Screening without a "gold standard": the Hui-Walter paradigm revisited

Area of Science:

• Biostatistics
• Epidemiology
• Medical Diagnostics

Background:

• Evaluating diagnostic tests often requires a gold standard, which is not always feasible.
• Previous Bayesian methods for two-test screening using single-population data have shown limitations in parameter estimation accuracy.
• Inferences may lack stability as sample size increases when data are insufficient to estimate all parameters.

Purpose of the Study:

• To examine the limitations of Bayesian inference for two-test screening with single-population data.
• To develop and present an improved statistical approach for situations lacking a gold standard.
• To enhance the accuracy and reliability of diagnostic test evaluations.

Main Methods:

• Analysis of Bayesian inference in the context of two screening tests.
• Comparison of single-population versus two-population sampling strategies.
• Development of a novel approach based on sampling from two distinct populations.

Main Results:

• Single-population sampling for two-test screening can lead to inaccurate inferences and unstable standard errors.
• Sampling from two populations resolves the issue of insufficient data for parameter estimation.
• The proposed two-population sampling method demonstrates increasingly accurate inferences with larger sample sizes.

Conclusions:

• The proposed two-population sampling approach offers a more robust method for evaluating diagnostic accuracy without a gold standard.
• This method ensures improved statistical validity and reliability in screening test assessments.
• The findings have implications for the design and interpretation of studies evaluating medical screening tests.