Midlife Alcohol Consumption and the Risk of Stroke in the Atherosclerosis Risk in Communities Study

Introduction

Stroke is a leading cause of mortality and disability worldwide, a major contributor to US healthcare costs, and is projected to increase in burden as the population ages.^1–3 As such, continued examination of modifiable risk factors and behaviors that may prevent disease occurrence is needed. One such factor may be alcohol, a prevalent exposure both in the United States and worldwide. Seventy percent of US adults report current drinking and more than one quarter report excessive drinking.^4,5

The relationship between alcohol intake and stroke has been widely studied, yet uncertainties remain; results from observational studies are inconsistent and randomized trials are infeasible. Light-to-moderate alcohol consumption (usually defined as 1–2 drinks per day in the United States) may reduce the risk of stroke, but some studies, particularly older ones, have not found beneficial effects.^6–8 Current meta-analyses suggest that moderate intake is protective for ischemic stroke (IS), but not intracerebral hemorrhagic (ICH), with possible differential dose–responses by sex.^6–8 Limitations of our understanding stem from (1) the assessment of alcohol intake late in life, a period that may not reflect the most critical exposure window for disease risk and that may be influenced by other medical conditions developing in later life; (2) alcohol measurement error and misspecification because of variations in drinking patterns; and (3) limited generalizability.^6–15 Furthermore, some studies lack adjustment for lifestyle and socioeconomic factors, which may account for protective effects in light drinkers, or combine former and never drinkers in a single referent group. In contrast to the evidence on light-to-moderate intake, consistent and convincing evidence supports the harmful effects of heavy consumption on stroke risk. The precise dose–response relationship, however, is unclear.

Despite the large body of work on alcohol and stroke, few studies have included substantial numbers of minority individuals. Blacks have higher stroke incidence and different drinking patterns from whites and, therefore, warrant investigation.³ In addition, few studies have accounted for the competing risk of mortality, which may be substantial in prospective studies. Subdistribution hazard estimation is particularly useful to public health scientists interested in assessing risks and benefits of alcohol in a population experiencing competing risks.¹⁶ In our study, we estimated the dose–response relationship between usual, midlife alcohol consumption and incident stroke among black and white adults in the Atherosclerosis Risk in Communities (ARIC) study, a population-based cohort drawn from 4 US communities.

Methods

Study Population

The ARIC study is a population-based cohort recruited using probability sampling of adults aged 45 to 64 years from 4 US communities: Forsyth County, NC; Jackson, MS; suburbs of Minneapolis, MN; and Washington County, MD. The ARIC study design and rationale are described in detail elsewhere.¹⁷ A total of 15 792 participants were enrolled at visit 1 (1987–1989) and underwent an in-home interview and physical examination, with 4 additional study visits during the subsequent 25 years. For analysis, we excluded race-ethnicities other than white or black (n=48), blacks from Minnesota or Washington County (n=55), participants missing alcohol intake (n=106), and those with prevalent stroke (n=284). After these exclusions, the analytic cohort totaled 15 305. We further limited our population to current or never drinkers (n=12 433) because of the heterogeneity in exposure among former drinkers in our population with regard to duration and quantity of consumption and time since cessation.¹²

Alcohol Assessment

Alcohol consumption was measured at baseline using an interviewer-administered questionnaire developed in accordance with the validated Willett 66-item food frequency questionnaire.¹⁸ Participants were asked to report whether they currently consumed alcoholic beverages and, if so, their usual intake of drinks per week; those reporting values under 0.5 were recorded as 0. Separate intake frequency questions were asked for standard drinks of wine (4 oz), beer (12 oz), and hard liquor (1.5 oz). Total drinks per week was calculated as the sum of standard drinks of each type.

Stroke Definition

Suspected stroke hospitalizations were ascertained by self-report, at study visits, during annual follow-up, and surveillance of local hospital discharge lists. Validation of suspected events and stroke diagnosis were conducted independently by a computer-derived algorithm and a physician reviewer using data abstracted from the medical record; differences were adjudicated by a second physician reviewer. Strokes were classified using criteria adopted from the National Survey of Stroke¹⁹ and required, at a minimum, evidence of sudden or rapid onset of neurological symptoms lasting >24 hours or leading to death in the absence of evidence for a nonstroke cause. Out-of-hospital fatal strokes (n=19) were not validated and were not included. Additional details of stroke subtype classification have been published.²⁰

Covariates

Confounders were identified based on substantive knowledge and directed acyclic graph analysis was used to obtain a minimally sufficient set of adjustment variables: age, sex, race, study center, baseline comorbid conditions, diet score, low-density lipoprotein cholesterol, education, smoking status, and marital status.²¹ Causal intermediates included blood pressure, high-density lipoprotein cholesterol (HDL-C), and atrial fibrillation.

Dietary factors were assessed using an interviewer-administered 66-item food frequency questionnaire measuring usual intake of foods during the past year. We calculated a diet score as described elsewhere²² based on quintile values for 6 dietary components: percent energy from trans and omega-3 fatty acids, fiber, folate, glycemic load, and polyunsaturated:saturated fatty acid ratio. Physical activity was measured using the Baecke physical activity questionnaire.²³ Baseline medical history included diabetes mellitus (self-reported physician diagnosis, fasting glucose ≥126 mg/dL, nonfasting ≥200 mg/dL, or self-reported pharmacological treatment) and coronary artery disease (ECG-adjudicated or self-reported myocardial infarction or any of self-reported heart/arterial surgery, coronary bypass, or angioplasty).

Statistical Methods

Descriptive statistics for participant characteristics were calculated according to alcohol intake category. Cox proportional hazards regression was used to estimate hazard ratios (HRs) and 95% confidence intervals for the association between alcohol and incident IS, ICH, and total stroke. Participants contributed person-time until the earliest of: incident stroke, death, loss to follow-up, or end of follow-up on December 31, 2011. The proportional hazards assumption was tested using interaction terms between exposure and time.

Alcohol consumption was categorized as drinks per week by examining the dose–response relationship as well as using a priori values selected to align with previous research and recommended guidelines. Results were robust to different category boundaries and are presented herein as ≤3, 4 to 17, and ≥18 drinks/wk, which reflect our data-informed categorization and cut points of previous studies.¹⁴ In a second alcohol assessment at year 6, 67% of participants were categorized the same as at baseline. We assessed possible nonlinear relationships using quadratic splines and polynomial terms. Knots were selected based on Akaike information criterion values compared across models with 2, 3, and 4 knots located at percentile values.²⁴ Secondary models for IS were stratified by sex–race group.

In secondary analysis, we estimated subdistribution (HR_SD) to assess the risk of stroke given the relatively high proportion of death (26%) during follow-up. Cause-specific hazard models, which censor deaths, yield estimates reflecting the relative rate of stroke. HR_SD, however, reflect the relative risk over a period of time.¹⁶ These were obtained using the SAS macro PSHREG based on the proportional subdistribution hazard models proposed by Fine and Grey.²⁵ All analyses were conducted using Statistical Analysis Software Version 9.2 (SAS Inc, Cary, NC).

Results

Over one third of the ARIC participants were light alcohol drinkers, consuming ≤3 drinks/wk (Table 1). Roughly one third reported lifetime abstention from alcohol, one quarter were moderate drinkers of 4 to 17 drinks/wk, and only 5% consumed >18 drinks/wk. Women comprised the largest proportion of abstainers and white men accounted for nearly three quarters of heavier drinkers. Light-to-moderate drinkers were of higher socioeconomic status—in the form of greater educational attainment, more managerial occupations, and higher family income—than heavier drinkers and lifetime abstainers. Current smoking was reported by 46% of heavier drinkers but only 23% of light drinkers. The prevalence proportion of diabetes mellitus was low overall (8.6%) and roughly twice as high in abstainers compared with current drinkers. Finally, blood pressure and HDL-C increased across alcohol consumption level.

Over a median follow-up of 22.6 and 22.7 years, respectively, there were 773 IS and 81 ICH. IS incidence rates per 100 000 person-years increased across alcohol intake categories: 251 for ≤3 per week, 313 for 4 to 17 per week, 435 for ≥18 per week, and 368 for abstainers (Table 2). Estimates were attenuated after adjustment for covariates; most of this attenuation occurred with model 1 covariates. In fully adjusted Cox models, light and moderate drinking were not associated with IS compared with lifetime abstention (HR, 0.98; 95% CI, 0.79–1.21 and HR, 1.06; 95% CI, 0.84–1.34, respectively; Table 2). Heavier drinking was associated with a 31% increase relative to abstainers (HR, 1.31; 95% CI, 0.92–1.86). Confidence intervals for these associations were wide and overlapped the null.

Associations of light-to-moderate drinking versus abstention were in opposite directions in whites compared with blacks and confidence intervals overlapped considerably and, therefore, we cannot conclude that HRs differ by sex–race group (Figure I in the online-only Data Supplement; P=0.5). HRs ranged from 0.78 (0.56–1.10) in white women to 1.20 (0.75–1.92) in black men.

ICH incidence rates ranged from 20 per 100 000 person-years among light drinkers to 41 among moderate-to-heavier drinkers. We found no clear association of light drinking compared with abstention (HR, 1.04; 95% CI, 0.56–1.94) after adjustment for age, race, sex, smoking status, and education (Table 3). Moderate-to-heavier drinking, however, was associated with higher ICH rates compared with abstention (HR, 1.99; 95% CI, 1.07–3.70); this estimate was unchanged when heavier drinkers were removed. The precision of these estimates was low because of small numbers of events, which also precluded adjustment for additional lifestyle factors. Effect estimates for total stroke were similar to those for IS alone: HR_light, 0.98; 0.80 to 1.20; HR_moderate, 1.13; 0.91 to 1.42; and HR_heavy, 1.36; 0.97 to 1.91.

Results from models with quadratic splines representing alcohol intake did not support a J-shaped relationship (Figure); higher-order spline terms were not significant, although these analyses were underpowered to estimate dose–response relationships. The log-HR of IS was roughly linear across intake (β=0.06 for a 1 drink/wk increase; HR, 1.06; 95% CI, 0.99–1.13; P=0.1) and also nonsignificant. Additional quadratic (P=0.3) and cubic terms (P=0.9) were not statistically significant compared with the linear model.

• Download figure
• Download PowerPoint

More than one quarter of the participants died during follow-up, ranging from 24% in light to 42% in heavier drinkers. The estimated HR_SD account for this underlying mortality and reflect the relative cumulative incidence in our population. As expected, effect estimates were attenuated slightly for heavy drinkers; light and moderate alcohol consumption were not associated with reduced IS risk compared with abstention and moderate intake may increase ICH risk (Tables 2 and3). Finally, we explored the impact of former versus never drinking on stroke. Estimates were nonsignificant with wide confidence intervals; interpretation of these estimates is difficult because of heterogeneity in consumption levels among former drinkers.

Conclusions

We did not find a protective effect of light-to-moderate midlife alcohol consumption on IS or ICH in this analysis of a biracial population-based cohort of US adults. Heavier intake tended to increase rates of IS and ICH compared with abstention, although confidence intervals were wide. The dose–response relationship we estimated for IS was imprecise, and did not support a clear J-shaped or linear relationship.

Alcohol consumed moderately in midlife may lower IS risk through beneficial alterations in vascular risk factors, including HDL-C, blood pressure, platelet aggregation, inflammatory markers and insulin sensitivity compared with no drinking.^7,26–28 In contrast, high doses of alcohol have well-established deleterious effects, including elevated blood pressure, inflammation, and development of atrial fibrillation. Low-dose alcohol may also increase the risk for hemorrhage through hemostatic changes that promote bleeding. Dose-specific effects may differ by sex; women attain higher blood alcohol levels than men because of different body composition and alcohol metabolism.²⁹

A J-shaped relationship between alcohol and IS is frequently reported in meta-analyses.^7,8,13–15 Consumption of 1 to 2 drinks/d was associated with 28% lower risk and >4 drinks/d with 69% increased risk.⁷ Meta-analyzed results may be limited by inclusion of studies with nonvalidated stroke events, that lacked adjustment for lifestyle factors, and that included causal intermediates (eg, blood pressure and HDL-C) in regression models. Our results, based on well-validated events with adjustment for important confounders, suggest no clear association between alcohol and stroke until heavier intake levels. Inconsistencies in estimates across studies could result from differential distributions of etiologic subtypes in these cohorts coupled with a different mechanistic role of alcohol in their pathogenesis (ie, larger protective effect for thrombotic versus cardioembolic strokes). Differences may also result from heterogeneity according to drinking pattern or sex–race or from measurement and selection biases.

Stronger protective effects of low-dose alcohol are reported for women compared with men (HR, 0.66; 95% CI, 0.61–0.71 versus HR, 0.80; 95% CI, 0.67–0.96).^7,8 Women consuming <2 drinks/d in the Nurses’ Health Study (a primarily white cohort) had a 12% to 18% lower risk of IS compared with nondrinkers.¹⁴ This aligns with results for white, but not black, women in the ARIC study. Our results for men were similar to those reported by the Health Professionals Follow-up Study that found no association for <1 drink/d and slightly elevated risk of ≥1 drinks/d.¹² We noted nonsignificantly higher HRs for blacks compared with whites that could be explained by different drinking patterns between whites and blacks.³⁰ Evidence suggests that moderate quantities of alcohol may be beneficial when consumed at moderate, but not high, frequency.¹²

Meta-analyzed data indicate that ICH risk increases log-linearly with increasing alcohol consumption for men, with possible J-shaped curves for women.^6,7,14 Consumption of ≤2 drinks/d is associated with 12% to 24% lower hemorrhage risk compared with abstention.^6,7,14 We did not find evidence of a protective effect of light drinking in our population. ICH rates were increased even at moderate drinking levels compared with abstention. Our results, however, were limited by small numbers of events.

Our results should be interpreted in light of several limitations. Participants may under-report alcohol consumption leading to misclassification. Although we were unable to quantify errors, construct and rank-order validity was supported by positive correlations of alcohol with both HDL-C and blood pressure. Residual confounding is possible. We had low power to estimate precise effects, particularly among heavier drinkers and for ICH. Finally, race-specific results may not generalize to the US population outside of the ARIC communities.

Strengths of our study include a prospective study design with >20 years of follow-up, a biracial population, and robust stroke ascertainment. Alcohol consumption was assessed using a validated instrument with beverage-specific questions (thus reducing under-reporting) that differentiated never from former drinkers. We had rich covariate data that permitted adjustment for lifestyle factors, smoking, and socioeconomic status.

Public health recommendations for alcohol consumption must consider both its benefits and risks. Although light-to-moderate intake may reduce the risk of some cardiovascular outcomes, other harmful effects exist even at low doses (eg, dependency and cancer). As such, the American Heart Association does not recommend initiation of drinking for disease prevention.³¹ Our results support this recommendation. We found no significant risk reduction for IS or ICH with light-to-moderate midlife alcohol consumption and increased risks at heavier intake levels. Understanding the alcohol–stroke relationship would be advanced by assessing dose-dependent exposure measurement errors, updating meta-analyses to include only high-quality studies and to explore modification by drinking pattern, pathogenic subtype, and sex–race.

Acknowledgments

We thank the staff and participants of the Atherosclerosis Risk in Communities study for their important contributions.

Footnotes