Improving Detection and Intervention for Social Isolation and Loneliness in Dementia Using NLP and Clinical Screening

Details

Dementia is a progressive neurocognitive condition that substantially reduces quality of life and life expectancy, and is frequently experienced as deeply isolating. Meta-analytic evidence indicates that over 40% of people with dementia report loneliness and more than 60% experience social isolation, both of which are associated with poorer cognitive and health outcomes. Despite this, social determinants are not routinely assessed or systematically addressed in memory clinic settings, limiting timely access to social prescribing and community-based support.

This project builds on prior work using Natural Language Processing (NLP) to analyse electronic health records (EHRs) across multiple NHS Trusts. In a dataset of over six million clinical documents from more than 30,000 patients, mentions of social isolation (15%) and loneliness (10%) were substantially under-recorded compared to population estimates. Importantly, these factors were associated with distinct cognitive trajectories: loneliness was linked to consistently lower cognitive scores, while social isolation predicted accelerated decline following diagnosis. These findings highlight a critical gap in clinical assessment and suggest that unrecognised social factors may contribute to increased healthcare costs and poorer outcomes.

The proposed PhD will address these gaps by combining quantitative, computational, and implementation-focused approaches. The project has two interconnected strands.

The first strand focuses on evaluating the feasibility and acceptability of implementing brief, validated screening tools for loneliness and social isolation (e.g., UCLA-3 and LSNS-6) in working-age memory clinics. This will involve analysing completion rates, identifying patients eligible for social prescribing, and examining referral uptake and engagement with services over a six-month follow-up period.

The second strand will extend existing NLP models to detect indicators of social isolation and loneliness directly from routine clinical text. The student will develop and benchmark models against validated screening measures, and explore how NLP-derived outputs can support clinical decision-making and personalised social prescribing. This includes evaluating acceptability among clinicians and patients, and identifying barriers and facilitators to implementation within NHS workflows.

Methodologically, the project will involve statistical modelling of longitudinal outcomes, NLP model development using transformer-based approaches, and mixed-methods evaluation informed by implementation science frameworks (e.g., CFIR).

Overall, this project will provide novel insights into how social determinants can be systematically identified and addressed in dementia care. It will generate evidence to support scalable, data-driven approaches to integrating social prescribing into routine clinical pathways, with potential to improve patient outcomes and reduce healthcare burden.