Skip to main content

Table 3 Evaluation metrics on the holdout test datasets of all hospitals aggregated across the prediction timeframe for the trained model and renal angina index (RAI)

From: Machine learning model for early prediction of acute kidney injury (AKI) in pediatric critical care

 

Sensitivity

PPV (TP:FP)

Any Stage AKI (%)

AKI Stage 2 or 3 (%)

Any stage AKI

AKI Stage 2 or 3

Model

41

58

47% (1:1)

22% (1:4)

RAI

5

3

4% (1:21)

1% (1:98)

  1. Evaluation metrics were calculated from AKI alerts, which were triggered when the model’s predicted AKI risk crossed the pre-determined risk threshold. Sensitivity is computed as the percentage of AKI episodes that were identified by the model among target AKI episodes. Positive predictive value (PPV) is shown alongside true positive (TP) to false positive (FP) ratios for clarity
Back to article page

Critical Care

ISSN: 1364-8535