Identifying older patients at high short-term risk of dementia and cognitive decline using routinely collected hospital electronic clinical and brain imaging data to improve care

Research aim To identify older hospital patients at high risk of developing dementia using hospital electronic patient records and brain scans to improve care Background Older people (>65 years) occupy 70% of UK hospital bed days with numbers still rising. Older in-patients are vulnerable, often multi-morbid or frail, and the stress of illness may reveal or precipitate cognitive fragility. Extensive electronic patient record (EPR) data are collected routinely during admission and many patients also have brain imaging usually with CT. Digital algorithms using these data could identify patients at high short-term risk of dementia thereby enabling targeted interventions, selection for trials and reducing costs. Aims and objectives We aim to produce risk scores for dementia and dementia subtype and accelerated cognitive decline in hospitalised older people (Digital Biomarkers for Dementia (DBD)s). The DBDs will be implemented as EPR algorithms or Apps to inform in-and out-patient care and results will be automatically transmitted to GPs. Our DBDs will be validated, feasible, user-friendly, reliable and acceptable to patients/carers, relevant to stakeholders, scalable for implementation across the NHS, and ready for commercialisation. Methods In pilot studies, we identified the predictors of dementia/dementia subtype after hospitalisation by linking our vascular disease cohort (Oxford Vascular Study-OXVASC, phase-1, 2002-2012, n=1369) to follow-up electronic hospital data. We included baseline brain imaging i) as a predictor and ii) to stratify patients for small vessel disease (SVD). Certain risk factors (eg infection) were stronger predictors in those with SVD and use of different models in those with/without SVD enhanced prediction. Using this work, we propose two DBDs: OxDBD1.0 will use hospital EPR data alone and OxDBD2.0 will use EPR data with measures from routinely acquired brain imaging. We will validate and refine our OxDBDs on two existing cohorts (OXVASC, phase-2 and UKBiobank) before validation in the acute hospital setting using existing and new data. In parallel, we will continue to develop our automated CT-brain analysis tool to extract quantified measures (eg of SVD) avoiding the need for time-consuming visual ratings, and enabling implementation of OxDBD2.0 at scale. Complimentary workpackages will deliver qualitative stakeholder research, and health economics with PPI woven throughout design, delivery and dissemination. Timelines for delivery Refinement/optimisation of our preliminary OxDBDs: end year 2; CT-brain analysis toolset available: end year 3. Validation of the OxDBDs in the acute hospital setting: end year 3 (using retrospective data) and by project end (prospective data); commercialisation and adoption plans for the OxDBDs and separately, the CT-brain analysis tool: by project end. Anticipated impact and dissemination Our OxDBDs will use only routinely acquired data with no additional input from patients or staff. The OxDBDs will enhance information given to patients, and carers and help avoid crisis re-admissions/institutionalisation through better pre-planning and carer support. Targeted interventions, prevention measures and timely diagnosis could reduce the population burden of dementia by 12-14% reducing both acute and long-term care costs. Commercialisation of the OxDBDs via eg pharma and healthcare providers, and separately, of the CT-analysis tool for embedding at point-of-care, has potential for global impact.