how to do survival analysis with very few events?

[fengyuwuzu]

 PROC LIFETEST data= adsl2 alphaqt=0.05 alpha=0.05 method=km outsurv=km_outsurv1 ;
     TIME aval * cnsr (1);
     STRATA   TRT01PN;
     ODS OUTPUT ProductLimitEstimates=kme_sd1;
 RUN;
ods listing;

got warning like this:

WARNING: The likelihood ratio test for strata homogeneity is questionable since some strata have no events.

In this case, is there a way to opt out the LRT? or is there a preferred way to handle this?

[Rick_SAS · Accepted solution]

I suppose the easiest way is to excluder the strata that do not have any events.

First run PROC FREQ:

proc freq data=adsl2;
tables TRT01PN * cnsr / norow nocol nopercent;
run;

Then look at the frequency table and find all the strata for which all entries are censored.

Then exclude those strata. For example, if the strata TRT01PN='A' does not have any events, then put

WHERE TRT01PN not in ('A');

into the procedure. If more than one strata does not have events, use

WHERE TRT01PN not in ('A', 'C', 'Z');

[fengyuwuzu]

Thank you Rick. Indeed one of the 3 arms has no event under the subgroup selection.