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

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Cluster-randomized trials (CRTs) with fragile populations often face complex attrition.
• Missing outcome data in CRTs can be heterogeneous, including participants with known survival, unknown survival, or those who have died.
• Existing methods struggle to jointly handle diverse missing data mechanisms and unknown survival status.

Purpose of the Study:

• To propose a novel Bayesian framework for estimating survivor average causal effects in CRTs.
• To develop a method that accounts for complex and heterogeneous missing data, including unknown survival status.
• To provide a generalizable approach for handling missingness in aging and palliative care research.

Main Methods:

• A Bayesian framework utilizing a multivariate outcome to jointly estimate causal effects.
• Distinguishing between individual-level and cluster-level survivor average causal effects in posterior estimates.
• Simulation studies to assess model performance across various missing data scenarios.

Main Results:

• The proposed Bayesian model demonstrated low bias and high coverage for key parameters in simulations.
• The framework successfully handles complex missing data, including unknown survival status and dropouts unrelated to mortality.
• The model was illustrated using data from a geriatric CRT, showing its practical applicability.

Conclusions:

• The developed Bayesian framework offers a robust solution for complex missing data in CRTs.
• This approach is particularly valuable for studies involving aging and palliative care populations.
• The methodology can be extended to various outcome types and multiple endpoints.