Addressing nonignorable missing data and heterogeneity in prognostic biomarker assessment

Area of Science:

• Biostatistics
• Biomarker Research
• Survival Analysis

Background:

• Time-dependent area under the curve (AUC) assesses biomarker discriminative performance for time-to-event outcomes.
• Missing biomarker data, especially nonignorable missingness, presents significant challenges in biomarker research.
• Covariates can influence biomarker accuracy, complicating performance evaluation.

Purpose of the Study:

• To estimate the impact of covariates on time-dependent AUC with nonignorable missing biomarkers.
• To propose a robust statistical method for handling complex missing data scenarios in biomarker studies.
• To validate the proposed method using simulation and real-world data.

Main Methods:

• Utilized instrumental variables to address identifiability issues caused by nonignorable missingness.
• Integrated inverse probability weighting with a pseudo partial likelihood score equation for parameter estimation.
• Established asymptotic properties of the proposed estimators for theoretical validation.

Main Results:

• The proposed method effectively estimates covariate effects on time-dependent AUC in the presence of nonignorable missing biomarker data.
• Simulation studies demonstrated the good finite sample performance of the developed estimators.
• The method was successfully applied to the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.

Conclusions:

• The developed statistical approach provides a reliable way to analyze biomarker performance with complex missing data.
• This method enhances the accurate evaluation of biomarkers in time-to-event studies, particularly when covariates are involved.
• The findings have implications for biomarker discovery and validation in clinical research, exemplified by the ADNI study.