Table 4. Performance of CAD4TB and human readers.
| Threshold for test positivity | Sens.1 [%] (95%CI) | Spec.2 [%] (95%CI) | PPV3 [%] (95%CI) | NPV4 [%] (95%CI) | PLR5 (95%CI) | NLR6 (95%CI) | |
| CAD4TB | ≥23 | 95 (91–98) | 33 (27–39) | 54 (49–60) | 89 (80–94) | 1.42 (1.29–1.56) | 0.16 (0.08–0.29) |
| ≥37 | 91 (86–94) | 52 (46–59) | 61 (55–67) | 87 (80–92) | 1.9 (1.65–2.19) | 0.18 (0.11–0.28) | |
| ≥56 | 85 (79–90) | 69 (62–75) | 69 (63–75) | 85 (79–89) | 2.71 (2.22–3.31) | 0.22 (0.15–0.31) | |
| ≥74 | 77 (71–83) | 79 (74–84) | 76 (69–82) | 81 (75–86) | 3.75 (2.88–4.88) | 0.29 (0.22–0.37) | |
| ≥89 | 62 (55–69) | 85 (80–89) | 77 (70–84) | 73 (67–78) | 4.12 (2.98–5.7) | 0.45 (0.37–0.54) | |
| ≥95 | 47 (40–54) | 94 (91–97) | 88 (80–93) | 68 (63–73) | 8.41 (4.86–14.56) | 0.56 (0.49–0.64) | |
| Expert reader | 4 | 59 (52–66) | 98 (95–99) | 96 (91–99) | 74 (69–79) | 27.62 (11.52–66.26) | 0.42 (0.35–0.49) |
| 3,4 | 78 (71–83) | 85 (80–89) | 81 (75–87) | 82 (77–87) | 5.18 (3.78–7.1) | 0.26 (0.2–0.34) | |
| 2,3,4 | 84 (78–89) | 72 (65–77) | 71 (65–77) | 84 (79–89) | 2.97 (2.4–3.67) | 0.22 (0.16–0.31) | |
| Clinical officer | 4 | 7 (4–12) | 97 (94–99) | 70 (46–88) | 56 (51–61) | 2.8 (1.1–7.15) | 0.95 (0.91–1) |
| 3,4 | 76 (69–82) | 65 (58–71) | 64 (58–70) | 76 (70–82) | 2.15 (1.78–2.61) | 0.37 (0.29–0.49) | |
| 2,3,4 | 97 (94–99) | 18 (13–24) | 50 (45–55) | 89 (77–96) | 1.19 (1.11–1.27) | 0.14 (0.06–0.35) |
1
sensitivity 2specificity 3positive predictive value 4negative predictive value 5positive likelihood ratio 6negative likelihood ratio. All parameters were assessed against group A and B as positive reference standard and group F as negative controls.