|Home | About | Journals | Submit | Contact Us | Français|
Objective To conduct a comprehensive systematic review and meta-analysis of studies assessing the effect of alcohol consumption on multiple cardiovascular outcomes.
Design Systematic review and meta-analysis.
Data sources A search of Medline (1950 through September 2009) and Embase (1980 through September 2009) supplemented by manual searches of bibliographies and conference proceedings.
Inclusion criteria Prospective cohort studies on the association between alcohol consumption and overall mortality from cardiovascular disease, incidence of and mortality from coronary heart disease, and incidence of and mortality from stroke.
Studies reviewed Of 4235 studies reviewed for eligibility, quality, and data extraction, 84 were included in the final analysis.
Results The pooled adjusted relative risks for alcohol drinkers relative to non-drinkers in random effects models for the outcomes of interest were 0.75 (95% confidence interval 0.70 to 0.80) for cardiovascular disease mortality (21 studies), 0.71 (0.66 to 0.77) for incident coronary heart disease (29 studies), 0.75 (0.68 to 0.81) for coronary heart disease mortality (31 studies), 0.98 (0.91 to 1.06) for incident stroke (17 studies), and 1.06 (0.91 to 1.23) for stroke mortality (10 studies). Dose-response analysis revealed that the lowest risk of coronary heart disease mortality occurred with 1–2 drinks a day, but for stroke mortality it occurred with ≤1 drink per day. Secondary analysis of mortality from all causes showed lower risk for drinkers compared with non-drinkers (relative risk 0.87 (0.83 to 0.92)).
Conclusions Light to moderate alcohol consumption is associated with a reduced risk of multiple cardiovascular outcomes.
Possible cardioprotective effects of alcohol consumption seen in observational studies continue to be hotly debated in the medical literature and popular media. In the absence of clinical trials, clinicians must interpret these data when answering patients’ questions about taking alcohol to reduce their risk of cardiovascular disease. Systematic reviews and meta-analyses have addressed the association of alcohol consumption with cardiovascular disease outcomes1 2 3 4 5 6 7 8 but have not uniformly addressed associations between alcohol use and mortality from cardiovascular disease, as well as the incidence and mortality from coronary heart disease and stroke. Additionally, further studies have been published since 2006, when the most recent reviews appeared. The continuing debate on this subject warrants an in depth reassessment of the evidence.
In this paper, we synthesise results from longitudinal cohort studies comparing alcohol drinkers with non-drinkers for the outcomes of overall mortality from cardiovascular disease, incident coronary heart disease, mortality from coronary heart disease, incident stroke, and mortality from stroke. Because of the many biological effects of alcohol consumption, we also examine the association of alcohol with mortality from all causes when this is reported in studies. We conducted meta-analyses for each of these outcomes and a sensitivity analysis with lifetime abstainers as the reference category to account for the heterogeneity within the reference group of non-drinkers. We also examined the effect of confounding on the strength of observed associations. In our companion paper,110 we link these cardiovascular outcomes with experimental trials of alcohol consumption on candidate causal molecular markers.
We performed a systematic review and meta-analysis following a predetermined protocol in accordance with the Meta-analysis of Observational Studies in Epidemiology (MOOSE) reporting guidelines.9 We identified all potentially relevant articles regardless of language by searching Medline (1950 through September 2009) and Embase (1980 through September 2009). Searches were enhanced by scanning bibliographies of identified articles and review articles, as well as reviewing conference proceedings from three major scientific meetings (American Heart Association, American College of Cardiology, and European Heart Congress) between 2007 and 2009. Experts in the field were contacted regarding missed, ongoing, or unpublished studies.
To search electronic databases, we used the strategy recommended for systematic reviews of observational studies.10 We specified three comprehensive search themes:
These three comprehensive search themes were then combined using the Boolean operator “and” in varying combinations.
Two individuals (SEB and PER) independently reviewed all identified abstracts for eligibility. All abstracts reporting on the association between alcohol intake and cardiovascular disease events were selected for full text review. This stage was intentionally liberal. We discarded only those abstracts that clearly did not meet the aforementioned criteria. The inter-rater agreement for this review was high (κ=0.86 (95% confidence interval 0.80 to 0.91)). Disagreements were resolved by consensus.
The same reviewers performed the full text review of articles that met the inclusion criteria and articles with uncertain eligibility. Articles were retained if they met the inclusion criteria for study design (prospective cohort design), study population (adults ≥18 years old without pre-existing cardiovascular disease), exposure (current alcohol use with a comparison group of non-drinkers), and outcome (overall cardiovascular disease mortality or atherothrombotic conditions, specifically incident coronary heart disease, coronary heart disease mortality, incident stroke, or stroke mortality). Both published and unpublished studies were eligible for inclusion. Authors were contacted if the risk profile of the cohort was unclear.
The primary exposure variable was the presence of active alcohol drinking at baseline compared with a reference group of non-drinkers. Because of the heterogeneity of this reference group, we identified the subset of studies using lifetime abstainers as the reference group and studies that distinguished former drinkers from non-drinkers. Whenever available, we extracted information on amount of alcohol consumed, using grams of alcohol per day as the common unit of measure. When a study did not specifically report the grams of alcohol per unit, we used 12.5 g/drink for analysis.11 We standardised portions as a 12 oz (355 ml) bottle or can of beer, a 5 oz (148 ml) glass of wine, and 1.5 oz (44 ml) glass of 80 proof (40% alcohol) distilled spirits. Volume of intake was categorised as <2.5 g/day (<0.5 drink), 2.5–14.9 g/day (about 0.5–1 drink), 15–29.9 g/day (about 1–2.5 drinks), 30–60 g/day (about 2.5–5 drinks), and >60 g/day (≥5 drinks).
The outcome variables of interest were defined as the presence or absence of death from cardiovascular disease (that is, fatal cardiovascular or stroke events), incident coronary heart disease (fatal or non-fatal incident myocardial infarction, angina, ischaemic heart disease, or coronary revascularisation), death from coronary heart disease (fatal myocardial infarction or ischaemic heart disease), incident stroke (ischaemic or haemorrhagic events), or death from stroke. A secondary analysis was performed within these selected studies to determine the association between alcohol consumption and the risk of death from all causes.
Both reviewers independently extracted data from all studies fulfilling the inclusion criteria, and any disagreement was resolved by consensus. We extracted the data elements of cohort name, sample size, and population demographics (country, percentage male, mean age or age range). We also extracted information for key indicators of study quality in observational studies proposed by Egger et al10 and Laupacis et al.12 Specifically, we evaluated the effect on each outcome of the number of potential confounding variables and the number of years participants were followed.
The relative risk was used as the common measure of association across studies. Hazard ratios and incidence density ratios were directly considered as relative risks. Where necessary, odds ratios were transformed into relative risks with this formula:
The standard error of the resulting converted relative risk was then determined with this formula:
Because these transformations can underestimate the variance of the relative risks derived from the odds ratios,14 15 we performed a sensitivity analysis that excluded four studies for which this transformation had been applied. All analyses were performed with Stata 10.0 (StataCorp, College Station TX, USA). The Stata “metan” command was used to pool the ln(relative risks) across studies according to the DerSimonian and Laird random effects model.16
In some studies, a single relative risk (or odds ratio) was not available for drinkers versus non-drinkers because the data were presented as only a dose-response (that is, several alcohol consumption levels relative to non-drinkers). In these cases, we first pooled across levels of intake within the study using a random effects model to derive a single relative risk for drinkers versus non-drinkers. The resulting single, study-specific relative risk was then pooled with those of other studies.
To visually assess the relative risk estimates and corresponding 95% confidence intervals across studies, we generated forest plots sorted by year of publication. Analyses were stratified by study quality criteria and by participant characteristics.
To assess heterogeneity of relative risks across studies, we inspected forest plots and calculated Q (significance level of P≤0.10) and I2 statistics.17 18 In the presence of heterogeneity, random effects models were used (rather than fixed effects models) to obtain pooled effect estimates across studies. Sensitivity analyses and stratified analyses were performed to assess the associations of selected study quality and clinical factors on cardiovascular risk, including number of confounding factors and duration of follow-up dichotomised at the median value. We also performed a sensitivity analysis excluding studies reporting only odds ratios. We conducted a cumulative meta-analysis of studies ordered chronologically to assess the sequential contributions of studies published over time.19 Finally, we assessed evidence of publication bias through visual inspection of funnel plots and Begg’s rank correlation test for asymmetry.20 21
Our initial search yielded a total of 4235 unique citations (fig 11).). After two rounds of reviews and searching citations of retained articles, we identified 131 studies as potentially relevant for analysis. We excluded studies of cardiovascular outcomes predefined as ineligible (such as chronic congestive heart failure or stable angina), non-atherothrombotic end points (such as arrhythmias), composite end points, or non-cardiovascular outcomes (such as cancer), and duplicate reports. This left 84 studies for our systematic review and meta-analysis. Table 11 provides details of the included studies.22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 Of these 84 studies, 34 (40%) reported on all-male cohorts, six (7%) reported on women only, and 44 (52%) included both men and women.
We evaluated two primary features of study quality—the number of years that participants were followed and adjustment for confounding. Duration of follow-up for study end points ranged from 2.5 to 35 years, with a mean follow-up of 11 years (standard deviation 6 years) (table 11).). Of the included studies, 13 (15%) had ≤5 years of follow-up. Similarly, studies varied in the degree of confounder adjustment, ranging from none to 18 variables, with a mean of six (SD 4). Most studies (68) presented adjusted estimates, but eight reported only unadjusted estimates and another eight adjusted only for basic demographic information. Methods of adjustment, effect measure, and confounding variables used in each study are presented in the appendix tables 1–5 on bmj.com for each of our primary outcomes.
For cardiovascular disease mortality and both end points for coronary heart disease, alcohol consumption was associated with lower risk, with relative risks of about 0.75 (table 22).). In general, relative risks derived from the more highly adjusted and from the less adjusted results were similar. Figures 2–4 reveal little visual evidence of heterogeneity despite statistical evidence of heterogeneity (P<0.001, I2=72.2%), probably driven by the large number of participants (>1 million). All the point estimates were <1.0 in studies, except for one study for cardiovascular disease mortality and two studies for coronary heart disease incidence and mortality.
In contrast, the overall associations of alcohol intake with stroke incidence and mortality were close to null, both in minimally adjusted and more highly adjusted models (table 22,, figs 55 and 66).). However, this null association seemed to obscure nearly significant but opposite associations with subtypes of incident stroke. Among the 12 studies on incident haemorrhagic stroke, the pooled relative risk for current alcohol drinkers compared with non-drinkers was 1.14 (95% confidence interval 0.97 to 1.34), whereas the eight studies on ischaemic stroke showed a moderate reduction in the pooled relative risk of 0.92 (0.85 to 1.00). Alcohol use was not associated with stroke mortality, but few studies assessed the risk of mortality from haemorrhagic or ischaemic stroke separately. Furthermore, only two studies reported relative risks on stroke end points for former drinkers compared with non-drinkers.
Analyses of the dose of alcohol consumed showed that 2.5–14.9 g alcohol (about ≤1 drink) per day was protective for all five outcomes compared with no alcohol (table 22).). For coronary heart disease outcomes, all levels of intake >2.5 g/day had similar degrees of risk reduction. For cardiovascular disease mortality as well as stroke incidence and mortality, the dose-response relations were less clear and more consistent with U or J shaped curves, suggesting an increased risk among drinkers of greater amounts of alcohol. Specifically, those who consumed >60 g/day were at a significantly increased risk of incident stroke compared with abstainers (relative risk 1.62 (1.32 to 1.98)).
In an analysis of differences in associations by sex, any amount of alcohol consumption relative to none was associated with greater reduction in cardiovascular disease mortality, stroke incidence, and stroke mortality for women than men. However, the association with stroke should be interpreted with caution, as the risk estimates for women are based on only three pooled studies. On the other hand, similar associations by sex were observed for coronary heart disease incidence and mortality (table 22).
Sensitivity analyses that were confined to only studies that controlled for the important confounders of smoking, age, and sex revealed generally similar results for all of the outcomes. Additional sensitivity analyses that account for the median number of confounding variables in the multivariable analyses of included studies revealed that those with fewer (less than the median) confounding variables generally reported slightly lower relative risk estimates. However, this pattern was inconsistent across the outcomes. Specifically, an increased risk of stroke mortality was observed for studies with limited adjustment for confounding. A similar trend was observed when considering the duration of follow-up. Using the pooled median number of years as the cut point, we found that studies with shorter follow-up reported a greater risk reduction for all outcomes except cardiovascular disease and coronary heart disease mortality (table 22).
Among those studies that used long term abstainers as the referent category, excluding former drinkers or evaluating them separately, the estimated association between drinking and both incidence and mortality estimates did not change substantively (table 22).). Among studies that evaluated former drinkers separately, the risk of death (from cardiovascular disease and coronary heart disease) was significantly higher in former drinkers than in drinkers. However, former drinkers did not have an increased risk of incident cardiovascular events (coronary heart disease or stroke).
Finally, a sensitivity analysis that excluded the few studies where only odds ratios instead of relative risks were presented had little effect on the results. In cumulative meta-analyses of cardiovascular disease and coronary heart disease outcomes (appendix figs 1–3 on bmj.com), there was little variation in the relative risk associated with alcohol consumption on cardiovascular disease mortality or incident coronary heart disease with addition of new studies after 1999; for coronary heart disease mortality, this plateau in incremental change from new studies occurred as early as 1992–3.
Of the 84 studies addressing alcohol and cardiovascular disease events, 31 also examined the association of alcohol consumption with all cause mortality. The pooled estimates from these studies showed a lower risk of all cause mortality for drinkers compared with non-drinkers (relative risk 0.87 (0.83 to 0.92)) (fig 77).). However, the association was J shaped, with the lowest risk for those consuming 2.5–14.9 g/day (relative risk 0.83 (0.80 to 0.86), 16 studies) and an elevated risk in those consuming >60 g/day (relative risk 1.30 (1.22 to 1.38), 8 studies).
Visual inspection of the funnel plot for each outcome did not show asymmetry, an indication that significant publication bias was not likely. This was further confirmed by a non-significant Begg’s test for each outcome (for cardiovascular disease mortality, P=0.40; incident coronary heart disease, P=0.75; coronary heart disease mortality, P=0.089; incident stroke, P=0.33; stroke mortality, P=0.59; all cause mortality, P=0.26).
In this review of 84 studies of alcohol consumption and cardiovascular disease, alcohol consumption at 2.5–14.9 g/day (about ≤1 drink a day) was consistently associated with a 14–25% reduction in the risk of all outcomes assessed compared with abstaining from alcohol. Such a reduction in risk is potentially of clinical importance, but consumption of larger amounts of alcohol was associated with higher risks for stroke incidence and mortality.
To our knowledge, this systematic review and meta-analysis is the most comprehensive to date. Although roughly similar estimates of lower risk were observed in previous meta-analyses of both coronary heart disease and stroke,1 2 3 4 5 6 7 8 our review extends the findings by assessing a broader array of relevant cardiovascular outcomes and adding several new important studies. Our review clarifies several discrepancies among prior reports. Corrao et al reported a J shaped relation between alcohol intake and coronary heart disease,2 whereas the review by Maclure described this relation as L shaped because he did not observe an increase in coronary heart disease risk associated with higher alcohol consumption.6 Our updated meta-analysis supports the latter association for coronary heart disease, with a 25–35% risk reduction for light to moderate drinking106 that also is present with heavier drinking.
Our analysis of multiple cardiovascular outcomes also shows the complexities inherent in the study of alcohol consumption. Modest alcohol intake was associated with lower stroke incidence and mortality, but the risk increased substantially with heavier drinking (that is, a J shaped relation). Furthermore, the association of alcohol consumption is complex and differs by stroke subtype, with a slightly lower risk of ischaemic stroke but higher risk of haemorrhagic stroke. These differential associations probably reflect the known antithrombotic effects of alcohol.107 Alcohol consumption, particularly at high doses, also seems to have an adverse association with blood pressure that may account, in part, for the higher risk of haemorrhagic stroke associated with heavier drinking.108 Additionally, our analysis does not consider other known detrimental effects of high alcohol consumption.3 Therefore, our findings lend further support for limits on alcohol consumption.106 109
Our review also highlights other important aspects of the relation between alcohol consumption and cardiovascular disease. Firstly, the lower risk of coronary heart disease associated with alcohol consumption was at least as strong for women as for men. Limited evidence suggests that the risk of stroke related to alcohol is lower for women than men, but this may only reflect lower alcohol intake among women. Secondly, inclusion of former drinkers did not seem to bias the association of alcohol consumption with cardiovascular disease. Thirdly, when studies were summarised chronologically, we found that the overall association between drinking and cardiovascular disease and coronary heart disease became apparent at least a decade ago, and ongoing studies have done little to revise the estimated associations.
From the extensive body of literature summarised here, the association between alcohol consumption and decreased cardiovascular risk is not in question, as additional research has not changed this conclusion. Rather, the lingering question is whether this association is causal. Clearly, observational studies cannot establish causation. However, when the present results are coupled with those from our companion review paper summarising interventional mechanistic studies focusing on biomarkers associated with cardiovascular disease,110 the argument for causation becomes more compelling. Indeed, the mechanistic biomarker review shows biological plausibility for a causal association by showing favourable changes in pathophysiologically relevant molecules.
Therefore, we can now examine the argument for causation based on Hill’s criteria.111 Beyond the biological plausibility argument discussed above, there is an appropriate temporal relation with alcohol use preventing cardiovascular disease. Secondly, we have observed a greater protective association with increasing dose, except that it seems to be offset somewhat by negative associations with the risk of haemorrhagic stroke. Thirdly, the protective association of alcohol has been consistently observed in diverse patient populations and in both women and men. Fourthly, the association is specific: moderate drinking (up to 1 drink or 12.5 g alcohol per day for women and 2 drinks or 25 g alcohol per day for men106) is associated with lower rates of cardiovascular disease but is not uniformly protective for other conditions, such as cancer.112 Lastly, the reduction in risk is notable even when controlling for known confounders (such as smoking, diet, and exercise). Any potential unmeasured confounder would need to be very strong to explain away the apparently protective association.
The results of our meta-analysis should be interpreted in context of the limitations of available data. Firstly, the quality of individual studies varied, with some studies having limited follow-up and limited adjustment for potential confounding. With respect to study follow-up, it is possible that misclassification of alcohol consumption may increase with study length because of changes in drinking habits over time. It is also possible that potential biological effects of alcohol vary with time of exposure. However, arguing against both these possibilities, the analysis stratified by length of follow-up did not show different associations between alcohol intake and outcome for shorter follow-up times versus longer times.
Secondly, only a limited subset of studies provided specific risk estimates for different beverages. Although there is great interest in differences between beer, wine, and spirits, alcoholic drinks generally have similar effects on high density lipoprotein cholesterol,113 and it is likely that any particular benefit of wine is prone to confounding by diet and socioeconomic status.114 115 None the less, this remains an interesting topic for further investigation.
Thirdly, we found only limited information on the relation between alcohol intake and mortality from subtypes of stroke, so this topic continues to be important for large observational cohort studies. Finally, we observed significant heterogeneity across studies for several of our pooled analyses. This may be due in great part to large study sample sizes, which can confer greater statistical power to heterogeneity tests, whereas the clinical relevance of this heterogeneity may be quite modest.10 Visual inspection of our various forest plots and the relative consistency of pooled relative risks across clinical and methodological variables suggest that there is considerable consistency in the relative risk findings across studies and across strata.
Given the consistency observed in our findings and compelling mechanistic data pointing to causation in our companion review, additional observational studies will have limited value except to elucidate more precisely the association of alcohol and stroke.116 Rather, debate should centre now on how to integrate this evidence into clinical practice and public health messages. In the realm of clinical practice, the evidence could form a foundation for proposing counselling for selected patients to incorporate moderate amounts of alcohol into their diets to improve their coronary heart disease risk. However, such a clinical strategy requires formal evaluation in pragmatic clinical trials that assess the questions of optimal patient selection, compliance, risks, and benefits. The focus of such trials would shift from assessing the association between alcohol and disease outcomes to evaluating the receptivity of both physicians and patients to the recommended consumption of alcohol for therapeutic purposes and the extent to which it can be successfully and safely implemented. In support of implementation trials, our two papers show that alcohol consumption in moderation has reproducible and plausible effects on markers of coronary heart disease risk.
With respect to public health messages, there may now be an impetus to better communicate to the public that alcohol, in moderation, may have overall health benefits that outweigh the risks in selected subsets of patients. Again, any such strategy would need to be accompanied by rigorous study and oversight of impacts. One approach would be to undertake public health messaging pilot studies on well defined target populations (such as a workplace or in a small jurisdiction) to permit detailed evaluation of effects on measures such as knowledge, attitudes, self reported drinking behaviours, and perhaps, secondarily, health outcomes.
The debate on how to integrate this evidence into clinical practice and public health messages will require integration of all possible effects of alcohol—from injury and violence to glucose metabolism and inflammation—and recognition that these effects may be distributed unequally across the population. For example, injury risk probably disproportionately affects younger individuals, whereas cardiovascular disease mainly affects older adults. Robust studies that examine multiple outcomes simultaneously are needed to identify those subsets of the population in which reduced cardiovascular risk might dominate against those for whom the risks of social and medical problems (including several cancers and injury112 117) are too great. Despite the latter concerns, results of our secondary analysis of overall mortality (fig 55)) support the notion that moderate alcohol consumption is associated with net benefit, at least in populations similar to those studied in the extant literature.
Our two systematic review papers summarise a surprisingly extensive body of literature on the relation between alcohol and cardiovascular disease. Our findings point to the need to define implications for clinical and public health practice. These reviews and the perspectives above provide a foundation for that dialogue.
Preliminary results from this manuscript were presented at the 32nd annual meeting of the Society of General Internal Medicine, Miami, Florida, 14 May 2009.
Contributors: All authors conceived the study and developed the protocol. PER and SEB conducted the search, abstracted the data for the analysis, and performed the statistical analysis. PER, SEB, and WAG wrote the first draft of the manuscript. All authors had access to the data, critically reviewed the manuscript for important intellectual content, and approved the final version of the manuscript. WAG will act as guarantor for the paper.
Funding: This work was supported by a contracted operating grant from Program of Research Integrating Substance Use Information into Mainstream Healthcare (PRISM) funded by the Robert Wood Johnson Foundation, project No 58529, with cofunding by the Substance Abuse and Mental Health Services and the Administration Center for Substance Abuse Treatment. PER is supported by a Frederick Banting and Charles Best Canada Graduate Scholarship from the Canadian Institutes of Health Research. SEB is supported by a Postdoctoral Fellowship Award from the Alberta Heritage Foundation for Medical Research. WAG is supported by a Canada Research Chair in Health Services Research and by a Senior Health Scholar Award from the Alberta Heritage Foundation for Medical Research. The study was conducted independently of funding agencies. None of the funding agencies played an active role in the preparation, review, or editing of this manuscript.
Competing interests: All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: support from the Robert Wood Johnson Foundation, the Substance Abuse and Mental Health Services, and the Administration Center for Substance Abuse Treatment (as detailed above) for the submitted work, no financial relationships with any organisations that might have an interest in the submitted work in the previous three years, no other relationships or activities that could appear to have influenced the submitted work.
Ethical approval: Not required.
Data sharing: Statistical code and datasets available from the corresponding author at firstname.lastname@example.org
Cite this as: BMJ 2011;342:d671