The assessment of safety is an important aspect of the evaluation of new therapies in clinical trials, with estimation of adverse event risk being an essential part of this. Standard methods for this such as the incidence proportion, that is, the number of patients with a specific adverse event out of all patients in the treatment groups, does not account for varying follow-up times between arms and competing risks. Similarly, it is well-known that simply using the 1 – Kaplan-Meier estimator to estimate AE risk in the presence of censoring overestimates that probability. In this talk we will introduce estimands (cause-specific hazard, event-free survival, and cumulative incidence function) in the presence of competing risk and discuss how to estimate them. Based on the work done within the SAVVY (Survival analysis forAdVerse events with VarYing follow-up times) project, an academia – pharma collaboration to which Roche contributed data of three RCTs, we will empirically illustrate how large biases of various estimators can become for estimation of AE probabilities in one arm and comparison of AE risk between two arms in a RCT.
Papers and code are linked on the SAVVY webpage.