Thirty studies were included (3,244 patients). Methodological quality was deemed to be moderate. None of the studies fulfilled all of the QUADAS items but all studies fulfilled at least four items and 22 studies met at least half of the items. Withdrawal data were reported clearly in all studies.
Pooled sensitivity was 77% (95% CI 72% to 81%; Ι²=78%) and pooled specificity was 79% (95% CI 74% to 84%; Ι²=78%); data were used from all 30 studies (covering 31 data sets). The area under the ROC curve was 0.85 (95% CI 0.81 to 0.88). Overall, statistical heterogeneity for the bivariate model was high (Ι²=96%, 95% CI 94% to 99%) and could not be explained by meta-regression analyses.
Evidence of significant publication bias was found.