09-23-2013 06:01 PM
I am using the inverse-probability method published by Cole et al (Adjusted survival curves with inverse probability weights) to create my weights however instead of using these weights in the COX procedure, proc phreg, I would like to use them in Proc Lifetest.
proc lifetest data=SDEHP3 plots=(s) graphics outsurv=ph2_survival;
time phage*phcensor1(1);strata dhCAT ;
The benefit is that I can use the medians with confidence intervals as my adjusted median survival times for each strata where as in Cox there are no ages with confidence intervals.
Does anyone see a problem with using inverse probability weights in the weight statement of proc lifetest to produce adjusted KM curves?
09-24-2013 08:57 AM
I don't know the answer to your question.
Another article deals directly with your question (Statistics in Medicine 24:3089-3110).
However, with PROC LIFETEST, you might consider the FREQ statement option, NOTRUNCATE,
that prevents truncation on non-integer weights and the use of Wilcoxon scores (=n_i) instead of the default
log-rank test scores (=1) so that these weights are incorporated in the estimation.