BioBERT-RxReadmit: Improving Hospital Readmission Predictions Through Clinical Text Analysis with BioBERT

This study introduces BioBERT-RxReadmit, a dual stage model designed to predict hospital readmissions using unstructured clinical data from the MIMIC-III dataset. The model's name reflects its dual focus: leveraging BioBERT (Bidirectional Encoder Representations from Transformers for Biomedical Texts) for analyzing medical text and Rx, symbolizing prescription and medical intervention, to address readmission risks. In the first stage, BioBERT identifies key clinical features such as symptoms, diagnoses, and treatments from free-text clinical notes. These features are then integrated with the complete clinical text in the second stage, where BioBERT is fine-tuned for classification to predict 30-day readmissions. This comprehensive approach improves the model's ability to recognize complex patterns in patient data, resulting in improved predictive accuracy. BioBERT-RxReadmit helps identify high-risk patients more effectively, reducing preventable readmissions, optimizing healthcare resources, and improving patient care, showcasing the transformative potential of advanced NLP mod-els in healthcare.