Analysis of Generalized Semiparametric Regression Models for Cumulative Incidence Functions with Missing Covariates

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Competing risks data present challenges in accurately quantifying event probabilities over time.
• Missing covariate data further complicate the analysis of cumulative incidence functions.
• Semiparametric regression models offer flexibility but require robust estimation methods for incomplete data.

Purpose of the Study:

• To develop and investigate generalized semiparametric regression models for cumulative incidence functions (CIFs) in the presence of missing covariates.
• To propose novel estimation procedures for these models, enhancing analytical capabilities for complex health data.
• To evaluate the performance and efficiency of the proposed methods through simulations and real-world data application.

Main Methods:

• Utilized generalized semiparametric regression models allowing for both parametric and non-parametric covariate effects on CIFs.
• Developed estimation procedures including direct binomial regression and inverse probability weighting (IPW) of complete cases.
• Established asymptotic properties of the proposed estimators and assessed finite-sample performance via simulations under various two-phase sampling designs.

Main Results:

• The proposed estimators demonstrated reliable performance in simulations, with efficiencies varying based on sampling designs.
• The methods effectively handle missing covariate data in the context of competing risks analysis.
• The developed approach provides a flexible framework for modeling CIFs with mixed covariate effects and link functions.

Conclusions:

• The generalized semiparametric regression models offer a robust approach for analyzing competing risks data with missing covariates.
• The developed estimation procedures are statistically sound and perform well in practice, as shown by simulation studies.
• Application to the RV144 vaccine trial highlights the utility of these methods in investigating associations between biomarkers and HIV-1 infection risk.