R help - Oct 2006 - Calculating the probability of an event a timeoint "t" from a Cox model fit

I would like to determine the probability of an event at a specific
timepoint given the linear predictor of a given Cox model.  For
instance, assume that I fit the following model:

 

data(pbc)

fit <- coxph(Surv(time, status)~ age, data=pbc)

 

To extract the value of the linear predictor for each patient in the
dataset:

 

prd <- predict(fit, newdata=pbc, type="lp")

 

However, what I am really interested in is the predicted probability
from the Cox model that an individual will experience an event by some
time t (say 2000 days in this example).  To my knowledge, calculating
this probability requires knowing one of the 3 baseline functions-the
survivor function or the hazard function or the cumulative hazard
function-estimated from the Cox model.  I would like to know :

 

1)     How do I extract the correct baseline hazard so that I can
predict the individual probablility of failure at 2000 days, given a
particular patient covariate pattern (in this case, a particular
patient's age?  Is the baseline hazard for the average patient
covariates the ideal one?

2)     Is there a better way of directly extracting these probabilities?

 

 

Brant Inman

 


	[[alternative HTML version deleted]]