Why did I get confidence intervals without a estimate?

[Amo89tw]

Hi all,

I was doing progression-free survival analysis. The software I am using is SAS EG (Version 7.15 HF2 (7.100.5.6112)). I noticed that one of my result was pretty weird to me. I got 75% quartile "NA", which was sensible to me because the survival function did not reach 25%. However, I got a lower limit of this point estimate. I could not get my head around how this happened. Many thanks in advance.

 

[FreelanceReinh · Accepted solution]

Hi @Amo89tw,

This is explained in section Breslow, Fleming-Harrington, and Kaplan-Meier Methods of the PROC LIFETEST documentation (see paragraph starting with "Sometimes, the confidence limits for the quartiles cannot be estimated."): The existence of the confidence limits does not directly depend on the Kaplan-Meier curve itself (dropping below 0.75, 0.5 or 0.25), but on the curves that represent the confidence limits (CL) for the survivor function. They envelop the Kaplan-Meier curve. Thus its well possible that the curve representing the lower CL already crosses the 0.25 level (so that the lower CL of the third quartile exists) while the Kaplan-Meier curve and the curve representing the upper CL remain above that level (hence neither the point estimate nor the upper CL of the third quartile exist). Similarly, the curve representing the lower CL and also the Kaplan-Meier curve can drop below the 0.5 level while the curve representing the upper CL is still above 0.5, so that just the upper CL of the estimated median survival time is missing (as is the case with your data).

[Amo89tw]

This is really helpful. Thank you so much @FreelanceReinh