Table 2.
Diagnostic performance of empirically derived threshold models for SARS-CoV-2 RT-PCR Positivity
| Accuracy | Sensitivity | Specificity | PPV | NPV | |
|---|---|---|---|---|---|
| Metric | |||||
| ≥ 40% | 0.818 (0.783–0.853) | 0.792 (0.735–0.842) | 0.850 (0.791–0.891) | 0.850 (0.799–0.894) | 0.792 (0.740–0.843) |
| > 10% | 0.776 (0.738–0.815) | 0.898 (0.852–0.934) | 0.646 (0.578–0.709) | 0.731 (0.680–0.782) | 0.855 (0.802–0.909) |
| > 80% | 0.774 (0.736–0.813) | 0.593 (0.528–0.656) | 0.968 (0.936–0.987) | 0.952 (0.918–0.987) | 0.689 (0.638–0.741) |
| Most accurate model (AI airspace opacity severity ≥ 40%) | |||||
| False Positive Rate | 0.155 (0.107–0.202) | LR+ | 5.13 (3.74–7.03) | RR | 4.06 (3.15–5.23) |
| False Negative Rate | 0.208 (0.156–0.259) | LR− | 0.246 (0.190–0.317) | OR | 20.9 (12.9–33.7) |
Bolded values indicate highest values for each category
SARS-CoV-2: severe acute respiratory syndrome coronavirus 2; RT-PCR: reverse transcription polymerase chain reaction; PPV: positive predictive value; NPV: negative predictive value; LR: likelihood ratio; RR: relative risk; OR: odds ratio