ABSTRACT
This chapter summarizes the key threats to validity most encountered in observational EHR epidemiology. The chapter begins by delineating the difference between the target, source, and study populations, with reference to the catchment of the EHR. Then, the chapter discusses the five mechanisms whereby an exposure could be associated with an outcome in etiologic research: true association, random chance, bias, confounding, and reverse causation. The last four comprise threats to internal validity and are discussed in detail in the chapter. Afterward, issues of external validity are framed as the generalizability or transportability of the study results to the target populations. Beyond understanding how these threats to validity may arise in EHR-based research, quantitative bias analysis techniques are promoted as a formalized approach to estimating their impact on the research findings.
