Comparing calibration and discrimination of CARM model to predict in-hospital mortality in development and validation datasets
| Dataset | Mean predicted risk: alive | Mean predicted risk: died* | Discrimination slope† | Scaled Brier score | AUC (95% CI) | Median imputed AUC (95% CI) |
| Development dataset | 0.047 | 0.229 | 0.183 | 0.175 | 0.874‡ (0.866 to 0.881) | 0.915 (0.888 to 0.941) |
| Validation dataset | 0.053 | 0.231 | 0.178 | 0.165 | 0.861 (0.852 to 0.869) | 0.900 (0.880 to 0.919) |
*Died in-hospital following emergency admission.
†Mean predicted risk difference between who discharged died and discharged alive.
‡Corrected optimism (original=0.874 and corrected=0.873).
AUC, area under the curve; CARM, computer-aided risk of mortality.