Meta-Analysis of the Accuracy of Two Diagnostic Tests Used in Combination: Application to the Ddimer Test and the Wells Score for the Diagnosis of Deep Vein Thrombosis

Abstract

ObjectivesIt is standard practice for diagnostic tests to be evaluated against gold standards in isolation. In routine clinical practice, however, it is commonplace for multiple tests to be used before making definitive diagnoses. This article describes a meta-analytic modeling framework developed to estimate the accuracy of the combination of two diagnostic tests, accounting for the likely nonindependence of the tests. MethodsA novel multicomponent framework was developed to synthesize information available on different parameters in the model. This allows data to be included from studies evaluating single tests or both tests. Different likelihoods were specified for the different sources of data and linked by means of common parameters. The framework was applied to evaluate the diagnostic accuracy of the Ddimer test and the Wells score for deep vein thrombosis, and the results were compared with those of a model in which independence of tests was assumed. All models were evaluated by using Bayesian Markov chain Monte Carlo simulation methods. ResultsThe results showed the importance of allowing for the (likely) nonindependence of tests in the meta-analysis model when evaluating a combination of diagnostic tests. The analysis also highlighted the relatively limited impact of those studies that evaluated only one of the two tests of interest. ConclusionsThe models developed allowed the assumption of independence between diagnostic tests to be relaxed while combining a broad array of relevant information from disparate studies. The framework also raises questions regarding the utility of studies limited to the evaluation of single diagnostic tests.