Bayesian and influence function-based empirical likelihoods for inference of sensitivity in diagnostic tests.

Area of Science:

• Statistics
• Medical Diagnostics
• Biostatistics

Background:

• Diagnostic test evaluation often focuses on sensitivity at a specific level of specificity.
• Current methods like normal approximation and empirical likelihood (EL) struggle with high specificity and may require parameter tuning.
• These limitations impact the reliability of diagnostic test performance assessments.

Purpose of the Study:

• To develop novel statistical methods for accurate sensitivity inference in diagnostic studies, particularly under high specificity.
• To address the limitations of existing normal approximation and empirical likelihood approaches.
• To introduce influence function-based empirical likelihood and Bayesian empirical likelihood methods.

Main Methods:

• Proposed an influence function-based empirical likelihood method.
• Developed Bayesian empirical likelihood methods.
• Conducted numerical studies to compare the proposed methods against existing ones using coverage probability and interval length.

Main Results:

• The proposed empirical likelihood and Bayesian methods demonstrated superior performance compared to existing approaches.
• The new methods showed improved coverage probability and shorter interval lengths in simulations.
• Analysis of the Alzheimer's Disease Neuroimaging Initiative (ANDI) dataset was performed using the proposed techniques.

Conclusions:

• The novel influence function-based and Bayesian empirical likelihood methods provide more reliable statistical inference for diagnostic test sensitivity, especially at high specificity.
• These methods overcome drawbacks of traditional techniques, offering enhanced precision and accuracy in medical diagnostic studies.
• The findings are validated through numerical simulations and application to real-world Alzheimer's disease data.