[Selection effects in epidemiological cohorts: nature, causes and consequences]

Abstract

Selection effects in cohort studies occur when the population observed over time differs from the target population due to the inclusion or follow-up of the subjects. Selection effects may bias the estimation of the frequency of the disease and of the exposure of interest, or the estimation of the association between disease and exposure. In industrial cohorts, where subjects are not volunteers, the reasons why they were hired and they stay working in a company are often linked with lifestyle factors or health problems that can be associated with the diseases or the exposures of interest. This phenomenon is known as the "healthy worker effect". When the cohort includes volunteers, other selection effects associated with the reasons why the individuals participate in the study and stay in the cohort may occur. The time course of both exposure of interest and health status during follow-up may also be associated with the probability of staying in the cohort. We describe the main components of the various selection effects associated with inclusion of the subjects at baseline, with their follow-up during the observation of the cohort, and we examine the main components of the selection effects and their consequences.