Table 3. Acute Kidney Injury Risk as Predicted by Models That Add and Do Not Add Intraoperative Data in Test Data Seta.
GBM Preoperative Model | No. (%) | ||
---|---|---|---|
GBM-Perioperative Modela | Total, No. | ||
Low Risk | High Risk | ||
Low Riskb | |||
Encounters | 6414 (94.4) | 381 (5.6) | 6795 |
Events | 283 (80.9) | 67 (19.1) | 350 |
Nonevents | 6131 (95.1) | 314 (4.9) | 6445 |
Proportion of encounters with events | 0.044 | 0.176 | 0.052 |
High Riskb | |||
Encounters | 381 (22.4) | 1318 (77.6) | 1699 |
Events | 52 (10.5) | 443 (89.5) | 495 |
Nonevents | 329 (27.3) | 875 (72.7) | 1204 |
Proportion of encounters with events | 0.136 | 0.336 | 0.291 |
Abbreviation: GBM, gradient boosting machine.
Risk stratification of GBM models in the test set for the outcome of acute kidney injury using preoperative and perioperative data in the test data set (nā=ā8494). For the GBM model using the perioperative model, the overall proportion of encounters with events was 0.300 and 0.049 for high- and low-risk groups, respectively.
High risk was defined as the top 20% of predicted risk. Low risk was defined as the bottom 80% of predicted risk.