Abstract:
|
Identifying treatment responders and non-responders in clinical trials via predictive signatures using baseline biomarkers and demographic/clinical variables is a first critical step towards precision medicine. Recent advances in statistical algorithms (tree-based methods, simple multiplicative/additive rule-based regression methods, composite score methods, etc.) have enabled researchers to retrospectively analyze the clinical trial data and identify a subgroup of patients that will benefit from a candidate drug. However, due to differences in the clinical trial designs, and the underlying biological and mechanistic relationships between the biomarkers and clinical outcome of interest, there is rarely a single algorithm that is optimal for all scenarios. In this talk, we summarize the results from simulation studies to investigate the performance of each algorithm under different scenarios, and provide recommendations on the future strategy for exploratory subgroup identification in precision medicine.
|