2020/06/16 by Aristidis K. Nikoloulopoulos, Nikoloulopoulos, Aristidis K.
Decision Sciences · Mathematics · #Applications (stat.AP) #FOS: Computer and information sciences #Methodology (stat.ME) #Optimal Experimental Design Methods #Statistical Methods and Bayesian Inference #Statistical Methods in Clinical Trials
paper · pdf · doi:10.48550/arxiv.2006.09278
openalex publication_date 2020/06/16 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
As the meta-analysis of more than one diagnostic tests can impact clinical decision making and patient health, there is an increasing body of research in models and methods for meta-analysis of studies comparing multiple diagnostic tests. The application of the existing models to compare the accuracy of three or more tests suffers from the curse of multi-dimensionality, i.e., either the number of model parameters increase rapidly or high dimensional integration is required. To overcome these issues in joint meta-analysis of studies comparing T >2 diagnostic tests in a multiple tests design with a gold standard, we propose a model that assumes the true positives and true negatives for each test are conditionally independent and binomially distributed given the 2T-variate latent vector of sensitivities and specificities. For the random effects distribution, we employ an one-factor copula that provides tail dependence or tail asymmetry. Maximum likelihood estimation of the model is straightforward as the derivation of the likelihood requires bi-dimensional instead of 2T-dimensional integration. Our methodology is demonstrated with an extensive simulation study and an application example that determines which is the best test for the diagnosis of rheumatoid arthritis.