R help - Apr 2010 - Problem with time in coxph/survfit

Hi!

I am having a few problems with coxph function, I had the same problem with
the use of survfit. Here it is:

when calling 'M22<-coxphw(Surv(V1,V2,V4)~V5, data=XTDV, id=XTDV$V3, *
prentice*= ~V5, robust=TRUE, censcorr=TRUE)'

knowning that my data is:
> XTDV[1:10,1:3]
   V1       V2   V3 V4        V5
1   0 36.39706   36  1  343.2224
2   0 36.39706  943  0  161.5931
3   0 36.39706   45  0  790.3714

I get this message: "Error in w[i] <- w[i] * w.sorted[fit$time
=obj$resp[i, 2]] : replacement has length zero". When I perform a debug I
find that the problem is that the time returned by obj$resp[i,2] is for a
certain i different from fit$time so all the vector is false....but in fact
the difference is minimal and it doesn't do that with precedent values...

Browse[2]> w.sorted[fit$time==obj$resp[2051,2]]
numeric(0)
Browse[2]> obj$resp[2051,2]
[1] 49.53238
Browse[2]> fit$time[23]
[1] 49.53238
Browse[2]> obj$resp[2051,2]-fit$time[23]
[1] -7.105427e-15

maybe it's because the time I use is not integer? Like I said, I had the
same problem using results from survfit but I fixed it by rounding values
before comparing them. Is there a way to change the function coxphw?

Thank you.

	[[alternative HTML version deleted]]

Your problem appears to be with coxphw, not coxph.  I have never before
heard of the coxphw function; you should identify what package it is
from in your request, and perhaps also contact the author.  If you have
found a similar issue in survfit I would be glad to see an example.

Terry Therneau
(author of coxph and survfit)