Future of Mental Health Theme

This call was the fourth of nine from the EPSRC Network Plus in Human Data Interaction (HDI), and it funded one large project:

Funded projects in the theme The Future of Mental Health

ExTRA-PPOLATE: (Explainable Therapy Related Annotations: Patient & Practitioner Oriented Learning Assisting Trust & Engagement)

Mat Rawsthorne et al, Nottingham University.


This project aims to examine the key components of trust in algorithm-mediated digital mental health (DMH) through a participatory design and dissemination study. We will co-create a collaborative machine learning decision support tool to help mental health practitioners and patients classify key processes in therapy transcripts. This will speed up rating of therapist fidelity and assessment of patient activation, thereby providing evidence for improving practice. The bigger question is: does the engagement of patients and practitioners in the design process make the AI application more credible and trustworthy?

To answer this question, ExTRA-PPOLATE will:

  1. Adapt responsible research and innovation (RRI) methods to manage the coproduction of an interdisciplinary mental health data science initiative.
  2. Prototype a person-centred semi-supervised model training process to refine definitions, expose and explore tacit and latent knowledge in assessment of psychotherapy.
  3. Identify the key factors contributing to trust in the model pipeline (data, processing, deployment) by examining domain expert requirements for the qualities of an engaging, interactive feedback interface and eliciting wider concerns about its acceptability.
  4. Assess whether patient and practitioner involvement in the development of a digital mental health decision support tool increases trust in it.
  5. Foster a community of interest in HDI approaches applied to mental health.

This is a very exciting project that could have significant impact on the process of quality-assuring therapeutic interventions. We are looking forward to seeing progress and research outputs.