Table 3.
Score results
| AIMS65 | Pre-endoscopic Rockall (PERS) | Glasgow-Blatchforda | Modified Glasgow-Blatchforda | |
|---|---|---|---|---|
| Overall performanceb | ||||
| Brier score | 0.0464 | 0.0647 | - | - |
| Scaled Brier score (%) | 20.1 | 19.7 | - | - |
| Discrimination (AUROC) (95% CI) Primary outcome | 0.684 (0.657–0.711) | 0.647 (0.618–0.675) | 0.774 (0.750–0.798) | 0.782 (0.759–0.805) |
| Discrimination (AUROC) (95% CI) Secondary outcome (inhospital death) | 0.786 (0.744–0.827) | 0.719 (0.663–0.776) | 0.685 (0.631–0.740) | 0.702 (0.649–0.755) |
| Calibrationb (95% CI) | ||||
| Observed inhospital death | 5.0% (3.9–6.1) | 5.0% (3.9–6.1) | - | - |
| Predicted inhospital death | 1.9% (1.77–2.01) | 16.28% (15.67–16.89) | ||
| Calibration-in-the-large, P-value | <0.001 | <0.001 | ||
| Slope | 2.355 (1.899–2.811) | 0.348 (0.260–0.436) | ||
| Intercept | 0.005 (−0.008 to 0.019) | −0.006 (−0.024 to 0.012) | ||
AUROC, area under the receiver operating characteristic; CI, confidence interval.
a
As the Glasgow-Blatchford score did not predict a probability of the outcome, the Brier score and calibration were not feasible.
b
As AIMS65 and PERS predict inhospital death, Brier score and calibration were assessed for the secondary outcome.