Original Article | Published:

Carbohydrates, insulin resistance and diabetes mellitus

Chocolate intake and diabetes risk in postmenopausal American women

European Journal of Clinical Nutrition volume 71, pages 10881093 (2017) | Download Citation



Recent long-term prospective cohort studies found inverse associations between chocolate consumption and the risk of type 2 diabetes, but provided conflicting evidence on the nature of the association among women. To assess this association in a large cohort of American women.


Multivariable Cox regression was used with the data from 92 678 postmenopausal women in the prospective Women’s Health Initiative study. Chocolate intake was assessed by food frequency questionnaire. Incidence of type 2 diabetes was determined by self-report of the first treatment with oral medication or insulin.


Among women free of diabetes at baseline, there were 10 804 cases, representing an incidence rate of 11.7% during 13.1 years and 1 164 498 person-years of follow-up. There was no significant linear association between long-term chocolate intake and type 2 diabetes risk, but there was significantly reduced risk at moderate levels of intake. Compared to women who ate 1 oz. of chocolate <1 time per month, those who ate this amount 1–<1.5 times per month, 1.5–<3.5 times per month, 3.5 times per month to <3 times per week and 3 times per week had hazard ratios of 0.97 (95% confidence interval: 0.92, 1.04), 0.92 (0.87, 0.98), 0.93 (0.88, 0.98) and 0.98 (0.92, 1.04) (P for linear trend=0.79). There was only evidence of such inverse associations for women with below-median physical activity (P for interaction <0.0001) and those with age<65 years (P=0.01).


We only found an inverse association between chocolate consumption and type 2 diabetes at moderate levels of consumption in two subgroups of postmenopausal women in the Women’s Health initiative cohort.


Type 2 diabetes mellitus (diabetes) is a widespread burgeoning chronic disease1 with severe clinical sequelae,2 and chocolate is highly popular. There has, therefore, been keen interest in the finding of significant inverse associations between chocolate consumption and diabetes risk in three recent long-term prospective epidemiological cohort studies. All three studies conducted Cox regression survival analyses. Matsumoto et al.3 found an inverse association in 18 235 males in the Physicians' Health Study with a 9.2-year follow-up period. Oba et al.4 used the data from 13 540 free-living Japanese persons with a 10-year follow-up period, and found a significant inverse association for males but not for females. Greenberg5 analyzed the data from 7 802 participants in the Atherosclerosis Risk in Communities cohort with a 13-year follow-up period, and found a significant inverse association with no evidence of effect modification by gender.

Our objective was to explore whether the association between chocolate intake and diabetes risk is inverse in postmenopausal women in the United States by using the data from the Women’s Health Initiative (WHI) Observational Study and Clinical Trials.

Materials and methods

Analytic sample

The design and methods of the WHI have been described in detail elsewhere.6 In brief, 161 808 postmenopausal women of 50–79 years of age were enrolled between 1993 and 1998 into the Observational Study (OS) or four overlapping clinical trials (CTs).

Of the original 161 808, 93 676 OS and 68 132 CT participants, we included all OS and CT-control participants (Figure 1). There were 120 191 of these participants at baseline from whom we excluded those with implausible FFQ energy intakes, defined as mean intakes <600 or >5000 kcal per day. The exclusion rate was 3.4%, leaving 116 088 women at year 0, from whom we excluded 4902 women (4.2%) who reported pre-existing diabetes and 126 women with missing data on pre-existing diabetes. Of the 11 1060 remaining women, we excluded a further 18 382 women, or 16.6%, for having missing values on any exposure, outcome or confounder variable. This left 92 678 women who provided the data for our analysis. The missing rate was lower than 1.5% for most of the exposure, outcome and confounder variables, and below 5% for all of them, except a family history of diabetes, for which the missing rate was 5.3%.

Figure 1
Figure 1

Participant flow chart.

Chocolate intake

We derived our exposure variable, chocolate consumption, from the WHI food frequency questionnaire (FFQ). The question was how often over the past 3 months the participant ate a medium portion of chocolate candy or candy bars. A medium portion was defined as 1 oz. or a small bar. We created five levels of intake frequency of a 1 oz serving of chocolate as follows: <1 time per month, 1 time per month to <1.5 times per month, 1.5 times per month to <3.5 times per month, 3.5 times per month to <3 times per week and 3 times per week. We selected cut points from the 23 different reported values of the frequency of consumption of 1 oz. of chocolate that yielded strata, which were as close as possible to quintiles, given that the lowest level (<1 serving per month) contained more than 20% of cases.


We used the following two models in our analyses: (1) a model with age, race/ethnicity and WHI studyarm (a categorical variable, indicating whether the participant was in one or more clinical trials or the observational study); and (2) a full model with the following covariates: age in years; race/ethnicity—white, black, other; WHI studyarm—OS, CT-control; family history of diabetes—yes or no; smoking status—never, past, <15/day, >=15/day; recreational energy expended, in MET-hours per week; coffee consumption—cups per day; non-chocolate energy intake—kcal per day;7 Alternative Modified Healthy Eating Index;8 education—<high school, some high school–<college, some college–<postgraduate study, postgraduate study or degree; family income—8 levels; physical functional ability, based on the Rand Physical Functioning Construct;9 and emotional well-being.9 Demographic and health variables were assessed with standardized protocols, and the Alternative Modified Health Eating Index was calculated using the FFQ data.

In building our full model, we identified potential covariates based on our knowledge of prior findings regarding risk factors for diabetes. We included age because of its established importance as a risk factor, and WHI studyarm because of potential effects due to differences in clinical trial inclusion criteria and participation requirements. We used forward inclusion and backward elimination to add further covariates. Our inclusion criterion was a >10% change in the regression coefficient for chocolate intake. We used the same criterion to test polynomial and categorical versions of covariates. Variables that did not meet our inclusion criteria were as follows: marital status; employment status; depression;10 sleep disturbances;11 social functioning;11 illness symptoms11 and activities of daily living.11

Outcome variable

Our outcome variable was the first incidence of diabetes treated with oral diabetes medication or insulin injections, self-reported on semi-annual or annual survey questionnaires. This measure has been found to be a reliable indicator of diagnosed diabetes in the WHI data.12, 13

The diabetes diagnosis date was designated as having occurred halfway between the survey when the diagnosis was first reported and the previous survey. Participants were followed from year 0 (baseline) to diagnosis of diabetes, death, loss to follow-up, or 30 September 2013, whichever occurred first. Participants lost to follow-up were censored.

Statistical methods

Cox regression14 was used to estimate the hazard ratios for diabetes in different levels of chocolate intake frequency, and for a 1 oz per day increment in chocolate intake. Interaction tests were conducted by inserting a cross-product of the predictor and interaction variables into our full model. All interaction variables were specified a priori, and subgroup analyses were conducted for each interaction variable that yielded a significant interaction effect. The proportional-hazards assumption was investigated by means of Schoenfeld residuals. The effects of potential nonproportionality on hazard ratio estimates were assessed by using time-dependent cox regression. This latter analysis showed that the potential nonproportionality caused essentially no changes in hazard ratio estimates or 95% confidence intervals.

Two-sided P<.05 was considered significant. Cox Regression analyses and the descriptive statistics in Table 1 were produced using SAS (v. 9.4, SAS Institute Inc, Cary, NC, USA). The manuscript was prepared using the Strobe guidelines for observational studies.15

Table 1: Baseline characteristics of postmenopausal women in different levels of chocolate consumption in the women’s health initiative


The WHI study protocol (available at www.whi.org) was approved by institutional review boards at each participating institution, and all participants provided written informed consent. Our procedures were in accordance with the ethical standards of the responsible institutional committee on human experimentation and in accordance with the Helsinki Declaration of 1975 as revised in 1983. WHI is registered at ClinicalTrials.gov NCT00000611.

Code availability

The SAS code used for cox regression analyses is available from the corresponding author.


Baseline characteristics

More frequent chocolate intake was associated with the following: (1) higher BMI and higher levels of caloric intake; (2) greater likelihood of being white and smoking >=15 cigarettes per day; (3) lower likelihood of having three or more close relatives with diabetes; and (4) lower levels of recreational physical activity, emotional wellness and healthy eating (Table 1).

Chocolate intake and incidence of diabetes

After exclusions, there were 92 678 women at baseline and 10 804 cases of diabetes during 1 164 498 person-years of follow-up, representing an incidence rate of 11.7%. The mean follow-up period was 13.1 years for censored participants in our survival analysis with all covariates.

Table 2 shows that after adjusting for age, race/ethnicity and WHI studyarm, there was a pattern of higher intake of chocolate being associated with higher hazard ratios (HRs) for diabetes risk. In our full model analysis, the HRs were closer to the null, and as chocolate intake frequency (1 oz servings) increased above 1.5 time per month, the hazard ratios (HRs) for diabetes decreased and were significantly lower than 1.00 for intake levels between 1.5 times per month and <3 times per week, but not for 3 times per week. The nadir was between 1.5 and <3.5 times per month. Results were similar when the sample was limited to women without pre-existing myocardial infarction, stroke, cancer and other major clinical diseases that may have caused changes in dietary habits prior to baseline.3

Table 2: Chocolate intake, hazard ratios and 95% confidence intervals for incident diabetes showing the effects of excluding serious chronic disease in the women’s health initiative

Secondary analyses

We specified a priori interaction tests for age, physical activity and BMI in our full model analysis (in Table 2), because these variables could affect dietary habits.3 After conducting the interaction tests, we stratified for age (P for interaction=0.01) and physical activity (p for interaction <.0001) because both variables exhibited significant interaction effects. We did not stratify for BMI as it did not exhibit a significant interaction effect (p for interaction=0.14). The stratifications (Table 3) produced HRs significantly lower than 1.00 for women with below-median physical activity, but not for women with above-median physical activity. Also, there were HRs significantly below 1.00 for women with age <65 years, but not those with age65 years.

Table 3: Chocolate intake, multivariable hazard ratios & 95% confidence intervals for incident diabetes stratified by age and physical activity in the women’s health initiative

We also repeated our full model analysis for all women in our analytic sample (Table 2) with the addition of BMI as a covariate because of the established role of BMI as an important risk factor for diabetes. The results were essentially unchanged. Compared to women who ate 1 oz. of chocolate <1 time per month, those who ate this amount 1-<1.5 times per month, 1.5–<3.5 times per month, 3.5 times per month–<3 times per week and 3 times per week had hazard ratios of 0.97 (0.91, 1.03), 0.89 (0.84, 0.95), 0.91 (0.86, 0.96) and 0.95 (0.89, 1.01).


Our main finding is that although there was no significant linear association between long-term chocolate intake and diabetes risk in the WHI cohort, there was evidence of a significantly reduced risk at moderate levels of intake. This finding was robust in that it was unchanged after we excluded women with pre-existing myocardial infarction, stroke, cancer and other major clinical diseases.

All three prior epidemiological analyses of the long-term association between chocolate intake and diabetes risk3, 4, 5 yielded inverse associations with a significant linear trend after excluding participants with pre-existing serious chronic disease. Similar exclusions in our analysis produced evidence of a significantly reduced risk at moderate levels of intake with a nonsignificant linear trend. We observed significant interaction effects for physical activity and age, with evidence of an inverse chocolate–diabetes association only among women with below-median physical activity and those with age< 65 years. Matsumoto et al.3 found similar effects for age, but no significant interaction for physical activity. Matsumoto et al.’s analysis did yield a significant interaction for BMI, with an inverse chocolate–diabetes association and a significant linear trend for BMI<25 kg/m2. Greenberg5 did not report any significant interaction or effect modification results. Our nonsignificant interaction for BMI and our secondary analysis finding that adding BMI as a covariate in our main full model analysis did not much alter any hazard ratios provides evidence that the association between chocolate and diabetes risk was not materially affected by BMI. Matsumoto et al. and Oba et al. did, but Greenberg did not, adjust for BMI in their main analyses.

Our study has several strengths. First, the WHI dataset is the result of careful data gathering, editing and validation using up-to-date, empirically-based protocols.6, 16, 17 Second, this large dataset provided adequate statistical power with which to perform a variety of subgroup analyses. Our analysis also has limitations. First, our chocolate intake variable was derived from the self-reported FFQ data.18 The FFQ data are prone to both random and systematic measurement error,19 which can distort hazard ratio estimates.20 Second, we did not have the data on the types of chocolate consumed by our participants. Different types of chocolate contain different levels of calories, sugar and flavanols;21 and as flavanols could be the cocoa compounds that decrease diabetes risk,22, 23 different types of chocolate could have yielded different results in our analysis. Also, we conducted three a priori interaction tests and three stratified analyses without adjusting for multiple comparisons, so that the results of these latter analyses should be regarded with caution.

In conclusion, we found that although there was no significant linear association between long-term chocolate intake and type 2 diabetes risk in the Women’s Health Initiative cohort, there was evidence of a significantly reduced risk at moderate levels of intake. The reduced risk only applied to the following two subgroups of women: those with below-median physical activity and those with age<65 years. These findings suggest that consuming chocolate is unlikely to reduce the risk of type 2 diabetes in postmenopausal women. Our results require confirmation.


  1. 1.

    , , , , , . Trends in lifetime risk and years of life lost due to diabetes in the USA, 1985-2011: a modelling study. Lancet Diabetes Endocrinol 2014; 2: 867–874.

  2. 2.

    . Addressing the burden of diabetes. JAMA 2014; 311: 2267–2268.

  3. 3.

    , , , , . Chocolate consumption and risk of diabetes mellitus in the Physicians’ Health Study. Am J Clin Nutr 2015; 101: 362–367.

  4. 4.

    , , , , , et al. Consumption of coffee, green tea, oolong tea, black tea, chocolate snacks and the caffeine content in relation to risk of diabetes in Japanese men and women. Br J Nutr 2010; 103: 453–459.

  5. 5.

    . Chocolate intake and diabetes risk. Clin Nutr 2015; 34: 129–33.

  6. 6.

    WHI Study Group. Design of the Women’s Health Initiative clinical trial and observational study. The Women’s Health Initiative Study Group. Control Clin Trials 1998; 19: 61–109.

  7. 7.

    , , , , , et al. Dietary fat and coronary heart disease: a comparison of approaches for adjusting for total energy intake and modeling repeated dietary measurements. Am J Epidemiol 1999; 149: 531–540.

  8. 8.

    , , , , , et al. Alternative dietary indices both strongly predict risk of chronic disease. J Nutr 2012; 142: 1009–1018.

  9. 9.

    , , . The RAND 36-Item Health Survey 1.0. Health Econ 1993; 2: 217–227.

  10. 10.

    , , , . Short version of the CES-D (Burnam screen) for depression in reference to the structured psychiatric interview. Psychiatry Res 2001; 103: 261–270.

  11. 11.

    (WHI) Women's Health Initiative 2013. Women's Health Initiative (WWW document).

  12. 12.

    , , , , , et al. Women Health Initiative Investigators. validity of diabetes self-reports in the Women’s Health Initiative: comparison with medication inventories and fasting glucose measurements. Clin Trials 2008; 5: 240–247.

  13. 13.

    , , , , , et al. Validity of diabetes self-reports in the Women’s Health Initiative. J Clin Epidemiol 2013; 66: 349–350.

  14. 14.

    . Regression models and life-tables. J Royal Stat Soc Series B-Statistical Methodology 1972; 34: 187–220.

  15. 15.

    , , , , , . The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: guidelines for reporting observational studies. Ann Intern Med 2007; 147: 573–577. PMID: 17938396.

  16. 16.

    , , , , , . The Women’s Health Initiative Observational Study: baseline characteristics of participants and reliability of baseline measures. Ann Epidemiol 2003; 13: S107–S121.

  17. 17.

    , , , , , et al. Women’s Health Initiative Investigators. Women’s Health Initiative Investigators. A diet high in low-fat dairy products lowers diabetes risk in postmenopausal women. J Nutr 2011; 141: 1969–1974.

  18. 18.

    , , , , , . Measurement characteristics of the Women's Health Initiative food frequency questionnaire. Ann Epidemiol 1999; 9: 178–187.

  19. 19.

    , , , . Adjustments for regression dilution in epidemiological regression analysis. Ann Epidemiol 1998; 8: 56–63.

  20. 20.

    . The proof and measurement of association between two things. Am J Psychol 1904; 15: 72–101.

  21. 21.

    , , . Cocoa and chocolate flavonoids: implications for cardiovascular health. J Am Diet Assoc 2003; 103: 215–223.

  22. 22.

    , , . Cocoa and chocolate in human health and disease. Antioxid Redox Signal 2011; 15: 2779–2811.

  23. 23.

    , , , , , et al. Cocoa Flavanol Intake and Biomarkers for Cardiometabolic Health: A Systematic Review and Meta-Analysis of Randomized Controlled Trials. J Nutr. 2016; 146: 2325–2333.

Download references


We thank the WHI Investigators (see Online Supplement) for their efforts in the collection of the WHI data.

Author contributions

JG, JM, LT, MN, LG and MV conceived the work that led to the submission. JM, LT, MN, LG and MV acquired the data. JG performed the statistical analysis. JG, JM, LT and MN played an important role in interpreting results. JG drafted the manuscript. JG, JM, LT, MN, LG, MV and LP contributed to the revision of the manuscript. JG had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors approved the final version.

Author information


  1. Department of Health and Nutrition Sciences, Brooklyn College of the City University of New York, Brooklyn, NY, USA

    • J A Greenberg
  2. Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA

    • J E Manson
  3. Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, WA, USA

    • L Tinker
    •  & M L Neuhouser
  4. Department of Public Health Sciences, Davis, CA, USA

    • L Garcia
  5. Department of Epidemiology and Prevention, Division of Public Health Sciences, Wake Forest School of Medicine, Medical Center Boulevard, Winston-Salem, NC, USA

    • M Z Vitolins
  6. Clinical Studies Center, Atlanta VA Medical Center, Decatur, GA, USA

    • L S Phillips
  7. Division of Endocrinology and Metabolism, Department of Medicine, Emory University School of Medicine, Atlanta, GA, USA

    • L S Phillips


  1. Search for J A Greenberg in:

  2. Search for J E Manson in:

  3. Search for L Tinker in:

  4. Search for M L Neuhouser in:

  5. Search for L Garcia in:

  6. Search for M Z Vitolins in:

  7. Search for L S Phillips in:

Competing interests

JM and colleagues at Brigham and Women’s Hospital, Harvard Medical School are recipients of funding from Mars Symbioscience for an investigator-initiated randomized trial of cocoa flavanols and cardiovascular disease. JG is the recipient of funding from the City University of New York Research Award Program to conduct a pilot randomized trial of cocoa compounds and appetite. LP has served on Scientific Advisory Boards for Boehringer Ingelheim and Janssen within the past several years, and has or had research support from Merck, Amylin, Eli Lilly, Novo Nordisk, Sanofi, PhaseBio, Roche, and the Cystic Fibrosis Foundation. In the past, he was a speaker for Novartis and Merck. He is also a co-founder of a company, Diasyst LLC, which aims to develop and commercialize diabetes management software programs. These activities involve diabetes, but have nothing to do with this manuscript. The remaining authors declare no conflict of interest.

Corresponding author

Correspondence to J A Greenberg.

Supplementary information

About this article

Publication history







Supplementary Information accompanies this paper on European Journal of Clinical Nutrition website (http://www.nature.com/ejcn)

Further reading