Predicting the 30-day readmissions of congestive heart failure patients from diagnosis records

Congestive heart failure (CHF) poses a major challenge to public health worldwide, marked by high readmission rates. As a result, early and accurate prediction and proactive intervention are crucial for effectively reducing readmission rates. Electronic health records (EHRs) serve as an essential resource for accessing patient health data and play a vital role in medical research. This study proposed a deep learning model named DeepCHF-RH to predict 30-day readmissions for patients with CHF using EHRs. Initially, we gathered clinical data from 13,018 CHF patients sourced from the Medical Intensive Care Database (MIMIC-IV). We then presented global and local feature extraction methods to identify the relevant features. Finally, the multilayer perceptron (MLP) served as the classifier to predict the likelihood of 30-day readmission. Experimental results showed that our model, DeepCHF-RH, achieved 90.27% accuracy and 97.07% AUC, surpassing all state-of-the-art prediction models across every evaluation metric.