Re: Simulation of survival data to produce sample size estimate.

[Terry Therneau <therneau@mayo.edu>]

 Why not use the actual survival curve?
 
 Assume a male, enrolled at age 50 in 1990:
 
> tdata <- data.frame(age=50*365.25, sex=1, year=chron('1/1/1990'))
> esurv <- survexp(~1, data=tdata, ratetable=survexp.us, times=(0:50)*365.25)
> plot(esurv$time/365, esurv$surv)

  This gives a survival curve, now grab random survival times off of it:
  
> ntime <- 300
> times <- approx(esurv$surv, esurv$time, runif(ntime))$y

 Splus has the US, US by race, Minnesota, Florida, and Arizona tables
built in.  (Since the rate tables are stored in terms of rate/day, all
use of them has to be in days.  Hence the multiplication/division by
365.25).  The approx function just does linear interpolation.

        Terry Therneau