Around 670,000 people in the UK are currently living with dementia, and this number is expected to double over the next twenty years. At present there are no available treatments that alter the course of the disease. We now understand that abnormal proteins build up in dementia patients years before someone develops symptoms, so it is possible that treatments do not work because the condition is too severe by the time we give them to patients. We therefore need to find a way of identifying people who are currently healthy but are at risk of getting dementia in the future. We ideally need to do this using only the sort of information that is available to GPs, to avoid doing invasive and expensive tests on lots of healthy people.
We have developed two risk prediction models one to predict dementia from any cause and one to predict Alzheimers disease dementia, using data from Wales, UK. We aim to use CPRD data to test these models in a different population (external validation). We will also test an existing dementia risk prediction model developed by Walters et al (2016), and will compare the performance of these two models within CPRD data.
In this study we aim to externally validate three dementia risk prediction models in CPRD data.
We have developed two 10-year risk prediction models (one for all-cause dementia and one for Alzheimers disease dementia) using linked primary care, hospital admissions and mortality data from the Secure Anonymised Information Linkage (SAIL) Databank in Wales. The Dementia Risk Score (DRS), developed by Walters et al. (2016) using data from The Health Improvement Network (THIN), aims to predict 5-year all-cause dementia risk.
To externally validate these models, we will use linked primary care, hospital admissions, deprivation and mortality data to create a cohort study, in which follow-up for eligible participants begins on 1/1/2008 (study start date) and ends at the earliest of: dementia diagnosis, death, loss to follow-up or end of follow-up. Predictors will be derived from primary care and hospital admissions data, and dementia outcomes will be ascertained using primary care, hospital admissions and mortality data. Participants will be eligible if aged 60-79 years at study start and are registered in a practice in England or Scotland with up-to-standard follow-up. We will exclude participants with an existing all-cause dementia code in any dataset prior to the study start date. To validate the models we will use the same modelling techniques employed during model derivation Royston-Parmar flexible parametric models, accounting for the competing risk of death in the models derived from SAIL data, and Cox regression in the DRS model.
We will produce calibration plots and calculate performance statistics for each of the models. We will estimate overall model fit using Cox-Snell R2 and Royston R2D techniques. We will use Harrells C-statistic and the D-statistic to estimate model discrimination. For calibration, we will calculate the calibration slope, produce calibration plots, and calculate expected/observed outcome probabilities at 5 and 10 years.
Diagnosis of all-cause dementia; Diagnosis of Alzheimers disease; death (as a competing risk)
Tim Wilkinson - Chief Investigator - University of Edinburgh
Tim Wilkinson - Corresponding Applicant - University of Edinburgh
Cathie Sudlow - Collaborator - University of Edinburgh
Christian Schnier - Collaborator - University of Edinburgh
Kym Snell - Collaborator - Keele University
Richard Riley - Collaborator - Keele University
HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation