6GRADE Tables

Accuracy estimates were not pooled because of the considerable heterogeneity among studies. Based on sensitivity median = 61%, specificity median = 92%

28 studies involving smear-negative patients were included; 21/28 (75%) used a cross-sectional design and 7/28 (25%) used a case-control design.

Study limitations were assessed using the QUADAS tool. 17/28 (61%) studies recruited patients in a consecutive manner; 18/28 (64%) studies were conducted in an outpatient setting. Blinding of the commercial test result was reported in 18/28 (64%) studies.

Diagnostic accuracy was considered a surrogate for patient-important outcomes (see ^A3). Indirectness was regarded as a greater concern if a commercial serological test is used as an ‘add on’ test, therefore this was downgraded one point. 16 (57%) were conducted in low/middle-income countries limiting generalisability to these settings.

Heterogeneity was assessed visually and statistically. There was significant heterogeneity in accuracy estimates: sensitivity range 29 to 77%, I-square = 72.5%; p = 0.0000; specificity range 77 to 100%, I-square = 72.1%; p = 0.0000. Subgroups were pre-specified by identity of commercial test, antibody detected, and smear status to decrease heterogeneity. Differing criteria for patient selection and greater duration and severity of illness of the study populations may have introduced variability in findings among studies. The heterogeneity between studies could also be explained by use of different cut-offs for positivity, a factor that could not be addressed.

Accuracy estimates were not pooled. The 95% confidence intervals were very wide for many individual studies; however, this factor was not downgraded as there were a large number of studies and 2 points had already been subtracted for inconsistency.

Data included in the systematic review did not allow for formal assessment of publication bias using methods such as funnel plots or regression tests (see ^A6). Therefore, publication bias cannot be ruled out and it was considered prudent to assume a degree of publication bias as studies showing poor performance of commercial tests were probably less likely to be published.

Based on pooled sensitivity = 76% (95% CI 63 to 87%), pooled specificity = 92% (95% CI 74 to 98%)

7 studies were included in smear-positive patients that evaluated Anda-TB IgG (Anda Biologicals, Strasbourg), an A60-based ELISA.

Study limitations were assessed using the QUADAS tool. None of the studies was considered to have a representative spectrum (only 2/7 studies were conducted in an outpatient setting; 1/7 studies used a cross-sectional study design; and 1/7 studies reported selecting subjects in a consecutive manner). In 2/7 studies the index test was blinded and in 5/7 studies differential verification was avoided.

Diagnostic accuracy was considered a surrogate for patient-important outcomes (see ^A3); only 1/7 studies was conducted in low/middle-income countries limiting generalisability to these settings.

Heterogeneity was assessed by visual inspection of forest plots of sensitivity and specificity estimates. Sensitivity in the studies varied from 54% to 85% and specificity varied from 68% to 100%. However, except for two studies by the same author, the sensitivity estimates were consistent. Specificity estimates were more variable. Heterogeneity between studies could be explained by use of different cut-offs for positivity, a factor that could not be addressed.

Accuracy estimates were pooled by bivariate meta-analysis. Pooled sensitivity and specificity had relatively wide confidence intervals: sensitivity 76% (95% CI 63% to 87%); specificity 92% (95% CI 74 to 98%).

Data included in the systematic review did not allow for formal assessment of publication bias using methods such as funnel plots or regression tests. Therefore, publication bias cannot be ruled out and it was considered prudent to assume a degree of publication bias as studies showing poor performance of commercial tests were probably less likely to be published. This in turn may have introduced ‘optimism bias’ in the pooled estimates of sensitivity and specificity; nevertheless this factor was not downgraded.

Based on pooled sensitivity = 59% (95% CI 10 to 96%), pooled specificity = 91% (95% CI 79 to 96%)

Four studies were included of smear-negative patients that evaluated Anda-TB IgG (Anda Biologicals, Strasbourg), an A60-based ELISA.

Study limitations were assessed using the QUADAS tool. None of the studies was considered to have a representative spectrum (only one study was conducted in an outpatient setting; 2/4 studies used a cross-sectional study design; and 0/4 studies reported selecting subjects in a consecutive manner). In 1/4 studies the index test was blinded and in 1/4 studies differential verification was avoided.

Diagnostic accuracy was considered a surrogate for patient-important outcomes (see ^A3). Indirectness was regarded as a greater concern if Anda-TB were used as an add-on test; this factor was therefore downgraded by one point. No studies were conducted in low/middle-income countries limiting generalizability to these settings.

Heterogeneity was assessed by visual inspection of forest plots of accuracy estimates. The sensitivity varied from 35 to 73% and the specificity varied from 88 to 93%. However, except for one study, sensitivity was consistent and this factor was therefore not downgraded.

Accuracy estimates were pooled by bivariate meta-analysis. Pooled sensitivity had very wide confidence intervals: sensitivity 59% (95% CI 10 to 96%); specificity 91% (95% CI 79 to 96%).

Data included did not allow for formal assessment of publication bias using methods such as funnel plots or regression tests. Therefore, publication bias cannot be ruled out and it was considered prudent to assume a degree of publication bias as studies showing poor performance of commercial tests were probably less likely to be published. This in turn may have introduced ‘optimism bias’ in the pooled estimates of sensitivity and specificity; nevertheless, this factor was not downgraded.