Table 3. Results of the Random Forest model.
Model | Dataset | Stays | % of AKI | auROC [iqr] | Sensitivity | Specificity | Median lead time (h) (AKI after the first 24h of ICU stay) | LR+ | LR- |
---|---|---|---|---|---|---|---|---|---|
Random Forest | Internal Test | 1’749 | 15.2 | 0.884 [0.864–0.905] | 82.0% | 80.0% | 15.5 | 4.10 | 0.24 |
Random Forest | eICU | 6’985 | 15.5 | 0.877 [0.868–0.888] | 83.0% | 81.3% | 14.0 | 4.44 | 0.23 |
Random Forest | Margherita Tre | 1’025 | 6.1 | 0.911 [0.882.0.936] | 82.5% | 83.2% | 23.0 | 4.91 | 0.20 |
auROC = area under receiving operator curve; LR + = likelihood positive ratio; LR- = likelihood negative ratio.