Atrial fibrillation and atrial flutter (AF/F) are both health conditions that occur when there are problems with the electrical signals in the heart causing it to beat irregularly. If left untreated, AF/F can damage the heart leading to stroke, heart failure and other serious health conditions, and thus the early detection and prevention of AF/F and its progression is of significant importance.
Knowing what information (such as age, blood pressure etc) might predict which people develop AF/F might help identify previously undiagnosed individuals. It may also help identify individuals who are likely to benefit from measures to prevent progression of the disease.
With this in mind, this study aims to use an anonymised primary care database (the Clinical Practice Research Datalink) with patients who are already diagnosed with AF/F. Their data will be used by a computer program to identify common characteristics that describe a patient with AF/F. This computer program will then be tested on the data to see how accurate it is in identifying those who actually have AF/F. The program will learn from what it got wrong and then improve its accuracy until it has a very good description of what a patient with AF/F looks like.
The aim of this retrospective cohort study is to utilise the Clinical Practice Research Datalink (CPRD) to identify patients diagnosed with AF/F in the UK, and use their data to identify risk factors that predict AF/F. This data will be used to develop risk predication equations.
Univariate and multivariate methods of statistical analysis will be used to identify risk factors that predict AF/F and summary descriptive statistics will be generated characterising patient demographics, clinical and treatment characteristics and medication use, in relation to stratification of risk and incidence of AF/F.
Risk predication equations generated by the CPRD dataset will then be compared against existing published risk equations to evaluate the performance of existing models in describing the association between AF/F incidence and patient-level risk factors.
Machine learning approaches to complement standard regression models will be evaluated; including approaches incorporating, kernel methods, random forests, deep neural networks and Gaussian processes. The resulting set of risk equations will be reflective of current UK patients and practices and may be informative to clinical and policy decision making.
The outcome of these methods will be an algorithm that can use routinely collected data to accurately describe the predictive characteristics of a patient with AF/F.
Incident cases of atrial fibrillation and atrial flutter (AF/F)
Phil McEwan - Chief Investigator - Health Economics & Outcomes Research Ltd ( HEOR Ltd )
Phil McEwan - Corresponding Applicant - Health Economics & Outcomes Research Ltd ( HEOR Ltd )
David Clifton - Collaborator - University of Oxford
Jason Gordon - Collaborator - Health Economics & Outcomes Research Ltd ( HEOR Ltd )
Mark O’Neill - Collaborator - Guy's & St Thomas' NHS Foundation Trust
Matthew Lumley - Collaborator - Pfizer Ltd - UK
Nathan Hill - Collaborator - Bristol-Myers Squibb Pharmaceuticals Limited - UK ( BMS )
Steven Lister - Collaborator - Bristol-Myers Squibb Pharmaceuticals Limited - UK ( BMS )
ONS Death Registration Data