Heart failure (HF), atrial fibrillation (AF) and acute coronary syndromes (ACS) are three of the most common heart diseases affecting the worlds population. They frequently coexist, but are studied separately in terms of risk factors and outcomes. There is variation in definitions of these diseases across guidelines and study designs. Machine learning (ML) is a field of computer science that uses artificial intelligence to learn relationships or patterns from the data with or without the need to define them a labelling information. ML has been used to identify novel disease definitions, but these analyses have been restricted to single diseases and smaller samples which may not be representative of the whole population.
This study will investigate if it is possible to discover new subtypes in patients with overlapping HF, AF and ACS. ML approaches will be used to assign patients into groups according to their clinical features (e.g. diagnoses, lab results).
Better characterisation of the overlap syndrome may improve prediction of the risk of particular outcomes, as well as providing novel insights into mechanism of action of current and new therapies. Definitions identified in this project can be validated in other electric health record (EHR) studies for routine use in research and practice.
The project intends to work on complex data to better understand how to define phenotypes and discover sub-phenotypes of three most common cardiovascular diseases: heart failure (HF), atrial fibrillation (AF) and acute coronary syndromes (ACS). Unsupervised machine learning approaches such as clustering have been adopted with the intention of recognising disease subtypes. However, those analyses are mostly confined to individual diseases and specific patient sub-groups, which may lead to inaccurate clustering and consequently inaccurate diagnostics. Furthermore, these three heart diseases are frequently risk factors for each other, and sometimes co-exist, which are currently understudied as an overlap syndrome. Therefore, we aim:
1) To use combined unsupervised and supervised statistical learning to improve discoveries of sub-phenotypes of HF and AF.
2) To define a broader, joint phenotype of HF or AF for subphenotype.
3) To cluster longitudinal phenotypes of HF and AF in order to resolve the inter-related phenotypes.
4) To use machine learning (supervised, unsupervised or combined) to identify clusters in the overlap syndromes between HF, AF and ACS, in other words, to characterize what phenotypes there are in clusters where three diseases overlap.
Heart Failure
- All-cause mortality
- Cardiovascular disease
- Coronary artery disease
- Cardiovascular mortality
- GP appointments
- Atrial fibrillation
- Stroke
- Hospital admissions
- Thromboembolism
- Emergency department attendances
- Percutaneous coronary interventions
- Coronary artery bypass graft surgery
Amitava Banerjee - Chief Investigator - University College London ( UCL )
Amitava Banerjee - Corresponding Applicant - University College London ( UCL )
Daniel Swerdlow - Collaborator - University College London ( UCL )
Folkert Asselbergs - Collaborator - University College London ( UCL )
Ghazaleh Fatemifar - Collaborator - University College London ( UCL )
Harry Hemingway - Collaborator - University College London ( UCL )
Laura Pasea - Collaborator - University College London ( UCL )
Mohamed Mohamed - Collaborator - University College London ( UCL )
Muhammad (Ashkan) Dashtban - Collaborator - University College London ( UCL )
Qianrui Li - Collaborator - Sichuan University
Riyaz Patel - Collaborator - Barts Health and UCLH NHS Trusts
Sheng-Chia Chung - Collaborator - University College London ( UCL )
Spiros Denaxas - Collaborator - University College London ( UCL )
Suliang Chen - Collaborator - University College London ( UCL )
Tom Lumbers - Collaborator - University College London ( UCL )
HES Accident and Emergency;HES Admitted Patient Care;HES Diagnostic Imaging Dataset;HES Outpatient;ONS Death Registration Data;Practice Level Index of Multiple Deprivation