Alcohol Intake, Drinking Pattern, and Risk of Type 2 Diabetes in Three Prospective Cohorts of U.S. Women and Men Free

Excessive alcohol consumption is well documented to be associated with various adverse health outcomes, including injuries, cancer, mental health issues, certain cardiovascular diseases, and liver diseases. It can be especially harmful for specific populations, such as pregnant women and young adults (1). The World Cancer Research Fund and the Dietary Guidelines for Americans both recommend limiting alcohol consumption and advise against initiating drinking for any reason (2,3). In contrast, numerous epidemiologic studies have observed a lower risk of type 2 diabetes (T2D) associated with light to moderate alcohol consumption (≤2 drinks or 30 g ethanol per day for men and 1 drink or 15 g ethanol per day for women). Key concerns for interpreting this inverse association include potential reverse causation by including former drinkers who quit due to (subclinical) illness as nondrinkers as well as confounding by socioeconomic status (SES) and lifestyle factors. There has been some evidence based on study designs that were less affected by these biases. A meta-analysis of 55 cohort studies found that compared with lifetime abstainers, alcohol consumption of 0.1 to 49 g/day was associated with a lower risk of T2D in women (4). Adjusting for SES and comprehensive lifestyle factors, we previously found that alcohol consumption was associated with lower HbA_1c, which could be seen as a surrogate end point of diabetes (5). A meta-analysis of 14 randomized controlled trials showed 1 to 5 drinks/day reduced fasting insulin (6). Those trials with a longer duration as months to years showed these benefits consistently.

The concerns about reverse causation and confounding have driven the conduct of Mendelian randomization (MR) studies, with most reporting null findings for genetically predicted alcohol level and T2D risk (7). While the validity of findings from MR studies largely depends on whether three key assumptions hold, only half of published MR studies evaluated all three assumptions (7). The relevance of MR studies to this research question is further damped by the use of genetic variants that are more related to alcoholism rather than light to moderate drinking and by the limited ability of these genetic variants to reflect drinking pattern such as frequency and drinking with meals (8). An early exploration in the Health Professionals Follow-up Study (HPFS), based on 1,571 cases of T2D, actually suggested that drinking days per week, not the quantity consumed, drove the inverse relationship between alcohol and diabetes, although this study was subject to the limited number of cases and an exclusively male population (9). Only two other studies have examined frequency and quantity jointly, and one of them yielded inconsistent conclusions with HPFS (10,11). Of note, both studies were conducted in Japanese men with much higher alcohol intake and were not controlled for confounding due to diet or energy intake.

In the current analysis, we investigated further the association of alcohol intake and drinking pattern with incident T2D in the Nurses’ Health Study (NHS), Nurses’ Health Study II (NHSII), and HPFS, with 18,000 additional T2D cases. The investigation in women is important considering differences in ethanol metabolism and in alcohol intake range and pattern between men and women. The large number of cases and repeated dietary assessments over 30 years facilitate a better control for reverse causation through exclusion criteria and sensitivity analyses and sufficient power for joint analysis of alcohol intake and drinking frequency. We hypothesized that moderate alcohol consumption is associated with a lower risk of T2D and that this association is mainly driven by drinking frequency instead of quantity consumed.

We used the data from three prospective cohorts involving both men and women in the U.S. NHS included 121,700 female nurses aged 30–55 years at enrollment in 1976 (12). NHSII included 116,429 female nurses aged 25–42 years at enrollment in 1989 (12). HPFS included 51,529 male health professionals aged 40–75 years at enrollment in 1986 (13). Participants were followed biennially through mailed questionnaires to collect lifestyle and medical information. Retention rates exceeded 90% for all cohorts. The study protocols were approved by the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health institutional review boards. All participants provided written consent.

Our analyses defined the baseline as the year when alcohol intake data were first collected (1980 for NHS, 1991 for NHSII, and 1986 for HPFS). In that baseline questionnaire, participants were also asked whether their alcohol intake had greatly decreased or increased during the past 10 years. We excluded those who were nondrinkers at baseline and reported a great decrease in alcohol intake (n = 16,096) to reduce the impact of misclassification and potential reverse causation. We excluded participants who had a baseline history of diabetes, cancer, stroke, myocardial infarction, angina, or coronary artery bypass grafting (n = 22,121). We also excluded participants with incomplete food frequency questionnaires (FFQs) or implausible energy intake (n = 47,142) as well as those with missing baseline alcohol intake (n = 369). These criteria resulted in a final population of 200,969 participants (77,326 from NHS, 86,718 from NHSII, and 36,925 from HPFS).

Alcohol intake was assessed via validated FFQs every 4 years since baseline (14). Each FFQ asked participants about their average frequency over the previous year of consuming one standard serving of each alcoholic beverage, with the options ranging from “never or less than once per month” to “6 or more times per day.” Alcohol intake, in grams per day, was then calculated as the sum of daily number of servings multiplied by the ethanol content for one serving of each alcoholic beverage (14.0 g of alcohol for a 1.5 oz/44 mL serving of liquor, 12.8 g for a 12 oz/355 mL serving of regular beer or 11.3 g for light beer, and 11.3 g for a 4 oz/118 mL serving of wine). The serving size for wine was increased to 5 oz/148 mL starting in 2003, and the alcohol content was adjusted proportionately to reflect the change. The FFQ-measured alcohol intakes have been validated against multiple diet records (correlation coefficients r = 0.80, 0.81, and 0.83 for liquor, beer and wine in women, and 0.86, 0.76, and 0.70 in men, respectively) and showed the expected linear association with blood levels of HDL-cholesterol (5,15).

In addition to the quantity, participants were also asked the frequency of alcohol use in a typical week in 1986, 1988, 1996, 2000, and 2004 in NHS, in 1989, 2005, and 2009 in NHSII, and in 1986, 1988, 1998, 2002, 2004, 2006, 2008, and 2012 in HPFS. This drinking days per week measure correlated highly with that from diet records as well (r = 0.79) (15). In 1994, participants in HPFS were additionally asked the proportion of alcohol consumed with meals, with options ranging from “<25%” to “>75%.” Similarly, the 1996 HPFS questionnaire and 1997 NHSII questionnaire asked the proportion consumed with meals for each type of beverage.

Incident T2D cases were first identified by self-report through the main biennial questionnaire and then confirmed by a supplementary questionnaire on the symptoms, diagnostic tests, and treatment of diabetes. Before 1998, cases were confirmed if at least one of the National Diabetes Data Group criteria was indicated in the supplementary questionnaire (16): 1) one or more classic symptoms (excessive thirst, polyuria, weight loss, excessive hunger) plus fasting glucose concentrations ≥7.8 mmol/L or random glucose concentrations ≥11.1 mmol/L; 2) at least two elevated glucose concentrations on different occasions (fasting concentrations ≥7.8 mmol/L, random glucose concentrations ≥11.1 mmol/L, and/or concentrations of ≥11.1 mmol/L after ≥2 h shown by oral glucose tolerance testing) in the absence of symptoms; or 3) treatment with hypoglycemic medication (insulin or oral hypoglycemic agent). After 1998, cases were confirmed using the American Diabetes Association criteria, which lowered the threshold for fasting glucose from 7.8 to 7.0 mmol/L (17). The high validity of the self-reported T2D confirmed by the supplementary questionnaire was previously demonstrated by comparison with medical records (18,19).

Total energy intake and overall diet quality represented by the Alternative Healthy Eating Index (AHEI)-2010 were computed using data from FFQs and the Harvard University Food Composition Database (20,21). Neighborhood SES was assessed based on the addresses reported in biennial questionnaires and linked to 1990 and 2000 U.S. Census data on education, income, house value, and employment status (22).

Person-years of follow-up accrued from the return of the baseline questionnaire until the earliest time of T2D diagnosis, death, loss to follow-up, or the end of follow-up (30 June 2020 for NHS, 30 June 2019 for NHSII, and 31 January 2020 for HPFS), whichever came first.

To investigate the extent to which alcohol intake, drinking frequency, and drinking with meals were associated with incident T2D, we used Cox proportional hazards models with time-varying exposures and covariates to estimate hazard ratios (HRs) and 95% CIs. The models were stratified jointly by age in months and calendar time in 2-year groups and adjusted for race, smoking, physical activity, total energy intake, AHEI-2010 without the alcohol component, coffee intake, multivitamin use, family history of T2D, history of hypertension and hypercholesteremia, SES, and menopausal status and hormone use (among women). Tests for linear trend were performed using the continuous alcohol intakes and drinking days. Tests for nonlinearity were evaluated by three-knot restricted cubic splines and likelihood ratio tests comparing the model with and without the cubic spline terms.

In the main analyses, we used the cumulative averages of total alcohol intakes and dietary covariates to better represent long-term diet and reduce measurement errors. In the sensitivity analyses, we examined the associations alternatively using baseline alcohol intake, cumulative averages of the most recent three assessments prior to T2D diagnosis, and cumulative averages excluding the most recent assessment prior to T2D diagnosis. To further evaluate the impact of potential reverse causation, we conducted latency analyses assuming years of 0–4, 4–8, 8–12, 12–16, 16–20, and 20–24 latency for the associations of total alcohol intake and drinking frequency with T2D. For example, in the analysis with a latency period of 4–8 years, we used the alcohol intake in 1980 for cases that occurred between 1984 and 1988.

It is possible that participants at higher risk of T2D might be more health conscious, drink less alcohol, undergo T2D screenings more often, and have more cases diagnosed. In this case, a higher proportion of participants would be diagnosed before experiencing symptoms, and thus, we restricted the analysis to symptomatic cases, identified by at least one symptom of diabetes (e.g., excessive thirst) in the supplementary questionnaire, to check the robustness of our findings. BMI might be a confounder and a mediator in the alcohol-T2D relationship. We adjusted for baseline BMI and waist circumference (WC) in the main analyses to obtain a more conservative estimate of any potential protective associations. The estimates without adjusting for BMI and WC and those adjusting for the time-varying variables were given in sensitivity analyses. Although former regular drinkers were already excluded from baseline nondrinkers, we further tested the association by taking lifetime abstainers as the reference and thereby excluding former drinkers with any level of consumption.

Finally, we examined whether the association between drinking frequency and T2D differed by age (<60, ≥60 years), smoking (never, ever), BMI (<25, 25 to <30, ≥30 kg/m²), AHEI-2010, physical activity, neighborhood SES (below median, above median), and family history of diabetes and history of hypertension (yes, no). Interactions were tested through likelihood ratio tests comparing the model with and without the interaction terms. SAS 9.4 software (SAS Institute) was used for all analyses. We considered two-sided P < 0.005 to be statistically significant and 0.005 ≤ P < 0.05 to be suggestively significant.

The data described in the article, code book, and analytic code will be made available upon request pending approval by the Channing Division of Network Medicine at Brigham and Women’s Hospital and Harvard Medical School. Further information including the procedures to obtain and access data from the NHS and the HPFS is described at https://www.nurseshealthstudy.org/researchers (contact e-mail: [email protected]) and https://hsph.harvard.edu/research/health-professionals/resources/for-external-collaborators/.

The median follow-up ranged from 24 to 33 years across the three cohorts. Over a total of 5,560,412 person-years, we documented 9,635 incident T2D cases in NHS, 7,476 cases in NHSII, and 3,440 cases in HPFS. Approximately 35% of participants in NHS and NHSII were baseline nondrinkers, while only 12% of participants in HPFS were baseline nondrinkers. HPFS also had the highest average alcohol intake among drinkers at 14.7 g/day (approximately one drink), whereas NHSII had the lowest average alcohol intake at 5.3 g/day (Supplementary Fig. 1). Alcohol consumption was positively associated with smoking (Table 1). Other than that, alcohol drinkers (especially moderate drinkers) generally had healthier lifestyles, such as higher physical activity and better diet quality, a lower prevalence of T2D family history, and a higher SES than nondrinkers.

We observed an inverse association between total alcohol intake and risk of T2D in the three cohorts, either using nondrinkers or the lowest intake category (0.1–4.9 g/day) as the reference (Table 2). The significant inverse association extended to 75 g/day in NHS and HPFS and then became nonsignificant for intakes above (HR ∼0.4 with wide CIs, data not shown). The lower risk of T2D in drinkers than in nondrinkers persisted when using alternative modeling of exposure (Supplementary Table 1 and Supplementary Fig. 2), when assuming a latency period up to 20 years (Supplementary Table 2), when using lifetime abstainers as the reference, and when restricting to symptomatic T2D cases (Supplementary Table 3). Table 1 shows lower BMI in female drinkers compared with nondrinkers and a flat distribution of BMI in men. Not adjusting for BMI and WC resulted in stronger inverse associations in all cohorts. Adjusting for time-varying instead of baseline BMI and WC led to slightly attenuated estimates, but the inverse association persisted (Supplementary Table 3).

Most alcohol drinkers consumed it 1–2 days per week. People who drank more often tended to be older, had higher SES, healthier lifestyles except for smoking, and less T2D family history (Supplementary Table 4). Higher drinking frequency also positively correlated with total alcohol intake (coefficient = 0.6 in women and 0.8 in men). After accounting for these covariates, data suggested a lower T2D risk with higher drinking frequency following a linear trend (Table 3 and Supplementary Fig. 3) For example, compared with drinking 1–2 days per week, HRs (95% CIs) for drinking 5–6 days were 0.73 (0.65, 0.83), 0.73 (0.62, 0.86), and 0.76 (0.67, 0.86) in NHS, NHSII, and HPFS, respectively, and each day increment was associated with a 6–7% lower risk. The association remained for latency periods up to 20 years (Supplementary Table 5), when restricting to symptomatic cases and after adjustment for time-varying BMI and WC (Supplementary Table 6) in NHS and HPFS. The association seemed to be stronger in the subgroups with overweight and obesity, although P values for multiplicative interaction were not significant (Supplementary Table 7). Notably, the additional adjustment for drinking frequency in the multivariable model largely diminished the association of higher alcohol intake with lower T2D risk, and nearly all risk estimates became nonsignificant (Table 2 and Supplementary Fig. 4). The adjustment for frequency also substantially attenuated the association across all alcoholic beverage types (Supplementary Tables 8–10).

We categorized participants jointly by drinking frequency and the quantity of alcohol consumed per drinking day (calculated as total alcohol intake divided by drinking frequency) (Supplementary Figs. 5–7). Results showed drinking frequency primarily drove the inverse relationship with T2D (Fig. 1 and Supplementary Table 11). The risk of T2D in men who drank 1–2 days per week did not significantly differ from that in nondrinkers, regardless of the quantity consumed. The lower risk of T2D began at 3–4 days per week and was strongest for drinking ≥5 days per week, whether it was <10 g or ≥30 g per drinking day (HR ∼0.73). In women, lower risk of T2D was seen in almost all joint categories. On average, the greatest benefit was observed for drinking ≥5 days per week, followed by drinking 3–4 days, and then 1–2 days per week. Modest variations across quantity existed; for instance, drinking ≥30 g/day resulted in less benefit than its moderate counterparts in some drinking frequency strata.

Among the subcohort of NHSII (25,000 participants) and HPFS (13,000) who were asked about drinking with meals, most consumed alcohol outside of meals. Compared with consuming <25% of alcohol with meals, consuming 25–75% of alcohol with meals was associated with a lower risk of T2D in women (and in men only for liquor). Consuming >75% of alcohol with meals showed a weaker inverse association in women and a null association in men (Supplementary Table 12). This association was not substantially affected by drinking frequency or quantity of the beverage consumed.

Based on three large U.S. cohorts with former drinkers excluded, we found a significantly lower risk of T2D among alcohol drinkers compared with nondrinkers in both men and women. This finding was robust to extended latency periods, alternative modeling of exposure, and restriction to symptomatic cases. The lower risk of T2D among drinkers was largely driven by the drinking frequency instead of the quantity consumed. The inverse association was apparent at 1–2 drinking days per week in women and 3–4 days per week in men and was strongest for ≥5 days per week, regardless of drinking <10 g or ≥30 g per drinking day. Drinking with meals had inconsistent but possibly modest inverse associations with T2D.

While extensive studies have been conducted on total alcohol intake and T2D, limited research has focused on drinking pattern and the relative importance of frequency versus quantity. The Melbourne Collaborative Cohort Study reported that men who consumed 210 g of alcohol over 1–3 days (quantity much higher than in our cohorts) had a fivefold risk of developing T2D than nondrinkers, while spreading over the same amount over more days did not increase the risk (23). Our estimate of a 20–30% lower T2D risk comparing ≥5 drinking days with 1–2 drinking days per week was comparable in magnitude with the same association in a Danish cohort (24), and our estimate of a 6% lower risk for each day increment was consistent with that (7%) in the previous HPFS study (9). Apart from that study, the only two joint analyses of drinking frequency and quantity with T2D so far were from two cohort studies in Japanese men, both using a single assessment at baseline for alcohol consumption (10,11). The Kansai Healthcare Study enrolled 10,631 participants aged 40–55 years (11). Similar to our findings in men, they observed no association for 1–3 drinking days per week and a decreased T2D risk for 4–7 days per week, regardless of whether consuming <2 or 2–4 drinks on each occasion (HR ∼0.74). The other study, Toranomon Hospital Health Management Center Study 11 (TOPICS11), enrolled 1,650 participants and reported “usual quantity per occasion was more important than weekly drinking frequency” (10). However, this conclusion was drawn from the observation that 6–10 U.S. drinks per occasion consistently increased the risk of T2D, irrespective of drinking frequency. In contrast, our study focused on moderate intakes instead of binge drinking. Also, their finding of lower risks with increased drinking frequency (HR 0.6 comparing 4–6 days with 0–1 day per week) was in line with ours indeed.

Nonetheless, we do not exclude the possibility that the association of alcohol intake or pattern with T2D could vary across ethnicities and sexes. It is among Asian men that an increased risk of T2D with alcohol intake has been most frequently reported, probably due to the prevalent binge drinking culture and genetic polymorphisms involving ethanol metabolism (25). A study from South Korea reported that regular nonbinge drinking was associated with a higher risk in men but a lower risk in women (26). The role of BMI or adiposity in the relationship between alcohol and T2D has not been fully elucidated, but it has been hypothesized to mediate a larger proportion of the relationship in women than in men (4). Similar to the current study, previous studies have noted BMI and WC adjustment resulted in greater attenuation of the association in women (24,27). Alcohol consumption was actually found to be related to higher or similar weight in men but to lower weight in women (28–31). A possible explanation is energy expenditure substantially increases in women yet only moderately changes in men after consuming alcohol (4).

Other potential mechanisms include the influence of alcohol on cardiometabolic biomarkers. Meta-analyses of randomized controlled trials have shown that alcohol consumption decreases HbA_1c and fasting insulin and increases adiponectin levels (6,32). Early experimental studies suggest alcohol may reduce HbA_1c by suppressing the acute postprandial rise in blood glucose and enhancing early insulin secretion (33,34). This is also supported by findings that alcohol consumption attenuates the positive association between glycemic load and T2D risk (35). Ethanol also increases the expression of the ADIPOQ gene in adipose tissue, which elevates adiponectin levels—an insulin sensitizer that helps reduce the risk of T2D (36).

Why T2D risk is especially lower in regular or frequent drinkers remains less clear. Previously, we found the lower level of C-peptide and adiponectin associated with alcohol intake was more profound among regular drinkers (≥3 days per week) than nonregular drinkers (5). It is possible that the effect on these biomarkers occurs over a short duration and thus makes frequent consumption more beneficial. Future studies are warranted to replicate our findings and further explore these mechanisms.

The current study is the first to examine the joint pattern of alcohol drinking frequency and quantity with T2D risk among women. We also provided robust evidence on decreased risk of T2D among moderate drinkers by minimizing the impact of reverse causation and confounding through exclusion of former drinkers, a series of sensitivity analyses, repeated assessment, and adjustment for time-varying confounders. The repeated assessment also reduced measurement error and assisted in capturing the long-term habit as well as the dynamic nature of alcohol consumption.

However, we acknowledge the limitation that residual confounding from other lifestyle and socioeconomic factors cannot be entirely excluded. Additionally, not all variables were collected at the same time or with the same interval. For example, drinking frequency was only assessed three times in NHSII, with the longest interval as 16 years, which might have partially contributed to the less robust results in NHSII beyond the young age or low alcohol intake of its participants. Also, we had to estimate the quantity consumed per drinking day by dividing total alcohol intake by drinking frequency. Future studies with assessment with a separate question would be ideal to reduce potential measurement error. Our study participants consisted of middle-aged White health professionals with generally healthy lifestyles and moderate alcohol consumption. Therefore, our findings are most applicable to populations with similar characteristics. Future studies in populations of other races/ethnicities, higher consumption levels and different alcohol-confounders pattern are warranted to extend generalizability, evaluate residual confounding, and confirm these findings.

In summary, we found light to moderate alcohol consumption, especially regular light drinking, was associated with a lower risk of T2D among both men and women. However, we do not suggest initiating alcohol consumption as a means of preventing diabetes, especially among younger adults. The detrimental effects of high alcohol consumption on cancer, accidents, and mental health issues cannot be ignored when balancing risks and benefits of alcohol consumption.

Acknowledgments. The authors thank the participants and staff of the NHS, NHSII, and HPFS for their valuable contributions.

Funding. The NHS is supported by National Institutes of Health (NIH) National Cancer Institute (NCI) grants UM1 CA186107 and R01 CA49449. The NHSII is supported by NIH NCI grants U01 CA176726 and R01 CA67262. The HPFS is supported by NIH NCI grant U01 CA167552. This work was in addition supported by American Cancer Society Clinical Research Professor grant CRP-23-1014041 (to E.L.G.), NIH NCI grant R00 CA215314 (to M.S.), American Cancer Society Mentored Research Scholar grant MRSG-17-220-01-NEC (to M.S.), a research grant from the Ottogi Ham Taiho Foundation (to J.H.), and the National Research Foundation of Korea grant RS-2024-00349274 funded by the Ministry of Science and Information and Communication Technology (to J.H.).

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. The authors assume full responsibility for the analyses and interpretation of these data. The funding sources played no role in the study design, data collection, data analysis, and interpretation of results, or the decisions made in preparation and submission of the article.

Duality of Interest. No potential conflicts of interest relevant to this article were reported

ICMJE Statement. All authors affirm that authorship is merited based on the ICMJE authorship criteria (www.icmje.org/disclosure-of-interest/).

Author Contributions. X.L. conducted analyses and drafted the manuscript. J.H. verified the data. X.L., J.H., and E.L.G. conceived and designed the study. S.A.S.-W., M.S., L.L., K.J.M., and E.B.R. reviewed and edited the manuscript. All authors helped to interpret the data and critically revised the manuscript for important intellectual content. E.L.G. provided supervision. J.H. controlled the decision to publish and attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. J.H. is the guarantor of this work and, as such, 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.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Cheryl A.M. Anderson and Cuilin Zhang.