Table 3.
Performance of the Germany and US predictive models on the corresponding test sets, evaluated by positive predictive value, sensitivity, and specificity using the 90th, 95th, and 99th percentiles as a threshold and area under receiver operator characteristic curve.
| Performance metric and threshold (percentile) | Germany predictive model (this study), % (95% CI) | US predictive model (adapted from [9]), % (95% CI) | |
| PPVa | |||
| 90% | 5.7 (4.9-6.4)b | 9.4 (8.9-9.8) | |
| 95% | 7.5 (6.3-8.8)b | 13.6 (12.9-14.3) | |
| 99% | 15.0 (10.7-19.3)b | 26.2 (23.7-28.5) | |
| Sensitivity | |||
| 90% | 40.3 (35.1-45.4) | 41.9 (40.0-43.9) | |
| 95% | 26.9 (22.3-31.1) | 30.3 (28.6-32.0) | |
| 99% | 10.6 (7.4-14.0) | 11.7 (10.5-13.0) | |
| Specificity | |||
| 90% | 90.4 (90.1-90.8) | 90.8 (90.6-91.0) | |
| 95% | 95.3 (95.1-95.6) | 95.6 (95.6-95.7) | |
| 99% | 99.1 (99.0-99.2) | 99.3 (99.2-99.3) | |
| AUCc | |||
| — | 0.749 (0.721-0.777) | 0.778 (0.769-0.788) | |
aPPV: positive predictive value.
bNonoverlapping 95% CI between Germany and US predictive models.
cAUC: area under the receiver operator characteristic curve.