CODEWITHSUNDEEP
r
JourneyMan
JourneyMan

Reputation: 21

survfit.coxph ; Predicting Survival using newdata and ID option

I am attempting to use surfit.coxph to predict an estimate of the survival function using the newdata and Id option. I am aware of the limitations of this; the baseline hazard is defined as the average of all covariates and what constitutes a typical patient, but please can we put this aside for one moment;

I am fitting the model;

Model.Cox <- coxph(Surv(Start,Stop, censor) ~ baseline,data = data)

I then try to use; 

summary(survfit(Model.Cox, newdata = data,id = Id ))

to predict new data. However, both and

summary(survfit(Model.Cox, newdata = data,id = Id ))$time
summary(survfit(Model.Cox, newdata = data,id = Id ))$surv

give different times than the original data? I would expect predictions for the times in the original dataset, is there a time when this would not be the case?

Upvotes: 2

Views: 3400

Answers (1)

IRTFM
IRTFM

Reputation: 263381

If times is missing (the default) and censored=FALSE (also its default) then you get only predictions at event times. If your expectation is for predictions only for a limited number of individuals, but at all the times in the original dataset ,then you need to provide a vector of times to the times parameter.

allT <- data$Stop
summfitID <- summary(survfit(Model.Cox, newdata = data,id = Id ), times=allT)
summfitID$time
summfitID$surv

Looking at the code I wondered if the same effect could be had just by setting censored-TRUE in the summary.survfit arguments.

Upvotes: 2

Related Questions