Proportion experiencing event under competing-risks regression

[Megan Moreton]

Hi,

I am looking at the results of a three-arm clinical trial, where a secondary outcome is the occurrence of a specific adverse event. The trial was run for 96-weeks. I would like to analyse this secondary outcome using time-to-event with competing risks (ie, early d/c & death). The event was coded as 0=censored; 1=event; 2=competing risk.

I use the following competing risks code:

stset weeks, id(id) failure(event==1)
stcrreg i.randgroup, compete(event==2)


         failure _d:  event == 1
   analysis time _t:  weeks
                 id:  id

Competing-risks regression                       No. of obs       =        287
                                                 No. of subjects  =        287
Failure event  : event == 1                      No. failed       =         43
Competing event: event == 2                      No. competing    =         62
                                                 No. censored     =        182

                                                 Wald chi2(2)     =       2.79
Log pseudolikelihood = -234.85681                Prob > chi2      =     0.2483

                                   (Std. Err. adjusted for 287 clusters in id)
------------------------------------------------------------------------------
             |               Robust
          _t |        SHR   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
   randgroup |
TDF/FTC/DTG  |   1.102586   .3762002     0.29   0.775     .5649167    2.151993
TDF/FTC+EFV  |   .5667937   .2387167    -1.35   0.178     .2482699    1.293975
------------------------------------------------------------------------------

As all individuals were followed-up for the same time, would it also be reasonable to present the results simply as "proportion experiencing the event by Week 96 [including censoring for competing risks]"

tab event randgroup, col

+-------------------+
| Key               |
|-------------------|
|     frequency     |
| column percentage |
+-------------------+

           |       randomisation group
     event |        G1         G2         G3 |     Total
-----------+---------------------------------+----------
  censored |        59         67         56 |       182 
           |     60.20      67.68      59.57 |     62.54 
-----------+---------------------------------+----------
     event |        16         18          9 |        43 
           |     16.33      18.18       9.57 |     14.78 
-----------+---------------------------------+----------
 competing |        23         14         29 |        66 
           |     23.47      14.14      30.85 |     22.68 
-----------+---------------------------------+----------
     Total |        98         99         94 |       291 
           |    100.00     100.00     100.00 |    100.00

So 16%, 18%, and 10%, for each group. Is there a post estimation command for stcrreg that I could use to generate % experiencing event by Wk96, adjusted for baseline characteristics?


Thanks,
Megan

[Megan Moreton]

Would it be suitable to run a logit model & the margins command?

gen eventbin = event==1
logit eventbin i.randgroup
margins i.randgroup