Semiparametric regression and risk prediction with competing risks data under missing cause of failure.

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Cohort studies with competing risks often suffer from incomplete failure data.
• Existing methods for semiparametric proportional cause-specific hazards models with missing data focus only on regression coefficients.
• Inference for covariate-specific cumulative incidence functions, crucial for risk prediction, is lacking in current approaches.

Purpose of the Study:

• To develop a unified framework for robust statistical inference in the presence of competing risks and missing failure data.
• To enable estimation of both regression coefficients and covariate-specific cumulative incidence functions.
• To provide reliable risk prediction tools for modern medicine.

Main Methods:

• A novel, computationally efficient maximum pseudo-partial-likelihood estimation method is proposed.
• The framework handles missing at random cause of failure.
• Modern empirical process theory is used to derive asymptotic properties of estimators.

Main Results:

• The proposed method provides accurate inference for both regression coefficients and covariate-specific cumulative incidence functions.
• Simulations demonstrate good performance even with a high proportion of missing data.
• The regression coefficient estimator shows improved efficiency over existing methods like augmented inverse probability weighting.

Conclusions:

• The unified framework offers a significant advancement for analyzing complex survival data in cohort studies.
• It enhances the accuracy of risk prediction by providing reliable estimates of cumulative incidence functions.
• The method is validated through simulations and applied to real-world HIV and bladder cancer data.