Research on a Hypotension Prediction Model for Hemodialysis Based on LSTM Algorithm

Abstract

Addressing the challenge of predicting hypotension during hemodialysis, this study proposes a forecasting model based on Long Short-Term Memory (LSTM) networks. By integrating multidimensional physiological data collected during hemodialysis—including blood pressure, heart rate, and ultrafiltration volume—the model achieves efficient prediction of hypotensive events. Compared to traditional statistical models, LSTM better captures temporal features in the data, thereby reducing prediction errors. Experimental results demonstrate that the proposed LSTM model achieves accuracies exceeding 88%, 85%, and 85% on the training, validation, and test sets respectively, with an F1-score approaching 0.81 and an AUC of 0.91. These metrics outperform other common algorithms such as Support Vector Machines (SVM) and Decision Trees (DT). These findings validate the model's potential for hemodialysis hypotension prediction, providing reliable technical support for clinical decision-making.