A systematic review 1 of rapid diagnostic tests for the detection of tuberculosis infection included 110 data sets relating to nucleic acid amplification tests (NAAT). Of the 106 data sets on respiratory specimens, 52% used culture alone as the reference standard, 44% combined culture with clinical symptoms (with or without an assessment of response to anti-TB therapy or chest X-ray), 2 used culture plus treatment response and 3 used clinical diagnosis plus treatment response only. The sensitivity and specificity pairs for both commercial and in-house NAAT tests were generally all in the top left-hand quadrant of the ROC plot, but were very heterogeneous. Pooled sensitivity of NAATs in diagnosing active pulmonary TB was 86% and specificity 96% (+LR 21.5, -LR 0.15) for sputum tests. For serum tests (4 data sets), sensitivity was 53% and specificity 95% (+LR 10.6, -LR 0.49). In diagnosing extrapulmonary TB, the sensitivity was 78% and specificity 96% (+LR 19.5, -LR 0.23). When analyses were restricted to the better designed studies, overall pooled accuracy was still reasonably high, although sensitivity in particular is probably not high enough to permit accurate ruling out of TB infection (79.4%). NAAT test accuracy progressively decreases in sputum smear-negative patients compared with sputum smear-positive, and in sputum culture-negative patients compared with culture-positive patients. Test accuracy and in particular test sensitivity are therefore related to the bacterial burden in clinical specimens.
Comment: The quality of evidence is downgraded by inconsistency (variability in results across studies, heterogeneity in interventions and outcomes).
Primary/Secondary Keywords