stci does not provide some upper bounds of confidence intervals

[Dorothea Ekoka Mbassi]

Dear community,

I am a bit puzzled and could not find the exact answer to my problem:
I am looking at survival times and running a simple command:

stci, by(group)

I am running this on different subsets of my data. Sometimes it gives me all values wanted (median, SE and 95% CI) for every group, but often there is no upper bound of the confidence interval provided:

. stci, by(rand)

        Failure _d: censored10qPCR
  Analysis time _t: diffhours_round

             | Number of
rand         |  subjects         50%      Std. err.    [95% conf. interval]
-------------+-------------------------------------------------------------
    Drug       |        16         144             .           16          .
     Placebo |        13          96      14.58038           40        120
-------------+-------------------------------------------------------------
       Total |        29          96      24.57327           48          .

You can see that for the Drug, SE and upper bound of CI are not shown, while for total only upper bound is missing.
In my understanding, as the lower bound could be found, the SE should be known and thus the upper bound should be calculated? e.g. with CI(lower)=x-1,96*se(x) and CI(upper)=x+1,96*se(x), in the illustrated example the SE would be very roughly 64 and the upper limit of CI roughly 267 (which would be longer than the follow-up in the present clinical trial)...

How can I obtain the upper bound of my confidence intervals?

[Leonardo Guizzetti]

Basically, your curve doesn't extend below the median time in the Drug group to estimate the upper bound. See #2 here for a more detailed explanation.

[Dorothea Ekoka Mbassi]

Thank You for this reference. I must admit that I do not understand all of the explanation - But do I understand that there is a mathematical reason that I cannot get the upper bound? Not something I could fix with a different command?

[Leonardo Guizzetti]

You might be able to estimate it with a parametric regression model (Kaplan-Meier estimates are non-parametric). But the form it would take and how good it will fit strongly depend on what your data look like, and I can't see that from here.