An evaluation of computational methods for aggregate data meta-analyses of diagnostic test accuracy studies

• DOI: 10.1186/s12874-024-02217-2
• Publication Date: 2024-05-10T10:01:43Z
• Authors: Yixin Zhao; Bilal Khan; Zelalem F. Negeri
• Keywords: Laplace's method

Abstract Background A Generalized Linear Mixed Model (GLMM) is recommended to meta-analyze diagnostic test accuracy studies (DTAs) based on aggregate or individual participant data. Since a GLMM does not have closed-form likelihood function parameter solutions, computational methods are conventionally used approximate the likelihoods and obtain estimates. The most commonly Iteratively Reweighted Least Squares (IRLS), Laplace approximation (LA), Adaptive Gauss-Hermite quadrature (AGHQ). Despite being widely used, it has been clear how these compare perform in context of an data meta-analysis (ADMA) DTAs. Methods We compared evaluated performance three for - IRLS, LA, AGHQ, via comprehensive simulation study real-life examples, ADMA By varying several parameters our simulations, we assessed terms bias, root mean squared error, confidence interval (CI) width, coverage 95% CI, convergence rate, speed. Results For scenarios, especially when meta-analytic were sparse (i.e., there no negligible with perfect diagnosis), comparable estimation sensitivity specificity. However, LA had largest bias error pooled specificity sparse. Moreover, AGHQ took longer time converge relative other two methods, although best rate. Conclusions recommend practitioners researchers carefully choose appropriate algorithm fitting do sets. either IRLS can be regardless characteristics