Assessing diagnostic tests by a strictly proper scoring rule

Area of Science:

• Medical Diagnostics
• Biostatistics
• Clinical Chemistry

Background:

• Traditional error rate measures for diagnostic tests have limitations.
• Proper scoring rules offer a statistically robust alternative for evaluating diagnostic test performance.
• Assessing posterior probability of disease is crucial for accurate diagnosis.

Purpose of the Study:

• To evaluate univariate quantitative diagnostic tests using strictly proper scoring rules.
• To compare the power of scoring rules against traditional error rate measures.
• To demonstrate the application of proper scoring rules in a clinical setting.

Main Methods:

• Estimating the posterior probability of disease as a function of test values.
• Utilizing training observations for model estimation and test samples for scoring.
• Applying the bootstrap procedure for standard error estimation and bias correction.
• Demonstrating the method with serum bile acids and bilirubin in liver disease patients.

Main Results:

• Proper scoring rules provide a more rigorous evaluation of diagnostic tests compared to error rates.
• The bootstrap procedure effectively addresses bias and estimates standard errors.
• The method is applicable to real-world clinical data, such as liver disease markers.

Conclusions:

• Strictly proper scoring rules are a valuable and recommended alternative for evaluating quantitative diagnostic tests.
• This approach offers enhanced statistical rigor and clinical utility over traditional error rates.
• The findings support the adoption of scoring rules in diagnostic test assessment.