Multilayer Perceptron-Based Real-Time Intradialytic Hypotension Prediction Using Patient Baseline Information and...

Area of Science:

• Nephrology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Intradialytic hypotension (IDH) is a frequent complication during hemodialysis, posing significant risks to patients.
• Existing IDH prediction methods often lack real-time applicability due to reliance on static patient data.

Purpose of the Study:

• To develop and validate a real-time IDH prediction model using time-series heart rate (HR) data and static patient information.
• To investigate the impact of HR differences and HR slope on the accuracy of real-time IDH prediction.

Main Methods:

• A multilayer perceptron (MP) model was developed incorporating time-series HR data (60 minutes prior to IDH) and static data (age, diabetes, hypertension, ultrafiltration).
• Clinical data from 80 hemodialysis patients were analyzed, with each patient undergoing approximately 12 sessions.
• Model performance was evaluated based on accuracy, true positive rate, and positive predictive value.

Main Results:

• The optimal MP model, utilizing 60 minutes of HR data and two hidden layers, achieved an accuracy of 81.5%.
• The model demonstrated a true positive rate of 73.8% and a positive predictive value of 87.3% for predicting IDH.
• HR differences and HR slope were identified as significant factors influencing real-time IDH prediction.

Conclusions:

• The proposed MP-based model effectively predicts intradialytic hypotension in real-time using continuous HR monitoring.
• This approach offers a promising tool to support medical staff in managing hemodialysis patients and preventing IDH.
• Future improvements may involve incorporating additional parameters to further enhance prediction accuracy.