Exploring the effect of discharge summaries for the prediction of 30-day unplanned patient readmission to the ICU

Healthcare is transforming into a data-intensive industry with the expectation to double its own data every 73 days by 2020. Electronic Health Records hold a vast amount of information that has the potential of improving care delivery ranging from management tasks in hospitals to inferring diagnoses from X-ray images. The massive volume of data, such as demographic data, diagnoses, tests, prescribed medications, and procedures, can be used to predict health risk or diagnose diseases. But few pay attention to the medical notes which contain abundant and critical information written by healthcare service providers during a patient’s stay or visit to the hospital. Because of the unstructured feature in these notes, they are usually underutilized to build prediction models. This project incorporates medical notes (e.g., discharge notes) along with demographic data available in the MIMIC-III dataset, to visualize patterns and finally train a prediction model for readmission of patients in the ICU.