HbA1c as a Predictor of Diabetesand as an Outcome in the DiabetesPrevention Program: ARandomized Clinical TrialDiabetes Care 2015;38:51–58 | DOI: 10.2337/dc14-0886

OBJECTIVE

Glycated hemoglobin (HbA1c), a standard measure of chronic glycemia for man-aging diabetes, has been proposed to diagnose diabetes and identify people atrisk. The Diabetes Prevention Program (DPP) was a 3.2-year randomized clinicaltrial of preventing type 2 diabetes with a 10-year follow-up study, the DPP Out-comes Study (DPPOS). We evaluated baseline HbA1c as a predictor of diabetes anddetermined the effects of treatments on diabetes defined by an HbA1c ‡6.5% (48mmol/mol).

RESEARCH DESIGN AND METHODS

We randomized 3,234 nondiabetic adults at high risk of diabetes to placebo,metformin, or intensive lifestyle intervention and followed them for the devel-opment of diabetes as diagnosed by fasting plasma glucose (FPG) and 2-h postloadglucose (2hPG) concentrations (1997 American Diabetes Association [ADA] crite-ria). HbA1c was measured but not used for study eligibility or outcomes. We nowevaluate treatment effects in the 2,765 participants who did not have diabetes atbaseline according to FPG, 2hPG, or HbA1c (2010 ADA criteria).

RESULTS

Baseline HbA1c predicted incident diabetes in all treatment groups. Diabetes in-cidence defined by HbA1c ‡6.5% was reduced by 44% by metformin and 49% bylifestyle during the DPP and by 38% bymetformin and 29% by lifestyle throughoutfollow-up. Unlike the primary DPP and DPPOS findings based on glucose criteria,metformin and lifestyle were similarly effective in preventing diabetes defined byHbA1c.

CONCLUSIONS

HbA1c predicted incident diabetes. In contrast to the superiority of the lifestyleintervention on glucose-defined diabetes, metformin and lifestyle interventionshad similar effects in preventing HbA1c-defined diabetes. The long-term implica-tions for other health outcomes remain to be determined.

The Diabetes Prevention Program (DPP), including its long-term follow-up DiabetesPrevention Program Outcomes Study (DPPOS), was a randomized clinical trial eval-uating metformin and an intensive lifestyle (ILS) weight-loss intervention to preventor delay type 2 diabetes (1–3) defined by 1997 American Diabetes Association (ADA)criteria for fasting plasma glucose (FPG) and 2-h postload glucose (2hPG) (4).

Corresponding author: William C. Knowler,dppmail@bsc.gwu.edu.

Received 8 April 2014 and accepted 22 Septem-ber 2014.

Clinical trial reg. nos. NCT00004992, clinicaltrials.gov, and NCT00038727, clinicaltrials.gov.

This article contains Supplementary Data onlineat http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-0886/-/DC1.

*A complete list of the members of the DiabetesPrevention Program Research Group can befound in the Supplementary Data online, anda complete list of the members of the WritingGroup is available in the APPENDIX.

© 2015 by the American Diabetes Association.Readers may use this article as long as the workis properly cited, the use is educational and notfor profit, and the work is not altered.

Diabetes Prevention Program Research

Group*

Diabetes Care Volume 38, January 2015 51

EPIDEM

IOLO

GY/H

EALTH

SERVICES

RESEA

RCH

Current diagnostic criteria define diabe-tes using HbA1c $6.5% (48 mmol/mol)(5,6). HbA1c ,6.5% but $6.0% (42mmol/mol) was recommended to iden-tify persons at high risk of developingdiabetes who should be offered preven-tive interventions (5). The lower limitdefining high risk has also been set at$5.7% (39 mmol/mol) (6). To comparethese different diagnostic criteria andevaluate HbA1c as a risk indicator, weevaluated HbA1c as a predictor of diabe-tes and as an alternate outcome in theDPP and DPPOS.

RESEARCH DESIGN AND METHODS

Participants, Treatment, and Follow-upThe methods and primary findings havebeen described (1–3), and protocols areavailable at https://dppos.bsc.gwu.edu/web/dppos/home. The trial is registeredas NCT00004992 (DPP) and NCT00038727(DPPOS). The3,234nondiabeticparticipantshad the following risk factors: BMI $24kg/m2, FPG $5.3 mmol/L (95 mg/dL) and,7.0mmol/L (126mg/dL), and 2hPG$7.8mmol/L (140 mg/dL) and ,11.1 mmol/L(200mg/dL). There were minor exceptionsto these criteria: FPG,7.8 mmol/L beforethis diagnostic level was lowered with the1997 ADA criteria, no lower limit of FPG inthe American Indian centers, and BMI$22kg/m2 in Asian Americans (1).Participants were randomly assigned

to one of three treatment groups: pla-cebo, metformin 850 mg twice per day,or ILS (1,2). A tentative diabetes diagno-sis was made if FPG was $7.0 mmol/L(126 mg/dL) at a semiannual examina-tion or 2hPG during the annual 75-g oralglucose tolerance test was $11.1mmol/L (200 mg/dL), according to ADAcriteria (4). A diagnosis required confir-mation on a repeat of the same test(FPG or oral glucose tolerance test) asthat triggering the tentative diagnosis(1). Confirmed diagnoses were reportedto the participants and their health-careproviders, but study metformin or pla-cebo was still provided unless hypergly-cemia worsened to FPG $140 mg/dL(7.8 mmol/L) during DPP or HbA1c

$7.0% (53 mmol/mol) during DPPOS. Atthis point, the study drug was discontin-ued, and diabetes management, includ-ing metformin or other drug treatment,was transferred to the participant’s ownhealth-care provider. Otherwise, HbA1cresults were not used for determining el-igibility or outcomes.

Diabetes incidence rates during theDPP were 11.0 cases/100 person-yearsin theplacebogroup, 7.8 in themetformingroup, and 4.8 in the ILS group, represent-ing reductions in diabetes incidence of31% and 58% with metformin and ILScompared with placebo (1). Followingdrug unmasking and release of these re-sults in 2001, all participants were offeredlifestyle intervention (3). Metformin wascontinued in the original metformingroup, placebo was discontinued, andthe original ILS group was offered addi-tional lifestyle support. Of the original co-hort, 2,766 (88% of those alive andenrolled at the end of the DPP whetheror not they had developed diabetes)were enrolled in the DPPOS long-termfollow-up (3). During the DPPOS, afterall study participants had been offeredlifestyle intervention, diabetes rates inthe metformin and former placebogroups fell to rates similar to those ofthe original ILS group, which remainedrelatively stable (3).

HbA1c was measured at baseline,6 months, and 12 months and then an-nually in the level 1 central laboratory atthe University of Washington. The high-performance liquid chromatographymethod was aligned to the NationalGlycohemoglobin Standardization Pro-gram. At the start of the study in 1996,analyses were performed using the Var-iant Classic instrument (Bio-Rad Labora-tories, Inc., Hercules, CA) with an overallinterassay coefficient of variance (CV) of2.3%. In 1999, the Variant Classic wasreplaced by the Variant II instrumentfrom the same manufacturer with anoverall interassay CV of 1.7%. In 2004,the Variant II was replaced by the TosohG7 analyzer (Tosoh Bioscience, Inc.,San Francisco, CA), with an overall inter-assay CV of 0.9%. Each transition wasmonitored by parallel measurementsof patient and quality control samplesto ensure no difference in the measure-ments among instruments. Additionally,to monitor for possible assay drift dur-ing the course of the study, five bloodpools having HbA1c levels of 5%, 6%, 7%,8%, and 9% (31, 42, 53, 64, and 75mmol/mol, respectively) were preparedin the laboratory. The pools were ali-quotted and stored under liquid nitro-gen and analyzed for several days everymonth, and the mean values were plot-ted against their assigned values. Therewas consistently low variation around

the target values with no evidence ofassay drift over time. Some HbA1c assaysare sensitive to hemoglobinopathies, ofwhich HbS and HbF are especially com-mon in African Americans, but the re-sults are not affected by HbS. When asuspected S variant was detected, itwas confirmed by an independentmethod. If HbF was above the instru-ment threshold, HbA1c results werenot reported.

Analysis of HbA1c as a Predictor andas a Study OutcomeThe current report includes the partici-pants who did not have diabetes atbaseline according to FPG, 2hPG, andHbA1c (2010 ADA criteria), that is, FPG,7.0 mmol/L, 2hPG ,11.1 mmol/L,and HbA1c ,6.5% (48 mmol/mol). Ofthe 3,234 individuals randomized, weexcluded 54 with FPG $7.0 mmol/L(enrolled before the change in ADA diag-nostic criteria), 7 with missing HbA1c,and 408 with HbA1c $6.5%, leaving ananalysis set of 2,765 participants. Theparticipants were grouped by baselineHbA1c ,5.5%, 5.5% to ,6.0%, or 6.0%to ,6.5% (,37, 37 to ,42, or 42 to,48 mmol/mol, respectively) to deter-mine the predictive value for diabetes de-velopment defined by glucose or HbA1c.Supplementary Fig. 1 shows the follow-up of these participants. HbA1c resultswere not confirmed with repeat tests;therefore, for these analyses, a singleHbA1c $6.5% was considered diagnostic.

Statistical MethodsThe intention-to-treat analysis com-pared each intervention group withthe placebo group on the modifiedproduct-limit life-table distribution us-ing the log-rank test statistic. Treatmentgroups and study time periods were alsocompared with incidence rates in cases/100 person-years. Person-years weresummed over all participants in a groupof time to follow-up before a diagnosis orto end of follow-up if diabetes did notdevelop during the period of interest. Di-abetes hazard rates were stratified byage, sex, and self-reported race/ethnicity,and the covariate effects were assessedby simultaneously evaluating indicatorterms for each major group comparedwith a predefined comparison groupwith the likelihood ratio test. Risk reduc-tion and interactions between treat-ment assignments and covariates were

52 HbA1c in the Diabetes Prevention Program Diabetes Care Volume 38, January 2015

assessed by proportional hazardsregression.We present results for the 2,765 par-

ticipants separately for the 3.0-yearmedian (interquartile range 2.5–3.7)follow-up in the DPP before the studyresults were announced and the proto-col was modified and for the totalfollow-up period (DPP and DPPOS) fromeach participant’s randomization until acommon closing date of 27 August 2008(median 9.9 years, interquartile range9.0–10.5). Statistical tests evaluatingboth periods must be interpreted whilerecognizing that they are not indepen-dent of each other (the first is containedin the second). Both periods are of inter-est: the DPP period because interventioneffects on diabetes incidence were maxi-mal in this period and it was the onlydouble-blindperiod (for placebo andmet-formin) and the total follow-up period toassess effects for as long as possible.

RESULTS

Baseline characteristics of the 2,765 par-ticipants are shown in Table 1. None ofthe diabetes risk factors, includingHbA1c and glucose measures, differedamong the treatment groups.HbA1c at baseline was a strong pre-

dictor of the development of glucose-defined diabetes during theDPP and totalfollow-up periods (Fig. 1A and B). Duringthe DPP and total follow-up periods, the

incidence of glucose-defined diabeteswas positively related to baseline HbA1c,and stratified by baseline HbA1c, the in-cidence was reduced by metformin ver-sus placebo (P , 0.001) and by lifestyleversus placebo (P , 0.001), and the re-duction by lifestyle was greater than thatbymetformin (P, 0.001). These relation-ships were continuous, with no evidenceof an HbA1c threshold. In neither periodwas there a significant interaction ofbaseline HbA1c with treatment group onincidence of diabetes; that is, treatmenteffect (as a percent rate reduction) wasindependent of baseline HbA1c. The abso-lute effect in reducing diabetes incidencewas greater in those with higher baselineHbA1c, however, among whom the inci-dence rates were higher regardless oftreatment assignment.

Incidence rates by treatment areshown for diabetes defined by HbA1c$6.5% in Fig. 1C and D. As withglucose-defined diabetes (Fig. 1A and B),baseline HbA1c strongly predictedHbA1c-defined diabetes, and treatmenteffects did not differ significantly bybaseline HbA1c. During the DPP and totalfollow-up periods, the incidence of dia-betes defined by HbA1c $6.5% was pos-itively related to baseline HbA1c, andstratified by baseline HbA1c, the inci-dence was reduced by metformin versusplacebo (P , 0.0001) and by lifestyleversus placebo (P , 0.0001), but the

reductions by metformin and lifestyledid not differ significantly from eachother. There was a significant interac-tion (P , 0.01) between baselineHbA1c and the lifestyle versus placeboeffect, the effect being greater at higherbaseline HbA1c. Indeed, this outcomewas so infrequent in those with baselineHbA1c ,5.5% that treatment effectscould not be estimated precisely in thisgroup.

Incidence rates by treatment arecompared for the outcomes of glucose-and HbA1c-defined diabetes (Fig. 2).Incidence rates of glucose-defined dia-betes were lower in this subset than forall participants as reported previously(2) because of exclusion of the highestrisk group with baseline HbA1c $6.5%,but the treatment effects persisted,with a reduction by 29%with metforminand by 51% with ILS during the DPP andby 21% with metformin and 28% withILS in the total follow-up period. Duringthe DPP period, the incidence of diabe-tes by glucose criteria was reduced bymetformin versus placebo (P = 0.0013)and by lifestyle versus placebo (P ,0.0001), and lifestyle intervention re-duced it more than metformin (P =0.0023). During the total follow-up pe-riod, the incidence of diabetes by glucosecriteria was reduced bymetformin versusplacebo (P = 0.0014) and by lifestyle ver-sus placebo (P , 0.0001), but the re-ductions by metformin or lifestyleintervention did not differ significantlyfrom each other. By contrast, for inci-dence rates of HbA1c-defined diabetes,metformin and ILS resulted in nearly thesame rate reductions as each other dur-ing the DPP period (44% and 49%, respec-tively) and during total follow-up (38%and 29%, respectively). During the DPPand total follow-up periods, the incidenceof HbA1c $6.5% was reduced by metfor-min versus placebo (P , 0.0001) and bylifestyle versus placebo (P , 0.0001) butdid not differ significantly between themetformin and lifestyle interventions.

There were significant race/ethnicityeffects on the incidence of glucose-defined (during total follow-up) and HbA1c-defined diabetes (during the DPP andtotal follow-up periods). Across all treat-ment groups combined, incidence rateswere highest among African Americans(Supplementary Table 1) during thetotal follow-up period when defined byglucose (P = 0.005) and when defined by

Table 1—Baseline characteristics at DPP randomization

Overall Placebo Metformin Lifestyle

Men 873 (31.6) 289 (31.0) 305 (33.5) 279 (30.3)

Women 1,892 (68.4) 643 (69.0) 606 (66.5) 643 (69.7)

Race/ethnicityCaucasian 1,621 (58.6) 534 (57.3) 548 (60.2) 539 (58.5)African AmericanHispanic 422 (15.3) 135 (14.5) 147 (16.1) 140 (15.2)American Indian 150 (5.4) 53 (5.7) 47 (5.2) 50 (5.4)Asian American 125 (4.5) 53 (5.7) 30 (3.3) 42 (4.6)

HbA1c,5.5% 532 (19.2) 186 (20.0) 164 (18.0) 182 (19.7)5.5–5.9% 1,200 (43.4) 385 (41.3) 421 (46.2) 394 (42.7)6.0–6.4% 1,033 (37.4) 361 (38.7) 326 (35.8) 346 (37.5)

Age (years) 50.3 (10.6) 50.2 (11.3) 50.4 (10.2) 50.1 (10.4)

Weight (kg) 93.3 (19.8) 93.1 (20.3) 93.4 (19.4) 93.3 (19.6)

BMI (kg/m2) 33.7 (6.5) 33.6 (6.6) 33.6 (6.4) 33.8 (6.5)

Waist circumference (cm) 104.3 (14.1) 104.5 (14.5) 104.1 (13.9) 104.4 (13.8)

FPG (mg/dL) 105.4 (7.4) 105.3 (7.3) 105.2 (7.4) 105.6 (7.4)

2hPG (mg/dL) 163.9 (16.9) 163.5 (16.7) 164.3 (17.0) 163.8 (16.9)

HbA1c (%) 5.8 (0.4) 5.8 (0.4) 5.8 (0.4) 5.8 (0.4)

Data are n (%) or mean (SD). None of the variables differed significantly among treatmentgroups.

care.diabetesjournals.org Diabetes Prevention Program Research Group 53

HbA1c during the DPP and total follow-up periods (all P , 0.001) in modelsadjusted for sex, age, HbA1c at baseline,FPG and 2hPG at baseline, and treatmentassignment.Results are presented stratified by

baseline age and sex in Fig. 3. In eachage stratum among men and for eachoutcome (diabetes defined by glucoseor HbA1c), rates were lower in the met-formin and ILS groups than in the pla-cebo group. Among women $60 yearsold, the incidence of glucose-defined di-abetes was higher in the metforminthan in the placebo group, but the in-cidence of HbA1c-defined diabetes waslower in the metformin group. Thethree-way interaction of sex 3 age 3treatment was not statistically signifi-cant, however, indicating that these dis-parities could be due to chance. Therewere no significant sex 3 treatment in-teractions for incidence of glucose-defined diabetes during DPP or the totalfollow-up period; however, there were

several significant sex 3 treatment in-teractions in incidence of HbA1c $6.5%.During the DPP, risk reduction by ILSversus placebo was greater in menthan in women (70% and 38%, respec-tively; P = 0.010 for sex3 treatment in-teraction). Reduction in risk of reachingHbA1c$6.5% during total follow-up wasalso greater in men (52% for ILS vs. pla-cebo and 54% for metformin vs. placebocompared with 15% and 29% risk reduc-tions, respectively, in women; both P ,0.05 for sex 3 treatment interaction).

There were significant (P, 0.05) inter-actions between treatment and baselineage in the incidence of glucose-defineddiabetes during the DPP and totalfollow-up periods, with ILS having agreater advantage over metformin atolder ages. Although there was a ten-dency for such an age interaction onincidence of HbA1c $6.5%, the interac-tion was not as pronounced and not sta-tistically significant in the DPP or totalfollow-up periods.

The coincidence of diabetes definedby glucose and HbA1c criteria was exam-ined. At the 1,059 examinations that ledto a confirmed diabetes diagnosis basedon glucose criteria during the totalfollow-up period, HbA1c was ,6.5% in779 (74%) participants, was $6.5% forthe first time in 105 (10%), and had been$6.5% at a previous examination in 175(17%). Conversely, at the first examina-tion after baseline at which HbA1c was$6.5% (750 occurrences), there was noconfirmed glucose-based diagnosis of di-abetes in 341 (45%), a confirmed glucose-based diagnosis triggered at the samevisit in 105 (14%), and a previously con-firmed glucose-based diabetes diagnosisin 304 (41%).

CONCLUSIONS

HbA1c is recommended for identifyingpersons at high risk of developing diabe-tes and as a diabetes diagnostic criterion(5,6). In this report, we evaluated base-line HbA1c as a predictor of diabetes and

Figure 1—Incidence of diabetes (new cases/100 person-years) by baseline HbA1c, where diabetes was determined by 1997 ADA criteria using FPGand 2hPG concentrations or by HbA1c$6.5% (48 mmol/mol). Results are shown for the original masked treatment phase (DPP with mean follow-upof 3.0 years) (A and C) and for the DPP plus long-term follow-up (total follow-up withmedian follow-up of 9.9 years) (B and D). Met, metformin; Plac,placebo.

54 HbA1c in the Diabetes Prevention Program Diabetes Care Volume 38, January 2015

analyzed the DPP and DPPOS as if HbA1c$6.5% had been the sole outcome.Baseline HbA1c predicted developmentof glucose- and HbA1c-defined diabetesduring the DPP and total follow-up peri-ods, confirming findings in other pre-vention trials (7–9) that HbA1c belowthe diagnostic level of 6.5% is directlyassociated with risk of developing dia-betes. The risk relationship is continu-ous with baseline HbA1c as previouslysuggested (5), confirming that selectionof high-risk cut points of 6.0% (5) or5.7% (6) is arbitrary. The optimal selec-tion of high-risk characteristics for indi-viduals offered diabetes preventioninterventions will depend on available

resources, health benefits of preventivemeasures, and their comparative effec-tiveness and costs.

In the DPP and DPPOS, diabetes wasprevented or delayed withmetformin orILS aimed at weight loss and increasedphysical activity. The ILS was substan-tially more effective than metformin inpreventing glucose-defined diabetes(2,3). By contrast, if HbA1c $6.5% hadbeen the outcome, we would have con-cluded that both interventions weresimilarly effective. We would have alsoaccrued fewer events, confirming thatHbA1c $6.5% alone defines fewer per-sons as having diabetes than does thecombination of FPG or 2hPG, as found

in the National Health and Nutrition Ex-amination Survey (10) and the FinnishDiabetes Prevention Study (7). TheHbA1c diagnostic cut point was purpose-fully chosen to favor specificity over sen-sitivity, recognizing that it would usuallylead to fewer diagnoses compared withthe 1997 glucose-based criteria (5).

These results add to previous DPP re-ports that metformin and ILS were sim-ilar in affecting FPG, whereas ILS wasmore effective for 2hPG-defined diabe-tes (2,3). Metformin lowered FPG (11)consistent with its suppression of en-dogenous glucose production by theliver (12).

Do these results indicate that the twointerventions will have equivalenthealth benefits in DPP participants;that is, is preventing diabetes definedby HbA1c as clinically important as pre-venting diabetes defined by FPG or2hPG? HbA1c and FPG represent usualglycemia better than does 2hPG, whichmeasures response to a nonphysiologicchallenge. All three measures, however,have similar associations with microvas-cular disease (13–16). Therefore, therelative importance for other healthoutcomes of these glycemic measuresis not clear. The current analyses donot address the relative importance ofreducing diabetes based on glucose orHbA1c levels. Better understanding ofthe relative long-term health effectsof the two interventions should comefrom further follow-up, during whichintervention effects on microvascular-neuropathic outcomes and cardiovasculardisease risk factors will be assessed (3).The analyses performed to date of thetreatment effects on other outcomesduring DPP, including lipids, blood pres-sure, and hemostatic factors, suggestthat ILS achieves better or similar resultswith less medication use (3,17,18).Weight loss per se, which was greaterwith ILS, might have benefits beyond di-abetes prevention.Neither themetforminnor the ILS had a significant interactionwith a multigene diabetes risk score inpredicting glucose-defined diabetes inthe DPP (19), providing little support forconsidering genetic risk in choosing one ofthese treatment approaches. Evidencesuggests that genotype at candidate lociinfluences success in weight loss interven-tion (20,21). Such developments andmore comprehensive research on the ge-netics of response to metformin and

Figure 2—Comparison of treatment effects on the incidence of diabetes diagnosed by glucosecriteria or by HbA1c$6.5% (48mmol/mol). Results are shown for the original masked treatmentphase (DPP) (A) and for the total follow-up period (B). Met, metformin; Plac, placebo.

care.diabetesjournals.org Diabetes Prevention Program Research Group 55

lifestyle interventions may ultimately leadto tools for selecting optimal interven-tions for diabetes prevention.Not only did treatment effects on

glucose-definedandHbA1c-defineddiabetesdiffer, but the participants diagnosed byeach criterion did not fully overlap. Only26% of those diagnosed by FPG or 2hPGhad a previous or simultaneous HbA1c$6.5%. Conversely, 55% of those first at-taining an HbA1c $6.5% had a current orprevious diagnosis of diabetes by glucosecriteria. We cannot determine whetherthose diagnosed by FPG or 2hPG with

HbA1c ,6.5% would have subsequentlymet this level had they not been diag-nosed by glucose levels and diabetesmanagement subsequently initiated.

There were significant treatment in-teractions with age and sex. We previ-ously reported that the effects of theactive DPP interventions differed byage, with ILS being exceptionally effectiveand metformin ineffective among partic-ipants$60 years old at baseline (22). Fur-thermore, among those developingdiabetes during DPP, the older groupwas more likely than the young or

middle-aged groups to be diagnosed by2hPG, on which metformin may haveless effect. In the current analysis ofHbA1c-defined diabetes, the greater ef-fect of ILS with older age was maintained,and metformin was effective in all age-groups, particularly in men. The apparentadverse effect of metformin on incidenceof glucose-defined diabetes in olderwomen was not observed for HbA1c-defined diabetes.

Racial differences in the relationshipsbetween HbA1c and FPG and 2hPG havebeen reported (23,24). Other studies,however, suggested that interracial dif-ferences in HbA1c parallel differences inother measures of chronic glycemia (25)and that the relationships of HbA1c withretinopathy (26) and macrovascular dis-ease and death (27) are the same inAmerican blacks and whites. Moreover,differences in HbA1c between races arenot explained by ancestry-informativegenetic markers (28) or by allele fre-quency differences in genes associatedwith HbA1c (29). In the current analysesover the total follow-up period, the in-cidence of diabetes in the African Amer-icans was higher than in the other race/ethnicity groups whether diabetes wasdefined by glucose or HbA1c. In addition,treatment effects on HbA1c-defined di-abetes were similar among the race/ethnicity groups. Despite potentialrace/ethnicity differences in HbA1c,these data suggest treatment efficacyin all the race/ethnicity groups.

There are three important limitationsto these analyses. First, we cannotstrictly compare the performance ofthe different tests for diabetes becauseby protocol, diagnoses made by FPG or2hPG required confirmation and theHbA1c tests did not. Second, we cannotdetermine to what extent elevation ofHbA1c to $6.5% was prevented or de-layed by the diagnosis of diabetes byglucose criteria and subsequent man-agement or behavioral changes. Third,the eligibility criteria, including adultswith BMI $24 kg/m2, with elevatedFPG and 2hPG but without limitationson HbA1c, limit generalizability. Amongpersons not selected by BMI and glucosecriteria as in the DPP, it is not known towhat extent HbA1c would predict diabe-tes or response to interventions. For ex-ample, persons with HbA1c in the rangeof 6.0–6.4% who do not meet the othercriteria may be at lower risk of diabetes

Figure 3—Incidence of diabetes diagnosed by glucose criteria or by HbA1c$6.5% (48mmol/mol)by baseline age and treatment assignment in men (A–D) and women (E–H). Results are shownfor the original masked treatment phase (DPP) and for the total follow-up period. Met, metfor-min; Plac, placebo.

56 HbA1c in the Diabetes Prevention Program Diabetes Care Volume 38, January 2015

than indicated herein. Given the unifor-mity of treatment effects according tobaseline FPG and 2hPG reported previ-ously (2) and HbA1c in the current study,we suspect that the study interventionswould reduce diabetes risk similarlyregardless of these baseline factorsbut that the absolute risk reduction(or numbers of cases prevented pernumber treated) would be lower inpersons with lower levels of risk fac-tors. Because eligibility criteria in pre-vention trials have been so restricted,we do not know how best to select per-sons who should be offered preventiveinterventions (30).In summary, HbA1c measured at DPP

entry predicted incidence of diabetes,and study treatment effects were uni-form with respect to baseline HbA1c

(i.e., there were no significant baselineHbA1c by treatment interactions). Bycontrast, although ILS was superior tometformin for preventing the develop-ment of glucose-defined diabetes, theeffects of the two study treatmentswere similar in preventing diabetes de-fined by HbA1c. The health implicationsof these treatment and diagnostic dif-ferences await further assessment oflong-term health outcomes.

AppendixThis article was prepared by William C. Knowler(National Institute of Diabetes and Digestive andKidneyDiseases, Phoenix, AZ), Sharon L. Edelstein(Biostatistics Center, George Washington Univer-sity, Rockville, MD), Ronald B. Goldberg (Univer-sity of Miami, Miami, FL), Ronald T. Ackermann(Northwestern University, Chicago, IL), Jill P.Crandall (Albert Einstein College of Medicine,New York, NY), Jose C. Florez (MassachusettsGeneral Hospital and Harvard Medical School,Boston, MA), Sarah E. Fowler (Biostatistics Cen-ter, George Washington University, Rockville,MD), William H. Herman (University of Michigan,Ann Arbor,MI), Edward S. Horton (Joslin DiabetesCenter and HarvardMedical School, Boston,MA),Steven E. Kahn (VA Puget Sound Health CareSystem and University of Washington, Seattle,WA), Kieren J. Mather (Indiana University, Indian-apolis, IN), and David M. Nathan (MassachusettsGeneral Hospital and Harvard Medical School,Boston, MA).

Acknowledgments. The Research Groupgratefully acknowledges the commitment anddedication of the participants of the DPP andDPPOS.Funding. During the DPPOS, the National In-stitute of Diabetes and Digestive and KidneyDiseases (NIDDK) of the National Institutes ofHealth provided funding to the clinical centersand the coordinating center for the design and

conduct of the study and collection, manage-ment, analysis, and interpretation of the data.The Southwestern American Indian Centerswere supported directly by the NIDDK, includ-ing its Intramural Research Program, and theIndian Health Service. The General ClinicalResearch Center Program, National Center forResearch Resources, supported data collectionat many of the clinical centers. Funding wasalso provided by the Eunice Kennedy ShriverNational Institute of Child Health and HumanDevelopment; the National Institute on Aging;the National Eye Institute; the National Heart,Lung, and Blood Institute; the Office of Re-search on Women’s Health; the National Insti-tute on Minority Health and Health Disparities;the Centers for Disease Control and Preven-tion; and the ADA.

The primary sponsor, the NIDDK, was repre-sented on the steering committee and played apart in study design, management, and publi-cation. The sponsors were not members of thewriting group, although all members of thesteering committee had input into the article’scontents and reviewed the manuscript.Duality of Interest. Bristol-Myers Squibb andParke-Davis provided additional funding andmaterial support during the DPP, Lipha (MerckSante) providedmedication, and LifeScan Inc. do-nated materials during the DPP and DPPOS. Noother potential conflicts of interest relevant tothis article were reported.

The opinions expressed are those of the inves-tigators and do not necessarily reflect the viewsof the funding agencies. A complete list ofcenters, investigators, and staff can be foundin the Supplementary Data online.Author Contributions. The Writing Group,W.C.K., S.L.E., R.B.G., R.T.A., J.P.C., J.C.F., S.E.F.,W.H.H., E.S.H., S.E.K., K.J.M., and D.M.N., hadaccess to all data. W.C.K. is the guarantor of thiswork and, as such, had full access to all the datain the study and takes responsibility for theintegrity of the data and the accuracy of thedata analysis.Prior Presentation. Parts of this study werepresented in abstract form at the 70th ScientificSessions of the American Diabetes Association,Orlando, FL, 25–29 June 2010.

References1. The Diabetes Prevention Program ResearchGroup. The Diabetes Prevention Program. De-sign and methods for a clinical trial in the pre-vention of type 2 diabetes. Diabetes Care 1999;22:623–6342. Knowler WC, Barrett-Connor E, Fowler SE,et al.; Diabetes Prevention Program ResearchGroup. Reduction in the incidence of type 2 di-abetes with lifestyle intervention or metformin.N Engl J Med 2002;346:393–4033. Knowler WC, Fowler SE, Hamman RF, et al.;Diabetes Prevention Program Research Group.10-year follow-up of diabetes incidence andweight loss in the Diabetes Prevention Pro-gram Outcomes Study. Lancet 2009;374:1677–16864. American Diabetes Association. Report ofthe expert committee on the diagnosis and clas-sification of diabetes mellitus. Diabetes Care1997;20:1183–1197

5. International Expert Committee. Interna-tional Expert Committee report on the role ofthe A1C assay in the diagnosis of diabetes. Di-abetes Care 2009;32:1327–13346. American Diabetes Association. Diagnosisand classification of diabetes mellitus. DiabetesCare 2010;33(Suppl. 1):S62–S697. Pajunen P, Peltonen M, Eriksson JG, et al.;Finnish Diabetes Prevention Study. HbA(1c) indiagnosing and predicting type 2 diabetes inimpaired glucose tolerance: the Finnish Dia-betes Prevention Study. Diabet Med 2011;28:36–428. SaitoT,WatanabeM,Nishida J, et al.; ZensharenStudy for Prevention of Lifestyle DiseasesGroup. Lifestyle modification and preventionof type 2 diabetes in overweight Japanesewith impaired fasting glucose levels: a random-ized controlled trial. Arch InternMed 2011;171:1352–13609. Ramachandran A, Snehalatha C, SamithShetty A, Nanditha A. Predictive value ofHbA1c for incident diabetes among subjectswith impaired glucose tolerancedanalysis ofthe Indian Diabetes Prevention Programmes.Diabet Med 2012;29:94–9810. Cowie CC, Rust KF, Byrd-Holt DD, et al. Prev-alence of diabetes and high risk for diabetesusing A1C criteria in the U.S. population in1988-2006. Diabetes Care 2010;33:562–56811. Kitabchi AE, Temprosa M, Knowler WC,et al.; Diabetes Prevention Program ResearchGroup. Role of insulin secretion and sensitivityin the evolution of type 2 diabetes in the diabe-tes prevention program: effects of lifestyle in-tervention and metformin. Diabetes 2005;54:2404–241412. Hundal RS, Krssak M, Dufour S, et al. Mech-anism by which metformin reduces glucose pro-duction in type 2 diabetes. Diabetes 2000;49:2063–206913. McCance DR, Hanson RL, Charles M-A, et al.Comparison of tests for glycated haemoglobinand fasting and two hour plasma glucose con-centrations as diagnostic methods for diabetes.BMJ 1994;308:1323–132814. Tapp RJ, Tikellis G, Wong TY, Harper CA,Zimmet PZ, Shaw JE; Australian Diabetes Obesityand Lifestyle Study Group. Longitudinal associa-tion of glucose metabolism with retinopathy:results from the Australian Diabetes Obesityand Lifestyle (AusDiab) study. Diabetes Care2008;31:1349–135415. Colagiuri S, Lee CMY, Wong TY, Balkau B,Shaw JE, Borch-Johnsen K; DETECT-2 Collabora-tion Writing Group. Glycemic thresholds fordiabetes-specific retinopathy: implications fordiagnostic criteria for diabetes. Diabetes Care2011;34:145–15016. Selvin E, Ning Y, Steffes MW, et al. Glycatedhemoglobin and the risk of kidney disease andretinopathy in adults with andwithout diabetes.Diabetes 2011;60:298–30517. Haffner S, Temprosa M, Crandall J, et al.;Diabetes Prevention Program Research Group.Intensive lifestyle intervention or metformin oninflammation and coagulation in participantswith impaired glucose tolerance. Diabetes2005;54:1566–157218. Orchard TJ, Temprosa M, Barrett-Connor E,et al.; Diabetes Prevention Program OutcomesStudy Research Group. Long-term effects of the

care.diabetesjournals.org Diabetes Prevention Program Research Group 57

Diabetes Prevention Program interventions oncardiovascular risk factors: a report from theDPPOutcomes Study. DiabetMed 2013;30:46–5519. Hivert MF, Jablonski KA, Perreault L, et al.;DIAGRAM Consortium; Diabetes Prevention Pro-gram Research Group. Updated genetic scorebased on 34 confirmed type 2 diabetes loci isassociated with diabetes incidence and regres-sion to normoglycemia in the Diabetes Preven-tion Program. Diabetes 2011;60:1340–134820. Franks PW, Jablonski KA, Delahanty LM,et al.; Diabetes Prevention Program ResearchGroup. Assessing gene-treatment interactionsat the FTO and INSIG2 loci on obesity-relatedtraits in the Diabetes Prevention Program. Dia-betologia 2008;51:2214–222321. Pan Q , Delahanty LM, Jablonski KA, et al.;Diabetes Prevention Program Research Group.Variation at the melanocortin 4 receptor geneand response toweight-loss interventions in theDiabetes Prevention Program. Obesity (SilverSpring) 2013;21:E520–E526

22. Crandall J, Schade D, Ma Y, et al.; DiabetesPrevention Program Research Group. The in-fluence of age on the effects of lifestyle mod-ification and metformin in prevention ofdiabetes. J Gerontol A Biol Sci Med Sci 2006;61:1075–108123. Herman WH, Ma Y, Uwaifo G, et al.; Diabe-tes Prevention Program Research Group. Differ-ences in A1C by race and ethnicity amongpatients with impaired glucose tolerance inthe Diabetes Prevention Program. DiabetesCare 2007;30:2453–245724. Ziemer DC, Kolm P, Weintraub WS, et al.Glucose-independent, black-white differences inhemoglobin A1c levels: a cross-sectional analysisof 2 studies. Ann Intern Med 2010;152:770–77725. Selvin E, Steffes MW, Ballantyne CM,Hoogeveen RC, Coresh J, Brancati FL. Racial dif-ferences in glycemic markers: a cross-sectional-analysis of community-based data. Ann InternMed 2011;154:303–30926. Tsugawa Y, Mukamal KJ, Davis RB, TaylorWC, Wee CC. Should the hemoglobin A1c

diagnostic cutoff differ between blacks andwhites? A cross-sectional study. Ann Intern Med2012;157:153–15927. Selvin E, Steffes MW, Zhu H, et al. Glycatedhemoglobin, diabetes, and cardiovascular risk innondiabetic adults. N Engl J Med 2010;362:800–81128. Maruthur NM, Kao WHL, Clark JM, et al.Does genetic ancestry explain higher values ofglycated hemoglobin in African Americans? Di-abetes 2011;60:2434–243829. Grimsby JL, Porneala BC, Vassy JL, et al.;MAGIC Investigators. Race-ethnic differencesin the association of genetic loci with HbA1clevels and mortality in U.S. adults: the thirdNational Health and Nutrition Examina-tion Survey (NHANES III). BMC Med Genet2012;13:3030. Knowler WC. Prevention of type 2 diabetes:comment on “Lifestyle Modification and Pre-vention of Type 2 Diabetes in Overweight Japa-nese With Impaired Fasting Glucose Levels.”Arch Intern Med 2011;171:1361–1362

58 HbA1c in the Diabetes Prevention Program Diabetes Care Volume 38, January 2015