The STEPP Approach to Assessing Treatment-Effect Heterogeneity for Relative and Absolute Endpoints

• Activity Number: 335
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ann Lazar*+ and Marco Bonetti and Bernard F. Cole and Wai-ki Yip and Richard D. Gelber
• Companies: University of California, San Francisco and Bocconi University and University of Vermont and Harvard School of Public Health and Harvard School of Public Health/Dana-Farber Cancer Institute
• Keywords: Competing risks ; Biomarkers ; Interaction ; Permutation-based inference ; Personalized medicine ; Overview studies

Abstract

Investigators of modern clinical trials often collect baseline data to explore whether treatment differences depend on patient characteristics or covariates (i.e., heterogeneity of treatment effects or HTE). It has been suggested that HTE detection may depend on the type of endpoint (e.g., absolute vs. relative) and test statistic. An HTE method known as Subpopulation Treatment Effect Pattern Plots (STEPP), a graphical method that constructs overlapping subgroups of patients as a function of continuous covariates, has been extended to include a variety of absolute and relative survival and competing risks endpoints, including the subdistribution hazard of a competing risk based on our proposed generalization of the "observed minus expected" estimation. The performance of STEPP was assessed through simulation for these endpoints. For small sample sizes, the tests did not adequately recover the alpha level. We therefore propose an alternative STEPP approach that balances subgroups based on the number of events, and through simulation we found that the alpha level of the tests were recovered accurately even for small sample sizes regardless of the type of endpoint.