Atrial fibrillation (AF) is a major cardiovascular health problem: it is common, chronic and incurs substantial health-care expenditure as a result of stroke, sudden death, heart failure and unplanned hospitalisation. The incident prevalence of AF has increased substantially over the last decades, however, the disease trajectories of patients hospitalised with AF are not fully understood. This project aims to study these disease trajectories using big health data, machine learning and health informatics techniques.
Explaining the science
In this project data analysis has been carried out using data provided by ResearchOne, a longitudinal, anonymised study looking at cardiovascular, non-cardiovascular and frailty outcomes in old people, with a cohort of 536,955 patients, gathering data from January 2003 to December 2015.
Primary data analysis obtained descriptive statistics to better understand the population means of the complete patient cohort. 'Survival analysis' was used to predict atrial fibrillation onset in patients in the complete cohort, which accounted for risk factors associated with atrial fibrillation such as drug prescriptions and patient level variable for example sex, ethnicity, BMI and socioeconomic status (Index of Multiple depravation Ranking).
This study will characterise disease trajectories using unsupervised machine learning techniques to classify the outcome events from re-admission. In addition, frailty models will be used to model all-cause mortality, while multi-state models will be used to investigate multiple cardiovascular and non-cardiovascular outcomes. The temporal trends of incident events and anticoagulation uptake will be quantified through trend analysis.
The aim of this project is to study, using big health data and health informatics techniques, the disease trajectories of patients with atrial fibrillation. The project will identify patients with atrial fibrillation from Hospital Episode Statistics (meaning an estimated analytical cohort of circa 1.2 million). Patients hospitalised with atrial fibrillation as primary diagnosis between 2003 and 2018 will be included in the study as cases. For each case, 5 patients hospitalised with other conditions will be randomly selected to match for age, sex and admission month and year.
This work has shown that patients diagnosed with conditions such as hypertension, chronic obstructive pulmonary disease, chronic kidney disease and ischaemic heart disease have an increased probability of developing incident AF, which further reinforces these diseases as prominent risk factors for developing AF. Furthermore, the learning from this study suggests factors, such as direct oral anticoagulant medication, can decrease the probability of a patient developing AF. Thereby suggesting that changes can be made to the patient’s lifestyle and can lead to better population health management.