Abstract
Recently natural language processing (NLP) tools have been developed to identify important risk factors in electronic health records (EHRs). Sentiment analysis, a field of NLP defined as an individual’s attitude towards a topic, has been widely used in non-medical industries for improving decision making. In this study, we undertook a first-of-its-kind annotation project to extend sentiment analysis to psychiatry, defining it as a clinician’s prognosis of a patient at a fixed time point. The sentiment analysis models trained using the results from this project will be incorporated in a feature engineering pipeline for a machine learning model that predicts inpatient readmission risk.