Commentary: Decreasing response rates require investigators to quantify and report the impact of selection bias in case-control studies

Abstract

A high response rate has long been regarded as one of the hallmarks of a ‘good’ epidemiological study. However, the validity of results from studies that require participants to complete questionnaires, have an interview or provide a biological specimen is becoming increasingly jeopardized by decreasing participation. One analysis showed that cooperation rates (participants interviewed/number eligible that were contacted) in published population-based case–control studies declined by 3.33% per year in cases and 5.15% per year in controls from 1991 to 2003. In such studies, cooperation rates of 70% and response rates [participants interviewed/(participants interviewedþ eligible non-participantsþ people of presumed but unconfirmed eligibility)] of 50% are not uncommon today. Furthermore, and maybe because of declining rates, only a minority of studies published sufficient information for either cooperation or response rates to be calculated. A fundamental tenet of a case–control study design is that the controls should be selected randomly, and independently of exposure status, from the source population from which the cases were drawn and thus be representative of the exposure experience of the source population. Poor response rates by themselves do not create selection bias any more than high response rates guarantee unbiased estimates, but by endangering representativeness, low participation can provide more opportunity for bias to occur. Selection bias develops if the selection probabilities are different for cases and controls based on their exposure status. Since odds ratios maybe biased due to low participation, should we discard case–control studies as a valid study design?