Imputation approaches for estimating diagnostic accuracy for multiple tests from partially verified designs

Area of Science:

• Biostatistics
• Medical Diagnostics
• Health Services Research

Background:

• Estimating diagnostic accuracy (sensitivity and specificity) is crucial for new tests and rater groups.
• Gold standard evaluations are often expensive or invasive, necessitating alternative methods.
• Semilatent class modeling has been used for diagnostic accuracy estimation in such scenarios.

Purpose of the Study:

• To present and evaluate imputation approaches for diagnostic accuracy and prevalence estimation.
• To compare the imputation method with existing semilatent class modeling techniques.
• To demonstrate the application and robustness of imputation in a real-world diagnostic study.

Main Methods:

• Utilized imputation techniques for estimating diagnostic accuracy and prevalence.
• Compared imputation with semilatent class modeling in terms of simplicity, robustness, and efficiency.
• Applied the imputation approach to a study on digital radiography for gastric cancer detection.

Main Results:

• Imputation provides a simpler and more robust method for diagnostic accuracy and prevalence estimation compared to semilatent modeling.
• The imputation approach exhibits only moderate efficiency loss relative to correctly specified semilatent models.
• Simulations and analysis confirmed the feasibility and robustness of imputation.

Conclusions:

• Imputation is a viable and advantageous alternative to semilatent class modeling for diagnostic accuracy estimation.
• The method is robust to modeling assumptions and practical for real-world applications.
• Digital radiography for gastric cancer serves as a practical example of imputation's utility.