All Times EDT
Keywords: Bayesian Hierarchical Model, Subgroup Analysis, Drug Trial Snapshot
The conventional subgroup analysis is often inadequate to address heterogeneous treatment effects across subgroups due to limited data in each subgroup. The total variability in the sample estimates from the conventional subgroup analysis is the sum of the within subgroup variability of sample estimator and the across subgroups variability in underlying/true parameter values. As such, the sample estimates are susceptible to random highs and random lows.
This presentation will focus on using Bayesian hierarchical model (BHM) to derive shrinkage estimates of subgroup treatment effects. In the BHM approach, the conventional subgroup analysis was conducted first, then the estimated sample means, and standard errors were used in the hierarchical modeling by assuming the subgroup sample means are random samples of a normal distribution. For a given subgroup, information from other subgroups is also used to estimate its treatment effect. Outcomes from all subjects are relevant, with an outcome from a subject in the given subgroup more relevant than the outcome of a subject not in the given subgroup. A shrinkage estimate of a treatment effect for a subgroup is a “weighted” average of sample estimate and overall estimate. The weights are based on the ratio of the between subgroup variability to the within subgroup variability. The greater that ratio the smaller the weight on the overall estimate (the less the shrinkage). Shrinkage estimates of subgroup treatment effects use more information. Therefore, they are more precise, closer to the true subgroup treatment effects than the sample estimates.
Several cases studies with published shrinkage estimate of subgroup treatment effects in drug trial snapshot (DTS) for approved new molecular entities will be used to illustrate the application of BHM in regulatory decision. The case studies will cover continuous, dichotomous, count or time to event endpoints, and different subgroups, such as region, race, disease subtypes etc.