Evergreen A
Power Analysis for Testing Treatment-Biomarker Interaction in Two-Phase Design (306578)
Jianwen Cai, University of North Carolina at Chapel Hill*Poulami Maitra, GlaxoSmithKline
Xiaofei Wang, Duke University
Donglin Zeng, University of North Carolina at Chapel Hill
Keywords: Case-cohort design, Log-rank test, Interaction, Expensive biomarker, Cost-efficient
In the new age of cancer research, investigators are increasingly interested in discovering new biomarkers that can help predict the best treatment for a patient. For time to event data, finding new biomarkers is often time-consuming and costly, which might lead to the study being infeasible, especially when the event is rare. However, leveraging the abundant information and biospecimen collected in the completed clinical trials can lead to shortened discovery cycle. We consider a two-phase study design utilizing the already collected clinical data as the first phase data and the available biospecimen for measurement of potential biomarker as the second phase. We derive explicit form for power and sample size calculation for testing the interaction between a treatment and an expensive biomarker under the two-phase design. A formula for the bounds of the power is presented. Simulation studies were conducted to examine the efficiency of the test under the two-phase design. The cost efficiency of the two-phase design compared to a simple random sample design was also examined. We illustrate the use of the formula based on information from the pooled databases of LACE.