New to Competing Risk Regression Model - Needs Advice

[Richard Lin2]

Hello,

I am just starting to use the competing risk regression model for survival analysis and would like advice on the two observations:

I have a dataset of 300 patients where "nrmyear" is the survival endpoint where there is a competing risk (nrm=2), and I did the following to look at the effect of one independent variable "intensity".

. stset nrmyear, failure(nrm==1) scale(1)

failure event: nrm == 1
obs. time interval: (0, nrmyear]
exit on or before: failure

------------------------------------------------------------------------------
330 total observations
0 exclusions
------------------------------------------------------------------------------
330 observations remaining, representing
102 failures in single-record/single-failure data
786.125 total analysis time at risk and under observation
at risk from t = 0
earliest observed entry t = 0
last observed exit t = 11.90137

. stcrreg intensity, compete(nrm==2)

failure _d: nrm == 1
analysis time _t: nrmyear

Iteration 0: log pseudolikelihood = -558.27385
Iteration 1: log pseudolikelihood = -557.90741
Iteration 2: log pseudolikelihood = -557.90692
Iteration 3: log pseudolikelihood = -557.90692

Competing-risks regression No. of obs = 330
No. of subjects = 330
Failure event : nrm == 1 No. failed = 102
Competing event: nrm == 2 No. competing = 104
No. censored = 124

Wald chi2(1) = 7.98
Log pseudolikelihood = -557.90692 Prob > chi2 = 0.0047

------------------------------------------------------------------------------
| Robust
_t | SHR Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
intensity | .4403543 .1278474 -2.82 0.005 .2492724 .7779116
------------------------------------------------------------------------------

. stcurve, cif at1( intensity=1 ) at2( intensity=2 )

I think I did it correctly but had two questions:

1. Why the two CIF curves look the same in the graph?

2. How do I derive the 95% CI from CIF at each time point? In COX model you can do a Life Table but is there such a thing in competing risk model?

Thanks a lot,
Richard

[Clyde Schechter]

The two curves have a similar overall appearance (they are clearly not "the same") because -stcrreg- imposes the proportional (sub)hazards assumption, juast as -stcox- does.

I do not know how to get confidence intervals for the CIF at each time point. I hope somebody else does and will respond to that question.

[Richard Lin2]

Thank you Clyde,

Does anyone know the answer to my second question?

Thanks,
Richard

[Steve Samuels]

Install stpm2 and stpm2cif from SSC. stpm2cif has an option to estimate confidence intervals for CIFs.

See: Sally Hinchliffe and Paul Lambert,Extending the flexible parametric survival model for competing risks, The Stata Journal (2013) 13, Number 2, pp. 344–355

Added: There is another approach: download the command stcrprep from SSC.
According to Slide 6 of Paul Lambert's presentation (link below)

After expansion and weighting of the data,
• sts graph, failure will plot CIF.
• sts test will perform test for differences in CIFs
• stcox will fit a Fine and Gray model (same as stcrreg).
• estat phtest can be used to assess proportional subhazards.
• streg, stpm2 can be used to fit parametric models for CIF.

Reference:
Paul Lambert's Stata Meetings UK13 presentation Estimating and modelling cumulative incidence functions using time-dependent weights

[Steve Samuels]

I should have added: the ci option of sts graph should plot the confidence intervals.

[Richard Lin2]

Thank you Steve, these commands seem to work well.

Best,
Richard