Developing a deep learning-based model to predict rehabilitation outcomes in underserved groups living with multiple long-term conditions (360G-Wellcome-341874_Z_25_Z)

This project aims to develop a deep learning–based model to predict successful completion of exercise-based rehabilitation amongst people living with multiple long-term conditions (MLTC). MLTC are increasingly prevalent among ethnic minority and socioeconomically disadvantaged populations, experiencing reduced quality of life, lower physical function, and higher healthcare utilisation. Although rehabilitation improves outcomes across many long-term conditions, adherence is low and early drop out is common. The research will begin by establishing a consensus-based definition of rehabilitation "completion". This definition will be used alongside real-world datasets from exercise-based rehabilitation programmes, to identify subgroups of participants based on demographic, socioeconomic, clinical and functional factors that associate with completion. This large dataset will be used to develop and train a deep learning model capable of identifying individuals at risk of non-completion. Following this, the views and experiences of people with MLTC who were deemed "non-completers" of exercise rehabilitation will be explored using qualitative interviews. This will identify personal, social and service level barriers and facilitators to programme completion, aiming to guide the development of personalised support strategies to improve adherence and promote equity in rehabilitation services. Keywords: multiple long-term conditions, rehabilitation, deep learning, health inequalities, adherence, prediction.