Comparison of Different Methods for the Meta-Analysis of Diagnostic Test Accuracy Studies-A Simulation Study

Meta-analysis of diagnostic test accuracy studies aggregates information from multiple studies on sensitivity and specificity. Classical approaches select a single pair of sensitivity and specificity per study (single threshold methods, STM), ignoring additional information if studies report results on multiple diagnostic thresholds. Recently, models have been proposed that consider all available information and enable inference on all diagnostic thresholds (multiple threshold methods, MTM). We compare five STM and six MTM to each other in a simulation study, evaluating their performance in various situations. Covering a broad range of real-life settings, we vary eight parameter dimensions in the data-generation mechanisms, including continuous or ordinal outcome type of an index test, and different numbers of diagnostic thresholds available per study. While model performances are comparable regarding bias, empirical coverage, and convergence, we observe a logit GLMM of the MTM type to perform best in many situations. Model performances depend strongest on the outcome type, while the number of thresholds only has a minor impact. We thus find the main advantage of using MTM by getting threshold-dependent estimates of sensitivity and specificity. Additionally, we illustrate differences between model estimates in two real-data examples on diagnosing type 2 diabetes using the continuous biomarker HbA1c and screening for any anxiety disorder using the ordinal questionnaire HADS-A. The applications reveal variations in model estimates within and between STM and MTM, which can be reduced by adjusting for the bias in the simulation settings resembling the real-data situation most closely.