research

letter

Diabetes, Obesity and Metabolism 14: 375–378, 2012.Published 2011. This article is a U.S. Government work and is in the public domain in the USA.research letter

Low HDL predicts differential blood pressure effects fromtwo weight-loss approaches: a secondary analysis of bloodpressure from a randomized, clinical weight-loss trial

Examining predictors of blood-pressure (BP) response to weight-loss diets might provide insight into mechanisms and help guide clinical care.We examined whether certain baseline patient characteristics (e.g. diet, medical history and laboratory tests) predicted BP response to twoweight-loss diet approaches that differ in macronutrient content. One hundred and forty-six overweight adult outpatients were randomized toeither a low-carbohydrate diet (N = 72) or orlistat plus a low-fat diet (N = 74) for 48 weeks. Predictors of BP reduction were evaluated usinga structured approach and random effects regression models. Participants were 56% African-American, 72% male and 53 (±10) years-old. Ofthe variables considered, low baseline high-density lipoprotein (HDL) predicted greater reduction in BP in those patients who received thelow-carbohydrate diet (p = 0.03 for systolic BP; p = 0.03 for diastolic BP and p = 0.02 for mean arterial pressure). A low HDL level mayidentify patients who will have greater BP improvement on a low-carbohydrate diet.Keywords: clinical trial, dietary intervention, dyslipidaemia, hypertension, obesity therapy, weight-loss therapy

Date submitted 4 July 2011; date of first decision 8 August 2011; date of final acceptance 1 November 2011

IntroductionAlthough weight loss improves blood pressure (BP), less isunderstood about how alterations in dietary macronutrientcontent (with the intention of weight loss) interact with patientcharacteristics such as demographics or metabolic profile tolead to BP improvement [1]. Our group previously foundthat a low-carbohydrate ketogenic diet (LCKD) was moreeffective than orlistat plus a low-fat diet (OLFD) for lower-ing BP (SBP/DBP −5.9/−4.5 mmHg vs. +1.5/+0.4 mmHg,p < 0.001) and reducing the need for BP medications (dosedecreased in 47 vs. 21%), despite similar weight loss betweengroups over 48 weeks [2]. The current study’s aim was to iden-tify baseline patient characteristics that predicted differentialeffects of diet on BP.

MethodsThe parent study was a randomized study that compared theeffects of a diet (LCKD) against a drug-diet combination(OLFD) on body weight over 48 weeks; BP was a pre-specified secondary endpoint. Further details of the study havebeen published [2]. The study was approved by the DurhamVeteran Affairs Institutional Review Board, and all participantsprovided informed consent.

Participants had to be 18–70 years-old and eligible fora weight-loss medication [3]. Relevant exclusion criteria

Correspondence to: Christy Boling Turer, Department of Pediatrics, University of TexasSouthwestern, 5323 Harry Hines Boulevard, Dallas, TX 75390-9063, USA.E-mail: Christy.Turer@UTSouthwestern.edu

included BP ≥160/100 mm Hg and elevated serum creatinine(>1.5 mg/dl for men, >1.3 mg/dl for women). Diet-specificinterventions were delivered in group meetings held biweeklyfor 24 weeks, then monthly through 48 weeks. LCKD partici-pants were instructed to restrict carbohydrate intake (initially<20 g/day). OLFD participants were instructed to restrictdietary fat (<30%); take orlistat (120 mg three times a day)and reduce caloric intake (500 kcal deficit).

BP was obtained at 19 time points over 48 weeks usinga calibrated, automated sphygmomanometer and appropriatecuff sizes [4]. BP was measured twice after participants satquietly for 5 min. Measures differing by ≥10% were repeateduntil the final two measures differed by <10%; analyses usedthe mean of the final two measures. Waist circumference(WC) was measured twice (and averaged) using a non-elastictape measure applied at the umbilicus. Fasting blood sampleswere processed by the Durham VA clinical laboratory usingstandardized techniques.

Analyses explored the relationship between BP at 48 weeksand potential predictors assessed at baseline. Diet measure-ments were averaged from 4-day food records completedbefore the intervention. Daily activity was measured usingthe Framingham Physical Activity score [5]. Medical his-tory and laboratory analyses included (i) updated criteria formetabolic syndrome (MetS), individually and summed as acontinuous score (0 to 5 criteria); (ii) high-sensitivity C-reactive protein (hsCRP); (iii) uric acid and (iv) haemoglobinA1C(HbA1C) [6]. MetS criteria included: WC >102 cm inmales, >88 cm in women; high-density lipoprotein (HDL)cholesterol <40 mg/dl in males or <50 mg/dl in women

research letter DIABETES, OBESITY AND METABOLISM

Table 1. Sample characteristics by weight-loss intervention and predictors of mean arterial blood pressure lowering.

Sample characteristics Predictor analysis

N (%) or mean (SD)‘A’ predicts the effect of treatment ‘B’ on BP when A is notcorrelated with B∗, and A × B × Time is significant†:

Type ofcharacteristic Potential predictor LCKD N = 72 OLFD N = 74

A × B × Timecoefficient p

Predictionsuggested

Baseline model Treatment group × time (in weeks) — — −0.13 <0.0001 NADemographics

and activityMean age, in years 53 (10) 52 (9) −0.001 0.8 NoGender, % female 20 (28) 21 (28) −0.02 0.7 NoRace, % African-American 40 (56) 41 (55) −0.02 0.7 NoMean physical activity score 30 (4) 30 (4) 0.3 0.3 No

Medical history Mean metabolic syndrome criteria 4 (1) 4 (1) −0.07 0.05 BorderlineWaist circumference criteria, % 72 (100) 74 (100) — — —Blood sugar criteria, % 55 (76) 53 (72) −0.03 0.7 NoTriglyceride criteria, % 35 (49) 31 (42) −0.04 0.6 NoBP criteria, % 64 (89) 66 (89) −0.1 0.2 NoHDL criteria, % 58 (81) 53 (72) −0.2 0.02 Yes

Laboratory data Mean hsCRP 0.6 (0.5) 0.8 (0.7) 0.01 0.7 NoMean haemoglobin A1C, % 6.3 (1.1) 6.4 (1.3) −0.02 0.5 NoMean uric acid, mg/dl 6.3 (1.6) 6.4 (1.6) −0.04 0.05 Borderline

Daily foodintake

Mean dietary fat,‡ g/day 105 (50) 100 (41) −0.01 0.4 NoMean dietary protein,‡ g/day 96 (30) 97 (35) −0.02 0.4 NoMean dietary fibre, g/day 17 (7) 16 (6) −0.03 0.4 NoMean dietary sodium, mg/day 4,082 (1,505) 3,988 (1,361) −0.003 0.3 NoMean calories, kcal/day 2,3882 (873) 2,185 (715) −0.0001 0.2 NoMean dietary carbohydrate,‡ g/day 262 (102) 223 (87) −0.0004 0.2 No

LCKD, low-carbohydrate ketogenic diet; OLFD, orlistat low-fat diet; BP, blood pressure; SD, standard deviation; HDL, high-density lipoprotein cholesterol;hsCRP, high-sensitivity C-reactive protein; MAP, mean arterial pressure.∗No baseline predictors were correlated with treatment assignment.†p = 0.05 tested significance; A = predictor, B = treatment, T = time in weeks.‡Proportion of baseline dietary intake from fat, protein and carbohydrate did not differ between groups, and was not predictive of MAP lowering (datanot shown).

or taking an HDL-elevating medication (i.e. nicotinic acid andfibrate); triglycerides (TG)≥150 mg/dl or taking a TG-loweringmedication (i.e. fish oil); BP ≥130/85 mmHg or a diagnosisof hypertension requiring medication and fasting blood glu-cose ≥100 mg/dl, taking diabetes medication, or diagnosis ofdiabetes.

Analysis

The primary outcome was BP at 48 weeks assessed as meanarterial pressure (MAP), systolic BP (SBP) and diastolic BP(DBP). MAP was adjusted for heart rate [7]. Predictors weretested using the structured analytic plan suggested by Kraemeret al. [8] and random effects models (random effects includedintercept and slope terms) across 19 time points. A vari-able was defined as a predictor when it preceded treatment(i.e. assessed at baseline), was not correlated with treatmentassignment, and the predictor’s interaction with treatment andtime (predictor × treatment × time) was statistically signifi-cant (p < 0.05).

A two-sided alpha-level of 0.05 and the 146 randomized par-ticipants provided 80% power to detect a treatment differencein �SBP of 6 mmHg. SAS Enterprise-Guide Version 4.2 (SAS,Cary, NC, USA) was used for analyses.

ResultsParticipants in the two interventions were similar in age, genderand race (Table 1). The majority (92%) met ATPIII criteria forMetS.

Given that MAP incorporates both SBP and DBP, the detailsof the MAP analysis are displayed in Table 1. The resultsof the SBP and DBP analyses were similar. All predictorsmet the first two criteria for prediction. Each was measuredat baseline (and therefore preceded the treatments) and nopredictor was correlated with treatment assignment. Onlyone variable met the third of the prediction criteria: theinteraction of baseline low HDL with both treatment andtime was significant (SBP p = 0.02, DBP p = 0.03, MAPp = 0.02). Figure 1 shows the separation in blood-pressureresponse over time for the Low HDL × Treatment subgroups.The modelled mean change (Figure 1A) accounts for baselinedifferences in initial MAP. Actual MAP changes are displayed inFigure 1B and C.

Uric acid and the sum of the MetS criteria did not meet thecut-point for statistical significance but the directions of theassociations indicate that BP may improve more on an LCKDfor patients with higher vs. lower levels of uric acid, and agreater as compared with a lower number of MetS criteria atbaseline. The analysis of baseline food records suggested no

376 Turer et al. Volume 14 No. 4 April 2012

DIABETES, OBESITY AND METABOLISM research letterA

B

C

Figure 1. Modelled and actual mean change in mean arterial pressure(MAP) over time and by high-density lipoprotein (HDL) cholesterolsubgroups (low and normal) and treatment groups [low-carbohydrateketogenic diet (LCKD) or orlistat low-fat diet (OLFD)]. (A) Modelledmean change in MAP, (B) Actual mean change in MAP in low HDL groupsand (C) Actual mean change in MAP in normal HDL groups.

differential effect of baseline sodium or baseline macronutrientintake by treatment assignment (e.g. BP improvement oneither weight-loss diet did not differ by initial sodiumintake).

DiscussionThis study identified low HDL as a predictor of BPimprovement on an LCKD compared to OLFD, suggesting thatlow HDL may identify those patients likely to experience morepronounced BP lowering on an LCKD. Prospective studies areneeded to validate the potential role of HDL as a biomarkerthat could be used to target use of this intervention.

The reason for the relationship between low HDL andBP response to the LCKD is unclear. Carbohydrate-restricted

diets elevate HDL, and HDL improves endothelial function byincreasing nitric oxide, decreasing vascular inflammation andinhibiting thrombosis [9]. Increasing the HDL of subjects withlow HDL may, therefore, restore endothelial function [10].

The predictor analysis also suggested that uric acid elevationsand MetS criteria may identify patients likely to experiencemore pronounced BP lowering on an LCKD. Characteristicsthat did not appear to predict the impact of treatment on BPincluded sodium intake and baseline BP control.

This study was strengthened by the prospective randomizedclinical trial design, intention-to-treat modelling methodology,minority representation and the complexity of obesity-relateddisease in participants. Subjects had numerous comorbidities,and were taking multiple chronic medications. Generalisabilityof clinical trials requires inclusion of subjects such as these.

Several limitations should be acknowledged. This is a sec-ondary data analysis; future randomized clinical trials mayconsider stratifying by low HDL. Several predictors weretested—the identified predictor could be a random findingin the setting of multiple testing. This seems unlikely givenresults were robust for SBP, DBP and MAP. BP medicationchanges were not assessed. Controlling for BP medicationchanges may have accentuated the noted effect, however,because more LCKD than OLFD participants reduced BPmedications. Unfortunately, there is no standard method toaccount for BP medication changes over time. Finally, detectinglongitudinal predictors is rare and requires a large sample size;some predictors may have not achieved statistical significanceas a result of inadequate power [11].

In conclusion, this study suggests that an LCKD may beparticularly useful for lowering BP in overweight patients withlow HDL. Further study is needed to evaluate the physiologiclink among BP, carbohydrate intake, and HDL level andpossibly other markers of cardiometabolic risk, over time.

C. B. Turer1,2, I. H. Bernstein,3,4 D. E. Edelman5,6 &W. S. Yancy, Jr5,6

1Department of Pediatrics, University of Texas Southwestern,Dallas, TX, USA

2Department of Internal Medicine, University of TexasSouthwestern, Dallas, TX, USA

3Department of Clinical Sciences, University of TexasSouthwestern School of Biomedical Sciences, Dallas, TX, USA

4Department of Allied Health Administration, University ofTexas Southwestern School of Biomedical Sciences, Dallas, TX,

USA5Center for Health Services Research in Primary Care,

Department of Veterans’ Affairs Medical Center, Durham, NC,USA

6Department of Medicine, Duke University Medical Center,Durham, NC, USA

AcknowledgementsFunding for the parent study was provided by the Departmentof Veterans Affairs (CLIN-5-03F). Dr W. S. Y. was supported

Volume 14 No. 4 April 2012 doi:10.1111/j.1463-1326.2011.01531.x 377

research letter DIABETES, OBESITY AND METABOLISM

by a VA Health Services Research Career Development Award(RCD 02-183-1). Dr. C. B. T. was a Health Services Researchfellow during the writing of this manuscript.

Conflict of InterestThe authors have no conflicts of interest to disclose.

C. B. T. conceived of and designed the secondary analysis,performed the statistical analysis, and drafted the manuscript.I. H. B. oversaw the statistical analysis and participated indrafting the manuscript. D. E. E. participated in the design ofthe analysis and drafting the manuscript. W. S. Y. conceivedof and designed the original clinical trial, collected the data,participated in the design of the analysis, oversaw the statisticalanalysis, and participated in the drafting of the manuscript. Allauthors read and approved the final manuscript.

References1. Brunner E, Rees K, Ward K, Burke M, Thorogood M. Dietary advice

for reducing cardiovascular risk. Cochrane Database Syst Rev 2007; 4:CD002128.

2. Yancy WS, Westman EC, McDuffie JR et al. A randomized trial of a low-carbohydrate diet vs orlistat plus a low-fat diet for weight loss. Arch InternMed 2010; 170: 136–145.

3. NIH; NHLBI NAASO. The Practical Guide: Identification, Evaluation, andTreatment of Overweight and Obesity in Adults. Rockville, MD: NationalInstitutes of Health, 2000, NIH Publication No. 00-4084.

4. Chobanian AV, Bakris GL, Black HR et al. National Heart, Lung, and BloodInstitute Joint National Committee on Prevention, Detection, Evaluation,and Treatment of High Blood Pressure. National High Blood PressureEducation Program Coordinating Committee: The Seventh Report ofthe Joint National Committee on Prevention, Detection, Evaluation, andTreatment of High Blood Pressure. The JNC7 Report. JAMA 2003; 289:2560–2572.

5. Kannel WB, Sorlie P. Some health benefits of physical activity: theFramingham Study. Arch Intern Med 1979; 139: 857–861.

6. Alberti KGMM, Eckel RH, Grundy SM et al. Harmonizing the metabolicsyndrome: a joint interim statement of the International DiabetesFederation Task Force on epidemiology and prevention; NationalHeart, Lung, and Blood Institute; American Heart Association; WorldHeart Federation; International Atherosclerosis Society; and InternationalAssociation for the Study of Obesity. Circulation 2009; 120: 1640–1645.

7. Razminia M, Trivedi A, Molnar J et al. Validation of a new formula formean arterial pressure calculation: the new formula is superior to thestandard formula. Catheter Cardiovasc Interv 2004; 63: 419–425.

8. Kraemer HC, Wilson GT, Fairburn CG, Agras WS. Mediators and moderatorsof treatment effects in randomized clinical trials. Arch Gen Psychiatry2002; 59: 877–883.

9. Lowenstein CJ, Cameron SJ. High-density lipoprotein metabolism andendothelial function. Curr Opin Endocrinol Diabetes Obes 2010; 17:166–170.

10. Bisoendial RJ, Hovingh GK, Levels JH et al. Restoration of endothelialfunction by increasing high-density lipoprotein in subjects with isolatedlow high-density lipoprotein. Circulation 2003; 107: 2944–2948.

11. McClelland GH, Judd CM. Statistical difficulties of detecting interactions andmoderator effects. Psychol Bull 1993; 114: 376–390.

378 Turer et al. Volume 14 No. 4 April 2012