a new prevention paradox: the trade-off between reducing incentives for risk selection and...

9
A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers Tim A. Kanters a, * , Werner B.F. Brouwer a , René C.J.A. van Vliet a , Pieter H.M. van Baal a, b , Johan J. Polder b, c a Institute of Health Policy & Management, Erasmus University Rotterdam, P.O. Box 1738, 3000 DR Rotterdam, The Netherlands b National Institute for Public Health and the Environment, P.O. Box 1, 3720 BA Bilthoven, The Netherlands c Tranzo Department, Tilburg University, P.O. Box 90153, 5000 LE Tilburg, The Netherlands article info Article history: Available online 3 November 2012 Keywords: The Netherlands Risk equalization Prevention Chronic Disease Model abstract The Dutch risk equalization scheme has been improved over the years by including health related risk adjusters. The purpose of the Dutch risk equalization scheme is to prevent risk selection and to correct for predictable losses and gains for insurers. The objective of this paper is to explore the nancial incentives for risk selection under the Dutch risk equalization scheme. We used a simulation model to estimate lifetime health care costs and risk equalization contributions for three cohorts (a smoking; an obese; and a healthy living cohort). Financial differences for the three cohorts were assessed by subtracting health care costs from risk equalization contributions. Even under an elaborate risk equalization system, the healthy living cohort was still most nancially attractive for insurers. Smokers were somewhat less attractive, while the obese cohort was least attractive. Lifetime differences with healthy living individuals (revenues minus costs) were modest: V4840 for obese individuals and V1101 for smokers. Under a simple form of risk equalization these differences were higher, V8542 and V4620 respectively. Improvement of the risk equalization scheme reduced the gap between costs and revenues. Incentives for undesirable risk selection were reduced, but simultaneously incentives for health promotion were weakened. This highlights a new prevention paradox: improving the level playing eld for health insurers will inevitably limit their incentives for promoting the health of their clients. Ó 2012 Elsevier Ltd. All rights reserved. Introduction Many countries struggle to improve public health. In many developed countries, lifestyle related risk-factors, such as smoking and obesity, are responsible for considerable morbidity and mortality. Prevention of such behavior may therefore be considered to be an important strategy in improving public health (OECD, 2005). When health insurers are given a prominent role in prevention, it is important to consider what incentives they have to promote healthy life styles. An important, though not the only, determinant for insurers incentives for prevention is the nancial balance between revenues (e.g. from premiums) and costs for their insureds. If this balance is most favorable for healthy living indi- viduals, insurers would have a clear and direct nancial incentive to engage in preventive actions to reduce unhealthy lifestyles. If such incentives indeed exist, however, insurers also have an incentive for risk selection. Insurers might select insureds with a healthy life style rather than improving habits of unhealthy living insureds. It is well-known that risk selection may threaten soli- darity, quality and efciency of the health care system and as such is discouraged in many countries (Van de Ven, Van Vliet, & Lamers, 2004). In the Netherlands a relatively sophisticated risk equaliza- tion scheme is used. In the Dutch scheme, insurers are nancially compensated for having insureds with an increased risk of high health care costs. Ideally, this happens in such a way that the incentives for risk selection are eliminated and that, regardless of the risk prole of their insureds, in principle, insurers all act on a level playing eldin competing with each other. This implies that there is a clear tension between the aim of having insurers engage in more preventive action in order to improve public health and the wish of reducing incentives for risk selection. In the case of perfect risk equalization, the expected nancial gains on healthy and unhealthy living insureds would be equal. Then, obviously, the insurersnancial incentives to improve the lifestyle of their insureds will disappear completely. On the * Corresponding author. Tel.: þ3110 408 29 17; fax: þ31 10 408 90 81. E-mail address: [email protected] (T.A. Kanters). Contents lists available at SciVerse ScienceDirect Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed 0277-9536/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.socscimed.2012.10.019 Social Science & Medicine 76 (2013) 150e158

Upload: johan-j

Post on 27-Jan-2017

213 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

at SciVerse ScienceDirect

Social Science & Medicine 76 (2013) 150e158

Contents lists available

Social Science & Medicine

journal homepage: www.elsevier .com/locate/socscimed

A new prevention paradox: The trade-off between reducing incentives for riskselection and increasing the incentives for prevention for health insurers

Tim A. Kanters a,*, Werner B.F. Brouwer a, René C.J.A. van Vliet a, Pieter H.M. van Baal a,b, Johan J. Polder b,c

a Institute of Health Policy & Management, Erasmus University Rotterdam, P.O. Box 1738, 3000 DR Rotterdam, The NetherlandsbNational Institute for Public Health and the Environment, P.O. Box 1, 3720 BA Bilthoven, The Netherlandsc Tranzo Department, Tilburg University, P.O. Box 90153, 5000 LE Tilburg, The Netherlands

a r t i c l e i n f o

Article history:Available online 3 November 2012

Keywords:The NetherlandsRisk equalizationPreventionChronic Disease Model

* Corresponding author. Tel.: þ31 10 408 29 17; faxE-mail address: [email protected] (T.A. Kanters)

0277-9536/$ e see front matter � 2012 Elsevier Ltd.http://dx.doi.org/10.1016/j.socscimed.2012.10.019

a b s t r a c t

The Dutch risk equalization scheme has been improved over the years by including health related riskadjusters. The purpose of the Dutch risk equalization scheme is to prevent risk selection and to correctfor predictable losses and gains for insurers. The objective of this paper is to explore the financialincentives for risk selection under the Dutch risk equalization scheme.

We used a simulation model to estimate lifetime health care costs and risk equalization contributionsfor three cohorts (a smoking; an obese; and a healthy living cohort). Financial differences for the threecohorts were assessed by subtracting health care costs from risk equalization contributions.

Even under an elaborate risk equalization system, the healthy living cohort was still most financiallyattractive for insurers. Smokers were somewhat less attractive, while the obese cohort was leastattractive. Lifetime differences with healthy living individuals (revenues minus costs) were modest:V4840 for obese individuals and V1101 for smokers. Under a simple form of risk equalization thesedifferences were higher, V8542 and V4620 respectively.

Improvement of the risk equalization scheme reduced the gap between costs and revenues. Incentivesfor undesirable risk selection were reduced, but simultaneously incentives for health promotion wereweakened. This highlights a new prevention paradox: improving the level playing field for healthinsurers will inevitably limit their incentives for promoting the health of their clients.

� 2012 Elsevier Ltd. All rights reserved.

Introduction

Many countries struggle to improve public health. In manydeveloped countries, lifestyle related risk-factors, such as smokingand obesity, are responsible for considerable morbidity andmortality. Prevention of such behavior may therefore be consideredto be an important strategy in improving public health (OECD,2005). When health insurers are given a prominent role inprevention, it is important to consider what incentives they have topromote healthy life styles. An important, though not the only,determinant for insurer’s incentives for prevention is the financialbalance between revenues (e.g. from premiums) and costs for theirinsureds. If this balance is most favorable for healthy living indi-viduals, insurers would have a clear and direct financial incentive toengage in preventive actions to reduce unhealthy lifestyles.

: þ31 10 408 90 81..

All rights reserved.

If such incentives indeed exist, however, insurers also have anincentive for risk selection. Insurers might select insureds witha healthy life style rather than improving habits of unhealthy livinginsureds. It is well-known that risk selection may threaten soli-darity, quality and efficiency of the health care system and as suchis discouraged in many countries (Van de Ven, Van Vliet, & Lamers,2004). In the Netherlands a relatively sophisticated risk equaliza-tion scheme is used. In the Dutch scheme, insurers are financiallycompensated for having insureds with an increased risk of highhealth care costs. Ideally, this happens in such a way that theincentives for risk selection are eliminated and that, regardless ofthe risk profile of their insureds, in principle, insurers all act ona ‘level playing field’ in competing with each other.

This implies that there is a clear tension between the aim ofhaving insurers engage in more preventive action in order toimprove public health and the wish of reducing incentives for riskselection. In the case of perfect risk equalization, the expectedfinancial gains on healthy and unhealthy living insureds would beequal. Then, obviously, the insurers’ financial incentives to improvethe lifestyle of their insureds will disappear completely. On the

Page 2: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

T.A. Kanters et al. / Social Science & Medicine 76 (2013) 150e158 151

contrary, however, if insurers do have a financial interest inengaging in prevention (i.e. healthy living persons are more prof-itable than unhealthy living individuals), this automatically impliesthe existence of incentives for risk selection. So, an interesting newprevention paradox emerges: preventing risk selection requiresperfect risk equalization, but preventing unhealthy lifestyles maybenefit from imperfect risk equalization. (The traditional preven-tion paradox describes the situation in which most health gainsfrom preventive action are gained in patients who are low riskindividuals and only a small proportion of health gains are gainedin high risk individuals, simply because most people have a lowrisk. This means that in order to gain substantial health gains ona macro level, especially low risk individuals should engage inprevention, for whom preventive action on an individual level isexpectedly marginal.)

In this paper, we will explore the financial incentives for riskselecting three distinct lifestyle groups (smokers, obese and healthyliving individuals), and investigate how the recent improvementsin the Dutch risk equalization scheme have changed these incen-tives. With our results, we can draw conclusions whether therepotentially is a prevention paradox between risk equalization andhealth promotion, especially with respect to interventions forreducing obesity and smoking. This holds since reduced incentivesfor risk selection, ceteris paribus, translate in reduced incentives forprevention of unhealthy life styles (if such life styles are moreexpensive for health insurers). In addressing these issues, we willuse Dutch data to highlight the current incentives for prevention(and risk selection) and the influence of the more recentimprovements in the Dutch risk equalization scheme in thatcontext. Given the Dutch health care system with its relativelysophisticated risk equalization scheme, Dutch data provides a goodexample of the more general point highlighted in this paper. Wewill demonstrate the financial incentives for reducing unhealthylifestyles in insureds for health insurers and show how improvedrisk equalization affects this.

The paper is structured as follows. In the background section, wewill briefly discuss the risk equalization scheme as used in theNetherlands as of 2006. Subsequently, we will discuss the method-ologyof thestudy. In the results sectionwewill examine theoutcomesof the analyses. Some final remarks on the results and appliedmethods will be made in the discussion, which concludes this paper.

Dutch health insurance context

The Dutch health insurance system is characterized by universalcoverage, open enrollment, and community rated premiums (Vande Ven & Schut, 2008). The system should guarantee accessibleand affordable health care for all citizens (Maarse & Bartholomée,2007).

Fig. 1. Funding of the Dutch h

In order to prevent risk selection e the possible negative conse-quence of community rated premiumse a risk equalization schemeisused in theNetherlands,whichadjusts the revenuesof the insurersin such a way that it, ideally, reflects the financial risk they run ontheir pool of insureds. As the predicting power of the risk equaliza-tion formula increases, the expected profits and losses on subgroupsconverge. This decreases insurers’ incentives to engage in riskselection (VandeVen, Beck, VandeVoorde,Wasem, & Zmora, 2007).The financial flows of the Health Insurance Act are given in Fig. 1.

Health insurance is obligatory in the Netherlands. The contractduration is one year after which insureds are free to switch insurersand there again is open enrollment. The number of switchers isrelatively small, but non-negligible. In 2007, 6% of the generalpopulation switched health insurer, while in 2008 this percentagewas 4% (Rooijen, De Jong, & Rijken, 2011). Every year theMinistry ofHealth, Welfare and Sport (MinVWS) estimates the total amount ofpublic health expenditure. The Health Care Insurance Boardsubsequently determines the nominal insurance premium, which isset in such a way that it covers 50 per cent of the health expendi-ture. In 2008, this annual nominal insurance premium was set atV970 per inhabitant (MinVWS, 2007). In order to allow premiumcompetition, insurers are allowed to determine the actual premiumthey charge to their customers at their own discretion. In 2008, theaverage Dutch premium for the basic health insurance packagewasV1100, with a compulsory deductible of V150 (Van de Ven & Schut,2008). Premium differences between insurers reflect the relativeefficiency (especially in terms of being prudent purchasers of care)of insurers. Basically, this role of competing insurers as prudentpurchasers of care on behalf of their insureds, forms the heart of theDutch health care system.

The other half of insurers’ revenues comes from the riskequalization fund (hereafter abbreviated as REF). The REF is filledby means of income dependent contributions by all adult citizens.In addition, the government contributes to the REF for children(Van de Ven et al., 2007; Van de Ven & Schut, 2008). The funds aresubsequently distributed among insurers on the basis of the riskprofile of their insureds. To determine this risk profile, currentlyseven parameters are used (Van Vliet et al., 2007):

(1) Age;(2) Gender;(3) Pharmacy costs groups (PCGs);(4) Diagnostic costs groups (DCGs);(5) Socio-economic status (SES) and age;(6) Region-based clusters; and(7) Source of income and age.

The fifth risk adjuster consists of the interaction between SESand age, while the seventh risk adjuster represents an interaction ofage and source of income.

ealth insurance scheme.

Page 3: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

T.A. Kanters et al. / Social Science & Medicine 76 (2013) 150e158152

The direct health indicators PCGs and DCGs both refer to yeart � 1 and serve as predictors for extramural and intramural careexpenditures in year t, respectively. PCGs and DCGs are comple-mentary to each other in predicting health care expenses. TheDutch risk equalization formula is relatively accurate in predictinghealth care expenses per individual, explaining approximately 22%of the variance in health care expenses (Van de Ven et al., 2004).Risk equalization is only applied to the basic insurance package,which covers medically necessary health care, and is equal for allDutch insureds (Van de Ven & Schut, 2008). Risk equalization onlyapplies to health care costs; loading fees, administrative costs andprofits are not included in the risk equalization flows.

The contributions from the REF to the insurers are determinedon an ex ante basis, using the seven risk adjusters. In addition to exante risk equalization, also several types of ex post risk equalizationmechanisms are currently used in the Dutch context, such asmacro-recalculation (equalization of the entire health care budget),generic or insurer recalculation (equalizing the difference betweenactual costs and predicted costs, between individual insurers,independent from the REF), and high costs equalization (individ-uals exceeding costs of V20,000 in 2008 are covered for 90% by theREF and only 10% by the insurer). Ex post equalization is claimed tolimit insurers’ incentives to be efficient (Van de Ven, 2011). TheDutch government therefore aims at improving the ex ante riskequalization mechanisms while simultaneously reducing ex postequalization mechanisms. Consequently, insurer’s financialaccountability will increase (Van de Ven & Schut, 2008). In the yearsafter the introduction of the new system of regulated competitionin 2006, insurers’ costs and revenues did not yet level out. This wasmainly due to fierce premium competition. In 2007, for instance,insurers’ costs on average were 4.8% higher than their revenues(CVZ, 2008), implying large collective losses.

Methods

We describe the financial incentives for risk selection (andreducing unhealthy lifestyles in insureds) experienced by healthinsurers in the Dutch setting in relation to recent improvements inthe REF scheme. Differences between revenues and expendituresrelated to having unhealthy living (smoking or being obese) orrather healthy living (non-obese, non-smokers) insureds wereestimated, from the perspective of the insurance company. Thegroup with the largest positive balance (revenues minus costs) wasseen as the financially most attractive for the insurer. When thedifference between these expenditures and revenues (i.e. netrevenues or expenditures) is more favorable for healthy than forunhealthy living individuals, risk selection may be profitable.Differences between the three cohorts were compared on a cross-sectional basis, over the entire lifespan of insureds, and peraverage life year. The situations with a simpler and a moreadvanced REF scheme were compared, to examine the influence ofimproving the risk equalization mechanism on incentives for riskselection at the level of the insurer. Ethical approval was notneeded, as the analyses were not based on data from newindividuals.

Since panel data on lifestyle, health status, and health careconsumption over a long time period were not available, we usedaggregate disease prevalence and health care expenditure predic-tions produced with the RIVM Chronic Disease Model (CDM). TheCDM is a Markov type simulation model that predicts survival andchronic disease incidence, prevalence andmortality as a function ofrisk factors such as smoking and obesity. The model parameters ofthe CDM were estimated from a variety of data sources (e.g. inci-dence and prevalence fromGP and cancer registries, mortality fromrecord linkage studies, relative risks from observational studies).

The CDM has been used before to compare these groups (smokers,obese and healthy living people) more generally (Rappange,Brouwer, Rutten, & Van Baal, 2010; Van Baal, Hoogenveen, DeWit, & Boshuizen, 2006; Van Baal et al., 2008).

Chronic Disease Model

In order to be able to calculate costs and revenues related to thedifferent cohorts, we used outcomes from the RIVM ChronicDisease Model (CDM; Hoogenveen, Van Baal, & Boshuizen, 2010;Van Baal et al., 2006; Van Baal et al., 2008). The CDM has been usedbefore to predict lifetime costs of obese, smoking and healthy livingindividuals (Van Baal et al., 2008). For our current purposes, weused predictions of this model for three cohorts each consisting of500men and 500women, whowere either: (1) smoking; (2) obese;or (3) healthy living people. All men and women were initially 20years of age. It should be noted that the predictions of the CDM canbe seen as aggregate level data and for our analysis we chose 500men and 500 women purely for convenience. As the predictions ofthe RIVM-CDM were made at the group level, not at the individuallevel and, therefore, we were unable to perform statistical tests.

The model used smoking and obesity as risk factors to simulatethe prevalence of 22 related diseases. All other diseases wereassumed to be unrelated to smoking or obesity in these analysesand were summed in one variable. The prevalence of unrelateddiseases was simulated for this variable and was only dependent onthe numbers of survivors in the cohorts. The outcomes of the CDM,which consists of number of survivors, disease prevalence andhealth care costs specified by disease and health care provider(hospitals, medication, etc.), were linked to the risk adjusters of theREF.

Health care costs

The Dutch Cost of Illness study 2003 determined disease coststhrough a topedown attribution of total health care costs in theNetherlands (RIVM, 2011; Slobbe et al., 2006). The patternsobserved in the Dutch Cost of Illness study are similar to the healthcare costs patterns observed in other countries (Heijink, Noethen,Renaud, Koopmanschap, & Polder, 2008). In this dataset healthcare costs for insurers can be subtracted from total costs of healthcare per disease category. Health care expenditures predictions ofthe CDM were made by attaching costs per patient per disease tothe predicted disease prevalence (for details see Van Baal, Feenstra,Hoogenveen, De Wit, & Brouwer, 2007 and Van Baal, Feenstra,Polder, Hoogenveen, & Brouwer, 2011).

Health care costs for insurers as used in the Cost of Illness studyof 2003 were adjusted for the demography of the 2008 Dutchpopulation. Subsequently, total health care costs for theNetherlands on which the Costs of Illness study is based was setequal to the 2008 macro budget for health care costs for insurers inthe Netherlands. The macro budget equaled V30.5 billon for 2008(MinVWS, 2007). Themacro budget only comprises insurer’s healthcare costs within the basic revenues package. Constant hospitalcosts and curative mental health care costs are components of themacro budget that were not incorporated in the CDM. Theseelements were therefore removed from the macro budget beforeequalizing Cost of Illness figures with the macro budget. Conse-quently, this study was based on total health care expenditure ofV22.8 billion.

Risk equalization model

We used the four main risk adjusters of the Dutch risk equal-ization system, namely age, gender and the two direct health

Page 4: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

T.A. Kanters et al. / Social Science & Medicine 76 (2013) 150e158 153

proxies PCGs and DCGs that relate to disease prevalence. Together,these four risk adjusters predict 17% of the variance in total healthcare costs, while as indicated, all seven risk adjusters in the Dutchrisk equalization formula explain 22% (Van de Ven et al., 2004; VanVliet et al., 2007). For each individual, total REF contribution wasequal to the summation of the contributions for the separate riskadjusters. In our analyses risk equalization contributions persimulation year are computed as follows: Risk equalizationcontribution per simulation year ¼ (Demography con-tribution*number of persons of corresponding age and sex) þ (PCGcontribution*number of persons with corresponding PCG-classification) � (PCG0 contribution*number of PCG0persons) þ (DCG contribution*number of persons with corre-sponding DCG-classification) � (DCG0 contribution*number ofDCG0 persons). The minimum and maximum contributions aregiven in Table 1.

All insureds were attributed to one or more of the 21 pharmacy-based cost groups. Twenty PCGs are related to 20 specific diseases,for which certainmedications have been prescribed (for at least 180daily doses a year) in the preceding year. The additional PCG (PCG0)captures those individuals who are not included in any of the PCGs,and contains 84% of the population. PCG0 corresponds witha negative risk equalization contribution, that is, insurers have tomake a payment into the REF for classified in PCG0 (Van Vliet et al.,2007).

Moreover, all insureds were categorized into one of 14 diag-nostic cost groups. Insureds with multiple DCG classifications werecategorized into the DCG with highest compensation. DCG classi-fications are based on diagnoses from clinical discharges in thepreceding year. DCG0 contains those individuals that are notcaptured by any other DCG. 98% of the Dutch population is classi-fied as DCG0. Table 1 shows that DCG0 is also linked to a paymentinto the REF (Van Vliet et al., 2007).

Prevalence of PCG’s and DCG’s related to smoking and obesity

Contributions from the REF associated with the risk adjusters in2008 (Table 1) are given in Van Vliet et al. (2007). Data on predictedvolumes of PCG and DCG classifications in the Netherlands wereobtained from the Health Care Insurance Board (CVZ). The NationalMedical Registration (2005) on hospital discharges was used todetermine the unrelated and related proportions of DCGs. Preva-lence rates were taken from the National Public Health Compass(RIVM, 2008) and other sources. A detailed explanation of theestimation of these prevalence rates can be found elsewhere (VanBaal, Engelfriet et al., 2011).

Disease prevalence as provided by the National Compass cannotbe directly used to calculate expected numbers of PCGs (and DCGs),because not all prevalent cases will be treated. Hence prevalenceand predicted volumes of PCG for the Dutch population (by CVZ) donot fully coincide. Obviously, since additional revenues for aninsurer are based only on treated individuals we adjusted the

Table 1Risk factors and equalization contributions in the 2008 Dutch risk equalizationsystem (Van Vliet et al., 2007).

Risk factor Number ofclasses

Smallestcontribution

Largestcontribution

Age*sex 38 V818 V3260Pharmacy costs groups 21 V�313 V21,057Diagnostic costs groups 14 V�98 V52,931Socio-economic status*age 12 V�201 V518Region 10 V�98 V60Source of income*age 20 V�304 V1019

prevalence rates to account for less than 100% treatment. This wasdone by dividing the predicted volumes of PCG for the Dutchpopulation by the national prevalence rates from the NationalCompass. This indicates howmany prevalent cases end up in a PCGclassification, i.e. the average number of PCG classifications perprevalent patient. This ratio was linked to the prevalence rates inthe CDM to find the number of PCG for each disease. If, for instance,160,000 diabetes PCG’s were predicted by CVZ for a prevalentpopulation of 200,000 patients, the ratio of PCG to patient would be0.8. For every 10 diabetes patients in the model, 8 diabetes PCGwould then be issued.

Similarly, we adjusted the number of hospital discharges asprovided by the National Medical Registration using the volumes ofDCG for the Dutch population as predicted by CVZ. Using hospitaldischarge data from the National Medical Registration (2005), theratio of DCG classification and disease was computed. This resultedin the number of DCGs per patient. Then, this number wascombined with the disease prevalence rates from the CDM, and thevolumes of DCGs in the three cohorts were found.

Prevalence of unrelated PCG’s and DCG’s

The occurrence of PCGs and DCGs unrelated to the 22 diseases inthe CDM was assumed to depend only on the number of survivorsin each cohort. The shares of related and unrelated PCG and DCGclassifications were computed based on the real number in 2008according to the risk equalization data of CVZ in order to find thenumber of unrelated PCGs and DCGs for each cohort.

With the prevalence of the related (CDM) diseases and total(PCG) diseases, the share of related diseases in total diseases wasdetermined. The proportion of unrelated PCGs was used to estab-lish the number of unrelated PCGs per inhabitant. This number ofunrelated PCGs per capita was subsequently used in combinationwith the number of survivors in the three cohorts to find the totalnumber of unrelated PCGs in the cohorts.

The National Medical Registration (2005) provided informationon the volume of DCGs and the volumes of related (CDM disease),and unrelated (non-CDM disease) DCGs. This ratio was linked to thepredicted total number of DCGs. The proportion of unrelated DCGswas used to find the number of CDM-unrelated DCGs per person.The per capita proportion was subsequently used to find thenumber of unrelated DCGs in the three cohorts.

The numbers of unrelated PCGs and DCGswere corrected for ageand sex. The related PCGs and DCGs were not corrected for age andsex, as the CDM already includes age and sex as risk factors onprevalence rates of the CDM (related) diseases.

For healthy insureds, that is, those not classified into PCG groupsand/or DCG groups, insurers have to contribute to the REF. Todetermine this payment into the REF for PCG0s and DCG0s, thevolumes of PCG0 and DCG0 classified persons in each cohort wereneeded. The number of people without diseases was calculated bymultiplying the disease-free fractions for all diseases with thenumber of survivors in the cohorts.

Results

Our findings resemble the results of Van Baal et al. (2008) in thesense that over the entire lifespan, health care costs were highestfor healthy living insureds (Table 2A; this paper exclusively exam-ines health insurers’ costs which explains differences compared toVan Baal et al. (2008), who included also long term care expendi-ture). Healthy living people died at older ages than unhealthy livingpeople, and therefore incurred costs over a longer period of time.Additional costs for healthy living people were especially associ-ated with diseases related to old age. This is shown in Fig. 2A, in

Page 5: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

Table 2Per person revenues, total costs and balances for the three cohorts (in Euros).

Life expectancyat age 20

Per personrevenues

Per personcosts

Balanceper person

A. LifetimeHealthy living 121,503 123,121 �1617Smokers 103,716 106,435 �2719Obese 111,663 118,120 �6457B. Per life year (on average)Healthy living 84.4 1886 1911 �25Smoking 77.4 1807 1855 �47Obese 79.9 1865 1973 �108

T.A. Kanters et al. / Social Science & Medicine 76 (2013) 150e158154

which positive values depict higher costs for healthy living people.Until just over the age of 70, health care costs were highest for theunhealthy cohorts due to the high prevalence of lifestyle relateddiseases in those cohorts. In the long run, the healthy living cohortbecame most costly. During their thirties, average costs for womenare higher due to health care costs associated to pregnancies, whichis reflected in the figure.

Annual health care costs per person were almost equal amongthe three cohorts until the age of 40. From then on, annual costs perperson were highest for smokers and lowest for healthy livingpeople. If smokers/obese lived to the age of 80, their annual costs

Cost differences for smokers and obese people com

-400000

-200000

0

200000

400000

600000

800000

1000000

1200000

1400000

20 40 60

Age (year

Addi

tiona

l cos

ts (E

uros

)

Differences in REF revenues of smokers and obes

-200000

0

200000

400000

600000

800000

1000000

1200000

20 40 60

Age (yea

Addi

tiona

l ben

efits

(Eur

os)

A

B

Fig. 2. Healthy living cohort compared to smoking and obese cohorts. A. Cost differences frevenues of smokers and obese people compared to healthy living people.

were about V500 higher than for healthy living survivors of thesame age. The differences with Fig. 2A are explained by the inclu-sion of life expectancy in this case.

Not only costs but also lifetime revenues for the insurer werehighest for the healthy living cohort. Fig. 2B shows REF contribu-tions for the cohorts by age, with positive values indicating higherrevenues for the healthy living cohort compared to the unhealthyliving cohorts. Fig. 2A and B show that revenues and costs followedsimilar patterns according to age. The negative values until the ageof 70 were much smaller than the positive differences for higherages. Revenues over the entire lifespanwere therefore larger for thehealthy living cohort than for the unhealthy living cohorts.

Fig. 2B shows the REF revenues for the cohorts as a whole. Fig. 3shows the same REF revenues for the three cohorts per survivor.The stepwise pattern is explained by the use of age-classes in thedemographic risk adjuster. Until the age of 40, revenues persurvivor were approximately equal for the three cohorts, and wereincreasing with age. From the age of 40 until the age of 90 insurersreceivedmost REF contributions for smoking survivors and least forhealthy living survivors. This difference mainly stemmed fromrelated PCGs and DCGs (included in the CDM as being related tosmoking and/or obesity). At the age of ninety, revenues were at themaximum of V4500 per survivor, and almost equal for the threecohorts. Beyond the age of 90, healthy living survivors were

pared to healthy living people

80 100

s)

Smokers

Obese

e people compared to healthy living people

80 100

rs)

Smokers

Obese

or smokers and obese people compared to healthy living people. B. Differences in REF

Page 6: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

Fig. 3. Risk equalization revenues per survivor by age class for smokers, obese and healthy living people.

T.A. Kanters et al. / Social Science & Medicine 76 (2013) 150e158 155

compensated more than smoking survivors, particularly due to therevenues associated with unrelated PCGs (i.e. unrelated tounhealthy living).

REF revenues for obese survivors were higher than revenues forhealthy living survivors at all ages. This was mainly the result ofdifferences in contributions for related PCG revenues.

The balance (revenues minus costs) for the three cohorts showssimilar patterns. This is shown in Fig. 4, in which a positive balanceindicates that revenues exceed costs. From around the age of 75, thebalance between costs and revenues became negative for each ofthe three cohorts. Within the age-classes, revenues were relativelyconstant due to the use of 5-year age-classes in the risk adjustmentscheme.

Until the age of 82, the healthy living cohort had the highestbalance of the three groups, i.e. was most profitable. For higherages, the smoking cohort showed to be the least unprofitable forinsurers. The obese cohort was financially least attractive forinsurers until the age of 85. After that, the net loss for the obesecohort was smaller than for the healthy living cohort.

Including life expectancy in the analyses led to superiorattractiveness for healthy living survivors, regardless of age. At themaximum, an insurer lost V350 more per year on an unhealthy

Fig. 4. Total balance (revenues min

living survivor, smoking or obese, compared to a healthy livingsurvivor of the same age.

Table 2 (panel A) shows the costs and revenues per cohort overthe entire lifespan. It demonstrates that healthy people incurredmost costs during their lives. Smokers were ‘cheapest’, while obesepeople took an intermediate position. This rank orderwas similar tothat presented for total health care costs (Van Baal et al., 2008).Table 2 also presents the total revenues for insurers for the differentcohorts, having the same rank order. Comparing costs and revenuesindicated that over the entire lifespan an insurer incurred losses onall three cohorts. Total lifetime health care costs for all 3000 peoplein the analyses combined were 3.2% larger than revenues for thosepeople. Over the entire lifetime, the healthy living cohort was mostattractive for the insurer, as losses were smallest for this cohort. Asshown in Table 2, an insurer could expect to loseV1617 on a healthyliving person; V2719 on a smoking person; and V6457 on an obeseperson (assuming that these persons joined at age 20 and remainedinsured until death). Obese people therefore were financially leastattractive. Table 2B shows that after adjusting for differences in lifeexpectancy, the healthy cohort remained the most attractive,whereas the obesewere still least attractive for health insurers. Thisimplies that insurers still have some financial incentive to engage in

us costs) for the three cohorts.

Page 7: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

Table 4Short term incentives (5 years) for prevention/risk selection (in Euros).

Rank Limited RE Broader RE

Group Profit/loss Group Profit/loss

1 Healthy 30e34 þ649 Healthy 70e74 þ2792 Smoking 30e34 þ641 Smoking 30e34 þ1993 Obese 30e34 þ595 Healthy 30e34 þ1994 Healthy 25e29 þ477 Obese 30e34 þ1775 Smoking 25e29 þ472 Healthy 55e59 þ171. . .

41 Healthy 85e89 �1878 Healthy 85e89 �95142 Obese 90þ �2036 Obese 90þ �100043 Smoking 90þ �2131 Smoking 90þ �111544 Smoking 85e89 �2348 Smoking 85e89 �123245 Obese 85e89 �2367 Obese 85e89 �1282

T.A. Kanters et al. / Social Science & Medicine 76 (2013) 150e158156

preventive activities (or, alternatively, risk selection), but also thatthe incentives are rather modest.

Next, in order to study the effects of an improvement in riskequalization on the incentives for prevention, the results from thecurrent Dutch system were compared to a less sophisticatedmechanism, using only age and gender to adjust for risk. All otherthings, such as lifetime costs for the three cohorts and total reve-nues (adding up revenues across the three cohorts), were heldconstant (implying that total costs still exceeded total revenueswith 3.2%).

Table 3 shows that the healthy living cohort was most profitablefor an insurer in a system with only age and gender as riskadjusters. The differences compared to the healthy living cohortindicated the incentives for risk selection, and consequently, theincentives for prevention. The financial incentive for prevention ofsmoking (obesity) was more than V4600 (V8500) in the limitedrisk equalization scheme. As expected, these differences decreasedwhen a more sophisticated risk equalization scheme includinghealth proxies was applied; the financial incentive for smokingprevention equaled V1100 and for obesity the financial incentivewas over V4800. This obviously discourages risk selection whenmore risk factors are added, while simultaneously incentives toengage in prevention are reduced. The financial incentive forprevention was reduced with approximately V3500, for bothsmoking and obesity.

We also considered the relative attractiveness of differentgroups per age class of 5 years to study insurers’ short-termincentives (rather than considering the life-time perspective as inTables 2 and 3). The results in Table 4 indicate that in case of limitedrisk equalization, age was an important factor to identify profitablegroups. Young people were profitable, old people were associatedwith considerable losses irrespective of their lifestyle. When amoresophisticated model was in place, lifestyle became an importantidentifier for profitability as well. Older people remained associatedto losses however, although absolute losses for these groupsdecreased. These results demonstrate clearly that in the short runselection of younger insureds irrespective of their lifestyle is muchmore profitable than investing in prevention to improve lifestyles.

Discussion

In this paper we have shown that insurers can expect differ-ences in health care costs and risk equalization revenues betweenhealthy living, smoking and obese people. Aggregated over theentire lifetime, both costs and revenues are highest for the healthyliving cohort. Subtracting costs from revenues indicates thathealthy insureds are still most attractive, followed by smokers andfinally obese individuals. These findings indicate that thereremains a (rather modest) incentive for insurers to engage in riskselection, especially in the case of obesity. The improvement of theREF by introducing more risk adjusters, related to the health ofinsureds, reduced these incentives, by reducing the differencesbetween the cohorts substantially. At the same time, improvement

Table 3Differences in incentives for different risk equalization (RE) schemes (in Euros).

LimitedRE

Comparedto healthyliving

BroaderRE

Comparedto healthyliving

Decrease inincentive: broaderRE compared tolimited RE

Healthyliving

790 �1617

Smoking �3831 �4620 �2719 �1101 3519Obese �7752 �8542 �6457 �4840 3702

also reduces the incentive to engage in preventive action, hencehighlighting the paradox between improvement of risk equaliza-tion and sustained incentives for prevention for insurers in thepost-reform Dutch context. This shows that countries that rely oninsurers (to some extent) to engage in preventive activities, shouldbe aware of the fact that this paradox exists. Improvement of therisk equalization scheme under specific circumstances, such as thecase described here, can reduce the financial incentives forinsurers to improve the lifestyle and therefore health of theirinsureds.

The analyses show that per life year healthy living people aremost attractive (i.e. the least unprofitable). In this respect, ourresults confirm the findings of Stam and Van de Ven (2008), whodemonstrated that people with lower health states are under-compensated in the Dutch risk equalization scheme. In our studyabsolute differences in life year attractiveness were small, so thatinsurer’s incentives to promote health are limited and costs ofprevention programsmay often exceed the potential revenues to begained.

Costs and benefits were highest for people in high age classes. Itis important to note however, that the number of people in theseclasses is limited. This is also reflected in Fig. 2. The decline in totalcosts and revenues observed in older age groups is especially due tothe limited number of survivors in those age groups.

Given the aim of the study, the insurer’s perspective wasadopted throughout our analyses. Therefore, only health care costsand revenues relevant to the insurers were taken into account.Importantly, therefore, we did not include any costs falling on othersectors or budgets. For instance, many long term nursing carefacilities in the Netherlands are financed through another financingscheme (i.e. Exceptional Medical Expenses Act; AWBZ). This isimportant to note, since it stresses that results presented here arelikely to be country specific to some extent. Moreover, costs andeffectiveness of various prevention programs targeted at reducingsmoking and obesity are beyond the scope of this paper. Wemerelyconcentrated on the potential incentives for health promotion byhealth insurance companies.

An important implication of our findings is that if countries wishto improve their risk equalization scheme, which in many cases isessential for an adequate functioning of a more competitive healthinsurance market, policy makers need to be aware of the fact thatsuccess in this respect necessarily implies reduced incentives forprevention of unhealthy lifestyles. When the incentives forprevention become relatively small, additional (financial or other)incentives may be required in order to stimulate insurers to engagein prevention. While we have focused here on prevention ofunhealthy lifestyles, it is likely that other types of prevention (oreven health improvement in general) may be affected in a similarway. Targeted financial incentives (for instance on the basis of

Page 8: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

T.A. Kanters et al. / Social Science & Medicine 76 (2013) 150e158 157

reported activities or, better still, success rates) may be required inthose cases.

It is interesting to note, moreover, that, while our study hasfocused on the Dutch context, the question of where the financialrisks (or revenues when such behavior saves money) of unhealthybehavior by insureds exactly fall, seems to be an understudiedissue. In case of imperfect or absent risk equalization, openenrollment and obligatory acceptance of insureds, this risk largelyfalls on insurers. They cannot charge high risks more, nor are theycompensated for high risks caused by individual behavior in sucha context. In the case of completely (and perfectly) risk-adjustedprivately paid premiums, the financial risk of unhealthy behaviorfalls on the individual, since the insurer will translate this risk intothe premium paid by individuals. In the case of perfect risk-equalization and fully community rated premiums, the risk isborne collectively (i.e. jointly by all individuals). In that case, indi-vidual behavior is translated into higher premiums for all. Moreresearch investigating what an ‘optimal’ way of dividing thefinancial risks related to unhealthy behavior would be, seemswarranted. However, in this respect, it is important to note that thepurpose of risk equalization is to correct for predictable losses andgains for insurers; the purpose is not per se to stimulate insurers toengage in prevention. So the gain of a better level playing field forcompeting health insurers comes at the prices of reduced incen-tives for prevention. An important question is whether and howrisk equalization schemes should be improved to avoid risk selec-tion of specific groups in terms of preventable, unhealthy behavior.This discussion strikes at the heart of what risk-equalization aims todo. It should create a level playing field, while at the same timemaintaining appropriate incentives for improving efficiency. Policymakers therefore face a trade-off between efficiency and riskselection in improving risk equalization. For instance, by simplyusing a variable ‘smoker’ or ‘obese’ in the risk equalization scheme,one could in principle eliminate risk selection incentives on thesegroups. At the same time, this would reduce all incentives forpreventive effort (and potentially for efficient purchasing of carerelated to these risk factors). How to strike an optimal balance inthis respect to some extent remains a political choice, one that willalso depend on other incentives (e.g. other incentives for preven-tion separate from the REF).

Some limitations of our study are important to note. First of all,we have combined data from different sources. The REF contribu-tions are based on themacro budget and the Costs of Illness study isbased on figures from Statistics Netherlands. For consistency, thefigures from the Costs of Illness studywere therefore linearly scaledback to the macro budget level. The figures from StatisticsNetherlands are composed differently than the macro budget,however. Scaling back the Costs of Illness numbers incorrectlyincludes deductibles and supplementary insurances in the costs. In2008, 94% of insureds had a supplementary insurance and allinsureds were faced with a minimal yearly deductible of V150.

As the CDM does not present estimates on an individual level,we could not estimate the number of diseases per individual. TheCDM is programmed as a deterministic Markov model, i.e. thesimulation model calculates the expected outcomes in one runwhich differs from a so-called micro-simulation or Monte Carlosimulation model. The predictions of the CDM are made at thegroup level, not at the individual level. Therefore, we could notperform statistical tests that individual level data would allow. Asunit of analysis, we chose cohorts of 500 men and 500 women,purely for reasons of convenience. Furthermore, in reality, comor-bidity is often larger than the statistical correlation used in ouranalyses to estimate the paybacks into the REF for ‘healthy’ PCG0and DCG0 individuals. This implies that the disease-free fraction ofthe population was underestimated; actual PCG and DCG

classification are concentrated among fewer people, and morepeople are disease-free in reality.

Under the Health Insurance Act, insureds can maximally becategorized into one (the highest) DCG. This could not be repro-duced in the CDM, because of the absence of data at the individuallevel. Insureds were categorized in all DCGs corresponding to theirdiseases, which might have resulted in an overestimation of DCGclassifications.

Moreover, this study only considered the four most importantrisk adjusters of the Dutch risk equalization formula (age, gender,PCGs and DCGs), which collectively explain 17% of the total 22% ofthe variance of health care costs that is captured by the actual2008-formula. The CDM does not provide data on the other riskadjusters e region, social economic status and source of income e

which therefore could not be incorporated. The relation betweenthese social factors and unhealthy lifestyles has been demonstratedin several publications (for instance Stronks, Van de Mheen,Looman, & Mackenbach, 1997; Van Lenthe, Droomers, Schrijvers,& Mackenbach, 2000; Van Lenthe et al., 2004). The small per lifeyear differences between the cohorts, shown in Table 2B are smallerthan the risk equalization contributions for socioeconomic status(SES). The influence of SES on these results therefore has to beexamined in future research. However, health care utilization isalso related to socioeconomic factors, and the direct health proxiesPCGs and DCGs therefore incorporate a large share of the explan-atory power of the social risk adjusters. Discussing improvementsof the current risk equalization model is beyond the scope of thecurrent paper. It is good to highlight, however, that such discus-sions are ongoing (e.g. Stam, Van Vliet, & Van de Ven, 2010; VanKleef & Van Vliet, 2012). Moreover, it is good to emphasize in thatcontext that our findings may be taken as illustration of the moregeneral problem of an insufficiently sophisticated REF that leads topredictable losses in specific groups of patients. It is important torecognize the consequences of such problems and to find adequatesecond best solutions, where possible (e.g. Glazer & McGuire,2002).

Some features of our findings need to be emphasized too.Overall, we found a negative balance between total costs and totalREF revenues for each of the three cohorts. The negative balance islikely to be explained by two factors. One element is that there is animbalance between revenues and costs in the actual Dutch healthcare sector, which is even larger than the 3.2% difference found inthis study (CVZ, 2008). The introduced competition appears to havelead to insurers incurring losses on health insurance products.These losses are corrected for by ex post equalization mechanismsin the Dutch system. Therefore, ex post equalization mechanismsfurther decrease the incentives for prevention, as it reduces thefinancial risks the insurer bears. Ex post risk equalization could notbe included in these analyses, particularly because of the aggre-gation level of the classes distinguished in the CDM. AlthoughDutch policymakers aim to cut down on these ex post risk equal-ization mechanisms, currently these mechanisms are very impor-tant in the Dutch risk equalization scheme.

Another possible explanation is the difference in demographiccomposition of the models used in our analyses and the demog-raphy of the actual Dutch population, which may have resulted inthis discrepancy between revenues and expenditures. In addition,some risk adjusters were not incorporated in our study. Risk-adjusters such as region may, to a small extent, also pick up ondifferences in lifestyles between regions, influencing our results.

In decision making on prevention, the costs and effects ofprevention programs are obviously taken into account. The differ-ences of per life year net revenues and costs between the cohortsare quite small, as shown in Table 2 (Section B). Consequently, theincentive for prevention is also limited; many prevention programs

Page 9: A new prevention paradox: The trade-off between reducing incentives for risk selection and increasing the incentives for prevention for health insurers

T.A. Kanters et al. / Social Science & Medicine 76 (2013) 150e158158

cost more per year than the V22 (V83) saved on smokers (obesepeople). For instance, for face-to-face smoking cessation programs,costs per quit attempt range fromV12 toV282 (Feenstra, Hamberg-van Reenen, Hoogenveen, & Rutten-van Mölken, 2005). Costs fordietary advice and physical advice range from V14 to V721 (Bogerset al., 2010). In that sense, cheap forms of risk selection may bemore attractive. On the other hand, the low mobility of insuredsprovides the insurer with a larger incentive to engage in preven-tion, as the benefits are likely to be accrued in the longer run.

It is also important to note that in addition to financial incen-tives, there might be other motives for insurers to engage inprevention. For instance, an insurer can experience reputationaldamage when not engaging in preventive activities. Moreover,(reimbursing) specific preventive activities may attract new(health-conscious) insureds and may thus be (and have been) usedas marketing tool. Governments may (legally) require certainactivities and insurers might engage in prevention for marketingreasons as well. Premium differentiation according to lifestyle isnot allowed in the Netherlands.

Concluding, we hypothesized that discouraging risk selectioncomes at the prices of reduced incentives for prevention; a newprevention ‘paradox’. This paper has shown that improvement ofrisk equalization indeed reduces incentives for prevention.Governments aiming for preventive activities by insurers as well assophisticated risk equalization schemes need to be aware of thisfact. Achieving both goals is likely to require additional efforts.

Acknowledgments

The authors would like to thank the participants of the 8thEuropean Conference on Health Economics 2010 (ECHE) for theiruseful comments on an earlier version of the paper. The authorswould like to thank Hans van de Hoek for his comments during ourstudy. Werner Brouwer gratefully acknowledges Netspar funding inthe context of the project ‘Living longer in good health’.

References

Bogers, R. P., Barte, J. C. M., Schipper, C. M. A., Vijgen, S. M. C., De Hollander, E. L.,Tariq, L., et al. (2010). Relationship between costs of lifestyle interventions andweight loss in overweight adults. Obesity Reviews, 11, 51e61.

CVZ. (2008). Health care figures, quarterly communication, 2007-4th quarter. Diemen:Health Care Insurance Board (in Dutch).

Feenstra, T. L., Hamberg-van Reenen, H. H., Hoogenveen, R. T., & Rutten-vanMölken, M. P. M. H. (2005). Cost-effectiveness of face-to-face smoking cessationinterventions: a dynamic modeling study. Value in Health, 8(3), 178e190.

Glazer, J., & McGuire, T. G. (2002). Setting health plan premiums to ensure efficientquality in health care: minimum variance optimal risk adjustment. Journal ofPublic Economics, 84, 153e173.

Heijink, R., Noethen, M., Renaud, T., Koopmanschap, M., & Polder, J. (2008). Cost ofillness: an international comparison: Australia, Canada, France, Germany andthe Netherlands. Health Policy, 88(1), 49e61.

Hoogenveen, R. T., Van Baal, P. H. M., & Boshuizen, H. C. (2010). Chronic diseaseprojections in heterogenous ageing populations: approximating multi-statemodels of joint distributions by modelling marginal distributions. Mathemat-ical Medicine and Biology, 27(1), 1e19.

Maarse, H., & Bartholomée, Y. (2007). A public-private analysis of the new Dutchhealth insurance system. European Journal of Health Economics, 8, 77e82.

MinVWS. (2007). Calculations division model 2008. The Hague: Ministry of Health(in Dutch).

OECD. (2005). Health at a glance, OECD indicators 2005. Paris: OECD Publishing.Rappange, D. R., Brouwer, W. B. F., Rutten, F. F. H., & Van Baal, P. H. M. (2010).

Lifestyle intervention: from cost savings to value for money. Journal of PublicHealth, 32(3), 440e447.

RIVM. (2008). Nationaal Kompas Volksgezondheid. http://www.nationaalkompas.nl/Accessed 15.07.08.

RIVM. (2011). Cost of illness in the Netherlands. http://www.costofillness.eu/Accessed 30.03.12.

Rooijen, M. R., De Jong, J. D., & Rijken, M. (2011). Regulated competition in healthcare: switching and barriers to switching in the Dutch health insurance system.BMC Health Services Research, 11, 95.

Slobbe, L. C. J., Kommer, G. J., Smit, J. M., Groen, J., Meerding, W. J., & Polder, J. J.(2006). Costs of illness in the Netherlands 2003. Bilthoven: Rijksinstituutvoor Volksgezondheid en Milieu, Nationaal Kompas Volksgezondheid (inDutch).

Stam, P. J. A., & Van de Ven, W. P. M. M. (2008). Evaluation risk equalization amonghealth insurers. Tijdschrift voor Gezondheidswetenschappen, 86, 92e100, (inDutch).

Stam, P. J. A., Van Vliet, R. C. J. A., & Van de Ven, W. P. P. M. (2010). A limited-samplebenchmark approach to assess and improve the performance of risk equaliza-tion models. Journal of Health Economics, 29, 426e437.

Stronks, K., Van de Mheen, H. D., Looman, C. W. N., & Mackenbach, J. P. (1997).Cultural, material, and psychosocial correlates of the socioeconomic gradient insmoking behavior among adults. Preventive Medicine, 26, 754e766.

Van Baal, P. H., Engelfriet, P. M., Hoogenveen, R. T., Poos, M. J., Van den Dungen, C., &Boshuizen, H. C. (2011). Estimating and comparing incidence and prevalence ofchronic diseases by combining GP registry data: the role of uncertainty. BMCPublic Health, 11, 163.

Van Baal, P. H. M., Feenstra, T. L., Hoogenveen, R. T., De Wit, G. A., & Brouwer, W. B. F.(2007). Unrelated medical care in life years gained and the cost utility ofprimary prevention: in search of a ‘perfect’ cost-utility ratio. Health Economics,16, 421e433.

Van Baal, P. H. M., Feenstra, T. L., Polder, J. J., Hoogenveen, R. T., & Brouwer, W. B. F.(2011). Economic evaluation and the postponement of health care costs. HealthEconomics, 20, 432e445.

Van Baal, P. H. M., Hoogenveen, R. T., De Wit, G. A., & Boshuizen, H. C. (2006).Estimating health-adjusted life expectancy conditional on risk factors: resultsfor smoking and obesity. Population Health Metrics, 4, 14.

Van Baal, P. H. M., Polder, J. J., De Wit, G. A., Hoogenveen, R. T., Feenstra, T. L.,Boshuizen, H. C., et al. (2008). Lifetime medical costs of obesity: prevention nocure for increasing health expenditure. PLoS Medicine, 5, 2.

Van de Ven, W. P. P. M. (2011). Risk adjustment and risk equalization: what needs tobe done? Health Economics Policy and Law, 6, 147e156.

Van de Ven, W. P. P. M., Beck, K., Van de Voorde, C., Wasem, J., & Zmora, I. (2007).Risk adjustment and risk selection in Europe: 6 years later. Health Policy, 83,162e179.

Van de Ven,W. P. P. M., & Schut, F. T. (2008). Universal mandatory health insurance inthe Netherlands: a model for the United States? Health Affairs, 27(3), 771e781.

Van de Ven, W. P. P. M., Van Vliet, R. J. C. A., & Lamers, L. M. (2004). Health-adjustedpremium subsidies in the Netherlands. Health Affairs, 23(3), 45e54.

Van Kleef, R. C., & Van Vliet, R. C. J. A. (2012). Improving risk equalizationusing multiple-year high cost as a health indicator. Medical Care, 50(2),140e144.

Van Lenthe, F. J., Droomers, M., Schrijvers, C. T. M., & Mackenbach, J. P. (2000). Socio-demographic variables and 6 year change in body mass index: longitudinalresults from the GLOBE study. International Journal of Obesity and RelatedMetabolic Disorders, 24(8), 1077e1084.

Van Lenthe, F. J., Schrijvers, C. T. M., Droomers, M., Joung, I. M. A., Louwman, M. J., &Mackenbach, J. P. (2004). Investigating explanations of socio-economicinequalities in health. European Journal of Public Health, 14, 63e70.

Van Vliet, R. J. C. A., Van Asselt, M. M., De Groot, N., Mazzola, G. J., Notenboom, A., &Goudriaan, R. (2007). Calculation risk equalization contributions risk equalizationmodel 2008: WOR322b. Den Haag: APE, bv (APE-rapport nr 519) (in Dutch).