Chocolate intake and risk of type 2 diabetes: prospective cohort studies Authors: Binkai Liu, Geng Zong, Lu Zhu, Yang Hu, JoAnn E Manson, Molin Wang, Eric B Rimm, Frank B Hu, Qi Sun Published: 2024 ABSTRACT AbstractObjectiveTo prospectively investigate the associations between dark, milk, and total chocolate consumption and risk of type 2 diabetes (T2D) in three US cohorts.DesignProspective cohort studies.SettingNurses’ Health Study (NHS; 1986-2018), Nurses’ Health Study II (NHSII; 1991-2021), and Health Professionals Follow-Up Study (HPFS; 1986-2020).ParticipantsAt study baseline for total chocolate analyses (1986 for NHS and HPFS; 1991 for NHSII), 192 208 participants without T2D, cardiovascular disease, or cancer were included. 111 654 participants were included in the analysis for risk of T2D by intake of chocolate subtypes, assessed from 2006 in NHS and HPFS and from 2007 in NHSII.Main outcome measureSelf-reported incident T2D, with patients identified by follow-up questionnaires and confirmed through a validated supplementary questionnaire. Cox proportional hazards regression was used to estimate hazard ratios and 95% confidence intervals (CIs) for T2D according to chocolate consumption.ResultsIn the primary analyses for total chocolate, 18 862 people with incident T2D were identified during 4 829 175 person years of follow-up. After adjusting for personal, lifestyle, and dietary risk factors, participants consuming ≥5 servings/week of any chocolate showed a significant 10% (95% CI 2% to 17%; P trend=0.07) lower rate of T2D compared with those who never or rarely consumed chocolate. In analyses by chocolate subtypes, 4771 people with incident T2D were identified. Participants who consumed ≥5 servings/week of dark chocolate showed a significant 21% (5% to 34%; P trend=0.006) lower risk of T2D. No significant associations were found for milk chocolate intake. Spline regression showed a linear dose-response association between dark chocolate intake and risk of T2D (P for linearity=0.003), with a significant risk reduction of 3% (1% to 5%) observed for each serving/week of dark chocolate consumption. Intake of milk, but not dark, chocolate was positively associated with weight gain.ConclusionsIncreased consumption of dark, but not milk, chocolate was associated with lower risk of T2D. Increased consumption of milk, but not dark, chocolate was associated with long term weight gain. Further randomized controlled trials are needed to replicate these findings and further explore the mechanisms. Introduction The global prevalence of type 2 diabetes (T2D) has increased noticeably over the past few decades, with an estimated 463 million people affected worldwide in 2019 and projected to increase to 700 million by 2045.1 T2D is a multifactorial disease characterized by insulin resistance and impaired insulin secretion, which can lead to numerous severe complications such as cardiovascular disease, renal failure, and loss of vision.2 A growing body of research has highlighted the importance of lifestyle factors, including healthy diets, in the prevention and management of T2D.3 4 Higher consumption of total dietary flavonoids, as well as specific flavonoid subclasses, has been associated with a decreased risk of T2D.5 6 In randomized controlled trials, these flavonoids exerted antioxidant, anti-inflammatory, and vasodilatory effects that might confer cardiometabolic benefits and reduce the risk of T2D,7 8 9 although data were not consistent.10 Chocolate, derived from the beans of the cacao tree (Theobroma cacao), is among foods with the highest flavanol content and is a popular snack globally.11 12 13 However, the association between chocolate consumption and risk of T2D remains controversial owning to inconsistent findings in observational studies.14 15 Furthermore, most previous studies have primarily focused on total chocolate intake, without considering the potential differences in health effects between chocolate subtypes (ie, dark, milk, and white chocolate).16 17 These subtypes differ in cocoa content and proportions of other ingredients such as sugar and milk, which may influence the association with risk of T2D.18 Using data from three prospective cohort studies that repeatedly assess participants’ diet during longitudinal follow-up, we investigated the association between subtypes of chocolate intake and risk of T2D, as well as change in bodyweight, which is a strong predictor for risk of T2D.19 Methods Study population This study used data from three large prospective cohorts: the Nurses’ Health Study (NHS), initiated in 1976 and comprising 121 700 female registered nurses20; the Nurses’ Health Study II (NHSII), launched in 1989 and comprising 116 340 female nurses21; and the Health Professionals Follow-Up Study (HPFS), initiated in 1986 and comprising 51 529 male health professionals.22 For analyses on total chocolate consumption, the study baselines (baseline 1) were 1986 for NHS and HPFS and 1991 for NHSII, when diet was first comprehensively assessed using a validated semiquantitative food frequency questionnaire.23 24 25 For analyses on chocolate subtypes, the study baselines (baseline 2) were 2006 for NHS and HPFS and 2007 for NHSII, coinciding with the availability of data on chocolate subtypes in the food frequency questionnaire. In the primary analyses, we excluded participants who had prevalent diabetes, cardiovascular disease (myocardial infarction, coronary artery bypass surgery, stroke), or cancer (except non-melanoma skin cancer) at baseline; had missing information on baseline age and chocolate intake, or an unusual total energy intake (ie, <600 or >3500 kcal/d for women and <800 or >4200 kcal/d for men); only completed the baseline food frequency questionnaire; or had an undetermined diagnosis date for T2D. In a secondary analysis of change in bodyweight by levels of chocolate intake, we excluded participants with self-reported diabetes, cardiovascular disease, or cancer at baseline. Additional baseline exclusions were extreme total energy intake and missing data on chocolate intake or bodyweight. During follow-up, we censored individuals who reported cardiovascular disease, cancer, or a diagnosis of diabetes. We also censored individuals with missing data on chocolate intake or bodyweight during follow-up. The flowchart in supplementary figure S1 shows the selection process for participants. The study protocol was approved by the Human Research Committee of Brigham and Women’s Hospital and the Harvard T.H. Chan School of Public Health. Participants provided informed consent by completing and returning study questionnaires. Assessment of diet Diet was assessed every four years using a semiquantitative food frequency questionnaire.23 Participants reported their average frequency of consumption for a standard food portion size (eg, one chocolate bar/pack or 1 oz) in the past year, choosing from nine levels ranging from “never, or less than once per month” to “≥6 per day.” Questions about chocolate consumption were included from 1980 in NHS (study baseline was set at 1986 for total chocolate when the comprehensive food frequency questionnaire was first administered in NHS), 1991 in NHSII, and 1986 in HPFS. Questions about specific subtypes of chocolate consumption, specifically “How often do you consume milk chocolate (bar or pack),” and “How often do you consume dark chocolate,” were added to the food frequency questionnaire from 2006 for NHS and HPFS and from 2007 for NHSII, which were study baselines for the analyses by chocolate subtypes. Information on consumption of chocolate subtypes was collected in 2006 and 2010 for NHS; 2007, 2001, and 2015 for NHSII; and 2006, 2010, 2014, and 2018 for HPFS. Nutrient intakes were measured based on the US Department of Agriculture food composition database. We calculated the average daily intake of nutrients by multiplying the frequency of consumption of each food item that contains the nutrient by the nutrient content, and then summing across all food items. The intake level of flavan-3-ols was calculated by summing the individual components (catechin, epicatechin, gallocatechin, epigallocatechin, epicatechin 3-gallate, and epigallocatechin 3-gallate). The intake level of total flavonoids was further calculated by summing intakes of flavonols, flavones, flavanones, flavan-3-ols, total theaflavins, and polymers proanthocyanidins. Total energy intake was derived from the food frequency questionnaires using the same algorithm. The validity and reproducibility of the food frequency questionnaires have been reported in detail previously.23 24 25 In validation studies among subgroups of participants in NHS and HPFS, the correlation coefficient between chocolate intake assessed by diet records over seven days and food frequency questionnaire was, respectively, 0.53 and 0.65 for dark chocolate and 0.42 and 0.47 for milk chocolate.26 Additionally, the food frequency questionnaire shows reasonable consistency of food intake measurements over time, with mean intraclass correlation coefficients of 0.71 (dark: 0.71; milk: 0.59) in NHS and 0.72 (dark: 0.66; milk: 0.60) in HPFS.26 Assessment of covariates In all three cohorts, information on participants’ personal and lifestyle factors, as well as diseases or medical conditions, was collected biennially through questionnaires. Variables assessed included race/ethnicity, bodyweight, waist circumference, smoking status, alcohol consumption, multivitamin use, menopausal status and postmenopausal hormone use (women only), oral contraceptive use (NHSII only), hypertension, hypercholesterolemia, and family history of diabetes. Specialty of health professionals was assessed in HPFS, and academic or professional degrees earned were assessed in NHS. Physical activity was measured using a validated questionnaire,27 28 29 which derived metabolic equivalent tasks in hours per week based on time spent on 10 recreational activities. Height was measured in the baseline questionnaire of these three cohorts, and time varying body mass index (BMI) was calculated as biennially updated bodyweight in kilograms divided by the square of height in meters. The Alternate Healthy Eating Index-2010 (AHEI) was calculated based on participants’ responses in the food frequency questionnaire to reflect overall diet quality, which emphasizes foods and nutrients, such as whole grains, nuts, red and processed meats, sugar sweetened beverages, and alcohol, that predict the risk of chronic disease.30 For NHS and NHSII, we further calculated z scores of standardized neighborhood socioeconomic status using census tract variables from the Neighborhood Change Database. More details on computation are described elsewhere.31 Assessment of T2D T2D was self-reported in the biennial follow-up questionnaires, and the diagnosis was further confirmed by study doctors through a supplementary questionnaire, which collected data on factors such as glucose concentration, glycated hemoglobin (HbA1c) concentration (after 2010), symptoms and treatments for T2D. People with confirmed T2D, which constituted the outcome of this analysis, were defined according to the National Diabetes Data Group criteria before 1998,32 where at least one of the following items was reported: raised fasting plasma glucose level (≥7.8 mmol/L (140 mg/dL)) or plasma glucose (≥11.1 mmol/L (200 mg/dL)), and at least one T2D symptom (excessive thirst, polyuria, weight loss, hunger); no T2D symptom, but had at least two raised plasma glucose measurements on different occasions; or treatment with insulin or other hypoglycemic drug (insulin or oral hypoglycemic agent). For people with T2D diagnosed after 1998, the threshold for raised fasting plasma glucose level was lowered to 7.0 mmol/L (126 mg/dL) according to the criteria of the American Diabetes Association.33 For people with a diagnosis after January 2010, we further added HbA1c ≥6.5% to the diagnostic criteria.34 The validity of the supplementary questionnaire was established in two previous studies in NHS and HPFS through blinded medical record reviews, with T2D diagnosis confirmed in 98% and 97% of people, respectively.35 36 Ascertainment of weight change Bodyweight was self-reported at baseline and in the biennial follow-up questionnaires, and it has been previously validated in these cohorts, with correlation coefficients of 0.97 in NHS and HPFS and 0.84 in NHSII.37 38 We calculated the outcome of interest, weight change every four years, by subtracting bodyweight at the start of each four year interval from bodyweight at the end of the four year interval. Statistical analysis We present participants’ characteristics at baseline 1 (1986 for NHS and HPFS, 1991 for NHSII) and baseline 2 (2006 for NHS and HPFS, 2007 for NHSII), stratified by consumption levels of total, dark, and milk chocolate and further categorized by study cohort. This breakdown allowed for a more nuanced understanding of the characteristics of participants, as individuals consuming similar levels of dark or milk chocolate might have distinct baseline characteristics across cohorts. To reflect the long term usual intake of total chocolate at baseline 2, we calculated cumulatively averaged total chocolate intake at or before baseline 2 (1980-2006 for NHS, 1991-2007 for NHSII, and 1986-2006 for HPFS). Person years were calculated as the duration between the return date of the baseline food frequency questionnaire to the date of T2D diagnosis, date of death, date of the last return of a valid follow-up questionnaire, or end of follow-up (2018 for NHS, 2021 for NHSII, and 2020 for HPFS), whichever came first. To reduce potential reverse causality that participants changed their diet after diagnosis of certain diseases, we stopped diet updates once participants reported myocardial infarction, stroke, coronary artery bypass graft, or cancer during follow-up. The rates for missing covariates were low in the cohorts. Any missing values for physical activity, alcohol intake, total energy intake, food and nutrients intake, BMI, AHEI, or z scores for socioeconomic status during follow-up were first replaced by valid values in the previous questionnaire cycle. For the remaining missing data of individual food items (<0.5%), we replaced the missing values with the median intake in the cohort. For missing values of smoking status, oral contraceptive use, menopausal status, and postmenopausal hormone use, we used the missing covariate indicator method. In primary analyses, we used Cox proportional hazards models to estimate the hazard ratios and corresponding 95% confidence intervals (CIs) for the association between consumption levels of total, dark, and milk chocolate and risk of T2D.39 To maximize statistical power and reduce within person variations, we cumulatively averaged total chocolate intake from baseline 1 and dark and milk chocolate intake from baseline 2. We categorized regular chocolate consumption into four groups of levels, including never or <1 serving/month (reference level), 1 serving/month to <1 serving/week, 1-4 servings/week, and ≥5 servings/week. The Cox models were stratified by age and calendar time. Fully adjusted models accounted for age, calendar year, ethnicity (white, African American, Asian, and other ethnicities), smoking status (never, former, current (cigarettes/day: 1-14, 15-24, or ≥25), or missing), alcohol intake (g/day: 0, 0.1-4.9, 5.0-14.9, and ≥15.0 in women, 0, 0.1-4.9, 5.0-14.9, 15.0-29.9, and ≥30.0 in men, or missing), family history of diabetes (yes/no), menopausal status and postmenopausal hormone use (premenopausal, postmenopausal (never, former, or current hormone use), or missing, for women), use of oral contraceptives (yes, no, NHSII only), physical activity (metabolic equivalent task hours (MET-h)/week: <3, 3.0-8.9, 9.0-17.9, 18.0-26.9, ≥27.0, or missing), baseline BMI (<21.0, 21.0-22.9, 23.0-24.9, 25.0-26.9, 27.0-29.9, 30.0-32.9, 33.0-34.9, ≥35.0, or missing), multivitamin use (yes/no), baseline hypertension, baseline hypercholesterolemia, total energy intake (continuous, kcal/day), and AHEI (fifths). For analyses of chocolate subtypes, we further adjusted total chocolate intake before baseline in the models (as cumulative averages for 1980-2002 (NHS), 1991-2003 (NHSII), 1986-2002 (HPFS)). Analyses were run separately in each cohort first, and then data were pooled to calculate overall study estimates, with cohort origin further adjusted in the model. Sensitivity analyses were conducted using the pooled dataset: using baseline chocolate intake level instead of time varying intake; using most recently updated data on chocolate intake instead of cumulatively averaged intake; adjusting for baseline BMI as a continuous instead of categorical variable; adjusting for time varying BMI instead of baseline BMI to explore the mediation effect of BMI; adjusting for time varying waist circumference instead of baseline BMI; adjusting for consumption of specific food and beverages high in flavan-3-ols (eg, tea, red wine, apples, pears, blueberries, bananas, and peppers) instead of AHEI; adjusting for consumption of red and processed meat, fruit and vegetables, sugar sweetened beverages, and whole grains instead of AHEI; further adjusting for each individual intake of flavonoids; further adjusting for saturated fat intake; and further adjusting for added sugar intake. In the pooled dataset, we examined potential effect modification by age (<70 years, ≥70 years), sex (male, female), baseline BMI (<30, ≥30), physical activity (