Search

Unplanned hospital readmission is a problem that affects hospitals worldwide and is due to different factors. The identification of those factors can help determine which patients are at greater risk of hospital readmission for early intervention. Our end goal is to predict and identify patterns to (i) feed a decision support system for efficient management of patients and resources and (ii) detect patients at high risk of 30-days readmission enabling preventive actions to improve management of hospital discharges. This study aims to analyze whether natural language processing and specifically keyword extractions tools and sentiment analysis can support 30-days readmission prediction. Features extracted from medical history notes and discharge reports were used to train a Logistic Regression model. The resulting model obtains an AUC of 0.63 indicating that the sentiment polarity score of the discharge report and several of the extracted keywords are representative features to consider.

The usage of context open data, such as environmental data, socio-economic resources, etc., in clinical settings has been limited till now. But in recent years the number of studies that consider these factors is increasing. However, for clinicians, researchers or other stakeholders to be able to use this information in clinical practice, appropriate visualizations tools must be developed. At the same time, the data are generally displayed in formats that are difficult to interpret. The use of geolocalized visualizing tools through maps provides a better understanding in most cases. The proposed tool aims to contribute to the development of systems that can help to the exploitation of clinical and context open data, making them more easily interpretable.