Background: One diagnostic test is rarely sufficient to complete a diagnosis. Despite this, most diagnostic accuracy studies focus on the evaluation of an individual test, and therefore approaches for the meta-analysis of this diagnostic literature have also focused exclusively on estimating the accuracy of individual tests. Where the accuracy of sequences of tests has been modelled previously, the often dubious assumption of test independence has been made (1). Hence, there is a need to develop meta-analytic approaches to correctly estimate the accuracy of test sequence strategies.
Objective: To develop a modelling approach to simultaneously synthesise studies of individual and multiple diagnostic test data in order to estimate the accuracy of test sequence strategies.
Methods: The methodological framework developed is broadly applicable to contexts where multiple tests are of interest. Here we describe the approach using the motivating example of the Ddimer and Wells tests for diagnosing deep vein thrombosis. This random-effects modelling framework allows: 1) the inclusion of studies evaluating either test singularly and in combination (complete or partial reported data) and thus incorporates all available evidence; 2) for the dependency between diagnostic tests; and 3) the incorporation of tests with multiple thresholds (Wells score).
Results: When results of this synthesis are compared to simpler, but invalid, modelling which assumes the performance of both tests is independent, considerably different estimates for test strategies are observed.
Conclusions: Accurate estimation of the accuracy of test sequences is critical for evidence-based decision making. For the first time a meta-analytic approach to do this has been developed.
1. Novielli N, et al. How is evidence on test performance synthesised in economic decision models of diagnostic tests? Value in Health 2010; 13(8): 952-957.