Bootstrap estimation of diagnostic accuracy with patient-clustered data.

Area of Science:

• Biostatistics
• Medical Diagnostics
• Statistical Modeling

Background:

• Accurate estimation of diagnostic test performance metrics like sensitivity and specificity is crucial.
• Receiver operating characteristic (ROC) curve analysis is a standard method for evaluating diagnostic tests.
• Multisite test outcome data, common in clinical studies, present statistical challenges due to potential correlations.

Purpose of the Study:

• To present a straightforward bootstrap methodology for calculating confidence intervals.
• To estimate sensitivity, specificity, and area under the ROC curve (AUC) for multisite test data.
• To offer a robust statistical approach for correlated diagnostic test results.

Main Methods:

• A simulation study was conducted to assess the performance of bootstrap estimates.
• Bootstrap estimates were compared against traditional analytic estimates.
• The methodology was illustrated using real-world data from a comparative angiographic study.

Main Results:

• Bootstrap and analytic methods yielded comparable coverage rates for 95% confidence intervals.
• Bootstrap estimates showed slightly superior coverage compared to analytic estimates when numerous sites per patient were involved.
• Bootstrap percentile intervals demonstrated better coverage performance than asymptotic normal bootstrap intervals.

Conclusions:

• The bootstrap method is a valuable tool for estimating confidence intervals for AUC, sensitivity, and specificity.
• This approach is particularly useful when dealing with correlated data from multisite test outcomes.
• The study validates bootstrapping as a reliable statistical technique in diagnostic test evaluation.