It is becoming increasingly clear that many cardiovascular diseases (e.g. heart disease) are not as homogenous as previously thought. This can be seen in the multifactorial causes of many of these diseases (where no single risk factor is necessary nor sufficient), and also in the increasing use of risk prediction models to rank patients based on their future risk of unfavourable outcomes. At present we cannot classify these risk factors into discrete categorical subtypes and typologies. Secondly, various risk prediction models are used at different times along the course of disease progression, but these have not been united into a unified model of how individuals progress from full health to death along various clinical encounters.
We plan to analyse data from a dataset called CALIBER, which links records for millions of patients from General Practitioners, hospitals and national statistics, using various visualization and statistical techniques. We aim to generate new hypotheses about possible underlying subtypes of cardiovascular disease progression, which we will then test with genetic data, if possible. Our findings may lead to a more personalized approach to the prevention and management of cardiovascular diseases, efficient use of healthcare resources, and the application of our research methods to the study of other chronic diseases in the future.
Objectives: 1) To visualize the pathways through which individuals transition from full health to death, along intermediating cardiovascular diagnoses. 2) To use the visualization to identify potential prognostic subtypes. 3) To describe the predictors of these subtypes, especially the role of biomarkers and comorbidities/multimorbidity. 4) To evaluate the causality of any biomarkers identified.
Participants: Baseline participants (ca 2 million patients from CPRD, who at entry between 1997-2010 are aged 30+) are followed up for morbidity and mortality with CPRD, HES, ONS and MINAP databases, to identify 22 cardiovascular outcomes (including coronary, stroke, and arrythmic events) and one non-cardiovascular outcome (major haemorrhage).
Methods: 1-2) Alluvial diagrams will be used to describe the progression of participants from health to death. 3) Classical Cox survival models will be adapted to identify which predictors describe each subtype. 4) Mendelian Randomization will be applied to data (probably the UK Biobank) for causal inference.
We are not aware of previous work that used electronic healthcare record data, to identify subtypes of disease progression. Thus if our results are validated by future studies, it could hold methodological value in demonstrating the utility of these methods for analyzing large-scale EHR datasets to identify phenotypic subtypes of disease.
Health Outcomes to be Measured:
Fatal outcomes: coronary heart disease death; cardiac arrest; other cardiovascular disease death; non-cardiovascular disease death
Non-fatal outcomes: Cardiovascular disease states: stable angina; unstable angina; STEMI; NSTEMI; other coronary heart disease; transient ischaemic attack; ischaemic stroke; subarachnoid haemorrhage; intracerebral haemorrhage; other stroke; mitral valve disease; aortic valve disease; other valve disease; atrial fibrillation ; other conduction disorder; peripheral vascular disease; abdominal aortic aneurysm; heart failure ; miscellaneous cardiovascular diseases; Net use of healthcare services
HES Admitted;MINAP;ONS;Patient IMD;Practice IMD (Standard)