Date/Time Date(s) - 26/03/20261:30 pm - 2:30 pm
Speaker: Zelalem Negeri (University of Waterloo)
Location: Hamilton Hall, Room 217
Title: New approaches for diagnostic test accuracy meta-analysis with study-variant thresholds
Abstract: In this talk, I will discuss new latent class modelling (LCM) approaches for meta-analysis of diagnostic test accuracy (DTA) studies with varying thresholds and no gold-standard or perfect reference standard. The current methods for DTA meta-analysis do not account for an absent gold standard or an imperfect reference standard, and they yield clinically meaningless pooled accuracy estimates that do not correspond to a specific cut-off when the primary studies report different test positivity thresholds. The proposed approaches explicitly account for study-specific thresholds by modelling an underlying continuous diagnostic marker and naturally accommodate multiple thresholds and absent or imperfect reference standards. Furthermore, the new approaches yield analytically tractable pooled sensitivity, specificity, SROC curves, and AUCs, yield interpretable DTA parameters at any threshold, and provide more reliable SROC geometry, particularly when thresholds vary widely across the primary studies in a meta-analysis. I will demonstrate the performance of the proposed and standard methods using simulated and real-life datasets.