Online Program

Guidance for Bayesian Analyses of Heterogeneity of Treatment Effect

*Carlos O. Weiss, Michigan State University 
Thomas A. Louis, Johns Hopkins Bloomberg School of Public Health 
Chenguang Wang, Johns Hopkins University  
Ravi Varadhan, Department of Oncology, Johns Hopkins University 

Keywords: Heterogeneity of treatment effect, Bayesian, patient-centered outcomes research

Assessment of HTE is essential in patient-centered outcomes research. While the Bayesian approach is arguably more suited to study HTE than the frequentist approach, methods guidance is lacking. Based on feedback from an expert panel of statisticians, we developed guidance for modeling HTE with Bayesian regression models: 1) State the goal of the Bayesian HTE analysis clearly; 2) Use evidence from external studies whenever available as the basis for developing a prior distribution of key parameters; 3) Pay careful attention to the six major components of the methodology for Bayesian HTE analysis: (i) model specification, (ii) prior specification, (iii) estimation and convergence diagnostics, (iv) model evaluation, (v) posterior summarization and (vi) sensitivity analysis; 4) Report results with transparency and reproducibility including goals of the HTE analysis, aspects of the analytic plan were chosen before the data was collected, how external evidence was considered or used, which models were examined, computational parameters and convergence diagnostics, aspects of model evaluation, posterior summaries, and sensitivity to varying models and prior distributions.