Table 1.
Machine learning studies on acute kidney injury (AKI) prediction.
| Study | Design | Setting | N | AKI Definition | Timing of AKI | ML Algorithm | Predictive Ability | Sensitivity; Specificity; Confidence Interval | External Validation |
|---|---|---|---|---|---|---|---|---|---|
| Kate et al. (2016) | retrospective | medical and surgical | 25,521 | AKIN | during hospitalization | naïve Bayes; support vector machine; decision trees; logistic regression |
AUROC 0.654 AUROC 0.621 AUROC 0.639 AUROC 0.660 |
75%; 61%; - |
no |
| Thottakkara et al. (2016) | retrospective | surgical | 50,318 | KDIGO | post surgery | naïve Bayes; generalized additive model; logistic regression; support vector machine |
AUROC 0.819 AUROC 0.858 AUROC 0.853 AUROC 0.857 |
77%; -; - |
no |
| Davis et al. (2017) | retrospective | medical and surgical | 2003 | KDIGO | during hospitalization | random forest; neural network; naïve Bayes; logistic regression |
AUROC 0.730 AUROC 0.720 AUROC 0.690 AUROC 0.780 |
-; -; 95% CI |
no |
| Cheng et al. (2018) | retrospective | medical and surgical | 60,534 | KDIGO, AKIN, RIFLE | during hospitalization | random forest; AdaBoostM1; logistic regression |
AUROC 0.765 AUROC 0.751 AUROC 0.763 |
69%; 71%; - |
no |
| Ibrahim et al. (2018) | prospective | contrast | 889 | KDIGO | pre and post intervention | logistic regression | AUROC 0.790 | 77%; 75%: - |
no |
| Koola et al. (2018) | retrospective | medical and surgical | 504 | KDIGO | during hospitalization | logistic regression; naïve Bayes; support vector machines; random forest; gradient boosting |
AUROC 0.930 AUROC 0.730 AUROC 0.900 AUROC 0.910 AUROC 0.880 |
87%; 76%; - |
no |
| Lin et al. (2018) | retrospective | ICU | 19,044 | KDIGO | during hospitalization | support vector machine | AUROC 0.860 | - | no |
| Koyner et al. (2018) | retrospective | medical and surgical | 121,158 | KDIGO | 24 h post admission | gradient boosting | AUROC 0.900 | 95% CI | no |
| Huang et al. (2018) | retrospective | PCI | 947,091 | AKIN | during hospitalization | gradient boost; logistic regression |
AUROC 0.728 AUROC 0.717 |
-; -; 95% CI |
no |
| Huang et al. (2019) | retrospective | PCI | 2,076,694 | AKIN | pre and post intervention | generalized additive model | AUROC 0.777 | -; -; 95% CI |
no |
| Tomašev et al. (2019) | retrospective | medical and surgical | 703,782 | KDIGO | during hospitalization | recurrent neural network | AUROC 0.921 | 95%; 70.3%; - |
no |
| Adhikari et al. (2019) | retrospective | surgical | 2901 | KDIGO | post surgery | random forest | AUROC 0.860 | 68%; -; - |
no |
| Flechet et al. (2019) | prospective | ICU | 252 | KDIGO | during hospitalization | random forest | AUROC 0.780 | -; -; 95% CI |
no |
| Parreco et al. (2019) | retrospective | medical and surgical | 151,098 | KDIGO | during hospitalization | gradient boosting; logistic regression; deep learning |
AUROC 0.834 AUROC 0.827 AUROC 0.817 |
-; -; 95% CI |
no |
| Xu et al. (2019) | retrospective | medical and surgical | 58,976 | KDIGO | during hospitalization | gradient boosting | AUROC 0.749 | - | no |
| Tran et al. (2019) | prospective | burn | 50 | KDIGO | during hospitalization | k-nearest neighbor | AUROC 0.920 | 90%; -; - |
no |
| Zhang et al. (2019) | retrospective | ICU | 6682 | KDIGO | 24 h post admission | gradient boosting | AUROC 0.860 | -; -; 95% CI |
no |
| Zimmerman et al. (2019) | retrospective | ICU | 46,000 | KDIGO | 72 h post admission | logistic regression; random forest; neural network |
AUROC 0.783 AUROC 0.779 AUROC 0.796 |
68%; 34%; - |
no |
| Rashidi et al. (2020) | retrospective and prospective | burn and trauma | 50/51 | KDIGO vs New Biomarkers | 1st week post ICU admission | recurrent neural network | AUROC 0.920 | -; -; 92% CI |
no |
AKI-acute kidney injury, AKIN-acute kidney injury network, AUROC-area under the receiver operating characteristic curve, ICU-intensive care unit, KDIGO-kidney disease improving global outcomes, ML-machine learning, PCI-percutaneous coronary intervention, RIFLE-risk, injury, failure, loss of kidney function, end-stage kidney disease.