A new diagnostic accuracy measure and cut-point selection criterion

Area of Science:

• Biostatistics
• Medical Diagnostics
• Machine Learning

Background:

• Existing diagnostic accuracy measures are limited to binary or three-stage diseases.
• Current methods for multi-stage diseases, like generalized Youden index, have limitations in cut-point selection and data utilization.

Purpose of the Study:

• To propose a novel diagnostic accuracy measure for diseases with k stages, named maximum absolute determinant.
• To establish the maximum absolute determinant as a criterion for optimal cut-point selection in multi-stage disease diagnosis.
• To provide geometric and probabilistic interpretations of the new measure.

Main Methods:

• Development of the maximum absolute determinant measure for k-stage diseases.
• Geometric and probabilistic interpretation of the proposed measure.
• Power and simulation studies to evaluate performance.
• Application to Alzheimer's Disease Neuroimaging Initiative data.

Main Results:

• The maximum absolute determinant utilizes all classification information, unlike existing measures.
• It serves as an effective criterion for selecting optimal diagnostic cut-points in multi-stage diseases.
• Performance evaluations demonstrate its utility in diagnostic accuracy assessment and cut-point selection.

Conclusions:

• The maximum absolute determinant is a comprehensive and effective measure for evaluating diagnostic accuracy in multi-stage diseases.
• This new measure overcomes limitations of existing methods, particularly in cut-point selection.
• The proposed method shows promise for real-world applications, as demonstrated by the Alzheimer's disease data analysis.