JMIR Publications

ABSTRACT

Background:

Acute kidney injury (AKI) is common in ICU patients and has significant impact on morbidity and mortality. AKI diagnosis is based on elevated serum creatinine and decreased urine output which is seen first after renal injury. There are no treatments to reverse or restore renal function once AKI has developed. Early prediction of AKI enables proactive management and may improve patient outcomes.

Objective:

To develop a machine learning algorithm for the prediction of Acute Kidney Injury (AKI), based on routinely collected intensive care unit data, designed to be implemented in European intensive care units.

Methods:

The machine learning algorithm was developed using an ensemble model consisting of a Random Forest and XGBoost, based on the Massachusetts Institute of Technology Lab for Computational Physiology MIMIC-IV Clinical Database, focusing on intensive care unit patients aged 18 years or older. Twenty-two variables are used for hourly predictions of AKI, defined in accordance with kidney disease: Improving Global Outcomes (KDIGO) guidelines.

Results:

The developed algorithm NAVOY® Acute Kidney Injury uses four hours of input and can with high accuracy predict patients with high risk of developing AKI. The prediction performance competes well with previously published prediction algorithms designed to predict AKI onset in accordance with KDIGO diagnosis criteria, as measured by the area under the receiver operating characteristics curve (AUROC) and the area under the precision-recall curve (AUPRC). NAVOY® Acute Kidney Injury yields AUROC = 0.91 and AUPRC = 0.75 for predictions 12 hours before AKI onset. The predictive performance is externally validated on hold-out test data, where NAVOY® Acute Kidney Injury is confirmed to predict AKI with high accuracy.

Conclusions:

NAVOY® Acute Kidney Injury has excellent predictive properties and only uses variables routinely collected in intensive care units.