Least squares regression methods for clustered ROC data with discrete covariates.

Area of Science:

• Biostatistics
• Medical Diagnostics
• Statistical Modeling

Background:

• Receiver Operating Characteristic (ROC) curves are crucial for evaluating diagnostic test accuracy with continuous or ordinal results.
• Clustered data from multiple tests on the same subject present challenges for traditional ROC analysis.
• Existing least squares methods for ROC curves do not adequately address clustered data or its statistical properties.

Purpose of the Study:

• To develop and evaluate least squares ROC methods specifically designed for clustered data with discrete covariates.
• To investigate the statistical properties and efficiency of these novel methods compared to existing nonparametric approaches.
• To provide a robust tool for accurate diagnostic test evaluation in complex, clustered data settings.

Main Methods:

• Development of least squares ROC methods allowing flexible baseline and link functions.
• Adaptation of methods to accommodate clustered data structures and discrete covariates.
• Derivation of asymptotic properties for the proposed statistical methods.

Main Results:

• The proposed least squares methods generate smooth ROC curves that respect the continuous nature of underlying true curves.
• Simulation studies demonstrate superior efficiency of the least squares methods over nonparametric ROC methods under specific model assumptions.
• The methods were successfully applied to a real-world example of detecting glaucomatous deterioration.

Conclusions:

• The developed least squares ROC methods effectively handle clustered data and discrete covariates, offering an advancement in diagnostic test evaluation.
• These methods provide a more efficient and statistically sound approach compared to existing techniques for complex datasets.
• The findings have significant implications for accurately assessing diagnostic test performance in clinical research and practice.