Table 5. Accuracy statistics of each tool and each dataset, using all available drugs.
All values are percentages. The best value for each statistic is shown in bold. Drugs used for each dataset are as follows. Training: Amikacin, Capreomycin, Ciprofloxacin, Ethambutol, Isoniazid, Kanamycin, Moxifloxacin, Ofloxacin, Rifampicin, Streptomycin; Global and Prospective: Amikacin, Capreomycin, Ciprofloxacin, Ethambutol, Isoniazid, Kanamycin, Moxifloxacin, Ofloxacin, Pyrazinamide, Rifampicin, Streptomycin. VME, very major error rate (false-negative rate); ME, major error rate (false-positive rate); PPV, positive predictive value; NPV, negative predictive value.
| (a) Training dataset | ||||||
|---|---|---|---|---|---|---|
| Tool | Sensitivity | Specificity | VME | ME | PPV | NPV |
| ARIBA | 90.04 | 97.91 | 9.96 | 2.09 | 85.59 | 98.62 |
| KvarQ | 80.81 | 98.03 | 19.19 | 1.97 | 84.96 | 97.38 |
| MTBseq | 82.68 | 97.65 | 17.32 | 2.35 | 82.91 | 97.62 |
| Mykrobe | 91.64 | 98.21 | 8.36 | 1.79 | 87.57 | 98.84 |
| TB-Profiler | 89.71 | 95.02 | 10.29 | 4.98 | 71.25 | 98.53 |
| (b) global dataset | ||||||
| Tool | Sensitivity | Specificity | VME | ME | PPV | NPV |
| ARIBA | 87.30 | 96.22 | 12.70 | 3.78 | 91.64 | 94.11 |
| KvarQ | 79.99 | 95.81 | 20.01 | 4.19 | 90.05 | 90.99 |
| MTBseq | 81.34 | 96.03 | 18.66 | 3.97 | 90.67 | 91.56 |
| Mykrobe | 88.12 | 96.16 | 11.88 | 3.84 | 91.58 | 94.47 |
| TB-Profiler | 87.39 | 92.89 | 12.61 | 7.11 | 85.36 | 93.95 |
| (c) prospective dataset | ||||||
| Tool | Sensitivity | Specificity | VME | ME | PPV | NPV |
| ARIBA | 90.10 | 98.87 | 9.90 | 1.13 | 78.74 | 99.54 |
| KvarQ | 84.11 | 99.21 | 15.89 | 0.79 | 83.19 | 99.26 |
| MTBseq | 85.82 | 99.00 | 14.18 | 1.00 | 80.05 | 99.34 |
| Mykrobe | 90.22 | 99.16 | 9.78 | 0.84 | 83.39 | 99.54 |
| TB-Profiler | 83.01 | 97.24 | 16.99 | 2.76 | 58.33 | 99.19 |