Summary of test performance predicting one or more unplanned admissions within 1 year of discharge for different case finding algorithms
| Method | Country | Model | Data sources | Total separations (patients) | Test separations (patients) | Target: 1 or more unplanned in 12 months | ||
| Auroc (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) | ||||||
| HURT | Australia | XGBoost | IP,ED, OP, ABS | 206 714 (125 743) | 34 801 (27 216) | 74.9% (74.2 to 75.5) | 39.2% (38.2 to 40.2) | 90% (89.5 to 90.4) |
| Billings et al25 2013 | UK | Weighted score | IP, ED, OP, GP | N/A (1 836 099) | N/A (N/A) | 78%* (N/A) | 42%† (N/A) | N/A (N/A) |
| QAdmission Score26 | UK | Weighted score | IP, ED, OP, GP | 12 957 648 (4 870 488) | N/A (N/A) | Women: 77.3% (77.1 to 77.8) Men: 77.6% (77.4 to 77.8) | 39% | N/A |
| SPARRA V427 | UK | Ensemble (ANN, RF, XGB, GLM, NB) | IP, ED, OP, GP | 12 957 648 (4 870 488) | 4 300 000 (N/A) | 80.1% (SE: 0.023) | ~52% ‡ (N/A) | ~90%‡ (N/A) |
*Patient-level performance.
†Recall was 78%.
‡Estimated from figure 2 (a).27
ABS, Australian Bureau of Statistics; ANN, Artificial Neural Network; ED, emergency department; GLM, Generalised Linear Model; GP, general practitioner; HURT, Hospital Unplanned Readmission Tool; IP, inpatient; N/A, not available; NB, Naive Bayes; OP, outpatient; RF, Random Forest; XGBoost, Extreme Gradient Boosting.