Machine learning model to predict hypotension after starting continuous renal replacement therapy.

Area of Science:

• Nephrology
• Critical Care Medicine
• Data Science

Background:

• Hypotension following continuous renal replacement therapy (CRRT) initiation is linked to adverse patient outcomes.
• Predicting CRRT-induced hypotension is challenging due to complex, interacting risk factors.
• Accurate prediction is crucial for timely intervention and mitigating negative consequences.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting hypotension post-CRRT initiation.
• To compare the predictive performance of ML models against established disease-severity scores.

Main Methods:

• Utilized a dataset of 2349 adult patients with acute kidney injury starting CRRT.
• Trained and tested ML models including Support Vector Machine (SVM), Deep Neural Network (DNN), Light Gradient Boosting Machine (LGBM), and Extreme Gradient Boosting (XGB).
• Defined hypotension as a mean arterial pressure (MAP) reduction ≥ 20 mmHg within 6 hours; performance assessed using Area Under the Receiver Operating Characteristic Curves (AUROCs).

Main Results:

• The XGB model achieved the highest AUROC (0.828), outperforming other ML models and traditional scores.
• DNN and LGBM models also demonstrated high predictive accuracy (AUROCs 0.822 and 0.813, respectively).
• All ML models significantly outperformed disease-severity scores (e.g., SOFA, APACHE II) with AUROCs < 0.6.

Conclusions:

• Machine learning models, particularly XGBoost, demonstrate superior ability to predict hypotension after CRRT initiation.
• These ML models offer a significant advancement over conventional scoring systems for hypotension risk assessment.
• The developed models can form the basis for clinical decision support systems to predict and prevent CRRT-related hypotension.