Treatment Response Subgroup Identification Using Penalized Regression Spline and Mixture Models

• Activity Number: 659
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Hongjie Zhu* and Lin Wang and Lynn Wei and Hui Quan
• Companies: Sanofi and Sanofi and Sanofi and Sanofi
• Keywords: biomarkers ; subgroup identification ; clinical trials ; regression spline ; mixture model ; precision medicine

Abstract

The mechanisms of action of biologic agents sometimes suggest that baseline biomarkers can be related to the treatment outcomes. Objectives of clinical trials may include analyzing biomarkers for their potential value to predict treatment responses, and identifying response subgroups based on predictive biomarkers if such biomarkers exist. Practical considerations for designing approaches include the needs to 1) analyze different types of primary and secondary outcomes, 2) capture potentially non-linear effects of the biomarkers while preventing over-fitting the models, 3) allow co-existence of both prognostic and predictive biomarkers, 4) handle multiple treatment groups, and 5) adjust for several covariates. This talk will discuss two approaches customized to achieve the objectives while meeting all the requirements: a penalized regression spline-based approach directly modeling biomarkers' effects to treatment outcomes, and a mixture of regression model using biomarkers to identify latent classes of patients with differential response patterns. Empirical powers of the approaches were evaluated through simulations.