AI System for the Swiss Cantonal Psychiatric Clinic Call Center to Help DetermineHospitalization Needs

AI systems are increasingly popular across various sectors of society. Many real-life problems have been addressed through AI, such as personalized medicine in healthcare, extreme weather prediction in environmental science, fraud detection in finance, and so on. Our research proposes to develop an AI system that assists the Swiss cantonal psychiatric clinic (OSC — Organizzazione Sociopsichiatrica Cantonale) in making more informed decisions about whether hospitalization is needed for a caller. The project is based on approximately 30,000 fully anonymized questionnaires completed by the OSC contact center which takes calls from individuals in need of psychological/psychiatric attention. Our goal is to select the most informative questions from the questionnaires that lead to the decision of hospitalization, with most of the answers being binary and the rest in text form. The research leverages text embeddings, a widely used tool in AI, by encoding text data into numerical vectors for understanding the semantic and emotional content of the text responses in the questionnaires. The methodology employs Information Imbalance, a recently developed non-parametric method for variable selection, to rank the relevance of the questions, and extends to the application of topic modelling for organizing and categorizing the calls, as well as the Gini index to assess the goodness of classification. By incorporating text embeddings into Information Imbalance, our research aims to identify the most relevant questions in the final classification task for hospitalisation, and thereby simplify the questionnaire and reduce the length of emergency calls by eliminating redundant questions that call center agents typically ask under a standardized protocol.