Bayesian Inference for Cluster-Randomized Trials With Multivariate Outcomes Subject to Both Truncation by Death and Missingness.

Cluster-randomized trials (CRTs) on fragile populations frequently encounter complex attrition problems where the reasons for missing outcomes can be heterogeneous, with participants who are known alive, known to have died, or with unknown survival status, and with complex and distinct missing data mechanisms for each group. Although existing methods have been developed to address death truncation in CRTs, no existing methods can jointly accommodate participants who drop out for reasons unrelated to mortality or serious illnesses, or those with an unknown survival status. This paper proposes a Bayesian framework for estimating survivor average causal effects in CRTs while accounting for different types of incomplete data. Our approach uses a multivariate outcome that jointly estimates the causal effects, and in the posterior estimates, we distinguish the individual-level and the cluster-level survivor average causal effect. We perform simulation studies to evaluate the performance of our model and found low bias and high coverage on key parameters across several different scenarios. We use data from a geriatric CRT to illustrate the use of our model. Although our illustration focuses on the case of a bivariate continuous outcome, our model is straightforwardly extended to accommodate more than two endpoints as well as other types of endpoints (e.g., binary). Thus, this work provides a general modeling framework for handling complex missingness in CRTs and can be applied to a wide range of settings with aging and palliative care populations.