PREDHIC: Predicting heart failure readmission and mortality using natural language processing

While natural language processing performance has increased in recent years from a technical point of view, it is important to investigate how it contributes to a decision-making task that helps a specialist based upon information found in text. In that context, the specialized domain limits the availability of in-domain training data and emphasizes the importance of terminology and a priori knowledge; and typically not only text, but also structured data need to be taken into account. PREDHIC investigates associated research directions in state-of-the-art neural text representation and classification methods, and evaluates their impact on a real task, with real data, with a multicentric design: the assessment of the risk of readmission and mortality of heart failure patients after hospital discharge, a question of high clinical interest. It associates NLP specialists from two computer science teams to medical information and heart failure specialists from two hospitals.