Fifty-two studies met the inclusion criteria (n=2,399): 32 case series; 16 cohort studies; and four case-control studies. Of the 52 studies, 87% recruited a representative population, 81% clearly defined inclusion criteria, 98% avoided partial and differential verification bias, 88% blinded interpreters of the index test to the reference standard results; 60% avoided progression bias; <50% did not report sufficient details to allow replication of the tests; and only 38% reported on uninterpretable results.
Only eight prospective studies provided sufficient data for 2x2 tables of test performance to be constructed (n=295, range 21 to 64). These studies were of higher quality and passed most of the QUADAS criteria.
Sensitivity of uMb was 38% (one study), 47% (one study), 86% (one study) and 100% (five studies); specificity ranged from 15% to 91%. Confidence intervals were wide. LR+ were <2.5 (six studies), 4.3 (one study) and 8.6 (one study). PPVs were <58% in all studies.