'Adherence bias' in nutritional epidemiology

摘要

In 1980, the Coronary Drug Project (CDP) Research Group published data to warn against the practice of analyzing randomized trial treatment group data by adherence status for evidence of the efficacy of the treatment (1). Using data from the trial's placebo group (Table 1), they found that participants taking <80% of the per-protocol placebo dose (nonadherers) had a 5-y total mortality rate nearly 90% higher than that of participants taking 80% of the dose (adherers), a differential that could not be explained by the potential confounding variables (20) examined. Whereas this story is of obvious relevance to the clinical trialist, it is also a cautionary tale for the nutritional epidemiologist. In any sample of persons enumerated as part of a cohort study, there will be a percentage of persons who would be nonadherers were they participants in a randomized trial. Their presence could lead to substantial confounding if nonadherer status is associated with the exposure of interest.