Heart and circulatory diseases (cardiovascular disease) are the main cause of death in the UK. In 2018, nearly 7 million people lived with cardiovascular disease and it was responsible for one-in-four of all UK deaths.
People with existing heart disease are up to 5 times more likely to have a stroke, up to half of them develop a second heart attack and are 6 times more likely to die compared to those without heart disease. Currently available tools predict the risk for developing future cardiovascular events or death in people without cardiovascular disease but there are no tools in the UK for people who already have existing cardiovascular disease, with no means to predict their future outcomes.
Using anonymised medical data routinely collected during general practice visits between 2006-2019 in England, we will develop a tool that predicts the risks for developing future cardiovascular events or death, in people who had a heart attack in the past between 2006-2014. We will examine the impact of the timing of the past heart attack on predicting future events within 1 and 5 years, considering the effect of other important factors and their changes over time. In order to validate our results, we will replicate our analyses using medical data from another similar UK primary care database. Our proposed tool will aid general practitioners to find patients at higher risk for future heart disease events and death, which could help them provide management plans that are tailored to individual patient needs.
Background
Cardiovascular (CV) disease (CVD) accounts for 25% of all UK deaths. Several CV risk prediction tools exist for incident (new onset) CVD (e.g. QRISK, Framingham). However, no such tools are available for people with prevalent (existing) CVD despite the clinical need for these tools for risk stratification.
Methods
Using electronic records from the UK Clinical Practice Research Datalink (CPRD) Aurum, we will develop a CV risk tool in a prevalent cohort of people with acute myocardial infarction (AMI) between 2006-2014. Multivariable regression models will be used to estimate 1- and 5-year risk for all-cause mortality (primary outcome), and CV mortality, heart failure, stroke, and recurrent myocardial infarction (MI) (individually and as composite secondary outcomes). We will adjust for covariates' changes (age, CV risk factors, comorbidities) using dynamic models. We will examine the association of the length of the look-back window of the previous AMI event with both outcomes.
Furthermore, we will use supervised Machine Learning (ML) techniques for competing risks analysis (death as a competing event) and single risk analysis (combination of deaths and CVD outcomes). The ML approaches will be Deep Learning models and Random Forests. For the Deep Learning models the data will be randomly divided in a training dataset (80%) and a validation dataset (20%), while for the Random Forests will be used the internal validation based on bootstrap sampling. The ML techniques will be assessed by using the AUROC in the validation dataset. The problems of model overfitting and missing data will be tackled as well.
The tool will be validated using CPRD GOLD database. Patients registered in >1,000 CPRD practices will be used.
This proposed CV risk stratification tool will help primary care providers find patients with CVD who are at higher risk from future adverse events that may benefit from more targeted interventions.
The primary outcome will be all-cause mortality. The secondary outcomes will be CV mortality, incident heart failure, stroke and recurrent MI, as well as a composite of the secondary outcomes. We will differentiate between ischaemic and haemorrhagic stroke events. All-cause mortality (primary outcome) will be identified from primary care and ONS data; CV mortality (secondary outcome) from ONS data; and CVD events (heart failure, stroke and recurrent MI) from primary care and HES APC data.
Salwa Zghebi - Chief Investigator - University of Manchester
Salwa Zghebi - Corresponding Applicant - University of Manchester
Azfar Zaman - Collaborator - Newcastle upon Tyne Hospitals NHS Foundation Trust
Corneliu Arsene - Collaborator - University of Manchester
Darren Ashcroft - Collaborator - University of Manchester
Evangelos Kontopantelis - Collaborator - University of Manchester
Mamas Mamas - Collaborator - Keele University
Rosa Parisi - Collaborator - University of Manchester
Steven Scholfield - Collaborator - University of Manchester
2011 Rural-Urban Classification at LSOA level;HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation