Covariate adjustment in continuous biomarker assessment

Area of Science:

• Biostatistics
• Medical Diagnostics
• Biomarker Research

Background:

• Continuous biomarkers are crucial for disease screening and diagnosis, requiring threshold setting for clinical decisions.
• Covariates (age, race, sample conditions) can influence biomarker distribution and confound disease association, necessitating adjustment.
• Existing covariate adjustment methods often target the entire biomarker distribution, risking model misspecification for specific clinical utility.

Purpose of the Study:

• To develop a covariate adjustment method for biomarker evaluation that specifically targets desired sensitivity/specificity levels.
• To propose a parsimonious quantile regression model for localized biomarker analysis at controlled sensitivity.
• To extend the local model for global covariate adjustment across the receiver operating characteristic (ROC) curve continuum.

Main Methods:

• A parsimonious quantile regression model is proposed for the diseased population, focusing locally on the controlled sensitivity level.
• Specificity is assessed with covariate-specific control of sensitivity, using sample-based and bootstrap approaches for variance estimates.
• The local model is extended to a global one for covariate adjustment of the ROC curve, ensuring monotonicity and computational efficiency.

Main Results:

• The proposed method demonstrates computational efficiency and restores monotonicity in estimated covariate-adjusted ROC curves.
• Asymptotic properties of the proposed estimators are established, indicating theoretical soundness.
• Simulation studies confirm the favorable performance of the proposed covariate adjustment method.

Conclusions:

• The novel quantile regression approach provides accurate covariate adjustment for biomarker evaluation at specific sensitivity levels.
• This method enhances the clinical utility of diagnostic tests by precisely controlling performance metrics.
• The approach is illustrated effectively in biomarker evaluation for aggressive prostate cancer.