-- Begin included message /* Combinations of Risk Factors */ data test2; input sex treat; DATALINES; 0 0 1 0 0 1 1 1 ; run; /* Survival estimates for the above combinations */ proc phreg data = pudat2; model withtime*wcens(0) = sex treat /ties = efron; baseline out = surv2 survival = survival lower = slower upper = supper covariates = test2 /method = ch nomean cltype=loglog; run; /* Survival estimates at 1 year */ proc print data = surv2 noobs; where withtime = 364;
fit1 <- coxph(Surv(withtime,wcens)~ sex+treat,data=pudat2) my.fit <- survift(fit1) summary(my.fit) Unfortunately, the estimates shown do not match those of SAS. -- End inclusion You need to add a newdata=test2 argument to the survfit call. Like the SAS code, test2 is a data set containing the list of subjects for which you want an estimate. You can also add a times= argument to the summary function. Last, if you look at the documentation you will see that the default handling for ties in the baseline hazard is the Fleming-Harrington method in R, whenever the "Efron" approx was used in computing the coefficients. (Same approximation, 2 different names when used in the two contexts, for more details see the chapter in my book). The "CH" or Aalen approx for a baseline hazard is the same math as the Breslow approx for the coefficients. So to match the chimeric mixture you used in the SAS code you will need to add type="aalen" to the survfit call. The numeric effect of this is usually neglible, but when I see "do not match" in a query I assume you are looking to have all the digits the same. As to competing risks, I have some code for R but documentation is yet to be done -- on my to do list. Walter Kremers (kremers.wal...@mayo.edu) has a nice SAS macro for this case, however. You might want to drop him a line. Terry Therneau ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.