Deep Learning for 30-Day Hospitalization Prediction in Dialysis Patients

This study delves into the application of Deep Learning (DL) for the prediction of hospitalization in dialysis patients within a 30-day timeframe. We comprise multivariate time series measurements from dialysis sessions, which hold significant value for individuals with diminished kidney function. This study conducted an analysis involving various models, including random forest, k-nearest neighbour, and three neural network models (feedforward, recurrent, and convolutional). The primary objective was to ascertain the most effective model for forecasting hospitalization. While the Feedforward Neural Network (FNN) demonstrated the highest performance among the neural network models, the random forest model surpassed all others, boasting an AUC of 0.692. Therefore, this study underscores the potential of DL in enhancing patient outcomes and optimizing resource allocation. Through the anticipation of hospitalizations, healthcare providers can proactively mitigate suffering and minimize costs.