Preventing psychosis relapse through crosslinguistic prediction models

Rationale for the workshop

Individuals with psychosis can experience atypical perceptions (e.g., auditory hallucinations), hold anomalous ideas (e.g., delusions), and suffer the deterioration of thought processes. Psychosis typically shows a course moving from periods characterized by minor symptoms severity and tolerable cognitive and social functioning (i.e., remissions) to episodes marked by an increased in symptoms severity (i.e., relapses). Critically, individuals with a relapsing course of psychosis have reduced chances of sustaining relationships, higher risks of unemployment and more severe functional and cognitive decline.

Accurate and timely interventions to prevent psychosis relapse can result in better psychological wellbeing and social functioning in these individuals, even reducing costs related to medication and (re)hospitalization to treat psychosis relapse. Yet, preventing psychosis relapse is a highly challenging task, given the heterogeneity in the clinical presentation of psychosis relapse and the inherent difficulty of monitoring citizens outside the clinic. This is why developing a remote monitoring system that incorporates valid and trustworthy predictors of psychosis relapse based on a combination of speech measures, statistical models, and artificial intelligence (AI) might improve the prevention of psychosis relapse. Crucially, a system of this nature should incorporate the medical needs and preferences of care providers and service users, while adhering to the emerging legal and ethical European framework to regulate the clinical use and commercialization of AI-based medical devices.

Aims of the workshop

The first aim will consist in determining how assessing the validity and reliability of speech-derived features to prospectively predict psychosis relapse can take place. The second aim will be to discuss strategies that can test the generalizability of speech-derived features across multiple languages when prospectively predicting psychosis relapse. The third aim of the workshop will consist in mapping the requirements of designing, developing, and implementing a remote monitoring system based on speech-derived features to prospectively predict psychosis relapse. The fourth aim will consist in foreseeing challenges and solutions related to carrying out an international randomized control trial to test the clinical value of speech-derived features in prospectively predicting psychosis relapse, while integrating the feedback of end-users to improve the design and functioning of the monitoring system.