Alcohol and coronary heart disease: a meta-analysis

Objective. To estimate parameters of the function relating alcohol consumpt ion with the risk of coronary heart disease and to identify the sources of heterogeneity in the parameter estimates. Methods. A search of the epidemio logical literature from 1966 to 1998 was performed using several bibliograp hic databases. Meta-regression models were fitted to evaluate non-linear ef fects of alcohol intake on the relative risk. The effects of some character istics of the studies, including an index of their quality, were considered as putative sources of heterogeneity of the estimates. Publication bias wa s also investigated. Findings. Among the 196 initially reviewed articles, 5 1 were selected. Since qualitative characteristics of the studies were sign ificant sources of heterogeneity, the pooled dose-response functions were b ased on the 28 cohort studies with higher quality. Risk decreased from 0 to 20 g/day (RR = 0.80; 95% CI: 0.78, 0.83); there was evidence of a protecti ve effect up to 72 g/day (RR = 0.96; 95% CI: 0.92, 1.00) and increased risk above greater than or equal to 89 g/day (RR = 1.05; 95% CI: 1.00, 1.11). L ower protective effects and harmful effects were found in women, in men liv ing in countries outside the Mediterranean area and in studies where fatal events were used as the outcome. Evidence of publication bias for moderate intakes and of heterogeneity of the estimates across studies for higher int akes were found. Conclusions. The degree of protection from moderate doses of alcohol should be reconsidered. Further research investigating the effec t of drinking patterns on the risk of coronary heart disease should be perf ormed. Caution in making general recommendations is needed.