Hi Statalist,

I am interested how to use Stata to calculate a number needed to treat statistic following competing risk regression.

In brief, I am examining a propensity-score matched cohort of patients, half of whom were exposed to an intensification of the drug of interest and the other half of whom were not. The primary outcome is occurrence of an adverse drug event and the competing risk is death. The study follow up is 30 days, so patients are censored at death or at 30 days.

To run the model, I used the stcrreg program:

Code:
stset adv_event_days_30, failure (event_adv_event_30==1)
stcrreg intensification if pscohort==1, compete(event_adv_event_30==2)

         failure _d:  event_adv_event_30 == 1
   analysis time _t:  adv_event_days_30

Iteration 0:   log pseudolikelihood = -1587.6318  
Iteration 1:   log pseudolikelihood = -1587.6308  
Iteration 2:   log pseudolikelihood = -1587.6308  

Competing-risks regression                       No. of obs       =      4,050
                                                 No. of subjects  =      4,050
Failure event  : event_a~30 == 1                 No. failed       =        192
Competing event: event_a~30 == 2                 No. competing    =        168
                                                 No. censored     =      3,690

                                                 Wald chi2(1)     =       5.47
Log pseudolikelihood = -1587.6308                Prob > chi2      =     0.0193

---------------------------------------------------------------------------------
                |               Robust
             _t |        SHR   Std. Err.      z    P>|z|     [95% Conf. Interval]
----------------+----------------------------------------------------------------
intensification |   1.407947   .2059497     2.34   0.019     1.057001    1.875414
---------------------------------------------------------------------------------
Where event_adv_event_30=1 is the primary outcome event and event_adv_event_30=2 is death

I am interested in calculating the number needed to treat (or in this case harm) with 95% confidence intervals at 30 days.

For cox proportional hazard models I understand I can simply take the inverse of the absolute risk reduction, as described below, but I am not sure how to go about this for a competing risk regression approach where the outcome is not exactly binary as it would be for cox proportional hazard models.

HTML Code:
https://www.statalist.org/forums/forum/general-stata-discussion/general/415025-calculating-number-needed-to-treat-with-confidence-intervalls-when-using-multiple-imputation
Another reference: Altman DG, Andersen PK. Calculating the number needed to treat for trials where the outcome is time to an event. BMJ 1999; 319(7223):1492–1495. [PubMed]

Appreciate your insight!

- Tim Anderson