Table 3. The performance of the developed models for the differentiation of COVID-19 from influenza, and community-acquired pneumonia with similar symptoms.
| Classification models | Model performance (Mean and 95% CI) | ||||||
| Accuracy (%) | F1 score | MCC | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | |
| PCR | 92.0 (73.9 - 99.1) | 0.66 | 0.67 | 100 (75.8 - 100) | 91.3 (71.9 - 98.9) | 50.1 (21.1 - 78.9) | 100 |
| CT | 84.0 (63.9 - 95.5) | 0.67 | 0.64 | 100 (79.8 - 100) | 80.9 (58.1 - 94.6) | 50.1 (29.3 - 70.7) | 100 |
| Integrated model-training set | 92.0 (73.9 - 99.1) | 0.81 | 0.78 | 100 (89.8 - 100) | 90.5 (69.6 - 98.8) | 86.7 (74.8 -92.2) | 100 |
| Integrated model-internal validation | 96.0 (79.6 - 99.9) | 0.86 | 0.85 | 92.1 (89.4 - 99.4) | 88.2 (83.9 - 100) | 92.3 (85.4 – 100) | 94.5 (79.4 - 99.1) |
| Integrated model-external validation | 92.7 (82.4 – 97.9) | 0.80 | 0.76 | 88.9 (51.8 – 99.7) | 93.5 (82.1 – 98.6) | 72.7 (46.6 – 89.1) | 97.7 (87.1 – 99.6) |
The models were trained by the random forest algorithm. 95% CI: 95% confidence interval; MCC: Matthews correlation coefficient; PPV (%): positive predictive value; NPV (%): negative predictive value.