Efficient Noninferiority Testing Procedures for Simultaneously Assessing Sensitivity and Specificity of Two...

Area of Science:

• Biostatistics
• Medical Diagnostics
• Statistical Methods

Background:

• Sensitivity and specificity are key metrics for evaluating binary diagnostic tests.
• Existing asymptotic methods for simultaneous testing of sensitivity and specificity may have poor type I error control in small samples.
• Unconditional approaches offer an alternative for robust statistical inference.

Purpose of the Study:

• To compare three unconditional statistical approaches for simultaneously testing sensitivity and specificity of diagnostic tests.
• To evaluate the performance of these methods, particularly regarding type I error control and statistical power.
• To identify the most suitable method for simultaneous assessment of diagnostic test accuracy.

Main Methods:

• Comparison of three unconditional approaches: estimation, maximization, and estimation-maximization.
• Evaluation of type I error rates and statistical power across different sample sizes.
• Focus on exact unconditional methods for simultaneous hypothesis testing.

Main Results:

• The estimation approach shows satisfactory type I error control, though not guaranteed exact.
• Both maximization and estimation-maximization approaches provide exact type I error control.
• The estimation-maximization approach demonstrates superior statistical power compared to the maximization-only approach.

Conclusions:

• Unconditional approaches are recommended for simultaneous testing of sensitivity and specificity, especially in small to medium sample sizes.
• The estimation-maximization method is preferred due to its balance of exact type I error control and higher statistical power.
• This study provides guidance for selecting appropriate statistical methods in diagnostic test performance evaluation.