Table 3. Suggested Score Threshold Performancea.
Performance measure | Suggested medium-risk threshold | Suggested single threshold | Suggested high-risk threshold |
---|---|---|---|
Goal | Emphasize sensitivity | Maximize F1 score (balance sensitivity and PPV) | Emphasize PPV |
Score threshold | 37.4 | 63.9 | 68.3 |
F1 score, % (95% CI) | 4.1 (4.0-4.1) | 9.3 (9.3-9.3) | 8.9 (8.8-8.9) |
Sensitivity, % (95% CI) | 50.2 (49.6-50.8) | 10.8 (10.5-11.2) | 7.9 (7.6-8.3) |
Specificity, % (95% CI) | 84.5 (84.5-84.6) | 99.2 (99.2-99.2) | 99.5 (99.5-99.5) |
PPV, % (95% CI) | 2.1 (2.1-2.1) | 8.2 (7.9-8.4) | 10.0 (9.6-10.4) |
NPV, % (95% CI) | 99.6 (99.6-99.6) | 99.4 (99.4-99.4) | 99.4 (99.4-99.4) |
NNE | 47.6 (46.6-48.1) | 12.3 (11.9-12.7) | 10.0 (9.6-10.4) |
Abbreviations: NNE, number needed to evaluate; NPV, negative predictive value; PPV, positive predictive value.
Observation-level measures are shown (4 183 737 total predictions; deterioration prevalence of 0.7%).