BSET (Bayesian Surrogate Evaluation Test) is an
R package for assessing the validity of surrogate markers
in clinical trials. It provides hypothesis testing tools to evaluate
whether a surrogate can reliably estimate the causal effect of a
treatment on a primary outcome. The package implements the
imputation-based Bayesian methodology of Carlotti and Parast (2026),
extending the frequentist rank-based approach of Parast et al. (2024).
BSET addresses key limitations of the frequentist method, including the
lack of causal interpretability and the inability to adjust for
covariates in the estimation process.
The package supports Bayesian testing both with and without baseline covariates. Additionally, it includes comprehensive simulation suites to replicate studies from both papers, enabling performance comparisons between Bayesian and frequentist approaches across diverse clinical scenarios.
For a detailed walkthrough and examples of the imputation-based methodology, please visit the BSET Tutorial.
If you use this package, please cite:
Carlotti, P. and Parast, L. (2026). A Bayesian Critique of Rank-Based Methods for Surrogate Marker Evaluation. arXiv preprint arXiv:2603.14381.
All code and instructions needed to replicate the results presented
in the paper can be found in the replication/ folder.