Affinity-based measures of biomarker performance evaluation

Area of Science:

• Biostatistics
• Medical Diagnostics
• Biomarker Research

Background:

• Existing diagnostic accuracy measures have limitations.
• There is a need for new summary statistics to evaluate biomarker performance.
• Biomarker accuracy assessment is crucial in medical research and clinical practice.

Purpose of the Study:

• To propose novel summary measures of biomarker accuracy.
• To develop a covariate-specific version for assessing conditional discrimination performance.
• To provide nonparametric Bayes estimators for these new indexes.

Main Methods:

• Conceptualization of summary measures based on Hellinger affinity and geometrical principles similar to Pearson correlation.
• Development of a covariate-specific summary index.
• Devising nonparametric Bayes estimators and deriving theoretical properties of priors.
• Performance assessment through a simulation study.

Main Results:

• The proposed summary measures offer a new way to assess biomarker accuracy.
• The covariate-specific index allows for conditional performance evaluation.
• Nonparametric Bayes estimators were successfully devised and their properties derived.
• Simulation studies confirmed the performance of the proposed methods.

Conclusions:

• The new summary measures provide valuable companions to existing diagnostic accuracy measures.
• The covariate-specific index enhances the assessment of biomarker discrimination performance.
• The proposed statistical methods are robust and applicable in practice.
• The methods are illustrated effectively using prostate cancer diagnosis data.