Simple methods for comparing two predictive values with incomplete data.

Area of Science:

• Biostatistics
• Medical Diagnostics
• Epidemiology

Background:

• Statistical methods for comparing binary diagnostic tests exist for complete data.
• Existing methods lack efficient handling of incomplete data, often requiring complex iterative algorithms.
• Simple methods for comparing sensitivities or specificities with incomplete data are available, but not for predictive values.

Purpose of the Study:

• To propose two novel, simple, and easily implemented statistical methods for comparing predictive values of two binary diagnostic tests with incomplete data.
• To address the limitations of existing methods that do not accommodate missing data or require complex computations.
• To provide efficient alternatives for analyzing paired diagnostic test data with missing observations.

Main Methods:

• Development of two new statistical methods based on minor modifications of existing weighted generalized score statistics.
• The proposed methods generate simple-to-compute test statistics for comparing predictive values.
• Comparative analysis using simulation studies to evaluate the efficiency of the proposed methods against an ad-hoc approach.

Main Results:

• The proposed methods are computationally simple and suitable for incomplete paired diagnostic test data.
• Simulation results indicate that the new methods are more efficient than the ad-hoc method, which excludes subjects with missing data.
• The methods demonstrated practical utility when applied to a real-world study on non-invasive endometriosis detection.

Conclusions:

• The proposed simple statistical methods offer an efficient and practical solution for comparing predictive values of binary diagnostic tests with incomplete data.
• These methods provide a valuable alternative to complex maximum likelihood approaches for handling missing data in diagnostic test evaluations.
• The findings have implications for improving the analysis of observational studies involving paired diagnostic tests, particularly in fields like medical diagnostics.