This repository contains the code and notebooks for the MSc thesis in Statistics and Data Science at University Carlos III of Madrid (UC3M).
Prediction of unplanned readmission cases using Natural Language Processing and Deep Learning Architectures
Unplanned hospital readmission is one of the main concerns for health care institutions, since it represents a patient's exposure to risk and a preventable waste of medical resources. Automatically identifying patients with a higher risk of Intensive Care Unit (ICU) readmission would help to efficiently reallocate the medical resources that would be unnecessary used during a readmission.
The analysis and prediction of unplanned ICU readmission will be studied by means of Machine Learning (ML), Deep Learning (DL) and Natural Language Processing (NLP) techniques. Including state-of-the-art approaches like transfer learning with pre-trained models. In particular, caregivers' text notes about patients discharge at the ICU will be analysed. For that purpose, de-identified database MIMIC-III (‘Medical Information Mart for Intensive Care’), of limited access provided by the MIT Lab for Computational Physiology, will be used. This large, single-center data-set includes high-resolution information associated to 61532 patients admitted to critical care units of the Beth Israel Deaconess Medical Center between 2001 and 2012.
Based on this data analysis, a performance comparison of the different DL-based strategies will be presented. Results show that simpler statistical models yield the best performance in comparison to DL-based models.