Codesign of Artificial Intelligence algorithm for personalised exercise videos to enhance community-based rehabilitation for people with stroke

Unmet Health Need: Stroke affects 118,000 people in the UK annually1 impacting quality-of-life/NHS resources. with costs of £17.4 billion yearly2. The most stressful transfer of care is from hospital to home. Many people with stroke (PwS)/carers feel afraid/unsupported/abandoned,3 without long-term NHS support.4 Rehabilitation exercises are crucial for recovery.5-6 Exercise prevents stroke reoccurrence,7 improves cardiovascular fitness,8 mobility,9 and strength.10 National-Clinical-Guidelines (2023)11 state PwS should engage in rehabilitation 3 hours/day. This increase may stretch NHS resources, and our application aims to address this challenge. We focus on providing cost-effective long-term support to PwS12-13 by making support accessible to PwS from diverse communities improving community stroke services. This aligns with recent stroke priorities including how community stroke services are best resourced (priority-5) and interventions to improve fitness/recovery and prevent further stroke (priority-9) .14 Proof-of-Concept: To address the gap, Kumar and Bristol-After-Stroke (BAS) developed an 8-week group exercise (1hr) and education (1hr) community programme, Next-Steps' that supported 250 PwS since 2016.15 Technology, e.g. exercise videos, can improve service provision11,16 is effective when compared to paper-based programmes in improving exercise adherence/functional outcomes.17-18 Next-Steps participants were prescribed exercise videos from the Stroke Association UK19, however, could not relate to these because demonstrations show a young/fit therapist. "Watching someone exercise who has had a stroke like me, who looks like me, who talks like me, will be more meaningful and would encourage me to participate" (Patient partners). Thus, BAS and Kumar s videos show PwS exercising under physiotherapist guidance.20 Next-Steps participants were prescribed 4-5 personalised videos to use at home. In a year, 35 PwS used these and 75% exercised daily for at least an hour. Everyone reported exercises were helpful/acceptable/easy to follow/confidence boosting. With Dhek Bhal (South Asian (SA) community) in Bristol, Kumar created similar exercise videos in SA languages to ensure inclusivity. Participants reported videos were meaningful, however, to ensure continuity of rehabilitation at home, patients need tools for clear understanding of exercise progression/regression. Innovation and Health Problem Addressed: Equipping patients to self-manage is advocated by the National-Clinical-Guidelines.11 We propose to co-develop with patients and physiotherapists an Artificial Intelligence (AI)21-22 driven algorithm that stratifies patients according to their functional abilities and provides goal specific exercises. The AI model employs a "Human-in-the-loop" approach by integrating therapist feedback to refine predictions. Therapists verify the model s assessments, ensuring accuracy and relevance. This iterative process allows the AI model to learn from real-world data, based on patient-reported measures, facilitating continuous improvement through re-training. This principle underpins the co-design and evaluation of the tablet-based app, ultimately enhancing treatment personalisation and effectiveness for patients. Public-and-Patient Involvement Engagement (PPIE): PPIE from BAS, Dhek Bhal (Bristol), Centre for Ethnic Health Research (Leicester) helped develop this study. Three separate online/in-person workshops were conducted with PwS (n=16) and carers (n=6) highlighted: • Exercise videos would benefit PwS. • Videos should be individualised/linked to patient goals allowing progression/regression. • Audio instructions are needed for visually-impaired people. We will further involve PPIE as steering-committee members. By prioritizing inclusivity, we will address varying experiences of individuals in the digital context.23-24