Project Description

Research Question: How can smartphones, social media and speech data (digital phenotypes – DPs) be used for identification and early intervention of emerging mood problems in young people in higher education (YPHE)?
Background: DPs can improve clinical assessment and predict symptom exacerbation in patients with depression and anxiety. YPHE have higher risk of anxiety and depressive symptoms than their peers. Despite promising results for DPs, there is resistance due to data privacy issues. Previous work has (a) focused on older adults despite young people at greatest risk of depression, (b) been unable to solve data privacy issues, and (c) used invasive and expensive methods. Our DPs are universally available, free, non-invasive and have no burden on participants.

Aims and Objectives: To co-develop and validate a privacy preserving AI tool that uses DP to predict anxiety and depressive symptom exacerbation in YPHE and test its utility for early identification.
Methods: Using a mixed-methods approach, we will use qualitatively explore students’ understanding of mental health, AI and DPs (work package 1), work with a PPI group address these issues (Work Package 2), develop AI pipelines based on feedback and carry out a longitudinal study with students to test the validity of our AI (WP3).
Analysis: Alongside our qualitative work, we will co-develop privacy preserving, decentralised AI models using a Federated Learning approach, and apply novel Deep Learning models (e.g. Long Short-Term Memory (LSTM)) to model DPs over time, incorporating patient views as part of our AI.