Hello,
I have am working with an administrative healthcare database and have constructed a competing risk regression to look at risk of having an ED visit within the first 30 days after being hospitalized, accounting for the competing risks of mortality and planned rehospitalization. I would like to compare outcomes at 30 days across a specific binary covariate - biologic sex.
If I was working in a logistic regression I would typically use the margins command to compare predicted probabilities of ED visits (e.g. any ED visit within 30 days). I am hoping for assistance on how to interpret the margins command after a competing risk regression as there is no clear documentation or examples in the documentation for stcrreg postestimation. I've provided example code below with (covariate details omitted for clarity).
How does one interpret a "predicted relative subhazard" in this instance? Would the predictive relative subhazard for males (0.48, 0.44-0.53) approximate a predicted probability at 30 days?
I have am working with an administrative healthcare database and have constructed a competing risk regression to look at risk of having an ED visit within the first 30 days after being hospitalized, accounting for the competing risks of mortality and planned rehospitalization. I would like to compare outcomes at 30 days across a specific binary covariate - biologic sex.
If I was working in a logistic regression I would typically use the margins command to compare predicted probabilities of ED visits (e.g. any ED visit within 30 days). I am hoping for assistance on how to interpret the margins command after a competing risk regression as there is no clear documentation or examples in the documentation for stcrreg postestimation. I've provided example code below with (covariate details omitted for clarity).
How does one interpret a "predicted relative subhazard" in this instance? Would the predictive relative subhazard for males (0.48, 0.44-0.53) approximate a predicted probability at 30 days?
Code:
stset tte_card, failure(status_card==1)
Survival-time data settings
Failure event: status_card==1
Observed time interval: (0, tte_card]
Exit on or before: failure
--------------------------------------------------------------------------
1,678,088 total observations
0 exclusions
--------------------------------------------------------------------------
1,678,088 observations remaining, representing
890,582 failures in single-record/single-failure data
32746613 total analysis time at risk and under observation
At risk from t = 0
Earliest observed entry t = 0
Last observed exit t = 31
stcrreg i.sex (other covariates in model omitted for clarity), compete(status_card==2) vce(cluster prvdr_num)
Failure _d: status_card==1
Analysis time _t: tte_card
Iteration 0: Log pseudolikelihood = -12409010
Iteration 1: Log pseudolikelihood = -12408413
Iteration 2: Log pseudolikelihood = -12408413
Competing-risks regression No. of obs = 1,678,088
No. of subjects = 1,678,088
Failure event: status_c~d == 1 No. failed = 890,582
Competing event: status_c~d == 2 No. competing = 206,206
No. censored = 581,300
Wald chi2(57) = 15894.90
Log pseudolikelihood = -12408413 Prob > chi2 = 0.0000
(Std. err. adjusted for 5,597 clusters in prvdr_num)
---------------------------------------------------------------------------------
| Robust
_t | SHR std. err. z P>|z| [95% conf. interval]
----------------+----------------------------------------------------------------
sex |
Female | .7362776 .0084726 -26.60 0.000 .7198575 .7530722
---------------------------------------------------------------------------------
margins sex
Predictive margins Number of obs = 1,678,088
Model VCE: Robust
Expression: Predicted relative subhazard, predict()
--------------------------------------------------------------------------------
| Delta-method
| Margin std. err. z P>|z| [95% conf. interval]
---------------+----------------------------------------------------------------
sex |
male | .4863794 .0227964 21.34 0.000 .4416994 .5310595
female | .3650681 .0167004 21.86 0.000 .332336 .3978003
--------------------------------------------------------------------------------
