Power Analysis and Sample Size Calculation of Statistical Interaction Model for Evaluation of Biomarker Predictive Effect

• Activity Number: 658
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Dung-Tsa Chen* and Hui-Yi Lin and Po-Yu Huang
• Companies: Moffitt Cancer Center and Moffitt Cancer Center and National Chung Hsing University
• Keywords: Predictive biomarker ; power analysis ; sample size calculation

Abstract

Predictive biomarkers have been proved as useful clinical utility to help physicians determine which treatment could improve clinical outcomes in specific subgroup of patients, such as the use of EGFR mutation status to guide non-small-cell lung cancer patients for the gefitinib or erlotinib treatment. From the statistical point of view, the predictive biomarker model can be translated into the statistical interaction effect model. The interaction model includes two variables, biomarker variable (positive/negative) and treatment variable (treatment/control), and the interaction term of the two variables. The key component of the interaction model is the interaction term which allows to assess the differential treatment effect when the biomarker status changes. A significant differential treatment effect will indicate the predictive value of the biomarker in predicting who will likely improve their clinical outcomes by receiving the treatment. In this study, we will evaluate various interaction effect models to formulate appropriate hypothesis of the biomarker predictive effect for power analysis and sample size calculation in survival outcome data.