Measuring and estimating diagnostic accuracy when there are three ordinal diagnostic groups.

Area of Science:

• Biostatistics
• Medical Diagnostics
• Machine Learning

Background:

• Assessing diagnostic accuracy with multiple ordinal categories is challenging.
• Existing methods may not fully capture performance across all classification thresholds.

Purpose of the Study:

• To develop and evaluate novel methods for measuring and estimating diagnostic accuracy with three ordinal diagnostic groups.
• To introduce receiver operating characteristic (ROC) surfaces as a tool for this assessment.

Main Methods:

• Utilized ROC surfaces to model classification probabilities across diagnostic thresholds.
• Proposed using the entire and partial volume under the ROC surface as accuracy measures.
• Developed maximum likelihood estimation and asymptotic variance for the ROC surface volume under normality assumptions.
• Investigated asymptotic confidence interval estimates and their performance via simulation.

Main Results:

• Introduced a novel measure of diagnostic accuracy based on ROC surface volume.
• Provided statistical methods for estimating this measure and its variance.
• Evaluated confidence interval performance and developed sample size determination methods.
• Demonstrated application in early-stage Alzheimer's disease diagnosis.

Conclusions:

• The proposed ROC surface methodology offers a robust approach to quantifying diagnostic accuracy for three ordinal groups.
• The statistical tools developed enable reliable estimation and validation of diagnostic performance.
• This framework has practical implications for clinical diagnosis and research, exemplified by Alzheimer's disease detection.