Domain Specific Transformers-Based Prioritization of Readmission for Patients in Healthcare

Abstract

Hospital re-admissions are not only resource-intensive but also a key indicator of healthcare quality. Therefore, there is a critical need to enhance the efficiency of healthcare systems by accurately identifying patients at a higher risk of readmission. While several prediction models were built before for predicting the re-admission of patients, no research study focused on the prioritization of patient readmissions. Thus, in this research, we focused on developing a predictive model to prioritize the readmissions of patients based on their discharge summaries. To enhance the model's performance and address class imbalance problems in this dataset, various data balancing techniques were investigated. BioBERT was utilized for feature extraction during model training. The capabilities of Optuna framework were leveraged for hyperparameter optimization. By utilizing this framework, we significantly reduced training time and computational resources. The predictive model exhibits remarkable performance when evaluated on the testing dataset. The experimental results signify the model's potential to effectively prioritize patient re-admissions, thereby contributing to more informed and efficient healthcare decision-making processes.