On nonparametric estimating ROC curve based on non-uniform rational B-spline

Area of Science:

• Statistics
• Biostatistics
• Medical Informatics

Background:

• Receiver operating characteristic (ROC) curves are crucial for evaluating diagnostic test efficacy.
• Accurate ROC curve estimation is vital for clinical decision-making.

Purpose of the Study:

• To introduce a novel, versatile method for ROC curve estimation using non-uniform rational B-splines (NURBS).
• To assess the performance of the NURBS-based estimator against existing methods via simulations and real-world data.

Main Methods:

• Utilized non-uniform rational B-splines (NURBS) with control points, weights, and knot sequences for ROC curve estimation.
• Applied linear constraints to NURBS basis function coefficients for smoothing and ensuring non-decreasing functions.
• Conducted Monte Carlo simulations and applied the method to metastatic kidney cancer and diffuse large B-cell lymphoma datasets.

Main Results:

• The NURBS-based estimator demonstrated strong performance in various simulation scenarios.
• Compared to empirical ROC, kernel-based ROC, and Bernstein polynomial estimators, the NURBS method showed competitive or superior accuracy in terms of averaged squared errors.
• The method was successfully applied to two real medical datasets, yielding promising results.

Conclusions:

• The NURBS method provides a powerful and accurate alternative for estimating ROC curves.
• This approach offers enhanced accuracy and flexibility in modeling diagnostic test performance.
• The findings support the utility of NURBS in biostatistical analysis and medical diagnostics.