From Summary ROC Curves to Clinical Implementation and Recognition: Bridging Meta-Analyses of Diagnostic Test Accuracy Studies and Clinical Usability

• Applicant: Professorin Dr. Annika Hoyer
• Subject Area: Epidemiology and Medical Biometry/Statistics
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 519901253

Meta-analyses and systematic reviews form the foundation of evidence-based medicine, guiding clinical decisions in treatment, diagnosis, and prevention, as well as informing healthcare policy. While statistical methods for meta-analyzing intervention trials are well established, meta-analysis of diagnostic test accuracy (DTA) studies has recently gained substantial attention. This is largely due to the increased complexity of diagnostic studies, which involve the bivariate outcome of sensitivity and specificity. This project addresses challenges in meta-analyzing DTA studies, particularly when multiple diagnostic thresholds per study are reported. Traditional meta-analyses often simplify data by selecting a single threshold per study, potentially losing valuable information. Recent guidelines recommend advanced statistical methods for complete-threshold analyses, which still possess limitations, including reliance on continuity corrections. To overcome these issues, we already developed alternative models: a meta-regression extending the bivariate generalized linear mixed model, and a novel bivariate time-to-event model that treats diagnostic test values as interval-censored survival data. This model accommodates heterogeneity and correlation via random effects and allows estimation of summary ROC curves. Extensions include flexible distributional assumptions using generalized F-distributions and semi-parametric piecewise-constant models to relax parametric constraints. Despite improvements, open questions remain on distributional assumptions and random effects modeling. In this research project, we focus on four main objectives: (1) developing new statistical methods for meta-analyzing ROC curves using multinomial distributions to model heterogeneity without Gaussian random effects, (2) creating an ensemble learning approach to combine multiple meta-analysis methods based on data characteristics, (3) developing a comprehensive R package including available methods and simulation workflows to facilitate practical application, and (4) translating these models into clinical practice for effective implementation. This research advances the meta-analytic methodology for DTA studies, aiming to improve evidence synthesis and guide practical application in medical diagnostics. We focus explicitly on transferring of state-of-the-art research methods into clinical practice, bridging the gap between method development and ready-to-use tools that can be used to perform meta-analyses of DTA studies without technical expertise.

• DFG Programme: Research Grants