Informed Bayesian survival analysis
František Bartoš
Department of Psychological Methods, University of Amsterdam
Dept. of Statistical Modelling, Institute of Computer Science of the Czech Academy of Sciences
Location: ZOOM
Meeting ID: 926 8545 1604
Passcode: 358199
Date: Thursday 10 February 2022
Time: 13:30 CET
Abstract:
Parametric survival analysis is a powerful method for parameter estimation, hypothesis testing, and survival extrapolation of censored event history outcomes. We outline and implement a Bayesian model-averaging framework for parametric survival analysis that allows us to incorporate historical data, test informed hypotheses, continuously monitor evidence, and incorporate uncertainty about the true data generating process. We illustrate the Bayesian framework by re-analyzing data from a colon cancer trial. In a simulation study, we compare the Bayesian framework to maximum likelihood estimation of survival models with AIC/BIC model selection in fixed-n and sequential designs. We find that the Bayesian framework (1) produces faster decisions in sequential designs, (2) has slightly higher statistical power and false-positive rates in fixed-n designs, and (3) yields more precise estimates of the treatment effect in small and medium sample sizes. We did not find a clear advantage in predicting survival.
Related pre-print: https://arxiv.org/abs/2112.08311
Personal Home Page:
References:
Kass, R. E., & Raftery, A. E. (1995). Bayes factors. Journal of the American Statistical Association, 90(430), 773-795.
Fragoso, T. M., Bertoli, W., & Louzada, F. (2018). Bayesian model averaging: A systematic review and conceptual classification. International Statistical Review, 86(1), 1-28.
Ibrahim, J. G., Chen, M. H., Sinha, D., Ibrahim, J. G., & Chen, M. H. (2001). Bayesian survival analysis (Vol. 2). New York: Springer.