An assessment of nucleic acid amplification testing for active mycobacterial infection

Incidence of TB based on WHO estimates from 2012: high incidence = > 100 cases per 100,000 people; medium incidence = 10–100 cases per 100,000 people; low incidence = ≤ 10 cases per 100,000 people

K = the number of studies; NAAT = nucleic acid amplification testing; TB = tuberculosis

There was greater variability in the specificity among studies, especially those that were conducted in countries with a high incidence of TB and that used in-house NAAT methodologies (Figure 42 in Appendix D). Meta-analysis of this subgroup showed that the pooled specificity was 79% (95%CI 67, 88), which was significantly lower than for the Xpert NAAT in these countries (96%; 95%CI 92, 96). In fact, the overall pooled specificity for all studies using in-house NAATs (91%; 95%CI 85, 95) was significantly lower than for those using the Xpert NAAT (97%; 95%CI 95, 98). Similar differences in the pooled estimates were seen for both the sputum and non-sputum subgroups, but the difference between in-house and commercial NAATs did not reach statistical significance in the non-sputum subgroup due to the wide CIs (Figure 15).

For specific non-sputum specimen types, the pooled specificity ranged from 90% to 97%, except for body fluids (such as synovial fluid and endometrial fluid), which had a pooled sensitivity of 69% (Figure 40 in Appendix D). It should be noted that although the pooled sensitivity of NAAT compared with culture was 97% for CSF specimens (compared with only 11% for AFB microscopy), the wide 95%CI (21, 100) indicated uncertainty in this estimate. As expected, the pooled specificity among different specimen types was much more varied, ranging from 71% for body fluids to 97% for both bronchial specimens and urine specimens.

Figure 16 Deek’s Funnel plot asymmetry test to assess publication bias for the diagnostic accuracy of NAAT compared with culture

Publication bias is assessed visually by using the inverse of the square root of the effective sample size (ESS) versus the log diagnostic odds ratio, which should have a symmetrical funnel shape when publication bias is absent (Light & Pillemer 1984). A regression slope coefficient, weighting by ESS, with p<0.05 indicates significant asymmetry (Deeks, Macaskill & Irwig 2005).

The LR scattergram in Figure 17 shows that the summary LR+ and LR– values for all studies investigating the ability of NAAT to correctly identify patients with and without TB, compared with culture, were mostly within the upper right quadrant of the graph (Figure 17A). Thus, a positive NAAT result was likely to correctly confirm the presence of MTB (as diagnosed by culture). As the summary estimates for exclusion were within the green shaded area close to the upper left quadrant, NAAT provided strong diagnostic evidence suggesting that patients who tested negative were more likely not to have culture-positive TB than to be falsely negative. Similar results were seen when studies that reported data for either sputum or non-sputum specimens were analysed separately (Figure 17B and C).