Identifying High-Risk Subgroups in Meta-Analyses with Hierarchical Bayesian Sparse Modeling

Music City Center 

To characterize the safety of new drugs, researchers rely on meta-analysis methods to assess the incidence of adverse events across studies. However, the risks of different adverse events vary by disease type and treatment. There is a clinical interest in identifying high-risk groups for closer toxicity monitoring, but this effort is hindered by sparsely observed outcomes and a high number of potential risk factors. Traditional meta-analysis methods, such as fixed and random effects models, fail to address this issue and often yield biased estimates for rare events. We frame the problem as a Bayesian variable selection approach to identify high-risk groups, utilizing shrinkage priors to address sparsity in linking adverse event probabilities to study-level covariates. Specifically, we propose a Bayesian model that selects both main and interaction effects using a hierarchical horseshoe prior. We illustrate the performance of our method using both simulated data and an application to meta-analysis of immunotherapy drugs.

Bayesian feature selection

Meta-analyses

Horseshoe prior

Categorical data analysis 

Section on Bayesian Statistical Science