|
Abstract:
|
Identifying subgroups, which respond differently to a treatment, both in terms of efficacy and safety, is an important part of drug development. In early phase exploratory clinical trials the well-known challenges of subgroup analyses are further amplified by a low sample size. We evaluate novel strategies to improve the common strategies for subgroup identification and treatment effect estimation in these settings in realistic simulation scenarios. We compare two subgroup identification strategies, one based on categorization of continuous covariates, one based on including continuous covariates and making use of splines. We then compare the sampling properties of several effect estimation approaches in subgroups identified with these two strategies, employing model averaging, resampling and lasso regression methods. Additionally we evaluate the influence of several parameters on the different estimators to further investigate their characteristics. Our results confirm that the naive estimates in subgroups suffer from selection bias and show that there are estimation methods available, which give much better estimates of the treatment effect in all considered scenarios.
|
