Enhancing Predictive Accuracy in Healthcare Readmission Rates: A Hybrid Deep Learning Approach Leveraging Temporal and Clinical Data Fusion

Abstract

Healthcare readmission rates represent a significant challenge for healthcare systems worldwide, impacting both patient outcomes and operational efficiency. This study proposes a novel hybrid deep learning approach that integrates temporal and clinical data to enhance the accuracy of readmission prediction. We leverage Long Short-Term Memory (LSTM) networks to model the temporal dynamics of patient health records and transformer networks to capture complex relationships within clinical features. Furthermore, we introduce a data fusion technique to effectively combine the learned representations from both modalities. Our experimental results, conducted on a large real-world healthcare dataset, demonstrate that the proposed hybrid model significantly outperforms traditional machine learning methods and existing deep learning approaches, providing a more accurate and robust solution for predicting healthcare readmissions. The insights gained from this research can contribute to improved resource allocation, personalized patient care, and ultimately, a reduction in preventable readmissions.