OBJECTIVES: Meta-analyses enhance precision in treatment effect estimates by aggregating data from multiple studies. Although well-established for continuous and binary outcomes, research on time-to-event outcomes often relies on Cox proportional hazard models. By integrating prior information and updating findings sequentially, Bayesian analysis provides a robust framework for meta-analyzing time-to-event outcomes using individual patient data (IPD). Unlike the Frequentist approach, the Bayesian method offers flexibility in interpreting treatment effects by incorporating multiple distributional assumptions without sample size constraints. This study aimed to meta-analyze data from two clinical trials in metastatic breast cancer patients using a Bayesian parametric survival model.

METHODS: The overall survival (OS) data was meta-analyzed using Bayesian parametric survival models in Stata. The "bayes: streg" command was used to fit models, incorporating clinical and statistical covariates from the trials. The Random-walk Metropolis-Hastings algorithm was used for efficient sampling using default normal priors. The pooled hazard ratio (HR) was estimated via a two-stage meta-analysis using Bayesian exponential and Weibull distributions.

RESULTS: Results from both statistical distributions were consistent. The Weibull model was preferred due to its superior goodness of fit statistic (i.e., Akaike Information Criterion; 2,124 vs 2,207 for Weibull and Exponential, respectively) and visual inspection of the fitted curve. The intervention of interest showed a statistically significant improvement in overall survival compared to the comparator (HR 0.74, p < 0.001) in the overall population after adjusting for covariates. These results were consistent with Frequentist meta-analyses using Cox models.

CONCLUSIONS: With IPD, the Bayesian approach is more reliable due to its ability to incorporate prior information. The Bayesian parametric survival model yielded robust and consistent treatment effect estimates, aligning with frequentist methods. This approach confirmed traditional Cox model findings and demonstrated the potential of Bayesian methods to enhance the precision and reliability of survival analyses in clinical research.