A statistical method was used for the meta-analysis of tests for latent TB in the absence of a gold standard, combining random-effect and latent-class methods to estimate test accuracy Mohsen Sadatsafavi a , Neal Shahidi b , Fawziah Marra c , Mark J. FitzGerald d , Kevin R. Elwood e,f , Na Guo a , Carlo A. Marra a, * a Faculty of Pharmaceutical Sciences, Collaboration for Outcomes Research and Evaluation, University of British Columbia, Vancouver, British Columbia, Canada b Faculty of Pharmaceutical Sciences, Center for Clinical Epidemiology and Evaluation, Vancouver Coastal Health institute, Vancouver, British Columbia, Canada c Faculty of Pharmaceutical Sciences, Collaboration for Outcomes Research and Evaluation, Pharmacy and Vaccine Services, British Columbia Center for Disease Control, Vancouver, British Columbia, Canada d Faculty of Medicine, Center for Clinical Epidemiology and Evaluation, Vancouver Coastal Health institute, University of British Columbia, Vancouver, British Columbia, Canada e Division of Tuberculosis Control, British Columbia Center for Disease Control, Vancouver, British Columbia, Canada f Division of Respiratory Medicine, Faculty of Medicine, University of British Columbia, Vancouver, British Columbia, Canada Accepted 20 April 2009 Abstract Objective: Because of the lack of a gold standard, the diagnostic performance of tests for the detection of latent tuberculosis infection (LTBI) is not known. However, statistical methods can be used to estimate the accuracy from the studies reporting the concordance among the tests. Study Design and Setting: We developed a random-effect latent-class model to estimate performance characteristics of three LTBI diagnostic tests: tuberculin skin test (TST, at 10-mm cutoff), QuantiFERON-TB gold (QFG), and TSPOT-TB from the studies evaluating agreement among the tests. Results: Nineteen studies were included. QFG had a sensitivity of 0.642 (95% confidence interval [CI]: 0.593e0.691) and specificity of 0.996 (95% CI: 0.989e1.000), TSPOT-TB had a sensitivity of 0.500 (95% CI: 0.334e0.666) and specificity of 0.906 (95% CI: 0.882e0.929), and TST had a sensitivity of 0.709 (95% CI: 0.658e0.761) and specificity of 0.683 (95% CI: 0.522e0.844). Results were not sensitive to the inclusion of any single study. When only the three studies that reported on TSPOT were removed, estimates for the other two tests varied minimally. Conclusions: Statistical methods can help estimate the accuracy of LTBI tests. Although the specificities were close to their reported values in the literature, the estimates for sensitivities were low; a finding that should be carefully evaluated. Ó 2010 Elsevier Inc. All rights reserved. Keywords: Tuberculosis; Meta-analysis; Sensitivity and specificity; Reproducibility of results; Statistical models; Predictive value of tests 1. Introduction Despite concerted global efforts, tuberculosis (TB) re- mains a major world health problem [1]. Of the millions of people annually infected with Mycobacterium tuberculosis (MTB), the bacteria responsible for the disease, 10% will eventually develop active TB [2]. Progression of a case of latent TB infection (LTBI) to active TB is associated with significant morbidity and mortality. In addition, new active cases of TB become the source of ongoing transmission of in- fection. On the other hand, LTBI is not transmittable, and its treatment is more efficient, safer, and less expensive [3]. Hence, the availability of an accurate diagnostic test for LTBI has important public health implications. Historically, the detection of LTBI has relied on the tu- berculin skin test (TST). The TST depends on a delayed- * Corresponding author. Faculty of Pharmaceutical Sciences, Collabo- ration for Outcomes Research and Evaluation, University of British Columbia, 2146 East Mall, Vancouver, BC V6T 1Z3, Canada. Tel.: þ1-604-806-8817; fax: þ1-604-875-5179. E-mail address: carlo.marra@ubc.ca (C.A. Marra). 0895-4356/10/$ e see front matter Ó 2010 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2009.04.008 Journal of Clinical Epidemiology 63 (2010) 257e269