Adjusting for intercurrent events using Bayesian joint models for longitudinal outcomes in clinical trials

Abstract

Background:

Intercurrent events in clinical trials can disrupt the interpretation and/or measurement of clinical endpoints. This article focuses on terminal intercurrent events that preclude complete measurement of a longitudinal outcome. When such events are related to the underlying outcome, particularly for physical signs, analyses based only on the available measurements can yield biased estimates. Consequently, a principled methodology is needed to effectively handle these intercurrent events.

Methods:

We propose a Bayesian joint modeling approach to account for terminal intercurrent events. Our model jointly analyzes longitudinal outcomes and terminal events using shared random effects. We employ multiple discrete-time survival submodels to accommodate different event types and evaluate operating characteristics through extensive simulations that resemble a clinical trial with recovery and death as competing events.

Results:

The proposed Bayesian joint modeling strategy demonstrates higher power than models that do not account for intercurrent events. Specifically, power increases by approximately 15% when bias due to intercurrent events is substantial and can be reduced by joint modeling.

Conclusion:

Our Bayesian joint modeling approach effectively addresses terminal intercurrent events in both the design and analysis phases of clinical trials. By explicitly accounting for event-related truncation of longitudinal follow-up, it improves the precision and reliability of treatment effect estimation when outcome measurement is incomplete.