Modelling multiple thresholds in meta-analysis of diagnostic test accuracy studies.

Area of Science:

• Biostatistics
• Medical Diagnostics
• Health Services Research

Background:

• Traditional meta-analyses of diagnostic test accuracy typically use only one sensitivity and specificity pair per study.
• Biomarker and questionnaire-based tests often yield multiple thresholds with associated true/false positive/negative values.
• This limits the comprehensive evaluation of test performance in existing meta-analysis frameworks.

Purpose of the Study:

• To introduce a novel meta-analysis approach that incorporates multiple thresholds for diagnostic tests.
• To leverage additional data from true/false positives/negatives at various thresholds for a more robust analysis.
• To improve the estimation of diagnostic test accuracy and identify optimal thresholds.

Main Methods:

• Estimating underlying biomarker or questionnaire distribution functions in diseased and non-diseased groups.
• Employing linear mixed-effects models on transformed data, assuming normal or logistic distributions.
• Accounting for across-study heterogeneity and the dependence between sensitivity and specificity.

Main Results:

• Generation of a summary receiver operating characteristic (SROC) curve.
• Calculation of pooled sensitivity and specificity across all thresholds.
• Identification of an optimal threshold via Youden index maximization.
• Demonstration with meta-analyses of B-type natriuretic peptide and procalcitonin.

Conclusions:

• The proposed method utilizes all available data for a more complete performance assessment.
• It provides estimations for both biomarker performance and the optimal threshold for diagnostic accuracy.
• This approach enhances the reliability and utility of meta-analyses for diagnostic tests.