Detecting rare adverse events in post-marketing studies: Sample size considerations

Identifying the causal relationships between drugs and rare, possibly serious adverse events is an increasingly important issue in post-marketing studies. The observational cohort study often is the design used to detect the causal relationships when it is unfeasible to conduct prospective, randomized clinical trials. The single-group study has the disadvantage of lacking a comparator group, and the two-group prospective study requires a prohibitively large sample size. A class of efficient hybrid designs is proposed and described in this dissertation that uses external data such as established databases and/or pre-NDA data. The proposed designs are intended to reduce the sample size requirements while maintaining the advantages of two-group designs, and enable safety decisions to be reached more quickly.; New sample size formulae based on the Poisson distribution are developed using both approximate and exact methods for the cases in which the incidence of certain adverse events for subjects treated with the compound is compared to an external control cohort not receiving the compound under study. The sample size reduction with the incorporation of an external control cohort is substantial compared to two-group study designs in which both groups are accrued concurrently and prospectively. The performance of both methods is compared. The results indicate that the exact test provides a more precise estimation and shows an improvement over the approximate method.; An approximate sample size formula is also developed for the design with the incorporation of pre-NDA data and external control cohort. The derivation is based on the random-effects model with compound Poisson-Gamma distribution. The extra variation, due to the inherent randomness of the data from multiple studies, outweighs the contribution of pre-NDA data in some situations. The design is generally recommended when the effect size is moderate (relative risk >2) and the inter-study heterogeneity is small to moderate. The simulation results indicate that the approximate formula is able to provide a satisfactory estimation in terms of type I error and power when the relative size of the PMS-treated cohort to the external control cohort is ≤1 and the sizes of the two populations do not differ greatly.