Table 2.
AUC for prediction of acute kidney injury (AKI), reintubation, mortality, and any outcome with 95% CIs for the test set (N = 11,996) for the ASA score, logistic regression (LR) models, deep neural networks predicting individual outcomes (DNN individual), and deep neural networks predicting all three outcomes (DNN combined).
| Score | AKIa | Reintubation | Mortality | Any outcome |
|---|---|---|---|---|
| ASA | 0.652 (0.636–0.669) | 0.787 (0.757–0.818) | 0.839 (0.804–0.875) | 0.76 (0.748–0.773) |
| RQIb | 0.652 (0.623–0.683) | 0.878 (0.842–0.909) | 0.907 (0.86–0.942) | 0.8 (0.778–0.821) |
| RSIc | 0.594 (0.571–0.615) | 0.829 (0.783–0.873) | 0.97 (0.944–0.99) | 0.597 (0.576–0.621) |
| Model type | AKIa | Reintubation | Mortality | Any outcome (stacked model) |
|---|---|---|---|---|
| LR OFS | 0.767 (0.748–0.785) | 0.856 (0.82–0.888) | 0.9 (0.865–0.93) | 0.843 (0.829–0.857) |
| LR OFS + MAP | 0.767 (0.749–0.785) | 0.855 (0.818–0.887) | 0.898 (0.863–0.93) | 0.843 (0.829–0.857) |
| LR RFS | 0.767 (0.748–0.785) | 0.862 (0.827–0.894) | 0.899 (0.864–0.93) | 0.843 (0.829–0.858) |
| DNN Individual OFS | 0.78 (0.763–0.796) | 0.879 (0.851–0.905) | 0.895 (0.854–0.93) | 0.866 (0.855–0.878) |
| DNN Individual OFS + MAP | 0.792 (0.775–0.808) | 0.876 (0.848–0.902) | 0.903 (0.871–0.933) | 0.874 (0.864–0.886) |
| DNN Individual RFS | 0.783 (0.766–0.799) | 0.879 (0.851–0.905) | 0.9 (0.865–0.931) | 0.866 (0.854–0.878) |
| DNN Combined OFS | 0.785 (0.767–0.801) | 0.858 (0.829–0.886) | 0.907 (0.872–0.938) | 0.865 (0.854–0.877) |
| DNN Combined OFS + MAP | 0.783 (0.765–0.8) | 0.84 (0.808–0.872) | 0.906 (0.87–0.937) | 0.86 (0.848–0.872) |
| DNN Combined RFS | 0.789 (0.772–0.806) | 0.842 (0.811–0.871) | 0.906 (0.87–0.937) | 0.852 (0.84–0.864) |
Each model was also evaluated for each feature set combination of original feature set (OFS), OFS + the minimum MAP features (OFS + MAP), and reduced feature set (RFS). Note that for the LR and individual models, there is one model per outcome and the predicted outcome probabilities from each model is stacked to predict any outcome. For the combined models, there is one model for all three outcomes and those probabilities are stacked to predict any outcome. Bold results indicate the best AUC for that measure.
aIt should be noted that AKI labels were only available for 4307 of the test patients, and so all AUCs reflect results for only those patients with AKI labels.
bRQI was calculated on 5591 test patients (63 reintubation, 38 mortality, and 491 any label); and on 2319 test patients with AKI labels (445 positive).
cRSI was calculated on 11,939 test patients (159 reintubation, 86 mortality, and 1066 any label); and on 4294 test patients with AKI labels (967 positive).