Alcohol use patterns and risk of diabetes onset in the 1979 National Longitudinal Survey of Youth Cohort

Abstract

One of the major limitations in studying alcohol’s effect on risk for diabetes is the issue of classifying drinking patterns across the life course prior to the onset of diabetes. Furthermore, this research often overlooks important life course risk factors such as obesity and early-life health problems that may complicate estimation of the relationship between alcohol and diabetes. This study used data from the US National Longitudinal Survey of Youth 1979 cohort of 14–21 year olds followed through 2012 (n=8,289). Alcohol use was captured through time-varying measures of past month volume and frequency of days with 6+ drinks. Discrete-time survival models controlling for demographics, early-life characteristics and time-varying risk factors of employment, smoking, and body mass index (BMI) group, stratified by sex and race/ethnicity, were estimated. Increased odds of diabetes onset was found among lifetime abstainers for women compared to the low volume reference group (odds ratio (OR) 1.57; 95% Confidence Interval (CI) 1.07–2.3). Increased odds of diabetes onset was also found among women who reported drinking 6+ drinks in a day on a weekly basis during the prior 10 years (OR 1.55; CI 1.04–2.31). Models interacting alcohol and BMI groups found increased odds of diabetes onset from lifetime abstention among overweight women only (OR 3.06; CI 1.67–5.60). This study confirms previous findings of protective effects from low volume drinking compared to lifetime abstention and harmful effects from regular heavy occasion drinking for women. Further, protective effects in this US sample were found to be limited to overweight women only.

INTRODUCTION

Population-based prospective cohort studies have shown a reduced risk of diabetes among low-to-moderate alcohol drinkers relative to abstainers and heavy drinkers (Marques-Vidal et al., 2015) or have found significant U-shaped relationships between alcohol and diabetes risk (Koloverou et al., 2015). Studies examining the effect of binge drinking have produced mixed results, with findings of both increased risk (Hodge et al., 2006) and no effect (Rasouli et al., 2013). Reviews and meta-analyses, including those considering misclassification bias, have consistently indicated that moderate alcohol consumption reduces diabetes risk, and found nonlinear dose-response relationships (Baliunas et al., 2009; Carlsson et al., 2005; Howard et al., 2004; Koppes et al., 2005; Pietraszek et al., 2010). Few studies have considered a lifecourse perspective on alcohol in relation to diabetes risk. When using lifetime abstinence (versus current abstinence), a study of the U.S. National Alcohol Survey found protective effects of moderate drinking compared to lifetime abstention for diabetes (Kerr and Ye, 2010). However, findings of increased risk among heavy drinkers are less consistent (Li et al., 2016). Where gender has been considered, a stronger protective effect of moderate alcohol use among women has been found (Hodge et al., 2006; Knott et al., 2015).

A few studies have considered the potential for interactions between obesity, measured via body mass index (BMI), and alcohol use patterns in diabetes risk. The possibility that the alcohol risk relationship with diabetes is confounded by diet was not supported by one of the few studies where detailed information on both was available (Imamura et al., 2009). A French study focused on wine and diabetes risk found protective effects for women only, and furthermore, the effects appeared to be limited to women who were overweight (Fagherazzi et al., 2014).

A European study found reduced risk among moderate and heavier drinking women compared to all non-drinkers with greater risk reductions among overweight and obese women compared to normal weight women (Beulens et al., 2012). A study from New Zealand also found reduced risk compared to no current drinking for both genders in the normal weight and overweight groups, but not among the obese (Metcalf et al., 2014). A European case-control study found increased risk from moderate to heavy alcohol use compared to light or no use among normal weight respondents and reduced risk among overweight respondents (Eckel et al., 2015). Although these studies results are somewhat conflicting and vary in the groups considered and methodological limitations, they indicate the potential importance of BMI, a major diabetes risk factor, for determining diabetes risks from alcohol use patterns.

Research on disparities in diabetes have found that Blacks and Hispanics have 50% to 100% higher rates of measured diabetes compared to Whites. While some studies reported differences even after controlling for risk factors including poverty (Gaskin et al., 2014), other studies suggest that these racial/ethnic disparities are eliminated after controlling for obesity, risk behaviors, insurance status, and other socioeconomic measures (LaVeist et al., 2009; Link and McKinlay, 2009). If these factors do explain racial/ethnic disparities in diabetes, then differences in drinking patterns could play a role in the disparities, however, little is known regarding differential alcohol risk relationships among U.S. racial/ethnic groups.

The current study utilizes data from the U.S. National Longitudinal Survey of Youth 1979 Cohort (NLSY79) collected through 2012 to evaluate the risk of diabetes onset in relation to lifecourse-defined alcohol pattern measures. The study design addresses some key limitations of prior alcohol-related health studies including a clearly defined lifetime abstainer group, the inclusion of repeated measures of alcohol volume and heavy drinking days, as well as repeated measures of other risk factors such as BMI and smoking, and attention to gender and racial/ethnic differences. Study goals focused on hypotheses regarding increased risk for diabetes onset among lifetime abstainers and heavy drinkers with an expectation of finding such effects among women only. Secondarily, we consider whether these effects vary across White, Black and Hispanic groups and the interplay of BMI with alcohol.

METHODS

Secondary data analysis was conducted using panel data from the NLSY79, an on-going study conducted by the U.S. Bureau of Labor. The NLSY79 used a stratified, clustered design to select a nationally representative sample of individuals born between 1957 and 1964. A sample of 6,111 non-institutionalized, civilian youths ages 14–21 and an oversample of 5,295 civilian Hispanic, Black, and economically disadvantaged youths were selected in 1979. Respondents were re-interviewed annually from 1979 through 1994 and every two years since then. The initial NLSY79 response rate was 90%, and retention rates during follow-up assessments were 90% or better during the first 16 survey waves and remained above 80% in more recent waves. This study used data up to the 2012 survey wave when respondents were aged 47 to 55 years. The study was approved by the Public Health Institute IRB #I14-007.Starting in 1998, as respondents turned 40 and then later 50, at the subsequent NLSY interview, they were given a specific module on general health and diagnoses of health conditions. Subjects who missed either of these health modules were excluded. Subjects who reported health problems before 1982 were also excluded because their alcohol consumption information was not obtained until the 1982 survey.

Measures

The outcome variable is the age of onset of diabetes. In the 40+/50+ Health Modules, respondents were asked, “Have you ever had, or has a doctor ever told you that you have diabetes or high blood sugar?” If respondents answered yes, then they reported the month and year of onset. Age of diabetes onset was calculated based on respondents’ month and year of birth provided at the baseline 1979 survey.

For alcohol consumption, respondents were asked about their alcohol use in survey years 1982–1985, 1988, 1989, 1994, 2002, and every other year starting in 2006 to 2012. Because of the intermittent years of alcohol availability, and changes in wording of alcohol questions across survey years, we created a categorical repeated measure of past month alcohol consumption that combined total volume and frequency of heavy drinking (HD) days (6 or more drinks in 1 sitting). Past-month heavy drinking frequency was categorized into: none, less than once a week and weekly or more. For each year, our final alcohol variable was based on a combined total volume and HD frequency to capture no alcohol, low volume with no HD, low volume with any HD, risky volume and monthly HD, risky volume and weekly HD, and high volume with any HD. Finally, to address concerns of appropriate non-drinking groups in prior alcohol-related health studies, we further distinguished the no alcohol group into lifetime abstainers and no current alcohol use/former drinkers. We describe our construction of lifetime abstainers elsewhere (Kerr et al., 2017). We use the low volume/non-heavy drinking group as the reference. To capture history of HD, we created two time-varying indicators to capture any monthly and any weekly HD of 6+ drinks in the ten years prior to the current year.

Alcohol volume calculations differed across the years due to changes in questions asked. For 1982–1985, volume was the sum of past week wine, beer and spirits drinks. For 1988–2012, total volume was past month usual quantity multiplied by usual frequency for all beverages combined. The 1982–1985 volume measure resulted in much higher and more variable numbers of drinks due to these measurement differences. For example, the mean monthly volume ranged from 57 to 68 drinks in 1982–1985, while the mean values are all less than 20 drinks for later survey years. Two adjustments were made to improve comparability across years. For 1988–2012, the measure of 6+ days was used to adjust the volume upward. Volume for 1982–1985 was adjusted downward, dividing by the factors 1.25, 1.5, 1.75, 2, 2.5, 3 and 4 for each 5-percentile increase starting from the 65th percentile to address outliers resulting from high numbers of drinks weekly. The resulting adjusted volume distribution from 1982–1985 is consistent with volume from the later years. We then created categories of total volume: zero, low (less than or equal to 14/7 (men/women) drinks per week), risky (more than 14/7 drinks and less than 28/14 drinks per week), and high (more than 28/14 drinks per week).

Covariates

We characterized respondents as foreign or U.S.-born. Self-reported race/ethnicity from 1979 was categorized into (1) White (reference group), (2) Black, (3) Hispanic, (4) Native American, and (5) Other racial/ethnic groups including Asian, Hawaiian, and Pacific Islander. Age was calculated based on month and year of interview from month and year of birth. Frequency of religious participation in 1979 was categorized into none up to more than once per week (0–5). Educational attainment by age 25 included less than high school, high school graduate, some college, and college degree or more.

To account for early life health conditions, we used a work-related health limitation construct from the 1979 survey. Respondents were asked three questions on work-related health limitations: (1) “(Are you/Would you be limited in the kind of work you (could) do on a job for pay because of your health?”; (2) “(Are you/Would you be) limited in the amount of work you (could) do because of your health?”; and (3) For those not working for pay, “Would your health keep you from working on a job for pay now?” Respondents reporting yes to any of these questions were considered to have had an early life health problem. Prior studies have found strong correlations of this construct with disability, health impairments, and chronic health conditions (Besen and Pranksy, 2014; Burkhauser et al., 2002; Walsemann et al., 2008), and we found this predictive of lifetime abstinence (Kerr et al., 2017).

To capture factors across the risk period, we include repeated measures of poverty status (yes/no), marital status (never married, married, separated, divorced, widowed), employment (unemployed, employed, out of the labor force, active armed forces), and whether they have children (yes/no). These measures were obtained from each survey year.

Finally, we accounted for health risk behaviors of smoking and obesity. A set of smoking questions was asked in 1992, 1994, 1998, 2008, 2010, and 2012 survey years, including ever smoked at least 100 cigarettes in lifetime, age starting to smoke daily, current smoking, and cigarettes smoked per day. Summarizing the available data, we created an age of onset of daily smoking and carried it forward from that age, and then grouped current smoking status for each year into non-smoker, prior daily smoker, and current daily smoker. We also accounted for obesity by calculating BMI (Centers for Disease Control and Prevention). Height was reported in survey years 1981, 1982, 1983, 1985, 2006, 2008, 2010, and 2012. Weight was reported in 1981, 1982, 1985–86, 1988–1990, and every other year after 1992 up to 2012. We carried forward height to calculate BMI scores for each available weight year. BMI was coded into underweight (<18.5), normal (18.5–24.9), overweight (25–29.9), and obese (≥30).

Analyses

Discrete-time survival models were used to model the onset of diabetes in relation to both fixed and time-varying predictors with person-year weights considering both sampling and attrition. Annual diabetes history (defined as the onset of diabetes or not) starting from 1982 was reconstructed for each subject. Subjects who reported onset of diabetes before or in 2012, were censored after that year. All other subjects were censored when they dropped out of the study or reached 2012. Time-varying covariates such as frequency of heavy drinking and employment were carried forward for the years without interview until data were available at the next interview year. Separate models were estimated for men and women, and for White, Black and Hispanic men and women. Further models included interactions with BMI groups of overweight and obesity for the lifetime abstainer (compared to low volume/non-heavy drinking group) and prior 10 year heavy per occasion drinking measures.

Our discrete-time survival models were implemented by a pooled logistic regression model (Hernán et al., 2000; Singer and Willett, 1993), treating each reconstructed person-year as an observation. To avoid instability of estimation, linear and quadratic terms for time were added into our discrete-time survival model with the other time variant and invariant covariates. Due to a substantial number of dropouts from the study over the years, an inverse probability weighted estimation (Hernán et al., 2000; Hernán et al., 2002) was combined with the pooled logistic regression. Specifically, we created a time-varying censoring indicator for year t and defined it to be 1 if a subject dropped out of the study or censored by the last survey year of 2012 and 0 otherwise. Then the subjects’ probabilities of being zero in year t by the censoring indicator were estimated by logistic regression with all relevant covariates including the main predictor variable of alcohol and other controlling covariates. The probabilities obtained from these logistic regressions were used as the denominator in the calculation of the weight (w_it) for the i^th subject in year t. For robustness, the probabilities of being zero, estimated by the same logistic regressions without including those covariates of main interest, such as alcohol, were used as the numerator in the calculation of w_it (Hernán et al., 2000; Hernán et al., 2002). The multiplication ${\Pi }_k^t{w}_{ik}$ up to year t was used to generate the censoring weights for each person-year. Survey weights are available for each year to adjust the sample to its original sampling frame so as to be representative of U.S. youth demographics in 1979. These survey weights and censoring weights were multiplied together and used in the IPW estimation of discrete-time survival analysis. All analyses were conducted in Stata version 14.2.

RESULTS

The final sample included 8,289 respondents. Table 1 presents the characteristics of the sample stratified by gender. A majority of respondents was U.S.-born. Among women, 61% were White, 15% Black, and almost 7% Hispanic. Most women had at least a high school diploma and 20% had a college degree. The racial/ethnic breakdown and educational attainment for men were similar to women. The main sex difference was that 7% of women reported a health limitation at baseline compared to 5% of men. However, the prevalence of diabetes was similar for women (11%) and men (10%).

Table 1.

Sample Characteristics, National Longitudinal Survey of Youth 1979 Cohort (N=8,289)

weighted % (N)
	Female (n=4,269)		Male (n=4,020)
Time-Invariant Variables
Age in 1982 (Mean)	21.79		21.65
Religious Attendance (Mean)	2.35		2.04
Race/Ethnicity
White	61.1%	(1,713)	63.4%	(1,667)
Black	14.8%	(1,266)	13.9%	(1,169)
Hispanic	6.5%	(764)	6.1%	(724)
Native American	6.1%	(196)	4.7%	(142)
Other	11.5%	(330)	11.9%	(318)
U.S. Born
Yes	95.7%	(3,987)	95.7%	(3,749)
Education Attainment
Less than High School	11.4%	(664)	13.9%	(767)
High School	44.8%	(1,885)	45.3%	(1,844)
Some College	23.7%	(1,047)	20.0%	(786)
College or more	20.1%	(682)	20.9%	(635)
Diabetes	9.4%	(482)	9.4%	(418)
Time-Varying Variables (1982–2012)
Has Children
Yes	68.2%		46.1%
Poverty Status
No	85.9%		89.5%
Yes	14.1%		10.5%
Marital Status
Never Married	22.5%		31.7%
Married	58.4%		54.3%
Separated	4.3%		2.9%
Divorced	13.8%		10.8%
Widowed	1.0%		0.3%
Employment Status
Employed	74.8%		85.8%
Unemployed	2.0%		2.9%
Out of Labor Force	22.9%		8.8%
Active Services	0.3%		2.5%

The fully adjusted discrete-time survival models predicting diabetes onset for women and men are presented in Table 2. Lifetime abstainers were found to have a significantly increased risk (OR=1.57) compared to the reference group who reported drinking less than 7 drinks per week on average and no 6+ heavy drinking days. No other significant relationships were found for current drinking patterns for women and no significant results were found for men. Prior heavy occasion drinking during the past 10 years was found to increase the risk of diabetes onset for women reporting prior weekly 6+ days (OR=1.55) and to reduce diabetes onset for men reporting prior monthly (but less than weekly) 6+ days (OR=0.71).

Table 2.

Adjusted Odds Ratios (OR) and 95% Confidence Intervals (CI) from Discrete-Time Survival Models Predicting Onset of Diabetes by Sex, National Longitudinal Survey of Youth 1979 Cohort (N=8,289) followed to 2012

	Female		Male
	OR	95% CI	OR	95% CI
Current Drinking Status (Ref Low Vol/Non-Heavy)
Lifetime Abstainer	1.57	(1.07, 2.30)	0.97	(0.53, 1.80)
Current Non-Drinker	1.29	(0.98, 1.71)	1.16	(0.85, 1.57)
Low Vol/Heavy Drinker	0.80	(0.46, 1.39)	1.06	(0.71, 1.60)
Risky Vol/Monthly Heavy Drinker	0.87	(0.40, 1.88)	0.80	(0.31, 2.08)
Risky Vol/Weekly Heavy Drinker	1.22	(0.47, 3.17)	0.85	(0.42, 1.70)
High Vol/Heavy Drinker	1.06	(0.47, 2.36)	0.88	(0.40, 1.94)
Prior Heavy Drinking (Past 10 Years)
Monthly Heavy Drinking	0.96	(0.70, 1.32)	0.71	(0.53, 0.95)
Weekly Heavy Drinking	1.55	(1.04, 2.31)	0.97	(0.71, 1.33)
Age	1.10	(1.05, 1.15)	1.20	(1.14, 1.27)
U.S. Born	1.01	(0.58, 1.76)	0.68	(0.39, 1.20)
In Poverty	1.32	(0.98, 1.77)	1.00	(0.65, 1.56)
Has children	0.88	(0.80, 0.97)	0.95	(0.83, 1.07)
Frequency of Religious Attendance 1979	1.03	(0.96, 1.11)	1.05	(0.98, 1.13)
Work Health Limitation 1979	1.54	(1.08, 2.19)	1.40	(0.87, 2.24)
Race/Ethnicity (Ref White)
Black	1.20	(0.91, 1.58)	0.92	(0.69, 1.24)
Hispanic	1.66	(1.19, 2.31)	1.13	(0.77, 1.67)
Native American	1.56	(0.97, 2.52)	1.15	(0.64, 2.09)
Other race	1.35	(0.88, 2.06)	0.85	(0.53, 1.38)
Smoking Status (Ref Non-smoker)
Previous Daily Smoker	0.86	(0.62, 1.18)	0.92	(0.66, 1.28)
Current Daily Smoker	1.05	(0.79, 1.40)	1.14	(0.84, 1.55)
Body Mass Index (Ref Normal Weight)
Underweight	3.23	(1.50, 6.93)	0.53	(0.07, 4.07)
Overweight	2.77	(1.90, 4.02)	2.67	(1.67, 4.28)
Obese	8.02	(5.75, 11.19)	8.04	(5.13, 12.59)
Marital Status (Ref Married)
Never Married	0.74	(0.53, 1.03)	1.53	(1.04, 2.25)
Separated	0.69	(0.46, 1.03)	1.09	(0.52, 2.29)
Divorced	0.95	(0.69, 1.31)	1.42	(0.94, 2.13)
Widowed	0.35	(0.15, 0.78)	4.57	(1.86, 11.22)
Employment Status (Ref Employed)
Unemployed	0.93	(0.46, 1.86)	1.44	(0.76, 2.72)
Out of Labor Force	1.36	(1.02, 1.81)	0.95	(0.59, 1.55)
In active forces	2.79	(0.40, 19.58)	0.73	(0.21, 2.57)
Education Attainment
HS Grad	1.22	(0.87, 1.69)	0.93	(0.66, 1.32)
Some College	1.74	(1.19, 2.55)	0.88	(0.57, 1.35)
College or more	1.26	(0.78, 2.06)	0.64	(0.38, 1.07)
Intercept	0.00	(0.00, 0.00)	0.00	(0.00, 0.00)

Notes:

^aOR = Odds Ratio

^bIndicates time-varying variable

^cCurrent drinking status is a combined total volume and frequency of heavy episodic drinking of 6+ drinks. Total volume is categorized into low volume (≤7/14 drinks per week for women and men, respectively), risky volume (>7/14 drinks per week), and high volume (>14/28 drinks per week). Frequency of 6+ heavy episodic drinking is grouped into any in the last month, monthly, and week

Results for other predictors include significantly increased risk for women with an early life health-related work limitation, Hispanic women, and women who attended some college but did not graduate. For men, having never married or being a widower significantly increased their risk compared to married men. Obesity was found to be important determinant of risk with similar odds ratios for men and women in the obese and overweight categories. Underweight was also a significant positive risk factor for women only.

Racial/ethnic and sex-stratified analyses of diabetes onset are presented in Table 3. Although few significant results were found, the differences in odd ratios across groups offers some insight into potential disparities in diabetes risk. Most notably, the increased risk from lifetime abstention is elevated (OR=1.77) and nearly significant for White women. Among men, elevated risks for lifetime abstainers among White and Hispanic men were countered by reduced risk among Black men, although wide confidence intervals exclude any inferences from this finding. Increased risks from heavier drinking patterns were seen for all groups of women but the risk relationships varied. For White and Hispanic women, the high volume drinking pattern showed the greatest risk, while for Black women, the risky volume pattern with weekly heavy drinking (6+) days had significantly elevated risk (OR=3.39). Having a history of weekly heavy drinking days in the prior 10 years also elevated the diabetes risk for Black and Hispanic women but not for White women. In general, these disaggregated models seem to shift the risk to current high volume and heavy drinking days compared to the model for all women where prior frequent heavy drinking days were the key risk group.

Table 3.

Adjusted Odds Ratios (OR) and 95% Confidence Intervals (CI) from Discrete-Time Survival Models Predicting Onset of Diabetes by Sex and Race/Ethnicity, National Longitudinal Survey of Youth 1979 Cohort

Female	White		Black		Hispanic
	OR	95% CI	OR	95% CI	OR	95% CI
Current Drinking Status (Ref Low Vol/Non-Heavy)
Lifetime Abstainer	1.77	(0.97, 3.20)	1.17	(0.71, 1.94)	1.03	(0.51, 2.08)
Current Non-Drinker	1.22	(0.78, 1.89)	1.59	(1.06, 2.38)	0.55	(0.29, 1.03)
Low Vol/Heavy Drinker	0.77	(0.31, 1.94)	1.82	(0.86, 3.82)	0.94	(0.43, 2.08)
Risky Vol/Monthly Heavy Drinker	0.57	(0.15, 2.09)	0.97	(0.29, 3.26)	1.89	(0.67, 5.31)
Risky Vol/Weekly Heavy Drinker	1.55	(0.33, 7.41)	3.39	(1.06, 10.81)	0.26	(0.03, 1.98)
High Vol/Heavy Drinker	2.04	(0.74, 5.65)	0.67	(0.20, 2.27)	1.93	(0.60, 6.20)
Prior Heavy Drinking (Past 10 Years)
Monthly Heavy Drinking	1.20	(0.75, 1.93)	0.73	(0.43, 1.24)	0.53	(0.31, 0.89)
Weekly Heavy Drinking	1.01	(0.51, 1.99)	1.48	(0.88, 2.51)	1.20	(0.58, 2.46)
Body Mass Index (Ref Normal BMI)
Underweight	3.39	(1.23, 9.34)	0.95	(0.12, 7.33)
Overweight	2.74	(1.55, 4.85)	1.79	(0.94, 3.40)	2.79	(1.32, 5.93)
Obese	9.76	(5.98, 15.94)	4.96	(2.83, 8.70)	8.38	(4.63, 15.19)
Male	White		Black		Hispanic
	OR	95% CI	OR	95% CI	OR	95% CI
Current Drinking Status (Ref Low Vol/Non-Heavy)
Lifetime Abstainer	1.33	(0.55, 3.19)	0.75	(0.27, 2.05)	1.37	(0.56, 3.37)
Current Non-Drinker	1.25	(0.81, 1.92)	1.22	(0.74, 2.02)	0.77	(0.43, 1.39)
Low Vol/Heavy Drinker	0.97	(0.52, 1.81)	1.57	(0.88, 2.80)	0.81	(0.42, 1.54)
Risky Vol/Monthly Heavy Drinker	0.79	(0.19, 3.27)	1.54	(0.42, 5.55)
Risky Vol/Weekly Heavy Drinker	1.42	(0.55, 3.66)	0.54	(0.15, 1.91)	1.06	(0.43, 2.59)
High Vol/Heavy Drinker	0.78	(0.20, 3.00)	1.37	(0.55, 3.44)	0.21	(0.04, 1.03)
Prior Heavy Drinking (Past 10 Years)
Monthly Heavy Drinking	0.76	(0.49, 1.17)	0.68	(0.42, 1.09)	0. 88	(0.51, 1.51)
Weekly Heavy Drinking	0.76	(0.45, 1.29)	0.97	(0.62, 1.50)	1.22	(0.74, 2.02)
Body Mass Index (Ref Normal BMI)
Underweight			3.04	(0.34, 27.08)	2.4
Overweight	2.16	(1.13, 4.14)	3.67	(1.76, 7.64)	7	(0.99, 6.15)
Obese	8.06	(4.42, 14.70)	10.75	(5.28, 21.88)	9.20	(3.87, 21.87)

Notes:

^aOR = Odds Ratio

^bIndicates time-varying variable

^cCurrent drinking status is a combined total volume and frequency of heavy episodic drinking of 6+ drinks. Total volume is categorized into low volume (≤7/14 drinks per week for women and men, respectively), risky volume (>7/14 drinks per week), and high volume (>14/28 drinks per week). Frequency of 6+ heavy episodic drinking is grouped into any in the last month, monthly, and week

Models including interactions terms between alcohol and obesity focused on the lifetime abstainer group to examine whether the increased risk for diabetes as compared to the low volume/non-heavy drinking group applied across the BMI categories. Results indicated differential risk relationships. Among those who are overweight (compared to normal weight), there was a significantly higher diabetes risk from lifetime abstention (compared to low volume non-heavy drinker group) for all women (OR 3.06; CI 1.67–5.60), white women (OR 3.66; CI 1.41–9.45) and white men (OR 4.17; CI 1.58–11.01). The models for all men and Hispanic women also showed elevated risk among overweight respondents only, albeit non-significant. No significant or substantially elevated risks were found for lifetime abstainers among obese respondents compared to those with normal weight.

DISCUSSION

A recent review of alcohol risk relationships for diabetes indicated that while many studies found benefits of moderate drinking, these benefits appear to be limited to women (Knott et al., 2015). Beneficial effects were also less evident in studies with lifetime abstainers as the control group. Heavy drinking occasions may have contributed to these mixed results since these were not measured in most studies (Rehm et al., 2017) and have been differentially implemented (Greenfield and Martinez, in press). These recent reviews highlight the importance of our diabetes risk analyses in which we capitalized on repeated measures of alcohol patterns taken over the lifecourse, utilized a clearly-defined lifetime abstainer group, categorized light-to-moderate drinking groups with measures of heavy drinking days, and accounted for a history of regular heavy drinking occasions. Further, we have controlled for early-life health limitations and repeated measures of overweight and obesity across the risk period, and considered interactions of alcohol use patterns with BMI categories. Results generally confirmed previous findings of increased diabetes risk associated with abstention from alcohol and heavy occasion drinking for women only. Results from stratified racial/ethnic groups suggest some potentially important differences in the beneficial effects of low volume drinking, given Black and Hispanic women did not appear to be at increased diabetes risk from lifetime abstention. The lack of consistent findings across sub-groups indicates some potential for confounding by unobserved risk factors and supports the need for high quality studies considering subgroup differences. Results from models interacting alcohol and BMI groups included strong findings of an elevated risk for diabetes onset among overweight women who are lifetime abstainers compared to overweight women who are moderate drinkers only. These models also indicated a similar relationship for overweight White men. Together with the few non-U.S. studies where these interactions have been considered (Beulens et al., 2012; Eckel et al., 2015; Metcalf et al., 2014), our results highlight the importance of considering how BMI and alcohol interact in future studies and for the interpretation of previous studies.

There are a number of limitations to this study, including a focus on a particular U.S. birth cohort in which their experiences are limited up to age 50. There are measurement issues affecting comparability across survey waves that have been addressed to the degree possible, but are likely to remain to some degree. The heavy occasion measure of 6 or more drinks in a day is high relative to the more typical 5+ drinks, or 4+ drinks for women, and does not fully capture patterns of heavy occasion drinking at these lower levels or at higher levels above 6 drinks that may also be relevant. The outcome measure was self-reported at age 40 or 50 and could have been mis-remembered to some degree. Even if correctly reported, the actual onset of diabetes may have occurred at an earlier age since some respondents may have had diabetes but were not aware of it.

CONCLUSIONS

The relationship between alcohol use patterns and diabetes is complex and there appear to be different relationships by sex and racial/ethnic groups. Regular light-to-moderate intake was associated with reduced risk among women, although this effect appears to be limited to overweight White women in the U.S. Overweight White men were also found to have reduced risk from low volume drinking. These findings limit the groups for whom moderate drinking would be expected to have beneficial impacts regarding diabetes onset. Heavy occasion drinking was associated with increased risk for women but results regarding the key elements of this relationship varied across racial/ethnic groups, with overall models indicating prior frequent heavy drinking as the key risk behavior, but within specific groups, current high volume or frequent heavy occasions were implicated. Future studies should utilize a lifecourse framework in evaluating diabetes onset risk and where possible consider drinking pattern details, sub-group differences and interactions with BMI groups.

Highlights.

• Regular light drinking reduces risk for diabetes onset only for overweight women.

• By race/ethnicity group, reduced risk was for overweight White women and men only.

• Heavy drinking in the prior 10 years increased risk for diabetes onset among women.