Table 2.
Comparison of model performance among the three data sets and statistical approaches. HJ23 had the highest model performance, yet lowest cohort size, while the opposite is true for eICU-CRD. For MIMIC and eICU-CRD, model performance is similar across all three methods. AUROC area under the receiver operator curve. 95% Confidence Intervals are provided in brackets.
| Dataset | Cohort size | Methods | Accuracy | AUROC |
|---|---|---|---|---|
| eICU | 3394 | Logistic regression | 78.65 (75.56–81.73) | 87.36 (84.86–89.86) |
| Random forest | 77.29 (74.13–80.44) | 84.36 (81.65–87.11) | ||
| Partial least square | 77.32 (74.17–80.47) | 83.70 (80.93–86.48) | ||
| MIMIC | 1295 | Logistic regression | 80.31 (75.47–85.15) | 87.41 (83.37–91.45) |
| Random forest | 80.23 (75.38–85.08) | 87.11 (83.03–91.19) | ||
| Partial least square | 79.54 (74.62–84.45) | 87.06 (82.97–91.15) | ||
| HJ23 | 172 | Logistic regression | 97.14 (91.54–100) | 98.69 (94.87–100) |
| Random forest | 80.00 (66.55–93.45) | 89.87 (79.73–100) | ||
| Partial least square | 88.57 (77.88–99.27) | 97.67 (92.59–100) |