NLP Based Prediction of Hospital Readmission using ClinicalBERT and Clinician Notes

Abstract

Hospital readmissions introduce a significant challenge in healthcare, leading to increased costs, reduced patient outcomes, and strained healthcare systems. Accurately predicting the risk of hospital readmission is crucial for implementing targeted interventions and improving patient care. This study investigates the use of natural language processing (NLP) techniques, specifically the ClinicalBERT model, to predict the risk of hospital readmission using the first 3-5 days of clinical notes, excluding discharge notes. We compare the performance of ClinicalBERT to other machine learning models, including logistic regression, random forest, and XGBoost, to identify the most effective approach for this task. This study highlights the potential of leveraging deep learning-based NLP models in the clinical domain to improve patient care and reduce the burden of hospital readmissions, even when utilizing only the initial clinical notes from a patient's hospitalization. It can also provide information early to allow Clinicians to intervene in patients who are at high risk. The results demonstrate that the ClinicalBERT model outperforms the other techniques, achieving higher accuracy, F1-score, and area under the receiver operating characteristic (ROC) curve. This study highlights the potential of leveraging deep learning- based NLP models in the clinical domain to improve patient care and reduce the burden of hospital readmissions.