1

Assume a parametric survival model:

surv <- Surv(time, event)
surv.reg.t <- survreg(surv ~ 1, dist = "t", data = myData)

It is possible to predict the percentile for one given case by

pct=1:9000/10000
predict.t = predict(surv.reg.t, newdata=data.frame(F = "X"), type='quantile', p=pct, se=TRUE)

This is the way described in How to predict survival probabilities in R and How to make prediction in survival analysis using R. In these answers the value of the survivor function is given, and the event time is predicted.

However, how to predict the survivor function given some event time x? Is there a way to use survreg.predict? If not how can I do this using the coefficients of surv.reg.t when the distribution dist="t" is used?

Community
  • 1
  • 1
Claude
  • 1,724
  • 3
  • 17
  • 46
  • You have x (outcome of the predict-operation) and y ( the `pct`-object) of the survival function ... so just find position of the closest `predict.t` and that will be the index to use into the the pct vector. – IRTFM Nov 22 '15 at 23:37
  • @42- That's the rocky path I would like to avoid. Is there not an easier way using `predict(..., type='response')`? – Claude Nov 23 '15 at 07:32
  • It didn't seem particularly "rocky" when I did it. – IRTFM Dec 15 '15 at 19:41

0 Answers0