In primary care, patients’ electronic health records can be searched to find information that indicates whether an individual is at increased risk or at the early stages of a specific condition, such as diabetes or stroke. This information can then be combined into a “risk prediction tool” often developed using a mathematical equation or “algorithm” incorporating scores for different risk factors such as age, sex, and other health conditions.
Dementia is now classed as a major public health issue. Worldwide, the numbers of people with dementia will triple by 2050. Despite dementia being a leading cause of disability and death, there is still no suitable tool to alert clinicians to those people at increased risk of dementia who might benefit from early intervention to delay or prevent it.
Also, we know that those from underserved communities, such as certain minorities and those on low incomes, are less likely to be identified as at risk of dementia at an early stage when preventive measures can be offered to reduce their risk. However, we currently do not know how to accurately calculate dementia risk in such communities and what approaches maybe welcomed and acceptable and, if so, how best to roll it out in primary care.
Further, we know that certain diseases, such as a history a cardiovascular disease and stroke, are associated with increased dementia risk. Yet, there is currently no externally validated risk model for predicting future dementia in people with disease related co-morbidity.
Background: Prediction models are widely used to risk stratify individuals for targeted intervention. Several dementia prediction models exist. However, these are characterised by low predictive accuracy and poor external validity. Further, to date, dementia risk model development has (1) almost exclusively focused on Caucasian participants; and (2) not utilized "big data", and thus has limited generalisability.
Aim: To develop and validate new approaches to assessing dementia risk, including their relevance to underserved communities and people with disease-related co-morbidity.
Study Design: Retrospective open cohort.
Setting: Primary care settings within the UK.
Participants: Patients aged 18 years and over between 1st January 1997 and 31st December 2021 with primary care records (CPRD GOLD and Aurum) linked to hospitalisation records (HES Admitted Patient Care).
Primary outcome: Incident diagnosis of dementia in either CPRD or HES data.
Methods: Using CPRD Aurum, competing-risk regression models with all-cause mortality as a competing risk will be used to develop dementia risk models. CPRD GOLD will be used to externally validate the models that are developed. The performance of the models will be assessed in terms of discrimination (predictive accuracy) using Harrell’s C-statistic, positive predictive value (PPV: percentage of patients with dementia correctly identified) and the model calibration using calibration slope by plotting agreement between predicted and observed events.
Outputs: Dementia risk prediction models using linked GP and hospitalisation data will ensure the developed models work well for (i) underserved communities; and (ii) people with disease related comorbidity including cardiovascular disease, depression, and stroke. This could inform the management of patients at greater risk of dementia within the UK general population.
Incident dementia diagnosis
Develop and validate a dementia risk prediction algorithm
Blossom Stephan - Chief Investigator - University of Nottingham
Ralph Kwame Akyea - Corresponding Applicant - University of Nottingham
Carol Brayne - Collaborator - University of Cambridge
Cathy Morgan - Collaborator - University of Manchester
David Reeves - Collaborator - University of Manchester
Louise Robinson - Collaborator - Newcastle University
Manpreet Bains - Collaborator - University of Nottingham
Nadeem Qureshi - Collaborator - University of Nottingham
HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation