Evaluating classification performance of biomarkers in two-phase case-control studies.

Area of Science:

• Biostatistics
• Epidemiology
• Medical Diagnostics

Background:

• Biomarkers are crucial for disease screening, early detection, and risk prediction.
• Two-phase case-control studies are common for biomarker evaluation.
• Biased sampling in phase two can invalidate biomarker classification accuracy inference.

Purpose of the Study:

• To develop inverse probability weighting-based estimators for biomarker classification performance.
• To address biased sampling in two-phase case-control studies.
• To improve the accuracy of receiver operating characteristic (ROC) curve measures.

Main Methods:

• Adopted inverse probability weighting for biased sampling correction.
• Developed estimators for ROC curve points, area under the curve (AUC), and partial AUC.
• Incorporated auxiliary variables for enhanced efficiency in weight estimation.

Main Results:

• Proposed weighted estimators demonstrated excellent performance in simulations.
• Traditional empirical estimators showed severe bias.
• Auxiliary variables improved efficiency for classification accuracy estimation.

Conclusions:

• Inverse probability weighting effectively corrects biased sampling in biomarker evaluation.
• The proposed method provides valid inference for biomarker classification accuracy.
• The approach is applicable to real-world studies, such as in renal artery stenosis and prostate cancer.