Long-Term Safety Evaluations in the Presence of Switching: Evaluation of Two Approaches

Evaluating the long-term safety of approved drugs for chronic indications is essential. This assessment ensures that the benefits continue to outweigh the risks beyond the follow-up observed in the clinical trials supporting market authorization. Consequently, these evaluations are mandated or recommended by multiple regulatory agencies and necessitate collecting and analyzing data from longitudinal cohorts in real-world settings post-market authorization. One challenge in analyzing and interpreting results using these sources is the complexity of long-term real-world drug utilization patterns in a competitive landscape, including switching between multiple drugs. Several methods have been developed to evaluate comparative long-term safety under real-world conditions. These methods include the experimental hierarchical approach, which extends the analytical follow-up period for the test drug to include the time after switching away, and the overlapping approach, which extends the follow-up period for both the test and comparator drugs to include the time after switching away. This paper uses multistate model methodology to consider initial and subsequent exposures in evaluating the estimators in these two methods. Our mathematical evaluations and simulations demonstrate that the estimators inflate the type-1-error across different switching and outcome incidence rate scenarios. Therefore, we propose a minor modification of the estimators to preserve the type-1-error. Currently used methods are simple but biased and lack clearly defined estimands. Methods based on multistate models may help identify and refine new estimands for evaluating long-term safety.