GRADE evidence profile: should commercial serological tests be used as a replacement test for conventional tests such as smear microscopy in patients of any age suspected of having pulmonary tuberculosis?
Based on sample size = 8,318, sensitivity median = 64%, specificity median = 91%. The quality of evidence was rated as high (no points subtracted), moderate (one point subtracted), low (two points subtracted), or very low (>2 points subtracted) based on five factors: study limitations, indirectness of evidence, inconsistency in results across studies, imprecision in summary estimates, and likelihood of publication bias. For each outcome, the quality of evidence started at high when there were randomized controlled trials or high-quality observational studies (cross-sectional or cohort studies enrolling patients with diagnostic uncertainty) and at moderate when these types of studies were absent. No points were subtracted when there were negligible issues identified; one point was subtracted when there was a serious issue identified; two points were subtracted when there was a very serious issue identified in any of the criteria used to judge the quality of evidence. Points subtracted are in parentheses. Publication bias was rated as “not likely,” “likely,” or “very likely” [23].
aWhat do these results mean given 10% or 30% disease prevalence among individuals being screened for TB?
bOutcomes were ranked by their relative importance as critical, important, or of limited importance. Ranking helped to focus attention on those outcomes that were considered most important.
cThe majority of studies lacked a representative patient population and were not blinded.
dAlthough diagnostic accuracy is considered a surrogate for patient-important outcomes, we did not downgrade.
eThere was considerable heterogeneity in study results.
fWe did not pool accuracy estimates. The 95% CIs were wide for many individual studies. We did not downgrade as there were a large number of studies and we had already taken off two points for inconsistency.
gData included in the 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. It is prudent to assume some degree of publication bias as studies showing poor performance of serological tests were probably less likely to be published. No points were deducted.