correct median follow up in stset?

[Amaranta Losey]

Hello all,
I am confused as to which is giving me the correct median follow up time in survival analysis- stsum or stdescribe.

. stdescribe

failure _d: recur2 == 1
analysis time _t: betweendx2recur/30

|-------------- per subject --------------|
Category total mean min median max
------------------------------------------------------------------------------
no. of subjects 175
no. of records 175 1 1 1 1

(first) entry time 0 0 0 0
(final) exit time 21.3101 .4333333 17.03333 93.33333

subjects with gap 0
time on gap if gap 0
time at risk 3729.2667 21.3101 .4333333 17.03333 93.33333

failures 120 .6857143 0 1 1
------------------------------------------------------------------------------

. stsum

failure _d: recur2 == 1
analysis time _t: betweendx2recur/30

| incidence no. of |------ Survival time -----|
| time at risk rate subjects 25% 50% 75%
---------+---------------------------------------------------------------------
total | 3729.266667 .0321779 175 13.33333 18.66667 26.9


They give me different median numbers- I am thinking stsum since stdescribe seems to be looking at the exit times- any insight? Thanks!

[Andrew Musau]

I am sympathetic to your cause as this is at least the third time you post the same question with no replies. First, I do not do survival analysis, but here is what I get from a quick read of the manual. For stsum

Methods and formulas The 25th, 50th, and 75th percentiles of survival times are obtained from S(t), the Kaplan –Meier product-limit estimate of the survivor function. The 25th percentile, for instance, is obtained as the minimum value of t such that S(t) ≤ 0.75.

The results of stdescribe on the other hand can be obtained from the summarize command. Based on some experimentation, with single-record survival data (number of records= number of subjects) , both stsum and stdescribe will give you approximately the same median value.

webuse page2, clear
stdescribe
stsum

Res.:

. stdescribe

         failure _d:  dead
   analysis time _t:  time

                                   |-------------- per subject --------------|
Category                   total        mean         min     median        max
------------------------------------------------------------------------------
no. of subjects               40   
no. of records                40           1           1          1          1

(first) entry time                         0           0          0          0
(final) exit time                     227.95         142        231        344

subjects with gap              0   
time on gap if gap             0   
time at risk                9118      227.95         142        231        344

failures                      36          .9           0          1          1
------------------------------------------------------------------------------

. 
. stsum

         failure _d:  dead
   analysis time _t:  time

         |               Incidence     Number of   |------ Survival time -----|
         | Time at risk       rate      subjects        25%       50%       75%
---------+---------------------------------------------------------------------
   Total |        9,118   .0039482            40        198       232       261

However, with multiple-record survival data (number of records ≠ number of subjects), you get different results.

webuse stan3, clear
stsum 
stdescribe

. stsum 

         failure _d:  died
   analysis time _t:  t1
                 id:  id

         |               Incidence     Number of   |------ Survival time -----|
         | Time at risk       rate      subjects        25%       50%       75%
---------+---------------------------------------------------------------------
   Total |     31,938.1   .0023483           103         36       100       979

. 
. stdescribe

         failure _d:  died
   analysis time _t:  t1
                 id:  id

                                   |-------------- per subject --------------|
Category                   total        mean         min     median        max
------------------------------------------------------------------------------
no. of subjects              103   
no. of records               172    1.669903           1          2          2

(first) entry time                         0           0          0          0
(final) exit time                   310.0786           1         90       1799

subjects with gap              0   
time on gap if gap             0           .           .          .          .
time at risk             31938.1    310.0786           1         90       1799

failures                      75    .7281553           0          1          1
------------------------------------------------------------------------------

I have no clue how things look like for multiple failure data and will not attempt to find out. As advertised above, here I show how you can reproduce the result of stdescribe using summarize. Since the data is per subject, we can use the contract command to obtain one record per subject.

webuse stan3, clear
contract id stime
sum stime, d

. sum stime, d

                    Survival time (Days)
-------------------------------------------------------------
      Percentiles      Smallest
 1%            2              1
 5%            3              2
10%            6              2       Obs                 103
25%           32              2       Sum of Wgt.         103

50%           90                      Mean           310.0777
                        Largest       Std. Dev.      427.9682
75%          427           1407
90%          979           1571       Variance       183156.8
95%         1386           1586       Skewness       1.725621
99%         1586           1799       Kurtosis       5.129539

If you ask for my opinion, I would favor reporting stsum's results derived from the survivor function, but I am no expert here.