A1997
Title: Subpopulation treatment effect pattern plot (STEPP): A review and a Bayesian extension for binary outcomes
Authors: Marco Bonetti - Bocconi University (Italy) [presenting]
Luca Benetti - University Bocconi (Italy)
Abstract: An overview of the STEPP methodology is provided, along with a new Bayesian extension for the specific case of binary outcomes. The focus is on the setting in which overlapping subpopulations are constructed according to a sliding window pattern, which has proven particularly useful in STEPP applications. The Bayesian approach employs a beta-binomial model to estimate the posterior distribution of subpopulation-specific treatment effects. The model assumes that overlapping intervals represent a mixture of parameters from adjacent non-overlapping intervals, with a uniform prior over the mixture weights. Estimation is performed using Metropolis within Gibbs sampling. The novel methodology is described and implemented, with application in the binary-outcome setting, and performance and sensitivity analyses regarding subgroup structure and the assumed dependence model are presented.
