Bayesian analysis of the causal reference-based model for missing data in clinical trials.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Causal Inference

Background:

• Missing data in clinical trials, often due to intercurrent events (ICEs), complicates statistical analysis.
• Reference-based imputation methods are common for handling ICEs but have limitations in variance estimation.
• Existing methods like Rubin's rules can be biased, and repeated sampling estimators may decrease with increasing ICEs.

Purpose of the Study:

• To propose a novel Bayesian reference-based causal model to address uncertainty in estimating treatment effects after ICEs.
• To provide a method that explicitly reflects uncertainty about the maintained treatment effect post-ICE.
• To estimate the treatment policy treatment effect in trials with limited post-ICE data.

Main Methods:

• Developed a Bayesian framework building upon the causal model of White et al. (2019).
• Introduced a prior distribution for the maintained effect parameter to quantify uncertainty.
• Employed simulations and an anti-depressant trial for validation and comparison.

Main Results:

• The proposed Bayesian approach offers inference that directly incorporates uncertainty regarding the maintained treatment effect.
• Demonstrated the ability to estimate treatment policy effects even with minimal post-ICE data.
• Simulations compared the frequentist properties of the new method against existing reference-based techniques.

Conclusions:

• The Bayesian reference-based causal model provides a robust method for handling missing data due to ICEs in clinical trials.
• This approach enhances the estimation of treatment effects by acknowledging and quantifying uncertainty.
• Applicable to trials with limited post-ICE data, offering improved statistical rigor.