﻿================================================================================

STUDY COHORT: hand-written Fine-Gray versus cmprsk::crr

================================================================================

n=1831, events=196, competing=7



          term  python_sHR  cmprsk_sHR  abs_diff_sHR  rel_diff_pct  python_se  cmprsk_se  python_p  cmprsk_p

   lac_2.1-3.0      1.1586      1.1585        0.0000        0.0006     0.1927     0.1925    0.4450     0.440

   lac_3.8-5.9      1.4788      1.4787        0.0000        0.0017     0.1787     0.1817    0.0286     0.031

lac_8.6 capped      2.1156      2.1155        0.0001        0.0045     0.2857     0.2928    0.0087     0.010



================================================================================

SIMULATED BENCHMARK: 1,084 events of interest, 2,084 competing events

================================================================================

term  python_sHR  cmprsk_sHR  abs_diff_sHR  rel_diff_pct  python_se  cmprsk_se

  x1      1.8355      1.8355           0.0        0.0000     0.0311     0.0297

  x2      0.7127      0.7127           0.0        0.0001     0.0641     0.0633



Cause-specific Cox on the same simulated data, for contrast:

covariate

x1    1.4833

x2    0.7767



The subdistribution and cause-specific estimates differ substantially

here, so matching cmprsk is a genuine test of the implementation rather

than a consequence of having too few competing events to distinguish them.



================================================================================

Largest relative discrepancy across both datasets: 0.004%

================================================================================

The implementations agree to within 1% on every coefficient.



Outputs: <local path>\analysis\results\gate8

