Prediction of incident mental health diagnoses at subsequent hospital admissions using large language models

New mental health diagnoses recorded during hospital care can affect assessment, treatment and discharge planning, but we do not know how reliably routinely collected records can identify people at increased risk before a later admission. We will use anonymised mental health records linked to national hospital records to evaluate statistical and artificial intelligence metwhods, particularly large language models, for predicting a first recorded mental health diagnosis at a subsequent admission and to determine whether performance varies between patient groups and outcome definitions.