Table 3 PO-AKI Prediction Performance (UCIMC)
From: InfEHR: Clinical phenotype resolution through deep geometric learning on electronic health records
Model | Default (0.5) | Optimal Rule-In | Optimal Rule-Out |
|---|---|---|---|
Clinical Heuristic | |||
Threshold | 0.500 | - | - |
Sensitivity | 0.131 | - | - |
Specificity | 0.994 | - | - |
PPV | 0.717 | - | - |
NPV | 0.904 | - | - |
Rule-In | 1.330 | - | - |
Rule-Out | 1.124 | - | - |
InfEHR prior | |||
Threshold | 0.500 | - | - |
Sensitivity | 0.540 | - | - |
Specificity | 0.982 | - | - |
PPV | 0.783 | - | - |
NPV | 0.947 | - | - |
Rule-In | 4.139 | - | - |
Rule-Out | 1.898 | - | - |
GRU-D | |||
Threshold | 0.500 | 0.253 | 0.061 |
Sensitivity | 0.605 | 0.739 | 0.854 |
Specificity | 0.973 | 0.964 | 0.888 |
PPV | 0.728 | 0.715 | 0.479 |
NPV | 0.953 | 0.968 | 0.981 |
Rule-In | 4.745 | 6.706 | 4.354 |
Rule-Out | 2.269 | 3.172 | 3.964 |
SeFT | |||
Threshold | 0.500 | 0.202 | 0.081 |
Sensitivity | 0.510 | 0.759 | 0.839 |
Specificity | 0.980 | 0.967 | 0.909 |
PPV | 0.751 | 0.733 | 0.525 |
NPV | 0.943 | 0.971 | 0.979 |
Rule-In | 3.415 | 7.671 | 4.861 |
Rule-Out | 1.778 | 3.447 | 3.935 |
InfEHR GNN | |||
Threshold | 0.500 | 0.384 | 0.195 |
Sensitivity | 0.655 | 0.730 | 0.829 |
Specificity | 0.979 | 0.975 | 0.896 |
PPV | 0.789 | 0.780 | 0.492 |
NPV | 0.959 | 0.968 | 0.978 |
Rule-In | 7.105 | 8.322 | 4.230 |
Rule-Out | 2.700 | 3.217 | 3.566 |
