The difference in restricted mean survival time (dRMST) at a specific time point is an appropriate measure to quantify the treatment effect between two arms in randomized clinical trials when the proportional hazards assumption does not hold. This is common in the context of immuno-oncology therapies. Several frequentist methods exist to estimate RMST based on modeling and integrating the survival function.
A more natural approach is to consider a regression model on the RMST directly using pseudo-observations which allows for a direct fit without modeling the survival function. Only two Bayesian methods exist, and both model the survival function with a nonparametric prior process.
We develop a new Bayesian method based on pseudo-observations and the generalized method of moments (GMM) that offers RMST estimation adjusted on covariates without the need to model the survival function, making it attractive compared to existing Bayesian methods.
A simulation study of 2-arm randomized clinical trials with different time-dependent treatment effects and covariate effects was conducted, demonstrating that this new approach yields valid results, consistent with existing methods and shows improved precision after covariate adjustment. For illustration, the methods were applied to analyze the PSA progression-free survival of the Getug-AFU 15 phase 3 trial in non-castrate metastatic prostate cancer.
In conclusion, we propose a new Bayesian survival method to provide a straightforward RMST estimation adjusted on covariates without the specifying the survival function.
Key words: Bayesian Survival Analysis; Restricted Mean Survival Time; Pseudo-observations; Generalized method of moments
Léa ORSINI is the winner of the first QuanTIM Webinar's PhD Student award.
