A Bayesian approach for instrumental variable analysis with censored time-to-event outcome

Area of Science:

• Biostatistics
• Epidemiology
• Econometrics

Background:

• Instrumental variable (IV) analysis is crucial for estimating causal effects when unobserved confounders or measurement errors are present.
• Existing IV methods are limited for time-to-event outcomes with censored data, necessitating new approaches.

Purpose of the Study:

• To develop a novel Bayesian instrumental variable (IV) approach for causal inference with time-to-event outcomes and censored data.
• To address limitations in current IV methodologies for survival analysis and complex observational data.

Main Methods:

• A two-stage linear model framework is employed within a Bayesian setting.
• Markov chain Monte Carlo (MCMC) sampling is utilized for parameter estimation, accommodating various error distributions.
• The method is validated through simulation studies and applied to real-world cohort data.

Main Results:

• The proposed Bayesian IV method significantly reduces bias in causal effect estimation.
• Coverage probabilities for the estimated causal effects are substantially improved compared to methods ignoring unobserved factors.
• The approach demonstrates robust performance across different model specifications.

Conclusions:

• The developed Bayesian IV method offers a powerful tool for causal inference in observational studies with censored time-to-event data.
• This methodology enhances the reliability of estimating treatment effects in the presence of unobserved confounding and measurement error.
• The application to large cohort studies highlights its practical utility in epidemiological and health research.