therneau/survival

Survival probability is not monotonically decreased across time

Closed this issue · 1 comments

L-1996-G commented

Hi,
I have panel-structured data (e.g. I observe one ID several time). I have built cox PH model with time-varying covariates (coxph(Surv(time1, time2, event)~., data=data)) and then make prediction with following code: predict(fit, data, type = 'survival'). Unfortunately, predicted survival probability is not decreased over time (e.g. For same ID, predicted survival probability for time2=100 is 0.995, while for time2=103 predicted survival probability is equal to 0.996). Conceptually survival probability must be monotonically decreasing.

what is problem and how to achieve monotony?

Thanks

therneau commented

You have a (time1, time2) data set. The resulting probability is for survival from time1 to time2, not for survival from 0 to time 2.
Or at least that is my best guess, since you showed neither data nor code.

  • 👍1