Multiple Imputation of Missing Covariates When Using the Fine-Gray Model

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• The Fine-Gray model is crucial for analyzing competing risks in survival data.
• Missing covariate data pose challenges in accurately estimating these risks.
• Existing imputation methods may not align with the Fine-Gray model's assumptions.

Purpose of the Study:

• To develop a multiple imputation method compatible with the Fine-Gray model for competing risks.
• To address missing covariate data in the context of estimating the risk of a single event.
• To improve the efficiency and accuracy of covariate association estimation.

Main Methods:

• Developed a novel multiple imputation approach leveraging parallels between Fine-Gray and Cox models.
• Incorporated imputation of potential censoring times for competing events.
• Utilized existing Cox model imputation methodology for missing covariates.

Main Results:

• The proposed method demonstrated good performance in estimating subdistribution log hazard ratios and cumulative incidences.
• It showed efficiency gains over complete-case analysis in simulations and a real-world example.
• Performance was satisfactory even when the proportional subdistribution hazards assumption was not strictly met.

Conclusions:

• The new imputation method is effective for the Fine-Gray model with missing covariate data.
• Accurate specification of proportionality on the correct scale is vital for individual cumulative incidence estimation.
• This approach offers a valuable tool for researchers analyzing competing risks data.