Statistical control in research on alcohol and tobacco: an example from research on alcohol and mortality.

Many of the major studies of deaths related to alcohol use report evidence for a U-shaped relationship between mortality and consumption, with abstainers having higher mortality than moderate drinkers. Some have suggested that this indicates a protective effect of moderate drinking; others have argued that pre-existing health problems and demographic and lifestyle differences explain the finding. In this report, we present evidence that the U-shaped function is largely an artefact of inadequate controls for smoking and drinking intake and history. Never smoking abstainers die at about the same rate as never smoking moderate drinkers. We stress the importance of using precise measures of drug use history and drug use status and make the point that, given the strong interaction between smoking and drinking, controlling for smoking may produce misleading results.