S41.2: Selection of the link function for survival regression models with the Brier score

✍ Scribed by Thomas Gerds; Martin Schumacher

• Publisher: John Wiley and Sons
• Year: 2004
• Tongue: English
• Weight: 69 KB
• Volume: 46
• Category: Article
• ISSN: 0323-3847
• DOI: 10.1002/bimj.200490143

✦ Synopsis

Commonly accepted tools for the selection of a regression model such as the mean squared error of prediction or R^2 are typically not applicable beyond the ordinary linear model. For right censored time-to-event data the Brier score can be used for the assessment and comparison of general survival regression models. In this talk we develop a graphical tool based on the Brier score for the assessment of estimated survival probabilities. Of interest is the assessment of the link function between the survival distribution function and the covariates, in particular the functional influence of a particular covariate. The potential use of the the estimated Brier score as a function of time for the construction of valid regression models is illustrated with an oncological data set. Bootstrap methods are considered to overcome the apparent error problem inherit to a data intensive model building process.