Including urinary output to define AKI enhances the performance of machine learning models to predict AKI at admission.

PURPOSE

Acute kidney injury (AKI) is a prevalent and detrimental condition in intensive care unit patients. Most AKI predictive models only predict creatinine-triggered AKI (AKI) and might underperform when predicting urine-output-triggered AKI (AKI). We aimed to describe how admission AKI prediction models perform in all AKI patients.

MATERIALS AND METHODS

Three types of models were trained: 1) pAKI, predicting AKI based on creatinine or urine output, 2) pAKI, predicting AKI based only on urine output, and 3) pAKI, predicting AKI based only on creatinine. We compared model performance and predictive features.

RESULTS

The pAKI models had the best overall performance (AUROC 0.673-0.716) and the most consistent performance across three patient cohorts grouped by type of AKI trigger (min AUROC of 0.636). The pAKI models had fair performance in predicting AKI (AUROCs 0.702-0.748) but poor performance predicting AKI (AUROCs 0.581-0.695). The predictive features for the pAKI models and pAKI models were distinct, while top features for the pAKI models were consistently a combination of those for the pAKI and pAKI models.

CONCLUSION

Ignoring urine output in the outcome during model training resulted in models that are unlikely to predict AKI adequately and may miss a substantial proportion of patients in practice.