Simulation shows undesirable results for competing risks analysis with time-dependent covariates for clinical outcomes

Table 1 Overview of the properties of logistic regression (LR) and the subdistribution models in a time-independent and time-dependent setting

Property	Time-independent		Time-dependent
	Fine&gray	LR	Beyersmann & Schumacher	LR
Allowance for censoring	Yes	No	Yes	No
Accounting for time-to-event	Yes	No	Yes	No
Ability to display cumulative incidence functions	Yes	Only plateau	No	Only plateau
Interpretation	Challenging	See text	Challenging	See text
Probability interpretation	Yes	Yes	No	Yes
Dependency on the infection hazard	.	.	Yes	No
Simulation performance:
Ability to capture no effect on cause-specific hazards (α₁₅=α₀₃,α₁₄=α₀₂)	Yes	Yes	No	Yes
Ability to capture negative effect on death hazard (α₁₅>α₀₃,α₁₄=α₀₂)	Yes	Yes	Yes, magnitude difficult to interpret	Yes
Ability to capture positive effect on death hazard (α₁₅<α₀₃,α₁₄=α₀₂)	Yes	Yes	Questionable	Yes
Ability to capture negative effect on discharge hazard (α₁₅=α₀₃,α₁₄<α₀₂)	Yes	Yes	Questionable	Yes
Ability to capture positive effect on discharge hazard (α₁₅=α₀₃,α₁₄>α₀₂)	Yes	Yes	Questionable	Yes

ISSN: 1471-2288