Multiply-robust estimator of cumulative incidence function difference for right-censored competing risks data.

Area of Science:

• Causal inference and statistical modeling
• Biostatistics and survival analysis

Background:

• Estimating average treatment effects (ATE) for competing risk outcomes is challenging due to confounding and competing events.
• Doubly robust estimators can fail if both propensity score and outcome models are misspecified.
• There is a need for methods accommodating multiple models to improve reliability in observational studies with competing risks.

Purpose of the Study:

• To propose a novel multiply robust (MR) estimator for cause-specific cumulative incidence functions (CIF) in the presence of right-censored competing risks data.
• To enhance the estimation of ATE in competing risks settings by providing stronger protection against model misspecification.

Main Methods:

• Integrated the pseudo-value approach with a multiply robust estimation framework.
• Transformed time-dependent, censored CIF into a complete-data outcome.
• Specified multiple candidate models for propensity scores and outcome regressions, ensuring consistency if at least one model is correct.

Main Results:

• Monte Carlo simulations showed the MR estimator has negligible bias and accurate coverage probabilities (95%) when at least one model is correctly specified.
• The estimator performed robustly across various censoring rates, with heterogeneous misspecifications outperforming homogeneous ones.
• Application to the Right Heart Catheterization dataset confirmed the method's practical utility and alignment with existing literature.

Conclusions:

• The multiply robust framework provides superior robustness and consistency for estimating cause-specific CIFs, even with high censoring.
• This methodology offers a resilient alternative to traditional estimators for treatment effect estimation in complex competing risks scenarios.