topic Proc Lifetest - Min and Max for duration of response in Statistical Procedures

Proc Lifetest - Min and Max for duration of response

Banoo — Sat, 23 Jan 2021 21:50:15 GMT

Hello,

I want to find minimum and maximum for duration of response. Is there a way to get those numbers from PROC LIFETEST procedure. Mean, median and 95% CI are straight forward from the procedure. Should I use Proc means or proc summary instead.

My SAS code :

ods output productlimitestimates = km (keep=stratum trtpn Timelist Survival)
Quartiles =_quartiles (keep=stratum trtpn percent estimate lowerlimit upperlimit percent
where=(percent eq 50))
Means = mean
CensoredSummary =_censorsum;
proc lifetest data=want
method=KM timelist=12 24 outsurv=surv;
time dor*dorcnsr(1); /* dor - duration of response */
strata trt;
run;

Any help is much appreciated. Thank you !!

Re: Proc Lifetest - Min and Max for duration of response

FreelanceReinh — Sun, 24 Jan 2021 16:57:33 GMT

Hello @Banoo,

@Banoo wrote:

I want to find minimum and maximum for duration of response. Is there a way to get those numbers from PROC LIFETEST procedure. Mean, median and 95% CI are straight forward from the procedure. Should I use Proc means or proc summary instead.

To be consistent with other quantile estimates (such as the median) using the Kaplan-Meier method -- see the "general formula for estimating the 100pth percentile point" in section Breslow, Fleming-Harrington, and Kaplan-Meier Methods of the PROC LIFETEST documentation and consider the cases p → 0 and p → 1 -- I would use

for the minimum: the smallest duration ("event time," i.e., excluding censored times)
for the maximum: the largest duration, unless it is censored, in which case the maximum cannot be estimated (cf. missing quartile estimates if the Kaplan-Meier curve doesn't drop far enough).

So, yes, you can use PROC MEANS or PROC SUMMARY, but taking the censoring flag (your variable dorcnsr) into consideration as described above.

Re: Proc Lifetest - Min and Max for duration of response

Banoo — Fri, 12 Mar 2021 15:12:55 GMT

Thanks for your input. I took the simpler approach by using Proc Means and appending them with appropriate censoring flag.