Keywords: Copula models, Bayesian
Bayesian copula models provide a flexible, interpretable framework to jointly model multivariate benefit and risk outcomes by separating the specification of the marginal outcome models from the copula model for dependence between outcomes. Several recent papers have examined joint Bayesian copula models in a clinical trial context. For example, Cunanan and Koopmeiners (2014) explore the ability of these models to identify the correct dose in phase I-II clinical trials and Costa and Drury (2018) investigate the ability of a copula model to find a dose with high posterior probability of achieving a specified benefit-risk profile. However, to the best of our knowledge, none of these recent papers have directly examined model accuracy.
In this poster, simulation studies are used to explore the calibration of both marginal and copula parameters estimated under the joint model as a function of continuous/discrete outcome type, correlation between outcomes, prior specification, and sample size. Because many confirmatory studies are designed primarily to estimate efficacy, with safety relegated to a secondary role, an additional aim is to determine circumstances under which estimates of the safety parameters can be improved by borrowing information from a related efficacy outcome.