Deep Learning Model for Real-Time Prediction of Intradialytic Hypotension.

Area of Science:

• Nephrology and Artificial Intelligence
• Machine Learning in Healthcare
• Predictive Analytics in Medicine

Background:

• Intradialytic hypotension (IDH) poses significant clinical challenges.
• Traditional statistical models struggle to predict IDH due to complex, interactive risk factors.
• Accurate prediction of IDH is crucial for patient safety during hemodialysis.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting the risk of intradialytic hypotension.
• To compare the performance of a recurrent neural network (RNN) against conventional models for IDH prediction.
• To leverage time-varying vital signs for real-time IDH risk assessment.

Main Methods:

• Utilized a large dataset of 261,647 hemodialysis sessions with over 1.6 million timestamps.
• Defined intradialytic hypotension based on systolic blood pressure and mean arterial pressure thresholds.
• Trained and tested a recurrent neural network model, comparing its performance (AUC, PRC, F1 score) against logistic regression, MLP, and LightGBM.

Main Results:

• The RNN model achieved superior performance in predicting intradialytic hypotension (IDH 1: AUC 0.94, IDH 2: AUC 0.87, IDH 3: AUC 0.79).
• RNN significantly outperformed other models across all evaluated metrics, including Area Under the ROC Curve (AUC), Area Under the Precision-Recall Curve (AUPRC), and F1 scores.
• The developed RNN models demonstrated high calibration for predicting intradialytic hypotension.

Conclusions:

• Deep learning, specifically RNNs, offers a powerful tool for predicting intradialytic hypotension.
• The proposed model enables real-time risk prediction of intradialytic hypotension.
• This approach can enhance patient management and safety during hemodialysis treatments.