PatientLevelPrediction icon indicating copy to clipboard operation
PatientLevelPrediction copied to clipboard

Add smooth calibration for survival models

Open rekkasa opened this issue 2 years ago • 1 comments

Is your feature request related to a problem? Please describe. Currently, the smooth calibration functions only assume binary outcomes.

Describe the solution you'd like Add a function that generates the smooth calibration for time-to-event outcomes using hazard regression or restricted cubic splines.

Describe alternatives you've considered The integrated calibration index (ICI) and the several E statistics are calculated correctly in getEvaluationStatistics_survival using hazard regression from polspline::hare. We could use this to generate the smooth calibration plots as well.

Additional context The code to implement this is available in Austin et al. (2020)

@jreps I can implement this if you agree

rekkasa avatar Apr 04 '22 15:04 rekkasa

This would be excellent! We definitely need a smooth calibration plot for survival models but I had no idea how to do it. Great job figuring out how to get this done.

jreps avatar Apr 05 '22 12:04 jreps