Learning-based biomarker-assisted rules for optimized clinical benefit under a risk constraint.

Area of Science:

• Biostatistics
• Medical Informatics
• Clinical Decision Making

Background:

• Improving clinical decision-making requires integrating novel biomarkers with existing clinical data.
• Effective statistical methods are crucial for combining diverse information to create targeted interventions.
• Balancing treatment benefits against potential harms is essential in medical interventions.

Purpose of the Study:

• To propose novel statistical approaches for constructing multiple-marker-based decision rules.
• To directly optimize a benefit function while controlling for harm.
• To develop methods applicable to disease detection, screening, surveillance, and prognosis.

Main Methods:

• Development of plug-in and direct-optimization-based algorithms.
• Construction of both nonparametric and parametric decision rules.
• Asymptotic property analysis of proposed estimators.

Main Results:

• Proposed methods demonstrate good clinical utility across various scenarios.
• Simulation results validate the effectiveness of the developed decision rules.
• The methods were successfully applied to a prostate cancer surveillance biomarker study.

Conclusions:

• Novel statistical approaches enable effective integration of multiple biomarkers for improved decision rules.
• The developed methods offer a robust framework for balancing benefit and harm in clinical settings.
• This work has significant implications for disease surveillance and personalized medicine.