Bivariate joint modelling for mixed continuous and binary responses in the presence of non-monotone, non-ignorable missingness: The case of prostate cancer

Abstract

AbstractJoint modelling for mixed longitudinal responses has played a prominent part in disease decision-making. It is based on a joint strategy of estimating joint likelihood with shared random effects. Non-ignorable missingness in outcomes increases complexity in joint model; a shared parameter model is proposed to incorporate non-ignorable missing data for joint modelling of longitudinal responses and missing data mechanism. Parameters are estimated under the Bayesian paradigm and implemented via Markov chain Monte Carlo (MCMC) methods with Gibbs sampler. To demonstrate the effectiveness of the proposed method, the joint model is applied to analyze a prostate cancer dataset. The objective is to assess whether there is an association between two mixed longitudinal biomarkers, which could have important implications for understanding disease progression and guiding treatment decisions. The dataset contains non-monotone missingness pattern. To evaluate the performance and robustness of the proposed joint model, simulation studies are conducted.