AdSOLVE Mental health
Research type
Research Study
Full title
Addressing socio-technical limitations of Large Language Models for use in mental health care planning and monitoring
IRAS ID
351415
Contact name
Domenico Giacco
Contact email
Sponsor organisation
Coventry and Warwickshire Partnership Trust
Duration of Study in the UK
0 years, 9 months, 22 days
Research summary
The growing need for mental health support in our society requires innovative solutions. Interventions assisted by Artificial Intelligence (AI) have proved promising in augmenting capacity of mental health services and options for treatment and mental health monitoring (Kazdin, 2021).
Standardised subjective measures are fundamental to mental health monitoring but have significant limitations: level self-awareness; willingness to complete questionnaires; limited choice of responses (Schwarz et al., 2023). Moreover, they do not consider information outside therapy sessions (e.g. social media data). Summaries that capture fluctuations in individuals' state-of-mind, based on heterogeneous sources, while emphasising key clinical concepts, can significantly assist in monitoring, prevention and early detection, augment clinician capacity, present alternatives to standard questionnaires and compensate for reduced access to mental health services. Related work has shown the potential for LLMs to create clinically meaningful summaries (Song et al., 2024). This monitoring systems could be used to augment reach and effectiveness of self-management interventions. Such interventions have significant beneficial effects, even for people with severe mental illness, reducing symptoms and length of admission in inpatient units, improving social functioning & quality of life beyond end of treatment (Giacco et al., 2017;2018; Lean et al., 2019; Islam et al., 2023). AI-assisted models can support mental health clinicians and service users in developing shared strategies for self-management, allowing for personalisation in monitoring progress.
Key challenges remain such as: how can monitoring and dialogue technology be combined? Can the possibility of fabricated or erroneous information (hallucinations) be minimised or ideally eliminated? How does the technology meet patient and clinician needs? Are these tools acceptable for patients and can they be integrated with services care and within practice guidelines (e.g. those developed by NICE)?
In this project, computer scientists, ethicists, mental health professionals and patients will collaborate, to co-develop and test an AI assisted tool for mental health monitoring and sf-management which may be integrated in standard care planning practices within NHS services.REC name
West of Scotland REC 5
REC reference
25/WS/0181
Date of REC Opinion
13 Nov 2025
REC opinion
Favourable Opinion