The Moses-Littenberg meta-analytical method generates systematic differences in test accuracy compared to...

Area of Science:

• Medical Statistics
• Diagnostic Test Evaluation

Background:

• Meta-analyses are crucial for synthesizing diagnostic test accuracy data.
• The Moses-Littenberg (ML) and hierarchical summary receiver operating characteristic (HSROC) models are common approaches for meta-analyses of diagnostic test accuracy.
• Differences in methodology can impact study conclusions.

Purpose of the Study:

• To compare diagnostic test accuracy meta-analyses using the Moses-Littenberg (ML) summary receiver operating characteristic (SROC) approach with the hierarchical SROC (HSROC) model.
• To evaluate the impact of different weighting methods within the ML model on accuracy estimates.

Main Methods:

• Reanalyzed 26 datasets from existing test accuracy systematic reviews using both ML (equal weighting and inverse variance weighting) and HSROC models.
• Estimated diagnostic odds ratios (DORs) and relative DORs (RDORs) with covariates.
• Compared models by calculating ratios of DORs and RDORs, and P-values for asymmetry and covariate effects.

Main Results:

• Moses-Littenberg models yielded lower DOR estimates at Q* (median 22% lower for equal weighting, 47% lower for inverse variance weighting) and central data points compared to HSROC.
• Differences in heterogeneity investigations were observed, with ML models showing smaller RDOR differences on average.
• Some instances showed ML models producing higher estimates than HSROC.

Conclusions:

• Moses-Littenberg meta-analyses may underestimate test accuracy and the magnitude of accuracy differences between subgroups.
• Hierarchical models offer a mathematically superior approach for diagnostic test accuracy meta-analyses.
• Recommend utilizing hierarchical model-based approaches for future diagnostic test accuracy meta-analyses.