Table 1.
Performance of the AKDa risk prediction models for elderly patients.
| Cohort and models | AUROCb (95% CI) | Cutoff | Sensitivity (95% CI) | Specificity (95% CI) | PPVc (95% CI) | NPVd (95% CI) | |
| Training cohort | |||||||
|
|
LRMe | 0.679 (0.660-0.698) | 0.749 | 0.701 (0.670-0.732) | 0.575 (0.556-0.594) | 0.353 (0.331-0.376) | 0.853 (0.836-0.869) |
|
|
XGBoostf | 0.756 (0.740-0.773) | 0.751 | 0.787 (0.758-0.813) | 0.605 (0.586-0.624) | 0.397 (0.374-0.421) | 0.895 (0.880-0.909) |
|
|
LightGBMg | 0.844 (0.831-0.857) | 0.724 | 0.788 (0.759-0.814) | 0.761 (0.745-0.777) | 0.522 (0.495-0.549) | 0.915 (0.903-0.927) |
|
|
MLPh | 0.734 (0.717-0.751) | 0.774 | 0.806 (0.778-0.832) | 0.539 (0.520-0.558) | 0.367 (0.345-0.389) | 0.894 (0.877-0.908) |
|
|
RFi | 0.814 (0.800-0.828) | 0.748 | 0.810 (0.783-0.836) | 0.671 (0.653-0.689) | 0.449 (0.425-0.474) | 0.914 (0.901-0.927) |
|
|
KNNj | 0.712 (0.694-0.730) | 0.789 | 0.742 (0.712-0.771) | 0.558 (0.539-0.577) | 0.357 (0.335-0.380) | 0.867 (0.850-0.883) |
| Internal validation cohort | |||||||
|
|
LRM | 0.669 (0.650-0.688) | 0.677 | 0.710 (0.679-0.740) | 0.566 (0.547-0.585) | 0.356 (0.333-0.378) | 0.853 (0.835-0.869) |
|
|
XGBoost | 0.684 (0.665-0.703) | 0.657 | 0.614 (0.582-0.647) | 0.663 (0.645-0.681) | 0.381 (0.356-0.407) | 0.836 (0.820-0.852) |
|
|
LightGBM | 0.853 (0.841-0.865) | 0.722 | 0.817 (0.791-0.842) | 0.759 (0.742-0.775) | 0.534 (0.507-0.560) | 0.925 (0.913-0.936) |
|
|
MLP | 0.719 (0.701-0.737) | 0.739 | 0.751 (0.722-0.779) | 0.587 (0.568-0.606) | 0.380 (0.357-0.403) | 0.875 (0.859-0.890) |
|
|
RF | 0.823 (0.809-0.837) | 0.745 | 0.844 (0.819-0.868) | 0.653 (0.634-0.671) | 0.450 (0.426-0.475) | 0.926 (0.913-0.937) |
|
|
KNN | 0.692 (0.674-0.711) | 0.789 | 0.731 (0.701-0.760) | 0.552 (0.532-0.571) | 0.355 (0.333-0.377) | 0.859 (0.841-0.875) |
| External validation cohort | |||||||
|
|
LRM | 0.763 (0.707-0.818) | 0.787 | 0.830 (0.738-0.899) | 0.586 (0.512-0.658) | 0.503 (0.422-0.584) | 0.872 (0.800-0.925) |
|
|
XGBoost | 0.736 (0.678-0.794) | 0.825 | 0.809 (0.714-0.882) | 0.613 (0.539-0.683) | 0.514 (0.430-0.596) | 0.864 (0.793-0.917) |
|
|
LightGBM | 0.755 (0.699-0.811) | 0.899 | 0.851 (0.763-0.916) | 0.597 (0.523-0.668) | 0.516 (0.435-0.597) | 0.888 (0.819-0.937) |
|
|
MLP | 0.724 (0.665-0.784) | 0.764 | 0.702 (0.599-0.792) | 0.683 (0.611-0.749) | 0.528 (0.437-0.618) | 0.819 (0.750-0.876) |
|
|
RF | 0.749 (0.692-0.806) | 0.778 | 0.798 (0.702-0.874) | 0.645 (0.572-0.714) | 0.532 (0.446-0.616) | 0.863 (0.795-0.916) |
|
|
KNN | 0.632 (0.566-0.699) | 0.789 | 0.670 (0.566-0.764) | 0.527 (0.453-0.600) | 0.417 (0.338-0.500) | 0.760 (0.677-0.831) |
aAKD: acute kidney disease.
bAUROC: area under the receiver operating characteristic curve.
cPPV: positive predictive value.
dNPV: negative predictive value.
eLRM: logistic regression model.
fXGBoost: Extreme Gradient Boosting.
gLightGBM: Light Gradient Boosting Machine.
hMLP: multilayer perceptron.
iRF: random forest.
jKNN: K-nearest neighbor.