Predictive factors for psychotropic medication prescription in child and adolescent psychiatric emergency care: a machine learning approach using a health data warehouse

Emergency department (ED) visits for child and adolescent psychiatric presentations have increased in recent years. Psychotropic treatments are frequently used during visit, and sometimes at discharge. However, there are no clear guidelines, and current practices are poorly described and understood in this domain. We propose exploring these practices and the factors leading to psychotropic medication in emergency settings in-visit and at discharge for children and adolescents. We conducted a retrospective cohort study of 1,527 patients seen at Necker Hospital (Paris, France) between 2021 and 2025, using clinical data warehouse (CDW) data. Sociodemographic and clinical variables, psychotropic administration during the ED stay, and prescriptions at discharge were analyzed using multilevel logistic regression and gradient boosting models, integrating structured data and SNOMED concepts extracted from free-text notes. Among 2,060 psychiatric ED visits, psychotropic medications were administered in 21.9% and prescribed at discharge in 13.9%. In-ED administration was mainly driven by markers of acute severity, including prolonged length of stay, older age, psychiatric diagnosis, and frequent ED use, while anxiety disorders were associated with lower odds. In contrast, discharge prescriptions were independently associated with in-ED psychotropic administration, psychiatric diagnosis, and socioeconomic context. Models using free text in addition to structured data significantly improved prediction, allowing for a comprehensive and global understanding of factors involved in prescribing. Psychotropic use in pediatric psychiatric emergencies reflects distinct processes of acute symptom management and discharge decision-making, influenced by different clinical and social factors. Leveraging unstructured clinical data enhances understanding of these practices and may inform future guidelines and continuity of care.