Table 3.
Indirect comparisons between each test and W4SS for the detection of tuberculosis in all participants (using culture as a reference standard)
| Number of studies | Number of participants | Sensitivity (95% CI) | Specificity (95% CI) |
Difference from W4SS* |
|||
|---|---|---|---|---|---|---|---|
| Sensitivity (p value) | Specificity (p value) | ||||||
| W4SS | 21 | 15 597 | 82% (72–89) | 42% (29–57) | .. | .. | |
| C-reactive protein | |||||||
| ≥10 mg/L | 5 | 3571 | 77% (61–88) | 74% (61–83) | 0·71 | 0·041 | |
| ≥8 mg/L | 5 | 3571 | 81% (68–89) | 70% (57–81) | 0·91 | 0·071 | |
| ≥5 mg/L | 5 | 3571 | 87% (77–93) | 60% (48–71) | 0·51 | 0·27 | |
| Chest x-ray | |||||||
| With any abnormality | 8 | 6195 | 72% (65–78) | 62% (51–71) | 0·26 | 0·13 | |
| Suggestive of tuberculosis | 8 | 6150 | 63% (57–70) | 78% (67–86) | 0·071 | 0·0049 | |
| Cough | |||||||
| Any | 21 | 15 568 | 56% (48–63) | 72% (65–79) | <0·0001 | 0·0006 | |
| Lasting ≥2 weeks | 17 | 10 906 | 38% (29–49) | 84% (77–90) | <0·0001 | <0·0001 | |
| Haemoglobin | |||||||
| <10 g/dL | 9 | 5116 | 43% (33–54) | 80% (73–85) | 0·0006 | 0·0013 | |
| <8 g/dL | 9 | 5116 | 12% (9–16) | 96% (93–97) | <0·0001 | <0·0001 | |
| BMI (<18·5 kg/m2) | 18 | 12 650 | 29% (22–38) | 89% (84–92) | <0·0001 | <0·0001 | |
| Lymphadenopathy | 4 | 2391 | 31% (14–55) | 90% (75–96) | 0·0023 | 0·0018 | |
| Parallel strategies† | |||||||
| W4SS and C-reactive protein (≥10 mg/L) | 5 | 3571 | 88% (63–97) | 31% (13–57) | 0·358 | 0·46 | |
| W4SS and chest x-ray with abnormal findings | 8 | 6186 | 94% (89–97) | 20% (10–37) | 0·0077 | 0·066 | |
| Sequential strategies† | |||||||
| W4SS then C-reactive protein (≥5 mg/L) | 5 | 3571 | 70% (31–92) | 75% (53–88) | 0·55 | 0·0405 | |
| W4SS then Xpert‡§ | 12 | 8557 | 58% (50–66) | 99% (98–100) | .. | .. | |
| Xpert for all‡§ | 12 | 8570 | 68% (57–76) | 99% (98–99) | 0·094¶ | 0·40¶ | |
Indirect comparisons are based on all studies that assessed at least one of the W4SS or relevant screening tests. BMI=body-mass index. W4SS=WHO-recommended four-symptom screen.
For Xpert for all, the comparator is W4SS then Xpert.
For parallel strategies, two screening tests are offered at the same time; for sequential strategies, a second screening test is offered only if the first screening test is positive.
Accuracy measures for entire algorithm using total Xpert (sputum or non-sputum sample Xpert result, or both); alternative algorithms are W4SS then single sputum Xpert (12 studies; 8556 participants; sensitivity 55% [95% CI 48–63], specificity 99% [99–100]) and single sputum Xpert alone (12 studies; 8569 participants; sensitivity 64% [53–74], specificity 99% [98–99]).
One study assessed Xpert and Xpert Ultra among 733 participants; sputum Xpert sensitivity was 57% (95% CI 47–67) and specificity was 99% (98–100), sputum Xpert Ultra sensitivity was 73% (62–81) and specificity was 98% (96–98), urine Xpert Ultra sensitivity was 27% (19–38) and specificity was 98% (96–99), and sputum and urine Xpert Ultra sensitivity was 75% (65–83) and specificity was 95% (94–97).
Bivariate model did not converge; results from a model assuming no correlation between sensitivity and specificity.