Table 1.
Comparison of competing risks regression models under different competing risks scenarios, with 250 subjects per group. Competing risks data were simulated from a bivariate exponential failure distribution with the hazard parameters indicated and independent censoring that resulted in about 33% of lifetimes being censored. Averages of estimated parameters are based on 3000 simulated datasets.
| Model Estimates (Group B/A ratio) | ||||||
|---|---|---|---|---|---|---|
| Specified Hazards | Cox | Fine and Gray | ||||
| Scenario | Group A | Group B | CHR | 95% CI | SHR | 95% CI |
| I: No difference for either event | ||||||
| Event 1 | λ11 = 1.00 | λ21 = 1.00 | 0.993 | 0.731–1.349 | 0.994 | 0.737–1.343 |
| Event 2 | λ12 = 1.00 | λ22 = 1.00 | 1.000 | 0.736–1.359 | 1.003 | 0.743–1.355 |
| II: Event 1 rate lower in Group A | ||||||
| Event 1 | λ11 = 0.50 | λ21 = 1.00 | 2.013 | 1.428–2.839 | 1.794 | 1.286–2.503 |
| Event 2 | λ12 = 1.00 | λ22 = 1.00 | 0.999 | 0.750–1.331 | 0.750 | 0.567–0.991 |
| III: Both event rates lower in Group A | ||||||
| Event 1 | λ11 = 0.50 | λ21 = 1.00 | 2.002 | 1.451–2.763 | 1.291 | 0.956–1.742 |
| Event 2 | λ12 = 0.50 | λ22 = 1.00 | 2.003 | 1.452–2.762 | 1.292 | 0.958–1.744 |
| IV: Large and small effects in Group A | ||||||
| Event 1 | λ11 = 0.50 | λ21 = 1.00 | 2.002 | 1.441–2.802 | 1.555 | 1.133–2.135 |
| Event 2 | λ12 = 0.75 | λ22 = 1.00 | 1.333 | 0.988–1.799 | 0.939 | 0.704–1.251 |