Why do only some cohort studies find health benefits from low volume alcohol use? A systematic review and meta-analysis of study characteristics that may bias mortality risk estimates.

Assumptions about alcohol's health benefits profoundly influence global disease burden estimates and drinking guidelines. Utilising theory and evidence, we identify and test study characteristics that may bias estimates of all-cause mortality risk associated with low volume drinking. We identified 107 longitudinal studies by systematic review with 724 estimates of association between alcohol consumption and all-cause mortality for 4,838,825 participants with 425,564 recorded deaths. "Higher quality" studies had a mean cohort age of ≤55 years, followed-up beyond 55 years, and excluded former and occasional drinkers from abstainer reference groups. "Low volume" alcohol use was defined as between one drink per week (>1.30g ethanol/day) and two drinks per day (<25g ethanol/day). Mixed linear regression was used to model relative risks (RRs) of mortality for subgroups of higher versus lower quality studies. As predicted, studies with younger cohorts and separating former and occasional drinkers from abstainers estimated similar mortality risk for low volume drinkers (RR=0.98, 0.87-1.11) as abstainers. Studies not meeting these quality criteria estimated significantly lower risk for low volume drinkers (RR=0.84, 0.79-0.89). In exploratory analyses, studies controlling for smoking and/or socio-economic status had significantly reduced mortality risks for low volume drinkers. However, mean RR estimates for low volume drinkers in non-smoking cohorts were above 1.0 (RR=1.16, 0.91-1.41). Studies with lifetime selection biases may create misleading positive health associations. These biases pervade the field of alcohol epidemiology and can confuse communications about health risks. Future research should investigate whether smoking status mediates, moderates or confounds alcohol-mortality risk relationships.