Table 3.
AKI prediction model for all care settings.
| Reference | Modeling data sources | Data volume | Model performance AUC value | External verification | Machine learning methods | Diagnostic criteria | ||||
|---|---|---|---|---|---|---|---|---|---|---|
| Data source | Data volume | Model performance (AUROC) | ||||||||
| Koyner et al. (29) | University of Chicago Hospital, US | 121,158 | 24 h in advance | 0.90 | No | Gradient boosting machine | KDIGO: Creatinine | |||
| 48 h in advance | 0.87 | |||||||||
| Dialysis 48 h in advance | 0.96 | |||||||||
| Mohamadlou et al. (32) | Stanford Medical Center, US | 48,582 | Onset of illness | 0.872 | MIMIC-III | 19,737 | Onset of illness | 0.841 | Boosting decision trees | NHS-AKI: Creatinine |
| 12 h in advance | 0.800 | 12 h in advance | 0.749 | |||||||
| 24 h in advance | 0.795 | 24 h in advance | 0.758 | |||||||
| 48 h in advance | 0.761 | 48 h in advance | 0.707 | |||||||
| 72 h in advance | 0.728 | 72 h in advance | 0.674 | |||||||
| He et al. (33) | University of Kansas Health System—KUHS | 96,590 | 24 h in advance | 0.744 | No | Logistic regression Random forest | KDIGO: Creatinine | |||
| Any time | 0.734 | |||||||||
| Day 1 after admission | 0.764 | |||||||||
| Day 2 after admission | 0.764 | |||||||||
| Kate et al. (31) | Aurora Health Care (15) hospitals | 25,521 | 0.743 | No | Logistic regression Support vector machines Decision trees Naive Bayes |
AKIN: Creatinine | ||||
| Cronin et al. (28) | US Department of Veterans Affairs | 1,620,898 | AKI I | 0.746–0.758 | No | Logistc regression LASSO regression Random forest | KDIGO: Creatinine | |||
| AKI II | 0.714–0.720 | |||||||||
| Dialysis | 0.823–0.825 | |||||||||
| Churpek et al. (34) | University of Chicago (UC) | 48,463 | 48 h in advance AKI II | 0.86 | NorthShore University Health System (NUS) | 246,895 | 0.86 | Gradient boosting machine | KDIGO: Creatinine | |
| Loyola University Medical Center (LUMC) |
200,613 | 0.85 | ||||||||
| Kim et al. (35) | Seoul National University Bundang Hospital, South Korea | 69,081 | AKI I | 0.88 | Seoul National University Hospital, South Korea | 72,352 | AKI I | 0.84 | Recurrent neural network Extreme gradient boosting |
KDIGO: Creatinine |
| AKI II | 0.93 | AKI II | 0.90 | |||||||
| Cheng et al. (36) | University of Kansas Medical Center - KUMC | 48,955 | 0.76 | No | Logistic regression Random forest Adaptive boosting |
KDIGO: Creatinine | ||||
| Tomasev et al. (30) | US Department of Veterans Affairs | 703,782 | Any AKI | 0.921 | No | Recurrent neural network | KDIGO: Creatinine | |||
| AKI II III | 0.957 | |||||||||
| AKI III | 0.980 | |||||||||
KDIGO, Kidney Disease: Improving Global Outcomes; AKIN, Acute Kidney Injury Network. NHS, National Health Service; MIMIC-III, Medical Information Mart for Intensive Care III.