|Home | About | Journals | Submit | Contact Us | Français|
Djoussé L: Designed study, conducted data analyses, and drafted the manuscript.
Lee IM: Collected data, critically reviewed the manuscript, supervised study (I take this to mean directing the WHS), and contributed to data analysis.
Buring JE: Collected data, critically reviewed the manuscript, and obtained funding.
Gaziano JM: Collected data, critically reviewed the manuscript, and supervised the study.
Although an association between moderate alcohol consumption and decreased cardiovascular disease (CVD) and mortality has been reported, limited data are available on potential mediating mechanisms. We examined the association between alcohol and CVD and mortality in 26,399 women and estimated the proportion of reduced risk of CVD/mortality explained by a series of intermediate factors.
Alcohol consumption was self-reported at baseline and CVD events and deaths were ascertained via follow-up questionnaires and medical records. Baseline levels of hemoglobin A1c, inflammatory markers, hemostatic factors, and lipids were measured. Blood pressure and hypercholesterolemia; and treatment for lipids were self-reported. During a mean follow up of 12.2 years, 1039 CVD events, 785 deaths (153 CVD deaths) occurred. There was a J-shaped relation between alcohol consumption and incident CVD, total and CVD mortality in a multivariable model. Compared with abstainers, alcohol intake of 5-14.9 g/d was associated with 26%, 35%, and 51% lower risk of CVD, total and CVD mortality, respectively, in a multivariable model. For CVD risk reduction, lipids made the largest contribution to the lower risk of CVD (28.7%), followed by hemoglobin A1c/diabetes (25.3%), inflammatory/hemostatic factors (5%), and blood pressure factors (4.6%). All these mediating factors together explained 86.3%, 18.7%, and 21.8% of the observed lower risk of CVD, total and CVD mortality, respectively.
These data suggest that alcohol effects on lipids and insulin sensitivity may account for a large proportion of the lower risk of CVD/mortality observed with moderate drinking under the assumption that the alcohol-CVD association is causal.
Several epidemiologic studies have reported U-or J-shaped relation between alcohol consumption and cardiovascular disease (CVD)1-5 and mortality6-8. HDL raising effect of alcohol drinking has been demonstrated in both observational9 and intervention studies10,11. Criqui et al.12 first reported that about half of the protective effect of alcohol on coronary disease and CVD mortality was mediated by HDL in a comprehensive assessment. These findings were confirmed later by other investigators9,13,14 or similar endpoint of non-fatal myocardial infarction. While health benefits of moderate drinking (about 1 drink per day for women and 1-2 drinks per day for men)15 may be mediated through HDL and non-HDL factors such as insulin sensitivity, inflammatory markers, hemostatic factors, or adiponectin16-20, investigation of non-HDL mediators of alcohol on CVD have received limited attention, especially with incident CVD or total/CVD mortality as endpoints. In addition, limited data are available on the joint influence of major mediators of moderate drinking on mortality and CVD. In a nested-case control study from the Nurses Health Study and the Health Professionals Follow-Up Study, Mukamal et al.21 demonstrated that alcohol consumption (at least 3-4 days per week) was associated with a lower risk of myocardial infarction in both sexes and that such relation was mediated by HDL, fibrinogen, and hemoglobin A1c.
Since alcohol consumption, even in moderate ranges may be associated with adverse health effects including raising blood pressure22,23 or increased risk of breast cancer24,25, it is important to understand the net contribution of physiologic effects of alcohol consumption on total mortality as well as CVD. Because most of the published mediation studies have concentrated on coronary artery disease9,14,21,26 and only few have focused on total CVD or all-cause and CVD mortality12,13, the current project sought to examine the association between alcohol consumption and CVD and mortality and determine the proportion of risk reduction explained by a series of potential mediating factors in the Women's Health Study.
We used data from the Women's Health Study (WHS) for which a detailed description of the WHS has been published previously27,28. Briefly, a total of 39,876 female health professionals aged 45 years and older at entry (1992-1995) were randomized to low-dose aspirin, vitamin E, or their corresponding placebos27,28. Each participant gave written informed consent and the Institutional Review Board at Brigham and Women's Hospital approved both study protocols.
For the present analyses, we included only 28,345 women (71.1%) who provided a blood sample at baseline. We then excluded women with a) missing data on biomarkers (n=738), b) missing alcohol information (n=6), c) pre-randomization reports of CVD that occurred before baseline (n=7), and d) missing data on potential confounders including body mass index, exercise, smoking, energy intake, fruits and vegetables, systolic blood pressure, and hypertension (n=1,195). Thus, a total sample of 26,399 women was used for current analyses. Characteristics between subjects who provided blood samples and those who did not were comparable (data not shown).
Alcohol consumption during the preceding 12 months was assessed by a questionnaire. Total alcohol intake was computed as the sum of the alcohol content in beer, wine, and spirits. We assumed that 360 ml (12 ounces) of beer contains 13.2 g of ethanol; 11.3 g for 360 ml (12 ounces) of light beer; 10.8 g for 120 ml (4 ounces) of wine; and 15.1 g for 45ml (1.5 ounces) of liquor. Additional details on alcohol assessment in the WHS have been published29.
Detailed description of ascertainment of CVD and mortality in the WHS has been previously described30. Briefly, for all cases of myocardial infarction, stroke, coronary revascularization and angioplasty, or cardiovascular death reported after enrollment, hospital records were obtained and reviewed by an Endpoint Committee. Myocardial infarction was confirmed if symptoms met World Health Organization criteria and the event was associated with elevated cardiac enzymes or characteristic ECG changes. Each stroke case was confirmed if the patients had a new neurological deficit with signs and symptoms persisting for > 24 hours (CT scans were available in most cases). Revascularization procedures and angioplasty were confirmed by hospital records. Deaths were confirmed by autopsy reports, death certificates, and circumstances of death. Incident CVD included myocardial infarction, coronary angioplasty and revascularization, ischemic stroke, and cardiovascular deaths.
Blood samples were collected at baseline in EDTA tubes and shipped cold overnight to a core laboratory, where they were centrifuged and frozen at −170°C (vapor phase liquid nitrogen) until analysis. Hemoglobin A1c concentration was measured by turbidometric immunoassay in red blood cells using the Hitachi 911 Analyzer (Roche Diagnostics, Indianapolis, IN). High-sensitivity C-reactive protein was measured using a high-sensitivity ELISA (Abbott Laboratories). Fibrinogen was measured using an immunoturbidimetric assay (Kamiya Biomedical, Seattle, WA) and soluble ICAM-1 was measured using an enzyme-linked immunosorbent assay (R & D Systems, Minneapolis, MN). Triglyceride levels were measured enzymatically, with correction for endogenous glycerol, using a Hitachi 917 analyzer and reagents and calibrators from Roche Diagnostics (Indianapolis, Indiana). Levels of total cholesterol and HDL were measured enzymatically on a Hitachi 911 autoanalyzer (Roche Diagnostics, Basel, Switzerland), and levels of LDL were determined directly (Genzyme, Cambridge, MA). These assays are approved for clinical use by the US Food and Drug Administration.
Demographic data were collected at baseline. In addition, information on blood pressure, history of hypertension, dyslipidemia or use of cholesterol lowering medications, history of diabetes, menopausal status, use of hormone replacement therapy, family history of premature myocardial infarction, smoking, physical activity was obtained at baseline. Dietary information was obtained by a food frequency questionnaire.
We classified each subject into one the following categories of total alcohol consumption: 0, 0.1-4.9, 5.0-14.9, 15.0-29.9, and ≥ 30 g/d. These cut points have been used previously in alcohol analyses in the WHS29 and make a distinction between moderate (5.0-14.9 g/d, which is about ½ to 1 standard drink per day) and heavy drinking among women (more than 2 drinks per day). We computed person-time of follow up from baseline until the first occurrence of outcome of interest (CVD or death) or censoring date, the date of receipt of the last follow-up questionnaire. We used Cox proportional hazard models to compute multivariable adjusted hazard ratios with corresponding 95% confidence intervals using subjects in the lowest category of alcohol consumption as the reference group. The initial model adjusted for age whereas the multivariable model controlled for age (continuous), body mass index (<25, 25-29, 30+ kg/m2), smoking (never, former, and current smokers), physical activity (quintiles of kilocalories per week), fruit and vegetable intake (≤ 1, 2-3, 4-5, and ≥ 6 servings per day), menopausal status (pre- or post-menopause, or uncertain), and family history of premature myocardial infarction. This model is subsequently referred to as the “basic model”. Additional adjustment for energy intake (quintiles) and other nutrients did not alter the results (data not shown), and so we did not adjust for these dietary variables in the basic model. P value for linear trend was obtained by fitting a continuous variable that assigned the median alcohol consumption in each alcohol category in a Cox regression model. For quadratic trend, we also added a squared term in the above model.
To examine the extent to which the observed lower risk of CVD or mortality associated with alcohol consumption is explained by a set of potential intermediate factors, we first estimated the hazard ratio from the basic model, comparing moderate drinkers (5.0-14.9 g/d of alcohol, or about ½ to 1 drink per day) to abstainers. We then estimated the same hazard ratio, adding sets of intermediate factors, one set at a time, to the basic model. Sets of intermediate factors included a) blood pressure factors (systolic blood pressure and use of antihypertensive medications), b) glucose metabolism factors (prevalent diabetes and HbA1c), c) inflammatory/hemostatic factors (hs-CRP, SICAM-1, and fibrinogen), d) and lipids (treatment for high cholesterol, HDL, LDL, triglycerides). Then, we considered the magnitude of change in the hazard ratio for the moderate drinkers compared with abstainers with and without addition of each set of intermediate factors of interest. A multivariate model that included a set of intermediate factors is referred to as the “intermediate factor”. A larger change in the hazard ratio toward the null implied a larger mediating effect of intermediate factors on the alcohol-related reduction in CVD or mortality risk. We used a previously defined equation31 to calculate the proportion of CVD/mortality risk reduction explained by each set of intermediate factors as follows: 100% x [HRfrom basic model – HRfrom intermediate factors] / [HRfrom basic model −1 ]. For the proportion of effect explained by a set of intermediate factors, we used the following equation that also control for confounding by other potential mediators: (HR3 - HR2) / (1 - HR1), where HR3 is the hazard ratio for moderate alcohol use from the model including all intermediate factors, HR2 is from the model including all but the intermediate factor of interest, and HR1 is from the basic model excluding all intermediate factors.
In sensitivity analyses, we examined the influence of breast cancer deaths on total mortality analyses (as moderate drinking may increase the risk of breast cancer). In addition, because some investigators have suggested that wine may confer greater health benefits than beer or spirits, we examined the association of beverage type and studied outcomes by including consumption of beer, wine, and spirits in the same model [each beverage type was grouped as 0 (reference), 1-7, and 8+ drinks/week]. Because consumption of one half to two drinks per day may still be considered as light-to-moderate drinking, we also repeated mediation analyses using corresponding intake of 5.0 to 30 g/d of alcohol. To examine the validity of self-reported alcohol, we examined the association between alcohol and plasma HDL-cholesterol. All analyses were completed using SAS, version 9.1 (SAS Institute, NC). Significance level was set at 0.05. The authors had full access to the data and take responsibility for its integrity. All authors have read and agree to the manuscript as written.
The mean age at randomization was 54.7±7.1 years (range 38.7 to 89.9 years) in the WHS. Table 1 presents baseline characteristics of the study participants. Alcohol consumption was associated with older age, smoking, higher levels of HDL, and lower concentration of C-reactive protein and fibrinogen as expected. Alcohol intake was strongly related to HDL in this cohort (age-adjusted means: 50.8, 53.9, 57.5, 59.4, 60.9, and 62.6 mg/dl from the lowest to the highest alcohol category, p for trend <0.0001).
During a mean follow up of 12.2 years, 1039 new CVD events, 785 confirmed deaths, and 153 cardiovascular deaths occurred. There was a J-shaped relation between alcohol consumption and CVD risk in a multivariable model adjusting for age, body mass index, smoking, physical activity, intake of fruits and vegetables, menopausal status, and family history of premature myocardial infarction, with the lowest risk observed in women consuming 5.0-14.9 g/d of alcohol [HR=0.74 (95% CI: 0.61-0.90), Table 2]. Additional adjustment for aspirin or vitamin E assignment did not alter the results. Similarly, there was a J-shaped relation between alcohol consumption and total and CVD mortality with the largest effect observed in women consuming 5.0-14.9 g/d [HR (95% CI) of 0.65 (0.51-0.82) for total death and 0.49 (0.27-0.89) for CVD death, Table 3].
The source of ethanol did not influence these findings. For example, multivariable adjusted hazard ratio for total mortality was 0.90 (0.74-1.10) for moderate consumption of wine compared to abstention from wine. Corresponding values were 0.83 (0.66-1.03) for beer and 0.86 (0.72-1.03) for liquor in a model that simultaneously controlled for other types of beverage. When moderate drinking was defined as 5 to 30 g/d of alcohol, similar results were observed with adjusted HR (95% CI) of 0.77 (0.64-0.91) for CVD; 0.71 (0.58-0.87) for all deaths, and 0.57 (0.35-0.94) for CVD deaths compared with abstainers. Exclusion of 54 cases of breast cancer deaths did not alter the results for total mortality [i.e. HR (95% CI) of 0.65 (0.51-0.82) with and 0.65 (0.51-0.82) without exclusion of breast cancer deaths in women consuming 5.0 to 14.9 g/d of alcohol].
For the incident CVD outcome, adding a set of lipid factors or glucose metabolism to the basic model each led to a largest attenuation of the hazard ratios (Table 2). Adding a set of inflammatory/hemostatic factors or blood pressure to the basic model had a minimal effect (Table 2). For total and CVD death analyses, adding any of the set of intermediate factors to the basic model had little influence on the hazard ratios (Table 3).
While additional adjustment for all sets of intermediate factors led to a complete elimination of the association between alcohol and CVD (Table 2), no major change was observed for total mortality, and a modest attenuation of the hazard ratios was seen for CVD mortality (Table 3). When examined as a set of risk factors, lipids factors were the largest contributors to the lower risk of CVD (28.7%), followed by diabetes/HbA1c (25.3%), inflammatory/hemostatis factors (5.0%), and blood pressure factors (4.6%), see Figure. Overall, 86.3% of the lower risk of CVD observed among moderate drinkers was explained by the potential mediators that we investigated (Figure), assuming a causal relation between alcohol and CVD. In contrast, investigated sets of mediators explained only 18.7% and 21.8% of the lower risk of total and cardiovascular mortality, respectively (Figure), under similar assumptions. For total and CVD mortality, it is noteworthy that none of the mediators individually explained a larger proportion of the observed relation, Figure. Using consumption of 5-30 g/d as moderate drinking did not change the inference: 93.2%, 21.7%, and 23.6% of the observed reduction in CVD, total and CVD deaths were accounted for by studied mediators, respectively, assuming a causal relation between alcohol and outcome.
In this large prospective study, we observed a J-shaped relation between alcohol consumption and CVD and mortality among women. Furthermore, we estimated that 86.3% of the lower risk of CVD observed in moderate drinkers was explained by alcohol effects on lipids, glucose metabolism, inflammatory/hemostatic factors, and blood pressure. Nearly 20% of the reduced risk of either total or CVD mortality among moderate drinkers was accounted for by these intermediate factors.
Our findings of a lower risk of CVD and mortality among moderate drinkers are consistent with data from the Nurses' Health Study, which reported a J-shaped relation between alcohol consumption and total and cardiovascular mortality32 and other cohorts that reported beneficial effects of moderate drinking on the risk of CVD33-35. As showed by others15,36,37, our data did not support differential associations for beer, wine, or spirits. The above interpretation of our data is only valid under the assumption that moderate drinking is causally related to CVD and mortality and that such causal relation may be mediated by studied factors. This is obviously a big assumption and we are far from providing proof for such relation in the absence of large randomized controlled trials (due to ethical reasons). It is possible that unmeasured or residual confounding could explain our findings.
There is a wide range of biological effects of alcohol20. In a randomized crossover trial of 63 postmenopausal women, consumption of 30 g/d of alcohol was associated with improved insulin sensitivity and lower fasting triglycerides after 8 weeks of intervention when compared to 0 g/d of alcohol38. In a trial of type 2 diabetes subjects who had previously abstained from alcohol, daily consumption of one alcoholic drink resulted in a reduction of fasting glucose after 30 days of intervention and such relation was stronger in subjects with higher HbA1c39. Other investigators have reported beneficial effects of moderate drinking on glucose metabolism40,41. In addition to the well-established HDL-raising effects of alcohol9-11, recent data showed that alcohol consumption has been associated with a favorable lipoprotein subclasses (larger particle size and fewer atherogenic small size LDL-particles)42. Other studies have provided evidence in support of beneficial effects of alcohol on adiponectin43,44, C-reactive protein18,45, fibrinogen18,45, and adhesion molecules46,47. The relation between alcohol intake and blood pressure has been inconsistent with most studies reporting an increase in blood pressure with heavy alcohol consumption22. These biologic paths of alcohol intake lend support to an important role of alcohol on glucose metabolism, lipids, and inflammatory markers in explaining about 86% of reduced risk of CVD observed in this study. This is consistent with a report showing that HDL, A1c, and fibrinogen explained a) 100% of the association between drinking and MI in the Health Professionals Follow up Study and b) 80% of the effects observed in the Nurses' Health Study21. In both men and women, HDL alone explained about half of the observed association21. In the Honolulu Heart program14, HDL accounted for about 50% of the lower risk of CHD observed with alcohol consumption whereas LDL and blood pressure had minimal influences. These data are consistent with other reports of mediation alcohol-CHD relation by HDL9,26.
We observed a minimal attenuation of the effects of moderate drinking on CVD upon control of hypertension variables. This finding is contrary to our primary hypothesis of blood pressure raising effect of any alcohol consumption. However, in heavy drinkers, we observed expected results in that the alcohol effect on CVD risk (17% lower risk) was greater (27% lower risk) upon control of blood pressure variables (Table 2).
In the present study, 18.7% of total and 21.8% of cardiovascular mortality was explained by a combination of studied intermediate factors. In the MRFIT study, HDL explained 45% of the association between alcohol intake and CHD mortality in men13. Consistent with our data, the Lipid Research Clinics Follow-up Study12 showed that moderate drinking in men and women was associated with a lower risk for CVD mortality and such relation was partially mediated by HDL and not LDL-cholesterol (e.g.; RR of 0.45 without and 0.54 with additional control for HDL in women). Of note is that there were only 48 CVD deaths in the LRC study12 and the study while not showing a significant effect was underpowered.
Contrary to the larger effect of alcohol explained by lipids, A1c, inflammatory and hemostatic factors in our study, a relatively small proportion of total and CVD mortality was explained by these factors. How might we explain such difference? It is possible that some of the beneficial effects of moderate drinking on CVD could be offset by excess risk of breast cancer as reported by others24,48. This scenario would attenuate the proportion of total deaths explained by alcohol effects in comparison to the proportion of CVD deaths or CVD events explained by alcohol intake. However, exclusion of deaths from breast cancer did not alter the findings on total mortality, suggesting limited influence of breast cancer deaths on our findings. Besides cancers, other conditions associated with heavy drinking such as cardiomyopathy, arrhythmias, cirrhosis, hypertension, and intoxication may also contribute to increased mortality and consequently diminish the proportion of reduced risk of death accounted for by alcohol effects. Since these conditions occur mostly with alcohol abuse, it is less likely that our observation among moderate drinkers was mainly due to toxic effects of alcohol. Nonetheless, it is possible that some heavy drinkers may have been misclassified as moderate drinkers due to underreporting of drinking habits. Alternatively, other biologic mechanisms not evaluated in the present study may be responsible for the alcohol-mortality relation. Furthermore, difficulties assigning the true cause of death in some cases may also contribute to such difference.
The current study has some limitations. It is possible that some of the covariates examined (i.e. hypertension or diabetes) may have influenced subsequent consumption of alcohol suggesting that they could be both intermediate factors and confounders. Thus, the use of alcohol and predictors of CVD assessed at a single time point does not allow to properly examine mediation in this setting and as consequence, our estimates of proportion of CVD explained may be biased. Generalizability of our data is limited by the fact that participants were female health professionals who may have different behaviors than males or the general population due to their higher educational attainment and a high socioeconomic status. We had a single measurement of intermediate factors which might have been influenced by factors unrelated to alcohol intake. Self-report on alcohol intake may have led to exposure misclassification (over- or underreporting). Due to the lack of repeated measures on alcohol consumption over time, we were unable to account for change in alcohol consumption over time in this cohort. Lastly, since individuals were not randomly assigned to alcohol consumption, we can not exclude chance, residual confounding, or confounding by unmeasured variables including psychological factors as a possible explanation of our findings. Nevertheless, our large sample size, the relatively long follow up, the standardized ascertainment of CVD events and deaths in the WHS, and the availability of various covariates are major strengths of this study.
In conclusion, our data suggest that a large proportion of the lower risk of CVD observed with moderate drinking may be accounted for by factors related to glucose metabolism, lipids, and inflammation/hemostasis in adult women under the assumption of a causal relation between moderate drinking and CVD. If confirmed in other populations, such information may help clarify the possible mechanisms underlying the observed association between moderate alcohol intake and CVD risk.
The Women's Health Study is supported by grants CA-047988, HL-43851, HL-080467 from the National Institutes of Health, Bethesda, MD.
Role of the Sponsor:
No influence on this paper.
Conflict of Interest Disclosures: