The NSC criteria ................... 109

Appendix 3 Management of impaired fasting glucose and impaired glucose tolerance ..................................................... 111

Health Technology Assessment reportspublished to date ....................................... 127

Health Technology Assessment Programme ................................................ 141

1,5-AG 1,5-antrydroglicitol

ACE angiotensin-converting enzyme

ACEI ACE inhibitor

ADA American Diabetes Association

AHRQ Agency for Healthcare Researchand Quality

ARB angiotensin receptor blocker

AUROC area under receiver operatingcharacteristic

BG blood glucose

BMI body mass index

BNF British National Formulary

CA cardiac arrest

CBG capillary blood glucose

CDC (US) Centers for Disease Controland Prevention

CHD coronary heart disease

CI confidence interval

CRS Cambridge Risk Score

CT computed tomography

CVD cardiovascular disease

DBP diastolic blood pressure

DCCT Diabetes Control andComplications Trial

DPP Diabetes Prevention Program

DPS Diabetes Prevention Study

DRS Diabetes Risk Score

EASD European Association for theStudy of Diabetes

ESRD end-stage renal disease

FBG fasting blood glucose

FPG fasting plasma glucose

FRA fructosamine

HbA1c glycosylated haemoglobin

HDL high-density lipoprotein

HOMA homeostasis model analysis

HRQoL Health-related quality of life

ICER Incremental cost-effectivenessratio

IFG impaired fasting glucose

IGT impaired glucose tolerance

IHD ischaemic heart disease

LADA latent autoimmune diabetes inadults

LDL low-density lipoprotein

LEA lower extremity amputation

LYG life-year gained

MI myocardial infarction

NCEP-ATP National Cholesterol EducationProgram Adult Treatment Panel

NDDG National Diabetes Data Group

NGSP National GlycohemoglobinStandardization Program

NGT normal glucose tolerance

NHANES National Health and NutritionExamination Survey

List of abbreviations


vii


continued

List of abbreviation continuedNICE National Institute for Health and

Clinical Excellence

NNS number-needed-to-screen

NSC National Screening Committee

OGTT oral glucose tolerance test

OHA oral hypoglycaemic agent

OR odds ratio

PCDP preclinical detectable phase

PCT Primary Care Trust

PG plasma glucose

PVD peripheral vascular disease

QALY quality-adjusted life-year

QWBI Quality of Wellbeing Index

RCT randomised controlled trial

RPG random plasma glucose

RR relative risk

SBP systolic blood pressure

SD standard deviation

SRQ Symptom Risk Questionnaire

STOP- Study To Prevent Non-insulin NIDDM Dependent Diabetes Mellitus

T2DM type 2 diabetes mellitus

UKPDS United Kingdom ProspectiveDiabetes Study

WHO World Health Organization

YHPHO York and Humber Public HealthObservatory

List of abbreviations

viii

All abbreviations that have been used in this report are listed here unless the abbreviation is well known (e.g. NHS), or it has been used only once, or it is a non-standard abbreviation used only in figures/tables/appendices in which case the abbreviation is defined in the figure legend or at the end of the table.


ix


BackgroundThe National Screening Committee (NSC) isresponsible for providing advice on screeningpolicy to all parts of the UK. A review of policy onscreening for type 2 diabetes is due shortly, andthis document was commissioned by the NHSR&D HTA Programme in order to support thatreview.

It is known that a proportion of people with type 2diabetes are undiagnosed. Blood glucose levelscan rise to diabetic levels with little or nothing inthe way of symptoms. Sometimes by the timepeople are diagnosed with diabetes, they havedeveloped complications such as the eye damageknown as retinopathy, due to an effect of diabeteson small blood vessels (microvascular disease).However, the main risk to health in undiagnosedtype 2 diabetes is an increased risk ofcardiovascular disease, in particular ischaemicheart disease, because of damage to the arteries(macrovascular disease). Early detection ofdiabetes would lead to measures to reduce the riskof heart disease, such as the use of statins to lowercholesterol, and also reduction of blood glucoselevels by, initially, diet and exercise, supplementedwith hypoglycaemic drugs if necessary.

Microvascular disease such as retinopathy isspecific to diabetes. However, the macrovasculardisease seen in diabetes is broadly the samedisease as seen in people without diabetes; thedifference in diabetes is the increased risk. Animportant issue when considering whether thereshould be screening for diabetes is that unlike withretinopathy, the increase in risk starts below thelevel of blood glucose used to define diabetes.There are groups of people who have higher thannormal blood glucose levels but who are notdiabetic. They are classified according to whethertheir blood glucose level is raised when fasting[impaired fasting glucose (IFG)] or is normal when fasting but raised after meals, or aftertesting with a 75-g glucose drink. The secondgroup are said to have impaired glucose tolerance (IGT).

The risk of heart disease is increased slightly inIFG but by about 60% in IGT.

Hence if reduction of heart disease is one of theaims of screening, then we should considerscreening not just for diabetes, but also for IGT.

ObjectivesThe objectives of this review were as follows:

● to reconsider the aims of screening forundiagnosed diabetes, and whether screeningshould be for other abnormalities of glucosemetabolism such as IGT, or the ‘metabolicsyndrome’

● to update the previous review for the NSC onscreening for diabetes, including reviewingchoice of screening test

● to consider what measures would be taken ifIGT and IFG were identified by screening, andin particular to examine evidence on treatmentto prevent progression to diabetes in thesegroups

● to examine the cost-effectiveness of screening,by a review of previous economic models, andby new modelling to take account of recentdevelopments in treatment such as the use ofstatins

● as part of the economic analysis, to considergroups at higher risk at which screening mightbe targeted

● to identify research needs.

MethodsThe literature searches (carried out up to the endof June 2005) and review concentrated onevidence published since the last review ofscreening, both reviews and primary studies. Thereview of economic studies included only thosemodels that covered screening. The newmodelling extended an existing diabetes treatmentmodel by developing a screening module.

The NSC has a set of criteria, which it applies tonew screening proposals. These criteria cover thecondition, the screening test or tests, treatmentand the screening programme. Screening fordiabetes was therefore considered using thesecriteria.

Executive summary

x

ResultsAs was known before this review, undiagnoseddiabetes can be detected by screening several yearsbefore it would become apparent after thedevelopment of symptoms. Earlier detection andtreatment reduces the development both ofspecific diabetes problems such as eye disease andof cardiovascular disease. Treatment to reduce therisk of cardiovascular disease has become muchless costly since the arrival of generic statins,which are now very cheap.

Detection of lesser degrees of glucose intolerancesuch as IGT is worthwhile, partly because the riskof cardiovascular disease can be reduced bytreatment aimed at reducing cholesterol level andblood pressure, and partly because some diabetescan be prevented. Several trials have shown thatboth lifestyle measures and pharmacologicaltreatment can reduce the proportion of peoplewith IGT who would otherwise develop diabetes.

Screening could be two-stage, starting with theselection of people at higher risk, based onprimary care records of age, weight and otherindicators of metabolic risk such as hypertension.Screening might be targeted at those above acertain body mass index threshold, whilerecognising that any cut-off would be an arbitraryline on a continuum of risk. The second-stagechoice of test for blood glucose remains aproblem, as in the last review for NSC. All offasting plasma glucose, the oral glucose tolerancetest and glycated haemoglobin would beacceptable, but none is perfect. The best test is theoral glucose tolerance test (OGTT), but it is themost expensive, is inconvenient and has weakreproducibility. Fasting plasma glucose would misspeople with IGT. Glycated haemoglobin does notrequire fasting, and may be the best compromise.It may be that more people would be tested anddiagnosed if the more convenient test was used,rather than the OGTT.

A review of previous economic models showed thatscreening for diabetes appeared to be cost-effective. The models differed in some aspects butreached broadly similar conclusions. The strongestand most comprehensive came from the USA, andthere were some doubts over their applicability tothe UK.

Five previous modelling studies examined thecosts and benefits of identification and screeningof people with IGT. All predicted that diabetesprevention measures would provide good value for

money. One was conducted from a UKperspective. Diet and exercise treatment is themost cost-effective option. Treatment withmetformin may be less cost-effective than lifestylechanges, but would be appropriate in somegroups. To some extent, the models may haveunderestimated benefit by focusing mainly onprevention of diabetes, and not taking full accountof the benefits of lifestyle changes on risk factorsfor cardiovascular disease.

Five economic studies assessed the costs and short-term outcomes of using different screening tests.None examined the long-term impact of differentproportions of false negatives. All considered thecosts that would be incurred and the numbersidentified by different tests, or different cut-offs.Results differed depending on differentassumptions. They did not give a clear guide as towhich test would be the best in any UK screeningprogramme, but all recognised that the choice ofcut-off would be a compromise between sensitivityand specificity; there is no perfect test.

The modelling exercise concluded that:

● Screening for diabetes appears to be cost-effective for the 40–70-year age band, more sofor the older age bands than the 40–49-yearband, but even in the 40–49-year age group, theincremental cost-effectiveness ratio forscreening versus no screening is only £10,216per quality-adjusted life-year.

● Screening is more cost-effective for people inthe hypertensive and obese subgroups.

● The costs of screening are offset in many groupsby lower future treatment costs.

● The cost-effectiveness of screening isdetermined as much by, if not more than,assumptions about the degree of control ofblood glucose and future treatment protocolsthan by assumptions relating to the screeningprogramme.

● The very low cost now of statins is an importantfactor.

Although the prevalence of diabetes increases withage, the relative risk of cardiovascular disease falls,reducing the benefits of screening.

Screening for diabetes meets most of the NSCcriteria, but probably fails on three:

● criterion 12, on optimisation of existingmanagement of the condition

● criterion 13, which requires that there should beevidence from high-quality randomised

Executive summary


xi


controlled trials showing that a screeningprogramme would reduce mortality ormorbidity

● criterion 18, that there should be adequatestaffing and facilities for all aspects of theprogramme.

It is uncertain whether criterion 19 – that all otheroptions, including prevention, should have beenconsidered – is met. The issue here is whether allmethods of improving lifestyles in order to reduceobesity and increase exercise have been sufficientlytried. The rise in overweight and obesity suggeststhat health promotion interventions have not sofar been effective.

ConclusionsThe case for screening for undiagnosed diabetes isprobably somewhat stronger than it was at the lastreview, because of the greater options forreduction of cardiovascular disease, principallythrough the use of statins, and because of therising prevalence of overweight and hence type 2diabetes. However, there is also a good case forscreening for IGT, with the aim of preventingsome future diabetes and reducing cardiovasculardisease.

Research needsOne key uncertainty concerns the duration ofundiagnosed diabetes, and whether the rise inblood glucose levels is linear throughout orwhether there may be a slower initial phase

followed by an acceleration around the time ofclinical diagnosis. This has implications for theinterval after which screening would be repeated.

Another uncertainty is the natural history of IGT,and in particular what determines progression todiabetes.

Research needs include the above, and

● Research into ways of reducing the prevalenceof insulin resistance. For example, what formsand amounts of exercise are required to preventor reduce insulin resistance?

● How can public health campaigns on lifestylemeasures be made more effective? Most cases oftype 2 diabetes are preventable. What balanceshould be struck between the public health,prevention by lifestyle approach, and the moremedical model of care focused on the individual?

● If screening were to be introduced, should it berepeated, and, if so, at what interval? More dataon the natural history of IGT may emerge fromcurrent research.

● If a decision were taken in principle thatselective screening should commence, furthermodelling as suggested in Chapter 5 could helpwith selection.

● A trial in which populations were clusterrandomised by practice to different screeningtests, with economic evaluation built in, might beuseful for showing which test was best in terms ofboth screening parameters and practicality.

A randomised controlled trial of the type requiredby NSC criterion 13 is under way but will notreport for about 7 years.

The purpose of this review is to underpinforthcoming discussions at the UK National

Screening Committee (NSC) on a review of policyon screening for type 2 diabetes mellitus (T2DM).The main aim was to look at evidence which hademerged since the last review, and so the first aimwas to examine recent reviews and any newprimary evidence not included in these reviews.However, as discussed in more detail below,screening for diabetes could, depending on thecut-off chosen for tests being positive, detect morepeople with lesser degrees of glucose intolerance,such as impaired glucose tolerance (IGT), thanwith diabetes. The main aim of screening is toreduce the burden of disease from cardiovasculardisease (CVD), to which people with diabetes aremore susceptible. Those with IGT are also atincreased risk, and although their relative risk(RR) is less than those with diabetes, there are farmore of them than there are people withundiagnosed diabetes, and so there will be morecardiovascular events in those with IGT than inthose with diabetes. They are also at risk ofprogression to diabetes.

Screening is therefore addressed from a somewhatwider perspective than in some previous reviews. A section has also been included that covers recentevidence on prevention of diabetes in those withwhat has been called ‘pre-diabetes’, although sincemost will not become diabetic, the term is notentirely satisfactory.

A key issue (details later) is that there is acontinuum of CVD risk across all levels of bloodglucose (BG). Hence there is no simple threshold at which people can be split into at riskand not at risk. In such circumstances, the finaldecision on whom to screen and treat (or at whatlevel of risk to do so) can be illuminated byeconomic analysis, since that can provide dataindicating when interventions become cost-effective.

Reviews of previous economic models of screeninghave been included. Considerable new modellinghas also been carried out.

Finally, the extent to which screening for diabetesand IGT meets the NSC criteria is also considered.

Screening for type 2 diabetes –the issuesThe NSC reviewed screening for T2DM a fewyears ago, and the current policy statement is onthe NSC website (www.nsc.nhs.uk). The case forscreening was assessed, as usual, against clearlydefined criteria, looking in turn at;

● the condition● the screening test● the treatment● the screening programme.

This review therefore does not start from a zerobase, but from the present NSC policy,underpinned by the work of Wareham and Griffin1

for the last policy review.

An assessment of the case for screening, judgedagainst the criteria, and based on the previousreview, is available on the NSC website. Many ofthe criteria were met, but there were concerns ordoubts over:

● whether all the cost-effective primaryprevention interventions had been applied(criterion 3)

● the screening test, and in particular whichthreshold of BG should be used, bearing inmind the need to focus more on large vesseldisease such as ischaemic heart disease (IHD)(criterion 6)

● whether screening and treatment should bespecifically aimed at diabetes, or a wider range of factors predisposing to IHD risk(criterion 11)

● whether treatment of existing diabetes wasoptimised (criterion 12)

● the lack of randomised controlled trials (RCTs)of screening (criterion 13)

● the balance between benefits and harms ofwidespread screening (criterion 15)

● the ability of the NHS to cope with a largenumber of people with newly diagnoseddiabetes (criterion 18).

Note that the NSC criteria are updated from timeto time, for example to cover new scenarios suchas genetic testing, and the numbering used above


1


Chapter 1

Background

reflects the current set, which are different fromthose used at the time of the last review. Thecurrent criteria are listed for convenience inAppendix 2.

In this review, the case for screening will beassessed against the criteria, but focusing mainlyon those where there were concerns last time, oron those where the evidence base may havechanged. We do not address type 1 diabetes,where screening is not required, or gestationaldiabetes, which was the subject of a previousreview.2

The NSC criteria now fall into groups as follows:

● the condition (criteria 1–4)● the test (criteria 5–9)● the treatment (criteria 10–12)● the screening programme (criteria 13–22).

Some of the criteria are not applicable to thisreview. Criteria 4, 9 and 22 deal with geneticscreening. Criteria 17, 18 and 20 are concernedwith the running of screening programmes andneed not be addressed until a decision in principleto provide it is taken.

The criteria that seem most important to thecurrent review are discussed below. The others thatare relevant, and the evidence which relates tothem, will be dealt with in Chapter 6.

The conditionNSC criterion 1 – the condition shouldbe an important health problemThe importance of T2DM has not diminished.Indeed, the trend in the prevalence is upwards,with a rise in all-age prevalence due todemographic change, and almost certainly a risein age-specific prevalences due to increasing levelsof obesity. It has been estimated that there will bean increase in prevalence of 16% for Englandbetween 2001 and 2010, based on ONS censusprojections and current obesity trends.3

However, it may be worth reflecting on what wewould be trying to achieve by screening for T2DM.People with T2DM are less prone than those withtype 1 diabetes to the acute metaboliccomplications such as diabetic ketoacidosis, whichstill causes deaths in the young. However, if goodglycaemic control is not achieved, they are at riskof the specific diabetic microvascularcomplications such as retinopathy and

nephropathy. An increasing proportion (in somecentres over half) of diabetic people on renaldialysis have T2DM.4

However, the biggest problem in T2DM is largevessel disease, and most people with T2DM die ofcoronary heart disease. Diabetes is also importantin peripheral vascular disease (PVD) and stroke.5,6

In most studies looking at the relationshipbetween BG and mortality in T2DM, the higherthe glucose level, the higher is the mortality, butthe rise per unit of glucose [e.g. per mmol/l offasting plasma glucose (FPG) or per 1% ofglycosylated haemoglobin (HbA1c)] is modest.7

The importance of large vessel disease can be seenin the end-points reported in the UK ProspectiveDiabetes Study (UKPDS). Table 1 shows thenumbers for most of the end-points in theconventionally treated group (UKPDS 33).8

Hence the majority of adverse events were due tolarge vessel disease. Most of the microvascularend-points were made up of retinalphotocoagulation for retinopathy, the risk of whichis probably less in the group who would bedetected by screening. It should be noted thatamongst the UKPDS patients, there were somewho might now be classified as latent autoimmunediabetes in adults (LADA), rather than trueT2DM.

However, if we take only those who were in theoverweight group who were randomised to

Background

2

TABLE 1 Numbers of end-points in UKPDS

Endpoint Number

MacrovascularFatal MI 91Non-fatal MI 101Sudden death 18Heart failure 36Angina 72Stroke 59Amputations and death from PVD 21MicrovascularDeath from renal disease 2Renal failure 9Blindness in one eye 38Vitreous haemorrhage 10Photocoagulation for retinopathy 117

All macrovascular 397All microvascular 176

MI, myocardial infarction.

metformin or conventional treatment if their FPGat 3 months was higher than 6.0 mmol/l, and whomay be more similar to those who would be foundby screening, then the picture is similar.Overweight was defined as more than 120% ofideal body weight. Table 2 shows the numbers ofend-points in the control group (UKPDS 34).9

Again, the end-points are dominated by largevessel disease.

Hence it could be argued (and has been in thepast – see reviews by Goyder and Irwig10 andJarrett11) that the most important reason forscreening for T2DM is in order to be able tointervene earlier with a view to reducing the riskof macrovascular disease, and mainly IHD.

What is diabetes?A digression into the underlying rationale for thedefinition of diabetes is now necessary. The only constant feature of diabetes is a raisedBG level. There may or not be any of the classicalsymptoms such as the passing of larger volumes ofurine and thirst. Many people with T2DM have nosymptoms when diagnosed. Conversely, manypeople without diabetes report similar symptoms,and so the symptoms, at least in milder forms, arenot specific to diabetes.12,13

However, the problem comes when defining whatis meant by ‘raised’.

Successive reports by working parties for theWHO14,15 and the American Diabetes Association(ADA)16,17 have examined the problems ofdiagnostic criteria for diabetes. The earlier historyhas been reviewed by Keen,18 who noted that theclassifications in the late 1970s, from the NationalDiabetes Data Group (NDDG) in the USA,19 and

the second report of the WHO ExpertCommittee20 were based on a hybrid approach,being primarily based on clinical descriptionaccording to treatment, but with some elementsbased on assumed aetiology. The classificationdivided diabetes mainly into insulin-dependentand non-insulin-dependent. However, thesereports did at least produce diagnostic criteria in terms of a threshold for true diabetes (a fasting blood glucose level of 7.8 mmol/l or over,and a 2-hour post-load level of 11.1 mmol/l orover), hence removing the uncertainty over what should be classed as diabetes. The term‘impaired glucose tolerance’ (IGT) was used todescribe the situation where BG was raised above normal but was below the threshold fordiabetes. This replaced terms such as ‘borderline’diabetes. However, the lower limit for IGT was leftsomewhat vague.18 The normal FPG level is up to5.6 mmol/l.

The key feature of the classifications was that thediagnosis of diabetes was based on the level atwhich the risk of retinopathy started. At the riskof some over-simplification, people with glucoselevels below the threshold did not get retinopathy;those above were at risk of retinopathy, with therisk increasing as glucose levels rose further. Thiswas based on three studies, described in the reportof the ADA’s expert committee.17

Despite the rationalisation which the WHO andADA classifications brought to a previouslysomewhat confused situation, some dissatisfactionsremained. One was the ‘hybrid’ nature of theclassification, and the confusion that arose becausemany people with non-insulin-dependent diabeteswere being treated with insulin in order to achievebetter control. A second was that the classificationwas too dependent on the oral glucose tolerancetest (OGTT), which is an inconvenient andunphysiological test (involving drinking 75 g ofglucose in water over a short period) with poorreproducibility. Another was that the twothresholds – 7.8 mmol/l for fasting and 11.1 mol/lfor 2 hours after a glucose load – had imperfectcorrelation, in that the fasting level implied agreater degree of hyperglycaemia than the 2-hourlevel.

The ADA and WHO14 groups reviewed theirclassifications. Their conclusions were fairlyconsistent (there was cross-representation betweenthe expert groups), and the new classificationsmade a number of changes. First, diabetes wassubdivided clearly, according to the need forinsulin, into:


3


TABLE 2 End-points in the overweight UKPDS group – controlsonly (N = 411)

End-point Number

MacrovascularAll IHD (MI, heart failure, angina) 121Stroke 25PVD 11MicrovascularRenal failure 3Blind in one eye 13Vitreous haemorrhage 3Photocoagulation 36

All macrovascular 157All microvascular 52

● Type 1 diabetes, where insulin treatment wasrequired for survival, because the pancreaticislets cells have been destroyed by the diseaseprocess; this covers the insulin-dependentgroup from the previous definitions.

● T2DM, where pancreatic insulin productioncontinues, but may be insufficient for control ofblood sugar levels; most of this group will betreated with diet alone, or with tablets, but somewill need insulin for control of BG. In theUKPDS patients, the trend was for glucose torise over time.21

Second, the ADA threshold for the FPG level atwhich diabetes was diagnosed was lowered to7.0 mmol/l, to be more compatible with the 2-hourlevel. It was anticipated that fasting glucose levelswould be used as the main method of diagnosis,being more convenient and more reproducible.The ADA group envisaged that FPG would be themain method of diagnosis, with OGTTs beingusually unnecessary.

The WHO group recommended the same fastingand 2-hour cut-off levels but, for epidemiologicaland screening purposes, preferred the 2-hourvalue.

The new ADA classification gives four groups;

● Those with both fasting and post-load levelsabove the thresholds – the diabetics, of whomthose with FPGs between 7.0 and 7.8 mmol/lcould be called the ‘new’ diabetics.

● Those with fasting above the upper limit ofnormal (6.1 mmol/l) but below 7.0 mmol/l; thisgroup is said to have impaired fasting glucose(IFG).

● Those with a normal fasting glucose under6.1 mmol/l, but with the post-load level above7.8 but under 11.1 mmol/l; this group isdescribed as having IGT.

● Those with fasting levels under 6.1 and post-load under 7.8 mmol/l, who are classed asnormal.

A collaborative project by the European DiabetesEpidemiology Group, the DECODE study,22

pooled data from 13 cohort studies, in order toexamine the risks of mortality in the variousgroups, relative to those with normal (defined asunder 6.1 mmol/l) glucose levels. Hazard ratioswere as given in Table 3.

Taking IFG and IGT statuses in combination, asomewhat simplified version of some of theirfindings is as given in Table 4.

Hence IFG alone, without IGT, is associated with aslight increase in mortality [although theconfidence intervals (CIs) overlap with noincrease), but IGT carries more risk. Similarfindings were reported from a meta-analysis byCoutinho and colleagues23 of 20 studiesexamining cardiovascular mortality (19 studies) ormorbidity (four studies). A fasting glucose level of6.1 mmol/l carried 1.33 times the risk of thereference one of 4.2 mmol/l; a 2-hour glucose levelof 7.8 mmol/l carried an RR of 1.58 comparedwith a 2-hour level of 4.2 mmol/l.

In both IFG and IGT, there is insulin resistance,but with different distributions. Pima Indians withIFG have higher fasting insulin levels than thosewith IGT, but the latter have higher post-prandialinsulin levels.24 There are also differences in othercardiovascular risk factors, with higher triglycerideand fibrinogen levels in IGT than IFG, reflectingthe higher insulin resistance in IGT than IFG.25

IGT is common – it affects 17% of Britons aged40–65 years.26 This has implications for choice ofa screening test – if FPG were to be used, a groupof people whose FPG is normal but who have IGTwould be missed. In the Rancho Bernardo study,27

the RR of a cardiovascular event in women aged50–89 years with normal FPG but IGT was 2.9.Similarly, a Paris study found that the heartdisease mortality rate in men with normal fastingglucose but IGT was double that of those with

Background

4

TABLE 3 DECODE – hazard ratios for different glycaemicstates

State Men Women

Normal 1.0 1.0IFG 1.21 1.08IGT 1.51 1.60‘New diabetes’ 1.81 1.79a

a Those with fasting glucose between 7.0 and 7.8 mmol/l.

TABLE 4 Relative risks of mortality

IGT status

IFG status Normal 2-hour IGTlevel

Normal fasting 1.0 1.56 (1.33 to 1.83)level

IFG 1.18 (0.99 to 1.42)a

a 95% confidence intervals.

normal glucose tolerance.28 However, an earlierpaper from Paris29 noted that fasting insulin levelswere a better predictor of future heart disease thanglucose levels, presumably reflecting the varyingdegrees of insulin resistance.

The Helsinki Police Study found the same.30 Inthe 22-year follow-up, 10% of those in the lowestquintile of baseline insulin had an IHD eventcompared with 25% of those in the highest insulinquintile. Also, in the San Antonio Heart Study,Haffner found that the baseline insulin levelpredicted not only future T2DM, but alsohypertension, low high-density lipoprotein (HDL)and high triglycerides.31

The DECODE group32 carried out a meta-analysisof 11 studies and compared the cardiovascularmortality in the highest and lowest quartiles ofplasma insulin levels. After adjustment for otherrisk factors, the risks were 1.54 in men and 2.66 inwomen in the highest quartiles.

In Japan, Tominaga and colleagues33 found thatIGT was a risk factor for CVD but that IFG wasnot. Numbers of deaths were fairly low and CIswide, but by the 7-year follow-up, the cumulativesurvival in those with IGT was significantly lowerthen that of those with normal glucose levels,whereas the survival in those with IFG was notdifferent from normal.

Unlike with retinopathy, there is no suddeninflection in the risk curve according to bloodglucose levels, but rather a continuum of risk.

Indeed, even within what is regarded as being theentirely normal range, higher BG levels correlatewith higher IHD rates. In the EPIC study inNorfolk,34 the relationship between HbA1c andcardiovascular risk started well within the non-diabetic range (Table 5).

A similar finding was reported by Piche andcolleagues in Quebec,35 although using 2-hourplasma glucose. They compared groups with lownormal 2-hour plasma glucose (PG)(<5.6 mmol/l), ‘high normal’ (5.6–7.7 mmol/l) andwith IGT. The high normal group were moreobese (as measured by waist, visceral fat bycomputed tomography (CT) scan, andsubcutaneous fat) than those with low normal PG.Their cholesterol:HDL ratio was also higher.

The same applies to PVD. Muntner andcolleagues36 reported data from the 1999–2002NHANES survey. The figures in Table 6 are after

multivariate adjustment. PVD was defined as anankle:brachial blood pressure ratio under 0.9.However, the CIs were wide and only the lastfigures had a 95% CI which did not overlap with 1.0.

There have been suggestions that the thresholdfor IFG should be reduced to 5.6 mmol/l.37 Thiswould greatly increase the prevalence of IFG. Forexample, in the DESIR study of French men andwomen aged 30–64 years, the prevalence of IFGwould rise from the 13% in men and 4% in womenseen using the old IFG threshold of 6.1 mmol/l to40% in men and 16% in women.38 Balkau andcolleagues38 noted that the risk of progression todiabetes is much less in old IFG than in new(Table 7). However the progression to IHD mayshow a different gradient.

It may be that IGT and IFG reflect differentaetiologies. Piche and colleagues39 followed acohort of men who initially had normal glucosetolerance for 6 years. Some developed IFG andsome IGT; most remained normal. Those whodeveloped IFG tended to be leaner than thosewho developed IGT [baseline body mass index(BMI) 26 versus 28], and on CT had much lessvisceral fat. Those who developed IFG tended tohave lower baseline fasting insulin, whereas thosewho developed IGT had higher levels. Hence IFGmay reflect a production problem with insulinwhereas IGT reflects insulin resistance. Numbersin this study were small: eight with IFG and 12with IGT.


5


TABLE 5 EPIC study – relative risks by bands of glycated HbA1c

HbA1c (%) RR of CVD

Men Women

<5% 1.0 1.05–5.4% 1.23 0.895.5–5.9% 1.56 0.986.0–6.4% 1.79 1.636.5–6.9 % 3.03 2.37>7% (newly diagnosed diabetes) 5.01 7.96Prior diabetes 3.32 3.36

TABLE 6 NHANES – relative risks of PVD by bands of HbA1c

HbA1c (%) RR of PVD

<5.3% 1.05.3–5.4 1.415.5–5.6 1.395.7–6.0 1.57

The metabolic syndromeThe metabolic syndrome has been described as acluster of cardiovascular risk factors associatedwith insulin resistance.40 An operational definitionwas devised by the National Cholesterol EducationProgram Adult Treatment Panel (NCEP-ATP),41

based on the presence of three or more of thefollowing:

● Abdominal obesity – waist circumference over40 inches (~100 cm) in men, and 35 inches(~90 cm) in women

● Triglycerides over 150 mg/dl (1.68 mmol/l)● HDL cholesterol <40 mg/dl (1.03 mmol/l) in

men, <50 µg/dl (1.29 mmol/l) in women● Blood pressure 130 or more systolic, 85 or more

diastolic● Fasting glucose over 110 mg/dl (6.1 mmol/l).

Hypertension and diabetes are commoncomponents, and about 40% of those with themetabolic syndrome have diabetes.40 Others willhave lesser degrees of glucose intolerance.However, some will have normal glucose tolerance.The question therefore arises as to whether oneshould be screening for the metabolic syndromebecause of the increased risk of CVD, rather thanjust for abnormalities of glucose metabolism.

Another issue concerns the choice of treatment forhypertension. People with hypertension have longbeen known to be at increased risk of diabetes,and in several studies it was noted that substantialproportions of newly diagnosed diabetic patientswere on anti-hypertensive drugs (see Jarrett andFitzgerald42 for a review). This raised thepossibility that some of the older drugs forhypertension such as thiazides might be causingsome diabetes.42

A meta-analysis of recent trials has cast new lighton this problem. Scheen43 pooled results from

eight RCTs of angiotensin-converting enzymeinhibitors (ACEIs) and reported that ACEIs andthe more specific angiotensin receptor blockers(ARBs) appear to reduce the incidence of diabetes,albeit only modestly, from 9.6% in those nottreated with an ACEI or ARB to 7.4% in those whowere. This finding applied whether the controlswere given placebo or older drugs such asthiazides. Those on thiazides or beta-blockers hadno greater incidence of diabetes than those onplacebo.

In a more recent meta-analysis, Gillespie andcolleagues44 included 14 trials of ACEIs or ARBsand found that these drugs reduced the incidenceof T2DM by 22% [odds ratio (OR) 0.78; 95% CI0.73 to 0.83].

Hence screening for diabetes might provide benefitto those with hypertension who are found to haveIGT or IFG; their hypertensive treatment could bechanged to an ACEI or ARB, which should reducethe risk of progression to diabetes.

The WHO definition of the metabolic syndrome iscouched more in terms of insulin resistance:

● T2DM, IFG or IGT, or normal glucose level buthigh insulin level plus two or more of

● elevated blood pressure – systolic 160 or over,diastolic 90 or over

● elevated triglycerides – 1.7 mmol/l or over● low HDL cholesterol – <0.9 mmol/l in men and

<1.0 mmol/l in women● obesity – BMI >30, or central obesity with

waist:hip ratio >0 .90 in men and, >0.85 inwomen

● microalbuminuria.

Scuteri and colleagues45 from the CardiovascularHealth Study in Americans over 65 years oldfollowed a cohort of 2175 subjects for just over

Background

6

TABLE 7 DESIR study – incidence of diabetes according to glycaemic status at baseline.

Incidence of diabetes per 1000 person-years

IFG status Glucose level (mmol/l) 30–44 years 45–54 years 55–64 years

MenNormal <5.6 2.3 1.7 1.1New IFG 5.6–6.0 4.9 8.5 11.5Old IFG 6.1–6.9 24.7 38.9 63.9

WomenNormal <5.6 0.4 1.4 0.7New IFG 5.6–6.0 5.5 7.0 5.9Old IFG 6.1–6.9 35.7 52.3 66.7

4 years. At baseline, 28% had the metabolicsyndrome by NCEP ATP III criteria and 21% hadit by WHO criteria; 81% of those considered tohave metabolic syndrome by one set of criteria alsohad it by the other. Using the metabolic syndromealone gave RRs for CVD of 1.9 and 1.89 for NCEPand WHO, respectively. Taking into account age,sex, family history of myocardial infarction (MI),smoking and low-density lipoprotein (LDL)cholesterol, increased the RR with the NCEPcriteria to 2.04 (95% CI 1.7 to 2.5) but reduced theRR with the WHO set to 1.63 (95% CI 1.3 to 2.0).

The increased risk of vascular disease in themetabolic syndrome has been reviewed recently byFord.46 The increase in risk depends on whichgroups are included. If those with diabetes areexcluded, the RR is less (1.58) than if those withdiabetes are included (2.02). Overall, the RR ofcardiovascular risk is about 1.7–1.9.

Data from the National Health and NutritionExamination Survey (NHANES) (unpublished butquoted by Haffner31) showed that in individualsaged over 20 years, the prevalences of IHD were:

● without metabolic syndrome – 8.5%● with metabolic syndrome by WHO criteria – 12.5%● metabolic syndrome by ATP III (NCEP) criteria

– 16.6%.

In the San Antonio Heart Study, Stern andcolleagues47 found that the metabolic syndrome asdefined by NCEP was not as good a predictor offuture diabetes as the Diabetes Predicting Model(which includes family history) or as good apredictor of heart disease as the Framingham RiskScore (which includes age, sex, smoking andcholesterol), but information on those factorscould easily be collected at the same time.

The prevalence of the metabolic syndrome hasbeen studied in middle-aged (40–69 years) peoplein different ethnic groups in London (Brent andSouthall) using both definitions. Tillin andcolleagues48 found the highest prevalence amongstSouth Asians (46% in men, 31% in women; WHOdefinition) and the lowest in European women(9%). However, the rates were not standardised forobesity. Central obesity was much commoner inthe Asian groups. Hence their higher prevalencemay reflect obesity not ethnicity (although thetendency towards central obesity could be due toeither genetic or lifestyle factors).

But would including screening for the metabolicsyndrome go far enough? Reaven49 pointed out

that many individuals who are both insulin-resistant and dyslipidaemic do not meet the NCEPcriteria, but are still at increased risk of vasculardisease. The answer may lie in economics – atwhat level of risk does screening and treatmentcease to be cost-effective?

The value of having a condition called themetabolic syndrome, rather than just dealing withthe individual components, has been called intoquestion by a joint statement from the ADA andthe European Association for the Study of Diabetes (EASD), published simultaneously inDiabetologia and Diabetes Care. The statementconcludes that:

Until much-needed research is completed, cliniciansshould evaluate and treat all CVD risk factors withoutregard to whether a patient meets the criteria fordiagnosis of the ‘metabolic syndrome’.

Kahn and colleagues (2005).50

The converse of the metabolic syndrome has alsobeen studied. Perry and colleagues51 from Cork inIreland identified factors, from previous studies,which protected against diabetes: BMI <25;waist:hip ratio <0.85 for women and 0.90 for men;never smoking; medium- to high-level physicalactivity; light drinking (3–5 to 7 units per week);and a prudent diet. In their sample of middle-aged Irish men and women, drawn from generalpractice populations, 7.5% had none of theseprotective factors. Insulin resistance was calculatedusing the homeostasis model analysis (HOMA)score (based on fasting levels of both insulin andglucose). Taking the 7.5% with no protectivefactors as the reference group, multivariateanalysis gave ORs for insulin resistance of 0.59with one protective factor, 0.48 with two, 0.14 withthree and 0.04 with four or more. About 13% hadfour or more protective factors.

Hence there is little doubt that lifestyle measurescould prevent most cases of T2DM. Weight losswould also benefit those who already havediabetes, or hypertension, and improvements canfollow even modest weight loss. Goldstein52

reviewed studies in which large and small amountsof weight were lost, and concluded that evenmodest weight reductions, of 10% or less, resultedin significant benefit in a substantial subset. Evenloss of a few kilograms can provide benefit.

In addition to the total calorie intake, the qualityof the diet can affect the incidence of diabetes. Inthe EPIC-Potsdam study, Heidemann andcolleagues53 found that a healthy pattern diet


7


(high intake of fresh fruit and low intake of high-calorie soft drinks, beer, red meat, poultry,processed meat) reduced the incidence of diabetes.ORs from the poorest to the best diet ranged from1.0 to 0.26. Adjusting for BMI reduced thedifference only slightly, implying that those whodo not succeed in dieting to lose weight could stillreduce their risk of diabetes by changing thecontent of the diet.

Decision pointHence if the aim of screening is to reduce heartdisease, it could be argued that one should looknot only at diabetes, but at all degrees of glucoseintolerance, and perhaps wider still for themetabolic syndrome.

This has implications for the test to be used, forthe choice of definitive test and for the treatmentsto be given to screen-positives. The aims oftreatment might be:

1. For those with definite diabetes, reduction ofthe risk of retinopathy and nephropathy, byreduction of PG to normal, initially trying dietand exercise, but using drug therapy whenindicated.

2. For those with PG levels in the IFG and IGTranges, prevention of progression to diabetes,by diet and exercise, or by drug therapy ifindicated.

3. For all of the above, measures to reducecardiovascular risk, by measures other than theglucose control measures already mentioned,such as qualitative improvements in diet;aspirin; cholesterol-lowering measures such asstatins; blood pressure control; and anti-obesitymeasures.

Decision analysis 1.Should the aim of screening be to identify people withdiabetes, or should there be a broader aim to reduce heartdisease in a wider group, or both? Diabetes would thenbe regarded as the higher glycaemic end of a spectrum ofglucose intolerance that is in turn part of a widermetabolic syndrome.

NSC criterion 2 – the epidemiology andnatural history of the condition,including development from latent todeclared disease, should be adequatelyunderstood and there should be adetectable risk factor, disease marker,latent period or early symptomatic stageIf the condition is diabetes, then this criterion ispartially met, although the terminology is notexactly right. Undiagnosed diabetes is not ‘latent’

in the sense that it can be causing damage (forexample, people with newly diagnosed diabetesmay already have retinopathy). It is latent only inthe sense of being asymptomatic (although somemay have mild symptoms such as polyuria withoutrealising the significance, or indeed that it isabnormal). The ‘disease marker’ is BG, measuredin various ways.

The main reason for using the word ‘partially’ isthat there is uncertainty about the duration ofdiabetes before diagnosis.

The duration of undiagnosed diabetesThe most quoted estimate of the duration ofdiabetes before diagnosis is from Harris andcolleagues,54 who used data from two population-based groups, one in the USA (Wisconsin) and the other in Australia, supplemented withdata from published studies from other areas orstudies, including UKPDS and Edinburgh. Inbrief, they plotted prevalence of retinopathyagainst known duration, and then extrapolatedback to estimate when diabetes started. Theprevalences at diagnosis were different – 21% inWisconsin and 10% in Australia. Possibleexplanations for the difference might be that theWisconsin group was very thoroughly studied byKlein and colleagues55 in the series of studies from the Wisconsin epidemiology of retinopathystudy, and so they may have detected more.Another contributor may have been that some(7.5%) of the Australian group were not diabeticby NDDG/WHO criteria of the time – someprobably had IGT.

In the UKPDS, the prevalence of retinopathy atdiagnosis (or, strictly, referral – but GPs were asked to refer newly diagnosed patients) was 24%, close to that in Wisconsin (even though,strictly, not all the UKPDS patients had diabeteseither – the entry criterion was an FPG of6.0 mmol/l). It might be worth noting that theUKPDS group was not population based, butrelied on referral to hospital clinics, and many hadhypertension, which would increase the risk ofretinopathy.

Extrapolating back gave estimates of onset ofdiabetes of 6.5 years prior to diagnosis inWisconsin and 4.2 years in Australia. The authors concluded, cautiously, that onset occurred “at least 4–7 years” before diagnosis.However, as they noted in the discussion, it takestime for metabolic changes to give rise todetectable retinopathy, so the onset is probablyearlier still.

Background

8

Ideally, studies of onset of retinopathy would startfrom the actual onset of diabetes. This requiresprospective studies, and there are not many ofthose. Jarrett56 followed up a cohort of men withIGT, with repeated OGTTs, so that the date ofdevelopment of diabetes in those who progressedcould be more accurately determined. Most of themen (180 of 240) did not develop diabetes over10 years. Of the 60 who did, none had detectableretinopathy in the first 5 years after onset. Thatsuggests that onset of diabetes could be 9–12 yearsbefore clinical diagnosis.

Thompson and colleagues57 used data from Egypt,and a similar backwards extrapolation method tothat of Harris and colleagues, but estimated thatthe onset of retinopathy was only 2.6 years beforeclinical diagnosis. The 95% CI was wide at 0.3 to8.4 years; they had data on 218 patients.Prevalence of retinopathy at diagnosis was 12%.They estimated that onset preceded diagnosis by7.6 years, using the data on duration of diabetesto retinopathy from Jarrett’s study.

One method used for estimating date of onsetuses the data from UKPDS on decline of beta cellfunction. UKPDS paper 1621 showed a steady risein HbA1c, irrespective of treatment, of about0.22% per year. The mean HbA1c level in thosesymptomatic at diagnosis was 9.6. The upper limitof normal HbA1c in UKPDS was 6.1. Hence, if it isassumed that the decline in beta cell function waslinear throughout, then one obtains an estimate of16 years from first elevation above normal (note:not the same as diabetes – there would be anintervening period of IFG and IGT).

Another UKPDS paper (UKPDS 6158) examinedinitial FPGs. The patients in the lowest tertile ofinitial FPG were more likely to be asymptomaticand more likely to be found by screening (such asat medical examinations for insurance oremployment purposes). However, theirprogression of glycaemia was similar over time tothat of patients with higher initial levels, so itappears that they may simply have been at anearlier stage in the disease. A shift forward of5 years would overlay the lowest tertile on thehighest, implying that there may be a gap of5 years between their initial HbA1c of 6.7% andthe 10% in the highest, more symptomatic group.That would suggest a rise per annum in HbA1cof 0.6%

Another possible method is to use GP recordspreceding the date of diagnosis. Gulliford andcolleagues59 used this approach to examine

contacts with and prescriptions from primary carein the UK in the 5 years before diabetes wasdiagnosed. However, they included only patientswho at some stage were treated with oralhypoglycaemic agents (OHAs); those on diet alonewere excluded. They then compared the OHAgroup with controls who never had any diabetesmedication. The diabetic group had higherconsultation rates 5 years before diagnosis andmore hypertension, hyperlipidaemia, obesity andIHD. However, they were not (as far as is known from the GP data) screened for diabetes, so some may have had undiagnosed diabetes orIGT. Records showed that 0.6% had IGT and 5%had ‘hyperglycaemia’ recorded. The authorsconcluded that none of the non-glycaemic factorswas sufficiently predictive of future diabetes –which for our purposes means that these factorscannot be used for retrospective estimates ofonset.

There are two problems with some of the aboveestimations.

First, the means for HbA1c for such groups willconceal variations within them. In particular, thoseleft on diet alone include both those who lostweight and took exercise and those who did not.UKPDS does not separate these subgroups. Along-term model of beta cell decline has beendevised by Bagust and Beale60 using data from the Belfast diet study, in which patients whomanaged to stay on diet alone for 6 years or morehad comparatively little decline in beta cellfunction.

Second, it is not known if the decline in beta cellfunction is linear throughout the course of thecondition. Is the decline in the ‘pre-diabetes’ andundiagnosed diabetes stages at the same rate?Bagust and Beale60 hypothesise that there are twostages – an initial slow decline over perhaps10 years or more, and then a rapid decline overthe next 10 years.

The rate of decline in the period before diagnosis (and hence the rate of increase of HbA1c and vascular damage) could haveconsiderable effects on the cost-effectiveness ofscreening. This will be examined in the economicmodel.

It may be worth considering the differencebetween the duration of undiagnosed disease inthe natural history situation and the undiagnosedduration in practice. The duration of undiagnoseddisease, that is, the interval between onset of the


9


metabolic problem and the appearance ofsymptoms, depends on the natural history and ispresumably fairly constant (unless othercontributory factors have changed over time, suchas levels of exercise). However, the undiagnosedduration may depend on clinical or patientbehaviour, which may have changed over timebecause of increasing awareness of risk factors,public concern about undiagnosed diabetes(prompted by campaigns such as the Diabetes UK‘missing million’), and easy accessibility of glucosetesting. Hence the proportion undiagnosed nowmay be less than in the past, although absolutenumbers may be greater, due to increasingprevalence.

If the condition includes impaired glucosetolerance, there is again reasonable understandingof the condition – it is known that it can progressto diabetes, and is associated with an increasedrisk of CVD. It is known that many people go backfrom IGT to normality. In the 10-year follow-up tothe Bedford study, far more men with IGTreturned to normal (128) than progressed todiabetes (36).61 And as discussed above, it is notknown if the decline in those who do progress islinear.

However, we know that a sizeable proportion willdevelop diabetes. Edelstein and colleagues62

analysed six studies, with 2389 individuals withIGT, followed up for between 2 and 27 years, andnoted that 23–62% progressed to diabetes overtime, at a rate of 4–9% per year. Admittedly, thehigher progression rates were seen in ethnicgroups at particularly high risk, such as the Pimasand Naurans. Progression was increased if BMIwas higher, but was not affected by age or familyhistory of diabetes.

The testNSC criterion 5. There should be a simple, safe,precise and validated screening testNSC criterion 6. The distribution of test values inthe target population should be known and asuitable cut-off level defined and agreedNSC criterion 7. The test should be acceptable tothe populationNSC criterion 8. There should be an agreedpolicy on the further diagnostic investigation ofindividuals with positive test results and thechoices available to those individualsScreening tests are addressed in Chapter 3. Theposition as regards these criteria depends partlyon what we are screening for – T2DM, IGT andIFG, or the metabolic syndrome.

TreatmentNSC criteria 10–12Treatments for diabetes itself are soundly evidence based, and will not be addressed in thisreview. However, Appendix 3 examines themanagement of lesser degrees of glucoseintolerance such as IGT, since any screeningprogramme would find more people with thatthan with diabetes.

Trends in overweight and obesityThe prevalence of T2DM is closely linked withthat of overweight. Data from the Health Surveyfor England63 show that the proportion of thepopulation which is overweight or obese (BMI 30or over) has been increasing in recent years(Table 8).

Background

10

TABLE 8 Health Survey for England: proportions of people who are overweight or obese trends over time

Men (%) Women (%)

BMI 25–30 BMI 30+ BMI 25–30 BMI 30+

1993 44.4 13.2 32.2 16.41994 44.3 13.8 31.4 17.31995 44 15.3 32.9 17.51996 44.6 16.4 33.6 18.41997 45.2 17.0 32.8 19.71998 45.5 17.3 32.1 21.21999 43.9 18.7 32.8 21.12000 44.5 21.0 33.8 21.42001 46.6 21.0 32.9 23.52002 43.4 22.1 33.7 22.8

The risks of being overweight are well known. Lewand Garfinkel64 reported mortality data from avery large study from the American Cancer Societywith over 750,000 men and women, classified interms of whether their weights were average,above or below. The reference group comprisedthose with weights 90–110% of average. Table 9shows the mortality ratios for IHD and diabetes.

However, these figures will reflect a mixture ofoverweight people – some just overweight, otherswith diabetes or hypertension. It may be more thediseases associated with overweight that increasethe risk. Obesity alone is less risky than obesitywith a co-morbidity, such as diabetes. Oldridgeand colleagues65 looked in the 1992 Health andRetirement Study in the USA (it is not clear fromthe results paper whether the study sampledpeople only in Michigan) at the effects of obesity,diabetes and hypertension in middle age(51–70 years). Compared with people with none ofthe three, those with obesity had a 24% increasedrisk of hospital admission, whereas those withdiabetes and obesity had double the risk. Thosewith all three had three times the risk. Some of theobese probably had undiagnosed diabetes. Thediagnosis in the study was made if respondentssaid that a doctor had previously told them thatthey had diabetes.

Hence the main risk of obesity may be from theincreased risk of diabetes, rather than obesityitself.

The prevalence of diabetes –diagnosed and undiagnosedThe York and Humber Public Health Observatory(YHPHO) (www.york.ac.uk/yhpho) has developed aprevalence model for diabetes. The currentestimated prevalence of diabetes (diagnosed andundiagnosed) is 4.37% of the population, 0.33%

under 30 years, 3.3% from 30 to 59 years and13.8% in those over 60 years. Prevalence alsovaries by ethnicity: whites 4.2%, blacks 5.6% andAsians 6.6% (not corrected for obesity). TheYHPHO estimates that there are currently about2.14 million people with diabetes in England, orabout 1 in 20 of the population.

By 2010, population ageing alone will haveincreased the proportion to 4.63%. If overweighttrends continue, the proportion due to ageing andweight combined will rise to 5.05%. If theprevalence of overweight and obesity were to fallto the levels before 1995, the expected prevalenceof diabetes would fall, to 4.24%, despite theageing change.

Rises in the prevalences of diabetes and IGT havebeen reported in other Western countries.Dunstan and colleagues in Australia66 reported aprevalence in the over 25-year-olds of diabetes of8% in men and 6.8% in women, with an additional17% of men and 15% of women having IGT orIFG. These data were based on a very largepopulation-based survey of 11,247 participantswho had the 75 g OGTT (the authors say that it isthe largest OGTT study in the world). Only 0.8%were indigenous Australians.

Half of those identified as having diabetes wereundiagnosed. The prevalence of IFG was 8% inmen and 3.4% in women; for IGT the figures were9% and 12%. The prevalence was much higherthan the 3.4% seen in two (smaller) previousAustralian studies (Busselton and Victoria, in 1981and 1992, respectively)67,68 and than self-reporteddiabetes in health surveys in 1989 and 1990 (2%).

The authors note that rises in obesity and theageing of the population explain part of the rise.However, they conclude that these do not fullyexplain the rise, and they hypothesise that twoother factors are reduced physical activity levels


11


TABLE 9 Mortality by weight

Weight as percentage of average

<80% 80–89 90–109 110–119 120–129 130–139 140+

IHDMen 0.88 0.90 1.0 1.23 1.32 1.55 1.95Women 1.01 0.89 1.0 1.23 1.39 1.54 2.07

DiabetesMen 0.88 0.84 1.0 1.65 2.56 3.51 5.19Women 0.65 0.61 1.0 1.92 3.34 3.78 7.90

and obesity at younger ages (and hence longerduration of obesity).

In Canada, a large study was carried out with theaid of a sample of physicians who looked after arepresentative sample of the population.69

Screening was sequential – a casual BG; those withresults over 5.5 mmol/l were asked to return for anFPG; those with FPG 6.1–6.9 mmol/l were invitedback for a 2-hour post-glucose load sample.However, screening was offered only to those whoattended for routine care for other reasons, whoby definition are less healthy than the rest of thepopulation, and may be at higher risk. That mayincrease the proportion with diabetes. Conversely,the number detected may have been reduced bythe failure of about one-quarter who had casualBG levels over 5.5 mmol/l to return for an FPG,and half those who did have an FPG did not havethe 2-hour post-load test.

It was believed that the prevalence of knowndiabetes in Canada was 5%. This study foundanother 2.2% with undiagnosed diabetes and atleast 3.5% with undiagnosed glucose intolerance(based on IFG and IGT combined).

Similar results were reported from the NHANESsurveys in the USA,70 summarised in a Centers forDisease Control and Prevention (CDC) report. Theprevalence of known diabetes in the over-20s was5.9%; testing by FPG found another 2.2%. Theprevalence of IFG was 6.1%. No 2-hour post loadtesting was done, so there may also have beenundiagnosed IGT.

The costs of diabetesSome cost of illness studies include both type 1and type 2 diabetes. Those which focus on type 2include a review by Raikou and McGuire,71 whonoted some of the problems in such studies:

● Cost of illness studies based only on diabetes asprimary diagnosis will underestimate costs,because most of the cost arises fromcomplications.

● Diabetes is under-recorded on hospitaldischarge data, so costs will be underestimated.

● When considering the cost of complications, theattributable proportion needs to be estimated,rather than the total cost. This can be done byestimating the excess treatment costs for peoplewith diabetes compared with a matched groupwithout diabetes (it could also be done using RR– if the RR of a heart attack is 4, then 75% ofthe cost could be attributed to diabetes).

Amongst the studies identified by Raikou andMcGuire71 as being of good quality, only one camefrom the UK, and it only looked at drug costs.Evans and colleagues in Tayside72 usedDARTS/MEMO data and showed that patientswith T2DM accounted for just over 7% of the totaldrugs budget. However, they noted that the excessprescribing costs were not just for diabeticmedications, and in particular that there was a10% higher use of cardiovascular drugs than inthe general population.

Raikou and McGuire71 appear to have missed thestudy by Bagust and colleagues,73 which reportsthe creation of a model, and the results fromrunning it, for T2DM. The model appears wellthought out and inevitably complex. The authorsestimate that the lifetime cost of healthcare after people have been diagnosed with T2DM isdouble that of the non-diabetic population.However, the economics of screening are notaddressed.

Williams and colleagues74 noted that much of thecost of healthcare for people with diabetes is notfor the diabetes itself, but for the complications.Using data from the Cost of Diabetes in EuropeType 2 study (CODE-2), they noted that 72% ofpatients in the study had at least onecomplication, of whom 34% had macrovasculardisease. In those with complications, the cost ofcare was increased by up to 250% compared withthose without complications. Hence there arefinancial reasons in addition to human factors forseeking to prevent the complications of diabetes,and the cost of complications has implications forthe cost-effectiveness of screening.

Background

12

Some of the more important (recent) reviews aresummarised below, and position statements

from the ADA and Diabetes UK are also included.

Wareham and Griffin (2001)1

This paper1 summarises a review done for the UKNSC, and is structured along the lines of the NSCcriteria (Appendix 2). Wareham and Griffin notedthat screening for T2DM meets many of thecriteria, but that a number of problems remain:

● The criterion that “cost-effective primaryprevention interventions should have beenimplemented” is not met, because there areknown methods of reducing the incidence ofT2DM, such as diet and exercise.

● There is no clear threshold at which the risk ofCVD takes off (unlike with microvasculardisease), which is a problem if the aim ofscreening is to reduce cardiovascular risk.

● The cost-effectiveness of screening wouldprobably depend mainly on that reduction incardiovascular risk, by a range of measures notspecific to diabetes, such as cholesterol andblood pressure lowering.

● Universal screening would be unlikely to beworthwhile, so target populations should beidentified.

● There are definite harms of screening, such asbeing labelled as diabetic.

● The NSC criterion on optimising present formsof care for the condition before finding newcases was not met.75

● The criterion that screening should besupported by evidence from RCTs was not met.

Wareham and Griffin concluded that on theevidence available in 2001, it was not clearwhether screening would be worthwhile.

Engelgau and colleagues (2000)76

Engelgau and colleagues76 carried out a series ofreviews of screening for diabetes. In the version of2000, they provide a thorough review examininghow well screening for T2DM matches the

screening criteria. Although this was a balancedand well-argued review, it cannot be classed assystematic; no details of search strategy, ofinclusion and exclusion criteria or indeed of anymethods were given.

They conclude that the criteria on whether thedisease is a significant health burden, the naturalhistory being known, and there being arecognisable asymptomatic but detectable stage,are met. That on whether early treatment hasadvantages over waiting until symptoms appear isprobably met, but with some doubts overcompliance, and they note the absence of RCTs ofscreening and early treatment.

They conclude that the criterion on a suitable testis met, while recognising that none of the tests areperfect. They regard urine testing as being oflimited value; consider that FPG is not as good aspost-prandial measurement because subjects withundiagnosed diabetes are more likely to exceed the2-hour cut-off than the FPG one; and concludethat HbA1c is acceptable, although not as good asthe post-load PG, but that there is a need tostandardise methods. They conclude thatquestionnaires are unsatisfactory, but they considerthese as a one-stop screening test, whereas they areprobably more useful as part of a two-stage screen.

They have considerable doubt as regards thecriterion on cost-effectiveness – possibly met withopportunistic screening, but not with population-based screening, even on a selective basis.

Their overall conclusion is that:

The effectiveness of screening for diabetes has notbeen directly demonstrated.

They accept that the early detection of diabetesand improved glycaemic control “may modestlyreduce” future microvascular disease, but thatthere is little evidence for benefits inmacrovascular disease. However, they do note thattreatment of hypertension and hyperlipidaemiamay be of benefit. Their review was written beforesome of the recent trials of statins in diabetes,such as CARDS,77 DALI78 and HPS,79 werepublished.


13


Chapter 2

Previous reviews

Previous reviews

14

Engelgau and colleagues76 noted that severalexpert bodies have recommended screening butthat none of the recommendations had beenformally evaluated.

Harris and colleagues (2003)80

Note: The first author of this review is RussellHarris, not to be confused with Maureen Harris,another noted contributor to the debate onscreening.

Harris and colleagues80 provided a high qualitysystematic review, with details on sources searchedand search strategy, study selection, dataextraction and synthesis and other aspects ofmethods. One weakness is that only MEDLINEwas searched. This is a common American failingand will lead to some trials being missed.81 Theyseem to have searched the Cochrane Library butonly for reviews, not for trials in CENTRAL.

The review was published as a journal article,80

but is also available at the full technical report onthe website of Agency for Healthcare Research andQuality (AHRQ).82

Harris and colleagues address several of thescreening issues. They recognise that there is anasymptomatic but detectable stage, but note thatthe true average length of this is unknown, andthat different people will have different durationsof this stage. They note that none of the tests areperfect; that the 2-hour PG is regarded as thereference standard but that it is less reproduciblethan the FPG; and that there are fewer logisticalproblems with HbA1c.

As regards the benefits of early treatment, theysummarise the trials of tighter control in T2DM(with the UKPDS being by far the largest), butcorrectly point out that these trials were mainlyabout the benefits in those whose diabetes hadbeen diagnosed, and hence that the findingsmight not be applicable to those detected byscreening and who would have lower glycaemialevels. This is a fair point. Even if there is stillbenefit at lower levels of glycaemia, the relativebenefits of lowering glucose might be similar interms of reducing microvascular complications,but since the absolute benefits would be less, thecost-effectiveness of intervention would be muchless.

The AHRQ review82 also addresses wider aspectsof risk reduction through screening, by

considering what benefits might accrue to thosefound to be diabetic through treatments forhypertension and hyperlipidaemia. The authorsnote that the diagnosis of diabetes can affect bothhow vigorously hypertension is treated and thechoice of drug. They report the findings of theHypertension Optimal Treatment trial,83 that theoptimal level of blood pressure may be lower forthose with diabetes than those without. They alsoreport on the controversy as to whether ACEIs aremore beneficial than other anti-hypertensiveagents, and conclude that they should still befavoured. This topic has been reviewed morerecently by Ravid and Rachmani,84 who alsoconcluded that ACEIs probably had advantagesand, as Harris and colleagues82 point out, theside-effects of ACEIs (such as cough) are less thanwith other anti-hypertensive drugs. They do notein an addendum that the ALLHAT trial, publishedin 2002,85 showed no advantage of ACEIs overcalcium channel blockers.

Another potential consequence of being diagnosedwith diabetes would be that the vascular risk scorewould be raised, and this could mean thedifference between being treated with a statin andnot, using the current consensus on levels of risk[National Institute for Health and ClinicalExcellence (NICE) guidance recommend thatpeople with a 10-year risk of 20% of CVD shouldbe treated with a statin].86

The same could apply to aspirin prophylaxis,which reduces the risk of heart disease in diabetes.

Hence one result of screening for diabetes mightbe the benefits from reduction of CVD by non-diabetic treatment, as a consequence of crossingthe threshold of risk at which these treatments arerecommended.

Harris and colleagues82 also review the harms oftreatment. They are unconvinced that ‘labelling’ isa significant problem, at least in terms of qualityof life (although noting possible problems with lifeinsurance costs) and they consider that the side-effects of drugs such as statins and ACEIs are nota great problem. They comment on thehypoglycaemic side-effects of diabetic drugs, butbase that on the UKPDS, where many patientswere on insulin. These results are less likely toapply to people found by screening, who would, atleast initially, have lower BG levels, controllable bydiet or metformin (which would be the drug ofchoice when a drug was needed, assuming thatscreening was selective and that a key selectioncriterion was BMI).

This review also considers cost-effectiveness, not inmonetary terms, but through estimation ofnumbers needed to screen to prevent one event.They use various assumptions (in the absence ofgood evidence) to show that to prevent one case ofblindness from retinopathy, by screen detectionthen tight glycaemic control, would have numbers-needed-to-screen (NNS) ranging from 4300 to61,400. However, they note that sincemacrovascular is much commoner thanmicrovascular disease, the NNS for that are muchlower, with a range in hypertensive people of 500 to7200 (which does not include the effect of statins).

Harris and colleagues82 came to no firmconclusions, perhaps because the purpose of theirreview was to provide a basis for the US PreventiveTask Force to do so, but the thrust of their reviewis that there are clear arguments in favour ofscreening, but also remaining doubts about cost-effectiveness.

The CDC Working Group87

The Primary Prevention Working Group of theCenters for Disease Control and Prevention(2004)87 carried out a review of primaryprevention of T2DM. They summarise the threerecent trials aimed at preventing IGT progressingto diabetes, as follows.

Da Qing Study88

This was an RCT over 6 years, with four groups –controls; diet; exercise; and diet plus exercise –with the incidence of diabetes reduced in the threeintervention groups by 31, 46 and 42%, respectively.

Finnish Diabetes Prevention Study(DPS)89

An RCT was performed over 3 years with twogroups – control versus lifestyle intervention(physical activity, weight loss, diet) – withincidence of diabetes reduced by 58% comparedwith controls.

Diabetes Prevention Program (DPP)90

This was an RCT of control versus metforminversus intensive lifestyle; incidence of diabetes wasreduced by 31% in the metformin group and 58%in the lifestyle group.

Fuller details of the trials are given in Appendix 3.

The CDC review has some useful thoughts onpolicy implications or challenges. These were asfollows:

1. Identification of subjects for preventiveinterventionsThe Working Group87 notes that there remainsuncertainty over best screening test, because ofthe inconvenience of the OGTT. They notethat there have been no trials of prevention ofprogression from IFG to diabetes, but that 24%of those with pre-diabetes have IFG.They then pose the question as to whether “theblood glucose criterion should be eliminated”.They note that many (61%) Americans areoverweight or obese, and that the lifestyleinterventions of the DPP would be beneficial toall. They wonder if the target should be notjust those with abnormal glucose levels, but allthe 47 million with the metabolic syndrome.However, they then wonder if intervening withsuch large numbers might dilute theintervention too much. No definite conclusion is reached.

2. The delivery of lifestyle interventionsThe Working Group87 note that lifestyleinterventions are effective, but that physiciansmay not be the best people to deliver them,because they are more accustomed to themedical role of disease management and drugprescribing. While physicians may have a rolein helping motivate people, the lifestyleinterventions might be better provided by non-medical staff such as dietitians, nurses orundefined “community health workers”. Butperhaps the key question is whether a morepublic health-based approach is required.Instead of trying to provide lifestyleinterventions as part of medical care, shouldthe aim be to reduce diabetes by “changing theunderlying environmental factors thatcontribute to obesity and sedentary behaviourin the general population”?

3. Are lifestyle interventions cost-effective?The review comments on the scarcity ofeconomic evaluations of preventing diabetes,but cites a 1998 reference in support of thisstatement, and may therefore have missedsome of the more recent economic studies (seeChapter 4 of this report). It notes that theincremental costs of the intervention arms ofthe DPP were modest at US$2191 and $2269for metformin and lifestyle interventions,respectively.

4. The ethical implicationsTwo main issues are considered by the WorkingGroup. The first is what strength of evidence isneeded for lifestyle interventions. For anycampaign aimed at the general public, or at ahigh-risk subset of it – but in either case atpeople who are not actually ill – should the


15


strength of evidence required be as good as orgreater than that required for conventionalmedical care?The second is about the ethics of extrapolation.If there is good evidence that lifestyleinterventions benefit those with IGT, can this beextrapolated to those with lesser risks ofprogression to diabetes or cardiovasculardisease, or are RCTs needed? As the review says:

Is it ethical to await results of a new, extensiveseries of randomised controlled trials to evaluateintervention efficacy in groups at lower risk ofdiabetes? Or is it acceptable to infer interventionefficacy in groups other than those defined by theeligibility criteria of the Diabetes PreventionProgram?

However the review makes no firmrecommendations but only summarises some ofthe evidence.

Other reviewsOne of the advocates of screening for diabetes hasbeen Maureen Harris, who has also been firstauthor of several studies on the prevalence ofundiagnosed diabetes in the USA. Her reviews areeditorials or commentaries rather than systematicreviews, but they deserve comment. In 1994, withMichaela Modan,91 she argued that there shouldbe a national programme in the USA, basing thisassertion on what was known about the prevalenceof undiagnosed diabetes and the presymptomaticphase, on the presence of complications in someat diagnosis, and the availability of effectivetreatments. They did not consider cost-effectiveness or the possible harms of treatment.They recommended the OGTT, althoughsimplified to a 2-hour glucose, as the best test,while accepting its drawbacks. They noted that ifcost and efficiency were to be considered, thenthere could be selection of people for screening onthe basis of obesity and hypertension.

In a later (non-systematic) review, Harris andEastman92 make similar points about the healthburden of diabetes, the latent phase and thedamage which can accrue during it, the naturalhistory and the effectiveness of treatments,updating the previous editorial with data on thebenefits of statins, and the lowered blood pressuregoal in those with diabetes. Another updateconcerns the new criteria of the ADA93 and therecommendations on which test to use. Harris andEastman make the point that although the 2-hourPG test is still better in theory (because it is more

sensitive), the FPG test may in practice detect agreater proportion because the simplicity makes itmore likely to be used. They do not considerHbA1c in this review.

Another recent review is by Borch-Johnsen andcolleagues.94 This was a short but good-qualityreview, with a clear search strategy and with theWHO screening criteria used for assessing the casefor screening. It focused mainly on the criteria, ona suitable test, on treatment and the indicationsfor that and on the ethical and psychologicalconsequences of screening.

They concluded that suitable tests were available,but that:

● FPG should not be used as the only test becauseone out of three cases of diabetes will be missed.

● Random blood glucose was of limited value, aswas urine testing.

● HbA1c would be useful when used as part of astepwise process, for example combined withFPG; they envisaged selecting individuals forOGTT testing.

● Questionnaires would help select high-riskpeople, but might need to be validated in othercountries.

Borch-Johnsen and colleagues94 had morereservations about treatment aspects. They notedthe beneficial results from trials in newlydiagnosed diabetes (UKPDS) and in the treatmentof hyperlipidaemia and hypertension. However,they identified a major gap in that there were notrials, then, of non-pharmacological interventionin the early stages of the disease, but that theADDITION (an Anglo–Danish–Dutch trial ofintervention in those with T2DM detected byscreening)95 had started. They concluded that:

Until such data are available, any recommendationregarding screening and early intervention willdepend on speculative interpretation of secondaryintervention studies.

As regards the ethical and psychologicalconsequences, they considered that there was aneed for further research. They also noted thatthere were shortfalls in care for many individualswith existing diabetes, which needed correctionbefore adding to the number by screening. Theirfinal conclusions were that T2DM met many of thecriteria but not all, that mass screening could notbe recommended, but that testing for diabetesshould be offered to those with hypertension,dyslipidaemia and CVD and those with many riskfactors for diabetes.

Previous reviews

16

Diabetes UKDiabetes UK issued a position statement on “Earlyidentification of people with type 2 diabetes” inNovember 2002.96 In brief, it advocated ascreening programme for both undiagnoseddiabetes, and also for IGT, with a view to reducingprogression to diabetes. It did not support generalpopulation screening, but recommended targetedcase finding, with screening for those with two ormore risk factors, such as:

1. White people aged over 40 years and peoplefrom black, Asian and minority ethnic groupsaged over 25 years, with:(a) a first degree family history of diabetes

and/or(b) who are overweight (BMI 25–20 and above)

and who have a sedentary lifestyle and/or(c) who have IHD, CVD, PVD or hypertension.

2. Women who have had gestational diabetes.3. Women with polycystic ovary syndrome who are

obese.4. Those known to have IGT or IFG.

Diabetes UK noted that there was a lack ofevidence on the best method of screening, but as regards tests, they favoured the 2-hour post-load glucose test (noting that a full OGTTwould be the best possible test but was impractical)followed by fasting PG measurement. Glycosuriawas also listed as having some value. HbA1c wasnot included in their list of recommended tests.

The American DiabetesAssociationThe ADA publishes clinical practicerecommendations each year, in a supplement toDiabetes Care. The 2005 edition97 has the followingstatement on screening for diabetes inasymptomatic adults:

1. Testing for diabetes should be considered in allindividuals at age 45 years and above, particularly inthose with a BMI of 25 or over, and if normal shouldbe repeated at 3-year intervals.2. Testing should be considered at a younger age orbe carried out more frequently in individuals who areoverweight (BMI �25) and have additional riskfactors, as follows:● are habitually physically inactive● have a first degree relative with diabetes● are members of a high-risk ethnic population● have delivered a baby weighing >9 lb or have been

diagnosed with GDM● are hypertensive (�140/90)

● on previous testing, had IGT or IFG● have history of vascular disease.

(ADA, 2005; abbreviated)97

The ADA said that either an FPG or a 2-hourOGTT (75 g) was appropriate, but that the FPGwas the preferred test for screening for pre-diabetes:

The FPG is more convenient to patients, morereproducible, and easier to administer than the 2-hOGTT. Therefore the recommended screening testfor non-pregnant adults is the FPG.

ExerciseA very useful collation of the evidence on physicalactivity and health was published as a book byHardman and Stensel.98 They examined theevidence for the benefits of exercise for variousconditions:

● CHD – exercise protects. It need not bevigorous (although vigorous exercise protectsmore) and so walking is sufficient to give someprotection. However, protection is lost shortlyafter exercise is stopped. The RR of heartdisease amongst the inactive is at least 2.

● Hypertension – the risk of developinghypertension amongst the inactive is about 1.5.Exercise is an effective way of reducing bloodpressure, although the reductions are modestcompared with pharmacological treatment –about 3.4 mmHg systolic and 2.4 mmHgdiastolic.

● Lipid levels – exercise increases HDL andlowers triglycerides.

● Coagulation of blood – exercise lowers plateletaggregation and fibrinogen.

● T2DM – cohort studies show a reduction in risk.Vigorous exercise four times per week cuts therisk of T2DM by 40% in men (from thePhysicians’ Health Study99). Taking more than7 hours of exercise a week reduces the risk ofdiabetes in women (from the Nurses HealthStudy100) by 30%. Having the combination of agood diet, a BMI under 25 and high physicalactivity (over 30 minutes per day of moderateor more strenuous exercise) in the NursesHealth Study gave an RR of diabetes of about0.1; Hu and colleagues100 suggest that 87% ofT2DM in avoidable. In those already havingT2DM, remission was seen after exercise in halfof those recently diagnosed with diabetes afterscreening (but numbers were small – only 41 intotal; from the Malmo feasibility study101).


17


Hardman and Stensel98 quote the National AuditOffice report on possible reasons for the reductionin physical activity:

● a reduction in occupational exercise● greater car ownership and use● a decline in walking as a mode of transport

● energy-saving devices in public places, such asescalators, lifts and automatic doors

● fewer opportunities for young people to takeexercise

● the substitution of physically active leisureactivities with sedentary ones such as computergames, television and the Internet.

Previous reviews

18

Organised screening would be a two-stageprocess, with the first stage being selection

from the general population (using generalpractice registers) of those likely to be more at riskthan average, and the second being testing ofglucose levels, usually in blood. However, somestudies have used urinary glucose testing as thesecond screen, followed by BG testing.

There is already some ad hoc screening on offer,such as by pharmacies, or urine testing at newpatient registration in general practice.

Stage 1 – selection by risk factorsTesting only people who are at higher thanaverage risk means that a higher proportion ofthose who will be tested for glucose will bepositive, the NNS to detect each true positive willbe lower and the whole programme will be morecost-effective.

1. Age is always a key factor, because the risk ofT2DM increases steeply with age. The cost-effectiveness of screening will be lower atyounger ages since the NNS to find each casewill increase, and also because the event ratefrom CVD will be lower.However, although the prevalence of diabetes isgreater in the older ages groups, the excessmortality may fall. Tan and colleagues102 foundthat in men diagnosed with T2DM over the ageof 65 years in Tayside, there was no excessmortality compared with the generalpopulation. This might represent a survivalselection effect. The situation in women wasdifferent, with an RR of death of 1.29 (95% CI 1.15 to 1.45). The implication of this might be that if the main aim of screeningis to reduce heart disease mortality andmorbidity, screening in men should not includethose aged over 65 years. Vijan and colleagues103 also looked at this intheir model of T2DM. Improving glycaemiacontrol improves life expectancy, but the gainsfrom a 2% improvement (HbA1c 11–9%) arevery different at different ages of onset(Table 10).

This modelling is based mainly onmicrovascular disease, and is more concernedwith poorly controlled patients with higherHbA1c levels than those who would be found byscreening. However, it does illustrate a generalprinciple that because most seriousmicrovascular disease comes on a decade or twoafter the onset of diabetes, those who becomediabetic at older ages have less risk and henceless to gain from screening. Hence one couldargue that screening at 65 years or over is notworthwhile. It should be noted that in Table 11the insulin-treated group are a mixture ofpatients with type 1 and type 2 diabetes.

2. BMI is the second factor, reflecting overweightand obesity. The risk of T2DM is greatlyincreased by excess weight. However, there isalso a link with the distribution of body fat, withabdominal (especially visceral) fat distributioncarrying a higher risk. Waist measurementcould be used as a risk factor, for example,more than 40 inches (approximately 100 cm) inmen or 35 inches (approximately 90 cm) inwomen. However, men tend to know their waistmeasurement, by which they buy trousers, morethan women, who use overall size.

3. Co-morbidities. The risk of diabetes isassociated with other aspects of the metabolicsyndrome such as hypertension andhyperlipidaemia, and with the presence ofvascular disease, such as peripheral vasculardisease or IHD.

4. Family history of diabetes, or of prematurevascular disease or hypertension, is a predictor.

5. Ethnicity is also a predictor, in that some ethnicgroups have a higher risk of T2DM thanothers, although this is less if adjustments aremade for BMI and fat distribution. In theManchester survey,104 the prevalence of knowndiabetes in a poor inner city area was 8 and3.7% in European men and women,


19


Chapter 3

Screening tests

TABLE 10 Life-years gained by age of onset103

Age of onset (years) Life-years gained

45 1.355 0.965 0.575 0.2

respectively. and 14 and 18.2% in Pakistanimen and women, respectively. The Pakistaniwomen had higher BMI than the Europeans(29.6 versus 27.2) and a higher waist:hip ratio(0.88 versus 0.81). Pakistani and European menhad similar BMIs (27.4 and 27.5, respectively),but the Pakistani waist:hip ratio was higher(0.96 versus 0.92; 95% CIs 0.94 to 0.97 and0.92 to 0.94). However, the most strikingdifferences were in physical activity. Theproportions taking at least 20 minutes ofexercise three times per week were 38 and 29%for European men and women, respectively,and 7 and 5% for Pakistani men and women,respectively. Physical activity reduces insulinresistance even if there is little or no weightloss.98

6. Drug treatments such as for hypertension, orcorticosteroids.

There are two ways in which such selection criteriacould be applied. Questionnaires could be sentout, albeit perhaps only to those above a certainage. However, increasingly, computer systems ingeneral practices will have much of the necessarydata – certainly age, drug treatment, co-morbidities and usually BMI. They are less likelyto have waist measurements and family history.However, it means that the first stage of anyscreening system could use existing data at littleextra cost. This was demonstrated by Baan andcolleagues105 in the Rotterdam study, where datafrom a population screening study were used inthree progressively more complex screening tools:(a) Level 1 had only data available in general

practice records, age, sex, use of anti-hypertensive and anti-hyperlipidaemiamedications, previous gestational diabetes,CVD and obesity.

(b) Level 2 had additional items which could beobtained from a simple questionnaire, familyhistory, smoking, cardiovascular symptomsand exercise (cycling).

(c) Level 3 added data that would need to beobtained from physical examination, bloodpressure and waist and hip measurements (forwaist:hip ratio).

Baan and colleagues105 found that the addition oflevel 3 data did not improve the predictive valueover level 2. Level 2 had little advantage over level1, despite physical activity having a strongpredictive power. In level 2, not cycling had an ORof 2.83, which was the highest OR of all thefactors.

Patients at higher risk could then be approachedopportunistically when they consulted for otherreasons. Those who did not consult within, say,2 years of the threshold age could be invited toattend for screening.

Hence one option for the first stage could be torely only on data already in general practicerecords, which would provide an inexpensivemethod. However, questionnaires could providemore data.

QuestionnairesSeveral systems have been used.

The Hoorn study106 used the Symptom RiskQuestionnaire (SRQ), which, despite the name,covered more than symptoms, age, height, weight,pain on walking, shortness of breath, frequentthirst, family history of diabetes, use of anti-hypertensive drugs and cycling. They then used acut-off of >6 mmol/l for selection for glucosetesting.

Herman and colleagues107 used data fromNHANES 2 to examine different combinations ofrisk factors. The combination of age, sex, previousdelivery of large infant, obesity, lack of exerciseand family history identified a high-risk groupwhich contained 79% of people with newly

Screening tests

20

TABLE 11 Prevalence of insulin- and non-insulin-treated diabetes per 1000 population, 1998

Age (years)

Diabetes Sex 16–24 25–34 35–44 45–54 55–64 65–74 75–84 85+

Insulin-treated Males 3.5 4.6 6.2 7.2 10 13.3 10.9 6.8Females 3.2 4.3 5.2 5.7 9.4 12.1 9.4 5.9

Non-insulin-treated Males 0.2 0.6 3.6 11.8 30.5 47.5 47.4 43.1Females 0.2 0.6 2.8 7.9 20.3 35.7 37.1 33.8

Source: Office of National Statistics.

diagnosed diabetes. Adding hypertension and ahistory of glucose intolerance made littledifference (83% sensitivity). Ethnic difference hadlittle effect once other factors such as weight hadbeen allowed for (although in this case the ethnicgroups were black people, Hispanics andAmerican Indians).

Burden and Burden108 reported that the ADAquestionnaire had low sensitivity (46%) in a studyin Leicestershire for detecting those with raisedcasual BG levels.

The Diabetes Risk Score (DRS) is designed to beself-administered, and so could be sent out bypractices to people at the appropriate age. It hasquestions on age, BMI, waist circumference,medications for hypertension, any history of highBG, physical activity and daily consumption ofvegetables, fruit or berries. The IGLOO group(Impaired Glucose Tolerance and Long-termOutcomes Observational study)109 used the DRSand compared the findings with the results ofOGTT, in people selected on the basis of havingone or more risk factors for CVD, and being aged55–74 years. Of 1377 patients, but without knowndiabetes, 15.4% had IGT, 11.1% had IFG, 11%had IGT and IFG and 17.4% had diabetes.

The cost per case detected, irrespective ofmethod, was low (under €23). However, theyapplied their results to scenarios of 1000 peoplewith different prevalences of undiagnosed T2DMand IGT. For the 10% prevalence (which is closerto the UK one), they concluded that using theDRS would reduce the proportion of patientsneeding a fasting blood glucose (FBG) test byabout 43%, and then OGTTs by about 90 people(from 270 to 180). The total cost would fall byabout one-third, and the cost per case detectedfrom €87 to €65, but there would be a drop insensitivity from 68 to 55%. This is mainly becausea DRS cut-off of 9 gives a sensitivity of 86% fordiabetes but 77% for diabetes + IGT. It is worthnoting that a DRS cut-off of 9 is more sensitive fordiabetes + IGT than an FBG with cut-off of6.1 mmol/l or more.

Griffin and colleagues110 developed a new riskscore for diabetes, the Cambridge Risk Score(CRS), based on data available in general practicerecords (assuming that family and smoking historyare recorded), and then validated it on anothergroup of people. Using a low threshold, 30% ofthe practice population in Ely would go on todiagnostic testing and 77% of those withundiagnosed diabetes would be detected. Using a

higher threshold, 15% would be called fordiagnostic testing but only 41% of theundiagnosed diabetics would be detected. Hence,as usual, there is a trade-off between sensitivityand cost, with cost including not just money butalso the adverse effects of screening amongst thefalse positives. Griffin and colleagues110 also notethe danger of normal results in those at high riskbecause of unhealthy lifestyles – they may see anormal result as justifying continuing with suchlifestyles.

A different approach was used by Spijkerman andcolleagues.111 They used a three-step approach,with step 1 being an SRQ, step 2 being an FPGand step 3 being an OGTT (unless the FPG was sohigh that an OGTT was not considered necessary).The aim of the study was not to assess theaccuracy of the SRQ (already done in a previousstudy), and those with scores <6 were not testedfurther. The usefulness for this review is two-fold.First, the group identified by screening had alowish HbA1c of 6.7% [standard deviation (SD)1.4]. Second, they had high prevalences ofhypertension (70%), high cholesterol levels (73%)and obesity (40%). Hence many would have beenidentified as being at high risk of CVD evenwithout diabetes screening. If the aim of screeningfor diabetes is so that measures to reducecardiovascular risk, such as statins, can be started,then in this case screening may have made littledifference.

Greaves and colleagues112 also based initialselection on data now available in general practicecomputer systems. They tested a simple systemusing age and BMI only, with four groups selectedby the combination of age (over 70, over 65, over60 and over 50 years) and BMI (over 33, 31.29and 27). Hence the oldest and fattest groupexcluded the other three, but the other threegroups could overlap with those above. They theninvited patients to attend for FPG measurement.Those with FPG over 6.1 mmol/l were invited backfor a repeat FPG. Those with FPG at 7.0 mmol/l orover on both occasions were classed as diabetic.Those with one or two levels between 6.1 and6.9 mmol/l were classed as IFG. The uptake ofscreening was 61%.

Greaves and colleagues112 then looked at theefficiency of screening in terms of the NNS ineach group, and found that for diabetes it rangedfrom an NNS of 22 in the age over 70 years/BMIover 33 group to 38 in the 50+ years/BMI 27group. For detecting diabetes or IFG, therespective NNS were 8 and 13. There was little


21


difference in the oldest/heaviest three groups.Screening became slightly less efficient in the over50 years/BMI over 27 group. However, the threeoldest groups were reported as comprising 6.66%of the practice populations (which seems low) andso restricting screening to them would detect fewerpeople with hyperglycaemia than screening allover 50 years with BMI over 27.

The authors do not present data separately for thegroup with ages between 50 and 59 years BMIover 27 and under 29. Hence, the marginalbenefits of applying each age/BMI cut-off couldnot be examined. It is possible that most of thepatients found in the most inclusive group (over50 years/BMI over 27) were actually from thehigher groups.

The Atherosclerosis Risk in Communities (ARIC)Study113 looked at the marginal advantages ofadding laboratory data to a clinical risk score. Interms of area under a receiver operatingcharacteristic (AUROC) curve, the results were asfollows:

● Clinical risk score alone [age, waist measurement,parental history, ethnic group, systolic bloodpressure (SBP)] gave an AUROC of 0.71.

● FPG alone gave 0.74.● Clinical plus FPG gave 0.78.● Adding triglycerides and HDL cholesterol

increased that to 0.80.● Scores based on the metabolic syndrome gave

AUROCs of 0.75 and 0.78.

Hence adding laboratory results improved thepredictive power, but only slightly.

Schmidt and colleagues113 also tested the SanAntonio clinical score (age, sex, family history, BMI,HDL cholesterol and hypertension) and found thatthat gave an AUROC, for whites only, of 0.80.

Stage 2 – glucose testingUrine testing has been advocated, but in mostreviews has been discounted because of lowsensitivity.1,76,94 It would in effect be anotherpreliminary screen, with screen positives going onto blood testing. Nevertheless, it has been shownto be of some use, especially as it can be done bypost.114 However, it would be much less useful ifscreening were to be for all degrees of elevatedBG. It also retains a place in the investigation ofpolyuria, but by definition people with symptomsare not included in the screening situation.

The tests for blood glucose include;

● casual (non-fasting) BG – more sensitive thanglycosuria but less specific

● FPG or FBG● glucose tolerance tests, combining fasting and

2-hour levels● HbA1c, which reflects BG over the previous

3 months (assuming red blood cells of normallongevity, and in the absence of haemoglobinvariants)

● fructosamine● 1,5-anhydroglucitol.

Casual BG is usually discounted because of itsvariability and poor sensitivity (at levels which giveacceptable specificity).115

The choice of test depends on what one isscreening for. FPG is reliable, in the sense ofshowing less day-to-day variability than OGTTs,and will identify people with diabetes and IFG.However, it will miss those with IGT, who have ahigher IHD risk than those with IFG.

The use of HbA1cThe use of HbA1c as a screening test has beenreviewed by various expert groups.

The WHO Expert Group14 did not recommendHbA1c for diagnosis, commenting that:

An alternative to the single blood glucose estimation or OGTT has long been sought to simplify the diagnosis of diabetes. Glycatedhaemoglobin, reflecting average glycaemia over aperiod of weeks, was thought to provide such a test.Although in certain cases it gives equal or almostequal sensitivity and specificity to glucosemeasurement, it is not available in many parts of theworld and is not sufficiently well standardised for itsuse to be recommended at this time.

However, screening was not addressed in this report.Its perspective was world-wide; HbA1c is routinelyavailable in the UK. Systems for standardisationare available such as the National GlycohemoglobinStandardization Program (NGSP),116 which aims tostandardise results to make them comparable withthe Diabetes Control and Complications Trial(DCCT) Trial. Most UK laboratories use thissystem to align their results to DCCT.

In a recent report, the ADA Expert Committee117

on the diagnosis and classification of diabetesmellitus summarised the advantages anddisadvantages of HbA1c for the diagnosis ofdiabetes. The Committee listed the advantages as:

Screening tests

22

● HbA1c measures average glycaemic levels over aperiod of 10 weeks or so, and is therefore morestable than FPG, and especially than 2-h OGTT.

● Fasting is not required, and the test can be doneat any time of day.

● The precision of HbA1c can be as good as thatof PG (Note the use of the word “can” – is therean implication that results may not be as goodin routine practice?).

● HbA1c is the test used for monitoring control ofdiabetes and correlates well with themicrovascular complications; it may be useful touse the same test for diagnosis and monitoring.

● It has been shown by meta-analysis that whenusing a statistical cut-point of 2 SDs above thenon-diabetic mean value, HbA1c is as good asFPG and 2-hour PG in terms of sensitivity(66%) and specificity (98%).

The disadvantages were identified as:

● Internationally, there is a profusion of assaymethods and reference ranges. However, thiscan be overcome by standardisation to theDCCT assay.

● HbA1c may be affected by other conditionswhich affect the life of the red blood cell; resultsmay then be misleading. This could be aparticular problems in ethnic groups in whichhaemoglobinopathy is common.

● A chemical preparation for uniform calibrationstandards has only recently become availableand is not universally available.

However, with the exception of the otherconditions, these disadvantages need not apply ina national screening system. There is therefore acase for using HbA1c as the screening test,particularly in view of its correlation withcardiovascular risk across a wide spectrum. In theNorfolk study, Khaw and colleagues118 noted thatthe rise in cardiovascular events with rising HbA1cstarted well below the diabetic range. Indeed, theypointed out that when both diabetes and HbA1care included in the statistical analysis, HbA1cdominates, as Gerstein119 argues in an editorial:

the glycosylated hemoglobin level is an independentprogressive risk factor for cardiovascular events,regardless of diabetes status.

So, if the aim of screening is to identify people athigh risk of vascular disease, is HbA1c more usefulthan diabetes?

Rohlfing and colleagues120 noted that two changeswould affect the performance of HbA1c as a

screening test. The first was the standardisationsystem (NGSP) mentioned above. The second wasthe change in the definition of diabetes, with thelowering of the FPG threshold to 7.0 mmol/l.Using data from NHANES 3, they showed that acut-off of 1 SD above the HbA1c mean would give83% sensitivity and 84% specificity, compared withFPG. Using a cut-off of 2 SD would give 63%sensitivity and 97% specificity. The 2 SD cut-offlevel would be an HbA1c level of 6.1%.

One weakness of this study was that most patientsdid not have an OGTT. Of those who had HbA1c�6.1%, 96% had non-diabetic FPGs. The 4% withFPGs under 7.0 mmol/l but HbA1c over 6.1% mayhave included people with IGT.

A large-scale trial of HbA1c as a screening test wascarried out in New Zealand, somewhatserendipitously as an add-on to a large screeningprogramme for hepatitis B.121 In the NorthIsland, 50,819 subjects being tested for hepatitis Balso had HbA1c measured. Of these, 300 who livedclose to the testing laboratories were invited toattend for a standard 75 g OGTT. Most (82%) ofthe participants were Maori.

Thirty-two were found to have undiagnoseddiabetes. Of these, 30 had HbA1c over 6.1% – anapparent sensitivity of 94% and specificity of 77%,if the OGTT is taken as the gold standard. A totalof 67 (22%) people had FG 6.1–6.9 mmol/l; only33 of these had HbA1c over 6.1%. To detect thosewith IFG, a lower HbA1c cut-off would be required.A cut-off of 5.6% would give a sensitivity of 98%but a poor specificity of 30%. A cut-off of 5.9%would give a more reasonable compromise withsensitivity 82% and specificity 73%.

McCance and colleagues122 compared FPG, 2-hourPG and HbA1c as diagnostic tests in Pima Indians,and found very little difference in the predictivepower for retinopathy of FPG, 2-hour PG andHbA1c.

Peters and colleagues123 carried out a meta-analysisof the use of HbA1c (and also HbA1) in thediagnosis of diabetes. They found that a cut-pointof 7.0% for HbA1c would give a group in whom89% were diabetic by the OGTT, 7% had IGT and4% were normal. Of those with diabetes by theOGTT, only 42% had an HbA1c over 7%, but theauthors argue that this cut-off identifies those whowould be likely to require pharmacologicaltreatment – although this is an assumption, basedon a position that those with HbA1c under 7% arerarely treated with drugs. They considered that


23


the inconvenience of the OGTT might be one ofthe reasons for under-diagnosis of diabetes.

Woerle and colleagues124 reported that the 2-hourPG explained more of rises in HbA1c than FPG.They also give data on further testing in thosewith HbA1c levels between 6.0 and 7.0%. Nonehad normal glucose tolerance (NGT); 43% hadIGT, 49% were diabetic and 8% had IFG. However,numbers were small (23, 26 and four, respectively).There was a steep rise in the proportion abnormalbetween the sixth and seventh deciles of HbA1c:

● Sixth decile: mean HbA1c 5.34 %; 30% hadabnormal glucose tolerance.

● Seventh decile: mean HbA1c 5.48%; 70% hadabnormal glucose tolerance.

Hence an HbA1c of 5.5% might be a suitable cut-off for screening.

One advantage of HbA1c is that it is used as thetarget for glycaemic control. The ADA PositionStatement of 200597 recommends that the aimshould be an HbA1c level under 7% (referenced toa non-diabetic range of 4.0–6.0%, using a DCCTaligned assay). However, the relationship betweencomplications and HbA1c is closer formicrovascular disease than macrovascular disease.

HbA1c has become part of the routine of diabetescare. However, some concerns remain, and auseful critical review by Jeffcoate125 summarisesthese as follows:

● Analytical variability – which should be reducedby standardisation methods.

● Biological variability, in particular that due todifferences in the longevity of the red bloodcell, but also uncertainties over the relativecontributions to HbA1c of preprandial (such asfasting) and postprandial BG elevations. Henotes an ADA conclusion that FPG is a closercorrelate than postprandial.

● Clinical variability – and in particular thepoorer correlation with macrovascular disease.

If postprandial glucose levels are better predictorsof macrovascular disease than FPG, and if HbA1cis more influenced by preprandial thanpostprandial PG, then using HbA1c as a screeningtest might be fine for predicting and preventingfuture diabetes, but not so good as a basis forpredicting and preventing future heart disease.

However, the relative contributions of fasting andpreprandial (taking fasting to be before breakfast

and preprandial to be before other meals) andpostprandial to overall glycaemia, as reflected inHbA1c, varies according to stage of disease.Monnier and colleagues126 reported that thehigher the HbA1c, the greater is the contributionof fasting and preprandial – which is logical, since most people spend most of the day in a non-postprandial state. At low levels of HbA1c,under 7.3%, postprandial has twice as much effect on HbA1c as fasting and preprandial. In the lowest quintile (mean HbA1c of 6.45%),postprandial glucose contributed 70% of theelevation in HbA1c.

This has implications for using HbA1c as ascreening method, since at the lower levelsexpected in those found by screening,postprandial will have more effect, and thereforethe level may reflect an IGT pattern more than anIFG pattern, which could be useful.

Do we actually need blood glucose testing?An interesting paper from the Hoorn study127

looked at those who were first-stage screen positiveby the CRS, but who did not have diabetes onglucose testing. Spijkerman and colleagues127

found that this group had almost as high a CVDrate as the true positives. Taking the RR in truenegatives as 1.0, the newly diagnosed diabeticshad an RR of 1.73 and the false positives had anRR of 1.56. Furthermore, the false positives madeup about half of the population, so their numberof events was much greater than that of the truepositives (175 versus 24 events). As the authorsnoted, screening might do harm to this group:

In current screening practice this large group ofpeople would be sent home feeling reassured aboutdiabetes but unaware of their increased risk ofmortality.

They might then feel less inclined to improvetheir lifestyles. The CRS includes remediablefactors in BMI and smoking. Given that this grouphave been identified by the CRS as being atincreased risk, they could be reminded of thatpost-screening and given appropriate advice.Would that lead to lifestyle changes? Perhapsresearch is needed.

Since 63% of the total deaths occurred in thefalse-positive group, the authors comment that:

It may be of greater public health benefit to intervenein the screen positive group as a whole rather thanonly in the relatively small group who on subsequentbiochemical testing have an increased glucoseconcentration.

Screening tests

24


25


If so, it could be argued that glucose testing is notnecessary. However, this is based on amacrovascular focus. Detection of diabetes wouldlead to screening for and intervention in the(admittedly less common) microvascular diseasesuch as retinopathy.

ConclusionsThere is no perfect test. The gold standard testmight be the OGTT (or the modified version withjust FPG and 2-hour PG) but repeated a weeklater, because of its imperfect reproducibility.However, it is impractical and, as Hanson andcolleagues128 pointed out, the emphasis on theOGTT may be part of the reason why so manypeople in the USA are undiagnosed. A slightly less

good test may in practice be more useful by beingapplied more frequently.

As regards evidence on tests, there is little tochange since the last NSC review by Wareham andGriffin in 2001.1 The FPG and the 2-hour PG areequally useful for assessing the risk ofmicrovascular complications such as retinopathy,but the 2-hour PG level is better for assessingmacrovascular risk, because of the difference inheart disease risk between IFG and IGT. HbA1chas advantages in terms of convenience andreproducibility compared with the OGTT or itsmodified form, the 2-hour PG. FPG is also morereproducible than the OGTT.

In practice, HbA1c may be the best test, but costsmore than FPG, although less than OGTT.

OverviewThe aim of this chapter is to identify and appraiseeconomic studies relevant to the decision ofwhether or not to screen for undiagnosed diabetesand or IGT. The chapter is divided into threesections based around three principal questions:

1. Is it a cost-effective use of resources to screenfor and treat people with undiagnosed diabetes?

2. Is it a cost-effective use of resources to screenfor and treat people with IGT?

3. If a screening programme were to beimplemented, what screening tests and cut-offpoints should be used?

First, modelling studies that assess the long-termcost and consequences of screening for T2DM arereviewed. Second, similar models that assess thelong-term costs and consequences of treatingpeople with IGT or IFG are reviewed. Finally,studies that consider the short-terms costs andoutcomes of alternative screening tests (and cut-offpoints) for diabetes and IGT/IFG are reviewed.

MethodsA systematic literature search (up to the end ofJune 2005) was undertaken to identify anyeconomic assessments of screening for diabetesand/or IGT/IFG. Databases searched wereMEDLINE, EMBASE, NHSEED and Science andSocial Science Citation Index; no time limit wasset but there was little prior to the first modelbeing published in 1998. Abstracts were reviewedand any studies that were potentially relevant toany of the three questions outlined above wereretrieved. Articles cited by other relevant studieswere also retrieved for review. In addition, abroader search was conducted for any economicmodels within the area of T2DM, to ascertain ifany such models had been used to address any ofthe questions of interest.

Inclusion and exclusion criteriaTo be included, studies generally had to assess thecosts and outcomes (long or short term) ofscreening strategies for either undiagnoseddiabetes or IGT/IFG. However, given the lack ofeconomic evaluations that had assessed screeningfor undiagnosed diabetes (question 1), studies that

assessed only the long-term outcomes [life-years orquality-adjusted life-years (QALYs)] or costs ofscreening (where relevant to the UK setting) werealso considered eligible for inclusion. In addition,no studies explicitly assessed the long-term costsand outcomes of screening for IGT/IFG, so studiesthat assessed the costs and outcomes of treatingpatients already identified as having IGT/IFG wereconsidered eligible for inclusion. Studies reportedin languages other than English were notincluded.

Data extractionData were extracted from the long-term modellingstudies of screening for diabetes or IGT under thefollowing headings:

1. author and year 2. decision problem (comparators, population,

setting, objectives) 3. cohort information (characteristics and

numbers)4. model structure, perspective and scope (basic

structure and assumptions) 5. modelling of disease progression (details of

structure and assumptions used for modellingdiabetes progression)

6. modelling of diabetes complications (details ofstructure and assumptions used to modelprogression of diabetes complications)

7. mortality (details of how mortality wasmodelled)

8. costs (details of costs considered in themodel).

9. outcomes (outcomes reported and methodsfor calculating life-years or QALYs)

10. findings (reported results of the base-caseanalysis)

11. sensitivity analysis (details and results of anysensitivity analysis conducted).

For the studies that assessed only the short-termcosts and outcomes of alternative screeningstrategies, data were extracted under the followingheadings:

1. author and year2. setting (country for which analysis was

conducted)3. objectives (specific questions addressed)


27


Chapter 4

Review of economic models and evaluations

4. strategies (alternative strategies considered)5. costs (cost included in the analysis, currency

and year)6. Outcomes considered7. time horizon and perspective of study8. results/authors’ conclusions (details of main

findings and authors’ conclusions).

Quality assessment of included studiesOne reviewer critically appraised all the economicevaluations included using the BMJ guidelines forreviewers of economic evaluations.129 Theeconomic models included in the review wereappraised by the same reviewer using a publishedchecklist for good practice in decision analyticmodelling in health technology assessment.130

Economic models assessing long-term costs and/or consequencesof screening for type 2 diabetesBackgroundProponents of screening for diabetes argue thatthe additional investments needed to implementscreening programmes are justified by thepotential reductions in future diabeticcomplications, mortality and associated treatmentcosts. However, although there is direct evidencethat various treatments are effective in reducingcomplications and mortality in people who havebeen clinically diagnosed with diabetes, there is nodirect evidence relating to the magnitude of anyfurther benefit that might be derived from startingthese treatments earlier, after detection byscreening. In the absence of direct evidence,economic models have been used to estimate thelong-term costs and outcomes associated withscreening. These models are based on availableevidence relating to the progression of diabetesand its complications and various assumptionsabout the impact of treating diabetes in its earlypreclinical phase.

Search resultsThe search for economic studies that assessed thelong-term costs and consequences of screening fordiabetes identified six potential studies. However,one of the models only considered the benefits ofscreening in terms of reduction in the incidence ofblindness, and another study considered only thelong-term costs of screening for a US population.This left four studies for inclusion in this sectionof the review – three that assessed the long-termcost-effectiveness of screening and one whichconsidered only the long-term outcomes ofimplementing a screening policy at the population

level. These are discussed and compared with eachother and previous diabetic modelling studiesbelow.

Statement of the decision problemThe characteristics of the four modelling studiesincluded in this review are summarised in Table 12.Three of the studies used similar Markov modelsto assess the long-term costs and outcomes ofscreening for T2DM.131–133 These models usedsimilar assumptions to a previously publishedmodel by Eastman and colleagues,134,135 butincorporated a screening module to assess theimpact of identifying and treating patients earlierthan they otherwise would have been. The Centerfor Disease Control Diabetes Cost-EffectivenessStudy Group (CDC) was the first to construct amodel assessing the impact of screening.131 Thismodel concentrated on the long-term costs andbenefits associated with the provision of glucose-controlling interventions in the early preclinicalphase of diabetes. Conventional treatments wereassumed to begin after early identification byscreening and were compared with the currentclinical practice in the USA at the time(conventional treatment commencing after clinicaldiagnosis). The authors assumed that screening,followed by early treatment for hyperglycaemia,would only impact upon the incidence ofmicrovascular complications and have no effect on CVD.

Chen and colleagues132 later used the same basicmodel structure as the CDC to investigate theimpact of repeated universal screening as opposedto once-off opportunistic screening. This studyalso addressed the question of how frequentlyscreening should be carried out.

More recently, Hoerger and colleagues133 updatedthe original CDC model to incorporate widerbenefits of screening. They used new data forvarious model parameters and incorporated thefinding that, for people with diabetes, tighterblood pressure control provides substantial benefit(relative to standard control) in terms of reducingCVD morbidity and mortality. The model assessedthe cost-effectiveness of targeting screening atpatients with hypertension compared with once-offopportunistic screening for all adults and thepractice of relying on clinical diagnosis.

The fourth study identified used a differentapproach to model the QALY gains associatedwith screening. Goyder and Irwig136 used adecision tree to weigh the potential benefitsagainst the potential harms of screening. The


28

baseline characteristics of the cohort were variedto ascertain in which populations the benefits ofscreening would be likely to outweigh the harm.They assumed that an early diagnosis withscreening would lead to improved treatment ofother CVD risk factors in addition to BG-controlling measures. Therefore, benefits in themodel (QALY gains) accrued from reductions inboth microvascular and cardiovascularcomplications. The potential harms of screeningincluded in the model were assumed reductions inquality of life associated with early diagnosis andadverse effects of the early treatment. None of theother studies considered potential harms ofscreening.

Cohort informationThe characteristics of the cohorts for which thesemodelling studies were conducted are summarisedin Table 13. The CDC group131 simulatedscreening for a hypothetical population with thedemographic characteristics of the US populationover 25 years of age. The population was assignedto either screening or current clinical practice. For 10,000 people with diabetes within thispopulation, the disease progression modelsimulated the development of complications fromonset of diabetes under each of the alternativeoptions (screening and current practice). Differentscreening cohorts defined by age (10-year age

groups), race and ethnicity are also consideredseparately in the model.

The cohort simulated by Hoerger and colleagues133

was based on 1997 population estimates projectedfrom the 1990 US Census and data on thedistribution of people with diabetes byhypertension, cholesterol level and smoking status.Different cohorts defined by age, sex, race andhypertension status were simulated through themodel using cohort simulation (all at once) ratherthan Monte Carlo simulation methods (onepatient at a time).

The cohort used by Chen and colleagues132

consisted of 30,000 individuals with demographiccharacteristics reflecting the population of Taiwan.

The decision tree analysis by Goyder and Irwig136

was conducted for a predominantly Caucasian UKcohort aged between 45 and 60 years. Theprevalence of diabetes in this cohort was based ona prevalence survey of the population of the Isle of Ely.

Model structure, scope and perspectiveThe CDC131 developed a semi-Markov model andused Monte Carlo simulation to model diseaseprogression, treatment costs, life-years and QALYsfor the two alternative options considered –


29


TABLE 12 Summary of screening models reviewed

Study Comparators Model type Economic Complications Benefits of outcomes modelled early treatment

modelled

CDC, 1998131 1. Clinical diagnosis (case Markov Cost per Retinopathy, Reduced finding) (Monte Carlo life-year and nephropathy, microvascular

2. Once-off opportunistic simulation) cost per QALY neuropathy, complicationsscreening all CVD

Chen et al., 1. Clinical diagnosis (case Markov Cost per Retinopathy, Reduced 2001132 finding) (Monte Carlo life-year and nephropathy, microvascular

2. Universal repeat simulation cost per QALY neuropathy, complicationsscreening assumed) all CVD

Hoerger et al., 1. Clinical diagnosis (case Markov Cost per QALY Retinopathy, Reduced 2004133 finding) (cohort analysis) nephropathy, microvascular

2. Once-off opportunistic neuropathy, and screening CHD (angina and cardiovascular

3. Targeted screening to MI), stroke complicationspeople with hypertension

Goyder and 1. Clinical diagnosis (case Decision tree QALYs gained All microvascular Reduced Irwig, 2000136 finding) (cohort analysis) complications microvascular

2. Screening for all (not distinguished), and 45–50-year-olds all CVD (including cardiovascular

angina, MI, chronic complicationsheart failure and stroke)

opportunistic screening of all adults versusreliance on clinical diagnosis. The diseaseprogression module consisted of a series ofMarkov processes, which were used tosimultaneously model disease progression on threediabetes complication pathways starting at onset ofdiabetes. The complication pathways modelledwere retinopathy, nephropathy and neuropathy.Individuals from the hypothetical cohort weresimulated one at a time through the screeningmodule, where they were first assigned variouscharacteristics weighted on the demographiccharacteristics of the US population. They werealso assigned a diabetes status (undiagnosed)based on the prevalence of undiagnosed diabetesin the US population (adjusted for age, race,ethnicity, obesity and hypertension status).Individuals with diabetes in the screening arm ofthe model could either be identified by screening(true positives) or missed (false negatives) basedon the sensitivity/specificity of the test (FPG testwith a cut-off value 6 mmol/l). A total of 10,000patients with diabetes, as assigned by the model,were then simulated one at a time through thedisease progression submodels. Patients whoreceived a diagnosis of diabetes as a result ofscreening were assumed to begin conventionalglucose-controlling treatment (diet and exercise)immediately, whereas those given a false-negativetest result began treatment at clinical diagnosis.All those with diabetes in the no-screening armwere assumed to begin treatment at clinicaldiagnosis. It was assumed that on average clinicaldiagnosis would occur 10.5 years after diabetesonset. This estimate was based on a study whichlooked at the relationship between the prevalenceof retinopathy and time from clinical diagnosisand used extrapolation to estimate the time at

which the prevalence would be zero (9–12 yearsbefore clinical diagnosis).54 Diagnosis by screeningwas assumed to occur 5 years earlier at 5.5 yearsafter onset (Figure 1). Those identified throughscreening were simulated to receive an extra 5 years of treatment compared with thosediagnosed clinically and, thus, have a slower rate ofprogression in the disease progression submodels.

The model used by Chen and colleagues132 hasthe same basic structure as the CDC model,although it is unclear if cohort simulation orMonte Carlo simulation was used for the analysis.Very little information was provided in thepublished paper regarding the structure of themodel and the methods for estimation ofparameter values and transition probabilities. Thislack of transparency makes it very difficult toassess the quality of the model. The authors usedan estimate of 1.1% for the annual incidence ofdiabetes, which does not seem to have been variedas a function of age, sex or any other patientcharacteristics. A hypothetical cohort of 30,000subjects was randomly allocated to one of twoscreening arms (screen every 2 years or screenevery 5 years) or a control arm (no screening).The screening test used was FBG but it is unclearfrom the paper what cut-off point was used andwhat sensitivity/specificity values were assumed.Presumably, screening was assumed to result indiagnosis of the incident diabetes cases at a timefrom onset consistent with the screening intervals.However, it was unclear whether or not theanalysts considered the impact that introducingmass screening would have on the undiagnosedprevalent cases in the population. Furthermore,the time from onset at which clinical diagnosis wasassumed to occur in the absence of screening was


30

TABLE 13 Cohort information used in the published screening models

Study Cohort demographic Source of cohort Cohort age range No. of patients characteristics based on information (years) in cohort

CDC, 1998131 US population National Health and 25 and over 10,000Nutrition Examination Survey II

Chen et al., 2001132 Taiwanese population Not stated 30 and over 30,000

Hoerger et al., 2004133 US population 1997 population Separate cohort Not statedestimates projected analyses for from the 1990 US screening at 35, 45, Census and data on 55, 65 and 75distribution of people with diabetes in the USA

Goyder and Irwig, Caucasian UK population Isle of Ely Diabetes 45–60 10,0002000136 Project

not stated. However, it was assumed that screeningwould result in early treatment that would slow thedevelopment of microvascular complications.

The model conducted by Hoerger and colleagues133

followed the same basic structure as that of theCDC but differed in number of important ways.First, cohort simulation was used to evaluate themodel as opposed to Monte Carlo simulation.This means that whole cohorts were passedthrough the model simultaneously, as opposed toone individual at a time. The model was evaluatedfor different screening cohorts defined by age,hypertension status and race/ethnicity separately.As shown in Figure 1, there were also slightdifferences in the assumptions about whendiagnosis would occur in the presence (average of5 years after onset as opposed to 5.5 years) andabsence of screening (average of 10 years afteronset as opposed to 10.5 years). Hoerger andcolleagues133 also made the assumption that upondiagnosis individuals would immediately receiveintensive glycaemic control, as defined in theUKPDS trial,8 as opposed to the conventionaltreatment assumed in the CDC model.Furthermore, all hypertensive individuals thatreceived a diagnosis of diabetes were assumed tochange from standard treatment for hypertensionto more intensive therapy [target diastolic bloodpressure (DBP) of 80 mmHg as opposed to90 mmHg].83 The disease progression module ofthe model used by Hoerger and colleagues133

differed from that of the CDC model with respectto added complication pathways for CHD andstroke.

The three Markov models all took the perspectiveof a single healthcare payer and incorporatedinputs consistent with this perspective. The inputsincluded were sensitivity/specificity of screeningtests, costs of the screening tests, transitionprobabilities relating to the progression ofdiabetes and its complications, diabetes- and non-diabetes-related mortality risks, utilities and datarelating to the effectiveness of glucose- andhypertension-controlling treatments. The modelsall adopted a lifetime horizon and assessed thelifetime incidence of diabetes-relatedcomplications, and also life-years and QALYs, forage- and race-specific cohorts under the differentscreening and no-screening scenarios.

The decision tree model developed by Goyder andIrwig136 was used to assess the short-term harmsand long-term benefits of once-off screening forpeople aged between 45 and 60 years. The cohortwas assigned to screening or no screening. Thoseassigned to screening were modelled to receive asingle FBG test (using a cut-off giving sensitivity90%), followed by a gold standard diagnostic testfor those screening positive. A prevalence ofdiagnosed diabetes of 4% was assigned to thecohort based on the findings of a prevalence studyconducted on the Isle of Ely.137 For every prevalent


31


CDC model131

Chen et al.132

Hoerger et al.133

Goyder and Irwig136

Time (years)

Diagnosis without screeningDiagnosis with screeningDiabetes onset(t = 0)

3 5 100

FIGURE 1 Screening module summaries (indicates time after onset at which diagnosis is assumed to occur in the presence andabsence of screening for each of the models)

case of diagnosed diabetes, it was assumed thatthere would be an undiagnosed case, based on ameeting of an expert committee on diagnosis andclassification of diabetes. Therefore, theprobability of diabetes being undiagnosed at thetime of screening was estimated to be 50%. Themodel assumed that treatment for diabetes andother cardiovascular risk factors (e.g. hypertensionor hyperlipidaemia) would be implemented upondiagnosis, thus reducing morbidity and mortalityfrom microvascular and cardiovascularcomplications. The authors made the assumptionthat early treatment for cardiovascular risk factorswould only reduce the incidence of cardiovascularevents within the time from screening to the timeclinical diagnoses would occur in the absence ofscreening. After the time of clinical diagnosis, itwas assumed that the incidence of cardiovascularevents would be the same regardless of whether ornot early treatment of risk factors was received.This may be an inappropriate assumption if theincidence of CVD is a function of the durationthat risk factors have been present.

Goyder and Irwig136 made two assumptionsregarding the impact that screening would haveon the time from onset at which diagnosis wouldoccur. First, they assumed that 50% of those whoare undiagnosed in the population would neverreceive a clinical diagnosis in the absence ofscreening – that is, they would die of mainlycardiovascular causes before receiving a diagnosisand before any end-stage microvascularcomplications would occur. Hence, screening inthis first group of patients would not result ingains from reductions in the incidence inmicrovascular complications, only gains due toreductions in CVD risks. This differs somewhatfrom the assumptions made in the Markovmodels, that all the undiagnosed patients in thepopulation would receive a diagnosis 10 yearsafter onset unless they died before this time. IfGoyder and Irwig’s136 assumption is correct, itmay be that the CDC model overestimates theimpact that screening would have on reducingmicrovascular complications. It adds weight to theargument that more benefit can be derived fromscreening by treating risk factors for CVD in thosescreening positive, rather than from reducingmicrovascular complications. However, thisassumes that it is more cost-effective to diagnosediabetes and then treat risk factors rather than justtreat risk factors in those who have them withoutfirst screening for diabetes. The secondassumption made by Goyder and Irwig136 is thatfor the 50% of undiagnosed patients who wouldreceive a clinical diagnosis in the absence of

screening, this would occur on average 6 yearsafter onset. This is shorter than the 10/10.5 yearsassumed in the other models. The effect is thatscreening results in a shorter window ofopportunity (3 years as opposed to 5 years) toprovide extra treatment from which patients cangain extra benefit (Figure 1).

Modelling of disease progressionIn the three Markov models,131–133 all patients areassumed to have no complications at onset ofdiabetes. In the CDC model, progression ofcomplications in the early prediagnosis phase ismodelled as a function of the prevalence ofcomplications observed at clinical diagnosis invarious epidemiological studies.54,91,138 Hoergerand co-workers133 only state that they madeassumptions about transition probabilities betweendisease states in the period before usual clinicaldiagnosis based on knowledge that progression isrelatively slow during this phase. After the timewhen clinical diagnosis would usually occur,disease progression is modelled in relation to thetime from normal clinical diagnosis based on thefindings of various epidemiological studies (seebelow). Some transition probabilities in theoriginal CDC model are also adjusted forrace/ethnicity to permit the estimation of cost-effectiveness of screening in different racial/ethnicgroups. Transition probabilities for race/ethnicitywere adjusted using multipliers derived byEastman and colleagues134,135 to reflect the higherincidence rates of complications observed inAfrican Americans, American Indians andHispanic Americans. In the analysis by Hoergerand colleagues,133 the cost-effectiveness ofscreening by different racial/ethnic groups was notexplored. Little detail was given on howprogression was modelled before the time ofnormal clinical diagnosis in the model by Chenand Colleagues,132 but the transition probabilitiesare reported to vary by time.

In the models by the CDC and Hoerger andcolleagues, the transition probabilities forprogression of microvascular complications werealso adjusted for glycaemic control – as measuredby HbA1c levels. Initiation of treatment afterscreening or clinical diagnosis reduces the HbA1clevel and thus reduces the transition probabilitiesfor disease progression. Adjustments for HbA1clevels were made using a power function derivedby Eastman and colleagues.134 As evidence relatingto the benefits of reducing HbA1c in T2DMpatients was lacking at the time of the CDCanalysis, the relative benefits (reduced incidence)per unit difference in HbA1c levels were derived


32

from a trial of treatments in patients with type 1diabetes.139 Hoerger and colleagues133 updatedthis using more recent data on the effects ofintensive glycaemic control on disease progressionin patients enrolled in the UKPDS trial of T2DM.8

This reportedly results in smaller reductions in theincidence of complications per unit difference inHbA1c level, compared with the original CDCmodel.

Figure 2 gives an indication of how the HbA1clevels were modelled over time in the studies bythe CDC131 and Hoerger and colleagues.133 Thefigure is based on our own calculations using dataprovided in the papers. In the original CDCmodel,131 the HbA1c levels were assumed to be6.8% at onset of diabetes with an annual increaseof 0.2% points resulting in a level of 8.9% atclinical diagnosis (consistent with data reported inthe UKPDS)140 and 7.8% at diagnosis by screening.From the authors’ report, it appears that theHbA1c level was assumed to drop by 2.1% pointson initiation of treatment and then increase slowlyagain by 0.156% points per year – again based onan earlier report of the UKPDS.141 The HbA1clevels were not allowed to go above 11% or below6% with treatment. Patients were modelled toreceive one of four treatment modalities: diet andexercise, oral hypoglycaemic agents, oralhypoglycaemic agents with insulin or insulin alone.The proportion using each treatment modality

varied by duration of diabetes, but exactly howeach treatment affected HbA1c levels and howHbA1c levels were used to adjust transitionprobabilities for disease progression were unclearfrom the report.

In the updated model by Hoerger andcolleagues,133 the same approach was used tomodel the HbA1c levels, except that more intensivetreatment upon diagnosis resulted in a greater fallof 2.9% points in HbA1c levels.8 In addition, afterthe initiation of treatment in the model byHoerger and colleagues,133 the HbA1c level isincreased by 0.2% points per year (as opposed to0.156% points) but is never allowed to rise above9%. Transition probabilities for the developmentof CHD and stroke were also adjusted for thepresence of hypertension and intensity ofhypertensive treatment in the model by Hoergerand colleagues (see below).

It was not possible to ascertain the approach usedto model the beneficial effects of implementingtreatment early in the model by Chen andcolleagues.132 The authors only stated thattreatment efficacy parameters were based on thoseused in Eastman and colleagues’ model andfindings from the UKPDS.

Goyder and Irwig136 populated their decision treemodel with probabilities derived from literature


33


4

5

6

7

8

9

10

11

1 6 11 16 21 26

Time (years)

HbA

1c (%

)

Clinical diagnosis (CDC model)

Screening diagnosis (CDC model)

Clinical diagnosis (Hoerger et al.)

Screening diagnosis (Hoerger et al.)

FIGURE 2 HbA1c levels over time for the screening and no screening strategies in the models by the CDC131 and Hoerger andcolleagues133

sources and various assumptions. Although littleinformation was given in the published journalpaper as to how these probabilities were derived, atechnical report was obtained from the authorsproviding more detail. The shorter duration ofundiagnosed diabetes (6 years) used in the modelwas estimated by dividing estimates of theprevalence of undiagnosed diabetes (cases per1000 individuals) by the incidence (cases per 1000per year) as estimated from various populationsurveys.142–145 However, the authors point out thatthe incidence estimates were based on clinicallydiagnosed cases, which will not accurately reflectthe true incidence of diabetes, since many caseswill never receive a clinical diagnosis. In order toreflect the uncertainty surrounding this parameter,the authors examined the impact of a wide rangeof estimates in sensitivity analysis.

In terms of the risks of microvascular complications,Goyder and Irwig136 assumed a lifetime risk of13% for any complication in intensively treated,clinically diagnosed individuals. This was takenfrom Eastman and colleagues’ model134 based onthe results from the DCCT trial of treatment intype 1 diabetes. This may be inappropriate givenmore recent findings that suggest intensiveglycaemic control is less effective in reducing thelifetime risks of complications in T2DM as it is intype 1 diabetes.8 The authors assumed that theeffectiveness of early treatment to reduce HbA1cwould be 50% relative to the effectiveness oftreatment later in the disease course. They alsoassumed that the maximum effect of treatment inthe early post-screening phase (before clinicaldiagnosis) would be to postpone microvascularcomplications by 3 years (the assumed length ofthe early treatment window). However, since it wasassumed that treatment during this early periodwould only be half as effective as treatment aroundthe time complications develop, avoidablemicrovascular complications were assumed to bepostponed by 1.5 years with early treatment.

For cardiovascular complications, Goyder andIrwig136 assumed that the risk of CVD inundiagnosed and diagnosed cases would be thesame. In the base case this was assumed to be 2%per year. This was varied in sensitivity analysis toassess the impact that screening might have ingroups at higher risk of CVD. It was assumed thatan RR reduction of 33% could be achieved duringthe early treatment period by implementingtighter treatment for hypertension and orhyperlipidaemia. This assumption is based on thefindings from several trials showing theeffectiveness of these treatments in clinically

diagnosed patients.146–150 The applicability ofthese findings to patients detected by screening isuncertain. The authors also assumed that the CVDevents avoided as a result of early treatment wouldhave been survived by 15 years on average andthat the RR of CVD due to early treatment wouldlast only for the period before clinical diagnosiswould normally be made. After the time of normalclinical diagnosis, the CVD risk is modelled asbeing the same, regardless of whether or not earlytreatment had been received.

In terms of the time horizon of the model, theauthors assumed a mean survival time of 17 yearsfrom the time of clinical diagnosis based on themodel of Eastman and colleagues.134

Implementation of early treatment is assumed toimpact upon morbidity only, and has no effect onlife expectancy in the model. The authorsestimated that the impact of screening and earlytreatment on the development of microvascularcomplications would occur after 15 years.Cardiovascular complications avoided due to earlytreatment were also assumed to last for 15 years.

Structure of complication submodels(Markov models only)NephropathyThe three studies using Markov models adoptedthe same structure for their nephropathysubmodels. Patients follow a path through aMarkov process with four consecutive states: nonephropathy, microalbuminuria, proteinuria andend-stage renal disease (ESRD). In the CDCmodel, Monte Carlo simulation is used todetermine whether patients progress from onestate to the next or not. Hoerger and colleagues133

use a cohort simulation where the whole cohortprogresses at once. The simulation method usedby Chen and colleagues132 is unclear. In the earlypreclinical phase, the models by the CDC131 andHoerger and colleagues133 use transitionprobabilities that yield the prevalence ofcomplications observed at clinical diagnosis inseveral epidemiological studies.54,91,138 It isunclear in the model by Chen and colleagues132

how transition probabilities were modelled duringthis period. After the time of normal clinicaldiagnosis (10 years after onset), the CDC modeluses the same transition probabilities fordeveloping nephropathy complications as thoseused by Eastman and colleagues.135 Chen andcolleagues132 also use these same transitionprobabilities, although it is not possible to tell ifthese have been used appropriately, that is, afterthe time clinical diagnosis would normally occur.Eastman and colleagues134,135 give details in their


34

papers on how transition probabilities forprogression to each state in the pathway werecalculated by duration since clinical diagnosis, andmade conditional on the patient being in theimmediately prior state. Transition probabilitiesare also adjusted for race, ethnicity and glycaemiccontrol, as discussed above.

In Hoerger and colleagues’ updated version of themodel,133 some of the transition probabilities inthe nephropathy module are the same (e.g.proteinuria to end-stage renal disease), but othershave been changed without explanation. Forexample, the transition probability formicroalbuminuria to proteinuria is substantiallylower. The transition probabilities formicroalbuminuria and proteinuria also varydepending on hypertension status and theintensity of hypertensive treatment received(standard before diabetes diagnosis, intensiveafter) based on data from the UKPDS.151

NeuropathyThe neuropathy submodel adopted by all threeMarkov models again follows the same structure asthat used in the study by Eastman andcolleagues.135 The states included are noneuropathy, symptomatic neuropathy, history oflower extremity amputation (LEA) and death fromLEA. An individual with neuropathy canexperience an LEA during any 1-year cycle, fromwhich they either die (transit to death from LEA)or survive (transit to history of LEA). Patients witha history of LEA can also experience a second LEAfrom which they can again die or survive.Transitions between these states, after time whennormal clinical diagnosis would have occurred, areagain based on the transition probabilitiescalculated by Eastman and colleagues.134 Theycalculated these from data on the incidence ofneuropathy and LEAs reported in the RochesterDiabetic Neuropathy Study.152–154 The prevalenceof neuropathy at clinical diagnosis was assumed tobe 3.5% based on data from the NHANES IIsurvey.155 An annual hazard rate was then assignedwhich would result in a cumulative incidence at8 years after diagnosis that was consistent with theRochester study.152,153 Transition probabilities forprogression from symptomatic neuropathy to firstLEA were again based on the cumulative incidenceof this event observed in the Rochester study andmade conditional on the presence of neuropathy.The neuropathy module in the updated study byHoerger and colleagues133 is essentially the samebut the transition probability for peripheralneuropathy is slightly higher as calculated fromdata reported in the UKPDS study.151

Hypertension and its control are assumed to haveno effect on the development of neuropathy.

RetinopathyThe CDC group based their retinopathy submodelon the model by Eastman and colleagues.134 Thestates included are no retinopathy, non-proliferative retinopathy, proliferative retinopathy,significant macular oedema and blindness. Thetransition probabilities used for these events arethose reported by Eastman and colleagues.135

These were calculated appropriately using datafrom the Wisconsin Epidemiologic Study ofDiabetic Retinopathy156 and were again madedependent on duration of diabetes since clinicaldiagnosis. In the model, people who developproliferative retinopathy and receive eyeexaminations are assumed to be treated withappropriate photocoagulation, thus reducing theirrisk of progression to blindness. However, it is notclear what the probability of being identified by aneye examination is.

The retinopathy submodel used by Chen andcolleagues132 follows the same basic structure asthat of the CDC model, but the transitionprobabilities used appear to be different. Nodetails were provided by Chen and colleagues132

on how these were calculated.

The retinopathy module in the model by Hoerger and colleagues133 differs from the othertwo models in that only three states are included.These are normal, photocoagulation andblindness. Transition probabilities are updatedwith data from the UKPDS trial and also varyaccording to hypertension status and treatment.

Cardiovascular diseaseSpecific cardiovascular complications of diabetesare not modelled in the original CDC analysis.However, cumulative incidence of all CVD andCVD mortality is modelled. It is not entirely clearfrom the paper how this is done, but presumablythe same approach as by Eastman and colleaguesis used.134,135 In the latter model, each person isassigned CVD risk factors (smoking status andmean SBP, total cholesterol and HDL cholesterol),based on their age, race and sex by sampling from probability distributions. A previouslypublished multivariate model based on theFramingham Heart Study is then used to calculate the incidence of all CVD.157 For patients who develop ESRD, 50% are assumed tohave CVD based on the observation that 50% ofdeaths in patients with diabetes-related ESRD aredue to CVD. The model assumes that glycaemic


35


levels have no impact on CVD. This may be aninappropriate assumption and more recentlypublished risk equations for CVD in patients withdiabetes include HbA1c levels as a predictorvariable.158

The model by Chen and colleagues132 also follows this same structure and makes the sameassumptions about CVD in patients with ESRD.

The model described by Hoerger andcolleagues133 (see http://www.annals.org/cgi/content/figsonly/140/9/689 for additional details)includes a fairly complex submodel for CHD andstroke. It is based on previous published models ofCHD.159,160 States in the CHD model includenormal, angina, history of CHD/MI and deathfrom CHD. Within a 1-year cycle, individuals inthe normal state can experience a CHD event,which could be angina, or a fatal or non-fatal MI(MI modelled as state). The Framingham riskequations are used to calculate the probabilitiesfor these events based on the demographic andepidemiological characteristics of the cohort being analysed.157 Following an initial MI, patients can either die within 30 days, in whichcase they transit to the state death from CHD, or they can survive over 30 days. Patients whosurvive over 30 days can either die from chronicconditions related to MI within the year,experience a second MI within the year (fromwhich they have probabilities of dying orsurviving), or continue on with no further eventsfor the remainder of the year. Those who survivemove to the state history of MI. Patients whosefirst CHD event is angina can either survive withangina for the remainder of the year, orexperience a MI within the year. Those survivingwith angina transit to the angina state, whereasthose who have subsequent MI within the year can either survive (transit to history of MI) or die within the year (death from CHD). In thesubsequent cycle of the model, patients in each of the states have further chances of staying where they are, experiencing an initial orsubsequent MI, or dying. The probabilities for all the events in this model are based onpublished multivariate equations, whichincorporate age, various CVD risk factors andCHD history.157,159,160

The effects on blood pressure control aremodelled as reductions in the risks of CHD events.The efficacy of intensive hypertension control inpreventing CHD comes from the HOT trial, whichfound that relative to standard control, it reducesthe risk of CHD by 51%.83

The stroke module included in the model byHoerger and colleagues133 includes three states:normal, history of stroke and death. Within each1-year cycle individuals can either stay in thenormal state or experience a stroke from whichthey die or transit to having a history of stroke.Again, no details are provided for how risks ofthese events are calculated in the published paperof the screening model. The technical reportdescribes how the Framingham equation is used tocalculate the transition probability from normal tostroke. The other transition probabilities werecalculated appropriately from another publishedsource.161 In the base-case analysis, it was assumedthat intensive hypertension control has no effecton the incidence of stroke relative to standardcontrol, but a 30% RR reduction was explored insensitivity analysis, based on the findings ofstudies showing such benefits for intensivehypertension control.151,162

MortalityMortality in the CDC screening study is modelledas a competing risk for each of the majorcomplications of diabetes. Increased mortalityrates among people with diabetes are attributed toincreased mortality due to CVD, ESRD, LEA andother causes. A mortality rate is assigned toindividuals in a sequential fashion. If, for example,a patient undergoes an LEA during the year, themodel will assign that person a mortality ratebased on the anatomical level of amputation. If apatient does not have an LEA or survives theoperation, the model assigns that person eitherthe mortality rate for ESRD, if it is present, or amortality rate determined from the increased riskof CVD mortality plus non-CVD morality. Datasources used for the mortality risks for ESRD andLEA are given in the report, but no details aregiven for any calculations required. Thecardiovascular mortality rate is estimated usingequations from the Framingham Heart Study83 asa function of age, sex, SBP, total cholesterol level,HDL levels and left ventricular hypertrophy. It isnot clear how these characteristics are assignedand modelled over time for individuals in thecohort, but presumably the same approach as byEastman and colleagues is taken whereby patientcharacteristics are sampled from probabilitydistributions. The authors of the CDC model131

also assume that diabetic patients have anincreased risk of non-CDV mortality 2.75 timesthat of non-diabetic individuals, which ispresumably used to adjust age- and sex-specificmortality rates published in life tables. Eastmanand colleagues135 originally introduced this factorto achieve consistency with a reported 5–10-year


36

reduction in life expectancy observed for middle-aged people with diabetes.

Chen and colleagues132 provide too little detail intheir paper to ascertain how mortality fromcompeting causes was modelled.

Hoerger and colleagues133 model mortality fromLEA, ESRD, CHD, stroke and other causes.Details of how these events were modelled wereprovided in the technical report only. Mortalityestimates for CVD/CHD are based on equationsoriginally published by Weinstein andcolleagues159 and updated by Hunink andcolleagues.160 These equations predictprobabilities of CHD-related deaths based on age,sex and history of CHD events. It is not clear ifthese risks are directly relevant to patients withdiabetes, as diabetic patients may have greaterrisks of death following CHD events than non-diabetic patients. Diabetic patients in the cohortwho experience a stroke, LEA or ESRD also havea higher mortality risk than patients without thesecomplications. The death rates for LEA wereobtained from published literature154 andappropriately transformed for incorporation intothe model. Mortality rates from ESRD are specificfor the cohort’s age, sex and race/ethnicity.Finally, patients in the cohort can experiencedeath from other causes from any state in themodel. These rates are taken from life tables anddo not appear to be adjusted in the same way asin the original CDC study.131

CostsThe costs included in the original CDC model arethose for screening, glucose-controlling treatment(including drugs, self-monitoring costs, outpatientvisits), costs associated with microvascularcomplications, costs associated with CVD andother routine medical costs not specific todiabetes. Costs are presented in 1995 US dollarsand are discounted at a rate of 3%. Assumptionsregarding extra resources required for screeningand treatment are described and justified in themodel, and the unit costs used are presented andreferenced. However, it is very difficult tocomment on the appropriateness of these unitcosts given the lack of detail on how they werecalculated.

Chen and colleagues132 use exactly the same costdata (1995 US dollars) as used in the CDCmodel131 and also discount at a rate of 3%.However, they do not provide any information onhow treatment regimens are assumed to changeover time.

In addition to the costs considered in the CDCmodel, Hoerger and colleagues133 also include thecosts of standard and intensive hypertensioncontrol for patients with hypertension. Also, theupdated study assumes that intensive glycaemiccontrol would be implemented upon diagnosis ofdiabetes, thus increasing resource use and costsassociated with treatment. Costs in the updatedstudy are provided in 1997 US dollars and unitcosts are derived from various sources in theliterature. In some cases the sources are the sameas in the original CDC study,131 but in other casesdifferent sources have been used. It is not clearwhy such choices have been made, but a numberof costs in the earlier study were derived fromstudies of patients with type 1 diabetes, so themore recent estimates in the updated model maybe more appropriate. Costs in the updated studyby Hoerger and colleagues133 are also discountedat 3%.

OutcomesThe primary outcomes in the CDC model131 wereadditional life-years and QALYs measured fromonset of diabetes. Outcomes were discounted at3%. The scarcity of data available for utility scoresat the time this model was constructed meant thatcertain assumptions had to be made. Utility scoreswere assumed to be 1 for each year of life livedwithout a major end-stage complication. Utilityadjustments were only made for years lived withblindness (0.69),163 ESRD (0.61)164 and LEA(0.8).165 No details are given in the paper as tohow these utility scores were calculated, so it is notpossible to comment on their suitability forestimating QALYs. The assumption was also madethat when a patient develops more than one majorcomplication, the lower utility score applies. Theseassumptions are unrealistic. First, reductions inutility are likely in the earlier phases of diabetesand its complications (possibly due to having thediagnosis or side-effects of treatments), andsecond, if two major complications areexperienced, it is not likely that a patient wouldvalue this the same as having only onecomplication. The model is therefore likely tounderestimate the quality of life impact ofdiabetes.

Chen and colleagues132 use the same utilityadjustments and discount QALYs at 3%. They donot provide any detail as to how they deal withpatients who experience multiple complications.

Hoerger and colleagues133 consider the sameoutcomes as the earlier model and use the sameutility values for blindness, LEA and ESRD. They


37


also add utility adjustment values for angina(0.947),135 history of MI (0.880)166 and history ofstroke (0.50).167 QALYs were also discounted at arate of 3%. The approach used for adjusting utilityvalues in patients with multiple complications isnot given in the paper.

A simplifying assumption of the decision treemodel by Goyder and Irwig136 is that nodistinction is made between different types ofmicrovascular complications and different types ofmacrovascular complications. It assumes that earlyintervention reduces the incidence of all of them.A utility weight for all microvascular complicationswas estimated by taking the average of the weightsfor blindness, LEA and ESRD as estimated in theGlobal Burden of Disease study.168 This was aweighted average based on the relative prevalenceof these complications as estimated by Eastmanand colleagues.135 For the macrovascular utilityweight the same method was used, taking theaverage of weights for angina, acute MI, chronicheart failure and stroke. For the disutility assumedto be associated with early diagnosis andtreatment, a value of 0.01 was used. This was takenfrom a study assessing the impact of treatment forhypertension given a lack of direct evidencerelating to the impact of diabetes. Benefits, interms of QALY gains, associated with avoidance ofmicro- and macrovascular complications, werediscounted at 3%.

Findings/conclusionsThe results for the four studies are as follows (costsin US dollars).

CDC Diabetes Cost-Effectiveness Group131

The incremental cost-effectiveness ratio (ICER)reported in the CDC study was $236,449 per life-year gained (LYG) and $56,649 per QALY forscreening of all adults over 25 years of age.Screening was found to be more cost-effective inyounger cohorts – $35,768 per LYG ($13,376 perQALY) for a 25–34-year-old age cohort comparedwith $64,878 per LYG ($18,707 per QALY) in35–44-year-olds and $681,989 per LYG ($116908per QALY) for those aged 55–64 years. This resultis explained by the fact that the CDC modelassumes that early treatment after screening onlybenefits patients by postponing occurrence ofmicrovascular complications. Screening, therefore,has little impact on mortality/life expectancy butdoes improve quality of life. More favourableICERs are observed in younger cohorts becausethey have more expected life-years in which toremain complication free. More favourable ICERswere also observed in African Americans due to

the increased prevalence of diabetes and increasedrisks of developing complications in this group.

Chen and colleagues132

Compared with control, both the screeningregimens are reported to reduce the incidence ofmicrovascular complications. The reportedreduction in incidence was particularly high forESRD (65%). This is much higher than thereduction reported in the CDC model (35%),despite the fact that a very similar modellingapproach was reportedly used. No clearexplanation is given for this anomaly. The ICERsreported for biennial and 5-yearly screening,compared with no screening, are $17,883 and$10,531 per QALY, respectively. Chen andcolleagues132 also found that screening youngercohorts was more cost-effective than screeningolder cohorts.

Hoerger and colleagues133

The authors found that in cohorts of all ages, theICERs were more favourable for screeningtargeted to people with hypertension thanuniversal screening – ICERs ranging from $87,096per QALY (35-year-old cohort) to $32,106 perQALY (75-year-old cohort) compared with noscreening. The equivalent ICERs for universalscreening were $126,238 and $48,146 per QALY.The ICERs for universal screening versus targetedscreening range from $143,839 per QALY (35-year-old cohort) to $443,433 per QALY (75-year-old cohort). Screening in general(targeted and universal) was found to be morecost-effective in older cohorts, a very differentfinding to that reported in the CDC study. Themain reason for this is that in the new model,most of the benefits of screening come fromreducing CHD events with intensive hypertensioncontrol as opposed to reducing microvascularcomplications with glycaemic control. Youngerpeople have lower risk for CHD and so benefitless. A second reason for the difference is that inthe new model, glycaemic control results insmaller reductions in the incidence ofmicrovascular complications. This is based onmore appropriate data from the UKPDS trial,which was not available at the time of the previousanalysis.

Goyder and Irwig136

The base-case analysis found that for every 10,000individuals screened, a net of 10 QALYs would begained: four from postponed microvascularcomplications and 17 from avoided cardiovascularcomplications balanced against a loss of 11 fromearly diagnosis due to screening. This seems to be


38

consistent with the findings of Hoerger andcolleagues,133 that the main benefits of screeningwill be obtained from reducing cardiovascularevents as opposed to microvascular complications.

Sensitivity analysisThe authors of the CDC study report that the cost-effectiveness of screening was sensitive to severalparameters including the screening test used, thesensitivity/specificity of the test and particularlythe length of prediagnosis interval and theintensity of treatment received after screening(more intensive treatments increasing the cost perQALY). This is an important finding since thestandard of care for T2DM patients should now beintensive treatment in the light of the results fromthe UKPDS study. A weakness of the CDC modelis the limited sensitivity analysis carried out. Onlyunivariate sensitivity analysis was conducted on asmall number of parameters, which does notadequately characterise the uncertainty inherent inthe modelling approach.

Chen and colleagues132 reported no sensitivityanalysis.

Hoerger and colleagues133 provided a much morethorough sensitivity analysis. In addition toperforming univariate analysis, they performedprobabilistic sensitivity analysis by simultaneouslyvarying 129 parameters over probabilitydistributions estimated from published 95% CIs.However, it is not clear if they took into accountinterdependence between parameters. Fromunivariate sensitivity analysis, it was found that thecost-effectiveness of targeted screening wasmoderately sensitive to the duration of theprediagnosis phase, and highly sensitive to theintensity of glycaemic treatment individualsreceive after diagnosis (ICERs halved ifconventional treatment was assumed). It is notentirely clear why this assumption had such agreat effect on the ICER. Results were also verysensitive to the risk reduction associated withintensive hypertension control. This is animportant finding given that the reduction used inthe model was taken from a single trial and thatadherence to drugs may be lower in generalpopulations. Based on 1000 iterations of second-order Monte Carlo simulation, the median ICERfor targeted screening of 55-year-olds was $34,229(95% of ICERs falling between $21,594 and$76,099 per QALY).

Goyder and Irwig136 conducted one-way sensitivityanalysis on most of the parameters included intheir model and found that with all else remaining

constant (at baseline values), the benefits ofscreening would no longer outweigh the harms ifthe annual risk of CVD fell below 0.8%, the RR ofCVD could not be reduced by 13% or more withearly treatment, the disutility of early diagnosis ismore than 2% or the annual discount rate appliedto QALY gains was greater than 7%. From this, theauthors concluded that the decision on who toscreen should be based on the baseline risk ofCVD, the presence of other treatable CVD riskfactors and the disutility that is assigned todiagnosis. In addition to their main analysis,Goyder and Irwig used these threshold values tocalculate the interval at which the benefits ofrepeat screening would be likely to outweigh theharms.

Conclusions of critical appraisalsThis section summarises the critical appraisals ofthe four modelling studies reviewed in this chapterin terms of a checklist provided for assessing thequality of decision analytic models.130

CDC Diabetes Cost-Effectiveness Group131

The authors provide a reasonable overview of thedecision problem and the structure andassumptions of their model, which are consistentwith the underlying theory of diabetes diseaseprogression. However, there is a lack oftransparency regarding the details of how specificcomponents of the model work. For example, it isnot clear how mortality risks are calculated or howthe glycaemic levels are used to adjust theprogression of complications.

Another problem is a lack of transparencyregarding the data incorporation process.Although sources are generally referenced well,values used in the model are not presented inmany cases and it is not always clear how/whetherdata from original epidemiological, clinical orcosting studies have been synthesised forincorporation into the model. The generalisabilityof US cost data to the UK is also questionable,especially since it is unclear how costs werecalculated. The limited sensitivity analysis is also aweakness, making it difficult to express confidencein the results presented.

Although there is no evidence presented thatdirectly assessed the internal and externalconsistency of the model, the model does notproduce any counterintuitive results and certaincomponents of the model were calibrated to beconsistent with data reported in epidemiologicalstudies (e.g. glycaemic levels and mortality). Insummary, due to the study weaknesses mentioned


39


above and the exclusion of any benefits in terms ofreduced risk of CVD (as a direct consequence ofglycaemic control or by initiation of treatmentsaimed at other CVD risk factors), the conclusionsof this model cannot be taken as authoritative.The model is somewhat outdated by the more up-to-date analysis which incorporates more recentand relevant data.

Chen and colleagues132

The model clearly outlines the decision problemof whether mass screening should be introduced,how often it should take place and who should bescreened. The objectives of the model are alsoclearly laid out. However, the main problem withthis model is the lack of transparency in the paperreporting it. Many assumptions have not beenmade explicit, making it impossible to ascertainthe appropriateness of the results. In general, the structure is the same as in the original CDCmodel, but there are differences in the screeningcomponent with regard to repeat screening. It isunclear if the authors have used the sameassumptions regarding the time after onset thatclinical diagnosis occurs in the absence ofscreening. It is also unclear whether they haveconsidered the impact that screening would haveon prevalent undiagnosed cases already present inthe population or only considered its impact onincident cases in a cohort, which starts with nodiabetes.

The data incorporation process is not recorded insufficient detail, with no information on thecalculation of transition probabilities, althoughmany of these were taken directly from a previousmodel. Another weakness is the fact that nosensitivity analysis was conducted, which isparticularly important given that assumptionsabout the uptake for screening and adherence tosubsequent treatment are likely to impact uponthe cost-effectiveness of mass screening for thewhole population.

The model also appears to deliver some surprisingresults, which seem inconsistent with the CDCmodel, despite the apparent similarities betweenthe two models. First, the model predicts aproportionally much larger reduction in theincidence of ESRD. Second, the ICERs reportedfor mass repeat screening are substantially lowerthan those reported by the CDC for once-offopportunistic screening. The opposite might beexpected if one considers that the cost of massrepeat screening is likely to be much higher, andmight not detect very many more cases.Furthermore, with repeated screening, the cost per

case detected is likely to increase with time, asthere will be fewer undiagnosed cases in thepopulation. Without knowing what theassumptions were regarding the prevalence ofundiagnosed diabetes in the cohort, the costs ofimplementing mass screening and the uptake forit, it is impossible to tell if these findings arecredible. With regard to the finding that massscreening was more cost-effective in youngercohorts, the same criticisms that applied to theCDC model also apply here.

Hoerger and colleagues133

This analysis is more comprehensive than theoriginal CDC model due to the inclusion ofsubmodels for CHD and stroke and theincorporation of benefits accruing from tightblood pressure control. However, there is a lack ofjustification given for excluding other potentiallycost-effective screening strategies from theanalysis. For example, screening for diabetes couldbe targeted at other high/higher risk groups usingrisk scores that take into account obesity, smokingstatus, cholesterol level, etc. Screening could bepotentially more cost-effective in such groups whohave a higher risk for diabetes and CVD thanpeople who have hypertension alone. The cost-effectiveness of repeated screening or organisedmass screening have not been explored either. Inaddition, the assumption in the screening moduleseems to be that implementing opportunisticscreening will result in 100% of those eligible for itbeing offered it, and 100% agreeing to uptake,which may be unrealistic. However, people alreadyattending their GP for other reasons are likely tobe a fairly accessible population.

Access to a technical report on the modelprovided a better understanding of the finerdetail. There is more transparency regarding thedata incorporation process in this analysis, withpoint estimates, distributions and sources beingpresented in tables. However, there was little detailgiven in the published paper about how certainparameters were derived from the literature. Thetechnical report gave a better account of thederivation of model parameters. However, some ofthe parameters used in the screening modelappear to be different to those published in thetechnical report, so it is still unclear how someparameter values used in the model were derived.A further advantage of the updated model is theincorporation of many parameter values asdistributions, which appear to have beenappropriately estimated from published 95% CIswhere available. This allowed second-orderprobabilistic sensitivity analysis to be conducted,


40

giving a better indication of the uncertaintysurrounding the cost-effectiveness estimate.

In terms of internal consistency, the modelappears to work in a predictable manner but doesproduce one slightly counterintuitive result. Thereis no clear explanation as to why the ICER forscreening is so much lower when conventionalglucose-controlling treatment, as opposed tointensive treatment, is assumed. However, this isconsistent with the findings of sensitivity analysisreported in the original CDC model. The resultsthat do differ from the previous version of themodel are adequately explained in terms of theinclusion of benefits relating to the intensivetreatment of blood pressure. So, given the findingsfrom the model, the authors conclude thatscreening of people with hypertension between theages of 55 and 75 years is likely to be a moreefficient strategy than universal screening. Thisconclusion seems reasonably valid given the datapresented in the model. However, the conclusionis based on an assumption that intensivehypertension control provides as much addedbenefit to people with preclinical diabetes as itdoes to people with clinical diabetes. There is alsothe same problem of the generalisability of UScost data to the UK.

Goyder and Irwig136

The authors of this model use a simpler modelstructure than the other analyses. This means thatmore simplifying assumptions were required andthese are not transparent and appropriatelyjustified in all cases. For example, the basis for theassumption that 50% of undiagnosed cases willnever be clinically diagnosed in the absence ofscreening is not clear. Furthermore, it is not clearhow long this group would be expected to live inthe absence of screening. It is difficult to assess theappropriateness of some of the assumptions giventhis lack of transparency and justification.

For parameters based on evidence from theliterature, details of sources and calculationsrequired for incorporation in the model wereprovided in the technical report. Appropriatemethods appear to have been used to identifysources and the data sources used are perhapsmore applicable to the UK setting than those usedin the other models. However, the prevalenceestimates are based on a sample that is notrepresentative of the UK population as a whole.Another problem is that where numerous datasources exist for one parameter, the choice ofestimate used in the model is not always justified.Although the impact of using a full range of

plausible values has been considered for mostparameters, only one-way sensitivity analysis wasused.

In terms of consistency, there is no evidence thatthe authors have tested the model prior toconducting their analysis. In general, the decisiontree approach used by Goyder and Irwig136

provides a less comprehensive and realistic way ofmodelling the impact of diabetes screening thanthe model by Hoerger and colleagues. Given thelarger number of simplifying assumptions, it isdifficult to assess the validity of the results.However, it is based on data that are perhaps morerelevant to the UK setting and it does producefindings that are broadly consistent with thefindings of Hoerger and colleagues.

Summary and recommendationsOf the four studies reviewed in this chapter, themodel of Hoerger and colleagues is the strongestand most comprehensive. However, it is difficult toassess the applicability of the results to the UKsetting given that the model was designed toreflect the epidemiology of, and the resource usepatterns for, diabetes in the USA. There arevarious reasons why the cost estimates used forlong-term complications in Hoerger andcolleagues’ model may not apply to the UK. First,the treatment protocols for various complicationsof diabetes may differ in the UK (i.e. be less ormore resource intensive). Second, the methodsused to estimate the costs to attach to resourcesused may be based on inappropriate methods fortransferral to the UK setting (i.e. user charges).Third, different price structures exist in the USAcompared with the UK. Although Hoerger andcolleagues conducted probabilistic sensitivityanalysis and showed that their findings were fairlyrobust to simultaneous variation in 192parameters (including all cost parameters), adefinitive conclusion on whether or not screeningfor diabetes would be cost-effective in the UK isnot possible from this review. On the other hand,the conclusion that selective screening is likely tobe more cost-effective than universal screening islikely to be applicable to the UK. However, exactlyhow screening should be targeted is still unclear,as the model by Hoerger and colleagues onlyconsidered one targeted screening strategy. Itwould be possible to target treatment atindividuals with cardiovascular risk factors otherthan hypertension (e.g. obese patients or thosewith elevated lipids).

Given the shortcomings of the models included inthis review and the difficulties surrounding the


41


generalisability of the results, it would be advisableto conduct a new modelling study to assess thecost-effectiveness of screening in the UK settingbefore drawing any conclusions on cost-effectiveness. This analysis should include allpossible screening options and in particular shouldlook at the cost-effectiveness of screening targetedat other groups at high risk of CVD (not justhypertensive patients). Modelling such strategieswill require some assumptions about the addedbenefit of providing a diabetes diagnosis on top ofthe other known cardiovascular risk factors. If it isassumed that diabetes diagnosis would result inbetter treatment, or better adherence to treatmentfor other CVD risk factors present (e.g. obesity orhyperlipidaemia), then targeted screening couldpotentially be cost-effective in these groups. Onthe other hand, the question remains as towhether or not it would be more cost-effective totry and improve treatment for CVD risk factors inall patients who have them, without screening fordiabetes. Thus a relevant comparator in models oftargeted diabetes screening could be no screeningbut improved treatment of all CVD risk factors inthose who have them.

Another potential screening option that has notbeen included in any of the screening models todate is the possibility of widening screening toidentify those with IFG or IGT. Several clinicalstudies have demonstrated the effectiveness oflifestyle interventions or pharmacologicaltreatments for reducing the incidence of T2DM inthose who are known to have IFG or IGT.89,90

Cost-effectiveness modelling studies have alsobeen conducted and estimate favourable cost perQALY and cost per LYG ratios for suchprogrammes.169–172 These models are reviewed indetail in the next section. It would be relativelystraightforward to adapt one of the existingscreening models, or another diabetes model, toestimate the extra costs and the added benefits ofdetecting and treating individuals with IGT or IFG.Such models would require data on the prevalenceof these conditions, the sensitivity/specificity oftests to detect them and the prevalence ofcardiovascular risk factors in these groups. Finally,a future analysis could also investigate the effectsof repeat screening as opposed to once-offscreening.

The main gap in the evidence relating to the cost-effectiveness of screening for diabetes is theuncertainty surrounding the effects of implementingor improving treatment for hyperglycaemia,hypertension and hyperlipidaemia in thepreclinical phase of diabetes.

Economic models assessing thecost-effectiveness of identifyingand treating people with impairedglucose toleranceBackgroundSeveral RCTs have demonstrated the effectivenessof lifestyle interventions or pharmacologicaltreatments for reducing the incidence of diabetesover the short to medium term.89,90,169 Thus, apotential benefit that could be gained fromscreening for hyperglycaemia would be to preventprogression to diabetes in those found to haveIGT or IFG. A within-trial cost-effectivenessanalysis, carried out alongside the largest diabetesprevention trial to date,170 estimated the cost perdiabetes case prevented and the cost per QALYgained for the lifestyle and metformininterventions compared with placebo. Thereported ICERs were $34,500 per case prevented($99,200 per QALY) for the metformin arm and$24,400 per case prevented ($51,600 per QALY)for the lifestyle arm. However, this analysis onlyconsidered the costs of treatment and the benefitsaccrued over the 3-year follow-up period of thetrial. It did not consider future IGT treatmentcosts, future costs avoided from delaying orpreventing the onset of diabetes or gains inquantity and/or quality of life associated with theavoidance of future complications of IGT ordiabetes. This section appraises and discusses thefindings of several modelling studies that havebeen carried out to assess the long-term costs andconsequences of preventing or delaying the onsetof diabetes.

Search resultsFive modelling studies were identified by literature searches for inclusion in this section.The characteristics of the five modelling studiesincluded in this section are summarised in Table 14.The studies all used modelling approaches broadlysimilar to those used in the studies assessing thelong-term costs and health impact of screening fordiabetes. The main difference is that the modelsreviewed in this section assessed the progression ofdisease from a time prior to the onset of diabetes,when subjects would have IGT or IFG. Three ofthe studies assessed the long-term costs andbenefits that would be expected with the use oflifestyle or pharmacological interventions for theprevention of diabetes,171–173 one study consideredlifestyle interventions and surgery174 and one onlyassessed the impact that delaying the onset ofdiabetes would have175 – without actuallyincorporating in the model the interventions that


42


43


TAB

LE 1

4Su

mm

ary

of p

reve

ntio

n m

odel

s re

view

ed

Stud

yC

ompa

rato

rsM

odel

typ

eEc

onom

ic

Com

plic

atio

ns

Ben

efit

s of

ear

ly

outc

omes

mod

elle

dtr

eatm

ent

mod

elle

d

Sega

l et

al.,

1998

174

1. In

tens

ive

diet

and

beh

avio

ur m

odifi

catio

n M

arko

v m

odel

C

ost

per

life-

year

Non

eRe

duce

d m

orta

lity

due

(ser

ious

ly o

bese

indi

vidu

als

with

and

with

out

(coh

ort

anal

ysis)

to p

reve

ntio

n of

dia

bete

sIG

T)

2. In

tens

ive

diet

and

beh

avio

ur m

odifi

catio

n (w

omen

with

pre

viou

s ge

stat

iona

l dia

bete

s)3.

Sur

gery

for

seve

rely

obe

se in

divi

dual

s 4.

Gro

up b

ehav

iour

al m

odifi

catio

n fo

r ov

erw

eigh

t an

d ob

ese

men

5.

GP

advi

ce

6. M

edia

cam

paig

n w

ith c

omm

unity

sup

port

(g

ener

al p

opul

atio

n)

Car

o et

al.,

200

4173

1. L

ifest

yle

inte

rven

tion

(DPS

tria

l)M

arko

v m

odel

C

ost

per

life-

year

Retin

opat

hy, n

euro

path

y,

Redu

ced

mor

talit

y du

e 2.

Met

form

in in

terv

entio

n (D

PP t

rial)

(Mon

te C

arlo

ne

phro

path

y, fo

ot u

lcer

s,

to p

reve

ntio

n of

dia

bete

s3.

Aca

rbos

e in

terv

entio

n (S

TO

P-N

IDD

M t

rial)

simul

atio

n)hy

pogl

ycae

mia

, str

oke,

4.

Con

trol

(no

inte

rven

tion)

tran

sient

isch

aem

ic a

ttac

ks,

MI a

nd a

ngin

a (o

nly

the

cost

s of

th

ese

even

ts w

ere

cons

ider

ed)

Palm

er e

t al

., 20

0417

21.

Life

styl

e in

terv

entio

n (D

PP t

rial)

Mar

kov

mod

el

Cos

t pe

r lif

e-ye

arN

one

Redu

ced

mor

talit

y du

e 2.

Met

form

in in

terv

entio

n (D

PP t

rial)

to p

reve

ntio

n of

dia

bete

s3.

Pla

cebo

(DPP

tria

l)

Her

man

et

al.,

2005

171

1. L

ifest

yle

inte

rven

tion

(DPP

tria

l)M

arko

v m

odel

C

ost

per

QA

LYRe

tinop

athy

, nep

hrop

athy

, Re

duce

d m

orta

lity

due

2. M

etfo

rmin

inte

rven

tion

(DPP

tria

l)(c

ohor

t an

alys

is)ne

urop

athy

, CH

D (a

ngin

a an

d to

pre

vent

ion

of d

iabe

tes

3. P

lace

bo (D

PP t

rial)

CA

/MI),

str

oke

and

QA

LY g

ains

from

redu

ced

diab

etes

com

plic

atio

ns

McE

wan

et

al.,

2005

175

No

inte

rven

tions

but

four

del

ay s

cena

rios

Sem

i-Mar

kov

Cos

t pe

r Q

ALY

Retin

opat

hy, n

ephr

opat

hy,

Redu

ced

mor

talit

y du

e as

sess

ed:

mic

rosim

ulat

ion

neur

opat

hy, M

I and

str

oke

to p

reve

ntio

n of

dia

bete

s 1.

no

dela

ym

odel

an

d Q

ALY

gai

ns fr

om

2. 2

-yea

r de

lay

redu

ced

diab

etes

3.

5-y

ear

dela

yco

mpl

icat

ions

4. 1

0-ye

ar d

elay

would be required to do this. The models used inthese studies are discussed in detail below.

Statement of decision problemSegal and colleagues174 conducted the first cost-effectiveness modelling study in the area of diabetesprevention. They developed their model to assessthe cost-effectiveness of several different types ofdiabetes prevention programme relative to acontrol. The objective was to ascertain which typesof programme were likely to be most cost-effective.The interventions considered in the model wereintensive diet and behaviour modification (targetedat all seriously obese individuals with and withoutIGT), intensive diet and behaviour modification(targeted at women with previous gestationaldiabetes), surgery for severely obese individuals(BMI >40), group behavioural modification foroverweight and obese men (mixed and with IGTonly), GP advice (targeted at high-risk adults) and amedia campaign with community support (generalpopulation). The study did not consider anypharmacological interventions for the prevention ofdiabetes.

A subsequent modelling study, conducted by Caroand colleagues,173 assessed the cost-effectiveness ofa lifestyle intervention, a metformin interventionand an acarbose intervention relative to nointervention. The lifestyle intervention was basedon that used in the DPS trial.89 This interventioninvolved seven dietician visits in the first year andfour visits per year thereafter, in addition toindividualised exercise programmes. For theacarbose and metformin interventions, it wasassumed that patients would be prescribed thedaily doses reported in various diabetespreventions trials.89,90,169 An additional analysiscarried out by the authors also assessed the costsof screening that would be required to identifypeople with IGT for treatment. This modelconcentrated on the benefits that could be gainedfrom preventing patients progressing from IGT todiabetes.

Palmer and colleagues172 used a similar approachto Caro and colleagues173 to extrapolate thefindings from the DPP trial over the lifetime ofpatients. They assessed the long-term costs andoutcomes of the lifestyle and metformininterventions, as reported in the DPP trial,90

relative to the control arm. The analysis wasconducted separately for an Australian, French,German, Swiss and UK population/setting. Itshould be noted that the lifestyle interventionused in the DPP trial was more resource intensivethan that used in the DPS trial (see below). The

metformin intervention consisted of a target doseof 850 mg twice per day and also includedstandard advice on diet and exercise. The controlarm of the trial was the same as the metforminarm but a placebo was given instead of metformin.

In a more recent study, Herman and colleagues171

also assessed the benefits that could be gainedfrom preventing progression to diabetes using thelifestyle and metformin interventions reported inthe DPP trial. They used the same model(Research Triangle International/CDC diabetesmodel) as that used by Hoerger and colleagues133

to assess the cost-effectiveness of screening fordiabetes. However, in this analysis the model wasmodified to predict disease progression prior tothe onset of diabetes.

In the most recent study reviewed here, McEwanand colleagues175 used their model to assess theimpact that delaying diabetes would have on costsand outcomes over a 20-year period. The reportwas only available as a poster presentation at timeof writing, so limited detail was available regardingthe model’s structure and assumptions. Theiranalysis was not an economic evaluation in thesense it did not assess the cost-effectiveness ofdifferent interventions relative to each other.However, it did assess the costs, clinical outcomesand QALYs that would accrue for four differentscenarios relating to the timing of diabetes onset.The four scenarios were no delay in diabetes onset(control), a 2-year delay in onset, a 5-year delay inonset and a 10-year delay in onset.

Cohort informationThe characteristics of the cohorts synthesised inthe modelling studies of diabetes preventionprogrammes are summarised in Table 15. Thecohorts modelled by Segal and colleagues174

varied for the different types of interventionsmodelled, which were targeted at different groups(see above). The authors did state that a single agegroup was assumed for all the different targetpopulations, but it was unclear what this was.

Caro and colleagues173 conducted their analysisfor a cohort of 1000 patients with thecharacteristics of those enrolled in a recentrandomised trial of interventions to preventdiabetes, the Study To Prevent Non-insulinDependent Diabetes Mellitus (STOP-NIDDM).169

Participants were predominantly Caucasians fromCanada, Germany, Austria, Norway, Denmark,Sweden, Finland, Israel and Spain, with a meanage of 54.5 years and a BMI between 25 and40 kg/m2. All participants had IGT defined as


44

2-hour plasma glucose concentration of 7.8 mmol/lor greater, and less than 11.1 mmol/l after a 75-gglucose load. Patients also had to have a fastingplasma glucose concentration of 5.6–7.7 mmol/l.For their screening analysis, the authors assumedthat the prevalence of IGT was 11% of thepopulation based on an epidemiological studyconducted in the USA.176

The cohort modelled by Palmer and colleagues172

was constructed to resemble the study populationof the DPP: 3234 patients over the age of 25 yearswith IGT and fasting glucose levels between 5.27and 6.94 mmol/l.90 The mean age of the cohortwas 50.6 years, the mean body weight 92 kg andthe mean BMI 34 kg/m2; 32% of the cohort weremen and 45% were from minority groups. Hencethe cohort modelled by Palmer and colleagues issubstantially different from that modelled by Caroand colleagues.173

Herman and associates171 also conducted theiranalysis for a cohort based on the characteristics ofpatients enrolled in the DPP trial. Thedemographic characteristics and modifiable riskfactors of the cohort modelled by McEwan andcolleagues175 were not reported in the availableposter but the model cohort consisted of 1000patients.

Model structure, perspective and scopeSegal and colleagues174 used a Markov model,with cohort analysis, to explore diabetic status and

survival in an intervention and control group foreach programme type they considered. Limiteddetails were provided in the published paper but itappears that their model had four main states:NGT, IGT, diabetes and death. Obesity and weightloss/gain also seem to have been modelled butthere is no detail provided of how this was done.Progression between the states was modelled tooccur on 5-yearly cycles over a period of 25 years.This does not seem an appropriate cycle lengthand no explanation or justification was providedfor this choice. However, the cohort was subject toan annual mortality risk so as to yield an annualtally of the number of survivors in each group, inaddition to expected life-years. The interventiongroup was divided into two streams in the model:those who would complete/adhere to treatmentand those who would fail to complete/adhere totreatment. The authors conservatively assumedthat those failing to complete treatment wouldexperience the same outcomes as the controlgroup. Beyond the 25-year period for which themodel was run, the remaining life expectancy ofsurvivors was calculated by adjusting the age-specific life expectancy of the general populationfor diabetic state and whether or not weight losshad occurred. The assumptions made regardingthe time horizon and continued effectiveness oftreatments, beyond the follow-up periods of theclinical studies on which effectiveness estimateswere based, were not made clear in the paper bySegal and colleagues.174 Although not explicitlystated, the model seems to have been conducted


45


TABLE 15 Cohort information used in the published prevention models

Study Cohort demographic Source of cohort Cohort age range No. of patients characteristics information (years) in cohort

Segal et al., 1998174 Varied by intervention being Not reported Not reported Not reportedassessed (see text)

Caro et al., 2004173 Predominantly Caucasians Patients enrolled in the 40–70 1000from Canada, Germany, STOP-NIDDM trialAustria, Norway, Denmark, Sweden, Finland, Israel and Spain. Mean age 54.5 years and BMI between 25 and 40 kg/m2

Palmer et al., 2004172 Mean age 50.6 years, mean Patients enrolled in the 25 and older Not reportedbody weight 92 kg and mean DPP trialBMI 34 kg/m2; 32% men and 45% from minority groups

Herman et al., 2005171 Mean age 50.6 years, mean Patients enrolled in the 25 and older Not reportedbody weight 92 kg and mean DPP trialBMI 34 kg/m2; 32% men and 45% from minority groups

McEwan et al., 2005175 Not reported Not reported Not reported Not reported

from the perspective of the health service,including only direct medical costs (apart from themedia campaign programme).

Caro and co-workers173 used Monte Carlosimulation to evaluate a Markov process with fourmain states that simulated patients could moveamongst on an annual basis. The four states wereIGT, NGT, diabetes and death. Individuals in thecohort began in the IGT state, from where theycould either transit to diabetes or regress back toNGT. Patients who regressed from IGT back toNGT could also revert back to IGT again. Forpatients who progressed to diabetes, it wasassumed to be irreversible and they were assignedthe corresponding lifetime costs and clinicaloutcomes associated with diabetes. The estimatedcosts and outcomes associated with diabetes werebased on a previous diabetes model developed bythe authors.177,178 The diabetes submodelconsisted of several Markov processes that wereused to model disease progression on severalcomplication pathways including retinopathy,nephropathy, neuropathy and CVD (stroke,transient ischaemic attacks, MI and angina).Individuals progressed through the model basedon their starting characteristics (age, sex, HbA1c)and were tracked until death. All patients whoprogressed to diabetes were assumed to bediagnosed at time of onset, as their BG would beregularly monitored if they were being treated forIGT. For the screening analysis, the authorsconsidered the cost of identifying individuals in astarting population with undiagnosed IGT. Theprevalence in the population was set at 11% basedon a prevalence survey.176 A once-off screen wasassumed and those testing positive were enteredinto the model to commence treatment. This wascompared with a scenario with no screening whereindividuals with IGT progressed to diabetes at thebase rate observed in the placebo arm of theSTOP-NIDDM trial. However, assumptions aboutthe time after diabetes onset when thisundiagnosed group would receive a diagnosis werenot made clear. If these patients were notdiagnosed with IGT, it is unlikely they wouldreceive a diagnosis of diabetes at onset, as dothose with diagnosed IGT. Therefore, individualswith undiagnosed IGT might not have theirdiabetes treated as early as those progressing todiabetes from diagnosed IGT. The effect might bethat complications would develop at a higher ratein this group. It is unclear whether this has beenconsidered in the model. If not, the model couldunderestimate the benefits associated withscreening for IGT. The interventions to preventprogression to diabetes were modelled to continue

for 5 years and then stop. It is not clear whetheror not this is an appropriate assumption and littlejustification is given for it in the paper. The modelfollowed up individuals in the cohort over aperiod of 10 years but if they progressed to thediabetes module, they accumulated lifetime costsand outcomes associated with diabetes. It isunclear why this 10-year period was chosen forfollow-up in the IGT model.

Palmer and co-workers172 also adopted a Markovmodel structure to explore the long-term costs andoutcomes that could be expected from the DPPinterventions. Three states, IGT, T2DM anddeceased, were included in the model. Simulatedpatients started in the IGT state and progressed todiabetes at different rates depending on thetreatment received. Transitions between stateswere modelled to occur on an annual basis thoughit is unclear whether cohort analysis or MonteCarlo simulation was used to model progression.State-specific annual mortality rates were appliedto patients with IGT and diabetes. In the first 8years after transition to diabetes, patients weresimulated to have a mortality risk higher than thatfor IGT, but lower than that for clinicallydiagnosed diabetes. This period reflects theestimated preclinical phase of diabetes.54,179 In thebase-case analysis, the DPP interventions wereassumed to continue for only 3 years (the trialperiod). Thereafter, patients were simulated toprogress from IGT to diabetes at the baseline rateand accrue no intervention costs. The model tookthe perspective of a single healthcare payer, andconsidered the direct medical costs and benefits(LYGs) over the lifetime of patients. The costsincluded were the costs of the DPP interventions,costs associated with side-effects of the interventions(gastrointestinal and musculoskeletal), the averageannual costs associated with diabetes and theaverage annual costs associated with IGT.

The study by Herman and colleagues171 used amodified version of the model used by Hoergerand colleagues133 to evaluate screening fordiabetes. A prediabetes module was added basedon data from the DPP on disease progression,costs and quality of life associated with IGT.Several parameters relating to the progression ofdiabetes and the quality of life associated withdiabetes were also updated using recentlypublished findings. In the prediabetes modulepatients were followed from IGT to diabetes ordeath, whichever came first. Unlike in the modelby Caro and colleagues,173 patients in this modelcould not revert to NGT. Patients were modelledto receive either the lifestyle, metformin or


46

placebo intervention of the DPP trial during theprediabetes phase. The interventions wereassumed to continue until onset of diabetes ordeath. This is perhaps a more realistic assumptionthan those of Caro and colleagues.173 and Palmerand colleagues.172 It was assumed by the authorsthat the microvascular and neuropathiccomplications would not progress during theprediabetes phase. However, hypertension anddyslipidaemia were modelled to progress at therates observed in the DPP and patients coulddevelop CHD and CVD. At onset of diabetes,patients move to the early diabetes module of themodel, where complications progress slowly for aperiod of 10 years (preclinical phase). After theearly preclinical phase, patients enter the normaldiabetes progression module, where progression ofcomplications occurs at the rate observed inepidemiological studies of clinically diagnoseddiabetes patients.

McEwan and colleagues175 used a modifiedversion of their model, which was originally usedto simulate the impact of treatments on theoccurrence of diabetes-related complications inpatients with clinically diagnosed diabetes. This model was based on the diabetes modeldeveloped by Eastman and colleagues.134,135

The complications included in the model werestroke, MI, PVD, peripheral neuropathy,nephropathy and retinopathy. The model wasmodified to assess the impact of delaying diabetesby allowing CVD events to occur before the onsetof diabetes. It was presumably assumed by theauthors that microvascular complications wouldnot progress during this period. The studyassessing the impact of delaying diabetes wasconducted from the perspective of the UK NHSover a 20-year time horizon and assumed theexisting pattern of care for people with diabetes.The costs that would be incurred in delaying theonset of diabetes do not appear to have beenincluded in the model. These would include thecosts of identifying those at risk and treating themappropriately with lifestyle and/orpharmacological treatments. Thus, this study doesnot address the question of whether or not itwould be cost-effective to screen for IGT and treatpeople found to be positive.

Modelling disease progressionSegal and colleagues174 conducted their analysisbefore results were available from any of the largediabetes prevention trials. Therefore, the transitionprobabilities between the NGT, IGT and diabetesstates were based on smaller trials andepidemiological studies. The authors provided very

little information on how transition probabilitiesbetween these states were calculated for each of theprogrammes considered. Only one example wasprovided in the paper of how transitionprobabilities were calculated for the behaviouralmodification programme aimed at overweight andobese men. A Swedish study, of an intensive weightloss and fitness programme for overweight personswith IGT, relative to a standard care group, wasused.101 The finding from the Swedish study wasthat at 5–6 years’ follow-up, persons in theintensive intervention group had a significantlyreduced rate of progression to T2DM comparedwith the control group (10.6% compared with21.4%). These figures were used directly as the 5-yearly probabilities of progressing from IGT toT2DM in the intervention and control groups ofthe model. No detail was provided on the qualityof the study or the precise characteristics of theparticipants, so it is unclear if these findings would be generalisable to the cohort considered inthe model. Furthermore, it is unclear howtransition probabilities were calculated for theother possible transitions in the model (IGT toNGT and NGT to IGT).

In the model by Caro and colleagues,173 thebaseline risk of developing diabetes, from a stateof IGT, was modelled based on the placebo arm ofthe STOP-NIDDM trial.169 Based on the results of this trial, the annual probability of progressingfrom IGT to diabetes was calculated at 0.063.Diabetes was defined in the model as twoconsecutive positive OGTTs. The annualprobability of transiting from IGT to NGT wasestimated to be 0.162 and back from NGT to IGT0.163. It is unclear exactly how these estimateswere derived. The impact of interventions wasincorporated in the model as relativereductions/increases in the risks of transitionbetween IGT, NGT and diabetes. The RRreductions were taken from the DPP, DPS andSTOP-NIDDM trials89,90,169 using the results forCaucasian subgroups where available. Theacarbose, metformin and lifestyle interventionsreduced the probability of transiting from IGT todiabetes by 36%,169 24%90 and 58%,89 respectively.In addition to reducing the risk of transiting todiabetes, acarbose, metformin and lifestyleinterventions were all modelled to increase thetransition from IGT to NGT by 9% and reducereversion back to IGT by 7%.169 On transition todiabetes, the progression of complications wasmodelled based on the patient’s HbA1c level and other characteristics at time of diagnosis.However, it is unclear from the paper how thesecharacteristics were assigned to simulated patients


47


and modelled to progress over time. From reviewof a previous version of the diabetes model used inthis analysis,177,178 it appears that patientcharacteristics were assigned by sampling fromprobability distributions. However, there is nomention in the current analysis of the distributionsused for the IGT cohort. Going by the previousversion of the diabetes model, HbA1c wasmodelled to increase by 0.15% points per yearregardless of treatment, but it is unclear whatHbA1c level was assumed at transition to diabetesfrom the IGT state. There was also a lack of clarityregarding assumptions about the standard of carethat individuals would receive for hyperglycaemiaand other CVD risk factors after the onset ofdiabetes. Regarding the time horizon and thecontinued effectiveness of treatment, the relativeeffects of the interventions were modelled to bethe same over their entire course (assumed to be5 years). However, the underlying risk ofprogressing to diabetes was assumed to increaseover time, reaching 20% by 10 years of follow-up.

The probability of transit from IGT to diabetes inthe model by Palmer and colleagues172 was basedon the rates per 100 patient years reported in theDPP.90 These were 10.8, 7.8 and 4.8 in theplacebo, metformin and lifestyle groups,respectively. The authors state that these rateswere converted into annual transition probabilitiesusing the ratetoprob function in DATA (the softwareused for the analysis). The side-effects of theinterventions were modelled using the rates ofadverse events reported in the DPP trial for thelifestyle, metformin and control group. Forgastrointestinal side-effects the rates were 30.7,77.8 and 12.9 events per 100 patient years for thecontrol, metformin and lifestyle interventions,respectively. The corresponding rates formusculoskeletal side-effects were 21.1, 20 and 14.1per 100 patient-years.

In the analysis by Herman and colleagues,171 theDPP was used to estimate an annual baselinehazard of diabetes onset of 10.8%.90 This is higherthan the hazard rate calculated by Caro andcolleagues,173 using data from the STOP-NIDDMtrial.169 The lifestyle and metformin interventionswere modelled to reduce the probability of transitto diabetes by 58 and 29.9%, respectively.90 Intheir analysis, Herman and colleagues171 assumedthat these treatments would continue until theonset of diabetes, and would continue to be aseffective. Patients were modelled already to havesome complications at diagnosis of IGT –microalbuminuria (6%), nephropathy (0.4%)neuropathy (8.5%), hypertension 28%,

dyslipidaemia (45%), history of CVD (1.1%) andhistory of MI (2%) – based on the participants inthe DPP at baseline.90 Microvascular complicationsand neuropathy were assumed not to progressduring the prediabetes phase, whereas CVD riskfactors and CHD and CVD were allowed toprogress to be consistent with the rates observedin the DPP. The authors estimated that theincidence of CHD and CVD in patients with IGTwould be 58 and 56%, respectively, of the ratesobserved in patients with diabetes. Theseestimates were based on two large epidemiologicalstudies.180,181 It was assumed that the non-diabetes-related mortality rate would be the samefor persons with IGT and diabetes. On entry tothe diabetes module, patients in Herman andcolleagues’ model were assumed to have an HbA1clevel of 6.4%, the level observed at diagnosis ofdiabetes in patients enrolled in the DPP.90 Patientswere assumed to be treated for diabetes during theearly preclinical phase (estimated to be 10 years)and their HbA1c levels were modelled to rise at arate resulting in the HbA1c level observed atrandomisation in the UKPDS, after the dietaryrun-in period (7.1%). Thus the annual rate ofincrease in HbA1c was estimated to be 0.07% peryear in the early preclinical phase of diabetes.During this 10-year period, microvascular andneuropathic complications were modelled toprogress slowly, to yield the prevalence observed atrandomisation in the UKPDS.8,182,183 Hypertensionand hyperlipidaemia were assumed to progress atthe same rates as observed in the DPP trial.183 Therisks of CHD and stroke were modelled using theUKPDS risk equations.184,185 These equationsestimate absolute risk of CHD and stroke based onrace, sex, age, smoking status, HbA1c levels, SBPand lipid ratios. However, they were derived usingdata for people diagnosed with clinical diabetes soit is unclear how appropriate they are formodelling risk in the period immediately afteronset of diabetes. It is possible that theyoverestimate the risk of CVD for patients withpreclinical diabetes. If this is the case, the cost-effectiveness of delaying the onset of diabetesmight decrease. This was not an assumption thatthe authors explicitly tested through sensitivityanalysis. McEwan and colleagues175 made thesame assumptions (see below). After the 10-yearpreclinical phase of diabetes, patients were assumedto receive intensive management of their diabetes,with changes in treatment and HbA1c levelsmodelled to reflect those observed in the UKPDS.8

An initial drop in HbA1c of 2.9% was observed inthe UKPDS, followed by increases of 0.2% peryear. Complications were also modelled to developat the rates observed in the UKPDS.8,151,183–185


48

The factors that have a direct effect on the eventssimulated in the model by McEwan andcolleagues175 include HbA1c, HDL cholesterol,total cholesterol, SBP, race or ethnicity, age,gender and tobacco use. The modifiable riskfactors were also programmed to increase/decreaseover the simulation period, depending ontreatments received. However, the assumptionsand methods used to model change over timewere not described in the available poster.Originally the model used the UKPDS riskequations to estimate the risk of stroke185 andCHD184 after onset of diabetes – based on thecharacteristics and modifiable risk factors ofsimulated diabetes patients. The authors adaptedthe model so that they could assess the impact thatdelaying diabetes would have on lifetime costs andoutcomes. They did this by incorporating theFramingham risk equations157 into the model, sothat cardiovascular events could be modelled priorto the onset of diabetes. This is an appropriate useof these equations since they were derived for anon-diabetic population. In order to account forthe increased risk of CVD that would be expectedwith the development of metabolic syndrome,prior to the onset of diabetes, the authors assumeda linear increase in the risk of CVD over the delayperiod – culminating in the risks predicted by theUKPDS equations184,185 at diabetes onset. Theauthors also addressed a methodologicaluncertainty in their model by assessing the impactthat a different set of risk equations for CHD andstroke would have on predicted costs andoutcomes.158 The UKPDS outcomes model is afurther development of the UKPDS risk engine.184

It predicts the occurrence and timing of diabetescomplications and death, and also life expectancyand quality-adjusted life expectancy. It was unclearfrom the available report how progression ofmicrovascular complications was modelled(perhaps using the transition probabilitiesestimated by Eastman and colleagues134,135) but itwas reported to be consistent between bothversions of the model assessed.

Diabetes complication submodelsThe diabetes complication submodels, of theprevention models that include them, are verysimilar to those used in the diabetes screeningmodels reviewed in the pervious section.Therefore, less detailed discussion is given to theindividual complication pathways here.

The model by Caro and colleagues173 includedcomplication pathways for retinopathy, neuropathy,nephropathy, foot ulcers, hypoglycaemia andmacrovascular complications (stroke, transient

ischaemic attacks, MI and angina). Little detailwas provided on the structure and assumptions ofthese submodels but the transition probabilitiesfor the various complications were based onobserved outcomes in various trials andepidemiological studies.139,141,156 The transitionprobabilities for the microvascular complicationswere made dependent on HbA1c levels using therisk gradients observed in the DCCT trial of type 1diabetes.139 The precise method of adjustment isnot reported, but the problems associated withapplying this risk gradient to patients with T2DMwere discussed in the previous section. Anotherpossible problem with the complication risks usedin this model is that they are based on studies ofclinically diagnosed patients. It is unclear whetheror not the transition probabilities wereappropriately adjusted to reflect the lower risksthat would be observed in the preclinical phase ofdiabetes. The risks of macrovascular complicationswere reportedly based on the Framinghamequations,157 but again too little detail is providedto assess if these have been used appropriately.The main problem with using the Framinghamequations to estimate CVD risks in people withdiabetes is that the equations only include diabetesas a dichotomous variable, and the risks forpeople with diabetes are based on only 337 peoplewith diabetes included in the study. Moreappropriate CVD risk equations for people withdiabetes are now available, as discussedabove.158,184,185

The model by Palmer and colleagues172 did notincorporate a diabetes submodel and so theincidence of complications associated withdiabetes was not assessed. However, the costs ofdiabetes and its associated complications werecaptured (see below).

Herman and colleagues171 used exactly the samediabetes complication submodels as used in theirprevious analysis of screening for diabetes.133

However, some were populated with more recentdata from the UKPDS. For example, theprogression of nephropathy was based on datafrom the UKPDS 64183 as opposed to data fromthe DCCT139 and UKPDS 38.151 The risks of CHDand stroke were based on the UKPDS 56184 and60185 as opposed to risk equations developed forpeople without diabetes.157,159,160

No details were available regarding the structureof the complication submodels used by McEwanand colleagues,175 but they are likely to have beensimilar to those used in other models described inthe previous section.


49



50

MortalityIn the study by Segal and colleagues,174 mortalitywas modelled by deriving annual age-specificmortality risks for each of the three states (NGT,IGT and T2DM). Two sets of mortality risks wereestimated for each intervention programme, onefor people who successfully achieve and maintainthe intervention targets and the other for thecontrol and those for whom the intervention isunsuccessful (i.e. do not reach or maintain targetweight). The mortality risks were calculated byadjusting age/sex-specific mortality rates for theAustralian population with the relative risksassociated with IGT, T2DM and obesity (weightloss). The relative risks were obtained frompublished epidemiological studies.186–190

The mortality risk associated with IGT in themodel by Caro and colleagues173 was obtained byadjusting age- and sex-dependent mortalityestimates from Canadian life tables. Mortality risksfrom the life tables were increased by 45% basedon the findings of an epidemiological study.191

This increased risk was assumed to be lost inpatients that reverted back to NGT. For those whodeveloped diabetes, mortality was presumablymodelled using the diabetes submodels andmortality risks appropriate to individualcomplications, but no detail was provided on this.

Mortality in the model by Palmer and colleagues172

was modelled in the same way. The averageannual probability of dying with IGT or diabeteswas calculated by adjusting age-, sex- and country-specific all-cause mortality (taken from appropriatelife tables) with published RRs for all-causemortality reported for patients with IGT or diabetesrelative to the normoglycaemic population.191

During the first 8 years after onset of diabetes,patients were assigned the RR of mortalityreported for people with undiagnosed diabetes inthe NHANES II survey (1.76). After the assumed8-year preclinical phase, patients were assignedthe RR associated with clinically diagnoseddiabetes (2.26). The RR of mortality for IGT wasestimated to be 1.37.

Herman and colleagues171 used a moresophisticated approach to model mortality. Theirapproach enabled them to keep track of differentcauses of death. In their model, patients could diefrom ESRD, LEA, CHD, stroke and other causes.For patients who developed diabetes, the CVDmortality risks were modelled using the UKPDSrisk equations.184,185 For the prediabetes submodel,CVD mortality risks were estimated by applyingRR reductions180,181 to the CVD mortality risks

predicted by the UKPDS risk equations. Themortality risks associated with LEA and ESRDwere obtained from published epidemiologicalstudies.154,192

The model of McEwan and colleagues175 uses thesame approach as that of Herman and colleaguesto model the CVD mortality associated withdiabetes. The Framingham equations157 and theassumptions about increasing risk with metabolicsyndrome (discussed earlier) were presumablyused to predict mortality in the period beforediabetes onset.

CostsSegal and co-workers174 used descriptions ofprogrammes reported in the literature to estimateresource use and then attached unit costs to reflectthe cost of implementing such programmes in anAustralian setting. Costs included in their analysiswere the intervention costs and average annualdiabetes management costs for those that transitto diabetes. Screening costs were also included forthe programmes targeted at obese and overweightmen with IGT and the GP advice programme.Potential cost savings from reducing the incidenceof CVD in patients with IGT were not included inthe analysis. Reducing CVD in patients with IGT isa potential benefit of such programmes, asidefrom preventing progression to diabetes. All costsin the model by Segal and colleagues wereexpressed in Australian dollars, but no cost yearwas specified. Future costs were reportedlydiscounted at 5% per annum.

The costs included in the model by Caro andcolleagues173 were treatment costs for the IGTinterventions, test costs to monitor progression ofhyperglycaemia, screening costs to identifyindividuals with IGT and lifetime diabetesmanagement costs for those who transit todiabetes. The costs of the pharmacologicalinterventions were based on the daily dosesreported in the diabetes prevention trials89,90,169

and the unit costs for these drugs in Canada. Thecost of the lifestyle intervention was modelled onthe resource use reported for the lifestyleintervention used in the DPS trial.90 This includedseven dietician visits in the first year of treatmentand four per year thereafter. Costs of two exerciseclasses per week were also incorporated using thecost of monthly gym membership in Canada. Thecost of an OGTT was based on the test cost andthe estimated physician time and all patients wereassumed to undergo one such test a year, and toundergo a subsequent OGTT if the first one waspositive. The unit costs for tests were based on

Canadian fee schedules. For those who developeddiabetes, the cost estimates for its managementwere based on previously developed cost profilesfor each of the complication pathways consideredin the diabetes submodel.193,194 For eachcomplication, the average direct cost of managingsuch an event from the acute phase until deathwas estimated. Costs included in these calculationsincluded hospitalisation costs, home healthcareservices, outpatient services, nursing home care,laboratory tests, drugs, emergency room visits anddiagnostic and therapeutic procedures. The mainsource of cost data was physician and laboratoryfee schedules. Use of such fees to estimate costmakes it difficult to assess generalisability. It is alsounclear what cost assumptions, if any, were madefor treatment of uncomplicated diabetes (e.g.intensive control of hyperglycaemia, treatment ofother CDV risk factors). All costs in the analysis byCaro and colleagues173 were reported in 2000Canadian dollars and discounted at a rate of 5%per annum.

Palmer and co-workers172 estimated the cost ofimplementing the DPP interventions in a primarycare setting in Australia, France, Germany,Switzerland and the UK. However, the costs ofidentifying people with IGT through screening didnot appear to have been incorporated in themodel. All costs were reported in 2002 euros, andonly the direct medical costs were considered. Theassumptions were that the lifestyle interventionwould require six review sessions with the GP inyear one, 16 lessons on diet and exercise with apractice nurse, and monthly reviews with the GP insubsequent years. The resource use for themetformin intervention was assumed to include850 mg of metformin twice per day per patient,one titration visit with the GP in year one, threesubsequent follow-up visits with the GP in year oneand four follow-up consultations with the GP insubsequent years. The resource use in the controlarm was assumed to be one annual visit with theGP for standard advice. It was assumed thatpatients who experienced side-effects would haveadditional consultations with their GPs and incurthe corresponding costs. Patients who progressedto diabetes in the model by Palmer and colleaguesincurred an annual cost that reflects the averageannual medical costs for people with diabetes.195

The annual direct medical costs associated withIGT were taken to be 46% of those for patientswith T2DM. This estimate was based on a singlestudy that found that medical costs 8 years prior todiabetes diagnosis, at the time patients would haveIGT, were 46% of the annual costs of a patient withT2DM.196 In this study, costs increased over the

8 years and peaked at the time of diabetesdiagnosis. Palmer and colleagues172 used thesefindings to model annual increases in coststhroughout the preclinical diabetes phase. Thecosts of the DPP interventions and diabetesmanagement in each of the country settingsconsidered in the model were estimated from avariety of sources appropriate to the country forwhich the analysis was being conducted. Sourcesincluded the Cost of Diabetes in Europe Study(CODE-2),195 which examined resource use inpeople with T2DM using population-based registersin eight European countries including the UK.

The costs included in the model by Herman andcolleagues171 were the costs of identifying andtreating individuals with IGT, costs ofcomplications (hypertension and CVD) associatedwith IGT and costs associated with T2DM. Theauthors also considered direct patient costs for aseparate analysis adopting a societal perspective.The costs of identifying people with diabetes werebased on screening a large and diverse populationwith OGTTs.170 The direct medical costs ofidentifying one person with IGT were calculatedby multiplying the number of OGTTs required toidentify one DPP participant by the unit cost of anOGTT. The OGTT unit cost was based onMedicare reimbursement rates. The costs ofinterventions were calculated based on theresources use profiles reported in the DPP trial.The costs of the metformin intervention includeddrug costs, costs of ensuring adherence and costsof monitoring and treating side-effects.170 Thecost of the lifestyle intervention was based on acore curriculum consisting of 16 one-on-onelessons covering diet, exercise and behaviourmodification.170 Monthly individual and groupsessions with case managers followed this. Thefuture costs of IGT associated complications werecalculated using a multiplicative cost modeldeveloped by the authors. Multipliers wereestimated and applied to the baseline costs of theDPP interventions to account for the costsassociated with incident hypertension and CVD. Asimilar, previously developed, cost model was usedfor the diabetes module.197 A multiplicativeprediction model was used to estimate annualdirect medical costs according to demographiccharacteristics, diabetes treatment (as per UKPDSintensive group), cardiovascular risk factors andmicrovascular and macrovascular complications.This a more comprehensive cost model than thatused by Caro and colleagues as it includes thecosts treating CVD associated with IGT. All costsare presented in 2000 US dollars and arediscounted at a rate of 3%.


51


The only detail available on the costs included inthe model by McEwan and colleagues175 was that only direct healthcare costs were considered.The total costs avoided under each delay scenario were presented in 2005 UK pounds. As mentioned earlier, the costs that would need to be incurred in order to delay the onset ofdiabetes do not seem to have been incorporated in the model. Costs were discounted at 3.5% per annum, the current recommended rate for the UK.

OutcomesIn terms of outcomes, Segal and colleagues174

modelled the number of diabetic years avoidedand the LYGs for each intervention programmecompared with its control. Outcomes werediscounted at 5% per annum.

The outcomes considered in the model by Caroand colleagues173 were the number of cases ofT2DM prevented over a 10-year period, life-yearsand years free of diabetes. However, it was unclearwhat assumptions were made about life expectancyin the absence of diabetes in order to calculatelife-years gained by preventing it. Life-years arereportedly discounted at 5% per annum.

Palmer and colleagues172 used their model toestimate the number of years free of diabetes, thepercentage of patients developing diabetes overtheir lifetime and life expectancy for each of thecomparators. The ICER of the interventionsversus the control arm was reported as cost perlife-year.

Herman and colleagues171 reported outcomes inQALYs. The utilities associated with IGT werebased on the Quality of Wellbeing Index (QWBI),which was administered to patients enrolled onthe DPP trial.170 The utility scores associated withdiabetes were based on another study that usedthe QWBI to assess utility in patients with bothtype 1 and type 2 diabetes recruited from amedical centre.198 The authors constructed anadditive health utility model to predict healthutility scores associated with IGT beyond the 3-year follow-up period of the trial. For eachcomplication that a patient experienced in themodel, a utility decrement was applied to theirutility score. The decrements were estimated byfitting a regression model of the QWBI4-derivedhealth utility scores to indicator variables fordiabetes and each demographic variable, treatmentand complication. The same process was used forassigning utility scores to diabetes and itscomplications.

McEwan and colleagues’175 model predicted theabsolute number of CHD, stroke, CVD deaths,retinopathy, nephropathy and neuropathy eventsfor each delay scenario considered. The modelalso predicted the number of QALYs gained foreach delay scenario versus the control. The sourceof utility scores for IGT and diabetes was notprovided in the available poster. In addition, theassumptions used for estimating utility scores formultiple complications were also not madeexplicit. The QALYs were discounted at 3.5% perannum.

FindingsThe programmes considered by Segal andcolleagues174 were estimated to yield between 43(GP advice for those with IGT) and 423 (surgeryfor seriously obese) life-years per 100 patientstreated. Relative to the control groups, theincremental cost per LYG varied frominterventions that both increased life expectancyand led to net savings (intensive diet andbehaviour change for seriously obese with IGT,group behaviour change intervention foroverweight men and the media campaign) to$12,300 per life-year (surgery for seriously obese).

Caro and colleagues173 predicted from their modelthat over the 10-year period considered, 70 of the1000 IGT patients in the cohort would beexpected to die in the absence of treatment and543 would develop diabetes. Intensive lifestylemodification was estimated to prevent 117 of thesecases and metformin and acarbose were estimatedto prevent 52 and 74 cases, respectively. Thelifestyle modification intervention was found to bethe most expensive in addition to being the mosteffective (ICER $749 per LYG compared with notreatment). The metformin and acarboseinterventions were found to be cost savingcompared with the no-treatment option. TheICERs reported for lifestyle modification versusmetformin and acarbose were $7725 and $9988per life-year, respectively.

Palmer and colleagues172 estimated that thelifestyle and metformin interventions wouldincrease both the number of years free fromdiabetes and life expectancy. The mean number ofyears free from diabetes was 8.14, 9.94 and 9.02for the control, lifestyle and metformin group,respectively. The corresponding improvements innon-discounted life expectancy for the lifestyleand metformin interventions were 0.22 and 0.11,respectively. They also estimated that in manycountries the interventions would be cost saving.However, for the UK setting they predicted that


52

the interventions would lead to small increases innet costs of €1021 and €378 for the lifestyle andmetformin interventions versus control,respectively. This translates to a net ICER of€6381 and €5400 per life-year for the lifestyle andmetformin interventions, respectively, in the UKsetting.

Herman and colleagues171 found thatapproximately 50% of their cohort would beexpected to develop diabetes within 7 years iftreated with placebo. In contrast, they estimated itwould take 18 years for 50% of lifestyle-treatedpatients to develop diabetes and 10 years for 50%of the metformin-treated participants to developdiabetes. These findings are broadly similar tothose of Caro and colleagues,173 despite thedifferences between the models and thepopulations for which they were conducted. Overa lifetime, Herman and colleagues171 estimatedthat 83% of participants treated with placebowould develop diabetes compared with 63% ofthose treated with lifestyle intervention and 75%of those treated with metformin. They alsoestimated that the lifestyle intervention wouldincrease life expectancy by 0.5 years and reducethe cumulative incidence of blindness by 38%,ESRD by 38%, LEA by 35%, stroke by 9% andCHD by 8%. The metformin intervention wasfound to increase life expectancy by 0.2 years andreduce the cumulative incidence of blindness by16%, ESRD by 17%, LEA by 16%, stroke by 3%and CHD by 2%. Compared with the placebointervention, the lifestyle intervention was foundto cost $635 more over a lifetime and produce again of 0.57 QALYs ($1100 per QALY). Themetformin intervention cost $3922 more over alifetime and resulted in a gain of 0.13 QALYs($31,300 per QALY). The lifestyle interventiondominated the metformin intervention.

As mentioned earlier, McEwan and colleagues175

evaluated their model using the original UKPDSrisk equations for CHD and stroke184,185 and alsousing the newer equations from the UKPDSoutcomes model.158 Using the original equations,the no-delay scenario resulted in discounted costsof £13,076 and 7.6 discounted QALYs over the 20-year time horizon. The total costs were foundto decrease to £11,291 per subject with a 2-yeardelay in diabetes, £9998 with a 5-year delay and£8395 with a 10-year delay. This translates todiscounted cost savings of £1785, £3078 and£4681 for the three delay scenarios, respectively. It should be noted that the cost savings woulddecrease or disappear if the costs required to delay diabetes were to be included in the analysis.

The predicted QALYs for the three delay scenarios were 8.3 for the 2-year delay, 9.1 for the5-year delay and 10.1 for the 10-year delay,representing gains relative to the no-delayscenario of 0.7, 1.5 and 2.5, respectively. When theequations from the UKPDS outcomes model158

were used, fewer cardiovascular and microvascularevents were predicted over the 20-year timeperiod. The cost savings associated with the threedelay scenarios were subsequently less using theseequations. However, the QALY gains were similarusing the equations from the UKPDS outcomesmodel.

Sensitivity analysisSegal and colleagues174 assessed the sensitivity oftheir findings to variations in several modelparameters using one-way sensitivity analysis. Theresults were found to be sensitive to the assumedsuccess rates of the programmes. However, evenwhen it was assumed that only 20% of patientswould achieve the weight loss/exercise targets forthe behaviour modification programmes, the costper life-year stayed within the range consideredfavourable.198 This was presumably due to theavoidance of high costs associated with diabetesmanagement.

Caro and colleagues173 conducted one-waysensitivity analysis on many of the parameters intheir model. They also found that the ICERs of allthe interventions remained in acceptable boundsrelative to no treatment. However, the interventionof choice changed when effectiveness results werevaried within the upper and lower bounds of theirCIs. For example, at the lower bound of the CI forthe lifestyle intervention, all else remainingconstant, the acarbose intervention would bedominant. However, no multivariate orprobabilistic sensitivity analysis was conducted tocharacterise better the uncertainty in the decisionbetween the options.

Palmer and colleagues172 conducted their analysisfor different subgroups to characterise the impactthat the heterogeneity in outcomes, observed inthe DPP trial,90 would have on their findings.They assessed the cost-effectiveness of theinterventions for cohorts of different age and BMIgroups using the findings reported for thesedifferent subgroups in the DPP trial. They foundthat the metformin intervention had a betterimpact on costs and life expectancy than thelifestyle intervention in younger more obesepatients but led to less benefit and higher coststhan the lifestyle intervention in the older cohort(aged 65 years at baseline). In a cohort with a BMI


53


<30 kg/m2, the metformin intervention led tominimal improvements in life expectancy andincreased costs, whereas the lifestyle interventionhad a greater impact on life expectancy and wascost saving. However, even in this group the ICERof metformin still remained attractive (€47,200 perlife-year) compared with other interventions fordiabetes.199 In a group with BMI >30 kg/m2, themetformin intervention led to a greaterimprovement in life expectancy and greater costsavings than the lifestyle intervention. Palmer andcolleagues also conducted extensive one-waysensitivity analysis on many of the individual costand probability parameters included in theirmodel. They found that the findings were mostsensitive to the probabilities of developingdiabetes, the RRs of mortality associated with IGTand diabetes, the annual costs of managingdiabetes and the costs of implementing the DPPinterventions. However, they found that whenthese parameters were varied within plausibleranges, the ICERs of the DPP interventionsremained in the range considered cost-effective byinternational standards.199 They also showed thatgreater improvements would be seen in diabetes-free years and life expectancy if the DPPinterventions were to continue, and continue to beeffective, for longer than the 3-year follow-upperiod of the DPP trial. They also found that thetotal lifetime net costs associated with the lifestyleintervention decreased as the assumed duration ofthe effect of the intervention increased. This wasdue to the higher number of diabetes cases beingprevented offsetting the costs of ongoing deliveryof the intervention.

Herman and colleagues171 conducted fairlyextensive univariate and bivariate sensitivityanalysis. First they ran the model for different agegroups and then they assessed the impact ofaltering various treatment effectiveness and costassumptions. They found that the lifestyleintervention would be cost saving in participantsyounger than 45 years and that metformin wouldnot be cost-effective in patients aged over 65 years(>$100,000 per QALY). The authors also reportedthat the cost-effectiveness of the interventionsimproved if they were modelled as they might beimplemented in routine clinical practice – groupsessions as opposed to one-on-one lessons for thelifestyle intervention and generic metformin forthose on pharmacological treatment. Even ifeffectiveness was assumed to be 50% less thanreported in the DPP trial, they estimated that bothinterventions would remain in the cost-effectiverange.198 The authors also conducted probabilisticsensitivity analysis where 81 parameters were

simultaneously varied over probability distributionsfor 500 iterations of the model. They found that95% of the cost per QALY ratios fell between $587and $9456 for the lifestyle intervention andbetween $16,509 and $84,583 for the metforminintervention.

Apart from assessing the methodologicaluncertainty surrounding the use of different riskequations in their model, McEwan andcolleagues175 reported no details of any furthersensitivity analysis conducted.

Conclusions of the critical appraisalsThe five modelling studies reviewed in thischapter were appraised using a quality assessmentchecklist for decision analytic models.130 Theresults of this process are summarised below.

Of the five models reviewed, four clearlyhighlighted the decision problem that they wereaddressing, and had objectives consistent with thisproblem.171–174 The fifth study, by McEwan andcolleagues,175 was not a decision analytic model inthat it did not consider a choice betweenalternative treatment options. In terms of thestructures of the models, most were broadly similarbut some were more comprehensive than others.Two of the models seemed rather simplistic in thatthey did not model individual complicationsassociated with IGT and diabetes172,174 or did notincorporate the quality-of-life impacts that thesecomplications would have.173 The models byHerman and colleagues171 and McEwan andcolleagues175 are preferred on these grounds. Themodel by Segal and colleagues,174 in particular,seemed over-simplistic, using just one set of 5-yearly transition probabilities to modeltransitions between three states. Furthermore, theassumptions regarding the probabilities oftransition were not made clear for many of theinterventions considered by Segal andcolleagues.174 There is also a lack of transparencyin the reporting of the model by Caro andcolleagues.173 In this model, some of theassumptions used to model life expectancy andLYGs were unclear and not explicitly justified. Inthe model by McEwan and colleagues,175 it wasnot possible to ascertain the methods used tomodel the microvascular complications. This wasbecause we only had access to a posterpresentation of this modelling study at the time ofwriting. The models by Palmer and colleagues172

and Herman and colleagues171 were reported verytransparently, allowing a clearer assessment oftheir quality and the appropriateness of the datathey incorporated.


54

In terms of the data incorporation process, noneof the models provided much detail on themethods used to identify sources for theparameters included. In addition, the details ofmethods used to synthesise data for incorporationinto the models were not provided in many of thereports. The exceptions were the models byHerman and colleagues171 and, to a slightly lesserextent, Palmer and colleagues.172 Herman andcolleagues provided all the details of how modelparameters were estimated in a technical reportpublished alongside the published paper.133 It wasclear from this report that, as far as possible,appropriate sources had been used andincorporated into the model correctly. This is notto say that there were not some potential problemswith the model as discussed in the previoussections of this review. Another strength of themodel by Herman and colleagues171 was theextensive use of sensitivity analysis. A strength ofthe model by Palmer and colleagues172 was thatgood-quality resource use and cost data wereincorporated, which was more relevant to the UKsetting. Although individual complications werenot included in the model, the costs of treatingthese were captured.

In terms of consistency, most of the modelsproduced results that made intuitive sense basedon the data that had been incorporated, and theyall produced findings that were broadly consistentwith one another.

In conclusion, the model by Herman andcolleagues171 was the most comprehensive andtransparent of the models reviewed. The model byMcEwan and colleagues175 was also verycomprehensive but the lack of detail available atthe time of writing made it difficult to assess itsquality and the appropriateness of its assumptions.The model by Palmer and colleagues,172 althoughsimple, was presented very transparently andmade assumptions that seemed reasonable in thelight of the decision that the model was designedto inform. The report on the model by Caro andcolleagues173 lacked transparency and made someassumptions that were not appropriately justified.Finally, the model by Segal and colleagues174 wasover-simplistic and lacked transparency.

Summary and recommendationsThe models reviewed all estimated long-term costsand outcomes associated with the delay orprevention of diabetes. Four of the studiesmodelled the short-term costs and effects ofdelivering interventions to prevent/delay the onsetof diabetes, and balanced these additional

treatment costs against expected future healthoutcomes and cost savings.171–174 Although thesemodels were of variable quality, and varied in theirstructure and assumptions, all predicted thatdiabetes prevention interventions would providegood value for money. The fifth study included inthe review only assessed the impact that delayingdiabetes would have on future health outcomesand medical costs.175 However, it also producedfindings that were broadly consistent with those ofthe other models – that delaying the onset ofdiabetes by even modest periods wouldsubstantially reduce the incidence of vascularcomplications (microvascular and cardiovascular),improve quality of life and avoid future medicalcosts. These results certainly seem to suggest thatif a screening programme were implemented forpeople at risk of developing diabetes, it would bea good use of resources to treat those found tohave IGT with lifestyle or pharmacologicalinterventions.

Based on the models reviewed, there appears tobe some uncertainty surrounding the preferredintervention for people identified as having IGT.First, the characteristics of the cohort seem to havesome bearing on which option is to be preferred,172

and second, the relative effectiveness of thealternative treatments in routine clinical practicemay also prove important. The evidence from thediabetes prevention trials and above modellingstudies suggests that lifestyle interventions arelikely to be the best option for most people withIGT. However, for select groups a pharmacologicaltreatment might be preferred.172 Furthermore, iflifestyle interventions prove very difficult toimplement effectively in routine clinical practice,and pharmacological treatments are well accepted,then these might prove the better option or beused in combination with lifestyle change. Anotherpoint worth mentioning is that lifestyleinterventions could have substantial impacts onother CVD risk factors such as obesity, bloodpressure and lipid ratios. It is not entirely clear ifthese potential benefits have been captured in themodels reviewed here. If not, they couldunderestimate the benefits of the lifestyleinterventions.

Some of the models reviewed also included thecosts that would be incurred in identifying peoplewith IGT (screening costs).171,173,174 Even whenthese costs were included, the reported ICERsremained favourable. However, none of themodels appeared to compare directly an IGTscreening and treatment scenario with a no-screening scenario. Although the model by


55


Herman and colleagues171 included costs ofidentifying patients for treatment, the control armin the model also assumed that patients would beidentified as having IGT. Therefore, patients inthe control arm also received intensive treatmentfor diabetes upon onset. If no screeningprogramme were in place to identify patients withIGT, then they would not receive a diagnosis ofdiabetes at onset, and would not receive treatmentuntil they were clinically diagnosed. In thisabsence of screening, individuals progressing fromIGT to diabetes would probably develop diabeticcomplications at a higher rate than thoseprogressing to diabetes after having beendiagnosed with IGT. Hence, if screening followedby treatment for IGT were to be compared with ano-screening scenario, then the cost-effectivenessestimates might be more favourable than thosereported in the studies reviewed here. However,there remains a question over who to screen thatneeds further clarification.

The prevalence of IGT will be highest in groupswho have other CVD risk factors. As the previoussection of the review concluded, screening fordiabetes is more likely to be cost-effective in thesegroups. The question is, with the cost-effectivenessof treating IGT having been demonstrated, shouldthe target population for screening be wider thanit would otherwise be if it were only those withundiagnosed diabetes who would benefit? Thisquestion requires a more thorough exploration ofthe cost-effectiveness of different screeningstrategies, where the benefits of treating thoseidentified with both IGT and undiagnoseddiabetes are incorporated.

In conclusion, several modelling studies allpredicted favourable ICERs for the identificationand treatment of people with IGT. Two of thestudies in particular were transparently presented,allowing a thorough assessment of quality. Both ofthese studies appeared to use appropriate dataand assumptions and used fairly extensivesensitivity analysis to characterise uncertainty.Both studies reported consistent findings and onewas conducted from the UK NHS perspective.Based on the findings of these two studies, lifestyleinterventions similar to those reported in the DPPtrial90 would appear to be the best option fortreating people identified with IGT. Their cost-effectiveness of the metformin interventionappears less certain, although there may beparticular groups for whom this interventionwould be the preferred option. Based on theavailable evidence, it seems reasonable to concludethat screening and treating people for IGT would

be cost-effective in the UK setting. However,before proceeding with such a programme, thequestion of who to screen requires furtherconsideration. A modelling approach could beused to assess the relative cost-effectiveness ofscreening targeted at different populations.

Studies assessing the costs andshort-term outcomes of diabetesscreening testsBackgroundIf a decision is made to implement a screeningprogramme for diabetes or IGT/IFG, then thereremains a question of what screening tests and cut-off points to use. One way to help inform thisdecision is to consider the costs of alternative teststrategies and the associated outcomes in terms ofthe numbers of true cases that would be detected,and also the numbers of false positives, falsenegatives and true negatives. In this section,studies that have assessed these costs andoutcomes for different tests and test cut-off pointsare reviewed.200

Search resultsThe searches identified five studies for inclusion inthis section of the review201–205 These studies aresummarised in Table 16 and discussed in detailbelow.

The studiesShirasaya and colleagues203 conducted an economicevaluation alongside a study to establish the efficacyof alternative screening tests for IGT and T2DM.They established the sensitivity and specificity ofthree indicators, which do not require fastingbeforehand, in order to identify the best optionfor screening in situations where fasting tests arenot practical. The three indicators compared were1,5-anhydroglucitol (1,5-AG), glycosylatedhaemoglobin (HbA1c) and fructosamine (FRA).The sensitivity and specificity of the threeindicators were assessed on a sample of 891 menwho had been assigned diabetes status according toWHO criteria. The optimum cut-off point for eachof the indicators was established by plotting receiveroperating characteristic curves and thendetermining the point on the curve that was closestto the point where sensitivity was 100% and thefalse-positive rate 0. The optimum cut-off points forthe three indicators were calculated for detectingdiabetes only and also diabetes and IGT together.

An economic evolution was then conducted usingthe sensitivities and specificities at the optimum


56

cut-off points for the three test indicators. Theauthors adopted a healthcare payer perspectiveand developed a model to estimate the cost-effectiveness of screening with the three indicatorsfor a hypothetical cohort of 1000 men with a givenprevalence of undiagnosed diabetes and IGT. Thenumbers of true positives, true negatives, falsepositives and false negatives were estimated basedon the previously calculated sensitivities andspecificities. For those who screened positive usingeach of the indicators, an OGTT was assumed tofollow. Those who screened negative (false andtrue negatives) were assumed to undergo the samescreening test in subsequent years (annualscreening assumed) and the model was run untilmore than 99% of diabetes cases in the cohortwould be identified. The cost per case detected foreach of the indicators was established for IGT anddiabetes together, and for diabetes alone. For theanalysis using the optimum cut-offs for IGT anddiabetes together, the authors also calculated thenumber of cases of IGT detected that wouldprogress to diabetes (over a 10-year period) to givethe prospective number of diabetes cases thatwould be detected. The prospective and presentcases were added together to give the totalnumber of diabetes cases that would be identifiedthrough each screening alternative.

The costs included in the analysis were the costs ofeach of the screening tests and the costs of theOGTT for those screening positive. Those whoscreened negative were assumed to incur the sametest costs in subsequent years. In the case of falsenegatives, individuals were assumed to incur thescreening test costs in subsequent years until theyscreened positive, in which case they would alsoincur the cost of an OGTT that year. The futurecosts were discounted at 5% per annum.

The results of the study were that, out of the threeindicators, HbA1c and 1,5-AG resulted in thehighest number of diabetes cases identified.However, when the optimum cut-off points forIGT and diabetes were used, 1,5-AG identified thehighest number of cases – closely followed by FRA.The authors concluded that in Japan, FRA wouldbe the most cost-effective strategy for screeningwhen fasting was not practical due to the low costof the test. However, the study findings were notclearly presented, making it very difficult toascertain the validity of this conclusion. Moreover,the results would depend on several assumptions.For example, one assumption was that testingwould be conducted on an annual basis and so the negative consequences/costs of falsenegatives would be minimised. In addition, the

costs and number of cases that would be detected,at the different cut-off points for the differenttests, were not presented. The authors did reportsome sensitivity analysis and reported that thechoice between indicators depended on the costsof the tests and the cut-off points chosen, butagain these analyses were poorly reported.Moreover, other important parameters such as thescreening interval were not addressed in sensitivityanalysis. Given the lack of transparency in thereporting of this study, it is impossible to commenton the applicability of the findings to the UKsetting.

In another similar study, Johnson and colleagues202

set out to assess the efficacy and cost of severaldifferent strategies for identifying diabetes thatwould not require fasting. They carried out theiranalysis for the US population aged between 45and 74 years, estimating that 72.6 million peoplewould be eligible for screening. The prevalence ofundiagnosed diabetes and IGT in the populationwas estimated to be 10 and 22%, respectively, atbaseline. Johnson and colleagues202 estimated thenumber of true positives, false positives and falsenegatives that would be obtained using threedifferent cut-off points (�100, �130 and�160 mg/dl) on a random plasma glucose (RPG)test, at 1-, 3- and 5-yearly intervals over a periodof 15 years. In addition they also assessed thescreening performance of a multivariate logisticequation, which incorporated RPG level,postprandial time, age, sex and BMI. Thesensitivity and specificity of the different cut-offs,and the logistic equation, were calculated byapplying them to a large dataset, which had all theappropriate information, including results ofOGTTs performed on consecutive days (goldstandard for diagnosis of diabetes). The totaldirect medical costs and patient costs associatedwith each strategy were calculated.

Johnson and colleagues202 found that for eachincremental improvement in sensitivity, using cut-offs below 130 mg/dl, there were substantialreductions in specificity. The multivariate equationwas also shown to be more sensitive than RPGalone at any given level of specificity and viceversa. The absolute difference in the number oftrue positive screening tests between the mostsensitive (RPG �100 mg/dl every year) and leastsensitive strategy (�160 mg/dl every 5 years) was4.5 million over the 15 years. The absolutedifference in the number of false positives betweenthe most sensitive and least sensitive strategy was476 million. Hence cut-off points with a higherspecificity were found to decrease minimally the


57


number of true-positive screening tests but todecrease substantially the number of false-positivetests. Based on these efficacy findings, the authorsconcluded that an RPG cut-off point of 130 mg/dl,or the multivariate equation, applied every 3 yearswould be the optimal screening strategy. It was notentirely clear what criteria the authors used toarrive at this conclusion. The costs of each of thestrategies were estimated by assessing the directmedical and patient costs associated with eachstrategy. It was estimated that 54.4 million of theeligible population would seek routine medicalcare within any given year, and would therefore beeligible for opportunistic screening. For thisgroup, the only direct medical costs were the costsof the screening test and, when necessary, adiagnostic test. The remaining population wereassumed to incur the cost of an arrangedoutpatient visit, in addition to the test costs. Basedon the baseline prevalence of undiagnoseddiabetes, the prevalence of IGT, an estimated rateof progression from IGT to diabetes, and thenumber of cases detected at each screeningexamination, the authors calculated the prevalenceof undiagnosed diabetes in the population foreach screening examination following baseline.The costs of the different screening strategies wereestimated to vary from US$6.9 to 42.7 billion. Aweakness was that no cost year and no discountingof future costs were reported. The strategies thatthe authors declared optimal based on sensitivityand specificity data (RPG cut-off point of 130 mg/dl and the multivariate equation appliedevery 3 years) cost US$11.1 and 9.7 billion,respectively. This translates into $642 and $563per true positive, respectively. However, it shouldbe noted that this was not an economic evaluationdesigned to identify the most cost-effectivescreening strategy. It is not possible to answer thatquestion without considering the longer term costsand consequences associated with false negativesrelative to the costs and consequences of true andfalse positives. The costs and consequences of falsenegatives will depend on the length of thescreening interval, the uptake of screening atsubsequent recalls and the rate of development ofdiabetes-related complications in undiagnosedindividuals (relative to those who are diagnosedand treated).

Another study, conducted by Icks and colleagues,201

compared several once-off screening strategies interms of cost per case of undiagnosed diabetesdetected. A simple decision analytic model wasdeveloped to compare four different strategiesover a time horizon of 1 year. The strategiesconsidered were a single FPG test (using a cut-off

of 7 mmol/l), an FPG test followed by an OGTTfor those with FPG �6.1 and <7.0 mmol/l(diabetes assumed if FPG �7.0 or OGTT�11.1 mmol/l), an OGTT alone (�11.1 mmol/l)and an HbA1c test followed by an OGTT for thosewith an HbA1c level >5.6%. For each of these fourstrategies, two different models for their deliverywere also assessed – screening of everyone in thepopulation and screening targeted at those atincreased risk (as defined by various risk factorsfor diabetes).

The authors estimated the total cost and thenumber of cases that would be detected for eachof the strategies. The characteristics of patients,the prevalence of the various levels ofhyperglycaemia (defined by each of the indicators)and the sensitivities/specificities of the various testswere estimated from a population prevalencesurvey of inhabitants (aged between 55 and74 years) of the German city of Augsburg.206 Theyalso used data from a UK-based study to reflectdifferent levels of uptake for the different teststrategies.200 The uptake of HbA1c testing wasestimated to be 100% based on the assumptionthat everyone in Germany between the ages of 55and 74 years would visit their physician at leastonce per year and so be accessible to testing. Thismay be an over-optimistic assumption. Theauthors conducted their analysis from both thehealthcare payer and societal perspectives,including direct medical costs and indirectproductivity costs associated with each strategy.

One of the main findings of this study was that thepopulation screening options dominated all thetargeted options. This was probably due to theassumption that the patient-level informationrequired for the targeted options would not beroutinely available, and so its collection wouldrequire additional assessments and incur extracosts. These costs were estimated to be greaterthan the extra costs it would take to screeneveryone in the population. However, theassumption seemed to be that in order to targetscreening, everyone in the population wouldrequire their BMI, blood pressure andtriglycerides to be measured. This would unlikelybe the case in reality, so this finding should not betaken as authoritative. In terms of the relative cost-effectiveness of the different strategies forpopulation-based screening, the strategiesrequiring FPG tests were dominated by the singleOGTT from the perspective of the healthcarepayer. The HbA1c test followed by an OGTT wasfound to identify more true cases but at greatercost (€771 per additional case detected). It was


58


59


more effective due to the assumption that uptakeof HbA1c testing would be 100% and that thosewith a positive HbA1c test would have a muchhigher uptake for the OGTT. Similar results werealso obtained when the cost-effectiveness of thestrategies was considered from the societalperspective. One of the weaknesses of the study byIcks and colleagues201 was the lack of detail givenregarding the estimation of the modeleffectiveness and cost parameters, making itdifficult to assess the internal and external validityof the results. A fairly comprehensive sensitivityanalysis was carried out, which showed that theresults were reasonably robust to changes in theprevalence of disturbed glucose metabolism,uptake for the various strategies and productivitycosts. However, the strategy considered to be mostcost-effective was sensitive to changes in theparticipation rates. The same weakness thatapplies to the other studies discussed in thissection also applies here. That is, by failing toconsider the relative long-term costs andconsequences of the different screening outcomes,it is not possible to answer the question of whichstrategy is best. Moreover, given the lack oftransparency regarding the parameter estimatesincluded in the model, and the use of charges asopposed to costs, it is not possible to ascertainwith confidence which of the strategies would bethe most efficient in the UK in terms of the costper case detected.

Zhang and colleagues204 addressed a similarquestion to that of Icks and colleagues.201

However, they assessed the efficacy, effectivenessand efficiency of various strategies for detectingboth cases of prediabetes (IGT and/or IFG) andcases of undiagnosed diabetes. Like Icks andcolleagues they assessed once-off screening for theUS population but assumed that only those whoseek healthcare would be screened. Based on datafrom the US 2000 census,207 the third nationalhealth and nutrition examination survey(NHANES III)208 and other published literature,they estimated that there would be 54.4 millionadults aged between 45 and 74 years who wouldbe eligible for screening. They further estimatedthat the eligible number would be 37.4 million ifscreening were to be restricted to those with a BMI�25 kg/m2. It was also estimated that there wouldbe 12.1 million cases of prediabetes and5.4 million cases of undiagnosed diabetes in thewhole population and 9.6 and 4.7 million cases inthe population with BMI �25 kg/m2. Themethods used for calculating these figures werenot clearly reported.

The screening strategies assessed by Zhang andcolleagues204 were as follows: (1) an OGTT foreveryone in the study population; (2) an FPG testfor everyone followed by an OGTT for those withfasting glucose levels above the cut-off used(95 mg/dl) but below the level used to diagnoseimpaired fasting glucose (110 mg/dl); (3) an HbA1ctest with an OGTT for those with an HbA1c levelabove 5%; (4) a capillary blood glucose (CBG) testwith an OGTT for those with capillary glucoseabove 100 mg/dl; and (5) a risk assessmentquestionnaire followed by an OGTT for those whoscored above a certain limit. The cut-off pointsused for the FPG, HbA1c and CBG tests werereportedly selected on the basis that they incurredthe lowest cost per case detected out of a range ofcut-off points tested, but the analysis used toinform this decision was not reported in the paper.

Zhang and colleagues204 assessed the effectivenessof the different strategies by estimating theproportion and number of prediabetes andundiagnosed diabetes cases that would be detectedusing each screening strategy. The sensitivities andspecificities of the various tests were taken frompublished sources and it was assumed that theOGTT was 100% sensitive and specific foridentifying IGT, IFG and undiagnosed diabetes.Some assumptions were also made about thesensitivity and specificity of the FPG test, but thesewere not entirely clear. The direct medical coststhat would be incurred through each screeningstrategy were calculated by using charges for thedifferent laboratory tests and physician timerequirements. The patient costs were calculatedbased on estimates of patient time required foreach strategy and transportation costs. Since theunit costs for the laboratory tests and physiciantime were based on charges, it is difficult togeneralise them to settings outwith the USA. It was unclear how patient time and travel costswere estimated.

The estimated number of cases that would bedetected with each of the screening strategiesapplied to the whole population ranged from 12.1to 17.5 million. The equivalent range for thetargeted screening option was 9.8 to 14.3 millioncases. In both cases the OGTT on its own wasfound to be the most effective strategy, followed byHbA1c, FPG and CBG. The total costs of each ofthe strategies from the single payer’s perspectiveranged from $2.16 billion (risk assessmentquestionnaire) to $3.44 billion (HbA1c with OGTT)for the whole population. This translates into costsper case averted ranging from $176 (CBG withOGTT) to $236 (HbA1c with OGTT). When

societal costs were included, all the strategies weremore costly and the costs per case averted rangedfrom $247 (CBG with OGTT) to $332 (HbA1c withOGTT). The cost per case identified was lowerwhen screening was restricted to those with a BMI�25 kg/m2 but the rank order remained the same.This finding differs from that of Icks andcolleagues,201 who estimated that targetedscreening would be dominated by populationscreening. This is probably due to the differingassumptions about the costs involved inidentifying people for the targeted strategy.

Based on these findings, the authors concludedthat the OGTT test alone was superior (moreeffective and less costly) to a strategy involvingHbA1c followed by OGTT. The choice between theother options, they concluded, would depend onthe objective of the screening programme becausethe less effective strategies were also more efficientin terms of the cost per case detected. However,these results were found to be sensitive toassumptions about uptake of the different tests. Inthe base case, 100% uptake was assumed for allthe tests, but when these assumptions were variedduring sensitivity analysis (75% uptake for FPGand 50% uptake for OGTT), the FPG test becamethe more effective and efficient strategy. However,the authors did not seem to consider thepossibility that uptake of OGTTs would be higherafter positive screening tests than it would bewithout. Had this been investigated, the strategyinvolving the HbA1c test may have appeared morefavourable. Due to the lack of transparencysurrounding some of the calculations andassumptions in this analysis, and the limitedsensitivity analysis, it is not possible to say withconfidence which of the strategies would be mosteffective and efficient (in terms of cost per caseidentified) in the context of a national screeningprogramme in the UK.

Zhang and colleagues conducted a subsequentanalysis205 using a very similar approach to theirprevious study. However, in this later study theyexamined the efficiency of a wider range of cut-offpoints for three different screening tests –presenting the cost per case detected for each cut-off point assessed. In addition, they looked athow the efficiency of the different cut-offs wouldchange if the goal of the screening programmewere to identify cases of undiagnosed diabetesalone, as opposed to both prediabetes andundiagnosed diabetes. The three tests theyassessed were FPG, HbA1c and random CBG. Themost efficient cut-off point on each of these testswas defined as the point that yielded the lowest

cost per positive case identified. The sensitivitiesand specificities for each of the cut-off pointsassessed in the analysis were obtained from thepublished literature,8,10 but little detail was givenon these studies. As in the previous analysis,subjects with an FPG above the cut-off, but belowthe value used to define prediabetes (orundiagnosed diabetes), were assumed to undergoan OGTT. For the other two screening tests, allsubjects above the different cut-offs assessed wereassumed to undergo an OGTT. The numbers ofindividuals above the different cut-off points werepresumably modelled based on the reportedsensitivities and specificities for the different cut-off points on the different tests. As before, the costsand effects of the three strategies were modelledfor all eligible adults in the US population –54.4 million adults aged between 45 and 74 yearswho seek healthcare at least once per year – fromboth single-payer and societal perspectives. Thecosts were all reported in 2000 US dollars.

Zhang and colleagues205 found that, if the purposeof screening was to identify both prediabetes andundiagnosed diabetes, the cost per case identifiedby cut-off value ranged from $125 to $321 for theCBG test, from $114 to $476 for the FPG test andfrom $153 to $536 for the HbA1c test from thesingle-payer perspective. If the purpose was toidentify only undiagnosed diabetes, the cost percase detected from the single-payer perspectiveranged from $392 to $671 for the CBG test, from$556 to $717 for the FPG test and from $590 to$817 for the HbA1c test. It was found that for allthree screening tests, the cost per case identifiedfirst decreased and then increased as the cut-offvalue of the screening test increased. This wasbecause, to begin with, the decreased sensitivity(decreasing number of cases detected) wasoutweighed by the decreased costs associated withfewer diagnostic tests. However, after a certainlimit, the decreased number of cases detectedoutweighed the decreased costs, resulting in a risein cost per case detected. The optimal cut-offpoint based on the cost per case detected, fordetecting both pre- and undiagnosed diabetes, was100 mg/dl for the FPG test, 5% for the HbA1c testand 100 mg/dl for the CBG test. For detectingundiagnosed diabetes only, the most efficient cut-off points were 110 mg/dl, 5.7% and 120 mg/dl,respectively. The authors did not present thenumber of cases that would be detected at thedifferent cut-off points or the additional costs ofdetecting extra cases by increasing sensitivity.Again, the decision as to which screening test touse in a screening programme will depend on thenumber of cases that would be identified, and not


60


61


TAB

LE 1

6Su

mm

ary

of s

tudi

es a

sses

sing

the

shor

t-te

rm c

osts

and

out

com

es o

f alte

rnat

ive t

est

stra

tegi

es fo

r the

det

ectio

n of

dia

bete

s or

IGT/

IFG

Stud

y an

d Po

pula

tion

Obj

ecti

ves/

Stra

tegi

esC

osts

Out

com

esT

ime

Res

ults

/aut

hors

’ C

omm

ent

sett

ing

ques

tion

sho

rizo

n an

d co

nclu

sion

spe

rspe

ctiv

e

Shiri

saya

et

al.,

1999

203

Japa

n

1000

men

(ran

ge o

fpr

eval

ence

valu

es fo

rpr

edia

bete

s an

dun

diag

nose

ddi

abet

esas

sess

ed)

Whe

n fa

stin

g te

sts

are

not

prac

tical

,w

hich

scr

eeni

ngst

rate

gy is

mos

tef

ficie

nt fo

r de

tect

ing

case

s of

und

iagn

osed

diab

etes

alo

ne, a

ndbo

th p

redi

abet

es a

ndun

diag

nose

d di

abet

esto

geth

er?

1. 1

,5-A

G +

OG

TT

for

thos

e ab

ove

cut-

off

2. H

bA1c

+ O

GT

Tfo

r th

ose

abov

ecu

t-of

f3.

FRA

+ O

GT

T fo

rth

ose

abov

e cu

t-of

f (di

ffere

nt c

ut-

offs

use

d fo

rde

tect

ing

diab

etes

alon

e an

ddi

abet

es a

ndIG

T/IF

G)

Curr

ency

and

yea

r:19

97ye

n (d

iscou

nt r

ate

5%)

Uni

t co

sts:

FRA

55

yen

HbA

1c95

0 ye

n1,

5-A

G 1

450

yen

OG

TT

200

0 ye

n(a

ssum

ed n

o ex

tra

phys

icia

n tim

ere

quire

men

ts)

Sour

ces:

char

ges/

fees

True

-pos

itive

case

sde

tect

ed

Pers

pect

ive:

Sing

le p

ayer

Tim

e ho

rizon

:U

ntil

>99

% o

fca

ses

in c

ohor

tar

e ev

entu

ally

diag

nose

d

FRA

had

low

est

cost

per

case

det

ecte

d in

Japa

n fo

rbo

th IG

T a

nd d

iabe

tes

toge

ther

and

for

diab

etes

alon

e. R

esul

ts fo

und

tobe

sen

sitiv

e to

cos

t of

test

s an

d H

bA1c

and

1,5-

AG

wer

e m

ore

effe

ctiv

e (n

o IC

ERs

repo

rted

)

Pres

enta

tion

ofre

sults

and

sens

itivi

tyan

alys

is no

ttr

ansp

aren

t.C

anno

tge

nera

lise

conc

lusio

ns t

oot

her

sett

ings

John

son

et a

l.,20

0520

2

USA

cont

inue

d

US

popu

latio

nbe

twee

n 45

and

74 y

ears

elig

ible

for

popu

latio

nsc

reen

ing

(72.

6m

illio

n)

To a

sses

s th

e co

sts

and

outc

omes

of

seve

ral s

cree

ning

stra

tegi

es t

o de

tect

T2D

M b

ased

on

RPG

test

s

RPG

cut

-offs

of 1

00,

130

and

160

mg/

dlan

d a

mul

tivar

iate

equa

tion

at 1

-, 3-

and

5-ye

arly

inte

rval

s ov

er15

year

s (O

GT

T o

rFP

G t

o co

nfirm

all

posit

ive

scre

ens)

Curr

ency

and

yea

r:20

00U

S$ (n

o di

scou

ntin

gre

port

ed)

Uni

t co

sts:

Phys

icia

n tim

e $5

1 pe

rvi

sitRP

G $

5.24

FPG

$5.

24O

GT

T $

17.2

2Pa

tient

tim

e $8

per

hour

(1 h

our

for

RPG

/FPG

, 2 h

ours

for

OG

TT

)Tr

avel

$7

per

trip

Sour

ces:

Cha

rges

True

pos

itive

s,fa

lse p

ositi

ves,

true

neg

ativ

esan

d fa

lsene

gativ

es

Pers

pect

ive:

Sing

le p

ayer

and

soci

etal

Tim

e ho

rizon

:15

yea

rs (b

utno

follo

w-u

p of

indi

vidu

alpa

tient

s)

RPG

�13

0m

g/dl

eve

ry3

year

s de

clar

ed o

ptim

alst

rate

gy b

ased

on

effic

acy

findi

ngs

The

tot

al c

osts

of t

hest

rate

gies

ran

ged

from

$6.9

to

$42.

7 bi

llion

Cas

es d

etec

ted

rang

edfr

om 1

4 to

18.

5 m

illio

n

RPG

�13

0 m

g/dl

eve

ry3

year

s co

st $

642

per

true

pos

itive

(no

ICER

sw

ere

repo

rted

for

the

othe

r st

rate

gies

)

Incr

ease

d co

sts

of id

entif

ying

mor

e ca

ses

bym

ovin

g to

mor

ese

nsiti

vest

rate

gies

not

pres

ente

d


62 TAB

LE 1

6Su

mm

ary

of s

tudi

es a

sses

sing

the

shor

t-te

rm c

osts

and

out

com

es o

f alte

rnat

ive t

est

stra

tegi

es fo

r the

det

ectio

n of

dia

bete

s or

IGT/

IFG

(co

nt’d

)

Stud

y an

d Po

pula

tion

Obj

ecti

ves/

Stra

tegi

esC

osts

Out

com

esT

ime

Res

ults

/aut

hors

’ C

omm

ent

sett

ing

ques

tion

sho

rizo

n an

d co

nclu

sion

spe

rspe

ctiv

e

Icks

et

al.,

2004

201

Ger

man

y

Popu

latio

n 13

53in

divi

dual

sbe

twee

n 55

and

74 y

ears

from

the

Ger

man

city

of A

ugsb

urg

To a

sses

s th

e co

stpe

r tr

ue c

ase

dete

cted

of s

ever

alop

port

unist

icsc

reen

ing

stra

tegi

esfo

r T

2DM

1. S

ingl

e FP

G t

est

(�7

mm

ol/l)

2. F

PG t

est

+O

GT

T fo

r th

ose

with

fast

ing

gluc

ose

�6.

1 an

d<

7 m

mol

/l3.

Sin

gle

OG

TT

(�11

.1 m

mol

/l)4.

HbA

1c+

OG

TT

for

thos

e w

ithH

bA1c

>5.

6%

(Eac

h st

rate

gyas

sess

ed fo

r th

ew

hole

pop

ulat

ion

and

also

for

pres

elec

ted

grou

pw

ith e

leva

ted

risk

for

diab

etes

)

Curr

ency

and

yea

r: 20

02eu

ros

(disc

ount

ing

not

appl

icab

le)

Uni

t co

sts:

Pres

elec

tion

test

ing

(incl

udin

g ph

ysic

ian

time

and

test

s) €

12.1

8 pe

rvi

sitFP

G t

est

€14

.78

HbA

1c€

16.0

0O

GT

T €

16.3

4Pa

tient

tim

e €

29.1

9 pe

rho

ur fo

rem

ploy

ed/€

5.37

for

thos

e no

t w

orki

ng(1

hour

for

HbA

1c/F

PG,

2ho

urs

for

OG

TT

)Tr

avel

$7

per

trip

Sour

ces:

Fee

s/ch

arge

s

True

-pos

itive

case

sde

tect

ed

Pers

pect

ive:

Sing

le p

ayer

and

soci

etal

Tim

e ho

rizon

:1

year

Popu

latio

n-ba

sed

scre

enin

g st

rate

gies

dom

inat

ed t

he t

arge

ted

stra

tegi

es

From

sin

gle-

paye

rpe

rspe

ctiv

e, t

he s

ingl

eO

GT

T t

est

dom

inat

edth

e te

sts

invo

lvin

g FP

G.

The

HbA

1c+

OG

TT

stra

tegy

iden

tifie

d m

ore

case

s bu

t w

as a

lso m

ore

cost

ly t

han

OG

TT

alo

ne

Resu

lts s

ensit

ive

toas

sum

ptio

ns r

egar

ding

the

part

icip

atio

n ra

tes

for

the

diffe

rent

tes

ts

A la

ck o

ftr

ansp

aren

cyre

gard

ing

the

estim

atio

n of

mod

elpa

ram

eter

s an

dth

e us

e of

char

ges/

fees

spec

ific

toG

erm

any

mak

eit

diffi

cult

toas

sess

the

appl

icab

ility

of

thes

e re

sults

to

a U

K s

ettin

g.A

lso,

incr

emen

tal

cost

s an

d ef

fect

sof

incr

easin

g th

ete

st s

ensit

iviti

esw

ere

not

repo

rted co

ntin

ued


63


TAB

LE 1

6Su

mm

ary

of s

tudi

es a

sses

sing

the

shor

t-te

rm c

osts

and

out

com

es o

f alte

rnat

ive t

est

stra

tegi

es fo

r the

det

ectio

n of

dia

bete

s or

IGT/

IFG

(co

nt’d

)

Stud

y an

d Po

pula

tion

Obj

ecti

ves/

Stra

tegi

esC

osts

Out

com

esT

ime

Res

ults

/aut

hors

’ C

omm

ent

sett

ing

ques

tion

sho

rizo

n an

d co

nclu

sion

spe

rspe

ctiv

e

Zha

ng e

t al

.,20

0320

4

USA

US

popu

latio

nbe

twee

n 45

an

d 74

yea

rsel

igib

le fo

rop

port

unist

icsc

reen

ing

(54.

4m

illio

n)

To a

sses

s th

e co

sts

and

cons

eque

nces

of

vario

us s

trat

egie

s to

iden

tify

indi

vidu

als

with

pre

diab

etes

and

undi

agno

sed

diab

etes

1. S

ingl

e O

GT

T fo

rev

eryo

ne2.

FPG

+ O

GT

T fo

rth

ose

with

IFG

>95

but

<11

0m

g/dl

3. H

bA1c

+ O

GT

Tfo

r th

ose

with

HbA

1c>

5%4.

CBG

+ O

GT

Tfo

r th

ose

with

CBG

>10

0 m

g/dl

5. R

isk a

sses

smen

tqu

estio

nnai

re +

OG

TT

for

thos

esc

orin

g �

10

Curr

ency

and

yea

r: 20

00U

S$ (d

iscou

ntin

g no

tap

plic

able

)

Uni

t co

sts:

Phys

icia

n tim

e $5

1 pe

rvi

sitFP

G $

5.24

CBG

not

rep

orte

dH

bA1c

not

repo

rted

OG

TT

$17

.22

Patie

nt t

ime

$8 p

erho

ur (3

.25

hour

s fo

r al

lst

rate

gies

exc

ept

FPG

4ho

urs)

Trav

el $

7 pe

r tr

ip

Sour

ces:

Cha

rges

Prop

ortio

nan

d nu

mbe

rof

tru

e ca

ses

iden

tifie

dfr

ompo

pula

tion

Pers

pect

ive:

Sing

le p

ayer

and

soci

etal

Tim

e ho

rizon

:1

year

The

cos

t pe

r ca

seid

entif

ied

rang

ed fr

om$1

76 t

o $2

36 fr

om t

hesin

gle-

paye

r pe

rspe

ctiv

e

Test

ing

all w

ith O

GT

Tw

as t

he m

ost

effe

ctiv

ebu

t C

BG t

est

and

risk

asse

ssm

ent

ques

tionn

aire

wer

e th

e m

ost

effic

ient

Resu

lts w

ere

sens

itive

to

assu

med

par

ticip

atio

nra

tes

for

the

diffe

rent

test

s

Lack

of

tran

spar

ency

surr

ound

ing

som

e of

the

calc

ulat

ions

and

assu

mpt

ions

,an

d on

ly li

mite

dse

nsiti

vity

anal

ysis,

mak

e it

diffi

cult

to a

sses

sth

e in

tern

al a

ndex

tern

al v

alid

ityof

the

find

ings

Zha

ng e

t al

.,20

0520

5

USA

US

popu

latio

nbe

twee

n 45

an

d 74

yea

rsel

igib

le fo

rop

port

unist

icsc

reen

ing

(54.

4m

illio

n)

To a

sses

s th

eef

ficie

ncy

of a

ran

geof

cut

-off

poin

ts o

nth

ree

diffe

rent

tes

tsfo

r de

tect

ing

undi

agno

sed

diab

etes

alon

e, a

nd b

oth

pred

iabe

tes

and

undi

agno

sed

diab

etes

toge

ther

1. F

PG +

OG

TT

for

thos

e w

ith IF

Gab

ove

the

cut-

off

but

<11

0 m

g/dl

2. H

bA1c

+ O

GT

Tfo

r th

ose

with

HbA

1cab

ove

cut-

off

3. C

BG +

OG

TT

for

thos

e w

ithC

BG a

bove

cut

-of

f

(Diff

eren

t cu

t-of

fsex

plor

ed t

o fin

d th

em

ost

effic

ient

cut

-of

fs fo

r de

tect

ing

diab

etes

alo

ne, a

nddi

abet

es a

ndpr

edia

bete

sto

geth

er)

Curr

ency

and

yea

r: 20

00U

S$ (d

iscou

ntin

g no

tap

plic

able

)

Uni

t co

sts:

Phys

icia

n tim

e $5

1 pe

rvi

sitFP

G $

5.24

CBG

not

rep

orte

dH

bA1c

not

repo

rted

OG

TT

$17

.22

Patie

nt t

ime

$8 p

erho

urTr

avel

$7

per

trip

Sour

ces:

Cha

rges

Cos

t pe

r ca

sede

tect

ed(c

osts

and

effe

ctiv

enes

sre

sults

not

repo

rted

sepa

rate

ly)

Pers

pect

ive:

Sing

le p

ayer

and

soci

etal

Tim

e ho

rizon

:1

year

Mos

t ef

ficie

nt c

ut-o

ffpo

ints

for

dete

ctin

gdi

abet

es a

lone

wer

e11

0m

g/dl

for

the

FPG

test

, 5.7

% fo

r th

e H

bA1c

test

and

120

mg/

dl fo

rth

e C

BG t

est

Cor

resp

ondi

ng c

ut-o

ffsfo

r de

tect

ing

pre-

diab

etes

and

dia

bete

sto

geth

er w

ere

100

mg/

dl,

5% a

nd 1

20 m

g/dl

Cos

ts a

ndef

fect

s no

tpr

esen

ted

sepa

rate

ly a

ndno

incr

emen

tal

cost

s pe

r ex

tra

case

s de

tect

edfr

om in

crea

sing

sens

itivi

typr

esen

ted

just the cost per case identified. It will also bedependent on the costs and consequences of falsenegatives, which will in turn be dependent on therate of progression of undiagnosed cases relativeto diagnosed cases, and the screening interval. If adecision was made to screen every 2 or 3 years,then the adverse consequences of false negativeswould be less severe than they would be ifscreening was to be just once off.

Summary and conclusionsThe studies reviewed in this section201–205 allassessed the short-term costs and consequences ofdifferent approaches to screening for T2DM, orIGT and IFG. All considered the costs that wouldbe incurred and the number of cases that wouldbe identified using different tests or different cut-off points on the same test. The studies all hadsimilar designs although estimates of cost per casedetected varied due to different assumptions aboutcosts, prevalence of diabetes/pre-diabetes,sensitivity/specificity of tests and participationrates. Moreover, the efficiency ranks for thedifferent tests varied from study to study and werealso found to be sensitive to assumptions withinstudies. Due to these uncertainties, it is notpossible to draw any conclusions as to which testwould be most efficient in a screening programmefor detecting diabetes and/or IGT/IFG in the UK.One thing that is clear is that when considering anappropriate cut-off point for a test, there is likelyto be a trade-off between the number of cases thatthe test can identify and the efficiency of that test.As the sensitivity of a test is increased beyond acertain limit, by lowering the cut-off point, thecost per case identified is also likely to increasedue to the larger number of false positives thatwill require a definitive diagnostic test. As Zhangand colleagues demonstrated,205 the oppositeeffect is seen when the specificity of a test isincreased beyond a certain limit. Therefore, thepreferred test and the preferred cut-off points fortests will vary depending on the precise objectives

and nature of the screening programme. If it isconsidered very important to detect as many truecases of disease as possible and minimise falsenegatives, then more sensitive strategies withhigher costs per case detected might beappropriate. If the false negatives are notconsidered to be such an important consequenceof screening, then a more specific and efficientscreening strategy could be used.

The main problem associated with all the studiesreviewed in this section is that by only consideringthe short-term costs and outcomes associated withthe detection process, and failing to consider thelonger-term impact of false negatives relative totrue positives, these types of studies cannot tell uswhich screening tests and cut-off point to use inscreening programmes for diabetes/IGT. If falsenegatives result in substantial increases in costs orreductions in life expectancy/quality of life, relativeto true positives, then more sensitive andexpensive test strategies are likely to be thepreferred option. However, if the long-termimpact of false negatives is not particularly severe,more specific strategies might be preferred. Thismight be the case if repeat screening were to beundertaken every 2 or 3 years.

In order to address these questions, the differentscreening tests and cut-off points need to beassessed in the context of models that consider thelifetime costs and outcomes of screening forT2DM. Such models were reviewed earlier in thisreport. By incorporating the costs andsensitivities/specificities of different screeningstrategies into such models, it would be possible toassess the longer term impacts that different teststrategies might have relative to each other. Thiswould provide a way of balancing the short-termcosts of the different screening strategies againstthe long-term costs and health outcomes, giving abetter indication of how to optimally screen fordiabetes.


64


65


IntroductionContext of researchThe aim has been to produce a basic screeningmodel linked to the existing fully developedtreatment model to investigate the order ofmagnitudes of effects given different scenarios fordiabetes screening policies.

BackgroundT2DM increases the risk of cardiovascular events inaddition to causing microvascular complications.The disease remains asymptomatic for a number ofyears during which the disease progresses andcardiovascular risk is elevated compared with thenormal population. It is estimated that up to onemillion people have undiagnosed T2DM in theUK.209 Screening for diabetes is potentiallybeneficial as early treatment may delay progressionof diabetes and reduce short- and long-termcomplications. Treatments include statins,antihypertensive and hypoglycaemic therapy andlifestyle change (dietary modification and increasedphysical activity).

Although there have been previous assessments ofthe cost-effectiveness of screening for diabetes,these have some limitations such as beingperformed before the use of statins.

ObjectivesIn order to quantify the trade-off between thecosts and benefits of screening and earlytreatment, a diabetes screening model wasdeveloped for this study and integrated with anexisting diabetes treatment model. This enabled:

● information derived from the screening literaturereview to be integrated with information on thecurrent and projected prevalence of undiagnoseddiabetes and information on the costs andeffectiveness of diabetes-related treatment

● the key parameters that will influence the cost-effectiveness of screening for diabetes and thesignificance of uncertainty around these keyparameters to be identified

● the characteristics of the screened population(including the proportion of diabetes

undiagnosed in the absence of systematicscreening), the characteristics and costs of thescreening and diagnostic tests to be varied

● the effectiveness and costs of subsequenttreatment to be varied

● initial estimates of the relative cost-effectivenessof a range of different screening options thatcould be used to inform an option appraisal fordiabetes screening to be produced.

Some existing limitations of previous publicationson screening modelling were also overcome, such as:

● modelling macrovascular risk reductions● incorporating the widespread use of generic

statins● accounting for existing policies such as diabetes

testing for those with existing CHD, and controlof hypertension in the general population.

Model scenariosBaseline scenarioThe baseline model assesses the cost-effectivenessof a single screening round of a population withan age of 40–70 years in order to identify andtreat diabetes before clinical diagnosis wouldoccur. The modelling covers a period of 40 yearsfrom the decision to screen or otherwise.

Other scenariosIn sensitivity analyses, the cost-effectiveness ofscreening different populations was assessed, suchas those in the 40–49-year age band, thehypertensive subgroup and the obese subgroup.Sensitivity analyses also tested the impact ofuncertainty in some of the parameters and modelassumptions, such as the costs of statins.

MethodsModel structureScreening modelThe principal approach to model-basedevaluations of screening interventions is to (1) develop a natural history model covering thepreclinical detectable phase (PCDP) and (2) then

Chapter 5

Modelling the cost-effectiveness of screening for type 2 diabetes

overlay relevant screening programmeinterventions (and also the ‘no screening’ optionin which diabetes is detected clinically).

The link between the screening programme andthe PCDP provides outcome measures includingthe proportion of persons detected prior toclinical presentation and the timing of the screen-detection. For diabetes, screening outcomes can bemeasured in terms of number of new casesdiagnosed, distribution of HbA1c levels atdiagnosis and distribution of ages at diagnosiscompared with ages at which diabetes wouldotherwise present (i.e. the reduction in time spentin the preclinical phase). These inputs feed intothe treatment model so it is possible to compareoverall lifetime costs and outcomes of a screenedcohort and an unscreened cohort.

Treatment modelThis simulates the development of microvascularand macrovascular complications in patients withT2DM. Microvascular risk is based on HbA1c andblood pressure, whereas risk of developingcardiovascular complications uses a modifiedversion of the UKPDS CVD Risk Engine (UKPDS56184 and UKPDS 60185), which is based oncardiovascular risk and outcomes in the UKPDStrial cohort (including treated and control arms).

The structure of the model is shown in Figure 3.

Note that the pathway after screening for thefollowing people is assumed to be the same as ifthey had not been screened (regardless of whetherthey have diabetes or not):

● patients who test negative at screening test● patients who test negative at diagnostic test.

Where these pathways are common regardless ofthe choice to screen or not, there is no need tomodel their long-term pathway because the costsand benefits cancel out. It is only necessary toaccount for the costs of the screening programmeitself.

Key model assumptionsScreening and diagnostic processA single HbA1c measurement was assumed as thescreening test, followed by a single OGTT if theHbA1c is above 5.7%. There is much debate aboutthe optimum test strategy201 and the model canincorporate more complex strategies, includingthe use of a questionnaire, followed by a screeningblood test before diagnostic testing. However,there is agreement that in the absence ofsymptoms (as is the case in screening) a secondglucose measurement is required before anyone isdiagnosed as diabetic.


66

Exit model

Exit model

Exit model

Undetected diabetics –enter progression modeluntreated*

Enter progression/complications model*

Screening

No screening

Negative test

Positive test

Diabetic

OGTT(diagnostic)

Non-diabetic

Complication-free

Complications/mortality

AnnualMarkovmodel

Clinical detectionand treatment

Annualupdate

Remain undetected

Complication-free

Complications/mortality

FIGURE 3 Model structure.* Matched patients in screening and ‘no screening’ model.

Natural history of HbA1c progressionThe potential benefit of earlier diagnosis andtreatment of hyperglycaemia will be closely relatedto the distribution of HbA1c results at screeningand its progression up to the point of clinicaldiagnosis.

Overview of methodTo obtain the duration between screening andclinical detection, HbA1c distributions weresampled at screen and clinical detection, andestimated rates of HbA1c progression used tocalculate to preclinical duration. This method waschosen as it provides some degree of variability inthe preclinical duration which was thought to bean important factor in determining the overallrisk profile of patients during the preclinicalperiod. Variability in the rate of HbA1cprogression, however, was considered toouncertain to model given that HbA1c progressionis the most uncertain aspect of the natural historyto estimate (see below).

Although there may be some correlation betweenHbA1c at screen detection and the HbA1c at whicha patient would be clinically detected, this wasthought to be low and not significant. Thereforethe screen and clinical HbA1c distributions weresampled independently.

HbA1c at screen detectionFirst, a distribution of HbA1c is assumed forscreen-detected patients. Results from screeningprogrammes are summarised in Table 17.

It was assumed that HbA1c values were reportedon the required DCCT aligned scale (the riskequations in our model are UKPDS-aligned). Datasuggest that the FPG test yields a higher HbA1cthan an OGTT test – this is expected becausepostprandial hyperglycaemia is often the more

dominant defect in early diabetes. Weighting hasbeen given to studies that used an OGTT test asthis test is the most sensitive, and it is assumedthat the mean HbA1c of the undiagnosed diabeticpopulation is 6.4%.

The HbA1c test is effectively a ‘hybrid’ of an FPG and an OGTT test, resulting in a sensitivitybelow 100%. To obtain a sample of HbA1c valuesfrom the undiagnosed population, samples weretaken from the upper region (starting from 1 – sensitivity), giving a mean HbA1c throughscreening of 6.7%.

The 1998 CDC screening evaluation131 used asimilar mean HbA1c of 6.8% at screen detection.An SD of 1.0 was assumed, as reported from theKORA survey,206 this is consistent withassumptions in the CDC screening modelling. The reviewers have assumed a skewed distribution,as the higher the HbA1c the more likely patientswould be symptomatic and be detected clinically,this assumption is supported by the HbA1cdistribution shown in Figure 3 of Hofer andcolleagues213 based on NHANES III data.

HbA1c at clinical detectionPatients were randomly assigned to an HbA1c levelthat would have been found in the absence ofscreening, that is, clinical detection using theHbA1c distribution from the UKPDS (which was onaverage 9% but was higher in symptomaticpatients). Another study also found thatsymptomatic patients had a slightly higher HbA1cof 9.9% at diagnosis.214 A higher HbA1c level of10.8% was observed in the Poole Diabetes Study,215

although the scale used here has not beenconfirmed. The assumption of an average HbA1cof 9% at clinical detection may be slightlyconservative in terms of estimating the duration ofundiagnosed diabetes.


67


TABLE 17 HbA1c at screen detection

Study Test Mean HbA1c (%) FPG (mmol/l)

KORA survey201 OGTT 6.2

Hoorn111 Mixed 6.7

NHANES III210 ADA criteria (FPG levels) 7.07WHO criteria (OGTT results) 6.58

EDIP211 (estimated from Figure 3) FPG + HbA1c 6.5 6.6

UKPDS – low FPG group of which 44% asymptomatic184,185 Unknown 6.7

Other UK studies:Ely,137 Coventry Diabetes Study212 Not available


68

Although there is some variation in HbA1 atdiagnosis by age group, as shown in the PooleDiabetes Study (their Figure 2),215 this isconsidered to be of minimal significance, becausethere will be little effect on the incremental riskbetween the two cohorts.

HbA1c progressionHaving established two HbA1c distributions atscreen and clinical detection, the remainingparameter required is the HbA1c progressionduring the preclinical phase – this will alsodetermine the additional time duration that would pass before clinical detection. Althoughindirect estimates of the duration have been made,these vary greatly from 6 to 11 years and mayinvolve unjustified assumptions about HbA1cprogression.

We believe that the best way to model thisuncertainty is firstly to consider the HbA1ctrajectory of an average person with diabetes:

● To review evidence on the rates of change inHbA1c where possible – results of patientsdeveloping diabetes in IGT studies can give anidea of the rate close to onset of diabetes; theUKPDS gives an idea of the rate of change atclinical detection. A modelling paper by Bagust and Beale60 using results from theBelfast Diet Study also suggests a possibleHbA1c trajectory.

● To explore a range of scenarios for the HbA1ctrajectory (for an average patient).

● To identify the scenario which has the best fitwith other sources of evidence on diseaseprogression.

Average HbA1c in non-diabetics is around 5.4%.We estimate HbA1c at onset of diabetes to bearound 5.85%. Although the DPP reported an HbA1c of 6.4% at onset of diabetes,90,171

(assumed onset as tested 6-monthly), this could be because those that progressed had more severebeta-cell dysfunction (rather than modifiableinsulin resistance), leading to a rapid rise inHbA1c.

There is little hard evidence on which to base therate of change of HbA1c. Some aspects of thenatural history can be inferred from varioussources – ‘triangulation’ using various sources of evidence is considered to be the only approach to estimating the natural history as there is no existing dataset that demonstrates the natural history of untreated diabetes. Various results from studies do seem to suggest

a lower rate of change in the earlier part of thepreclinical phase.

In the US Diabetes Prevention Program,90 38% ofplacebo patients became diabetic at 4 years. HbA1cchange in the placebo group during this periodwas 0.2%. For the placebo patients, even if it isassumed that HbA1c rose three times faster inthose who became diabetic, we estimate thatHbA1c would only have risen at an annual rate ofabout 0.1%.

In the Finnish Diabetes Prevention Study,89 HbA1cdid not change at least up to the end of year 3 inthe control group even though diabetes incidencewas significant (21, 23 and 42 at 3, 4 and 6 years,respectively), although the control group didreceive some dietary information makinginterpretation more difficult.

Of note is the HbA1c level at screen detection, thatis, between 6.0 and 6.5%. This is much closer tothe HbA1c level at onset (which is estimated to beabout 5.8–5.9%) than the level at clinical detection(9% per UKPDS). This suggests one or both of thefollowing could be happening:

● The rate of HbA1c progression increases eithergradually or increases more rapidly at a specificphase of progression such that patientsgenerally are less likely to be detected in themore advanced preclinical phase (because therewould be relatively less time in this phase).There is a biological study that supports thishypothesis.216

● Some slower progressors may not be detected inthe latter stages of the preclinical phase becausetheir raised CHD risk may result in mortalityduring this period – this would have the effectof detecting patients at relatively low HbA1clevels

The International Diabetes Center, Minneapolis,MN, produced a diagram217 showing ahypothesised trajectory for glycaemic progression– this appears to be frequently referenced and is consistent with an up-sloping HbA1ctrajectory.

The rate of change in HbA1c during the first fewyears of the UKPDS was around 0.2% p.a., despite antiglycaemic therapy.8 The expected rate of increase would be greater withouttreatment.

Taking account of the limited evidence above, it was assumed that, for the baseline analysis,

the rate of HbA1c change increases exponentiallyover time from 0.15 p.a. at diabetes onset to0.45 p.a. at clinical detection (assumed to be at9.0% HbA1c on average). This is clearly an area forfurther research. The impact of a more rapidexponential increase (and shorter preclinicalperiod) was examined in the sensitivity analyses.

Reduction in time to diagnosis through screeningEstimates of the expected lead time betweenscreen and clinical detection are an indirect output of this modelling approach and can becompared with the findings with other sources ofdata using other methods to estimate thepreclinical course.

The estimated average delay in diagnosis throughnot screening is approximately 6–7 years. Theestimated sojourn period between onset andclinical detection is around 11 years. Thedistribution of the delay obtained is shown inFigure 4.

Changes to other risk factors during thepreclinical periodThe assumptions made about baselinecharacteristics and changes to these during thepreclinical period of the unscreened cohort areimportant.

Systolic blood pressure (SBP)SBP in the UKPDS at baseline was 136 mmHg. Itwas assumed that there is reasonably tight SBPcontrol in undetected patients given current NHSincentives for GPs, although not as tight as therecommended level for people with diabetes. Itwas also assumed that SBP is at most 150 mmHgat baseline.

For annual changes in SBP, data from the HealthSurvey for England 200363 (Volume 2, Risk factorsfor cardiovascular disease, Figure 7C, p. 186) wereused.

Approximate changes (in the general population)were calculated as follows:

● Rise p.a. women = 118 to 137 mmHg from age40 to 60 years, or approximately 1 mmHg p.a.

● Rise p.a. men = 129 to 138 mmHg, from age40 to 60 years, or approximately 0.5 mmHg p.a.

It was assumed due to the higher prevalence ofhypertension in diabetes, that the SBP rises attwice the general population rate.

CholesterolTotal cholesterol was 5.4 mmol/l at baseline in theUKPDS.8


69


0

2

4

6

8

10

12Fr

eque

ncy

(%)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Time (years)

FIGURE 4 Distribution of reduction in time to diagnosis through screening

Changes in cholesterol vary by age and gender inthe general population. Further uncertainty arisesfrom some cholesterol monitoring and the impactof diabetes, so no changes in cholesterol levelshave been included (this is consistent withobservations in UKPDS218).

No annual changes in cholesterol level areassumed.

Modelling of treatment and outcomesRisk equationsThroughout the time horizon of the modelling(i.e. including the preclinical period), the UKPDSCHD and stroke risk equations were used tomodel the risk of cardiovascular events. Theequations published by Eastman for microvascularevents were also used.134,135

Treatment assumptions from the point ofdiagnosisModelling the no-screening comparator involvesinitially modelling patients without statin orfurther antihypertensive therapy or diet/exerciseor antiglycaemic therapy. At a patient’sdetermined point of clinical detection, they beginactive treatment in the model. The model assumesthat from diagnosis, both hyperglycaemia andother CHD risk factors such as hypertension anddyslipidaemia, are treated according to currentclinical guidelines. The widespread availability,falling cost and significant cardiovascular riskreduction (30%) of statins is a key factor in theeffectiveness of screening.

The model assumes that if proliferativeretinopathy or macular oedema develops, thenthis would be treated using photocoagulation,leading to a 45% reduction in risk of severe visionloss. For the purposes of costing retinopathyscreening, it was assumed that this takes placeannually.

An area of uncertainty is whether reductions inHbA1c lead to reductions in CHD. A recent reviewconcludes that the evidence suggests that chronichyperglycaemia is associated with increased CHDrisk.219 A recent analysis reported from the follow-up of DCCT participants (type 1 diabetes), theEpidemiology of Diabetes Interventions andComplications (EDIC) study, may be highlysignificant – tight HbA1c control (7% versus 9% forconventional control) reduced CVD events byabout 50%.220 For our base-case analysis, theresults reported in UKPDS 35221 were used,namely that each 1% decrease in the observedmean updated HbA1c is associated with a 14%

lower incidence of MI. Sensitivity analysis wasundertaken to test the impact of varying thisassumption.

A detailed description of the Sheffield Treatmentmodel should appear shortly as an on-lineDiscussion Paper222 until the full publication isavailable. The modelling of risk factors post-diagnosis is covered in detail in the ‘Backgroundto methods’ section at the end of this chapter.

Other assumptionsPathways following screeningA key assumption is that the only long-termmodelling required involves comparing theprogression of the true positives under screeningwith their progression if they had been detectedlater clinically. This assumes that a negativescreening test or a positive screening test followedby a negative diagnostic test does not have anyclinically significant long-term consequences,although screening itself will potentially have animpact on both quality of life (by provokinganxiety or reassurance) and risk behaviour (due toindividuals becomes more or less motivated tochange their diet or lifestyle) depending on thecontext of screening (these issues are welldescribed in papers by Speight223 and Edelmanand colleagues224). It is assumed that these effectscan be modified by providing appropriateinformation and support at the time of screeningand have not been included in the modelling. Themain impact of diabetes on health-related qualityof life (HRQoL) is through the development ofcomplications.

CostsThe model considers costs from an NHSperspective. The costs of screening tests and costsof subsequent treatment for both diabetes andassociated complications are included.

Costs borne by patients or society such as costs tothe individual of attending for glucose tests or thecosts to society of supporting an individual withdiabetes-related blindness are not included.

Key model parametersPopulation characteristicsThe age, sex and ethnicity structure of thescreened population is based on the PBS model developed by the Yorkshire and HumberPublic Health Observatory,225 available atwww.yhpho.org.uk.3 Details on how the populationprevalence figures were calculated for alternativesubgroups are in the ‘Background to methods’section at the end of this chapter.


70

The population size of subgroups based on agegroup, diagnosed hypertension and BMI werecalculated using information from the PBS modeland recent data from The Health Survey forEngland.

As the PBS prevalence figures are at 2001 levels,an uplift of 10% was applied to estimate theprevalence in 2007 based on assumptionspublished with the PBS prevalence model.226

The percentage of the diabetic population that isundiagnosed can be varied. The PBS prevalencemodel documentation suggests that the figure liesbetween one-third and half based on Wild andcolleagues227 and King and Rewers.228 InNHANES III,210 35 and 45% of diabetes caseswere undiagnosed by ADA and WHO criteria,respectively. More recent estimates, based on aNational Diabetes Audit,229 suggest that withcurrent levels of ad hoc screening activity,approximately 25% remain undiagnosed inEngland. The default figure was set to 35%.

Baseline complication ratesMacrovascular complicationsWhen evaluating the effectiveness of screeningolder age groups, the increased prevalence ofCHD at baseline is important given the increasedrisk of subsequent MI. However, patients with aCHD history should be screened for diabetes as anelement of secondary prevention and it wastherefore assumed that the baseline prevalence ofCVD amongst the cohorts for potential screeningwould be negligible.

Similarly, prevalence of prior stroke at diagnosis isassumed to be nil because these people have aCVD profile and in theory should be covered by abroader prevention programme other than apurely diabetic screening programme.

Undiagnosed CHD would be increasinglyprevalent at baseline as age increases. As both ageat diagnosis and duration since diagnosis arefactors in the UKPDS risk engine, this trend isaccounted for within the model.

Microvascular complicationsMicrovascular complications are related toglycaemia and blood pressure. There is debate,however, about the level of glycaemia needed forretinopathy to develop. A recent report from theADA Conference indicated that retinopathy ispresent in some prediabetic patients.230 Theassumption was used that the prevalence of (non-proliferative) retinopathy increases linearly from

nil at the average HbA1c at onset (5.85%; see endof paragraph) to 17% (330/1919 with Level 35 orworse)231 in the UKPDS at an HbA1c of 9%. This isclose to the 15% prevalence recently reportedfrom the Tayside Diabetes Network232 and the17% based on the Early Diabetes InterventionProgram (EDIP) study.211 Analyses from NHANESIII and other studies show a marked change inprevalence above an HbA1c of 5.9%,17 confirmingthat our assumption above is reasonable.

An individual’s presence or otherwise ofretinopathy at entry to the simulation isdetermined by their HbA1c at entry.

A similar relationship is assumed for baselineprevalence of nephropathy, with the prevalenceassumed at diagnosis at an HbA1c level of 9%being 6.5% for microalbuminuria and 0.7% formacroalbuminuria.183

At an average HbA1c of 6.7% at screen detection,estimated prevalence of retinopathy,microalbuminuria and macroalbuminuria is 4.6,1.8 and 0.2%, respectively. As the prevalence ofneuropathy at diagnosis was only 1.2%,233 it hasbeen assumed that this is negligible at screendetection.

Uptake of screeningThis is set to 50% based on uptake rates within thecurrent UK NSC’s diabetes screening pilots inEngland (the DHDS pilots), where patients arebeing invited for screening by letter. It has beenassumed there is no uptake bias and therefore thatthe screened population is similar to the generalpopulation that would be eligible for screening.

Screening and diagnostic testsIn the baseline model, the screening test assumedis an HbA1c test with a cut-off of 5.7%, assuggested by Zhang and colleagues,205 as theoptimal cut-off. They reported sensitivity andspecificity at different HbA1c cut-off levels, thesewere 66 and 83%, respectively, at a cut-off of 5.7%.

An OGTT is assumed to be performed to confirm(or otherwise) diagnosis in patients who areinitially screen positive using the HbA1c test.

CostsTable 18 shows the key costs associated withscreening and early treatment.

It was assumed that no additional costs of case-finding would be incurred on the grounds thatsuitability for testing would be evident at the time


71


of a routine GP appointment (i.e. assumingopportunistic screening rather than a newdedicated mass screening programme).

The costs for an HbA1c test and OGTT test areestimated to be approximately £5 and £20,respectively, including nurse time, transport andlaboratory processing costs. No societal costsassociated with patients’ lost time is included.Additional GP infrastructure costs are assumed tobe avoided through adoption of opportunisticscreening when patients visit GP practices for‘routine’ appointments. No costs of monitoringthe screening programme are assumed.

Costs of therapies were obtained from the NHSDrug Tariff234 (which are generally considerablylower than BNF prices).

The main additional cost of statins arising fromscreening occurs during the period in which theywould otherwise have been undiagnosed (somefurther additional costs are attributable to improvedsubsequent survival). It was assumed that 40 mg of(generic) simvastatin is sufficient to managecholesterol levels during this period and appliedthis cost throughout the model. Average costs ofstatins were therefore estimated at £ 0.17 per day.

Insulin cost was based on the average cost ofMonotard (£10.50 per 1000 units) and Lantus(Insulin Glargine) (£26 per 1000 units).

Costs of treatment of complications are as specifiedfor the Sheffield Diabetes Treatment Model.

Treatment costs are from an NHS perspective; forexample, no costs are included for carers ofpatients who develop blindness.

Costs are expressed in real terms (i.e. no inflationincluded).

Discount ratesCosts and benefits were discounted in line with thenew NICE guidance for England and Wales of3.5% for both.

Screening scenariosIn addition to baseline scenarios for the age bands40–49, 50–59 and 60–69 years, sensitivity analyseswere undertaken to demonstrate the relationshipbetween the key parameters and the modeloutcomes. One-way sensitivity analyses include theimpact of screening a higher risk population(hypertensive and obese populations), the impactof varying treatment regimes (more intensiveglycaemic control), treatment costs (statin costs)and the impact of varying the assumed naturalhistory (progression of hyperglycaemia prior toclinical diagnosis).

Model outputsFor a given population, the model predicts thenumber screened, the number of new cases thatwould be diagnosed by screening, cardiovascularevents (MI, and unstable angina and stroke) andmortality in the presence and absence ofscreening, microvascular complications and the overall costs and QALYs associated withscreening. Cost-effectiveness is presented in terms of:

● Marginal net benefit (i.e. the financial value,above that expected to have to be invested, torealise the QALY gains using an acceptabilitythreshold of £ 20,000 per QALY)

● The ICER – the ratio of incremental costs toincremental QALYs.


72

TABLE 18 Key cost inputs relating to screening and early treatment

Resource Unit Cost (£) Note

Case finding Per case 0

Screening test (HbA1c) Test 5.00 This is based on a conversation with laboratorycolleagues, suggesting a range of costs from about£3.50 to £20 depending on volume and speed ofresults; a figure has been taken near the lower endof that range assuming that screening would involvelarge numbers but with no need for speed

Diagnostic test (OGTT) Test 20.00 HTA Review of screening for gestational diabetes2

Cost of statins Per day 0.17 40 mg generic simvastatin234

Non-medication costs of monitoring/managing Per year 97.00 Calculation based on several sourcesdiabetes (based on two GP visits with HBA1ctests, and retinopathy screening)

The marginal net benefit is more readilyinterpretable, particularly where the ICERcalculation involves negative figures, as is often thecase in our results.

ResultsResults from baseline scenarioScreening outcomesScreening a cohort of patients between 40 and70 years of age in a Primary Care Trust (PCT)population of 100,000 individuals would meanthat 33,740 people would be invited for screening,of whom 16,870 would respond and take ascreening test; 3078 would need diagnostic testingand 283 (0.3%) would be new cases of diabetes.Based on our model, they would be detected anestimated 6–7 years earlier than in the absence ofscreening. In practice, only those with BMI over30 kg/m2 might be invited.

Cost-effectiveness of screening (per persondiagnosed)Overall £715 would be the cost of additionaltreatment and monitoring, largely resulting fromearlier diagnosis, and £880 would be saved fromthe reduction in complications and theirtreatment. The apportioned total cost of screeningand diagnosis per diagnosed case would be £516.This represents an overall additional cost of £351and a gain in QALYs of 0.155 per case detectedand treated, giving an incremental cost per QALYratio of £2266.

Sensitivity analysesNumbers screened and test results are given inTable 19. The relative proportions withundiagnosed diabetes and IGT would varydepending on cut-off level chosen. With lower cut-offs, IGT would be more common.)

The screening strategy (i.e. choice of test andHbA1c cut-off) and assumed uptake rate wouldleave 575 people in the 40–70-year age bandundetected per 100,000 total population; 428 ofthese would not have taken up screening, with theremainder being those who are screened butwhose HbA1c is below the 5.7% cut-off level for apositive result.

The numbers to screen are obtained from the PBSprevalence model by providing weighted averagesof age-specific prevalence within 5- or 10-year agebands. As the numbers are based on the mix ofthe total population of England, they representthe expected number from a ‘typical’ PCTpopulation.

The results of one-way sensitivity analyses for thetreatment of diagnosed patients at screendetection compared with clinical detection (i.e. excluding screening costs) are given in Table 20.

This modelling suggests that screening is cost-effective for all populations between 40 and70 year of age, and for all one-way parameter andtreatment sensitivity analyses around the basecase. It cannot be concluded without furtheranalysis, however, that this would be the case forsensitivity analyses around every population. Thesame applies if some of the sensitivity analysisassumptions were combined (e.g. highermonitoring costs and lower CHD reduction fromHbA1c lowering).

Age groupsTable 20 shows additional treatment costs andQALY gains per new case detected. All age groupsshow a QALY gain of at least 0.12. The 50–59-yearand 60–69-year age groups both show cost savingsfrom treatment also. After the costs of screening


73


TABLE 19 Numbers screened and test results

Screening scenario Undiagnosed Number to screen Detected and Detected and diagnosed prevalence (%) (per 100,000 diagnosed diabetic as IGT/IFG

population)a (approximate estimate)

All 40–70-year olds 2.50 16,869 283 17740–49 years of age 1.40 6,535 60 3750–59 years of age 2.20 6,051 88 5560–69 years of age 4.80 4,283 135 84Hypertensive (age 40–70 years) 4.40 6,083 176 110Obese (BMI >30 g/m2) 4.10 4,004 108 67Higher uptake (70%) 2.50 23,616 396 247

a Assuming 50% uptake of those invited to screening.

are taken into account, cost-effectiveness appearsgreatest for the 60–69-year age group (Table 21).This is partly because more cases are diagnosed inthe 60–69-year age band due to the higherprevalence (see Table 19), resulting in the effectivescreening cost per case detected being muchlower in the 60–69-year age group (£590 and £281per case detected for 50–59- and 60–69-year agegroups, respectively).

The 40–49-year age group yields additionaltreatment costs, lower QALY gains and higherscreening costs per case detected due to lowerbaseline risk of complications and lowerprevalence.

High-risk populationsPopulations at higher risk of diabetes are also athigher risk of cardiovascular complications in thepresence of diabetes, for example if older orhypertensive. The model suggests that QALYgains will be greater in some higher riskpopulations, as shown above for the hypertensivesubgroup.

PrevalenceThe degree of uncertainty in the prevalence ofdiabetes is unlikely to affect whether screening iscost-effective or not, unless several of the‘pessimistic’ assumptions in the sensitivity analysesapply.

Preclinical HbA1c trajectoryA more rapid rise in HbA1c prior to diagnosis willinfluence both the potential benefit and thepotential duration of additional treatment. Themodel suggests that this will reduce the additionalcosts largely by reducing the duration of additionaltreatment (to about 4 years under this assumption)while impacting on later complication rates.

CHD benefit of HbA1c reductionsThe assumption that reducing HbA1c will reduceCHD rates is one of the reasons for the significantdrop in CHD complications with earlier treatment.However, even after eliminating this benefit, thetreatment of associated risk factors (statins,antihypertensives, aspirin) will still produce asignificant benefit.


74

TABLE 20 Results of one-way sensitivity analyses for early treatment compared with treatment at clinical detection per casediagnosed (excludes screening costs)

Screening scenario Incremental costs Incremental QALYs compared with compared to

no screening (£)a no screening

All 40 to 70 year olds (baseline) –165 0.155

Age (years)40–49 305 0.12150–59 –198 0.16960–69 –75 0.179

Risk factorsHypertensive patients –56 0.166Obese patients –165 0.155

Natural historyFaster HbA1c progression prior to clinical detection (shorter delay from –353 0.175

screen to clinical detection)

Treatment effectivenessLower CHD risk reduction achieved from HbA1c reduction 34 0.096Lower CVD risk reduction from statins 217 0.156More intensive HbA1c control (adding high-cost drug as third-line

combination therapy) 432 0.09333% lower reduction from screening in microvascular complications –24 0.138

Treatment and monitoring costsMonitoring cost £129 instead of £97 p.a. 0 0.155Higher statin cost – average 40/80 mg 440 0.155Insulin cost 1/3 lower (to see effect of fewer patients requiring insulin in

long-term through screening) –121 0.155

a Negative incremental costs denote cost saving.


75


TAB

LE 2

1Re

sults

of o

ne-w

ay s

ensit

ivity

ana

lyse

s fo

r ove

rall

cost

s an

d co

st-e

ffect

ivene

ss o

f scr

eeni

ng v

ersu

s no

scr

eeni

ng p

er 1

00,0

00 p

opul

atio

n

Scre

enin

g sc

enar

ioN

umbe

r To

tal

Num

ber

Tota

l inc

rem

enta

l To

tal

Mar

gina

l net

IC

ER (

scre

enin

g sc

reen

edsc

reen

ing

dete

cted

and

tr

eatm

ent

incr

emen

tal

bene

fit o

f scr

eeni

ng

vs n

o sc

reen

ing)

cost

(£)

trea

ted

cost

(£)

aQ

ALY

s(p

er s

cree

ned

pers

on)

(£)b

All

40–7

0-ye

ar o

lds

(bas

elin

e)16

,869

145,

898

283

–46,

697

4446

2,26

6

Age

(ye

ars)

40–4

9 6,

535

55,7

8160

18,2

597

1110

,216

50–5

96,

051

52,1

4188

–17,

528

1544

2,32

460

–69

4,28

337

,981

135

–10,

103

2410

61,

152

Ris

k fa

ctor

sH

yper

tens

ive

patie

nts

6,08

353

,708

176

–9,8

2429

891,

505

Obe

se p

atie

nts

4,00

435

,234

108

–17,

791

1779

1,04

6

Nat

ural

his

tory

Fast

er H

bA1c

prog

ress

ion

prio

r to

clin

ical

16

,869

145,

898

283

–99,

861

5056

929

dete

ctio

n (s

hort

er d

elay

from

scr

een

to

clin

ical

det

ectio

n)

Trea

tmen

t ef

fect

iven

ess

Low

er C

HD

risk

red

uctio

n ac

hiev

ed fr

om

23,6

1620

4,25

739

613

,656

3823

5,70

6H

bA1c

redu

ctio

nLo

wer

CVD

risk

red

uctio

n fr

om s

tatin

s16

,869

145,

898

283

61,2

8144

404,

682

Mor

e in

tens

ive

HbA

1cco

ntro

l (ad

ding

hig

h-co

st

16,8

6914

5,89

828

312

2,24

726

1510

,214

drug

as

third

-line

com

bina

tion

ther

apy)

33%

low

er r

educ

tion

from

scr

eeni

ng in

16

,869

145,

898

283

–6,8

3739

383,

563

mic

rova

scul

ar c

ompl

icat

ions

Scre

enin

g te

st c

osts

Hig

her

HbA

1cte

st c

ost

(£10

) 16

,869

230,

241

283

–46,

697

4441

4,19

3Tr

eatm

ent

and

mon

itorin

g co

sts

Mon

itorin

g co

st £

129

inst

ead

of £

97 p

.a.

16,8

6914

5,89

828

365

4443

3,33

4H

ighe

r st

atin

cos

t –

aver

age

40/8

0m

g16

,869

145,

898

283

124,

503

4436

6,17

7In

sulin

cos

t 1/

3 lo

wer

16,8

6914

5,89

828

3–3

4,30

744

452,

549

a N

egat

ive

incr

emen

tal c

osts

den

ote

cost

sav

ings

.b

Ass

umes

a c

ost-

effe

ctiv

enes

s ac

cept

abili

ty t

hres

hold

of £

20,0

00/Q

AL Y

.

Treatment costsThe additional use of statins is a major driver ofassociated costs and benefits associated with earlytreatment of patients identified through screening.The potential for a wider range of generic statinsover the next 5–10 years and associated reductionsin the costs of all statins could potentially furtherincrease the cost-effectiveness.

The results of the sensitivity analysis of the effectof adding a high-cost drug as third-linecombination therapy can be explained as follows.If screening is undertaken, long-term HbA1ccontrol is improved. This means that morepatients remain below the HbA1c level at whichinsulin would be added (8.3% was assumed), butmany could, if recommended by guidelines, betaking triple combination drugs to control HbA1cto target HbA1c levels applicable to patients usingoral agents (7.5% in the model).

Detailed analysis of base-case resultsThe following analysis relates solely to the costsand benefits of treatment at screen detectioncompared with treatment only after clinicaldetection.

The QALY gain in the base case can be brokendown as follows:

Survival benefitsDiscounted LYGs during the preclinical 0.013

periodBenefit of postclinical detection from 0.030

improved survival during the preclinical period

Benefit from being in ‘no CVD’ state versus 0.048‘CVD’ at end of preclinical period (improved long-term survival)

Other –0.008Subtotal 0.083

Benefit from reduced complicationsCHD 0.021Retinopathy 0.027PVD 0.017Other 0.007Subtotal 0.072Total 0.155

The benefit from reduction in first CHD events isshown in Figure 5.

Table 22 shows cost and benefits of treatingscreened detected diabetics compared with notreatment until clinical detection. The results arebased on multiple simulations of 10,000 patients.The incremental costs arising from earliertreatment (i.e. at screen detection) can be

summarised as follows (note: negative costs denotecost savings):

£Additional therapy 193Additional monitoring 522

Fewer complications:Macrovascular –457Retinopathy –120Dialysis –147Other Microvascular –156Total –165

There is a significant reduction in microvascularcomplications as a result of screening attributableto:

● The long-term improved HbA1c controlpredicted resulting from earlier detection andtreatment (see Figure 6) as observed in theUKPDS.158 Similarly in the DCCT, thescreening HbA1c value was a major predictor ofsubsequent HbA1c levels.235

● The high sensitivity of retinopathy risk todifferent levels of HbA1c, even at moderatelyelevated levels.

● The microvascular risk equations take accountof historical HbA1c levels by using ‘meanupdated’ HbA1c, that is, the mean of the annualvalues. (The HbA1c-based hazard ratios used byEastman and colleagues134 are based on the riskgradients from the DCCT using mean updatedHbA1c values235). This is important because thesubstantial difference in HbA1c levels during thepreclinical period (see Figure 6) has some effecton long-term risk. The elevated long-term riskdespite improved glycaemic control shown inthe DCCT as pre-DCCT exposure (determinedby screening HbA1c value and diabetesduration) was a major predictor of subsequentcomplications.235

Note that the complications submodels (i.e. CHD,stroke, etc.) run in parallel to reflect the realitythat these are ‘competing risks’ – this means thatthe level of risk in, for example, the CHD modelaffects survival, thereby affecting the number ofpatients remaining at risk. This can be seen in theresults – although CHD events are lower in thescreened cohort, stroke incidence is very similar(despite the benefit of statins during thepreclinical period) and other-cause mortality ishigher in the screened cohort.

The initial drop in the screened curve is due topatients starting diet and exercise and medicationafter diagnosis.


76


77


TAB

LE 2

2To

tal l

ifetim

e in

cide

nce

of e

vent

s an

d ye

ars

spen

t in

hea

lth s

tate

s: c

osts

and

QAL

Ys a

risin

g th

roug

h tr

eatm

ent

per p

atie

nt

Per

patie

nt

Ope

ning

Prev

alen

ce a

t en

dIn

cide

nces

of a

ll‘S

tate

yea

rs’

Tota

l cos

ts (£

)U

tility

(util

ity lo

ss):

prev

alen

ceev

ents

(non

-fata

l eve

nts)

QA

LYs

Scre

enin

gSc

reen

ing

No

Scre

enin

gN

oSc

reen

ing

No

Scre

enin

gN

oD

iffer

ence

Scre

enin

gN

oD

iffer

ence

scre

enin

gsc

reen

ing

scre

enin

gsc

reen

ing

Scre

enin

g

Trea

tmen

tU

tility

of t

hose

aliv

e10

,000

453

446

10.6

3210

.549

0.08

3(u

npen

alise

d fo

r com

plica

tion)

:O

f whi

ch ‘C

ompl

icatio

n-fre

e’–

––

Stat

esN

o re

tinop

athy

9,27

622

915

917

8,28

614

7,44

5N

on-p

rolif

erat

ive re

tinop

athy

641

9086

1,54

03,

122

18,2

1730

,047

0.00

00.

000

0.00

0Pr

olife

rativ

e re

tinop

athy

845

861

193

1,86

92,

807

29

–7–0

.002

–0.0

030.

001

Sign

ifica

nt m

acul

ar o

edem

a–

5785

1,05

52,

112

8,64

721

,614

5111

3–6

3–0

.008

–0.0

220.

014

Seve

re v

ision

loss

–73

108

237

601

2,40

45,

741

3182

–50

–0.0

07–0

.019

0.01

2

No

hist

ory

of c

oron

ary

hear

t10

,000

313

299

190,

394

184,

206

Non

-fata

l MI

–14

014

72,

032

2,38

819

,029

23,4

481,

648

1,91

5–2

67–0

.044

–0.0

560.

012

Dea

th (f

rom

MI o

nly)

3,50

03,

558

3,50

03,

558

0

No

cere

brov

ascu

lar h

istor

y10

,000

166

161

189,

635

187,

118

Stro

ke o

r tra

nsie

nt is

chae

mic

atta

ck–

287

285

2,75

22,

801

19,7

8820

,535

3,89

94,

089

–190

–0.1

85–0

.194

0.00

9D

eath

due

to st

roke

onl

y70

870

970

870

90

No

neph

ropa

thy

9,68

126

122

617

5,29

616

4,32

0M

icroa

lbum

uria

281

142

135

2,12

42,

630

29,4

1734

,586

–0.0

18–0

.022

0.00

4G

ross

pro

tein

uria

3846

7136

372

64,

470

8,28

1–0

.003

–0.0

050.

002

Dial

ysis

–1

532

6497

207

103

250

–147

–0.0

00–0

.001

0.00

0N

o ne

phro

path

y –

post

-tran

splan

t–

29

1856

143

260

5598

–43

–0.0

00–0

.001

0.00

0D

eath

due

to n

ephr

opat

hy o

nly

534

534

No

prev

ious

PVD

10,0

0036

635

220

2,82

519

7,52

1PV

D–

8590

687

915

6,23

89,

353

108

179

–70

–0.0

19–0

.031

0.01

2Am

puta

tion

(A1

or A

2)–

24

7914

535

978

044

86–4

3–0

.005

–0.0

100.

005

Dea

th –

mul

tiple

co-

mor

bidi

ties

1920

1920

Dea

th –

invo

lving

oth

er c

ause

s5,

317

5,23

45,

317

5,23

4

cont

inue

d


78 TAB

LE 2

2To

tal l

ifetim

e in

cide

nce

of e

vent

s an

d ye

ars

spen

t in

hea

lth s

tate

s: c

osts

and

QAL

Ys a

risin

g th

roug

h tr

eatm

ent

per p

atie

nt (

cont

’d)

Per

patie

nt

Ope

ning

Prev

alen

ce a

t en

dIn

cide

nces

of a

ll‘S

tate

yea

rs’

Tota

l cos

ts (£

)U

tility

(util

ity lo

ss):

prev

alen

ceev

ents

(non

-fata

l eve

nts)

QA

LYs

Scre

enin

gSc

reen

ing

No

Scre

enin

gN

oSc

reen

ing

No

Scre

enin

gN

oD

iffer

ence

Scre

enin

gN

oD

iffer

ence

scre

enin

gsc

reen

ing

scre

enin

gsc

reen

ing

Scre

enin

g

Man

agin

g di

abet

es in

clud

ing

ther

apy

Dru

gs:

OH

Ms

1st l

ine

770

772n

d lin

e11

145

663r

d lin

e95

8214

4th

line

022

7–2

27Al

l oth

er0

00

Antih

yper

tens

ive39

836

434

Lipi

d-re

lated

774

546

228

Tota

l dru

g co

sts

1,45

51,

263

193

Mon

itorin

g1,

348

821

526

Adve

rse

even

t cos

ts41

4331

632

0–4

Tota

ls20

,567

25,3

501,

047,

111

1,03

8,26

89,

060

9,22

5–1

6510

.339

10.1

840.

156


79


ScreeningNo screening

5000

4000

3000

2000

1000

01 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39

6000

Inci

denc

e

Year

FIGURE 5 Cumulative Incidence of first CHD events

5

5.5

6

6.5

7

7.5

8

8.5

9

ScreeningNo screening

0 3 6 9 12 15 18 21 24 27 30 33 36 39

HBA

1c (%

)

Year

FIGURE 6 HbA1c trends (from time at which decide to screen or otherwise)

The HbA1c curves temporarily cross because of thesignificant short-term fall in HbA1c on initiationtherapy at clinical detection in the no-screening cohort. This is only a short-termeffect, with HbA1c resuming its upward coursewithin a year or two. As clinical detection isoccurring at varying times, the curve of averageHbA1c is flattened out over the period duringwhich most unscreened patients are detected.HbA1c is higher than for the screened cohort inthe long term.

DiscussionSummary of findingsIn the baseline scenario, offering screening toeveryone aged 40–70 years, screening appears torepresent a cost-effective intervention. This is dueto the cost reductions and QALY gains fromreductions in complications, largely from fewercardiovascular events (risk reduced largely bystatin treatment during the preclinical period) andfewer individuals developing retinopathy andother microvascular complications (risk reduced bysustained improved glycaemic control).

The main economic trade-off is between:

1. the costs of the screening and diagnostic tests2. the net cost of statins, other therapy and

monitoring costs (such as retinopathyscreening) during the period before clinicaldetection minus the reduction in treatmentcosts of cardiovascular events and retinopathy

3. the higher QALYs that result from earlierintervention.

In addition to reduced retinopathy during thepreclinical period, the decision to screen or notcreates a modest but clinically significantdifference in HbA1c that leads to sustaineddifferential retinopathy incidence rates after thepoint of clinical detection.

Sensitivity analyses show that whereas a prevalencehas a relatively minor impact on overall cost-effectiveness, the cost-effectiveness of earliertreatment is closely dependent on the costs andeffectiveness of subsequent treatment ofcardiovascular risk.

The trade-off between size of CVD risk reductionand duration of benefit is seen in the results inTable 20. The most cost-effective age groups toscreen and treat earlier appear to be the 50–59-year and 60–69-year groups, these have a

sufficiently high CVD risk to obtain a significantbenefit from earlier intervention while havingsufficient remaining life expectancy to realisemedium- to long-term benefits. Although seemingto be still cost-effective, the 40–49-year age groupis much less so than the other age groups as theuncertainty in the various parameters could affectwhether this group, in particular, is cost-effectiveto screen and treat.

Comparison with previous modelsOverall, the importance of key parametersidentified by previous models has been confirmedand some additional areas of uncertainty,particularly around the preclinical course ofdiabetes, have been highlighted. As expected, thecost-effectiveness of screening is strongly related tothe costs of screening/additional treatment and theeffectiveness of the additional treatment. The costsand usage of statins become key parametersbecause of the model assumption that one impactof earlier diagnosis will mean that individuals athigh risk of CVD are started on statins earlier.This is consistent with the CDC model, whichpredicted that screening would be most cost-effective in those with pre-existing hypertensionwho would receive intensive, and cost-effective,earlier treatment to reduce their cardiovascularrisk.131

Our modelling of the natural history gives anestimated total preclinical duration of around11 years, with 6–7 years being the average delaybetween screen detection and symptomaticdetection, this is close to the assumption withinthe CDC model. The average delay is probablygreater for younger age groups (the estimate isabout 8 years), assuming some correlation betweenage and HbA1c at detection.

Previous models have also assumed linear HbA1cprogression, which could also result in anunderestimate of the potential benefits of earliertreatment.

The large reduction in cardiovascular eventspredicted is due to use of statins from diagnosis.Some other diabetes screening effectivenessevaluations were performed either before statinswere in widespread use or before the recent sharpfalls in the price (largely through expiry ofpatents), and therefore would include either lowerrisk reductions or greater costs.

The population in the present model excludesthose with existing CHD at baseline and hasreflected the move towards tighter blood pressure


80

control in the general population. This means thatlong-term survival in the modelled population ishigher. As a result, differences in survival arisingthrough screening (up to the point at whichclinical detection is made) result in a greater long-term benefit than would otherwise have beenobtained.

Our model incorporates the relationship betweenHbA1c at diagnosis and long-term HbA1c control.Improved long-term HbA1c control arisingthrough screening is translated into long-termreductions in microvascular complications.

Other models have also not considered thepotential benefit of reductions in long-term HbA1clevels on CHD incidence. Although our base caseassumed a 14% CHD risk reduction per 1% fall inmean updated HbA1c as per UKPDS 35,221

sensitivity analysis indicates that screening wouldstill be highly cost-effective if this were only 5%.

Our test cost is low compared with other models.Treatment costs are largely UK based, accountingfor some of the difference with other models.

A utility of 0.77 was assigned to a person withdiabetes but no complications, whereas othermodels tend to use 1.0. The latter does not reflecteither the reduced age-related quality of life of thegeneral population or the diabetes-specific loss ofquality of life arising from taking medication,monitoring and so on. This explains why somemodels report a utility gain through screening ofsimilar order to our model, despite much lowerreductions in complications.

Strengths and weaknesses of analysisDue to the timescale involved, we have notproduced a state of the art model or made use ofmore sophisticated modelling methods toundertake evidence synthesis to structure orpopulate the economic model. It has beennecessary to make various assumptions andexploration of uncertainty has been limited tosimple sensitivity analyses. We have neverthelessproduced a screening model linked to the existingfully developed treatment model to investigate theorder of magnitude of effects given differentscenarios for diabetes screening policies.

The model results are dependent on the choice ofmodel parameters and, although more complexthan some previous models, there are a number ofareas where complete data are unavailable. Inparticular, the natural history of diabetes before

clinical diagnosis and the clinical characteristics ofcases detected by screening remain uncertain,despite review of the best available evidence. Forthis reason, a number of sources have been usedor different methods triangulated to estimate, forexample, the duration of disease betweenscreening and clinical detection and the HbA1ctrajectory.

Our modelling suggests that screening is cost-effective across all populations between 40 and70 years of age and across most one-way sensitivityanalyses. However, it cannot be concluded withoutfurther analysis that this would be the case forsensitivity analyses around every population. Thesame applies if some of the sensitivity assumptionswere combined (e.g. higher monitoring costs andlower CHD reduction from HbA1c lowering).Equally, there is some uncertainty about therelative cost-effectiveness of the alternativepopulations.

A probabilistic sensitivity analysis would berequired to determine adequately the optimumscreening policy (see the subsections ‘Optimumscreening age’ and ‘Rescreening interval’, p. 82),taking account of the impact of uncertainty andexploring the most significant parameters.Probabilistic sensitivity analysis would effectivelybe a more comprehensive multivariate extensionof the one-way sensitivity analyses that have beenconducted so far, and would quantify thelikelihood of a screening policy being costeffective. Methods have recently been developedthat reduce the heavy computational burden ofprobabilistic sensitivity analysis for patient-levelmodels based on the algebra of analysis ofvariance and Bayesian statistics,236 making arigorous uncertainty analysis feasible in reasonabletime. Our simulations have removed most ‘first-order variability’ from the results but anyprobabilitic sensitivity analysis should be largeenough to ensure that this is effectivelyeliminated.

Potential for further modellingFurther development of existing modelparametersAlthough this model has used the best availablepublished evidence on the natural history ofdiabetes prior to clinical diagnosis, there ispotential to develop more sophisticated models. A gradual upsloping HbA1c curve during thepreclinical period was assumed, but there are somesuggestions that there could be a slow linear phasefollowed by a more rapid linear phase (possiblywhere beta-cell failure becomes more significant


81


than insulin resistance). It would be useful toincorporate more primary data on aspects of thenatural history of the development and earlyprogression of diabetes. Potential sources of datainclude the Whitehall Study,237 the UKPDS,238 theADDITION trial95 and the DARTS TaysideRegional Diabetes Network.239

Screening for both prevention/delay and earlierdetection of diabetesThis model addresses screening and earliertreatment for diabetes. However, any screeningprogramme for diabetes will also identifyindividuals who would benefit from interventionsto delay or prevent the development of diabetes.This issue has been considered separately, but ifboth prevention and early detection are deemedto be cost-effective, it would be useful to model theoverall impact of screening on both preventionand early treatment.

Optimum screening ageOur modelling suggests that screening is cost-effective for all age groups between 40 and70 years. However, it may be that there is relativelylittle incremental benefit from screening patientsin their 40s compared with waiting until they arearound 50 years old (but any future modellingshould also consider possible interactions betweenage and rate of diabetes progression as discussedbelow in the subsection ‘Patient variability in thepreclinical phase’).

Rescreening intervalRescreening would not arise as an issue if a singlescreening intervention to identify prevalence casesof undiagnosed diabetes had not beendemonstrated to be cost-effective from a certainage. However, if a prevalence-round screeningstrategy is cost-effective, then clearly whetherrescreening should be done, and if so at what timeinterval, should also be modelled. In practice,once an individual has been given a screening test,they will expect to be rescreened unless they canassume they are no longer at a significant risk ofthe condition screened for. This is unlikely,particularly if they have already been offered ascreening blood test on the basis of the presenceof risk factors for diabetes.

Cardiovascular risk reductionIf screening for diabetes is considered in thecontext of screening for CVD, then it would belogical to quantify the additional value ofidentifying both individuals at risk of diabetesand individuals with diabetes within acardiovascular screening/risk reduction

programme. The model would still estimate theoverall costs and benefits of screening, but withthe objective of cardiovascular risk reduction. Suchan approach might reduce the potential for thosewithout diabetes but with other cardiovascular riskfactors to be ‘falsely reassured’ by screening. Ourmodel could be refined to restrict statin use to aperson diagnosed with diabetes once their CHDrisk exceeds a certain threshold.

Secondary CHD riskThe modelling has shown that the reduction,through earlier treatment, in the prevalence ofCHD (non-fatal MI), at the time at which clinicaldetection would otherwise have occurred, is asignificant driver of cost-effectiveness. It would bepossible to refine the secondary event rates so thatthey are dependent on risk factors.

Relationship between nephropathy and CHDCHD risk increases significantly with progressionfrom overt nephropathy to ESRD. The very highcosts of dialysis and renal transplantation arereduced to some extent by screening (throughbetter long-term glycaemic control), but it would beuseful to model the risk relationship more precisely.This is especially the case given the relatively highlong-term survival of our modelled population(having no symptomatic CHD at baseline and tightblood pressure control), because more patientscould live long enough to develop ESRD.

Patient variability in the preclinical phaseIt would be useful to consider the between-patientvariability in HbA1c changes – this may beimportant because slow progressors are more likelyto be detected through screening as they spendlonger in the preclinical phase. Furtherinformation on this variability might be obtainedusing data from trials treating patients with IGTand from UKPDS data.

The literature often refers to younger patientsexperiencing a faster beta-cell decline than olderpatients – it would be worth investigating thisfurther to see if this interaction is significant and,if so, to quantify the impact on cost-effectiveness.

Newer agentsAny future work should include new therapeuticdevelopments; for example, if glucagon-likepeptide (GLP) analogues or dipeptidyl peptidaseIV (DPPIV) inhibitors can stabilise or reverse beta-cell decline, the benefits of early detectionmay change. Inhaled insulin may change both thecosts and effectiveness of glycaemic control(through improved adherence).


82

ConclusionsThe modelling undertaken has led to thefollowing conclusions:

1. Screening for diabetes appears to be cost-effective for the 40–70-year age band subject tolimitations of the evidence available and withinthe one-way (parameter and treatment protocol)sensitivity analyses that we have carried out.

2. Screening for diabetes appears to be most cost-effective for the 50–59-year and 60–69-year agebands and less so for the 40–49-year age group.

3. Screening appears to be even more costeffective for the hypertensive and obesesubgroups.

4. Further analysis should explore whetherscreening is cost-effective under somecombined parameters and treatment sensitivityassumptions, particularly for the 40–49-yearage band.

5. Assumptions about degree of control andtreatment protocols after clinical detection areas important, if not more so, than assumptionsrelating to the screening programme itself indetermining the cost-effectiveness of screening.

6. Although the results suggest that there is acost-effective screening programme, theoptimum age and other criteria are notobtainable from these results and further workis needed to determine this. The optimumpolicy also needs to take account of theoptimum rescreening interval and the benefitsof treating patients detected with IFG or IGT.

7. Further uncertainty analysis (includingprobabilistic sensitivity analysis) is required toexplore the robustness of our preliminaryconclusions and to prioritise any areas forfurther research or modelling.

8. There appears to be limited quantifiedevidence in the public domain describing thenatural history of diabetes before clinicaldetection. Further work to understand thiswould increase confidence in the estimatedtime between potential screen detection andclinical detection.

9. Further work to refine central estimates forsome parameters (such as CHD risk reductionresulting from HbA1c reductions) would beuseful, possibly using expert elicitation.

Further work is also needed to establish the overallcost-effectiveness if diabetes screening isincorporated into a comprehensive risk reductionstrategy that includes assessment andmanagement of overall risk of CVD, includingassessment and management of IGT. However, itseems plausible that including assessment of

glycaemia as an element of a cardiovascular riskassessment and risk reduction intervention wouldprove cost effective:

1. given the additional benefits (from reduction inlong-term complications) from earlierintervention in individuals at significant risk ofCVD due to IGT and other associated riskfactors and

2. assuming the availability of cost-effectiveinterventions to offer to those identified asbeing at risk.

Background to methodsModelling of treatment and outcomespost-diagnosisWe have developed a patient-level Markovsimulation model following the lifetime clinicalpathways of T2DM patients from diagnosis.Patients’ attributes include age, sex, ethnicity,smoking and baseline levels of HbA1c, cholesteroland blood pressure.

Each period, patients have their metabolicvariables monitored (which can be varied, butusually 6-monthly or yearly) and if they exceedthresholds on blood pressure or glucose control,patients’ therapies are re-examined and switchedto the next therapy in a defined sequence; hencethe model can compare alternative therapysequence strategies in addition to alternativetighter or looser control and monitoring ofmetabolic variables.

The risks of fatal and non-fatal events associatedwith the key diabetes co-morbidities are examinedeach period using published evidence on risk. Theco-morbidity modules are heart disease, stroke,diabetic retinopathy, nephropathy and neuropathy.These work in parallel and independently, so thatany individual patient may have a number of co-morbidities at the same time.

Patients’ quality of life is measured based onpublished evidence of utility values of all of thehealth states concerned. This enables a final resultof overall survival and QALYs to be calculated.Similarly, costs of drugs, monitoring and relatedco-morbid clinical events are included andtracked, enabling a lifetime cost to be quantified.

Risk factors and treatment for screeningdetected cases in treatment modelModelling the screening arm involves running themodel in the usual way, with characteristics of


83


true-positive patients at screening forming thebaseline characteristics at diagnosis in thetreatment model.

● Cholesterol: following the results of the HeartProtection Study showing that statin therapy isbeneficial in diabetics regardless of baselinecholesterol levels,79 the model assumes that allpatients with diabetes will be prescribed statinson diagnosis, and benefit from a 30%cardiovascular risk reduction (Waugh N,University of Aberdeen; personalcommunication).

● BG: it is assumed that patients are initiallytreated with diet and exercise followed bymonotherapy, then combination oral agentsfollowed by the addition of insulin. Patientsprogress to the next line of therapy if theirHbA1c exceeds the specified threshold in thetreatment model, namely 7.5%. The annualHbA1c changes are based around (a) anequation derived from UKPDS results58 to linkHbA1c falls during 3 months’ diet and exercisetreatment to the level at diagnosis and (b) theequation provided in UKPDS 68,158 whichrelates annual changes to HbA1c levels atrandomisation and levels in the previous year.

● Smoking: smoking status at baseline is includedin the risk calculations. The UKPDS 68model158 gives a hazard ratio of 1.4 for currentsmokers versus non-smokers. We have noinformation on which to incorporate anychanges in smoking behaviour resulting from adiagnosis of diabetes. Adoption of interventionsto increase smoking cessation at diagnosis ofdiabetes would enhance the clinical benefits ofscreening but entail additional costs.

● Blood pressure: the same threshold is assumedfor starting antihypertensive therapy as forclinically detected patients, i.e. 135 mmHg.

● Aspirin efficacy: aspirin is assumed to be taken where the 10-year CHD risk exceeds 15%. The associated risk reduction achieved in primary CHD prevention in patients with

diabetes is uncertain, but analyses suggest some benefit.240 The present model uses amodest default 5% CHD risk reduction as thereis no clear evidence that the benefit observed innon-diabetic people is mirrored in diabeticpatients. If the risk is significantly greater, thiscould increase the effectiveness of screeningbecause it would provide further benefit during the preclinical period at negligibleadditional cost.

Use of relative risk reductionsWhere multiple RRs are applied to reflect theeffect of statins, HbA1c reductions, aspirin therapyand so on, these have been applied on amultiplicative rather than additive basis.

Calculation of diabetes prevalence inspecified populationThe total prevalence of T2DM for the populationspecified by the age, gender and ethnicityvariables is obtained from a link to the PBSPrevalence Model3 (Yorkshire and Humber PublicHealth Observatory). These calculations have beenadjusted to calculate only T2DM prevalence,excluding the type 1 diabetes population.

There are options to filter patients furtheraccording to hypertensive status and BMI levels.Adjustments to the prevalence from the PBSmodel are made using the following RRs:

● RR for diabetes if hypertensive = 1.67, which isestimated from the RRs in the reports in theKORA survey206 and the Inter99 Study.241

● RR for diabetes per unit BMI = 1.25 (based oninternal review242).Where both hypertensive and BMI criteria arespecified, these two RRs have been multipliedtogether and then multiplied by a factor of 0.67to make a rough adjustment for the correlationbetween hypertension and obesity, which wasestimated from the univariate and multivariaterelative risk reported in the Inter99 study.241


84

The aims of screeningAs discussed earlier, the aims of screening couldinclude:

● Detection of undiagnosed diabetes, with a viewto starting treatment aimed at lowering BGlevels and reducing the risk of the microvascularcomplications of diabetes. Treatment wouldstart with diet and exercise, with hypoglycaemicdrugs added if necessary.

● Detection of people at risk of developingdiabetes, with a view to starting measures toreduce progression to diabetes. Treatmentwould be with diet and exercise.

● Identification of a group, comprising both of theabove, who are at increased risk of macrovasculardisease, and then treatment to reducesubsequent CVD. Treatment would includereduction of BG as above, but control ofhypertension and reduction of hyperlipidaemiawould probably be more important. A furtherbenefit would arise from treatment ofhypertension with ACEIs and ARBs, which wouldprevent progression to diabetes in some people.

● An additional reason for screening might bethat even in those individuals already identifiedas at high cardiovascular risk, knowledge of theadditional risk due to diabetes/hyperglycaemiamay motivate individuals to act to reduce riskby increasing the perceived benefit (versusinconvenience/personal cost) of lifestyle changeor taking medication – that is, the value of theinformation to the individual rather than theclinician.

Diabetes and IGT confer an increased risk ofmacrovascular disease, but the increase with IFG ismuch less. Hence if the main aim is to reducefuture heart disease, detecting IFG is lessimportant. This has implications for the screeningtest. It makes FPG less attractive. The remainingoptions are then the OGTT (reduced to a 2-hourchallenge test after 75 g of glucose) or HbA1c.

In favour of HbA1c test is that it is simpler toperform than the 2-hour OGTT and is morereproducible, and that we have data from theNorfolk Study showing that it is associated withCVD across a wide spectrum.

As Goyder and Irwig10 pointed out, people withdiabetes detected by screening are at high risk ofmacrovascular disease but at comparatively lowrisk of microvascular disease. In the UKPDS, thosewhose diabetes was asymptomatic had lower FPGand fewer microvascular complications atdiagnosis and were at lower risk in the years whichfollowed (admittedly, not all had diabetes – theentry criterion was FPG over 6 mmol/l).

Could it be argued that in people known to be athigh risk of heart disease, because of factors suchas hypertension, high cholesterol levels andoverweight, that testing BG level would providelittle additional benefit, since they should alreadybe treated with antihypertensive agents, diet andstatins? (Diet because it helps reduce bothcholesterol and blood pressure.) Many of thisgroup would be able to control theirhyperglycaemia on diet alone.

Arguments against such a position is that they areat some risk of microvascular disease at diagnosis,that the natural history of the disease is that beta-cell failure appears progressive (UKPDS 16),243

that most require additional antihyperglycaemictherapy over time (UKPDS 16)243 and that themarginal cost of adding an FPG or an HbA1c testis low when fasting blood is being taken for lipidestimation anyway.

Screening intervalAssuming the first screen is at age 45 years, whenshould people be rescreened? Rather than have afixed interval for all, the second screen coulddepend on the result of the first. If an HbA1c levelof 6% is taken as the threshold for intervention (atthat level, with diet, exercise and monitoring overtime), and if it is assumed that HbA1c rises by 0.2%per year (in those in whom it does rise), then theinterval in years could be

6.0 – HbA1c at age 45/0.2

Those whose HbA1c had not risen at second screencould be reassured and not recalled. The recallinterval could be based on rate of rise. If the ageexpected to reach an HbA1c of 6% or more was


85


Chapter 6

Discussion

over 70 years, then perhaps they could also bedischarged (on the grounds that screening fordiabetes at age over 70 years is not worthwhile).102

However, the above arguments make anassumption that rise in HbA1c is linear. This is notknown, and given that the rate of rise may varyamongst patients, and that it may not be linear, itwould probably be safer to recall at a fixed intervalfor the first two rounds of screening. The resultingdata should then provide the evidence base forfuture recall policy.

At present there are insufficient data on which tobase any screening interval. A pragmatic approachmight be to screen by risk factors at age 45 years,screen by BG in those who were risk-factor positiveand then re-screen a random sample of BG-negativepeople at, say, 5 years as part of a research project.

Research needs1. Research is needed into ways of reducing the

prevalence of insulin resistance. More researchis also needed into which forms and amounts ofexercise are effective in reducing hyperglycaemiaand the effect on cardiovascular risk. Why dosome people become insulin resistant whenoverweight whereas others do not? Is exercisethe key factor?

2. The main need is to find ways of achievingcompliance with diet and other lifestylemeasures in order to reduce overweight andobesity. How can public health campaigns bemade more effective?

3. In the UKPDS, HbA1c rose in all treatmentgroups. However, the groups included thosewho gained and those who lost weight.Research is needed into whether the rise overtime is seen also in those who reduce weightand take exercise. Would return to normalweight, or close to it, coupled with regularexercise to reduce insulin resistance, abort thebeta-cell failure over time?

4. If screening is to be introduced, should it beone-off; if not, what should the interval be?Data are needed on the rise in glucose overtime, and whether it is linear.

5. If blood is being taken for glucosemeasurement, what other investigations wouldbe cost-effective at the same time? A lipidprofile for HDL, LDL and triglycerides?

6. Where should screening stop? Should it be forundiagnosed diabetes, or that plus IGT, or forthe wider metabolic syndrome? What cut-offshould be used, to strike the optimum balanceamongst sensitivity, specificity and what theNHS can cope with?

7. What is the optimum balance between the clinical approach of identifying andtreating individuals and the public healthapproach of intervention at population level?

Some of the public health questions will beaddressed by the Public Health appraisal functionnow taken over by NICE. The appraisals willinevitably pose another broad question:

8. Which public health or health promotioninterventions are not cost-effective and shouldbe stopped? These may include current ad hocscreening activities that could be stopped ortargeted more efficiently.

Research currently under way includes:

● The ADDITION study – do screening and earlydiagnosis and intervention reducecardiovascular risk? (Griffin S and Khunti K:personal communication, 2005).

● The Whitehall Study, which should yieldinformation on the potential benefits ofscreening (Brunner E: personal communication,2005).

● The Diabetes Heart Disease and StrokePrevention Study of screening in a routinesetting, including in less affluent areas, andthose with a high proportion of ethnicminorities, is already showing differences inuptake of screening, and in percentages ofpeople with new diabetes (Goyder E: personalcommunication, July 2005).

● The 1958 Birth Cohort Study, which isexamining risk scores and their marginalbenefits over simply using BMI, and alsoregional variations in prevalence (Power C:personal communication, 2005).

Does screening for diabetes andIGT meet the NSC criteria?The UK NSC criteria for evaluating screeningprogrammes were adapted from the WHO criteria published in 1968.244 The criteria arepublished by the NSC on their website(http://www.nsc.nhs.uk/pdfs/criteria.pdf).

This section applies the criteria to screening fordiabetes and lesser degrees of glucose intoleranceand summarises the evidence presented in theprevious chapters. Our view on the extent to whichthe criteria are satisfied is appended in bold at theend of the text on each criterion.

Discussion

86


87


The conditionCriterion 1. The condition should be animportant public health problemWhether the problem is defined as diabetes, alldegrees of glucose intolerance, the metabolicsyndrome or even wider as overweight and obesity,it is an important public health problem. The levelof individual risk will vary according to the scopeof the definition – in some cases at the margin, theindividual health problem, in terms of excess risk,will be slight. However, because of the very largenumbers involved, even slight increases in riskamount to a major public health problem. The‘Rose principle’ applies here – there is more to begained from a leftward shift in the population risk,even if the individual benefits are small, comparedwith only identifying those at the high-risk rightend of the population curve.245 However, bothapproaches can be applied.

Criterion met.

Criterion 2. The epidemiology and natural historyof the condition, including development fromlatent to declared disease, should be adequatelyunderstood and there should be a detectable riskfactor, or disease marker, and a latent period orearly symptomatic stageThis was discussed at length in Chapter 1. Thereare several uncertainties about natural history:● Is the progression to diabetes linear

throughout, or is there a slow initial phasefollowed by a faster decline in beta-cellfunction? This would affect screening interval.

● Why do some people with IGT progress todiabetes, whereas about half return tonormality? What do the half do that ensuresimprovement?

● Is the progression always IGT, then IFG, thendiabetes? Or are there two pathways?

● In people diagnosed with T2DM, is progressioninevitable as shown in UKPDS overall – or dothose who lose weight and take exercise havemuch slower progression, or indeed regressionto normal, as shown in the Malmo trial?

However, answers to all of these questions are notneeded before screening can start. Enough isknown about natural history on a population basis(for example, a sizeable minority of those withIGT will progress to diabetes; many withundiagnosed diabetes will be developingretinopathy; many people with diabetes will haveadvancing but asymptomatic heart disease, thefirst manifestation of which may be a fatal MI).

Criterion met.

Criterion 3. All the cost-effective primaryprevention interventions should have beenimplemented as far as practicablePrimary prevention of T2DM involves severalaspects:

● Public health campaigns to encourage the publicto avoid overweight and take exercise. A reviewof the evidence on interventions at populationlevel is outwith the scope of this review.

● Intervention on a personal basis in those whocan be identified as being at higher risk ofT2DM, such as those with IGT as alreadymentioned. There is good trial evidence that thiscan be effective, and our review of the economicevidence shows that intervention is cost-effective,but that refers to prevention of diabetes alone.

The main problem with public health measures isnot cost-effectiveness (most campaigns would beinexpensive because they are delivered to largenumbers) but clinical effectiveness. There havebeen many campaigns at varying levels, from massmedia campaigns about healthy eating to localinitiatives at health authority level such as tryingto encourage people to ‘walk about a bit’. Yet theprevalence of overweight and obesity continues torise. The research need arising from this has beenmentioned above. We know that lifestyle measurescan work; if they do they would be cost-effective(measures such as diet and exercise being largelycost free to the NHS).

Criterion met? For diabetes, yes. For IGT andIFG, unproven. Prevention certainly possible intheory.

Criterion 4. If the carriers of a mutation areidentified as a result of screening, the naturalhistory of people with this status should beunderstood, including the psychological implicationsNot applicable.

The testCriterion 5. There should be a simple, safe,precise and validated screening testAs discussed in Chapter 3, screening would be twostage – assessment of risk by ‘questionnaire’ butwith most of the necessary data already held ongeneral practice computer systems; and then BGmeasurement in those at higher risk. The decisionon the risk level at which to intervene would beinfluenced by the capacity of the service to copewith those newly diagnosed.

There have been various ways of achieving thefirst stage, and the debate is not about validation

but about the marginal benefits of adding orremoving additional items.

For the second stage, there are three acceptabletests – FPG, OGTT and HbA1c. All are safe,precise and validated. However, each has itsadvantages and disadvantages. OGGT isinconvenient, poorly reproducible and has to berepeated, and would probably not qualify assimple. FPG requires people to fast, compliancemay be imperfect and those with isolated IGTwould be missed. HbA1c is more expensive but canbe done at any time of day and reflects glycaemiaover a period of several months.

But the debate is about which test is best – all areacceptable.

Criterion met.

Criterion 6. The distribution of test values in thetarget population should be known and a suitablecut-off level defined and agreedThe distribution of test values is known frompopulation-based research studies. The cut-off leveldepends on whether screening is for diabetes,diabetes plus IGT or wider. The case for screeningfor metabolic syndrome is unproven pendingfurther research. Hence it could be assumed thatscreening is only for diabetes and IGT, using HbA1cor the 2-hour PG. The cut-off would be for debate,partly influenced by what it was felt the NHS couldcope with, but might be based more towards thelevels shown to increase risk in the EPIC Norfolk study, rather than the HbA1c of 7%suggested on the basis of being the level at whichpharmacological treatment might be necessary.

Criterion met.

Criterion 7. The test should be acceptable to thepopulationThe evidence is that the test is acceptable to many,but not all. For example, unpublished data fromthe Leicester ADDITION study shows a pooruptake (Griffin S and Khunti K, ADDITION StudyGroup, Cambridge and Leicester: personalcommunication, July 2005), although that was inthe context of a trial and may have been affectedby an information overload imposed by ethicscommittees. There seems to be a wish on behalf ofDiabetes UK, the key consumer organisation, forscreening to be provided.

In any case, screening would be offered, and thosewho did not wish it would not need to accept.

Criterion met.

Criterion 8. There should be an agreed policy onthe further diagnostic investigation of individualswith a positive test result and on the choicesavailable to those individualsThere is agreement on further investigation, with the key rule being that one abnormal glucose level should always be confirmed. Inasymptomatic people, diabetes should not bediagnosed on the basis of one result. If the initial screening test was HbA1c, the follow-up test could be an FPG, for simplicity. If that wasnormal, a 2-hour post-load PG could be sought.

Criterion met.

Criterion 9. If the test is for mutations, thecriteria used to select the subset of mutations to be covered by screening, if all possiblemutations are not being tested, should be clearlyset outNot applicable.

The treatmentCriterion 10. There should be an effectivetreatment or intervention for patients identifiedthrough early detection, with evidence of earlytreatment leading to better outcomes than latetreatmentThe question of which condition is to be screenedfor is less important here, because initial treatmentfor T2DM, IGT and the metabolic syndromewould be the same – diet and exercise. Statinsmight be added.

There is a good evidence base for treatment ofdiabetes itself. Several trials have shown that theprogression to diabetes from IGT can be reducedby lifestyle changes or drug treatment (seeAppendix 3). Lifestyle changes can also improvecontrol of diabetes. A meta-analysis of the effectsof exercise246 showed that it reduced HbA1cby a clinically significant amount of 0.66% (whichis not far off the difference between the intensiveand control groups in UKPDS). Interestingly, the benefit of exercise was achieved without weight loss.

Other aspects of management of diabetes includescreening for complications; the case for eyescreening is beyond doubt and has been addressedby previous NSC policy considerations.

The key issue is whether treatment earlier in thedisease process is better than later treatment oncethe condition has become symptomatic. The aimof treatment is to prevent the complications of

Discussion

88

diabetes. Some people have retinopathy atdiagnosis (UKPDS) but rarely to a sight-threatening stage, so earlier treatment is unlikelyto prevent more visual loss than later. Similarly,nephropathy takes time to develop.

However, vascular disease is the main risk in thegroup which would be affected by screening. IHD is often asymptomatic until an MI occurs,and the first MI may be fatal. The risk can bereduced by about one-third by statin therapy.Hence earlier treatment would have clearadvantages. Many of the people in this metabolicgroup will have other risk factors, but thediagnosis of diabetes or IGT may result in themcrossing the NICE 2% per year threshold, andbeing treated with a statin.

In the absence of RCTs of screening versus noscreening, we have to try to obtain evidence onwhether earlier treatment gives better outcomesfrom other studies. Data from the UKPDS havebeen used for this purpose58 by examiningoutcomes according to FPG at diagnosis. Those inthe lowest tertile of initial FPG were more likely tobe asymptomatic and found by screening (such asinsurance or employment medical examinations).They had less retinopathy at diagnosis and hadfewer complications during the study. However,this may just mean that they were at an earlierstage of the disease – lead time bias. It could alsobe because they had a milder form of the disease,but their rise in PG over time was similar to thatin the other groups, so that seems unlikely.Nevertheless, they were diagnosed with fewercomplications, and there is scope to reduce thefuture development of complications byintervention.

Criterion met.

Criterion 11. There should be agreed evidence-based policies covering which individuals shouldbe offered treatment and the appropriatetreatment to be offeredInitial treatment would be diet and exercise, andmonitoring for progression to levels at whichpharmacological treatments would be given. Forstatins, this might be at a risk level of 2% or moreper annum, aiming at a total cholesterol of5 mmol/l or less. For HbA1c, the aim might be tokeep it under 7% (ADA) or 7.5% (NICE), withmetformin the first-line drug. The meglitinideanalogues might be considered if the problem ismainly post-prandial hyperglycaemia.

Criterion met.

Criterion 12. Clinical management of thecondition and patient outcomes should beoptimised in all healthcare providers prior toparticipation in a screening programmeMany patients with T2DM are not well controlled, and many may not be on, for example,statin therapy. Only 43% of people with diabetes in Scotland for whom an HbA1c result wasavailable for the Scottish Diabetes Survey, had an HbA1c under 7.5%, the NICE guidelines target. These data also include peoplewith type 1 diabetes, but are numericallydominated by T2DM. However, guidelines formanagement of diabetes have been issued byNICE, as has guidance on statin use.86 It maytake some time for these to work through thesystem.

Hence there is more to be done to optimise careand outcomes thereof for existing patients, and itcould be argued that they should take first priority.Adding an additional load of new patients wouldmake it more difficult to achieve this.

Criterion not met.

The screening programmeCriterion 13. There should be evidence from high-quality randomised controlled trials that thescreening programme is effective in reducingmortality or morbidityAs yet, no RCTs of screening for diabetes havebeen reported. However, one is under way. TheADDITION Study is an RCT of systematicscreening and targeted cardiovascular riskreduction in primary care, being carried out inCambridge, Denmark and the Netherlands.95

However, the screening phase was not due to be completed until late 2006, after which there will be a 5-year follow-up period. Hence results will not be available for the next review of NSC policy.

Criterion not met.

Criterion 14. There should be evidence that thecomplete screening programme (test, diagnosticprocedures, treatment/intervention) is clinically,socially and ethically acceptable to healthprofessionals and the publicScreening would only be offered, not imposed,and details of follow-up investigations andtreatments would be included with the invitation.Hence only those who found the completeprogramme acceptable would attend for screening.

Criterion met.


89


Criterion 15. The benefit from the screeningprogramme should outweigh the physical andpsychological harm (caused by the test,diagnostic procedures and treatment)Any screening programme has the potential toconvert a healthy person into ‘a patient’. Beinglabelled as diabetic could cause anxiety, in additionto possibly having adverse effects on some forms ofemployment (although that is more of a problemwith insulin-treated diabetes). The anxiety woulddepend on the perceived seriousness of diabetes tothe newly diagnosed person.

In the Hoorn study,106 the psychological impact ofscreening was assessed by qualitative methods (semi-structured interviews) in 20 patients newlydiagnosed (mean age 62 years) and 20 at increasedrisk of diabetes but who tested negative. The mainlimitation of the study was the small numbersinvolved. The results indicate that the fact of thediagnosis caused little alarm. This may have been, asthe authors report, because only one of the newdiabetic patients regarded the disease as serious,and because about half felt that they had control, inthe sense that they believed they could take effectiveaction to deal with it, such as dieting. Their meanBMI was 28.6 kg/m2, so their belief seems justified.

More harm might ensue in the group who were atrisk but who tested negative. Adriaanse andcolleagues106 note that this group may have beenso reassured by the result that they saw no reasonto adjust their lifestyles.

Criterion met? Uncertain.

Criterion 16. The opportunity cost of thescreening programme (including testing, diagnosisand treatment, administration, training andquality assurance) should be economicallybalanced in relation to expenditure on medicalcare as a whole (i.e. value for money)The economic analysis suggests that screening forIGT and prevention of diabetes would be cost-effective, although not all costs have beenidentified. If screening was, as assumed, based inprimary care, there would presumably be a formof target or points-based systems to remuneratepractices. The cost of the programme woulddepend on the number of people invited forscreening and the response rate. Screening mightonly be offered to those with BMI 30 kg/m2 orover, although any such cut-off would be anarbitrary line in a continuum of risk. Themodelling exercise in Chapter 5 shows thatscreening would be more cost-effective in theobese or the hypertensive, but not dramatically so.

Criterion met.

Criterion 17. There should be a plan formanaging and monitoring the screeningprogramme and an agreed set of qualityassurance standardsNot applicable until a decision is made inprinciple to provide screening. A qualityassessment system for HbA1c exists.

Not applicable.

Criterion 18. Adequate staffing and facilities fortesting, diagnosis, treatment and programmemanagement should be available prior to thecommencement of the screening programmePrimary care services and diabetes clinics arealready under pressure. It is unlikely that theycould cope with a large extra load at present.Screening could be brought in slowly, and a highthreshold for positivity used, at least in the earlystages.

Criterion not met.

Criterion 19. All other options for managing thecondition should have been considered (e.g.improving treatment, providing other services), toensure that no more cost-effective interventioncould be introduced or current interventionsincreased within the resources availableThe main option other than screening would be apublic health programme to encourage people tokeep weight down and take exercise. The rise inoverweight and obesity implies that past andpresent public health campaigns have failed. Is itworth trying harder, or trying new options such as‘exercise prescriptions’? Is there a danger of‘medicalising’ unhealthy life style and reducing theemphasis on personal responsibility for health?

It was noted that the best way to identify peoplefor screening would be to identify them fromgeneral practice records. Instead of inviting themfor screening, the data could be used for targetedhealth promotion interventions such as for weightloss or exercise.

Criterion met? Uncertain.

20. Evidence-based information, explaining theconsequences of testing, investigation andtreatment, should be made available to potentialparticipants to assist them in making an informedchoice.The evidence exists and would be supplied whenpeople were being invited to attend.

Criterion met.

Discussion

90

Criterion 21. Public pressure for widening theeligibility criteria for reducing the screeninginterval, and for increasing the sensitivity of thetesting process, should be anticipated. Decisionsabout these parameters should be scientificallyjustifiable to the publicThe problem here might be that the decisions ofscreening policy, such as whom to invite, woulddepend partly on cost-effectiveness aspects, whichmay not be understood by the public.

Criterion met? Uncertain

Criterion 22. If screening is for a mutation theprogramme should be acceptable to peopleidentified as carriers and to other familymembersNot applicable.

SummaryOf the 22 criteria:

● 12 are met.● Three are not met.

● In three there is uncertainty.● Four are not applicable.

ConclusionsTargeted screening for T2DM meets most but notall of the NSC criteria. The main failure is thatthere is no RCT of screening. Screening alsoappears to be cost-effective.

However, several issues need to be resolved. Themain aim of screening would be to reduce CVD,and that is increased not only in diabetes but alsoin IGT, and to a lesser extent in IFG. Dependingon which test for BG was used and what cut-offwas used to define ‘positives’, many more peoplewith IGT would be found than with diabetes. Adecision is needed on what to do for them.

The first national policy decision might thereforebe whether to screen for diabetes, or to have abroader, integrated approach to reduction of CVD.The next decision might be the balance ofinvestment in public health/health promotionmeasures, versus individual screening and care.


91



93


We thank Professor John Jarrett, London, and Mrs Audrey Birt, Diabetes UK, for

commenting on the draft of most chapters, andRoberta Ara, operation research analyst, Sheffield,for reviewing the modelling chapter.

We thank Pamela Royle, University of Aberdeen,for literature searches and for commenting on thedraft of Chapters 1, 2, 3 and 6, and Dr WilliamSimpson, consultant in biochemical medicine, forcommenting on a draft of Chapter 3.

Contribution of authorsNorman Waugh (Professor of Public Health) wasresponsible for Chapters 1–3 and 6 and for thefinal draft of the whole report. Graham Scotland(Research Fellow) and Paul McNamee (SeniorResearch Fellow) reviewed the previous economicmodels in Chapter 4. Mike Gillett (Operational

Research Analyst), Elisabeth Goyder (SeniorLecturer in Public Health) and Alan Brennan(Director of Health Economics and DecisionScience) were responsible for the new economicmodelling in Chapter 5. Rhys Williams (Professorof Clinical Epidemiology) and Ann John(Specialist Registrar in Public Health Medicine)reviewed the literature on prevention of type 2diabetes in Appendix 3. Chapters were exchangedfor internal review. Elisabeth Goyder reviewedChapters 1–3. Graham Scotland and NormanWaugh reviewed Chapter 5.

Norman Waugh coordinated the review, which wascommissioned by the NHS R&D HTA Programmeon behalf of the National Screening Committee.This views expressed are those of the authors andnot necessarily those of the NHS R&D HTAProgramme.

Acknowledgements


95


1. Wareham NJ, Griffin SJ. Should we screen for type2 diabetes? Evaluation against National ScreeningCommittee criteria. BMJ 2001;322:986–8.

2. Scott DA, Loveman E, McIntyre L, Waugh N.Screening for gestational diabetes: a systematicreview and economic evaluation. Health TechnolAssess 2002;6:(11).

3. Yorkshire and Humber Public Health Observatory.PBS diabetes population prevalence model. 2005. URL:http://www.yhpho.org.uk/viewResource.aspx?id=7.Accessed 26 December 2006.

4. Ritz E. Hypertension and nephropathy in type IIdiabetes. Proc R Coll Physicians Edinb 1996;26:374–83.

5. Laing SP, Swerdlow AJ, Carpenter LM, Slater SD,Burden AC, Botha JL, et al. Mortality fromcerebrovascular disease in a cohort of 23 000patients with insulin-treated diabetes. Stroke 2003;34:418–21.

6. Waugh NR. Amputations in diabetic patients – areview of rates, relative risks and resource use.Community Med 1988;10:279–88.

7. Groeneveld Y, Petri H, Hermans J, Springer MP.Relationship between blood glucose level andmortality in type 2 diabetes mellitus: a systematicreview. Diabetic Med 1999;16:2–13.

8. UKPDS Study Group. Intensive blood-glucosecontrol with sulphonylureas or insulin comparedwith conventional treatment and risk ofcomplications in patients with type 2 diabetes(UKPDS 33). Lancet 1998;352:837–53.

9. Effect of intensive blood-glucose control withmetformin on complications in overweightpatients with type 2 diabetes (UKPDS 34). UKProspective Diabetes Study (UKPDS) Group.Lancet 1998;352:854–65.

10. Goyder E, Irwig L. Screening for diabetes: whatare we really doing? BMJ 1998;317:1644–6.

11. Jarrett RJ. Screening for type II diabetes mellitus.Journal Med Screen 2000;7:2–3.

12. Keen H. The Bedford survey: a critique ofmethods and findings. Proc R Soc Med 1964;57:196–202.

13. Welborn TA, Curnow DH, Wearne JT, Cullen KJ,McCall MG, Stenhouse NS. Diabetes detected byblood-sugar measurement after a glucose load:report from the Busselton survey, 1966. Med J Aust1968;2:778–83.

14. Alberti KG, Zimmet PZ. Definition, diagnosis andclassification of diabetes mellitus and itscomplications. Part 1: diagnosis and classificationof diabetes mellitus provisional report of a WHOconsultation. Diabetic Med 1998;15:539–53.

15. World Health Organization. Diabetes mellitus: Reportof a WHO Study Group. Technical Report Series727. Geneva: WHO; 1985.

16. Goldstein DE, Little RR, Lorenz RA, Malone JI,Nathan D, Peterson CM, et al. Tests of glycemia indiabetes. Diabetes Care 2004;27:1761–73.

17. Gavin II, Alberti KGMM, Davidson MB, DeFronzo RA, Drash A, Gabbe SG, et al. Report ofthe Expert Committee on the Diagnosis andClassification of Diabetes Mellitus. Diabetes Care1997;20:1183–97.

18. Keen H. Criteria and classification of diabetesmellitus. In Mann JI, Pyorala K, Teuscher A,editors. Diabetes in epidemiological perspective.Edinburgh: Churchill Livingstone; 1983.pp. 167–82.

19. Classification and diagnosis of diabetes mellitusand other categories of glucose intolerance.National Diabetes Data Group. Diabetes 1979;28:1039–57.

20. WHO Expert Committee on Diabetes Mellitus:second report. World Health Organ Tech Rep Ser1980;646:1–80.

21. UK Prospective Diabetes Study Group. UKProspective Diabetes Study 16. Overview of6 years’ therapy of type II diabetes: a progressivedisease. Diabetes 1995;44:1249–58.

22. European Diabetes Epidemiology Group. Glucosetolerance and mortality: comparison of WHO andAmerican Diabetes Association diagnostic criteria.The DECODE Study Group. European DiabetesEpidemiology Group. Diabetes Epidemiology:collaborative analysis of diagnostic criteria inEurope. Lancet 1999;354:617–21.

23. Coutinho M, Gerstein HC, Wang Y, Yusuf S. The relationship between glucose and incidentcardiovascular events. A metaregression analysis ofpublished data from 20 studies of 95,783individuals followed for 12.4 years. Diabetes Care1999;22:233–40.

24. Weyer C, Bogardus C, Pratley RE. Metaboliccharacteristics of individuals with impaired fastingglucose and/or impaired glucose tolerance. Diabetes1999;48:2197–203.

References

25. Festa A, D’Agostino R, Jr., Hanley AJ, Karter AJ,Saad MF, Haffner SM. Differences in insulinresistance in nondiabetic subjects with isolatedimpaired glucose tolerance or isolated impairedfasting glucose. Diabetes 2004;53:1549–55.

26. Davies MJ, Gray IP. Impaired glucose tolerance.BMJ 1996;312:264–5.

27. Barrett-Connor E, Ferrara A. Isolatedpostchallenge hyperglycemia and the risk of fatalcardiovascular disease in older women and men.The Rancho Bernardo Study. Diabetes Care1998;21:1236–9.

28. Charles MA, Balkau B, Vauzelle-Kervroedan F,Thibult N, Eschwege E. Revision of diagnosticcriteria for diabetes. Lancet 1996;348:1657–8.

29. Fontbonne A, Charles MA, Thibult N, Richard JL,Claude JR, Warnet JM, et al. Hyperinsulinaemia asa predictor of coronary heart disease mortality in ahealthy population: the Paris Prospective Study,15-year follow-up. Diabetologia 1991;34:356–61.

30. Pyorala K, Savolainen E, Kaukola S, Haapakoski J.Plasma insulin as coronary heart disease riskfactor: relationship to other risk factors andpredictive value during 91/2-year follow-up of theHelsinki Policemen Study population. Acta MedScand Suppl 1985;701:38–52.

31. Haffner SM. Update on diabetes and themetabolic syndrome. Adv Stud Med 2003;3:277–85.

32. Hu G, Qiao Q, Tuomilehto J, Eliasson M, Feskens EJ, Pyorala K, et al. Plasma insulin andcardiovascular mortality in non-diabetic Europeanmen and women: a meta-analysis of data fromeleven prospective studies. Diabetologia 2004;47:1245–56.

33. Tominaga M, Eguchi H, Manaka H, Igarashi K,Kato T, Sekikawa A. Impaired glucose tolerance isa risk factor for cardiovascular disease, but notimpaired fasting glucose. The Funagata DiabetesStudy. Diabetes Care 1999;22:920–4.

34. Khaw KT, Wareham N, Luben R, Bingham S,Oakes S, Welch A, et al. Glycated haemoglobin,diabetes, and mortality in men in Norfolk cohortof European Prospective Investigation of Cancerand Nutrition (EPIC-Norfolk). BMJ 2001;322:15–18.

35. Piche ME, Lemieux S, Perusse L, Weisnagel SJ.High normal 2-hour plasma glucose is associatedwith insulin sensitivity and secretion that maypredispose to type 2 diabetes. Diabetologia 2005;48:732–40.

36. Muntner P, Wildman RP, Reynolds K, Desalvo KB,Chen J, Fonseca V. Relationship between HbA1clevel and peripheral arterial disease. Diabetes Care2005;28:1981–7.

37. Genuth S, Alberti KG, Bennett P, Buse J, Defronzo R, Kahn R, et al. Follow-up report on thediagnosis of diabetes mellitus. Diabetes Care2003;26:3160–7.

38. Balkau B, Hillier T, Vierron E, D’Hour A, Lepinay P, Royer B, et al. Creating a pandemic ofprediabetes: the proposed new diagnostic criteriafor impaired fasting glycaemia. Diabetologia2005;48:801–2.

39. Piche ME, Despres JP, Pascot A, Nadeau A,Tremblay A, Weisnagel SJ, et al. Predictors of thedevelopment of impaired fasting glucose versusimpaired glucose tolerance are partly different inmen: a 6-year follow-up study. Diabetologia 2004;47:590–2.

40. Reynolds K, Muntner P, Fonseca V. Metabolicsyndrome: underrated or underdiagnosed?Diabetes Care 2005;28:1831–2.

41. NCEP. Executive Summary of The Third Report of The National Cholesterol Education Program(NCEP) Expert Panel on Detection, Evaluation,and Treatment of High Blood Cholesterol inAdults (Adult Treatment Panel III). JAMA 2001;285:2486–97.

42. Jarrett RJ, Fitzgerald AP. Non-insulin-dependentdiabetes mellitus, glucose intolerance, bloodpressure, hypertension, and antihypertensivedrugs. Diabetic Med 1994;11:646–9.

43. Scheen AJ. Renin–angiotensin system inhibitionprevents type 2 diabetes mellitus. Part 1. A meta-analysis of randomised clinical trials. DiabetesMetab 2004;30:487–96.

44. Gillespie EL, White CM, Kardas M, Lindberg M,Coleman CI. The impact of ACE inhibitors orangiotensin II type 1 receptor blockers on thedevelopment of new-onset type 2 diabetes. DiabetesCare 2005;28:2261–6.

45. Scuteri A, Najjar SS, Morrell CH, Lakatta EG. The metabolic syndrome in older individuals:prevalence and prediction of cardiovascularevents: the Cardiovascular Health Study. DiabetesCare 2005;28:882–7.

46. Ford ES. Risks for all-cause mortality,cardiovascular disease, and diabetes associatedwith the metabolic syndrome: a summary of theevidence. Diabetes Care 2005;28:1769–78.

47. Stern MP, Williams K, Gonzalez-Villalpando C,Hunt KJ, Haffner SM. Does the metabolicsyndrome improve identification of individuals atrisk of type 2 diabetes and/or cardiovasculardisease? Diabetes Care 2004;27:2676–81.

48. Tillin T, Forouhi N, Johnston DG, McKeigue PM,Chaturvedi N, Godsland IF. Metabolic syndromeand coronary heart disease in South Asians,African-Caribbeans and white Europeans: a UK

References

96

population-based cross-sectional study. Diabetologia2005;48:649–56.

49. Reaven GM. Insulin resistance, cardiovasculardisease, and the metabolic syndrome: how well dothe emperor’s clothes fit? Diabetes Care 2004;27:1011–12.

50. Kahn R, Buse J, Ferrannini E, Stern M. Themetabolic syndrome: time for a critical appraisal.Joint statement from the American DiabetesAssociation and the European Association for theStudy of Diabetes. Diabetologia 2005;48:1684–99.

51. Perry IJ, Villegas R, Salim A, Flynn A. Clusteringof protective factors for glucose intolerance andinsulin resistance: a cross-sectional study. DiabetMed 2005;22:1091–7.

52. Goldstein DJ. Beneficial health effects of modestweight loss. Int J Obes Relat Metab Disord 1992;16: 397–415.

53. Heidemann C, Hoffmann K, Spranger J,Klipstein-Grobusch K, Mohlig M, Pfeiffer AF, et al.A dietary pattern protective against type 2 diabetesin the European Prospective Investigation intoCancer and Nutrition (EPIC)–Potsdam Studycohort. Diabetologia 2005;48:1126–34.

54. Harris MI, Klein R, Welborn TA, Knuiman MW.Onset of NIDDM occurs at least 4–7 yr beforeclinical diagnosis. Diabetes Care 1992;15:815–19.

55. Klein R, Klein BE, Moss SE. How many steps ofprogression of diabetic retinopathy aremeaningful? The Wisconsin epidemiologic studyof diabetic retinopathy. Arch Ophthalmol 2001;119:547–53.

56. Jarrett RJ. Duration of non-insulin-dependentdiabetes and development of retinopathy: analysisof possible risk factors. Diabet Med 1986;3:261–3.

57. Thompson TJ, Engelgau MM, Hegazy M, Ali MA,Sous ES, Badran A, et al. The onset of NIDDMand its relationship to clinical diagnosis inEgyptian adults. Diabetic Med 1996;13:337–40.

58. Colagiuri S, Cull CA, Holman RR. Are lowerfasting plasma glucose levels at diagnosis of type 2diabetes associated with improved outcomes?: UK prospective diabetes study 61. Diabetes Care2002;25:1410–17.

59. Gulliford MC, Charlton J, Latinovic R. Increasedutilization of primary care 5 years before diagnosisof type 2 diabetes: a matched cohort study. DiabetesCare 2005;28:47–52.

60. Bagust A, Beale S. Deteriorating beta-cell functionin type 2 diabetes: a long-term model. QJM 2003;96:281–8.

61. Jarrett RJ. Do we need IGT? Diabetic Med 1987;4:544–5.

62. Edelstein SL, Knowler WC, Bain RP, Andres R,Barrett-Connor EL, Dowse GK, et al. Predictors ofprogression from impaired glucose tolerance toNIDDM: an analysis of six prospective studies.Diabetes 1997;46:701–10.

63. Department of Health. Health Survey for England2003: latest trends: Table 6: body mass index (BMI), by survey year, age and sex. 2005. URL:http://www.dh.gov.uk/assetRoot/04/09/89/15/04098915.xls. Accessed 26 December 2006.

64. Lew EA, Garfinkel L. Variations in mortality byweight among 750,000 men and women. J ChronicDis 1979;32:563–76.

65. Oldridge NB, Stump TE, Nothwehr FK, Clark DO.Prevalence and outcomes of comorbid metabolicand cardiovascular conditions in middle- andolder-age adults. J Clin Epidemiol 2001;54:928–34.

66. Dunstan DW, Zimmet PZ, Welborn TA, de Court,Cameron AJ, Sicree RA, et al. The risingprevalence of diabetes and impaired glucosetolerance: the Australian Diabetes, Obesity andLifestyle Study. Diabetes Care 2002;25:829–34.

67. Glatthaar C, Welborn TA, Stenhouse NS, Garcia-Webb P. Diabetes and impaired glucosetolerance. A prevalence estimate based on theBusselton 1981 survey. Med J Aust 1985;143:436–40.

68. Guest CS, O’Dea K, Hopper JL, Nankervis AJ,Larkins RG. The prevalence of glucose intolerancein aborigines and Europids of south-easternAustralia. Diabetes Res Clin Pract 1992;15:227–35.

69. Leiter LA, Barr A, Belanger A, Lubin S, Ross SA,Tildesley HD, et al. Diabetes Screening in Canada(DIASCAN) Study: prevalence of undiagnoseddiabetes and glucose intolerance in familyphysician offices. Diabetes Care 2001;24:1038–43.

70. Prevalence of diabetes and impaired fastingglucose in adults – United States, 1999–2000.MMWR Morb Mortal Wkly Rep 2003;52:833–7.

71. Raikou M, McGuire A. The economics ofscreening and treatment in type 2 diabetesmellitus. Pharmacoeconomics 2003;21:543–64.

72. Evans JM, MacDonald TM, Leese GP, Ruta DA,Morris AD. Impact of type 1 and type 2 diabeteson patterns and costs of drug prescribing: apopulation-based study. Diabetes Care 2000;23:770–4.

73. Bagust A, Hopkinson PK, Maier W, Currie CJ. An economic model of the long-term health careburden of type II diabetes. Diabetologia 2001;44:2140–55.

74. Williams R, Van Gaal L, Lucioni C. Assessing theimpact of complications on the costs of type IIdiabetes. Diabetologia 2002;45:S13–S17.


97


75. Audit Commission. Testing times: a review of diabetesservices in England and Wales. 2000. URL:http://www.audit-commission.gov.uk/Products/NATIONAL-REPORT/EB2CA6BA-C5E5-4B8F-A984-898D19E8C603/nrdiabet.pdf. Accessed 27 December 2006.

76. Engelgau MM, Narayan KM, Herman WH.Screening for type 2 diabetes. Diabetes Care2000;23:1563–80.

77. Colhoun HM, Betteridge DJ, Durrington PN,Hitman GA, Neil HAW, Livingstone SJ, et al.Primary prevention of cardiovascular disease withatorvastatin in type 2 diabetes in the CollaborativeAtorvastatin Diabetes Study (CARDS): multicentrerandomised placebo-controlled trial. Lancet 2004;364:685–96.

78. The effect of aggressive versus standard lipidlowering by atorvastatin on diabetic dyslipidemia:the DALI study: a double-blind, randomized,placebo-controlled trial in patients with type 2diabetes and diabetic dyslipidemia. Diabetes Care2001;24:1335–41.

79. Collins R, Armitage J, Parish S, Sleigh P, Peto R.MRC/BHF Heart Protection Study of cholesterol-lowering with simvastatin in 5963 people withdiabetes: a randomised placebo-controlled trial.Lancet 2003;361:2005–16.

80. Harris R, Donahue K, Rathore SS, Frame P, Woolf SH, Lohr KN. Screening adults for type 2diabetes: a review of the evidence for the USPreventive Services Task Force. Ann Intern Med2003;138:215–29.

81. Royle PL, Bain L, Waugh NR. Sources of evidencefor systematic reviews of interventions in diabetes.Diabet Med 2005;22:1386–93.

82. Harris R, Donahue K, Rathore SS, Frame P, Woolf SH, Lohr KN. Screening adults for type 2diabetes: a review of the evidence for the USPreventive Services Task Force. 2003. URL:http://www.ahrq.gov/clinic/3rduspstf/diabscr/diabrev.pdf. Accessed 27 December 2006.

83. Hansson L, Zanchetti A, Carruthers SG, Dahlof B,Elmfeldt D, Julius S, et al. Effects of intensiveblood-pressure lowering and low-dose aspirin inpatients with hypertension: principal results of theHypertension Optimal Treatment (HOT)randomised trial. HOT Study Group. Lancet 1998;351:1755–62.

84. Ravid M, Rachmani R. Treatment of diabeticnephropathy: anti-hypertensive treatment. InBoner G, Cooper ME, editors. Management ofdiabetic nephropathy. London: Martin Dunitz; 2003.pp. 105–28.

85. ALLHAT-LLT. Major outcomes in moderatelyhypercholesterolemic, hypertensive patientsrandomized to pravastatin vs usual care: the

Antihypertensive and Lipid-Lowering Treatmentto Prevent Heart Attack Trial (ALLHAT-LLT).JAMA 2002;288:2998–3007.

86. National Institute for Health and ClinicalExcellence. Statins for the prevention of cardiovascularevents: Technology Appraisal 94. 2006.www.nice.org.uk. Accessed 28 December 2006.

87. Williamson DF, Vinicor F, Bowman BA, Centers forDisease Control and Primary Prevention. Primaryprevention of type 2 diabetes mellitus by lifestyleintervention: Implications for health policy. AnnIntern Med 2004;140:951–7.

88. Pan XR, Li GW, Hu YH, Wang JX, Yang WY, An ZX, et al. Effects of diet and exercise inpreventing NIDDM in people with impairedglucose tolerance. The Da Qing IGT and DiabetesStudy. Diabetes Care 1997;20:537–44.

89. Tuomilehto J, Lindstrom J, Eriksson JG, Valle TT,Hamalainen H, Ilanne-Parikka P, et al. Preventionof type 2 diabetes mellitus by changes in lifestyleamong subjects with impaired glucose tolerance. N Engl J Med 2001;344:1343–50.

90. Knowler WC, Barrett-Connor E, Fowler SE,Hamman RF, Lachin JM, Walker EA, et al.Reduction in the incidence of type 2 diabetes withlifestyle intervention or metformin. N Engl J Med2002;346:393–403.

91. Harris MI, Modan M. Screening for NIDDM. Whyis there no national program? Diabetes Care 1994;17:440–4.

92. Harris MI, Eastman RC. Early detection ofundiagnosed diabetes mellitus: a US perspective.Diabetes Metab Rese Rev 2000;16:230–6.

93. Expert Committee on the Diagnosis andClassification of Diabetes Mellitus. Report of theExpert Committee on the Diagnosis andClassification of Diabetes Mellitus. Diabetes Care2000;23 Suppl 1:S4–19.

94. Borch-Johnsen K, Lauritzen T, Glumer C,Sandbaek A. Screening for type 2 diabetes –should it be now? Diabetic Med 2003;20:175–81.

95. Lauritzen T, Griffin S, Borch-Johnsen K, Wareham NJ, Wolffenbuttel BH, Rutten G. TheADDITION study: proposed trial of the cost-effectiveness of an intensive multifactorialintervention on morbidity and mortality amongpeople with type 2 diabetes detected by screening.Int J Obes Relat Metab Disord 2000;24 Suppl 3:S6–11.

96. American Diabetes Association. Position statement:early identification of people with type 2 diabetes.2002. URL: http://www.diabetes.org.uk/About_us/Our_Views/Position_statements/Early_identification_of_people_with_type_2_diabetes/.Accessed 26 December 2006.

References

98

97. American Diabetes Association. Standards ofmedical care in diabetes. Diabetes Care 2005;28Suppl 1:S4–36.

98. Hardman AE, Stensel DJ. Physical activity andhealth. London: Routledge; 2003.

99. Manson JE, Nathan DM, Krolewski AS, Stampfer MJ, Willett WC, Hennekens CH. Aprospective study of exercise and incidence ofdiabetes among US male physicians. JAMA 1992;268:63–7.

100. Hu FB, Manson JE, Stampfer MJ, Colditz G, Liu S, Solomon CG, et al. Diet, lifestyle, and therisk of type 2 diabetes mellitus in women. N Engl JMed 2001;345:790–7.

101. Eriksson KF, Lindgarde F. Prevention of type 2(non-insulin-dependent) diabetes mellitus by dietand physical exercise. The 6-year Malmo feasibilitystudy. Diabetologia 1991;34:891–8.

102. Tan HH, McAlpine RR, James P, Thompson P,McMurdo MET, Morris AD, et al. Diagnosis of type2 diabetes at an older age: Effect on mortality inmen and women. Diabetes Care 2004;27:2797–9.

103. Vijan S, Hofer TP, Hayward RA. Estimated benefitsof glycemic control in microvascular complicationsin type 2 diabetes. Ann Intern Med 1997;127:788–95.

104. Riste L, Khan F, Cruickshank K. High prevalenceof type 2 diabetes in all ethnic groups, includingEuropeans, in a British inner city: relative poverty,history, inactivity, or 21st century Europe? DiabetesCare 2001;24:1377–83.

105. Baan CA, Ruige JB, Stolk RP, Witteman JC,Dekker JM, Heine RJ, et al. Performance of apredictive model to identify undiagnosed diabetesin a health care setting. Diabetes Care 1999;22:213–19.

106. Adriaanse MC, Snoek FJ, Dekker JM, Van der Ploeg HM, Heine RJ. Screening for type 2diabetes: an exploration of subjects’ perceptionsregarding diagnosis and procedure. Diabetic Med2002;19:406–11.

107. Herman WH, Smith PJ, Thompson TJ, Engelgau MM, Aubert RE. A new and simplequestionnaire to identify people at increased riskfor undiagnosed diabetes. Diabetes Care 1995;18:382–7.

108. Burden ML, Burden AC. The American DiabetesAssociation screening questionnaire for diabetes. Is it worthwhile in the UK? Diabetes Care 1994;17:97.

109. Franciosi M, De Berardis G, Rossi MC, Sacco M,Belfiglio M, Pellegrini F, et al. Use of the diabetesrisk score for opportunistic screening ofundiagnosed diabetes and impaired glucosetolerance: the IGLOO (Impaired Glucose

Tolerance and Long-Term OutcomesObservational) study. Diabetes Care 2005;28:1187–94.

110. Griffin SJ, Little PS, Hales CN, Kinmonth AL,Wareham NJ. Diabetes risk score: towards earlierdetection of type 2 diabetes in general practice.Diabetes Metab Res Rev 2000;16:164–71.

111. Spijkerman AM, Adriaanse MC, Dekker JM,Nijpels G, Stehouwer CD, Bouter LM, et al.Diabetic patients detected by population-basedstepwise screening already have a diabeticcardiovascular risk profile. Diabetes Care 2002;25:1784–9.

112. Greaves CJ, Stead JW, Hattersley AT, Ewings P,Brown P, Evans PH. A simple pragmatic system for detecting new cases of type 2 diabetes andimpaired fasting glycaemia in primary care. Fam Pract 2004;21:57–62.

113. Schmidt MI, Duncan BB, Bang H, Pankow JS,Ballantyne CM, Golden SH, et al. Identifyingindividuals at high risk for diabetes: theAtherosclerosis Risk in Communities study. DiabetesCare 2005;28:2013–18.

114. Davies M, Alban-Davies H, Cook C, Day J. Selftesting for diabetes mellitus. BMJ 1991;303:696–8.

115. Qiao Q, Keinanen-Kiukaanniemi S, Rajala U,Uusimaki A, Kivela SL. Random capillary wholeblood glucose test as a screening test for diabetesmellitus in a middle-aged population. Scand J ClinLab Invest 1995;55:3–8.

116. NGSP. National Glycohemoglobin StandardizationProgram. 2006. URL: www.ngsp.org. Accessed 28 December 2006.

117. Expert Committee on the Diagnosis andClassification of Diabetes Mellitus. Report of theExpert Committee on the Diagnosis andClassification of Diabetes Mellitus. Diabetes Care2003;26 Suppl 1:S5–20.

118. Khaw KT, Wareham N, Bingham S, Luben R,Welch A, Day N. Association of hemoglobin A1cwith cardiovascular disease and mortality in adults:the European prospective investigation into cancerin Norfolk. Ann Intern Med 2004;141:413–20.

119. Gerstein HC. Glycosylated hemoglobin: finallyready for prime time as a cardiovascular riskfactor. Ann Intern Med 2004;141:475–6.

120. Rohlfing CL, Little RR, Wiedmeyer HM, England JD, Madsen R, Harris MI, et al. Use ofGHb (HbA1c) in screening for undiagnoseddiabetes in the US population. Diabetes Care2000;23:187–91.

121. Ellison TL, Elliott R, Moyes SA. HbA1c screeningfor undiagnosed diabetes in New Zealand. DiabetesMetab Res Rev 2005;21:65–70.


99


122. McCance DR, Hanson RL, Charles MA, Jacobsson LT, Pettitt DJ, Bennett PH, et al.Comparison of tests for glycated haemoglobin andfasting and two hour plasma glucoseconcentrations as diagnostic methods for diabetes.BMJ 1994;308:1323–8.

123. Peters AL, Davidson MB, Schriger DL, Hasselblad V. A clinical approach for the diagnosisof diabetes mellitus: an analysis using glycosylatedhemoglobin levels. Meta-analysis Research Groupon the Diagnosis of Diabetes Using GlycatedHemoglobin Levels. JAMA 1996;276:1246–52.

124. Woerle HJ, Pimenta WP, Meyer C, Gosmanov NR,Szoke E, Szombathy T, et al. Diagnostic andtherapeutic implications of relationships betweenfasting, 2-hour postchallenge plasma glucose andhemoglobin A1c values. Arch Intern Med 2004;164:1627–32.

125. Jeffcoate SL. Diabetes control and complications:the role of glycated haemoglobin, 25 years on.Diabetic Med 2004;21:657–65.

126. Monnier L, Lapinski H, Colette C. Contributionsof fasting and postprandial plasma glucoseincrements to the overall diurnal hyperglycemia oftype 2 diabetic patients: variations with increasinglevels of HbA1c. Diabetes Care 2003;26:881–5.

127. Spijkerman A, Griffin S, Dekker J, Nijpels G,Wareham NJ. What is the risk of mortality forpeople who are screen positive in a diabetesscreening programme but who do not havediabetes on biochemical testing? Diabetesscreening programmes from a public healthperspective. J Med Screen 2002;9:187–90.

128. Hanson RL, Nelson RG, McCance DR, Beart JA,Charles MA, Pettitt DJ, et al. Comparison ofscreening tests for non-insulin-dependent diabetesmellitus. Arch Intern Med 1993;153:2133–40.

129. Drummond MF, Jefferson TO. Guidelines forauthors and peer reviewers of economicsubmissions to the BMJ. The BMJ EconomicEvaluation Working Party. BMJ 1996;313:275–83.

130. Philips Z, Ginnelly L, Sculpher M, Claxton K,Golder S, Riemsma R, et al. Review of guidelinesfor good practice in decision-analytic modelling inhealth technology assessment. Health Technol Assess2004;8:(36).

131. CDC Diabetes Cost-Effectiveness Study Group.The cost-effectiveness of screening for type 2diabetes. CDC Diabetes Cost-Effectiveness StudyGroup, Centers for Disease Control andPrevention. JAMA 1998;280:1757–63.

132. Chen TH, Yen MF, Tung TH. A computersimulation model for cost-effectiveness analysis ofmass screening for type 2 diabetes mellitus.Diabetes Res Clin Pract 2001;54:S37–42.

133. Hoerger TJ, Harris R, Hicks KA, Donahue K,Sorensen S, Engelgau M. Screening for type 2diabetes mellitus: a cost-effectiveness analysis. AnnIntern Med 2004;140:689–99.

134. Eastman RC, Javitt JC, Herman WH, Dasbach EJ,Copley-Merriman C, Maier W, et al. Model ofcomplications of NIDDM. II. Analysis of thehealth benefits and cost-effectiveness of treatingNIDDM with the goal of normoglycemia. DiabetesCare 1997;20:735–44.

135. Eastman RC, Javitt JC, Herman WH, Dasbach EJ,Zbrozek AS, Dong F, et al. Model of complicationsof NIDDM. I. Model construction andassumptions. Diabetes Care 1997;20:725–34.

136. Goyder EC, Irwig LM. Screening for type 2diabetes mellitus: a decision analytic approach.Diabetic Med 2000;17:469–77.

137. Williams DR, Wareham NJ, Brown DC, Byrne CD,Clark PM, Cox BD, et al. Undiagnosed glucoseintolerance in the community: the Isle of ElyDiabetes Project. Diabetic Med 1995;12:30–5.

138. Harris MI. Undiagnosed NIDDM: clinical andpublic health issues. Diabetes Care 1993;16:642–52.

139. Diabetes Control and Complications TrialResearch Group. The effect of intensive treatmentof diabetes on the development and progression oflong-term complications in insulin-dependentdiabetes mellitus. N Engl J Med 1993;329:977–86.

140. UKPDS Study Group. UK Prospective DiabetesStudy (UKPDS). VIII. Study design, progress andperformance. Diabetologia 1991;34:877–90.

141. Turner R, Cull C, Holman R. United KingdomProspective Diabetes Study 17: a 9-year update ofa randomized, controlled trial on the effect ofimproved metabolic control on complications innon-insulin-dependent diabetes mellitus. AnnIntern Med 1996;124:136–45.

142. Welborn TA, Glatthaar C, Whittall D, Bennett S.An estimate of diabetes prevalence from a nationalpopulation sample: a male excess. Med J Aust1989;150:78–81.

143. McPhillips JB, Barrett-Connor E, Wingard DL.Cardiovascular disease risk factors prior to thediagnosis of impaired glucose tolerance and non-insulin-dependent diabetes mellitus in acommunity of older adults. Am J Epidemiol 1990;131:443–53.

144. Davies M, Day J. Screening for non-insulin-dependent diabetes mellitus (NIDDM): how oftenshould it be performed? J Med Screen 1994;1:78–81.

145. Andersson DK, Svardsudd K, Tibblin G.Prevalence and incidence of diabetes in a Swedishcommunity 1972–1987. Diabetic Med 1991;8:428–34.

References

100

146. Koskinen P, Manttari M, Manninen V, Huttunen JK,Heinonen OP, Frick MH. Coronary heart diseaseincidence in NIDDM patients in the HelsinkiHeart Study. Diabetes Care 1992;15:820–5.

147. Pyorala K, Pedersen TR, Kjekshus J, Faergeman O,Olsson AG, Thorgeirsson G. Cholesterol loweringwith simvastatin improves prognosis of diabeticpatients with coronary heart disease. A subgroupanalysis of the Scandinavian Simvastatin SurvivalStudy (4S). Diabetes Care 1997;20:614–20.

148. Design, rational, and baseline characteristics of theProspective Pravastatin Pooling (PPP) project – acombined analysis of three large-scale randomizedtrials: Long-term Intervention with Pravastatin inIschemic Disease (LIPID), Cholesterol andRecurrent Events (CARE), and West of ScotlandCoronary Prevention Study (WOSCOPS). Am JCardiol 1995;76:899–905.

149. Fuller J, Stevens LK, Chaturvedi N, Holloway JF.Antihypertensive therapy for preventing cardiovascularcomplications in people with diabetes mellitus. CochraneReview. Chichester: Wiley; 1997. Issue 4,CD 002188.

150. Curb JD, Pressel SL, Cutler JA, Savage PJ,Applegate WB, Black H, et al. Effect of diuretic-based antihypertensive treatment on cardiovasculardisease risk in older diabetic patients with isolatedsystolic hypertension. Systolic Hypertension in theElderly Program Cooperative Research Group.JAMA 1996;276:1886–92.

151. UK Prospective Diabetes Study. Tight bloodpressure control and risk of macrovascular andmicrovascular complications in type 2 diabetes:UKPDS 38. UK Prospective Diabetes Study Group.BMJ 1998;317:703–13.

152. Dyck PJ, Kratz KM, Karnes JL, Litchy WJ, Klein R,Pach JM, et al. The prevalence by staged severityof various types of diabetic neuropathy,retinopathy, and nephropathy in a population-based cohort: the Rochester Diabetic NeuropathyStudy. Neurology 1993;43:817–24.

153. Humphrey LL, Palumbo PJ, Butters MA, Hallett JW Jr, Chu CP, O’Fallon WM, et al. Thecontribution of non-insulin-dependent diabetes tolower-extremity amputation in the community.Arch Intern Med 1994;154:885–92.

154. Reiber GE, Boyko EJ, Smith DG. Lower extremityfoot ulcers and amputations in diabetes. In HarrisMI, Cowie CC, Stern MP, Boyko EJ, Reiber GE,Bennett PH, editors. Diabetes in America. Bethesda,MD: National Institute of Diabetes and Digestiveand Kidney Diseases of the National Institutes ofHealth. 1995. pp. 409–28.

155. Nelson RG, Knowler WC, Pettitt DJ, Hanson RL,Bennett PH. Incidence and determinants ofelevated urinary albumin excretion in PimaIndians with NIDDM. Diabetes Care 1995;18:182–7.

156. Klein R, Klein BE, Moss SE, Davis MD, DeMets DL. The Wisconsin Epidemiologic Studyof Diabetic Retinopathy. X. Four-year incidenceand progression of diabetic retinopathy when ageat diagnosis is 30 years or more. Arch Ophthalmol1989;107:244–9.

157. Anderson KM, Odell PM, Wilson PW, Kannel WB.Cardiovascular disease risk profiles. Am Heart J1991;121:293–8.

158. Clarke PM, Gray AM, Briggs A, Farmer AJ, Fenn P,Stevens RJ, et al. A model to estimate the lifetimehealth outcomes of patients with type 2 diabetes:the United Kingdom Prospective Diabetes Study(UKPDS) Outcomes Model (UKPDS no. 68).Diabetologia 2004;47:1747–59.

159. Weinstein MC, Coxson PG, Williams LW, Pass TM,Stason WB, Goldman L. Forecasting coronaryheart disease incidence, mortality, and cost: theCoronary Heart Disease Policy Model. Am J PublicHealth 1987;77:1417–26.

160. Hunink MG, Goldman L, Tosteson AN, Mittleman MA, Goldman PA, Williams LW, et al.The recent decline in mortality from coronaryheart disease, 1980–1990. The effect of seculartrends in risk factors and treatment. JAMA 1997;277:535–42.

161. Sacco RL, Shi T, Zamanillo MC, Kargman DE.Predictors of mortality and recurrence afterhospitalized cerebral infarction in an urbancommunity: the Northern Manhattan Stroke Study.Neurology 1994;44:626–34.

162. Schrier RW, Estacio RO, Esler A, Mehler P. Effects of aggressive blood pressure control innormotensive type 2 diabetic patients onalbuminuria, retinopathy and strokes. Kidney Int2002;61:1086–97.

163. Dasbach EJ, Fryback DJ, Thornbury JR. Healthutility preference differences [abstract]. Med DecisMaking 1992;12:4.

164. Klein R, Klein BE, Moss SE. Prevalence ofmicroalbuminuria in older-onset diabetes. DiabetesCare 1993;16:1325–30.

165. Eckman MH, Greenfield S, Mackey WC, Wong JB,Kaplan S, Sullivan L, et al. Foot infections indiabetic patients. Decision and cost-effectivenessanalyses. JAMA 1995;273:712–20.

166. Tsevat J, Goldman L, Soukup JR, Lamas GA,Connors KF, Chapin CC, et al. Stability of time-tradeoff utilities in survivors of myocardialinfarction. Med Decis Making 1993;13:161–5.

167. Wolf PA, D’Agostino RB, O’Neal MA, Sytkowski P,Kase CS, Belanger AJ, et al. Secular trends instroke incidence and mortality. The FraminghamStudy. Stroke 1992;23:1551–5.


101


168. Murray CJ, Lopez AD. The global burden of disease:a comprehensive assessment of mortality and disabilityfrom diseases, injuries and risk factors in 1990 andprojected to 2020. Boston, MA: Harvard School ofPublic Health; 1996.

169. Chiasson JL, Josse RG, Gomis R, Hanefeld M,Karasik A, Laakso M. Acarbose treatment and therisk of cardiovascular disease and hypertension inpatients with impaired glucose tolerance: theSTOP-NIDDM trial. JAMA 2003;290:486–94.

170. Diabetes Prevention Program Research Group.Within-trial cost-effectiveness of lifestyleintervention or metformin for the primaryprevention of type 2 diabetes. Diabetes Care 2003;26:2518–23.

171. Herman WH, Hoerger TJ, Brandle M, Hicks K,Sorensen S, Zhang P, et al. The cost-effectiveness oflifestyle modification or metformin in preventingtype 2 diabetes in adults with impaired glucosetolerance. Ann Intern Med 2005;142:323–32.

172. Palmer AJ, Roze S, Valentine WJ, Spinas GA, Shaw JE, Zimmet PZ. Intensive lifestyle changes ormetformin in patients with impaired glucosetolerance: modeling the long-term healtheconomic implications of the diabetes preventionprogram in Australia, France, Germany,Switzerland, and the United Kingdom. Clin Ther2004;26:304–21.

173. Caro JJ, Getsios D, Caro I, Klittich WS, O’Brien JA.Economic evaluation of therapeutic interventionsto prevent Type 2 diabetes in Canada. Diabetic Med2004;21:1229–36.

174. Segal L, Dalton AC, Richardson J. Cost-effectiveness of the primary prevention of non-insulin dependent diabetes mellitus. Health PromotInternation 1998;13:197–209.

175. McEwan P, Bergenheim K, Peters JR, Williams R,Currie CJ. Delaying the onset of type 2 diabetesmay have a substantial impact on macrovascularand microvascular events and patient costs. Posterpresented at the 65th Annual Scientific Session ofthe American Diabetes Association, 10–14 June2005.

176. Harris MI, Hadden WC, Knowler WC, Bennett PH.Prevalence of diabetes and impaired glucosetolerance and plasma glucose levels in USpopulation aged 20–74 yr. Diabetes 1987;36:523–34.

177. Caro JJ, Ward AJ, O’Brien JA. Lifetime costs ofcomplications resulting from type 2 diabetes in theUS. Diabetes Care 2002;25:476–81.

178. Caro JJ, Klittich WS, Raggio G, Kavanagh PL,O’Brien JA, Shomphe LA, et al. Economicassessment of troglitazone as an adjunct tosulfonylurea therapy in the treatment of type 2diabetes. Clin Ther 2000;22:116–27.

179. Thompson TJ, Engelgau MM, Hegazy M, Ali MA,Sous ES, Badran A, et al. The onset of NIDDMand its relationship to clinical diagnosis inEgyptian adults. Diabet Med 1996;13:337–40.

180. Qureshi AI, Giles WH, Croft JB. Impaired glucosetolerance and the likelihood of nonfatal stroke andmyocardial infarction: the Third National Healthand Nutrition Examination Survey. Stroke 1998;29:1329–32.

181. Barzilay JI, Spiekerman CF, Kuller LH, Burke GL,Bittner V, Gottdiener JS, et al. Prevalence ofclinical and isolated subclinical cardiovasculardisease in older adults with glucose disorders: theCardiovascular Health Study. Diabetes Care 2001;24:1233–9.

182. Kohner EM, Aldington SJ, Stratton IM, Manley SE,Holman RR, Matthews DR, et al. United KingdomProspective Diabetes Study, 30: diabeticretinopathy at diagnosis of non-insulin-dependentdiabetes mellitus and associated risk factors. ArchOphthalmol 1998;116:297–303.

183. Adler AI, Stevens RJ, Manley SE, Bilous RW, Cull CA, Holman RR. Development andprogression of nephropathy in type 2 diabetes: theUnited Kingdom Prospective Diabetes Study(UKPDS 64). Kidney Int 2003;63:225–32.

184. Stevens RJ, Kothari V, Adler AI, Stratton IM,United Kingdom Prospective Diabetes Study(UKPDS) Group. The UKPDS risk engine: amodel for the risk of coronary heart disease intype II diabetes (UKPDS 56). Clin Sci 2001;101:671–9.

185. Kothari V, Stevens RJ, Adler AI, Stratton IM,Manley SE, Neil HA, et al. UKPDS 60: risk ofstroke in type 2 diabetes estimated by the UKProspective Diabetes Study risk engine. Stroke2002;33:1776–81.

186. Balkau B, Eschwege E, Papoz L, Richard JL,Claude JR, Warnet JM, et al. Risk factors for earlydeath in non-insulin dependent diabetes and menwith known glucose tolerance status. BMJ 1993;307:295–9.

187. Simons LA, McCallum J, Friedlander Y, Simons J.Diabetes, mortality and coronary heart disease inthe prospective Dubbo study of Australian elderly.Aust N Z J Med 1996;26:66–74.

188. Lee IM, Manson JE, Hennekens CH, Paffenbarger RS Jr. Body weight and mortality. A 27-year follow-up of middle-aged men. JAMA1993;270:2823–8.

189. Rissanen A, Heliovaara M, Knekt P, Reunanen A,Aromaa A, Maatela J. Risk of disability andmortality due to overweight in a Finnishpopulation. BMJ 1990;301:835–7.

References

102

190. Manson JE, Stampfer MJ, Hennekens CH, Willett WC. Body weight and longevity. A reassessment. JAMA 1987;257:353–8.

191. Saydah SH, Loria CM, Eberhardt MS, Brancati FL.Subclinical states of glucose intolerance and risk ofdeath in the US. Diabetes Care 2001;24:447–53.

192. Dong F, Orians C, Manninen D. Economicevaluation of approaches to preventing diabetic end-stage renal disease. Prepared for the Centers for DiseaseControl and Prevention. Seattle, WA: Battelle-Centers for Public Health Research andEvaluation; 1997.

193. O’Brien JA, Shomphe LA, Kavanagh PL, Raggio G,Caro JJ. Direct medical costs of complicationsresulting from type 2 diabetes in the US. DiabetesCare 1998;21:1122–8.

194. O’Brien JA, Caro I, Getsios D, Caro JJ. Diabetes inCanada: direct medical costs of majormacrovascular complications. Value Health 2001;4:258–65.

195. Jonsson B, Advisory Board. Revealing the cost oftype II diabetes in Europe. Diabetologia 2002;45:S5–12.

196. Nichols GA, Glauber HS, Brown JB. Type 2diabetes: incremental medical care costs duringthe 8 years preceding diagnosis. Diabetes Care2000;23:1654–9.

197. Brandle M, Zhou H, Smith BR, Marriott D, Burke R, Tabaei BP, et al. The direct medical costof type 2 diabetes. Diabetes Care 2003;26:2300–4.

198. Laupacis A, Feeny D, Detsky AS, Tugwell PX. How attractive does a new technology have to beto warrant adoption and utilization? Tentativeguidelines for using clinical and economicevaluations. CMAJ 1992;146:473–81.

199. Klonoff DC, Schwartz DM. An economic analysisof interventions for diabetes. Diabetes Care 2000;23:390–404.

200. Lawrence JM, Bennett P, Young A, Robinson AM.Screening for diabetes in general practice: crosssectional population study. BMJ 2001;323:548–51.

201. Icks A, Haastert B, Gandjour A, John J, Lowel H,Holle R, et al. Cost-effectiveness analysis ofdifferent screening procedures for type 2 diabetes:the KORA Survey 2000. Diabetes Care 2004;27:2120–8.

202. Johnson SL, Tabaei BP, Herman WH. The efficacyand cost of alternative strategies for systematicscreening for type 2 diabetes in the US.population 45–74 years of age. Diabetes Care 2005;28:307–11.

203. Shirasaya K, Miyakawa M, Yoshida K, Takahashi E,Shimada N, Kondo T. Economic evaluation ofalternative indicators for screening for diabetesmellitus. Prev Med 1999;29:79–86.

204. Zhang P, Engelgau MM, Valdez R, Benjamin SM,Cadwell B, Narayan KM. Costs of screening forpre-diabetes among US adults: a comparison ofdifferent screening strategies. Diabetes Care 2003;26:2536–42.

205. Zhang P, Engelgau MM, Valdez R, Cadwell B,Benjamin SM, Narayan KM. Efficient cutoff pointsfor three screening tests for detecting undiagnoseddiabetes and pre-diabetes: an economic analysis.Diabetes Care 2005;28:1321–5.

206. Rathmann W, Haastert B, Icks A, Lowel H,Meisinger C, Holle R, et al. High prevalence ofundiagnosed diabetes mellitus in SouthernGermany: target populations for efficientscreening. The KORA survey 2000. Diabetologia2003;46:182–9.

207. Bureau of the Census. Population projections of theUnited States by age, race and Hipanic origin: 1995 to2050. Current Population Reports. Washington, DC:US Government Printing Office; 1996.

208. National Center for Health Statistics. Thirdnational health and nutrition examination survey,1988–1994, reference manual and reports. Manual formedical technicians and laboratory procedures used forNHANES III. Hyattsville, MD: Centers for DiseaseControl and Prevention; 1996.

209. Diabetes in the UK. 2004. URL:http://www.diabetes.org.uk/Professionals/Information_resources/Reports/Diabetes_in_the_UK_2004/. Accessed 26 December 2006.

210. Harris MI, Eastman RC, Cowie CC, Flegal KM,Eberhardt MS. Comparison of diabetes diagnosticcategories in the US population according to the1997 American Diabetes Association and1980–1985 World Health Organization diagnosticcriteria. Diabetes Care 1997;20:1859–62.

211. Perry RC, Shankar RR, Fineberg N, McGill J,Baron AD, Early Diabetes Intervention Program.HbA1c measurement improves the detection oftype 2 diabetes in high-risk individuals withnondiagnositc levels of fasting plasma glucose: theEarly Diabetes Intervention Program (EDIP).Diabetes Care 2001;24:465–71.

212. Simmons D. Parity, ethnic group and theprevalence of type 2 diabetes: the CoventryDiabetes Study. Diabet Med 1992;9:706–9.

213. Hofer TP, Vijan S, Hayward RA. Estimating themicrovascular benefits of screening for type 2diabetes mellitus. Int J Technol Assess Health Care2000;16:822–33.

214. O’Connor PJ, Rush WA, Engelgau M, Gregg EW,Stiffman MN. What a difference a year makes:changes in blood pressure, HbA1c, LDL, andweight within one year of a new diagnosis ofdiabetes. Presented at the Centers for Disease


103


Control. Diabetes Translation Conference, NewOrleans, LA, 2000.

215. Gatling W, Guzder RN, Turnbull JC, Budd S,Mullee MA. The Poole Diabetes Study: how manycases of type 2 diabetes are diagnosed each yearduring normal health care in a definedcommunity? Diabetes Res Clin Pract 2001;53:107–12.

216. Topp B, Promislow K, deVries G, Miura RM,Finegood DT. A model of beta-cell mass, insulin,and glucose kinetics: pathways to diabetes. J TheorBiol 2000;206:605–19.

217. Bergenstal RM, Kendall DM, Franz MJ,Rubenstein AH. The Natural History of Type 2Diabetes (Chart), in Management of type 2diabetes: a systematic approach to meeting thestandards of care. II: oral agents, insulin, andmanagement of complications. In DeGroot LJ,Jameson JL, editors. Endocrinology, Chapter 58.Philadelphia, PA: Saunders; 2001.

218. Davis TM, Cull CA, Holman RR. Relationshipbetween ethnicity and glycemic control, lipidprofiles, and blood pressure during the first 9years of type 2 diabetes: UK. Prospective DiabetesStudy (UKPDS 55). Diabetes Care 2001;24:1167–74.

219. Selvin E, Marinopoulos S, Berkenblit G, Rami T,Brancati FL, Powe NR, et al. Meta-analysis:glycosylated hemoglobin and cardiovasculardisease in diabetes mellitus. Ann Intern Med 2004;141:421–31.

220. National Institute of Diabetes and Digestive andKidney Diseases (NIDDK). US National Institutes ofHealth (NIH) News. 2005. URL:http://www.nih.gov/news/pr/jun2005/niddk-12a.htm. Accessed 26 December 2006.

221. Stratton IM, Adler AI, Neil HA, Matthews DR,Manley SE, Cull CA, et al. Association of glycaemiawith macrovascular and microvascularcomplications of type 2 diabetes (UKPDS 35):prospective observational study. BMJ 2000;321:405–12.

222. ScHARR. Health Economics and Decision Scienceswebsite – Discussion Papers. 2005. URL:http://www.shef.ac.uk/scharr/sections/heds/discussion.html. Accessed 26 December 2006.

223. Speight J. Assessing the impact of diabetesscreening on quality of life or quality of health?:semantics are important. Diabetes Care 2002;25:1893–4.

224. Edelman D, Olsen MK, Dudley TK, Harris AC,Oddone EZ. Impact of diabetes screening onquality of life. Diabetes Care 2002;25:1022–6.

225. Forouhi NG, Merrick D, Goyder E, Ferguson BA,Abbas J, Lachowycz K, et al. Diabetes prevalence in England, 2001 – estimates from anepidemiological model. Diabetic Med 2006;23:189–97.

226. PBS prevalence model documentation. 2005. URL:http://www.yhpho.org.uk/viewResource.aspx?id=7.Accessed 26 December 2006.

227. Wild S, Roglic G, Green A, Sicree R, King H.Global prevalence of diabetes: estimates for theyear 2000 and projections for 2030. Diabetes Care2004;27:1047–53.

228. King H, Rewers M. Global estimates forprevalence of diabetes mellitus and impairedglucose tolerance in adults. WHO Ad HocDiabetes Reporting Group. Diabetes Care 1993;16:157–77.

229. The National Clinical Audit Support Programme(NCASP). National Diabetes Audit. 2006. URL:http://www.icservices.nhs.uk/ncasp/pages/audit_topics/diabetes/default-new.asp. Accessed 26December 2006.

230. Conference Report on ADA: Diabetic RetinopathyDetected in Pre-Diabetes Patients. 2005. URL:http://www.medpagetoday.com/Endocrinology/2005ADAMeeting/tb/1185. Accessed 26 December2006.

231. Stratton IM, Kohner EM, Aldington SJ, Turner RC,Holman RR, Manley SE, et al. UKPDS 50: riskfactors for incidence and progression ofretinopathy in type II diabetes over 6 years fromdiagnosis. Diabetologia 2001;44:156–63.

232. Tayside Diabetes Network. The Tayside diabeteshandbook – screening and management of retinopathy.2005. URL: http://www.diabetes-healthnet.ac.uk/handbook/eyecomp.htm. Accessed 26 December2006.

233. Adler AI, Stevens RJ, Neil A, Stratton IM, Boulton AJ, Holman RR. UKPDS 59:hyperglycemia and other potentially modifiablerisk factors for peripheral vascular disease in type2 diabetes. Diabetes Care 2002;25:894–9.

234. Prescription Pricing Authority. NHS Drug Tariff.2005. URL: http://www.ppa.org.uk/edt/July_2005/mindex.htm. Accessed 26 December 2006.

235. Diabetes Control and Complications TrialResearch Group. The relationship of glycemicexposure (HbA1c) to the risk of development andprogression of retinopathy in the diabetes controland complications trial. Diabetes 1995;44:968–83.

236. O’Hagan A, Stevenson M, Madan J. Monte Carloprobabilistic sensitivity analysis for patient levelsimulation models. 2005. URL:http://www.tonyohagan.co.uk/academic/abs/MCPSA.html. Accessed 26 December 2006.

237. UCL Department of Epidemiology and PublicHealth. Whitehall II Study. 2006. URL:http://www.ucl.ac.uk/whitehallII/publications/index.htm. Accessed 26 December 2006.

References

104

238. The Oxford Centre for Diabetes EaMDTU. The UK Prospective Diabetes Study. 2006. URL:http://www.dtu.ox.ac.uk/index.php?maindoc=/ukpds/. Accessed 27 December 2006.

239. Morris AD, Boyle DI, MacAlpine R, Emslie-Smith A, Jung RT, Newton RW, et al. Thediabetes audit and research in Tayside Scotland(DARTS) study: electronic record linkage to createa diabetes register. DARTS/MEMO Collaboration.BMJ 1997;315:524–8.

240. Hennekens CH, Knatterud GL, Pfeffer MA. Use ofaspirin to reduce risks of cardiovascular disease inpatients with diabetes: clinical and researchchallenges. Diabetes Care 2004;27:2752–4.

241. Glumer C, Carstensen B, Sandbaek A, Lauritzen T,Jorgensen T, Borch-Johnsen K. A Danish diabetesrisk score for targeted screening: the Inter99study. Diabetes Care 2004;27:727–33.

242. Selective review of the literature on the relationshipbetween BMI and type 2 diabetes. [See Appendix 2 ofPBS Diabetes Population Prevalence Model –Phase 2 Briefing Document]. 2005. URL:http://www.york.ac.uk/yhpho/documents/pbsdpm/pbs%20phase%202/PBS%20Briefing%20Document%20-%20Phase2.pdf. Accessed 26 December 2006.

243. UK Prospective Diabetes Study. U.K. prospectivediabetes study 16. Overview of 6 years’ therapy oftype II diabetes: a progressive disease. UKProspective Diabetes Study Group. Diabetes 1995;44:1249–58.

244. Wilson JM, Junger G. Principles and practice ofscreening for disease. Public Health Paper Number34. Geneva: WHO; 1968.

245. Rose G. Strategy of prevention: lessons fromcardiovascular disease. BMJ (Clin Res Ed) 1981;282:1847–51.

246. Boule NG, Haddad E, Kenny GP, Wells GA, Sigal RJ. Effects of exercise on glycemic controland body mass in type 2 diabetes mellitus: a meta-analysis of controlled clinical trials. JAMA 2001;286:1218–27.

247. Williams R, Herman W, Kinmonth AL, Wareham N, editors. The evidence base for diabetescare. Chichester: Wiley; 2002.

248. Hamman RF. Prevention of type 2 diabetes. Theevidence base for diabetes care. Chichester: Wiley;2002. pp. 75–176.

249. Keen H, Jarrett RJ, Fuller JH. Tolbutamide andarterial disease in borderline diabetics. ExerptaMed Int Congr Ser 1968;312:588–602.

250. Keen H, Jarrett RJ, McCartney P. The ten-yearfollow-up of the Bedford survey (1962–1972):glucose tolerance and diabetes. Diabetologia 1982;22:73–8.

251. Paasikivi J. Long-term tolbutamide treatment aftermyocardial infarction. A clinical and biochemicalstudy of 178 patients without overt diabetes. ActaMed Scand Suppl 1970;507:1–82.

252. Sartor G, Schersten B, Carlstrom S, Melander A,Norden A, Persson G. Ten-year follow-up ofsubjects with impaired glucose tolerance:prevention of diabetes by tolbutamide and dietregulation. Diabetes 1980;29:41–9.

253. Engelgau M, Narayan KM. The evidence to screenfor type 2 diabetes mellitus. In Williams R,Herman W, Kinmonth AL, Wareham N, editors.The evidence base for diabetes care. Chichester: Wiley;2002. pp. 191–233.

254. Sherwin RS, Anderson RM, Buse JB, Chin MH,Eddy D, Fradkin J, et al. The prevention or delayof type 2 diabetes. Diabetes Care 2003;26(Suppl 1):S62–9.

255. American Diabetes Association. Screening for type2 diabetes. Diabetes Care 2004;27(Suppl 1):S11–14.

256. Davies MJ, Tringham JR, Troughton J, Khunti KK.Prevention of type 2 diabetes mellitus. A review ofthe evidence and its application in a UK setting.Diabetic Med 2004;21:403–14.

257. Williams R, Rapport F, Elwyn G, Lloyd B, Rance J,Belcher S. The prevention of type 2 diabetes:general practitioner and practice nurse opinions.Br J Gen Pract 2004;54:531–5.

258. Wylie G, Hungin AP, Neely J. Impaired glucosetolerance: qualitative and quantitative study ofgeneral practitioners’ knowledge and perceptions.BMJ 2002;324:1190.

259. Azen SP, Peters RK, Berkowitz K, Kjos S, Xiang A,Buchanan TA. TRIPOD (TRoglitazone In thePrevention Of Diabetes): a randomized, placebo-controlled trial of troglitazone in women withprior gestational diabetes mellitus. Control ClinTrials 1998;19:217–31.

260. Sjostrom L, Torgerson JS, Hauptman J, Boldrin M.XENDOS (XENical in the prevention of Diabetesin Obese Subjects): a landmark study. Presented atthe 9th International Congress on Obesity, SãoPaulo, 2002.

261. Gerstein HC. Emerging pharmacologicapproaches to diabetes prevention. Presented atthe 15th National Congress, Associazione MediciDiabetologi, Genoa, 18–21 May 2005.

262. Kanaya AM, Herrington D, Vittinghoff E, Lin F,Grady D, Bittner V, et al. Glycemic effects ofpostmenopausal hormone therapy: the Heart andEstrogen/Progestin Replacement Study. Arandomized, double-blind, placebo-controlledtrial. Ann Intern Med 2003;138:1–9.


105


263. Molitch ME, Fujimoto W, Hamman RF, Knowler WC. The diabetes prevention programand its global implications. J Am Soc Nephrol2003;14:S103–7.

264. Herman W, Brandle M, Zhang P, Williamson DF,Matulik MJ, Ratner RE, et al. Costs associated withthe primary prevention of type 2 diabetes mellitusin the diabetes prevention program. Diabetes Care2003;26:36–47.

265. Lindstrom J, Eriksson JG, Valle TT, Aunola S,Cepaitis Z, Hakumaki M, et al. Prevention ofdiabetes mellitus in subjects with impaired glucosetolerance in the Finnish diabetes prevention study:results from a randomized clinical trial. J Am SocNephrol 2003;14:S108–13.

266. Mensink M, Corpeleijn E, Feskens EJ, Kruijshoop M, Saris WH, de Bruin TW, et al. Studyon lifestyle-intervention and impaired glucosetolerance Maastricht (SLIM): design and screeningresults. Diabetes Res Clin Pract 2003;61:49–58.

267. Anand SS, Razak F, Vuksan V, Gerstein HC,Malmberg K, Yi Q, et al. Diagnostic strategies todetect glucose intolerance in a multiethnicpopulation. Diabetes Care 2003;26:290–6.

268. McEwan P, Williams R, Bergenheim K, Peters JR,Currie CJ. Economic benefits of delaying the onsetof type 2 diabetes. In preparation. 2005.

References

106

MEDLINE 1. exp *Diabetes Mellitus, Type 2/ 2. exp Mass Screening/ 3. (diabetes and screening).m_titl. 4. 1 and 2 5. 1 and 3 6. 4 or 5 7. limit 6 to (english language and yr="2000 –

June 2005")

EMBASE1. exp *Non Insulin Dependent Diabetes Mellitus/ 2. exp Mass Screening/ 3. (diabetes and screening).m_titl. 4. 1 and 2 5. 1 and 3 6. 4 or 5 7. limit 6 to (english language and yr="2000 –

June 2005")

The Cochrane Library 2005, Issue 2 – all sections #1 MeSH descriptor Diabetes Mellitus, Type 2

explode all trees in MeSH products #2 MeSH descriptor Mass Screening explode all

trees in MeSH products #3 (#1 AND #2) #4 screening in Record Title and diabetes in

Record Title in all products #5 (#3 OR #4)


107


Appendix 1

Search strategies

The UK NSC criteria for evaluating screeningprogrammes were adapted from the WHO

criteria published in 1966. The criteria arepublished by the NSC on their website(http://www.nsc.nhs.uk/pdfs/criteria.pdf).

The condition1. The condition should be an important health

problem.2. The epidemiology and natural history of the

condition, including development from latentto declared disease, should be adequatelyunderstood and there should be a detectablerisk factor, disease marker, latent period orearly symptomatic stage.

3. All the cost-effective primary preventioninterventions should have been implementedas far as practicable.

4. If the carriers of a mutation are identified as aresult of screening the natural history ofpeople with this status should be understood,including the psychological implications.

The test5. There should be a simple, safe, precise and

validated screening test.6. The distribution of test values in the target

population should be known and a suitablecut-off level defined and agreed.

7. The test should be acceptable to thepopulation.

8. There should be an agreed policy on thefurther diagnostic investigation of individualswith a positive test result and on the choicesavailable to those individuals.

9. If the test is for mutations the criteria used toselect the subset of mutations to be covered byscreening, if all possible mutations are notbeing tested, should be clearly set out.

The treatment10. There should be an effective treatment or

intervention for patients identified throughearly detection, with evidence of early

treatment leading to better outcomes than latetreatment.

11. There should be agreed evidence-basedpolicies covering which individuals should beoffered treatment and the appropriatetreatment to be offered.

12. Clinical management of the condition andpatient outcomes should be optimised in allhealthcare providers prior to participation ina screening programme.

The screening programme13. There should be evidence from high-quality

RCTs that the screening programme iseffective in reducing mortality or morbidity.

14. There should be evidence that the completescreening programme (test, diagnosticprocedures, treatment/ intervention) isclinically, socially and ethically acceptable tohealth professionals and the public.

15. The benefit from the screening programmeshould outweigh the physical andpsychological harm (caused by the test,diagnostic procedures and treatment).

16. The opportunity cost of the screeningprogramme (including testing, diagnosis andtreatment, administration, training andquality assurance) should be economicallybalanced in relation to expenditure onmedical care as a whole (i.e. value for money).

17. There should be a plan for managing andmonitoring the screening programme and anagreed set of quality assurance standards.

18. Adequate staffing and facilities for testing,diagnosis, treatment and programmemanagement should be available prior to thecommencement of the screening programme.

19. All other options for managing the conditionshould have been considered (e.g. improvingtreatment, providing other services), to ensurethat no more cost-effective intervention couldbe introduced or current interventionsincreased within the resources available.

20. Evidence-based information, explaining theconsequences of testing, investigation andtreatment, should be made available topotential participants to assist them in makingan informed choice.


109


Appendix 2

The NSC criteria

21. Public pressure for widening the eligibilitycriteria for reducing the screening interval,and for increasing the sensitivity of the testingprocess, should be anticipated. Decisionsabout these parameters should be scientificallyjustifiable to the public.

22. If screening is for a mutation, the programmeshould be acceptable to people identified ascarriers and to other family members.

Appendix 2

110

Early detection and primaryprevention of type 2 diabetesThis section starts by looking at the conclusions ofreviews published since the last NSC review of thetopic; then at primary studies published sincethen; and lastly at studies on the management of‘prediabetes’.

Background (reviews of studiespublished prior to 2002)Two of the reviews mentioned in this openingsection are chapters contained in Williams andcolleagues.247 They are not, in the true sense ofthe term, ‘systematic reviews’ in that they were notplanned and carried out according to guidelinessuch as those advocated by the CochraneCollaboration. Nevertheless, they provide areasonably comprehensive and critical coverage ofthe literature up until the date of publication ofthat text. The third is a consensus statementissued as part of the WHO (World HealthAuthority) diabetes programme. It was the resultof the deliberations of an expert group which metin May 2002 in Geneva.

‘Prevention of type 2 diabetes’, RF Hamman248

The work cited by Hamman is summarised inTable 23 and the studies that he cited in Table 24.His overall summary of this body of evidence wasas follows:

SUMMARYThe available data now provide a firm answer thattype 2 diabetes can be delayed or prevented in high-risk subjects. Trials currently underway shouldprovide substantial evidence about the impact ofspecific pharmacologic interventions on diabetesprevention.

With the current level of information, it is reasonableto recommend a programme of moderate levels ofphysical activity, weight maintenance or modestweight loss (for overweight people), and a low-fat,calorie-moderated diet for the positive effects this willhave on cardiovascular and other risks for largenumbers of people. In addition, subjects at high riskshould have specific risk factors for cardiovascular

disease treated (e.g. lipids, blood pressure). Therecent results from ACE-inhibitor trials suggest thatsuch agents may also lower diabetes risk as animportant benefit.

The phrase “… for the positive effects this willhave on cardiovascular and other risks for largenumbers of people.” is interesting in the light ofthe fact that very few of the studies identified hadcardiovascular risk factors as primary end-points.These included RCTs carried out as a follow-up tothe Bedford study249,250 in which tolbutamide(500 mg twice daily) with and without dietaryadvice was compared with placebo with andwithout dietary advice. No significant differenceswere found between any of the groups. Paasikivi251

investigated the same dose of tolbutamide versusplacebo and found that the difference in CVDmortality seen at 18 months did not persistbeyond that time. The Malmöhus study,252 whichmade similar comparisons, found no significantdifferences in CVD mortality.

‘The evidence to screen for type 2 diabetes’,Engelgau and Narayan253

Table 25 provides a summary of the results (costs inUS dollars and cost per QALY gained) derivedfrom the lifestyle simulation model described intheir reference 76 (CDC Diabetes Cost-Effectiveness Study Group, 1998).131 These resultsand scrutiny of other evidence as it was then led tothe following conclusions:

CONCLUSIONSThe effectiveness of diabetes screening has not beendirectly demonstrated. Indirect examination of thepotential benefits of screening using data from RCTsof treatment of diagnosed diabetes, observationalstudies and disease models lend some support to theidea that early improvement in glycaemic control mayhelp reduce the lifetime occurrence of microvasculardisease. There is little convincing evidence that therewill be macrovascular disease reductions. Thephysical, psychological and social effects of screeningand early diagnosis and treatment remain unclear.Thus, on balance, there is only modest evidence, atbest, supporting screening for type 2 diabetes. (Levelof evidence II-3; Strength of Recommendation B.)


111


Appendix 3

Management of impaired fasting glucose and impaired glucose tolerance

Appendix 3

112

TABLE 23 Search strategy and results

Search question:To identify existing guidelines for prediabetes/T2DM: prevention, screening, early identification

Keywords – thesaurus/free text/MeSH Prediabetes, prediabetic state, type 2 diabetes, non-insulin-dependent diabetes mellitus, prevention and control,primary prevention, screening, impaired glucose tolerance,impaired fasting tolerance

Publication types – guidelines, systematic reviews, press Studies, reviews or best evidence onlyreleases, conference proceedings, published statistics, etc.

LimitationsLanguage EnglishDates covered 1999 → Non-UK WorldwideOther limitations –

Sources of information

Evidence-based resources including Cochrane, Clinical Evidence and guidelines

Public health and health management and knowledge HMICbases (HMIC , NPHS Library catalogues

Clinical databases (MEDLINE, EMBASE, CINAHL) MEDLINE, EMBASE

Specialist databasesGrey Literature Sources (SIGLE, HMIC), specific Databases (Dissertation Abstracts, NRR Research Registers and ASSIA, CRD)

Internet – health gateways – OMNI

Specialised collections relevant to topic (Royal Colleges, professional Associations)

UK Health Departments

Affiliated organisations – HPA, HP

Wider NHS – NeLH, PHOs, HDA

Non-NHS – LAs, academia

European/international authoritative sources – EU, WHO

Expert opinionFrequently cited authors within specialist field, colleagues, E-bulletin boards (project work details)

Search resultsEMBASE

Search history Results

1 prediabetes.mp. or *Impaired Glucose Tolerance/ 1,1652 "PREVENTION AND CONTROL"/ or PRIMARY PREVENTION/ 5,3853 1 and 2 74 limit 3 to (human and english language and yr=1999–2004) 65 from 4 keep 1–3 36 *SCREENING/ 3,9077 Non Insulin Dependent Diabetes Mellitus/ or type 2 diabetes.mp. 32,8448 (7 or 1) and 6 219 8 2110 limit 9 to (human and english language and yr=1999–2004) 411 from 10 keep 1–2 2

continued


113


TABLE 23 Search strategy and results (cont’d)

MEDLINE


1 Mass Screening/ 20,5792 Prediabetic State/or prediabetes.mp. 3103 *DIABETES MELLITUS, TYPE II/pc [Prevention & Control] 6764 2 or 3 9725 *PRIMARY PREVENTION/ 1,6926 (2 or 3) and (5 or 1) 917 limit 6 to (human and english language and yr=1999–2004) 48

HMIC


1 prediabetes.mp. or exp PREDIABETIC STATE/ 22 exp DIABETES/ 8343 prevention.mp. or exp PREVENTIVE MEASURES/ 12,4894 exp SCREENING/ 2,7505 type 2.mp. 1246 1 and (3 or 4) 07 impaired glucose.mp. 17

TABLE 24 Studies cited by Hamman248

1. Intervention: combined lifestyle interventions (mainly dietary modification aimed at weight loss andincreased physical activity)

Type Study group Outcome No. of Referencesa

studies

Randomised trials Normal glucose tolerance Glucose 9 45–50, 52–54, 56, 63, 64, 67, 280Overweight, normal glucose tolerance Glucose 2 60, 62Hyperinsulinaemia Glucose 2 43, 44Glucose intolerance Diabetes 1 42IGT Diabetes 3 55, 59, 68, 69

Non-randomised Glucose intolerance Diabetes 1 73trials or cohort IGT Diabetes 1 57studies IGT Mortality 1 77

Family history of diabetes Diabetes 1 78,79

2. Intervention: physical activity alone


studies

Randomised trials None

Non-randomised Various (mostly mixtures of subjects Diabetes 19 76, 103–111, 113, 114, 116–122, trials or cohort with normal and abnormal glucose 125studies tolerance)

continued

Appendix 3

114

TABLE 24 Studies cited by Hamman248 (cont’d)

3. Intervention: dietary modification alone


studies

Randomised trials None

Non-randomised Nonetrials

Cohort studies Various (mostly mixtures of subjects Diabetes 20 5, 107, 114, 123, 140–143, 146–150, with normal and abnormal glucose 154, 155, 157–159, 161, 162, 238tolerance)

4. Intervention: pharmacological agents (with or without lifestyle modification)


studies

Tolbutamide vs Glucose intolerance Glucose, 5 181, 184–186, 190, 192, 194, 196, placebo diabetes or 197

mortality (one study190,192,194

– Bedford study – had CVD as outcome)

Tolbutamide vs Glucose intolerance Diabetes 1 187, 188phenformin vs placebo

Chlorpropamide vs Glucose intolerance Normalisation 1 195tolbutamide vs of glucose phenformin tolerance

Glicazide vs IGT, IFG or fasting hyperglycaemia Glucose 2 56, 199, 281placebo tolerance

Glibenclamide (G) Glucose intolerance Glucose 1 202vs Biguanide (B) vs toleranceG + B vs placebo

Phenformin vs Glucose intolerance Diabetes 1 204, 205placebo

Phenformin + diet Glucose intolerance Glucose 1 206tolerance

Metformin vs High waist:hip ratio; subjects with IGT Glucose 4 70–72, 207–210, 212placebo or IGT + overweight or obesity tolerance,

lipids, diabetes

Troglitazone vs Normal, overweight or IGT or GDM Glucose or 5 216, 218, 217, 221, 220placebo diabetes

Acarbose vs placebo IGT Diabetes 2 222, 223

Captopril vs Hypertensive patients Diabetes 2 224, 226‘conventional’ treatment (for hypertension)

Ramipril vs placebo Mixed normal and hypertensive Diabetes 1 225

Pravastatin vs Non-diabetic men with elevated Diabetes 1 240placebo cholesterol

a These are references numbered as in the original review,248 not in the present report.

‘Screening for type 2 diabetes. Report of a WHOand IDF meeting’The consensus report of this meeting (May 2002)led to the following conclusions andrecommendations:

CONCLUSIONS1. Screening for T2DM is important in terms of

individual health, day-to-day clinical practice andpublic health policy.

2. There is currently no direct evidence thatindividuals will benefit from the early detection ofT2DM through screening (direct evidence is thatfrom RCTs specifically designed to answer questionsrelated to early detection through screening).

3. Despite this lack of direct evidence, early detectionthrough screening is already taking place both byinviting individuals from the general population tocome forward for screening and, opportunistically,when individuals perceived to be at high risk ofdeveloping diabetes attend for healthcare (usuallyprimary healthcare) for other reasons.

4. These activities present opportunities for collectingobservational data, which, although no substitutefor direct RCT evidence, can provide important,circumstantial evidence.

5. Following a demonstration of any benefits ofscreening, the most important epidemiologicalconsiderations determining whether to screen inany given population will be (a) the prevalence ofundiagnosed T2DM in a population and (b) thedegree to which T2DM is associated with risk ofCVD and other important health outcomes in thatpopulation.

6. The most important health systems considerationswill be the capacity (a) to carry out the screening,(b) to provide effective healthcare for those whoscreen positive, (c) to address the psycho-socialneeds of all those who undergo screening and (d)to implement effective prevention in those who,although not confirmed to have diabetes at thetime, are at high risk of its future development.

7. The most important economic considerations are(a) the cost of early detection to the health systemand to the individual, (b) the extra costs of treatmentfollowing early detection and (c) the relative cost-effectiveness of early detection compared with thatof improving the care of clinically detected (asopposed to screen-detected) cases.

8. Screening for T2DM is a dynamic topic in whichnew evidence will become available and furtherconsiderations will arise over time.

RECOMMENDATIONS1. Health authorities and professional organisations

should formulate policies on screening for T2DMeven if the policy is that screening is not currentlyto be advocated.

2. There is an urgent need for direct RCT evidenceon the effects of early detection of T2DM throughscreening. (Such evidence should include healthoutcomes related to diabetes CVD, psychosocialoutcomes and economic considerations forindividuals, health systems and the wider society.Although RCTs directed to answering thesequestions may be costly and logistically difficult,there is, in the current state of knowledge, noethical reason why they should not be undertaken.)


115


TABLE 25 Effectiveness of screening for type 2 diabetes from lifestyle simulation models

Lifetime cumulative incidence (%) LYG QALYs Cost (US$)

ESRD Blindness LEAgained

Per LYG Per QALY gained

Total population age ��25 yearsWithout screening 3.5 9.1 4.6With screening 2.6 5.9 3.6Absolute risk reduction 0.9 3.2 1.0

NNT 111 31 100 0.02 0.08 236,449 56,649

Total population age 25–34 yearsWithout screening 19.2 32.4 19.0With screening 15.9 25.9 16.0Absolute risk reduction 3.3 6.5 3.0

NNT 30 15 33 0.12 0.35 35,768 13,376

Total population age ��65 yearsWithout screening 0.3 1.7 1.0With screening 0.2 1.1 0.7Absolute risk reduction 0.1 0.5 0.3

NTT 1000 200 333 0.00 0.01 NA 575,241

NA, not applicable; NNT, number-needed-to-treat.Adapted from Engelgau and Narayan (2002).253

3. Since the results of such RCTs will not be availablefor some time (if ever), there is also an urgentneed to develop a framework (or model) whichwould permit countries to evaluate the cost-effectiveness of the early detection of diabetescompared with other preventive and therapeuticinterventions.

4. The testing of apparently unaffected individuals atincreased risk of having diabetes when theseindividuals attend for healthcare for other reasons(sometimes called ‘opportunistic screening’) maybe justified provided that (a) the reasons fortesting are adequately explained to the individual,(b) the health system has the capacity for theclinical management of those who screen positive, (c) methods with adequate sensitivity andspecificity are available, (d) the psychosocial needsof those who screen positive and those who screennegative can be met and (e) the health system canimplement effective preventive strategies for thoseconfirmed to be at high risk for the futuredevelopment of diabetes.

5. If such opportunistic screening is being advocated,then this should be carried out according to apolicy which should (a) be clear and relevant in itsaims and objectives, (b) be based as far as possibleon sound evidence, (c) take into account theepidemiology of T2DM and related CVD risk inthe population and (d) be sensitive to competinglocal health priorities.

6. Where screening is already taking place, formalevaluation should be integral to these activities.The results of such evaluations could contribute tothe general assessment of the value of earlydetection and should, when appropriate,contribute to the modification or curtailment ofthe local activities being evaluated.

7. Given the dynamic nature of this topic, policies forscreening for T2DM must be reviewed from timeto time as new evidence accumulates.

Early detection and primary preventionof type 2 diabetes (review of workpublished from 2002 onwards withparticular emphasis on publishedguidelines for the detection of‘prediabetes’)Search strategyThe complete search strategy is given in Table 23.In brief, the search question was:

To identify existing guidelines for prediabetes/T2DM:prevention, screening, early detection

and was of publications, available in English andpublished in the major databases such asCochrane, HMIC, MEDLINE and EMBASE. The search was for items published from 1999onwards. This summary concentrates on thosepublished from 2002 onwards.

Summary of findingsReviews and position statementsThe ADA, in its review of the prevention or delay ofT2DM,254 concluded that there was substantialevidence that T2DM could be prevented or delayedbut commented that it was “not yet known” whetherthe interventions (lifestyle change and/orpharmaceutical interventions) that had been shownto be successful “will cost-effectively reduce themorbidity and mortality associated with diabetes”.However, it did comment that lifestylemodifications which were specifically directedtowards modest weight loss and increased physicalactivity were likely to have additional health benefitsover and above their specific effects in relation totype diabetes. It recommended “some” of thefollowing prevention policies (listed in Table 2 ofthe original publication – strength of the evidencegiven as A–E in descending order of certainty):

● Individuals at high risk of developing diabetesneed to become aware of the benefits of modestweight loss and participating in regular physicalactivity (A).

● Screening based on current screeningguidelines for diabetes [(reference 49 in theoriginal – Expert Committee on the Diagnosisand Classification of Diabetes Mellitus (2003)],men and women �45 years of age, particularlythose with BMI �25 kg/m2, are candidates forscreening to detect prediabetes (IFG or IGT).Screening should be considered in youngerindividuals with a BMI �25 kg/m2 who haveadditional risk factors [given in Table 3 of theoriginal] (B).

● In individuals with normoglycaemia,rescreening at 3-year intervals is reasonable (C).

● How to screen: screening should be carried outonly as part of a healthcare office [sic] visit.Either an FPG test or a 2-hour OGTT (75-gglucose load) is appropriate, and positive testresults should be confirmed on another day (B).

● Intervention strategy: patients with prediabetes(IFG or IGT) should be given counselling onweight loss and also instruction for increasingphysical activity (A).

● Follow-up counselling appears important forsuccess (B).

● Monitoring for the development of diabetesshould be performed every 1–2 years (E).

● Close attention should be given to, andappropriate treatment given for, other CVD riskfactors (e.g. tobacco use, hypertension,dyslipidaemia (A).

● Drug therapy should not be routinely used toprevent diabetes until more information isknown about its cost-effectiveness (E).

Appendix 3

116

Among the eight recommendations for furtherresearch, the most relevant to this review is thefirst mentioned [particularly the section in (our)italics]:

What is the cost-effectiveness of a DPP-like lifestyleintervention? Are there more cost-effective strategies, and how would they affect the morbidity and mortality associated with diabetes? The issue ofwhether the preventive or delaying strategies areeffective in preventing or delaying CVD is not directly tackled in this review, thus recognising the relative lack of evidence relating to the effect on CVD risk of interventions shown to be effective in preventing or delaying the onset of diabetes itself.

A further position statement issued by the ADA255

concludes that “there is sufficient indirect evidenceto justify opportunistic screening in a clinicalsetting of individuals at high risk”. The screeningtest advocated for this opportunistic screening isthe FPG. Again, no direct reference is made to themanagement of prediabetes in relation toreduction of CVD risk.

Davies and colleagues256 reviewed the evidencewith a specific focus on its relevance to the UKsetting. They point out that none of the availableRCTs have been carried out in the UK andquestion whether their findings can be replicated in the UK given possible difference in the pathophysiology of IGT (which, frankly,seems unlikely), a lower rate of progression todiabetes (which is possible) and low levels ofawareness of the importance of IGT (which are documented257,258). The authors conclude that it is unlikely that a study of the effects ondiabetes incidence or CVD outcomes of ‘upstream’ interventions (i.e. primary preventionmeasures in total populations) will ever beconducted.

Key publications 2002 onwardsIn addition to the studies listed in Table 24, keypublications from 2002 onwards are summarisedin Table 26.

Studies specifically dealing withmanagement of ‘prediabetes’In relation to reducing diabetes riskDavies and colleagues256 present a comparison, interms of number-needed-to-treat of theinterventions tested in the Da Qing,88 TRIPOD,259

Diabetes Prevention Programme (DPP),90 DiabetesPrevention Study (DPS),89 STOP-NIDDM169 andXENDOS260 trials. These are summarised in Table 27.

Further pharmacological interventions aresummarised by Gerstein.261 These include:

● Metformin: as in DPP mentioned above.● Acarbose: as in STOP-NIDDM mentioned above.● Orlistat: as in XENDOS mentioned above.● Thiazolidinediones: troglitazome (as in DPP,

withdrawn as a result of adverse effects) andTRIPOD.

● ACEIs: ramipril (as in HOPE); lisinopril (as inALLHAT); captopril (as in CAPP); enalapril (asin D-SOLVD); and others.

● Insulin and insulin secretagogues: of theoreticalbenefit.

● Statins and fibrates: pravastatin (as inWOSCOPS), significant reduction in conversionto diabetes; no effect found for atorvostatin(ASCOT), simvastatin (Heart Protection Study),pravastatin (LIPID).

● Oestrogen: reduction in conversion to diabetesfound by Kanaya and colleagues.262 Adverseeffects on CVD risk render this an unsuitabletherapeutic approach.

In relation to reducing cardiovascular disease riskThe only diabetes prevention RCT which hasreported a positive effect on CVD risk is theSTOP-NIDDM trial of acarbose. Chiasson andcolleagues169 reported a significant reduction innew cases of hypertension, MI and any CVD eventin those taking acarbose. Given the fact that 30%of those randomised to the active interventionstopped taking the drug and this analysis wasbased on intention-to-treat, the effect of acarboseon CVD incidence may well be even greater thanthat described. However, this unexpectedly largeeffect needs to be replicated in other trials and hasnot been reported as a result of lifestyle changesor in relation to other pharmacologicalinterventions.

Ongoing work with the DiabetesForcaster™ model (McEwan and colleagues, in preparation)provides numerical estimates of the effects on CVD morbidity and mortality of delaying the onset of type 2 diabetes. Shown in Table 28are the expected cardiovascular events, totalsimulation costs and QALYs obtained by runningthe model over a 20-year time horizon for twoscenarios: scenario 1 provides a linear increase inrisk over the time after the onset of diabeteswhereas scenario 2 provides a stepwise change inrisk at diagnosis. For scenario 1, a 10-year delay inthe onset of diabetes roughly halves the number ofdeaths from CVD, reduces direct healthcare coststo around one-third and provides an average gainin QALYs per person of 3.31.


117


Appendix 3

118 TAB

LE 2

6Ke

y st

udie

s pu

blish

ed fr

om 2

002

onw

ards

Gro

upSt

udy/

leve

l T

itle

Aut

hors

Year

Sour

cePo

pula

tion

In

terv

enti

on

Out

com

esFo

llow

-up/

Com

men

ts:

Trea

tmen

t ev

iden

ceC

ount

ryst

udie

dgi

ven

anal

ysis

blin

ding

, ef

fect

Mul

ti-c

entr

edgr

oups

tr

eate

d eq

ually

Inte

rven

tions

in IG

Tpo

pula

tion

RCT

II

The

Dia

bete

sPr

even

tion

Prog

ram

and

its g

loba

lim

plic

atio

ns

USA

Mul

ti-ce

ntre

d–

5

Mol

tich

et a

l.263

2003

J Am

Soc

Nep

hrol

14:S

103–

7

3234

�25

yea

rs,

mea

n51

year

sBM

I �24

,m

ean

34kg

/m2

IGT

68%

wom

en,

45%

BM

ethn

icgr

oups

1. L

ifest

yle

goal

s: 7

%w

eigh

t lo

ss,

150

min

utes

’ex

erci

se/w

eek

Beha

viou

rm

odifi

catio

nin

divi

dual

and

grou

p se

ssio

ns

2. M

etfo

rmin

3. P

lace

bo

Dev

elop

ing

DM

, ann

ual

OG

TT

Av.

3.8

yea

rsIT

T a

naly

sisD

oubl

e-bl

inde

d fo

rm

etfo

rmin

and

plac

ebo

Gro

ups

not

trea

ted

equa

lly, i

.e.

lifes

tyle

was

inte

nsiv

e

Av.

wei

ght

loss

5.6

kg in

Life

styl

egr

oup

Inci

denc

e D

MLi

fest

yle

58%

(95%

CI 4

8 to

66)

inci

denc

e re

duce

dM

etfo

rmin

31%

low

er (9

5% C

I 17

to 4

3) c

ompa

red

with

pla

cebo

Life

styl

e ef

fect

see

nac

ross

eth

nic

grou

ps

Trea

ting

6.9

peop

lefo

r 3

year

s,pr

even

ts 1

cas

ew

ith li

fest

yle

13.9

with

met

form

in

Econ

omic

s of

prim

ary

prev

entio

n

Cos

tsas

soci

ated

with

the

prim

ary

prev

entio

n of

T2D

M in

DPP

USA

Am

eric

anD

iabe

tes

Ass

ocia

tion26

4

2003

Dia

bete

sCa

re26

:37–

47

To d

escr

ibe

cost

s of

DPP

Dire

ctm

edic

alco

sts,

dire

ctno

n-m

edic

alco

sts

and

indi

rect

cost

s,pl

aceb

o,m

etfo

rmin

and

inte

nsiv

elif

esty

lein

terv

entio

ns

Dire

ct m

edic

alco

sts,

labo

rato

ryte

sts

to id

entif

y 1

pers

on w

ith IG

T$1

39

Ove

r 3

year

s, c

osts

per

part

icip

ant

rela

tive

to p

lace

bow

as $

2191

inm

etfo

rmin

, $24

12in

life

styl

e fo

r a

heal

thca

re s

yste

m

cont

inue

d


119


TAB

LE 2

6Ke

y st

udie

s pu

blish

ed fr

om 2

002

onw

ards

(co

nt’d

)

Gro

upSt

udy/

leve

l T

itle

Aut

hors

Year

Sour

cePo

pula

tion

In

terv

enti

on

Out

com

esFo

llow

-up/

Com

men

ts:

Trea

tmen

t ev

iden

ceC

ount

ryst

udie

dgi

ven

anal

ysis

blin

ding

, ef

fect

Mul

ti-c

entr

edgr

oups

tr

eate

d eq

ually

Con

sider

ed m

odes

tin

crem

enta

l cos

tsbu

t nee

d to

eval

uate

cos

tsre

lativ

e to

hea

lthbe

nefit

s to

dete

rmin

e re

alva

lue

Inte

rven

tions

in IG

Tpo

pula

tion

RCT

II

Prev

entio

n of

diab

etes

mel

litus

insu

bjec

ts w

ithim

paire

dgl

ucos

eto

lera

nce

in

the

Finn

ishD

iabe

tes

Prev

entio

nSt

udy:

res

ults

from

ara

ndom

ised

clin

ical

tria

lPr

even

tion

ofTy

pe 2

di

abet

esm

ellit

us b

ych

ange

s in

lifes

tyle

am

ong

subj

ects

with

impa

ired

gluc

ose

tole

ranc

eFi

nlan

dM

ulti-

cent

res

5

Lind

stro

m

et a

l.265

Tuom

ileht

oet

al.89

2003

2001

J Am

Soc

Nep

hrol

14:S

108–

13

N E

ngl J

Med

, 344

:13

43–5

0

522

men

and

wom

en,

67%

wom

en.

Av.

age

55ye

ars.

Av.

BM

I31

kg/m

2 .IG

T

1. L

ifest

yle.

N=

265,

die

t(r

educ

ed fa

tin

take

to

30%

ener

gyin

take

),ex

erci

se (o

ver

4ho

urs

per

wee

k), w

eigh

tlo

ss (a

t le

ast

5%) i

n 7

sess

ions

inye

ar 1

, the

n 4

times

per

year

, inc

lude

dnu

triti

onist

.Fo

od d

iarie

s

2. C

ontr

ol.

N=

257.

Ann

ual

info

rmat

ion

–ve

rbal

and

writ

ten

Inci

denc

e of

DM

by

WH

Ocr

iteria

.C

hang

es in

wei

ght,

diet

,ex

erci

se

3.2

year

s

90%

follo

w-

up

Ach

ieve

d4.

2kg

wei

ght

loss

in inte

rven

tion

grou

p

Peop

le w

ithbe

stco

mpl

ianc

eto

life

styl

ech

ange

had

grea

test

redu

ctio

n in

inci

denc

e

Life

styl

e in

cide

nce

rate

= 3

.2/1

00 P

Yve

rsus

con

trol

=7.

8/10

0 PY

Haz

ard

ratio

= 0

.4(0

.3–0

.7) f

orin

cide

nt D

M

ARR

= 4

.6/1

00 P

Y

Risk

of d

iabe

tes

redu

ced

by 5

8% in

inte

rven

tion

grou

pov

er t

he t

rial

(p<

0.00

1)

Also

impr

oved

CV

risk

fact

ors

–re

duce

d T

G a

ndbl

ood

pres

sure

cont

inue

d

Appendix 3

120 TAB

LE 2

6Ke

y st

udie

s pu

blish

ed fr

om 2

002

onw

ards

(co

nt’d

)

Gro

upSt

udy/

leve

l T

itle

Aut

hors

Year

Sour

cePo

pula

tion

In

terv

enti

on

Out

com

esFo

llow

-up/

Com

men

ts:

Trea

tmen

t ev

iden

ceC

ount

ryst

udie

dgi

ven

anal

ysis

blin

ding

, ef

fect

Mul

ti-c

entr

edgr

oups

tr

eate

d eq

ually

Scre

enin

g IG

Tin

hig

h-ris

kpo

pula

tion

Inte

rven

tions

in IG

Tpo

pula

tion

tofo

llow

Obs

erva

tiona

l,cr

oss-

sect

iona

l

IV RCT

II

Stud

y on

lifes

tyle

-in

terv

entio

nan

d IG

TM

aast

rict

(SLI

M):

desig

nan

d sc

reen

ing

resu

lts

The

Net

herla

nds

3 ce

ntre

s

Men

sink

et a

l.266

2003

Dia

bete

s Re

sCl

in P

ract

61:4

9–58

2820

age

dov

er 4

0+,

BMI

>25

kg/m

2

or fa

mily

hist

ory

DM

scre

ened

with

OG

TT

Iden

tifie

d22

6 (8

.3%

)T

2DM

, 215

(7.9

%) I

FG,

385

(14.

2%)

IGT

2nd

OG

TT

144

with

IGT

ent

ered

into

stu

dy

1. L

ifest

yle-

diet

ary

(aim

5–7%

wei

ght

loss

), ex

erci

se(a

im30

min

utes

�5

per

wee

km

oder

ate

phys

ical

activ

ity),

free

acce

ss t

otr

aini

ngpr

ogra

mm

e,in

divi

dual

advi

ce

2. C

ontr

ol –

oral

and

writ

ten

info

rmat

ion

rebe

nefit

s of

wei

ght

loss

and

exer

cise

.N

o ad

ditio

nal

appo

intm

ents

Resu

lts/y

ield

scre

enin

g61

08 in

vite

dov

er 4

0+,

BMI

>25

kg/m

2

or fa

mily

hist

ory

DM

3288

(53.

8%)

non-

resp

onde

rs,

mea

n ag

elo

wer

55.7

±0.

1vs

56.

8±

0.1

year

s,ge

nder

equi

vale

nt

1st

OG

TT

in28

20:

226

(8.3

%)

T2D

M, 2

15(7

.9%

) IFG

,38

5 (1

4.2%

)IG

T

379

for

2nd

OG

TT

In m

en a

nd w

omen

with

BM

I ove

r30

kg/m

2 , DM

3�

and

IFG

/IGT

2�

mor

e pr

eval

ent

than

tha

n if

BMI

belo

w 2

7kg

/m2

Prev

alen

ce o

fne

wly

dia

gnos

edD

M 2

�hi

gher

inm

en t

han

wom

enfo

r ea

ch B

MI/a

gegr

oup.

Posit

ive

rela

tion

betw

een

age

and

prev

alen

ce D

M a

ndIG

T, IF

G h

ighe

st55

–59

year

sSt

rong

upw

ard

tren

d fo

r ag

e an

dBM

I was

see

n fr

omN

GT

to

T2D

Mw

ith IG

T a

nd IF

Gin

bet

wee

nIG

T m

ore

prev

alen

t th

an IF

G,

limite

d ov

erla

p.Ta

bles

of y

ield

give

n

cont

inue

d


121


TAB

LE 2

6Ke

y st

udie

s pu

blish

ed fr

om 2

002

onw

ards

(co

nt’d

)

Gro

upSt

udy/

leve

l T

itle

Aut

hors

Year

Sour

cePo

pula

tion

In

terv

enti

on

Out

com

esFo

llow

-up/

Com

men

ts:

Trea

tmen

t ev

iden

ceC

ount

ryst

udie

dgi

ven

anal

ysis

blin

ding

, ef

fect

Mul

ti-c

entr

edgr

oups

tr

eate

d eq

ually

Scre

enin

g fo

rT

2DM

and

IFG

Obs

erva

tiona

lcr

oss-

sect

iona

lsu

rvey

IV

A s

impl

epr

agm

atic

syst

em fo

rde

tect

ing

new

case

s of

typ

e2

diab

etes

and

impa

ired

fast

ing

glyc

aem

ia in

prim

ary

care

UK

Gre

aves

et

al.11

220

04Fa

m P

ract

21:5

7–62

16 p

ract

ices

in S

omer

set

with

prev

alen

tD

M 2

.36%

100

patie

nts

from

eac

hpr

actic

e, 2

5fo

r ea

ch o

f 4gr

oups

step

ped

byag

e an

d BM

Icr

iteria

1287

patie

nts

recr

uite

d,39

.5%

mal

e

Com

pute

rised

sear

chin

g of

rout

inel

yco

llect

ed d

ata

as s

tart

ing

poin

t fo

r a

targ

eted

scre

enin

gpr

ogra

mm

e.

Sele

cted

patie

nts

rece

ived

invi

tatio

n an

dca

ll to

att

end

for

test

.W

eigh

t,he

ight

, age

,FB

GIf

first

IFG

abno

rmal

,re

peat

ed fo

rdx

Resu

lts y

ield

of s

cree

ning

RR t

o in

vite

60.6

% (9

5%C

I 55.

7 to

65.6

) lik

ebr

east

can

cer

scre

enin

g

BMI d

ata

avai

labl

e fo

r76

.8%

of

over

-50

popu

latio

n.

1287

atte

nded

scre

enin

gcl

inic

, 199

abno

rmal

FBG

, 199

atte

nded

2nd

test

55 (4

.3%

atte

nder

s)T

2DM

93 (7

.2%

atte

nder

s)IF

GN

o ge

nder

diffe

renc

es

For

thos

e at

tend

ing

NN

T t

o de

tect

IFG

/T2D

M in

ove

r70

s, B

MI

�33

kg/m

2w

as7.

7, 1

6.5

if on

ein

clud

es n

on-

atte

nder

sN

NT

if a

ge�

50ye

ars

and

BMI

�27

kg/m

212

.8bu

t gr

eate

rpr

opor

tion

ofpo

pula

tion-

scre

enin

g tr

ade-

off

Aut

hors

favo

urw

ide

scre

enin

gcr

iteria

for

over

-50

s, B

MI

�27

kg/m

2 cont

inue

d

Appendix 3

122 TAB

LE 2

6Ke

y st

udie

s pu

blish

ed fr

om 2

002

onw

ards

(co

nt’d

)

Gro

upSt

udy/

leve

l T

itle

Aut

hors

Year

Sour

cePo

pula

tion

In

terv

enti

on

Out

com

esFo

llow

-up/

Com

men

ts:

Trea

tmen

t ev

iden

ceC

ount

ryst

udie

dgi

ven

anal

ysis

blin

ding

, ef

fect

Mul

ti-c

entr

edgr

oups

tr

eate

d eq

ually

Eval

uatio

n of

stra

tegi

es t

osc

reen

pred

iabe

tes

Eval

uatio

nC

osts

of

scre

enin

g fo

rpr

edia

bete

sam

ong

US

adul

ts

Zha

ng

et a

l.204

2003

Dia

bete

sCa

re26

:253

6–42

Age

47–7

4ye

ars

visit

edhe

alth

care

prov

ider

�1

in p

revi

ous

year

with

no

prev

ious

diag

nosis

T2D

M=

54.4

mill

ion

With

BM

I�

25kg

/m2

elig

ible

popu

latio

n =

37.4

mill

ion

5 st

rate

gies

:1.

all

OG

TT

2. a

ll FB

G if

posit

ive

but

no IF

G t

hen

OG

TT

3. H

bA1c

ifpo

sitiv

eO

GT

T4.

CBG

ifpo

sitiv

eO

GT

T5.

risk

asse

ssm

ent

scor

e th

enO

GT

T

Effe

ctiv

enes

sof

eac

hst

rate

gyde

fined

by

prop

ortio

nof

cas

esid

entif

ied.

Ass

umed

OG

TT

and

FBG

100

%se

nsiti

ve fo

rde

tect

ing

IGT,

IFG

and

T2D

M

Eval

uate

d if

limite

dst

rate

gies

to

thos

e ag

ed45

–74

year

s,BM

I�

25kg

/m2

Med

ical

and

non-

med

ical

cost

s

Wea

knes

ses:

no-o

neun

der

45 o

rov

er75

year

s,fa

mily

hist

ory

not

incl

uded

Cos

ts o

fun

dete

cted

case

s no

tin

clud

ed

Prop

ortio

nid

entif

ied

of p

re-

DM

and

T2D

Mra

nged

69–

100%

Cos

t pe

r ca

seid

entif

ied

rang

ed$1

76–2

36 fr

omsin

gle-

paye

rpe

rspe

ctiv

e

Test

ing

all w

ithO

GT

T m

ost

effe

ctiv

e, C

BG a

ndris

k as

sess

men

tm

ost

effic

ient

If pe

ople

muc

h le

ssw

illin

g to

tak

e an

OG

TT

tha

n an

FPG

test

, the

n FB

Gst

rate

gy m

ost

effe

ctiv

e

Trad

e-of

f bet

wee

nef

fect

iven

ess

and

effic

ienc

yde

pend

ing

onw

heth

er a

im is

to

iden

tify

mor

e ca

ses

or p

ursu

e lo

wes

tco

st p

er c

ase

Scre

enin

gov

erw

eigh

tin

divi

dual

s ha

dlo

wer

cos

t pe

r ca

se

cont

inue

d


123


TAB

LE 2

6Ke

y st

udie

s pu

blish

ed fr

om 2

002

onw

ards

(co

nt’d

)

Gro

upSt

udy/

leve

l T

itle

Aut

hors

Year

Sour

cePo

pula

tion

In

terv

enti

on

Out

com

esFo

llow

-up/

Com

men

ts:

Trea

tmen

t ev

iden

ceC

ount

ryst

udie

dgi

ven

anal

ysis

blin

ding

, ef

fect

Mul

ti-c

entr

edgr

oups

tr

eate

d eq

ually

Det

ectio

n IG

Tin

eth

nic

popu

latio

n

IVD

iagn

ostic

stra

tegi

es t

ode

tect

glu

cose

into

lera

nce

ina

mul

tieth

nic

popu

latio

n

Can

ada

Ana

nd

et a

l.267

2003

Dia

bete

sCa

re26

:290

–6

936

Can

adia

ns o

fSo

uth

Asia

n,C

hine

se a

ndEu

rope

ande

scen

t

Usin

g RO

Ccu

rves

to

dete

rmin

ese

nsiti

vitie

s/sp

ecifi

citie

san

d cu

t-of

fva

lues

Sens

itivi

ty o

f AD

Acr

iteria

to

diag

nosis

T2D

M is

low

and

ther

e is

subs

tant

ial

varia

tion

betw

een

ethn

ic g

roup

s

FBG

and

HbA

1cca

nbe

use

d to

iden

tify

thos

e w

ith D

M b

utO

GT

T n

eede

d fo

rIG

T

Prev

alen

cest

udy

Cro

ss-

sect

iona

l

IV

Hig

hpr

eval

ence

of

undi

agno

sed

DM

inSo

uthe

rnG

erm

any:

targ

etpo

pula

tion

for

effic

ient

scre

enin

g. T

heKo

ra s

tudy

Ger

man

y

Rath

man

et

al.20

620

03D

iabe

tolo

gia

46:1

82–9

OG

TT

to

rand

omsa

mpl

e 13

53su

bjec

ts in

the

KORA

stud

y 20

00

Age

55–7

4ye

ars

Prev

alen

ces

(WH

O 1

999)

and

NN

TS

(scr

een)

to

iden

tify

1pe

rson

calc

ulat

ed

62%

agr

eed

to part

icip

ate;

‘hea

lthy

part

icip

ant

effe

ct’

Prev

alen

ces:

Kno

wn

DM

9%

Unk

now

n D

M9.

7%, I

GT

16.

8%,

IFG

9.8

% in

men

7.9,

6.9

, 16.

0,4.

5% in

wom

en

NN

T in

men

+2.

9if

have

abd

omin

alob

esity

, par

enta

lD

M a

ndhy

pert

ensio

n

Car

diov

ascu

lar

risk

fact

ors

wor

sen

amon

g gl

ucos

eto

lera

nce

cate

gorie

s? Ju

stify

scre

enin

g st

rate

gies

DM

, dia

bete

s m

ellit

us; I

TT,

inte

ntio

n-to

-tre

at; P

Y, p

erso

n-ye

ar; R

OC

, rec

eive

r op

erat

ing

char

acte

ristic

; TG

, trig

lyce

rides

.

Conclusions1. There is still no evidence from RCTs specifically

designed to show whether early detection throughscreening is worthwhile and that individuals willbenefit from the early detection of T2DM. Thedefinitive RCT is unlikely to be undertaken.

2. Given that opportunistic screening andscreening by invitation are happening inprimary care in the UK, standards andguidelines are required in order to achieveequitable access to good practice.

3. Screening activities, whether opportunistic orby invitation, may, depending on cut-offs used,

identify more people with IGT and/or IFG thanpeople with previously undiagnosed diabetes.

4. There is convincing RCT evidence thatinterventions relating to lifestyle change andsome pharmacological treatments (as yetunlicensed for this use) are effective in delayingtransition from IGT to diabetes, particularly inpeople who are overweight or obese.

5. It is, as yet, an unanswered question as towhether the efficacy of these interventions inthe RCT context can be translated intoeffectiveness (and cost-effectiveness) ineveryday clinical practice.

Appendix 3

124

TABLE 27 Number-needed-to-treat (NNT) to avoid one case of progression from IGT to T2DM in published trials

Study Cumulative incidence of Intervention NNT Duration T2DM vs placebo (%) (years)

Da Qing (Pan et al., 1997)88 66 vs 44 Lifestyle 4.5 6

TRIPOD (Azen et al., 1998)259 30 vs 14 Troglitazone 6 2.5

Diabetes Prevention Program 29 vs 14 Lifestyle 7 3(DPP Research Group, 2002)90

Diabetes Prevention Study 42 vs 32 Lifestyle 8 4(Tuomilehto et al., 2001)89

STOP-NIDDM (Chiasson et al., 2003)169 42 vs 32 Acarbose 11 4

Diabetes Prevention Program 29 vs 22 Metformin 14 3(DPP Research Group, 2002)90

XENDOS (Sjostrom et al., 2002)260 9 vs 6 Xenical 36 3

Adapted from Davies and colleagues (2004).256

TABLE 28 Summary of cardiovascular results, total costs and QALYs obtained over a 20-year time horizon running the model with nodelay in diabetes and 1, 3, 5 and 10 years’ delay for scenarios 1 and 2: values shown are expected cumulative events (when running themodel with no delay), and expected events/costs avoided and gain in QALYs

Events Events avoided (or QALYs per person gained)

No 1-year delay 3-year delay 5-year delay 10-year delaydelay

Scenario Scenario Scenario Scenario Scenario Scenario Scenario Scenario 1 2 1 2 1 2 1 2

CHD 421 24 15 44 61 81 95 121 181Stroke 269 18 20 49 56 88 84 151 159CVD death 271 17 10 40 47 73 70 133 131

Costs and QALYs Effect on costs and QALYs

Costs per subject:Non-discounted (£) 18,303 –£703.48 –839.69 –2119 –2237 –3441 –3616 –6026 –6604Discounted (£) 13,145 –605.91 –687.02 –1722 –1823 –2717 –2869 –4547 –4978

QALYs per subject:Non-discounted 9.74 0.41 0.37 1.13 1.19 1.85 1.85 3.17 3.31Discounted 7.53 0.34 0.32 0.93 0.98 1.48 1.49 2.43 2.53

Adapted from McEwan and colleagues (in preparation).268

6. It is also an unanswered question as to whetherdelaying transition to diabetes delays orprevents the cardiovascular outcomesassociated with diabetes.

7. It is highly likely that the effectiveness and cost-effectiveness of early identification of diabetes,IGT and IFG will be enhanced by combining

this with the early identification andmanagement of CVD risk.

8. A care pathway for the early detection andprevention of T2DM and associated CVD riskbased on the best evidence available (or at leastconsensus) is required to guide primary careprofessionals and others on current and futurepractice.


125



141

Health Technology AssessmentProgramme

Prioritisation Strategy GroupMembers

Chair,Professor Tom Walley, Director, NHS HTA Programme,Department of Pharmacology &Therapeutics,University of Liverpool

Professor Bruce Campbell,Consultant Vascular & GeneralSurgeon, Royal Devon & ExeterHospital

Professor Robin E Ferner,Consultant Physician andDirector, West Midlands Centrefor Adverse Drug Reactions,City Hospital NHS Trust,Birmingham

Dr Edmund Jessop, MedicalAdviser, National Specialist,Commissioning Advisory Group(NSCAG), Department ofHealth, London

Professor Jon Nicholl, Director,Medical Care Research Unit,University of Sheffield, School of Health and Related Research

Dr Ron Zimmern, Director,Public Health Genetics Unit,Strangeways ResearchLaboratories, Cambridge

Director, Professor Tom Walley, Director, NHS HTA Programme,Department of Pharmacology &Therapeutics,University of Liverpool

Deputy Director, Professor Jon Nicholl,Director, Medical Care ResearchUnit, University of Sheffield,School of Health and RelatedResearch

HTA Commissioning BoardMembers

Programme Director, Professor Tom Walley, Director, NHS HTA Programme,Department of Pharmacology &Therapeutics,University of Liverpool

Chair,Professor Jon Nicholl,Director, Medical Care ResearchUnit, University of Sheffield,School of Health and RelatedResearch

Deputy Chair, Dr Andrew Farmer, University Lecturer in GeneralPractice, Department of Primary Health Care, University of Oxford

Dr Jeffrey Aronson,Reader in ClinicalPharmacology, Department ofClinical Pharmacology,Radcliffe Infirmary, Oxford

Professor Deborah Ashby,Professor of Medical Statistics,Department of Environmentaland Preventative Medicine,Queen Mary University ofLondon

Professor Ann Bowling,Professor of Health ServicesResearch, Primary Care andPopulation Studies,University College London

Professor John Cairns, Professor of Health Economics,Public Health Policy, London School of Hygiene and Tropical Medicine, London

Professor Nicky Cullum,Director of Centre for EvidenceBased Nursing, Department ofHealth Sciences, University ofYork

Professor Jon Deeks, Professor of Health Statistics,University of Birmingham

Professor Jenny Donovan,Professor of Social Medicine,Department of Social Medicine,University of Bristol

Professor Freddie Hamdy,Professor of Urology, University of Sheffield

Professor Allan House, Professor of Liaison Psychiatry,University of Leeds

Professor Sallie Lamb, Director,Warwick Clinical Trials Unit,University of Warwick

Professor Stuart Logan,Director of Health & SocialCare Research, The PeninsulaMedical School, Universities ofExeter & Plymouth

Professor Miranda Mugford,Professor of Health Economics,University of East Anglia

Dr Linda Patterson, Consultant Physician,Department of Medicine,Burnley General Hospital

Professor Ian Roberts, Professor of Epidemiology &Public Health, InterventionResearch Unit, London Schoolof Hygiene and TropicalMedicine

Professor Mark Sculpher,Professor of Health Economics,Centre for Health Economics,Institute for Research in theSocial Services, University of York

Professor Kate Thomas,Professor of Complementaryand Alternative Medicine,University of Leeds

Professor David John Torgerson,Director of York Trial Unit,Department of Health Sciences,University of York

Professor Hywel Williams,Professor of Dermato-Epidemiology,University of Nottingham

Current and past membership details of all HTA ‘committees’ are available from the HTA website (www.hta.ac.uk)


Health Technology Assessment Programme

142

Diagnostic Technologies & Screening PanelMembers

Chair,Dr Ron Zimmern, Director ofthe Public Health Genetics Unit,Strangeways ResearchLaboratories, Cambridge

Ms Norma Armston,Freelance Consumer Advocate,Bolton

Professor Max Bachmann,Professor of Health CareInterfaces, Department ofHealth Policy and Practice,University of East Anglia

Professor Rudy BilousProfessor of Clinical Medicine &Consultant Physician,The Academic Centre,South Tees Hospitals NHS Trust

Ms Dea Birkett, Service UserRepresentative, London

Dr Paul Cockcroft, ConsultantMedical Microbiologist andClinical Director of Pathology,Department of ClinicalMicrobiology, St Mary'sHospital, Portsmouth

Professor Adrian K Dixon,Professor of Radiology,University Department ofRadiology, University ofCambridge Clinical School

Dr David Elliman, Consultant inCommunity Child Health,Islington PCT & Great OrmondStreet Hospital, London

Professor Glyn Elwyn, Research Chair, Centre forHealth Sciences Research,Cardiff University, Departmentof General Practice, Cardiff

Professor Paul Glasziou,Director, Centre for Evidence-Based Practice,University of Oxford

Dr Jennifer J Kurinczuk,Consultant ClinicalEpidemiologist, NationalPerinatal Epidemiology Unit,Oxford

Dr Susanne M Ludgate, Clinical Director, Medicines &Healthcare Products RegulatoryAgency, London

Mr Stephen Pilling, Director,Centre for Outcomes, Research & Effectiveness, Joint Director, NationalCollaborating Centre for MentalHealth, University CollegeLondon

Mrs Una Rennard, Service User Representative,Oxford

Dr Phil Shackley, SeniorLecturer in Health Economics,Academic Vascular Unit,University of Sheffield

Dr Margaret Somerville,Director of Public HealthLearning, Peninsula MedicalSchool, University of Plymouth

Dr Graham Taylor, ScientificDirector & Senior Lecturer,Regional DNA Laboratory, TheLeeds Teaching Hospitals

Professor Lindsay WilsonTurnbull, Scientific Director,Centre for MR Investigations &YCR Professor of Radiology,University of Hull

Professor Martin J Whittle,Clinical Co-director, NationalCo-ordinating Centre forWomen’s and Childhealth

Dr Dennis Wright, Consultant Biochemist &Clinical Director, The North West LondonHospitals NHS Trust, Middlesex

Pharmaceuticals PanelMembers

Chair,Professor Robin Ferner,Consultant Physician andDirector, West Midlands Centrefor Adverse Drug Reactions, City Hospital NHS Trust,Birmingham

Ms Anne Baileff, ConsultantNurse in First Contact Care,Southampton City Primary CareTrust, University ofSouthampton

Professor Imti Choonara,Professor in Child Health,Academic Division of ChildHealth, University ofNottingham

Professor John Geddes,Professor of EpidemiologicalPsychiatry, University of Oxford

Mrs Barbara Greggains, Non-Executive Director,Greggains Management Ltd

Dr Bill Gutteridge, MedicalAdviser, National SpecialistCommissioning Advisory Group(NSCAG), London

Mrs Sharon Hart, Consultant PharmaceuticalAdviser, Reading

Dr Jonathan Karnon, SeniorResearch Fellow, HealthEconomics and DecisionScience, University of Sheffield

Dr Yoon Loke, Senior Lecturerin Clinical Pharmacology,University of East Anglia

Ms Barbara Meredith,Lay Member, Epsom

Dr Andrew Prentice, SeniorLecturer and ConsultantObstetrician & Gynaecologist,Department of Obstetrics &Gynaecology, University ofCambridge

Dr Frances Rotblat, CPMPDelegate, Medicines &Healthcare Products RegulatoryAgency, London

Dr Martin Shelly, General Practitioner, Leeds

Mrs Katrina Simister, AssistantDirector New Medicines,National Prescribing Centre,Liverpool

Dr Richard Tiner, MedicalDirector, Medical Department,Association of the BritishPharmaceutical Industry,London

Current and past membership details of all HTA ‘committees’ are available from the HTA website (www.hta.ac.uk)

Therapeutic Procedures PanelMembers

Chair, Professor Bruce Campbell,Consultant Vascular andGeneral Surgeon, Departmentof Surgery, Royal Devon &Exeter Hospital

Dr Mahmood Adil, DeputyRegional Director of PublicHealth, Department of Health,Manchester

Dr Aileen Clarke,Consultant in Public Health,Public Health Resource Unit,Oxford

Professor Matthew Cooke,Professor of EmergencyMedicine, Warwick EmergencyCare and Rehabilitation,University of Warwick

Mr Mark Emberton, SeniorLecturer in OncologicalUrology, Institute of Urology,University College Hospital

Professor Paul Gregg,Professor of OrthopaedicSurgical Science, Department ofGeneral Practice and PrimaryCare, South Tees Hospital NHSTrust, Middlesbrough

Ms Maryann L Hardy, Lecturer, Division ofRadiography, University ofBradford

Dr Simon de Lusignan,Senior Lecturer, Primary CareInformatics, Department ofCommunity Health Sciences,St George’s Hospital MedicalSchool, London

Dr Peter Martin, ConsultantNeurologist, Addenbrooke’sHospital, Cambridge

Professor Neil McIntosh,Edward Clark Professor of ChildLife & Health, Department ofChild Life & Health, Universityof Edinburgh

Professor Jim Neilson,Professor of Obstetrics andGynaecology, Department ofObstetrics and Gynaecology,University of Liverpool

Dr John C Pounsford,Consultant Physician,Directorate of Medical Services,North Bristol NHS Trust

Dr Karen Roberts, NurseConsultant, Queen ElizabethHospital, Gateshead

Dr Vimal Sharma, ConsultantPsychiatrist/Hon. Senior Lecturer, Mental HealthResource Centre, Cheshire andWirral Partnership NHS Trust,Wallasey

Professor Scott Weich, Professor of Psychiatry, Division of Health in theCommunity, University ofWarwick

Disease Prevention PanelMembers

Chair, Dr Edmund Jessop, MedicalAdviser, National SpecialistCommissioning Advisory Group(NSCAG), London

Mrs Sheila Clark, ChiefExecutive, St James’s Hospital,Portsmouth

Mr Richard Copeland, Lead Pharmacist: ClinicalEconomy/Interface, Wansbeck General Hospital,Northumberland

Dr Elizabeth Fellow-Smith,Medical Director, West London Mental HealthTrust, Middlesex

Mr Ian Flack, Director PPIForum Support, Council ofEthnic Minority VoluntarySector Organisations, Stratford

Dr John Jackson, General Practitioner, Newcastle upon Tyne

Mrs Veronica James, ChiefOfficer, Horsham District AgeConcern, Horsham

Professor Mike Kelly, Director, Centre for PublicHealth Excellence, National Institute for Healthand Clinical Excellence, London

Professor Yi Mien Koh, Director of Public Health andMedical Director, London NHS (North West LondonStrategic Health Authority),London

Ms Jeanett Martin, Director of Clinical Leadership& Quality, Lewisham PCT,London

Dr Chris McCall, GeneralPractitioner, Dorset

Dr David Pencheon, Director,Eastern Region Public HealthObservatory, Cambridge

Dr Ken Stein, Senior ClinicalLecturer in Public Health,Director, Peninsula TechnologyAssessment Group, University of Exeter, Exeter

Dr Carol Tannahill, Director,Glasgow Centre for PopulationHealth, Glasgow

Professor Margaret Thorogood,Professor of Epidemiology,University of Warwick, Coventry

Dr Ewan Wilkinson, Consultant in Public Health,Royal Liverpool UniversityHospital, Liverpool


143Current and past membership details of all HTA ‘committees’ are available from the HTA website (www.hta.ac.uk)

Health Technology Assessment Programme

144Current and past membership details of all HTA ‘committees’ are available from the HTA website (www.hta.ac.uk)

Expert Advisory NetworkMembers

Professor Douglas Altman,Professor of Statistics inMedicine, Centre for Statisticsin Medicine, University ofOxford

Professor John Bond,Director, Centre for HealthServices Research, University ofNewcastle upon Tyne, School ofPopulation & Health Sciences,Newcastle upon Tyne

Professor Andrew Bradbury,Professor of Vascular Surgery,Solihull Hospital, Birmingham

Mr Shaun Brogan, Chief Executive, RidgewayPrimary Care Group, Aylesbury

Mrs Stella Burnside OBE,Chief Executive, Regulation and ImprovementAuthority, Belfast

Ms Tracy Bury, Project Manager, WorldConfederation for PhysicalTherapy, London

Professor Iain T Cameron,Professor of Obstetrics andGynaecology and Head of theSchool of Medicine,University of Southampton

Dr Christine Clark,Medical Writer & ConsultantPharmacist, Rossendale

Professor Collette Clifford,Professor of Nursing & Head ofResearch, School of HealthSciences, University ofBirmingham, Edgbaston,Birmingham

Professor Barry Cookson,Director, Laboratory ofHealthcare Associated Infection,Health Protection Agency,London

Dr Carl Counsell, ClinicalSenior Lecturer in Neurology,Department of Medicine &Therapeutics, University ofAberdeen

Professor Howard Cuckle,Professor of ReproductiveEpidemiology, Department ofPaediatrics, Obstetrics &Gynaecology, University ofLeeds

Dr Katherine Darton, Information Unit, MIND – The Mental Health Charity,London

Professor Carol Dezateux, Professor of PaediatricEpidemiology, London

Dr Keith Dodd, ConsultantPaediatrician, Derby

Mr John Dunning,Consultant CardiothoracicSurgeon, CardiothoracicSurgical Unit, PapworthHospital NHS Trust, Cambridge

Mr Jonothan Earnshaw,Consultant Vascular Surgeon,Gloucestershire Royal Hospital,Gloucester

Professor Martin Eccles, Professor of ClinicalEffectiveness, Centre for HealthServices Research, University ofNewcastle upon Tyne

Professor Pam Enderby,Professor of CommunityRehabilitation, Institute ofGeneral Practice and PrimaryCare, University of Sheffield

Professor Gene Feder, Professorof Primary Care Research &Development, Centre for HealthSciences, Barts & The LondonQueen Mary’s School ofMedicine & Dentistry, London

Mr Leonard R Fenwick, Chief Executive, Newcastleupon Tyne Hospitals NHS Trust

Mrs Gillian Fletcher, Antenatal Teacher & Tutor andPresident, National ChildbirthTrust, Henfield

Professor Jayne Franklyn,Professor of Medicine,Department of Medicine,University of Birmingham,Queen Elizabeth Hospital,Edgbaston, Birmingham

Dr Neville Goodman, Consultant Anaesthetist,Southmead Hospital, Bristol

Professor Robert E Hawkins, CRC Professor and Director ofMedical Oncology, Christie CRCResearch Centre, ChristieHospital NHS Trust, Manchester

Professor Allen Hutchinson, Director of Public Health &Deputy Dean of ScHARR,Department of Public Health,University of Sheffield

Professor Peter Jones, Professorof Psychiatry, University ofCambridge, Cambridge

Professor Stan Kaye, CancerResearch UK Professor ofMedical Oncology, Section ofMedicine, Royal MarsdenHospital & Institute of CancerResearch, Surrey

Dr Duncan Keeley,General Practitioner (Dr Burch& Ptnrs), The Health Centre,Thame

Dr Donna Lamping,Research Degrees ProgrammeDirector & Reader in Psychology,Health Services Research Unit,London School of Hygiene andTropical Medicine, London

Mr George Levvy,Chief Executive, Motor Neurone Disease Association,Northampton

Professor James Lindesay,Professor of Psychiatry for theElderly, University of Leicester,Leicester General Hospital

Professor Julian Little,Professor of Human GenomeEpidemiology, Department ofEpidemiology & CommunityMedicine, University of Ottawa

Professor Rajan Madhok, Consultant in Public Health,South Manchester Primary Care Trust, Manchester

Professor Alexander Markham, Director, Molecular MedicineUnit, St James’s UniversityHospital, Leeds

Professor Alistaire McGuire,Professor of Health Economics,London School of Economics

Dr Peter Moore, Freelance Science Writer, Ashtead

Dr Andrew Mortimore, PublicHealth Director, SouthamptonCity Primary Care Trust,Southampton

Dr Sue Moss, Associate Director,Cancer Screening EvaluationUnit, Institute of CancerResearch, Sutton

Mrs Julietta Patnick, Director, NHS Cancer ScreeningProgrammes, Sheffield

Professor Robert Peveler,Professor of Liaison Psychiatry,Royal South Hants Hospital,Southampton

Professor Chris Price, Visiting Professor in ClinicalBiochemistry, University ofOxford

Professor William Rosenberg,Professor of Hepatology andConsultant Physician, Universityof Southampton, Southampton

Professor Peter Sandercock,Professor of Medical Neurology,Department of ClinicalNeurosciences, University ofEdinburgh

Dr Susan Schonfield, Consultantin Public Health, HillingdonPCT, Middlesex

Dr Eamonn Sheridan,Consultant in Clinical Genetics,Genetics Department,St James’s University Hospital,Leeds

Professor Sarah Stewart-Brown, Professor of Public Health,University of Warwick, Division of Health in theCommunity Warwick MedicalSchool, LWMS, Coventry

Professor Ala Szczepura, Professor of Health ServiceResearch, Centre for HealthServices Studies, University ofWarwick

Dr Ross Taylor, Senior Lecturer, Department ofGeneral Practice and PrimaryCare, University of Aberdeen

Mrs Joan Webster, Consumer member, HTA –Expert Advisory Network

How to obtain copies of this and other HTA Programme reports.An electronic version of this publication, in Adobe Acrobat format, is available for downloading free ofcharge for personal use from the HTA website (http://www.hta.ac.uk). A fully searchable CD-ROM is alsoavailable (see below).

Printed copies of HTA monographs cost £20 each (post and packing free in the UK) to both public andprivate sector purchasers from our Despatch Agents, York Publishing Services.

Non-UK purchasers will have to pay a small fee for post and packing. For European countries the cost is£2 per monograph and for the rest of the world £3 per monograph.

You can order HTA monographs from our Despatch Agents, York Publishing Services by:

– fax (with credit card or official purchase order) – post (with credit card or official purchase order or cheque)– phone during office hours (credit card only).

Additionally the HTA website allows you either to pay securely by credit card or to print out yourorder and then post or fax it.

Contact details are as follows:York Publishing Services Email: [email protected] Box 642 Tel: 0870 1616662YORK YO31 7WX Fax: 0870 1616663UK Fax from outside the UK: +44 1904 430868

NHS libraries can subscribe free of charge. Public libraries can subscribe at a very reduced cost of £100 for each volume (normally comprising 30–40 titles). The commercial subscription rate is £300 per volume. Please contact York Publishing Services at the address above. Subscriptions can only bepurchased for the current or forthcoming volume.

Payment methods

Paying by chequeIf you pay by cheque, the cheque must be in pounds sterling, made payable to York PublishingDistribution and drawn on a bank with a UK address.

Paying by credit cardThe following cards are accepted by phone, fax, post or via the website ordering pages: Delta, Eurocard,Mastercard, Solo, Switch and Visa. We advise against sending credit card details in a plain email.

Paying by official purchase orderYou can post or fax these, but they must be from public bodies (i.e. NHS or universities) within the UK.We cannot at present accept purchase orders from commercial companies or from outside the UK.

How do I get a copy of HHTTAA oonn CCDD?

Please use the form on the HTA website (www.hta.ac.uk/htacd.htm). Or contact York PublishingServices (see contact details above) by email, post, fax or phone. HTA on CD is currently free of chargeworldwide.

The website also provides information about the HTA Programme and lists the membership of the variouscommittees.

HTA

Health Technology Assessm

ent 2007;Vol. 11: No. 17

Screening for type 2 diabetes: literature review and econom

ic modelling

Screening for type 2 diabetes: literature review and economic modelling

N Waugh, G Scotland, P McNamee, M Gillett, A Brennan, E Goyder, R Williamsand A John


HTAHealth Technology AssessmentNHS R&D HTA Programmewww.hta.ac.uk

The National Coordinating Centre for Health Technology Assessment,Mailpoint 728, Boldrewood,University of Southampton,Southampton, SO16 7PX, UK.Fax: +44 (0) 23 8059 5639 Email: [email protected]://www.hta.ac.uk ISSN 1366-5278

FeedbackThe HTA Programme and the authors would like to know

your views about this report.

The Correspondence Page on the HTA website(http://www.hta.ac.uk) is a convenient way to publish

your comments. If you prefer, you can send your comments to the address below, telling us whether you would like

us to transfer them to the website.

We look forward to hearing from you.

May 2007

nhs r&d hta programme hta · nihr health technology assessment programme. t. he health...

Documents