Combining Machine Learning and Urine Oximetry: Towards an Intraoperative AKI Risk Prediction Algorithm

Area of Science:

• Nephrology
• Cardiology
• Medical Technology

Background:

• Acute kidney injury (AKI) is a common complication in cardiac surgery patients, affecting up to 50%.
• Current AKI diagnostic methods (serum creatinine, urine output) result in delayed detection.
• Monitoring the partial pressure of oxygen in urine (PuO₂) offers a potential dynamic assessment of AKI risk.

Purpose of the Study:

• To evaluate the predictive performance of two machine learning algorithms for AKI in cardiac surgery.
• To compare a model incorporating PuO₂ monitoring with a model using only preoperative risk factors.
• To determine if intraoperative PuO₂ enhances AKI prediction accuracy.

Main Methods:

• Developed two machine learning models to predict AKI.
• One model integrated a feature derived from intraoperative PuO₂ monitoring.
• The other model exclusively used preoperative risk factors, selected via automated forward variable selection.

Main Results:

• The model incorporating PuO₂ data achieved a significantly higher area under the receiver operator characteristic curve (AUROC) of 0.78.
• The model relying solely on preoperative factors had an AUROC of 0.66.
• The difference in AUROC between the two models was statistically significant (p < 0.01).

Conclusions:

• Intraoperative PuO₂ monitoring, when integrated into a machine learning model, significantly improves the prediction of AKI in cardiac surgery patients.
• PuO₂-based models offer a more dynamic and potentially earlier assessment of AKI risk compared to traditional preoperative risk factor analysis.
• This approach may lead to timely interventions and improved patient outcomes in cardiac surgery settings.