Identifying Outlier Centers in Multi-center Clinical Trials: A Hierarchical Bayesian Model Approach

The conduction of clinical trials for new drug development requires extensive quality control (QC). As a part of this QC, data-driven identification of errors and misconduct in clinical trial centers often performed. There is a substantial need for unified statistical framework for such statistical inspection of clinical trial data, that should take into account data-generation process, allow to utilize prior knowledge, and allow for flexible accommodation to data amount, specifics or structure. Desired statistical methodology should be robust to noisy data and should generate interpretable scores, that can be put into action. The Bayesian hierarchical robust (robit) modeling with mixed posterior predictive and prior predictive tail p-values satisfies all the denoted requirements. The reasoning of selection of statistical procedure is provided in this work, as well as investigation of some motivating examples and comparison with more traditional random effects hierarchical (meta-analysis) modeling.