SYLFAEN-PSY: SYmptom Learning through Foundation models to Augment and ENhance measurement in PSYchosis

Dimensional symptom scales are sensitive measurements of the course of psychosis. While they are recommended for routinely monitoring psychotic symptoms, their use is inconsistent at best: Fewer than 30% of psychiatrists report even occasionally using them, while research studies often differ substantially in which and how many scales they assess, even when measuring the same symptoms. Thus, most people with psychotic disorders, either in the clinic or as research participants, only get their experiences partially translated (if at all) into quantitative dimensions. One main argument against dimensional scales is that they are lengthy to complete, and our project offers a pragmatic alternative. We will create a system, informed by lived experience and based on Large Language Models, to generate proxy measurements of psychotic symptoms by integrating multimodal data. These proxies will be validated against gold-standard scales in deeply characterised cohorts of individuals with psychotic disorders, and in research data from mental health NHS services. Through extensive assessments of uncertainty and bias, we will develop a new standard for AI-based information extraction and inference in mental health research, delivering a solution for assessing the experiences of people with psychosis at scale, and removing a long-standing barrier for symptom-based research.