Leveraging ClinicalBERT for Hospital Readmission Prediction Using Clinical Notes and Discharge Summaries
Bioinformatics Data Science PhD student
University of Delaware
Hospital readmissions pose a significant challenge in healthcare systems, leading to increased healthcare costs and patient burden. Accurate prediction of hospital readmissions can enable targeted interventions and improved healthcare outcomes. In this study, we explore the effectiveness of ClinicalBERT, a domain-specific language representation model for clinical text, on the MIMIC-III dataset. The aim is to predict 30-day hospital readmission at various time points of admission. Clinical notes and discharge summaries are preprocessed to extract relevant features for the readmission task. We compare the performance of ClinicalBERT with other machine learning models and evaluate its ability to accurately predict readmission using only unstructured data. The results demonstrate that ClinicalBERT exhibits strong predictive capabilities for hospital readmission. This study showcases the potential of utilizing ClinicalBERT for hospital readmission prediction using clinical notes and discharge summaries from the MIMIC-III dataset. The findings suggest that leveraging domain-specific language representations can significantly improve the accuracy of predictive models in the clinical domain, aiding healthcare providers in making informed decisions and reducing the rate of preventable hospital readmissions.
Omar is a Ph.D. student in the Bioinformatics Data Science program at UD. He received his bachelor’s degree in health informatics from King Saud University, Saudi Arabia. He also received a master’s degree in health informatics from the University of Pittsburgh. Currently, he is doing his research under the supervision of Dr. Vijay Shanker and Dr. Cathy Wu.