Analysis of Adverse Events in the Presence of Discontinuations

Abstract

Commonly, the occurrence of adverse events in clinical studies is summarized in terms of incidences. This approach has several drawbacks, such as the lack of reference to observation time and absence of information about discontinuations. If survival analysis methodology is used instead, it is rarely considered that the time to an adverse event and the time to discontinuation might not follow a random censorship model. In this situation, classical statistical methods like Kaplan-Meier estimates or log-rank tests are no longer appropriate. In this article, I investigate the applicability of several methods that account for a potential dependence between event times and discontinuation times on the analysis of adverse events.