Table 2.

Dependent competing risks: comparison of competing risks regression models under different scenarios

Model estimates (group B/A ratio)
Specified HazardsCoxFine–Gray
ScenarioGroup AGroup BCHR (95% CI)SHR (95% CI)
Scenario I: No difference for either event
 Event 1λ11 = 1.00λ21 = 1.001.006 (0.725–1.395)1.006 (0.730–1.387)
 Event 2λ12 = 1.00λ22 = 1.000.999 (0.720–1.385)0.998 (0.724–1.375)
Scenario II: Event 1 rate lower in group A
 Event 1λ11 = 0.50λ21 = 1.003.200 (2.083–4.914)3.097 (2.039–4.704)
 Event 2λ12 = 1.00λ22 = 1.000.789 (0.588–1.058)0.570 (0.428–0.760)
Scenario III: Both event rates lower in group A
 Event 1λ11 = 0.50λ21 = 1.001.997 (1.416–2.816)1.337 (0.971–1.840)
 Event 2λ12 = 0.50λ22 = 1.002.014 (1.427–2.842)1.350 (0.981–1.859)
Scenario IV: Large and small effects in group A
 Event 1λ11 = 0.50λ21 = 1.002.585 (1.762–3.791)2.132 (1.480–3.070)
 Event 2λ12 = 0.75λ22 = 1.001.130 (0.830–1.538)0.793 (0.589–1.067)

NOTE: Competing risks data were simulated from a bivariate exponential failure distribution with dependence between failure times and hazard parameters indicated and independent censoring that resulted in about 33% of lifetimes being censored. Averages of estimated parameters are based on 3,000 simulated data sets, each with sample size N = 250.