Few articles have evaluated the effects of non-response in cohort studies. The aim of this study was to assess if non-response, both to the baseline and to follow-up, may introduce selection bias on the association between occupational level and mortality. There were applied the non-response and mortality estimates of the Whitehall II study to a hypothetical cohort encompassing 110.000 person-years (occupational level: low - 50,000; high - 60,000). There were considered four scenarios: (1) full participation; (2) non-response to baseline; (3) non-response to follow-up; (4) non-response to baseline and follow-up. There were calculated mortality rate ratios (RR) considering the high occupation level as reference and absence of confounding or information bias. The first scenario RR was 1.79, with only the last scenario showing an important impact of the non-response on the rate ratio estimate (RR=1.2). The results suggest that combining non-response in different stages of cohort studies may cause selection bias.
cohort studies; refusal to participate; selection bias