Moderate alcohol consumption, types of beverages and drinking pattern with cardiometabolic biomarkers in three cohorts of US men and women

Abstract

Underlying mechanisms of the inverse relationship between moderate alcohol consumption and cardiometabolic disorders are unclear. Modification by types of alcoholic beverages consumed and drinking pattern remains understudied. We aimed to provide insight into the mechanisms by examining 14 insulinemic/glycemic, inflammatory and lipid markers. We used cross-sectional data from 15,436 women in the Nurses’ Health Study, 19,318 women in the Nurses’ Health Study II, and 6872 men in the Health Professionals Follow-up Study. Multivariable linear regression was used to estimate the percentage differences in biomarker concentrations according to alcohol intakes. The average alcohol intake in the combined cohort was 3.3 servings/week. We found a 1 serving/d increment in alcohol intake (14g ethanol, 44 ml liquor or 355 ml beer or 118 ml wine per day) was associated with a 0.6% lower level of HbA1c, 1.7% to 3.6% lower proinflammatory markers and 4.2% higher adiponectin, as well as 7.1% higher HDL-cholesterol and 2.1% lower triglyceride with a significant linear trend. Wine, especially red wine, was associated with lower inflammation in particular. Beer had weaker favorable to null associations with blood lipids and adiponectin. Liquor was associated with higher C-peptide and interleukin-6, yet equally associated with lower HbA1c and higher HDL-cholesterol as other beverages. Drinking 3 days or more per week was related to a better biomarker profile than nonregular drinking independent of intake levels. Drinking appeared to have similar associations irrespective whether done with meals or not. Our data indicated moderate alcohol intake, especially if consumed from wine and done regularly, was associated with favorable profiles of insulinemic/glycemic and inflammatory markers and blood lipids.

Introduction

Excessive alcohol drinking causes many adverse outcomes and increases total mortality, while moderate drinking (1 serving/d or less for women and 2 servings/d or less for men [1]) is often associated with a lower risk of cardiovascular diseases (CVD) and type 2 diabetes [2]. Inverse associations from previous observational studies have been viewed cautiously due to concern about potential confounding by lifestyle and socioeconomic status (SES), and “sick quitters” effects resulting from the inclusion of former heavy drinkers who quit due to sickness as nondrinkers. Underlying mechanisms of the inverse relationship between moderate drinking and cardiometabolic diseases are unclear. Randomized controlled trials (RCTs) studying short-term alcohol consumption and biomarkers have provided some clues, although evidence is more consistent regarding alcohol improving HDL-cholesterol and glycemic control yet less consistent for how alcohol influences inflammatory markers [3, 4]. Besides RCTs’ limited ability to establish dose-response relationships, other important questions remain such as modification by types of alcoholic beverages consumed and pattern of drinking. Some studies have suggested the benefit of wine or beer consumption over liquor [5, 6] and of distributing consumption to more days with meals over binge drinking during weekends on an empty stomach [7], but the totality of evidence does not provide a clear conclusion [3, 8] and the comparison between red and white wine was understudied.

Utilizing data from three large prospective cohorts of US adults, the Nurses’ Health Study (NHS), Nurses’ Health Study II (NHSII) and Health Professionals Follow-up Study (HPFS), a few earlier analyses have been done for several inflammatory markers and in diabetic patients with small sample sizes (≤1500) [9, 10]. In the present study, we expanded associations with moderate alcohol intake to a more comprehensive set of biomarkers related to insulin resistance/glycemic control and inflammation as well as blood lipids in apparently healthy individuals with a much larger sample size (41,626). We also examined the associations separately by types of beverages and various aspects of drinking pattern (frequency, binge, with meals).

Method

Study population

The NHS began in 1976 and enrolled 121,700 female registered nurses aged 30–55 y at baseline. The NHSII began in 1989 and enrolled 116,429 younger female nurses aged 25–42 y [11]. The HPFS began in 1986 and enrolled 51,529 male health professionals aged 40–75 y [12]. In each cohort, mailed questionnaires were conducted biennially to collect information on lifestyle and medical factors, with a follow-up rate exceeding 90% for each 2-y cycle. Blood samples were provided by 32,826 women in NHS between 1989 and 1990, by 29,611 women in NHSII between 1996 and 1999, and by 18,159 men in HPFS between 1993 and 1995. Participants who provided blood samples were generally similar with those who did not regarding demographics and lifestyle characteristics [13]. For the current study, we included participants who provided a blood sample and had their biomarkers measured when involved in prior nested case-control studies. We excluded participants who had a history of diabetes (n=1795), CVD (n=1315) or cancer (n=1489) at blood draw. We further excluded those with missing data on alcohol intake (n=1060). These criteria led to a final study population of 41,626 individuals (15,436 from NHS, 19,318 from NHSII and 6872 from HPFS). The study protocol was approved by institutional review boards of the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health. All participants provided written consent.

Assessment of alcohol intake and drinking pattern

Alcohol intake was assessed via validated food-frequency questionnaires (FFQs) administered every 4 years [14]. In each FFQ, participants were asked the average frequency over the previous year of consuming one standard serving/drink of each alcoholic beverage (1.5 oz/44 ml for liquor, 12 oz/355 ml for beer and 4 oz/118 ml for white wine and red wine), with the options ranging from “never or less than once per month” to “6 or more times per day”. Total wine was computed as the sum of white wine and red wine. Total alcohol was computed as the sum of liquor, beer and total wine. The type of drinkers was defined based on the beverage they consumed with the highest frequency. In main analyses, we used the average intake from the last 2 FFQs administered before blood collection (1986 and 1990 for NHS, 1995 and 1999 for NHSII, and 1990 and 1994 for HPFS). We also considered the change in alcohol intake between these 2 FFQs in secondary analyses. On average, the midpoint of these 2FFQs preceded blood collection by 18, 6 and 22 months in NHS, NHSII and HPFS, respectively.

Data on drinking pattern were collected in 1988 among NHS and HPFS participants and in 1989 among NHSII participants, where they were asked the usual number of days of alcohol consumption in a week, and the maximum number of drinks per drinking occasion (with the options ranging from “0” to “≥15”). In line with previous literature [15], we defined regular drinking as 3 or more days of consumption in a week and defined binge drinking as 4 or more drinks on an occasion for women or 5 or more drinks on an occasion for men. In 1994, participants in HPFS were additionally asked the proportion of alcohol consumed with meals, with options ranging from “<25%”, to “>75%”.

Participants were also asked how their alcohol intake had changed in the past 10 years, with the options “greatly increased”, “greatly decreased” and “no change” in 1980 NHS and 1986 HPFS questionnaires. In NHSII, alcohol consumption at ages 15–17, 18–22, 23–30 and 31–40 was instead collected in the 1989 questionnaire. We considered current nondrinkers who reported to have greatly decreased alcohol consumption (in NHS and HPFS) or whose consumption exceeded 1 serving/d at any younger ages (in NHSII) as former drinkers.

The FFQ-measured alcohol intakes have been validated against multiple one-week diet records in these cohorts [16]. Correlation coefficients for liquor, beer and wine intake were 0.80, 0.81 and 0.83 respectively in women, and 0.86, 0.76 and 0.70 in men.

Biomarker assessment

Procedures for blood collection, handling and storage have been described elsewhere [17]. In the current study, we focused on 14 insulinemic/glycemic, inflammatory markers and blood lipids that have been shown to be related to the etiology of type 2 diabetes or CVD and were available in our cohorts: C-peptide, HbA1c, C-reactive protein (CRP), interleukin-6 (IL-6), tumor necrosis factor receptor 1 (TNF-R1), tumor necrosis factor receptor 2 (TNF-R2), intracellular adhesion molecule 1 (ICAM-1), adiponectin, leptin, total cholesterol (TC), HDL-cholesterol (HDL-C), LDL-cholesterol (LDL-C), triglyceride (TG) and TG/HDL-C ratio. All biomarkers were measured using standard methods [17]. Quality-control samples were randomly interspersed among participants’ samples. Intra-assay coefficients of variation were lower than 20% for each biomarker. Considering biomarkers were measured in multiple batches and there might be laboratory drift over time, all biomarker concentrations were recalibrated to an “average batch” within each cohort using the method described by Rosner et al. [18].

Covariate assessment

Overall diet quality was represented by the Alternative Healthy Eating Index (AHEI)-2010 [19] (excluding the alcohol component) and calculated with average intake from the 2 FFQs closest to blood draw. A higher score indicates better diet quality. Based on updated information in biennial questionnaires, physical activity was calculated by multiplying weekly hours spent on various forms of exercise and the metabolic equivalent of task (MET) score of each activity and then summing up to obtain a total weekly MET-hours. We used the cumulative average physical activity and BMI (from 1986 to 1990 questionnaires for NHS, 1995 to 1999 questionnaires for NHSII, and 1990 to 1994 questionnaires for HPFS) in main analyses to reduce measurement error. For other covariates, we used the questionnaire closest to the blood draw. The correlation coefficient between questionnaire-based and diary-based activity scores was 0.58 and that between self-reported and technician-measured weights was 0.97 [20, 21].

Neighborhood SES for participants was assessed based on the census block of their addresses reported in biennial questionnaires and linked to 1990 and 2000 census data. Wealth (median household income, median value of housing units, % of households receiving interest or rent), education (% of adults with a high school degree and % with a college degree) and occupation (% employed) of the census block were standardized and then combined into this neighborhood socioeconomic score, with higher scores suggesting higher levels of neighborhood SES [22].

Statistical analyses

We constructed multivariable generalized linear models to evaluate associations between alcohol intake and biomarker concentrations. All biomarkers were natural log-transformed to improve normality, and thus results were presented as percentage differences calculated using the equation: [exp (β-coefficient) −1] × 100%. Alcohol intake was categorized as nondrinkers, >0 to 1, >1 to 2 and >2 servings/d for evaluation of the dose-response relationship. Median values of intake were assigned to each category to create a continuous variable and Wald test of the quadratic term of this variable was conducted to assess linearity.

Because the participants in the three cohorts were similar in race/ethnicity, lifestyle, occupation and SES, and the characteristics’ patterns according to alcohol intakes were similar, data from all three cohorts were combined for main analyses. Model 1 was adjusted for age at blood draw, sex, case-control and fasting status and total energy intake. Model 2 was additionally adjusted for physical activity, smoking, AHEI-2010 excluding alcohol, multivitamin use, regular use of aspirin or nonsteroidal anti-inflammatory drugs (NSAIDs), history of hypertension and hypercholesterolemia, use of cholesterol-lowering drugs, BMI, waist circumference, and menopausal status and menopausal hormone therapy (women only). In secondary analyses of 4-y changes in alcohol intake with biomarker, average total energy intake, physical activity, AHEI-2010 score and BMI were replaced with their baseline values and 4-y changes, and baseline alcohol intake was included in the model.

In sensitivity analyses, we restricted the analysis to participants who provided fasting blood, who were selected as controls in previous case-control studies and who reported no history of hypertension or hypercholesterolemia, and excluded former drinkers or all nondrinkers to test the robustness of results. We also checked the results after additionally adjusting for neighborhood SES.

We also examined associations between alcohol intake and biomarker concentrations stratified by cohort, sex, age at blood draw (<60/≥60 years), smoking (never/ever), physical activity (below/above median) and BMI (<25.0/25.0–29.9/≥30.0 kg/m²). Effect modification was assessed by likelihood ratio tests comparing the model with and without the product term between 1 serving/d of alcohol intake and each of the stratified variables above.

To explore potential heterogeneous associations of each alcoholic beverage with biomarkers, separate analyses for consuming 1 serving/d of liquor, beer, total wine, white wine and red wine were conducted. The multivariable models were adjusted for the same set of covariates as in Model 2 and were mutually adjusted for each other to account for consumption of multiple alcoholic beverages. We also estimated percentage differences in biomarkers of substituting 1 serving/d of each beverage with the same quantity of other alcoholic beverages using the partition model approach [23].

Finally, we examined associations between biomarkers and drinking pattern which was jointly defined by regular drinking and binge drinking, taking nondrinkers as the reference group. Associations with drinking with meals were examined in HPFS, taking <25% as the reference.We considered two-sided P<0.005 to be statistically significant and 0.005≤P<0.05 to be suggestively significant. SAS version 9.4 (SAS institute) was used for all statistical analyses.

Data and Resource Availability

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 Nurses’ Health Study and the Health Professionals Follow-Up Study is described at https://www.nurseshealthstudy.org/researchers (contact e-mail: nhsaccess@channing.harvard.edu) and https://sites.sph.harvard.edu/hpfs/for-collaborators/.

Results

A total of 34,754 women and 6872 men were included in our study. 71% of our population drank alcohol. The average alcohol intake was 3.3 servings/week (4.8 servings/week in drinkers) (Supplementary Figure 1). HPFS had the highest intake of 6.6 servings/week mainly contributed by liquor and beer drinking while NHSII had the lowest intake of 2.3 servings/week mainly contributed by white wine and beer. Within each sex, compared with nondrinkers and those who drank less than 1 serving/d, participants who drank more tended to be older, current smokers with greater pack-years, regular aspirin and NSAIDs users and have higher energy intake, physical activity, SES and lower BMI (Table 1). Biomarker concentrations according to alcohol intake categories are presented in Supplementary Table 1.

Table 1.

Characteristics of participants at blood draw according to total alcohol intake in the combined cohort^a

	Women (n=34,754)				Men (n=6872)
	Non drinkers	>0 to 1 serving/d	>1 to 2 servings/d	>2 servings/d	Non drinkers	>0 to 1 serving/d	>1 to 2 servings/d	>2 servings/d
	n=10,820	n=19,960	n=2674	n=1300	n=1248	n=3290	n=1317	n=1017
Age at blood draw, y	49.6 ± 8.9	49.7 ± 8.6	51.7 ± 8.8	53.5 ± 8.6	62.0 ± 8.6	62.0 ± 8.6	62.4 ± 8.1	62.5 ± 8.2
Total alcohol, grams/d	0 ± 0	3.8 ± 3.3	16.8 ± 3.7	35.8 ± 10.8	0 ± 0	5.3 ± 3.7	18.1 ± 3.9	40.1 ± 13.5
Total alcohol, servings/d	0 ± 0	0.3 ± 0.3	1.4 ± 0.3	2.9 ± 0.9	0 ± 0	0.4 ± 0.3	1.4 ± 0.3	3.2 ± 1.0
Liquor, servings/week	0 ± 0	0.5 ± 1.0	2.5 ± 3.2	6.0 ± 7.5	0 ± 0	0.8 ± 1.3	3.6 ± 3.3	9.5 ± 8.4
Beer, servings/week	0 ± 0	0.5 ± 1.0	2.1 ± 2.8	5.0 ± 7.7	0 ± 0	0.9 ± 1.2	3.1 ± 2.8	6.7 ± 7.4
Total wine, servings/week	0 ± 0	1.2 ± 1.4	5.1 ± 3.6	9.5 ± 8.4	0 ± 0	1.2 ± 1.4	3.3 ± 2.7	5.9 ± 6.9
White wine, servings/week	0 ± 0	0.8 ± 1.1	3.3 ± 3.2	6.5 ± 7.0	0 ± 0	0.7 ± 0.9	1.8 ± 1.9	3.2 ± 4.6
Red wine, servings/week	0 ± 0	0.4 ± 0.7	1.8 ± 2.4	3.1 ± 5.2	0 ± 0	0.5 ± 0.8	1.5 ± 1.7	2.7 ± 4.3
Days drinking in a week	/	2.0 ± 1.5	4.3 ± 2.0	5.5 ± 1.7	/	2.3 ± 1.6	4.6 ± 1.8	6.0 ± 1.3
Maximum drinks per occasion	/	2.5 ± 1.7	3.6 ± 2.1	4.7 ± 2.4	/	2.3 ± 1.7	3.2 ± 2.1	4.9 ± 2.6
Current smokers, %	7.6	10.4	14.5	22.7	4.6	5.2	5.2	12.0
Pack-years among ever-smokers	5.9 ± 13.5	7.8 ± 13.7	11.5 ± 15.9	17.3 ± 20.2	8.6 ± 16.2	10.5 ± 16.3	13.1 ± 16.7	19.3 ± 19.8
AHEI score excluding alcohol	46.1 ± 10.3	47.6 ± 9.6	48.3 ± 9.4	46.8 ± 8.8	48.2 ± 10.6	49.2 ± 10.2	49.4 ± 9.7	47.2 ± 9.1
Total energy intake, kcal/d	1775 ± 509	1803 ± 488	1887 ± 484	1981 ± 465	1959 ± 570	1947 ± 539	2043 ± 520	2191 ± 524
Physical activity, MET-h/week	13.9 ±15.9	16.8 ± 17.5	19.2 ± 18.9	17.4 ± 17.6	28.5 ± 22.3	30.8 ± 22.7	33.0 ± 22.7	31.9 ± 21.7
BMI, kg/m2	26.7 ± 5.9	25.5 ± 4.9	24.1 ± 3.8	24.5 ± 3.9	25.6 ± 3.4	25.6 ± 3.1	25.3 ± 2.7	25.7 ± 2.7
Waist circumference, cm	79.8 ± 10.2	78.2 ± 9.0	76.8 ± 8.1	78.1 ± 8.1	94.3 ± 8.1	94.6 ± 8.2	94.0 ± 7.0	95.4 ± 7.8
Hypertension, %	20.0	16.4	14.5	21.0	23.1	23.5	23.6	30.0
Hypercholesterolemia, %	31.2	29.4	25.1	28.4	36.5	37.3	39.0	42.1
Cholesterol-lowering drug use, %	3.6	3.0	2.7	2.9	6.0	7.2	8.3	7.9
Regular aspirin or NSAID use,b %	47.8	51.7	55.2	55.7	51.1	56.9	62.7	64.6
Multivitamin use, %	45.2	46.6	45.0	46.4	39.6	42.8	39.8	41.6
Neighborhood socioeconomic score	−0.9 ± 4.6	0.4 ± 4.8	1.5 ± 5.1	1.0 ± 5.2	−0.7 ± 4.4	0.2 ± 4.7	0.3 ± 4.7	0.0 ± 4.5
Postmenopausal, %	44.3	43.0	42.6	46.4	/	/	/	/
Current menopausal hormone use,c %	21.5	21.6	21.5	22.0	/	/	/	/

^aValues are means ± SDs for continuous variables and percentages for categorical variables. All variables were standardized by age at blood draw. MET, metabolic equivalent of task; AHEI, Alternative Healthy Eating Index; NSAID, nonsteroidal anti-inflammatory drug. /, non-applicable.

^bRegular users were defined as ≥2 tablets of aspirin (325 mg/tablet) or NSAIDs per week.

^cDefined in menopausal women only.

The dose-response relationships between alcohol intake and biomarkers were mostly monotonic (Supplementary Figure 2). P-values suggested overall absence of evidence for non-linearity or modest non-linearity and justified presenting results for 1 serving/d of intake as in Table 2. In the combined cohort, a 1 serving/d increment in alcohol intake was associated with favorable profiles of several biomarkers: 0.6% lower level of HbA1c, 1.7% to 3.6% lower proinflammatory markers and 4.2% higher adiponectin, as well as 1.4% higher TC (likely driven by 7.1% higher HDL-C) and 2.1% lower TG with a significant linear trend. Most of these associations were also observed for increasing alcohol intake for 1 serving/d during the 4 years prior to blood draw, although their magnitudes were smaller. Considering the possibility that some may had reduced alcohol consumption due to disease occurrence or preclinical phases, for the latter analyses, we further excluded those who developed hypertension during the 4-y period or who became cases in prior case-control studies within 4 years after blood draw; yet results remained essentially unchanged (data not shown).

Table 2.

Percentage of differences (95%CIs) in biomarker concentrations associated with 1 serving/d increase in average alcohol intake and in 4-year change of alcohol intake

	Alcohol intake	Change in alcohol intake
Insulinemic/glycemic markers
C-peptide
n	12,450	11,035
Model 1	−1.3 (−2.3, −0.3) *	−0.1 (−1.6, 1.3)
Model 2	−0.7 (−1.6, 0.3)	0.2 (−1.2, 1.6)
HbA1c
n	9958	8642
Model 1	−0.7 (−0.8, −0.5) ***	−0.2 (−0.5, 0.0)
Model 2	−0.6 (−0.8, −0.4) ***	−0.2 (−0.4, 0.0)
Inflammatory markers
CRP
n	16,403	14,350
Model 1	−3.9 (−5.8, −2.1) ***	−1.8 (−4.6, 1.1)
Model 2	−1.7 (−3.4, 0.1)	−1.0 (−3.6, 1.7)
IL-6
n	11,040	9678
Model 1	−2.0 (−3.4, −0.7) **	−2.0 (−4.0, 0.0)
Model 2	−0.6 (−1.9, 0.7)	−0.9 (−2.9, 1.1)
TNF-R1
n	3323	2892
Model 1	−4.2 (−5.2, −3.1) ***	−2.6 (−4.1, −1.0) **
Model 2	−3.5 (−4.5, −2.4) ***	−1.9 (−3.4, −0.2) *
TNF-R2
n	10,739	9364
Model 1	−3.9 (−4.4, −3.3) ***	−1.4 (−2.3, −0.6) ***
Model 2	−3.6 (−4.1, −3.0) ***	−1.2 (−2.0, −0.3) *
ICAM-1
n	6276	5501
Model 1	−0.8 (−1.4, −0.1) *	−0.3 (−1.3, 0.6)
Model 2	−1.7 (−2.3, −1.1) ***	−0.8 (−1.7, 0.2)
Adiponectin
n	13,943	12,230
Model 1	5.6 (4.6, 6.6) ***	2.5 (1.1, 3.9) ***
Model 2	4.2 (3.2, 5.2) ***	1.8 (0.4, 3.2) *
Leptin
n	9870	8806
Model 1	−4.6 (−6.2, −2.9) ***	−2.4 (−4.7, 0.0) *
Model 2	−0.9 (−2.3, 0.4)	0.5 (−1.4, 2.5)
Lipid markers
TC
n	15,009	13,103
Model 1	1.4 (1.0, 1.7) ***	0.4 (−0.1, 0.9)
Model 2	1.4 (1.1, 1.8) ***	0.8 (0.3, 1.3) **
HDL-C
n	8655	7563
Model 1	7.7 (7.0, 8.4) ***	4.5 (3.5, 5.5) ***
Model 2	7.1 (6.4, 7.8) ***	4.4 (3.4, 5.4) ***
LDL-C
n	7199	6339
Model 1	−0.5 (−1.2, 0.2)	−0.9 (−2.0, 0.1)
Model 2	−0.4 (−1.1, 0.2)	−0.3 (−1.3, 0.7)
TG
n	10,253	8980
Model 1	−3.5 (−4.7, −2.3) ***	−3.0 (−4.7, −1.2) **
Model 2	−2.1 (−3.2, −1.0) ***	−1.6 (−3.2, 0.1)
TG/HDL-C ratio
n	6115	5391
Model 1	−8.3 (−10.0, −6.6) ***	−6.0 (−8.5, −3.3) ***
Model 2	−7.5 (−9.1, −5.9) ***	−5.4 (−7.9, −2.9) ***

Model 1 was adjusted for age at blood draw (continuous), sex (male/female), case-control status (case/control), fasting status (fasting/not fasting) and total energy intake (quartiles). Model 2 was additionally adjusted for physical activity (quartiles), smoking status and intensity (never smoker/former smoker, pack-years<30/former smoker, pack-years≥30/current smoker, pack-years<30/current smoker, pack-years≥30), AHEI score excluding alcohol (quartiles), multivitamin use (yes/no), regular use of aspirin or nonsteroidal anti-inflammatory drug use (yes/no), hypertension (yes/no), hypercholesterolemia (yes/no), use of cholesterol-lowering drugs (yes/no), BMI (<23.0/23.0–24.9/25.0–27.4/27.5–29.9/≥30.0 kg/m²) and waist circumference (continuous), and for women, menopausal status (premenopausal/postmenopausal/unknown) and menopausal hormone therapy (never/past/current use).

In change analyses, the average total energy intake, physical activity, AHEI-2010 score and BMI were replaced with their baseline values and 4-y changes (all continuous), and the baseline alcohol intake (continuous) was further adjusted.

CRP, C-reactive protein; IL-6, interleukin-6; TNF-R1, tumor necrosis factor receptor 1; TNF-R2, tumor necrosis factor receptor 2; ICAM-1, intracellular adhesion molecule 1; TC, total cholesterol; HDL-C, HDL-cholesterol; LDL-C, LDL-cholesterol; TG, triglyceride. Ref, reference.

^*0.005≤P<0.05,

^**0.001≤P<0.005,

^***P<0.001.

Findings from main analyses generally remained the same when restricting the analysis to participants who provided fasting blood, who were selected as controls in previous case-control studies or who reported no history of hypertension or hypercholesterolemia (Supplementary Table 2). The proportion of former drinkers was 2.4%, 2.5% and 8.8% in NHS, NHSII and HPFS, respectively. Excluding these former drinkers or excluding all nondrinkers did not change the results substantially (Supplementary Table 2). The adjustment of neighborhood SES did not change the results either.

We observed the favorable associations between alcohol and biomarkers to be persistent in never smokers and slightly more pronounced in women and older individuals (Supplementary Figure 3 and 4). Although C-peptide and CRP were not significantly associated with alcohol intake in main analyses, we observed a lower C-peptide in women, never smokers and participants with obesity, and a lower CRP in older participants and never smokers.

There were slightly positive correlations between consumption of different types of beverages. The correlation coefficients were up to 0.1, except that between white wine and red wine around 0.2 to 0.3. In our population, wine drinkers had a healthier lifestyle and a higher SES than liquor and beer drinkers (Supplementary Table 3). After adjusting for these potential confounders and mutually adjusting for types of alcoholic beverages, wine, especially red wine, was more profoundly associated with lower concentrations of proinflammatory markers, while beer consumption had weaker favorable to null associations with blood lipids and adiponectin. Liquor was associated with higher C-peptide and interleukin-6, yet equally associated with lower HbA1c and higher HDL-C as other beverages (Figure 1). This pattern persisted when stratified by sex, age, smoking or alcohol intake level, and became even clearer in the substitution analyses (Supplementary Table 4).

Figure 1.

Percentage of differences (95%CIs) in biomarker concentrations associated with 1 serving/d of each type of alcoholic beverage. Estimates were adjusted for the same set of covariates as Model 2 in Table 2, with SES additionally adjusted.

Alcohol consumption levels differed across drinking pattern groups: the median was 8.8, 7.3, 2.2 and 1.0 servings/week for binge and regular, nonbinge and regular, binge and nonregular, and nonbinge and nonregular drinking groups, respectively. After further adjusting for the quantity of alcohol consumed, we found greater inverse associations for the two regular drinking groups than the two nonregular drinking groups (Table 3). Among regular drinkers, the biomarker concentrations were largely similar for binge and nonbinge drinkers, except for a lower TG and TG/HDL-C ratio were observed only in nonbinge drinkers. In HPFS, the proportion of alcohol consumed with meals did not show significant associations with biomarker concentrations (Supplementary Table 5).

Table 3.

Percentage of differences (95%CIs) in biomarker concentrations associated with drinking pattern

	Non drinkers (n=12,068)	Binge and nonregular drinking (n=4821)	Nonbinge and nonregular drinking (n=15,479)	Binge and regular drinking (n=4350)	Nonbinge and regular drinking (n=4908)
Insulinemic/glycemic markers
C-peptide
n	3086	983	4645	1469	2435
Model 1	0 (ref)	−3.1 (−6.7, 0.7)	−4.8 (−7.1, −2.5)	−7.5 (−11.2, −3.6)	−10.4 (−13.3, −7.4)
Model 2	0 (ref)	−2.7 (−6.1, 0.8)	−1.0 (−3.2, 1.2)	−3.0 (−6.7, 0.8)	−2.9 (−5.9, 0.1)
HbA1c
n	2838	977	3645	1187	1426
Model 1	0 (ref)	−0.4 (−0.9, 0.1)	−0.3 (−0.6, 0.1)	−0.8 (−1.3, −0.2)	−0.8 (−1.3, −0.3)
Model 2	0 (ref)	−0.4 (−0.8, 0.1)	−0.1 (−0.4, 0.3)	−0.4 (−1.0, 0.1)	−0.3 (−0.8, 0.2)
Inflammatory markers
CRP
n	4426	1498	6191	1934	2560
Model 1	0 (ref)	−1.6 (−7.7, 4.9)	−9.2 (−12.9, −5.4)	−18.0 (−23.8, −11.8)	−22.4 (−27.1, −17.5)
Model 2	0 (ref)	−0.8 (−6.4, 5.1)	−2.4 (−6.0, 1.3)	−6.7 (−12.7, −0.3)	−7.0 (−12.1, −1.5)
IL-6
n	3038	1128	4067	1320	1612
Model 1	0 (ref)	−4.6 (−8.6, −0.5)	−7.6 (−10.2, −4.9)	−14.3 (−18.5, −9.9)	−14.6 (−18.3, −10.8)
Model 2	0 (ref)	−3.5 (−7.4, 0.5)	−3.8 (−6.4, −1.1)	−8.3 (−12.7, −3.8)	−6.9 (−10.7, −2.9)
TNF-R1
n	924	385	1207	411	433
Model 1	0 (ref)	−3.0 (−6.0, 0.0)	−3.4 (−5.5, −1.3)	−6.6 (−10.1, −3.0)	−5.2 (−8.3, −2.1)
Model 2	0 (ref)	−3.1 (−6.0, −0.2)	−2.2 (−4.2, −0.1)	−5.1 (−8.6, −1.6)	−2.4 (−5.5, 0.8)
TNF-R2
n	2973	1074	4008	1278	1536
Model 1	0 (ref)	−3.0 (−4.7, −1.3)	−2.9 (−4.0, −1.7)	−8.0 (−9.9, −6.1)	−5.7 (−7.4, −4.0)
Model 2	0 (ref)	−3.1 (−4.8, −1.3)	−1.9 (−3.0, −0.7)	−6.7 (−8.6, −4.7)	−3.7 (−5.5, −2.0)
ICAM-1
n	1744	440	2453	571	1148
Model 1	0 (ref)	−0.1 (−2.5, 2.4)	−1.2 (−2.6, 0.3)	−2.1 (−4.8, 0.6)	−4.0 (−6.0, −1.9)
Model 2	0 (ref)	−1.6 (−3.9, 0.7)	−0.8 (−2.1, 0.6)	−2.1 (−4.6, 0.5)	−1.9 (−3.9, 0.2)
Adiponectin
n	3887	1359	5336	1605	1921
Model 1	0 (ref)	4.7 (1.8, 7.8)	5.1 (3.1, 7.1)	12.0 (8.1, 15.9)	11.4 (8.1, 14.7)
Model 2	0 (ref)	4.0 (1.2, 7.0)	2.3 (0.4, 4.1)	7.3 (3.7, 10.9)	4.9 (1.9, 8.0)
Leptin
n	2686	983	3813	935	1568
Model 1	0 (ref)	−4.1 (−9.2, 1.4)	−7.1 (−10.5, −3.7)	−17.9 (−23.3, −12.2)	−16.5 (−21.1, −11.7)
Model 2	0 (ref)	−0.8 (−5.0, 3.6)	3.1 (0.2, 6.1)	−3.2 (−8.2, 2.1)	1.8 (−2.6, 6.5)
Lipid markers
TC
n	4313	1620	5661	1640	1931
Model 1	0 (ref)	0.3 (−0.8, 1.3)	0.9 (0.2, 1.6)	1.7 (0.4, 3.0)	0.7 (−0.4, 1.9)
Model 2	0 (ref)	0.4 (−0.6, 1.3)	1.2 (0.5, 1.8)	2.1 (0.9, 3.3)	1.5 (0.4, 2.6)
HDL-C
n	2355	766	3151	1103	1388
Model 1	0 (ref)	2.4 (0.1, 4.7)	3.5 (2.0, 5.0)	10.5 (7.8, 13.3)	10.1 (7.8, 12.4)
Model 2	0 (ref)	2.3 (0.2, 4.5)	1.6 (0.3, 3.0)	7.6 (5.1, 10.2)	5.4 (3.4, 7.6)
LDL-C
n	1930	570	2668	819	1304
Model 1	0 (ref)	−2.1 (−4.7, 0.5)	0.7 (−1.0, 2.3)	−0.5 (−3.3, 2.4)	−0.7 (−3.0, 1.7)
Model 2	0 (ref)	−1.8 (−4.1, 0.7)	1.4 (−0.1, 2.9)	1.2 (−1.5, 4.0)	1.6 (−0.6, 3.8)
TG
n	3027	1133	3818	1132	1240
Model 1	0 (ref)	−3.6 (−6.9, −0.1)	−6.4 (−8.7, −4.2)	−6.2 (−10.3, −2.0)	−14.0 (−17.3, −10.6)
Model 2	0 (ref)	−2.0 (−5.1, 1.2)	−2.1 (−4.3, 0.0)	−0.4 (−4.4, 3.7)	−5.4 (−8.7, −1.9)
TG/HDL-C ratio
n	1644	529	2255	743	1019
Model 1	0 (ref)	−4.2 (−10.3, 2.3)	−8.9 (−12.6, −5.0)	−12.2 (−18.4, −5.5)	−21.7 (−26.3, −16.8)
Model 2	0 (ref)	−3.3 (−8.9, 2.7)	−3.4 (−7.0, 0.4)	−3.6 (−9.9, 3.1)	−10.0 (−14.9, −4.8)

Adjusted for the same set of covariates as in Table 2 main analysis. Alcohol intake level was additionally adjusted (continuous). CRP, C-reactive protein; IL-6, interleukin-6; TNF-R1, tumor necrosis factor receptor 1; TNF-R2, tumor necrosis factor receptor 2; ICAM-1, intracellular adhesion molecule 1; TC, total cholesterol; HDL-C, HDL-cholesterol; LDL-C, LDL-cholesterol; TG, triglyceride. Ref, reference.

Discussion

In three large US cohorts of individuals without major chronic diseases, we observed favorable profiles of insulinemic/glycemic, inflammatory markers and blood lipids for light to moderate alcohol consumption, starting from <1 serving/d, and extending to higher intake levels in a generally linear manner. The associations varied by types of alcoholic beverages and pattern of drinking. Wine, especially red wine consumption, had the strongest association with lower inflammation. Beer showed less to null benefit in blood lipids and adiponectin. Liquor might instead be related to insulin resistance (higher C-peptide) and higher inflammation (measured using IL-6), yet it was equally associated with lower HbA1c and higher HDL-C as other beverages. Regular drinking was associated with a better biomarker profile than nonregular drinking independent of the quantity consumed.

Different upper limits for moderate drinking across countries reflect uncertainty in the dose-response relationship between alcohol and health (e.g.,14 servings/week or 198g/week for men in the US, 100g/week in China, 210 and 280g/week in France and Poland) [24]. The intake in our population was modest, with the mean intake of 3 servings/d in the highest category and only 4% women and 15% men drank more than 2 servings/d. Thus, our data supporting favorable associations were restricted to this level, and could not provide information for higher intakes. Within this range, we found the relationship between alcohol and most biomarkers to be monotonic and statistically linear, although the non-linear associations with CRP and IL-6, complex interactions between biomarkers, and the opposite effects of biomarkers and acetaldehyde could possibly underlie non-linear associations between alcohol and clinical outcomes. Assuming linearity, the magnitude of our associations was larger than in other studies. For example, a study of 726 diabetic patients in HPFS found a 1 serving/d increment in alcohol intake was associated with a 2.05 mg/dL higher HDL-C [10], yet with a 3.83 mg/dL (7.1% * 54 mg/dL in nondrinkers) higher HDL-C in our population of over 41,000 healthy men and women. Two meta-analyses of RCTs in healthy individuals estimated 1 serving/d of alcohol increased HDL-C by 2.00 and 1.83 mg/dL among mostly young male participants [3, 25]. Besides residual confounding, the larger magnitude in our study may stem from the predominantly older female participants as opposed to young males. This aligns with our finding of stronger associations in women and older participants, which might be due to ethanol metabolism or beverage preferences [26, 27]. It is also likely that habitual alcohol intake might confer accumulated effects on biomarkers not fully captured by short-term interventions in those RCTs (mostly <8 weeks). HDL-C has been recognized as one of the most important biomarkers mediating the relationship between alcohol intake and CVD mortality [28]. The higher HDL-C associated with a 1 serving/d of alcohol consumption is greater than currently available pharmacological studies (e.g., one FDA-approved therapy for low HDL-C patients, fibrates, increased HDL-C by 2.6 mg/dL [29]), but we do not encourage intentionally increasing HDL-C through heavy consumption of alcohol, as higher risks of several cancer and injuries should also be taken into consideration.

Ethanol has been shown to increase the transport rate of lipoproteins and lipoprotein lipase activity to influence levels of HDL-C and TG [30]. It reduces HbA1c by suppressing the acute rise in blood glucose after meals and increasing early insulin response [31]. Also, adiponectin is an insulin sensitizer. Ethanol increases expression of ADIPOQ gene in adipose tissue [32] and suppresses the synthesis of proinflammatory cytokines and chemokines [33]. The associations with inflammation might be attributed to compounds other than ethanol in alcoholic beverages as well, such as polyphenols, which have anti-inflammatory and anti-oxidative potential [34]. Polyphenols are richer in red wine (101mg/100mL) than white wine (10mg/100mL) and beer (4mg/100mL) while almost undetectable in liquor [35], possibly contributing to the gradient observed across the alcoholic beverages with inflammatory markers.

Contrary to the more established relationship between alcohol consumption and HDL-C concentrations and glycemic control, more inconsistency exists in the relatively understudied area of alcohol and inflammation, and we hypothesize it is at least partially due to the influence of the type of alcoholic beverages consumed. For example, RCTs showed red wine but not gin (30g daily for 4 weeks) reduced CRP and ICAM-1 in healthy individuals [6], and IL-6 in individuals at high risk of CVD [36]. Similarly, white wine but not gin (30g daily for 4 weeks) reduced CRP, IL-6 and ICAM-1 [37]. Besides, RCTs comparing regular beer/red wine with their dealcoholized versions and water indicated both ethanol and non-ethanol contents contributed to anti-inflammation, likely in a synergistic way [36, 38]. In real-world settings, the type of alcoholic beverages mainly consumed is often impacted by SES and correlates with overall lifestyle. Wine drinkers in our population had a higher SES and healthier lifestyle than liquor and beer drinkers, consistent with an earlier report in the US [39]. However, variation by beverages remained when SES and lifestyle were adjusted, and our cohorts being quite homogeneous in these factors likely minimized the potential for residual confounding. Although drinking pattern could also be different across beverages, the percentage of regular drinkers among wine drinkers was actually lower than in drinkers of other alcoholic beverages (26.4 vs 37.9 and 30.6 for liquor and beer), which should have biased our results toward smaller differences between beverages.Drinking more frequently or distributing the total intake into more days and drinking with meals have been hypothesized to be healthier patterns of alcohol consumption, due to slower absorption and lower peak concentration of alcohol [40]. A prospective cohort study among UK Biobank participants found a higher drinking habit score, jointly defined by frequency ≥3 days/week and drinking with meals, was associated with lower all-cause mortality, CVD mortality and cancer mortality, during 9 years of follow-up [7]. Previous analyses in HPFS found, even stratified by total alcohol intake, drinking ≥3 days/week was associated with a lower risk of diabetes and myocardial infarction during 12 years of follow-up [41, 42]. However, neither HPFS study reported an apparent benefit of drinking with meals, in line with the current study.

The strengths of our study include the measurement of a comprehensive set of biomarkers relevant to multiple mechanisms in cardiometabolic disorders, detailed information collected on types of alcoholic beverages and various aspects of drinking pattern, and a large sample size comprising both men and women empowering these separate analyses. A series of sensitivity analyses provided evidence against alternative explanations such as “sick quitters” effects and confounding by SES, and supported the robustness of our findings. Cross-sectional designs usually have a limited ability to infer causality. However, the repeated assessment of alcohol and covariates gave us a unique opportunity to conduct the change analysis, and its consistency with the main analysis led to more confidence in true causality. Besides, alcohol affects blood biomarkers rather quickly, reducing the possibility of reverse causation, and also making cross-sectional analysis probably more relevant than follow-up for decades. Along with careful adjustment of confounders in models, our cohorts being homogeneous in race, SES and lifestyle mitigated the concern about confounding. However, the generalizability of our study was also limited to populations with similar characteristics and cannot be extrapolated to heavy drinking populations. It should be noted that there are intricate mediations/interactions among the assessed biomarkers and all of these biomarkers collectively contribute to cardiovascular risks. Therefore, when interpreting our findings, the specific estimate for each individual biomarker might be less important than the overall biomarker profile.

In conclusion, although heavy alcohol intake has been associated with several cancers and total mortality, moderate alcohol intake was associated with favorable profiles of insulinemic/glycemic and inflammatory markers and blood lipids. Drinking wine and drinking more than 3 days per week showed stronger associations. Future studies on modification by beverage type and drinking pattern are warranted to confirm our findings, and studies among younger, more diverse populations and heavy drinking populations are encouraged to extend our findings.