Table 2.
The performance of various models for segregating ATB from LTBI in discovery cohort
| Parameters | Training set (n = 669, 347 ATB, 322 LTBI) | Test set (n = 223, 121 ATB, 102 LTBI) | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| cforest | bart | gamboost | gbm | glmnet | lda | log_reg | svm | cforest | bart | gamboost | gbm | glmnet | lda | log_reg | svm | |
| AUC (95% CI) | 0.995 (0.991–0.998) | 0.986 (0.979–0.992) | 0.970 (0.959–0.981) | 0.985 (0.978–0.992) | 0.953 (0.937–0.968) | 0.926 (0.907–0.946) | 0.961 (0.947–0.975) | 0.981 (0.972–0.990) | 0.978 (0.962–0.993) | 0.976 (0.961–0.991) | 0.969 (0.949–0.988) | 0.968 (0.949–0.987) | 0.951 (0.926–0.977) | 0.917 (0.880–0.954) | 0.951 (0.924–0.978) | 0.950 (0.924–0.976) |
| Sensitivity (95% CI) | 98.85% (97.07–99.55%) | 92.22% (88.92–94.60%) | 89.05% (85.33–91.92%) | 93.37% (90.25–95.54%) | 86.17% (82.14–89.41%) | 78.10% (73.45–82.13%) | 87.90% (84.04–90.92%) | 91.35% (87.93–93.88%) | 93.39% (87.50–96.61%) | 89.26% (82.48–93.61%) | 85.12% (77.71–90.38%) | 87.60% (80.55–92.34%) | 83.47% (75.84–89.04%) | 80.99% (73.09–86.99%) | 87.60% (80.55–92.34%) | 85.12% (77.71–90.38%) |
| Specificity (95% CI) | 95.65% (92.84–97.39%) | 93.48% (90.24–95.70%) | 92.24% (88.79–94.69%) | 92.55% (89.15–94.94%) | 92.55% (89.15–94.94%) | 92.24% (88.79–94.69%) | 93.48% (90.24–95.70%) | 94.41% (91.34–96.44%) | 91.18% (84.08–95.29%) | 90.20% (82.89–94.59%) | 92.16% (85.28–95.97%) | 88.24% (80.55–93.14%) | 91.18% (84.08–95.29%) | 92.16% (85.28–95.97%) | 93.14% (86.51–96.64%) | 93.14% (86.51–96.64%) |
| PPV (95% CI) | 96.08% (93.53–97.65%) | 93.84% (90.77–95.94%) | 92.51% (89.18–94.88%) | 93.10% (89.94–95.32%) | 92.57% (89.18––94.96%) | 91.55% (87.83–94.21%) | 93.56% (90.35–95.75%) | 94.63% (91.67–96.57%) | 92.62% (86.57–96.07%) | 91.53% (85.10–95.33%) | 92.79% (86.42–96.30%) | 89.83% (83.06–94.09%) | 91.82% (85.18–95.64%) | 92.45% (85.81–96.13%) | 93.81% (87.76–96.97%) | 93.64% (87.44–96.88%) |
| NPV (95% CI) | 98.72% (96.75–99.50%) | 91.77% (88.29–94.28%) | 88.66% (84.81–91.62%) | 92.83% (89.48–95.18%) | 86.13% (82.09–89.37%) | 79.62% (75.24–83.40%) | 87.76% (83.86–90.81%) | 91.02% (87.47–93.64%) | 92.08% (85.14–95.93%) | 87.62% (79.96–92.62%) | 83.93% (76.02–89.59%) | 85.71% (77.76–91.15%) | 82.30% (74.24–88.24%) | 80.34% (72.23–86.53%) | 86.36% (78.71–91.56%) | 84.07% (76.22–89.68%) |
| PLR (95% CI) | 22.73 (13.62–37.95) | 14.14 (9.34–21.41) | 11.47 (7.86–16.74) | 12.53 (8.52–18.43) | 11.56 (7.85–17.03) | 10.06 (6.88–14.72) | 13.48 (8.90–20.42) | 16.34 (10.42–25.63) | 10.58 (5.66–19.79) | 9.10 (5.04–16.45) | 10.85 (5.56–21.20) | 7.45 (4.36–12.72) | 9.46 (5.04–17.74) | 10.33 (5.28–20.20) | 12.77 (6.23–26.17) | 12.40 (6.04–25.45) |
| NLR (95% CI) | 0.01 (0.00–0.03) | 0.08 (0.06–0.12) | 0.12 (0.09–0.16) | 0.07 (0.05–0.11) | 0.15 (0.11–0.19) | 0.24 (0.19–0.29) | 0.13 (0.10–0.17) | 0.09 (0.06–0.13) | 0.07 (0.04–0.14) | 0.12 (0.07–0.20) | 0.16 (0.11–0.25) | 0.14 (0.09–0.23) | 0.18 (0.12–0.27) | 0.21 (0.14–0.30) | 0.13 (0.08–0.21) | 0.16 (0.10–0.25) |
| Accuracy (95% CI) | 97.31% (95.79–98.29%) | 92.83% (90.62–94.55%) | 90.58% (88.13–92.57%) | 92.97% (90.78–94.68%) | 89.24% (86.66–91.37%) | 84.90% (81.99–87.42%) | 90.58% (88.13–92.57%) | 92.83% (90.62–94.55%) | 92.38% (88.13–95.19%) | 89.69% (85.00–93.03%) | 88.34% (83.47–91.92%) | 87.89% (82.96–91.54%) | 87.00% (81.95–90.79%) | 86.10% (80.94–90.03%) | 90.13% (85.52–93.39%) | 88.79% (83.97–92.29%) |
ATB: active tuberculosis; LTBI: latent tuberculosis infection; AUC: area under the ROC curve; PPV: positive predictive value; NPV: negative predictive value; PLR: positive likelihood ratio; NLR: negative likelihood ratio; CI: confidence interval