Bayesian dynamic power prior borrowing for augmenting a control arm for survival analysis.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Real-World Evidence

Background:

• Real-world data (RWD) can improve clinical trial efficiency by augmenting internal control arms.
• However, heterogeneity between RWD and internal controls can introduce bias.
• Existing Bayesian dynamic borrowing methods address bias for binary/continuous outcomes but not survival data.

Purpose of the Study:

• To extend Bayesian dynamic borrowing to survival analysis for estimating hazard ratios.
• To propose a novel method for quantifying borrowing based on empirical Bayes and log-hazard ratios.
• To develop a robust inference framework incorporating covariate adjustment and multiple imputation.

Main Methods:

• Developed an empirical Bayes method to estimate borrowing strength using log-hazard ratios between external and internal controls.
• Employed Bayesian bootstrap, covariate adjustment, and multiple imputation for comprehensive uncertainty quantification.
• Validated the approach through simulation studies and applied it to a real-world oncology dataset (CheckMate-057).

Main Results:

• The proposed method effectively incorporates external real-world data into survival analysis.
• Demonstrated robust performance in mitigating bias and accounting for uncertainty.
• Successfully applied to advanced non-squamous non-small cell lung cancer data, illustrating practical utility.

Conclusions:

• The novel Bayesian dynamic borrowing approach is suitable for survival analysis in clinical trials.
• This method enhances trial efficiency and reliability when using real-world data.
• The methodology is adaptable for various oncological endpoints and other disease areas.