ABSTRACT
If population-based disease registries, for example, cancer registries, and registries of the population are available, enabling a complete follow-up of the cohort, a cohort study is unlikely to be affected by selection bias, because participation in the study is determined before the observation period, that is, before any disease occurrence. Therefore, the remainder of this chapter focuses on the case-control design, although selection bias is also relevant for cross-sectional studies (see Chapter 2) that are often applied to investigate the association between symptom prevalence and electromagnetic field (EMF) exposure (see Chapter 16). In this chapter, reverse causality and detection bias are also briefly discussed from the perspective of selection bias. Additional considerations on reverse causality are made in Chapter 6 from the perspective of confounding.
