Table 3.
Accuracy of the biosignatures identified in the present study in the diagnosis of active TB.
| Biosignature | AUC (95% CI) | Accuracy in training set | Accuracy in test set or after leave-one-out cross-validation | Performance according to WHO TPP | |||||
|---|---|---|---|---|---|---|---|---|---|
| Sens (95%CI) | Spec (95%CI) | Sens (95%CI) | Spec (95%CI) | PPV (95%CI) | NPV (95%CI) | Sens (at Spec ≥70%) | Spec (at Sens ≥90%) | ||
| 4-marker biosignature (I-309, procalcitonin, CRP, PDGF-BB) identified in the Norwegian cohort | |||||||||
| 0.98 (0.96–1.00) |
92.9% (85.3–97.4) |
89.5% (66.9–98.7) |
91.8% (83.8–96.6) |
89.5% (66.9–98.7) |
97.5% (91.3–99.3) |
70.8% (54–83.4) |
100 | 89 | |
| Performance of the 4-marker biosignature in the South African cohort | |||||||||
| 0.90 (0.82–0.98) |
68.2% (45.1–86.1) |
93.2% (83.5–98.1) |
68.2% (45.1–86.1) |
91.5% (81.3–97.2) |
75% (55.3–87.9) |
88.5% (80.6–93.5) |
91 | 76 | |
| 5- marker biosignature (C1q, procalcitonin, CRP, PDGF-BB, Ferritin) identified in Pulmonary TB from the Norwegian cohort | |||||||||
| 1.00 (1.00–1.00) |
100% (89.9–100) |
100% (79.4–100) |
100% (89.9–100) |
100% (79.4–100) |
100% |
100% |
100 | 100 | |
| Performance of the 5-marker biosignature in the South African cohort | |||||||||
| 0.76 (0.65–0.88) |
77.3% (54.6–92.2) |
57.6% (44.1–70.4) |
63.6% (40.7–82.8) |
57.6% (44.1–70.4) |
35.9 (26.6–46.4) | 80.9 (70.1–88.5) | 59 | 44 | |
| 3-marker biosignature (MMP-9, IP-10, sCD40L) identified in the South African cohort | |||||||||
| 0.90 (0.83–0.97) |
68.2% (45.1–86.1) |
88.1% 77.1–95.1) |
68.2% (45.1–86.1) |
88.1% (77.1–95.1) |
68.2% (50.3–82) |
88.1% (80–93.2) |
86 | 68 | |
| Performance of the 3-marker biosignature in the Norwegian cohort | |||||||||
| 0.58 (0.44–0.72) |
49.4% (38.4–60.5) |
57.9% (33.5–79.7) |
48.2% (37.3–59.3) |
36.8% (16.3–61.6) |
77.4% (69.4–83.7) |
13.7% (7.9–22.9) |
38 | 16 | |
| Performance of the 3-marker biosignature in Pulmonary TB from the Norwegian cohort | |||||||||
| 0.50 (0.34–0.66) |
38.2% (22.2–56.4) |
68.4% (43.4–87.4) |
29.4% (15.1–47.5) |
31.6% (12.6–56.6) |
43.5% (29.6–58.5) |
20% (11.1–33.4) |
47 | 0 | |
| 5-marker biosignature (G-CSF, C3b/iC3b, procalcitonin, IP-10, PDGF-BB) identified in the Norwegian and South African cohorts combined | |||||||||
| 0.94 (0.89–0.99) |
84.2% (72.1–92.5) |
91.8% (80.4–97.7) |
72.7% (49.8–82.3) |
90.5% (69.6–98.8) |
88.9% (67.2–96.8) |
76% (61.2–86.4) |
93 | 78 | |
| 6-marker biosignature (RANTES, G-CSF, C1q, CC3, CFH, IP-10) identified in Pulmonary TB from the Norwegian and South African cohorts combined | |||||||||
| 0.93 (0.88–0.98) |
82.4% (65.5–93.2) |
83.7% (70.3–92.7) |
66.7% (38.4–88.2) |
81% (58.1–94.6) |
71.4% (49.1–86.6) |
77.3% (61.7–87.6) |
91.2 | 73.5 | |
Biosignatures generated from data obtained in the Norwegian cohort (pulmonary TB + extrapulmonary TB), South African cohort (pulmonary TB) and when the cohorts (Norway and SA) were combined are shown. The Leave-one-out cross-validation method was applied to the Norwegian cohort (N = 104) and South African cohort (N = 81), whereas the training and test set method was used when both cohorts were combined (N = 185) to determine the predictive accuracy of biosignatures. The performance of biosignatures were also evaluated against the minimum sensitivity and specificity values of the minimum WHO target product profile (TPP) for a triage test. AUC, Area under the ROC curve; Sens, sensitivity; Spec, specificity; PPV, Positive Predictive Value; NPV, Negative Predictive Value; CI, Confidence Interval.