Step 2 | |||
---|---|---|---|
Step 1 | Update 1 | Update 2 | |
Discrimination | |||
AUC (95% CI) | 0.80 (0.79 to 0.82) | 0.80 (0.79 to 0.82) | 0.80 (0.79 to 0.82) |
Sensitivity | 1% | 85% | 85% |
Specificity | 100% | 47% | 53% |
PPV | 100% | 92% | 93% |
NPV | 12% | 30% | 32% |
Calibration (H&L statistic*) | |||
χ2 | 178.6 | 8.6 | 8.9 |
p Value | <0.001 | 0.38 | 0.35 |
Overall performance (LR-test) | |||
χ2 | − | − | 1.18† |
p Value | − | − | 0.28 |
Reclassification | |||
NRI‡ | − | 0.31§ | 0.05† |
p Value | − | 0.09 | 0.32 |
Parameters | |||
Model intercept¶ | −6.263 | −3.607 | −5.359 |
Calibration intercept** | − | 2.656 | 0.904 |
Calibration slope†† | − | − | 1.33 |
Step 1: Original prediction model applied to untreated residents.
Step 2: Updating the original model.
Update 1: recalibrating intercept.
Update 2: recalibrating intercept and slope.
*H&L statistic: non-significant p values indicate adequate fit.
†Compared with update 1.
‡Cut-off point 80%.
§Compared with step 1.
¶Intercept of the new model, after updating.
**Deviation from original intercept (−6.263).
††To recalibrate the original regression coefficients, all original regression coefficients are multiplied by the calibration slope or shrinkage factor.
AUC, area under the curve; H&L statistic, Hosmer and Lemeshow goodness-of-fit statistic; LR-test, likelihood ratio test; NPV, negative predictive value; NRI, net reclassification index; PPV, positive predictive value; χ2: Chi-square.