Machine learning in myocardial infarction to improve diagnosis, risk prediction, and treatment decisions

Artificial intelligence has the potential to transform the way that we practice medicine. Our aim is to harness routine data from electronic health records, using signal processing and statistical machine learning, to develop clinical-decision support tools that will aid in the diagnosis and targeting of treatments for patients with myocardial infarction.

display of binary code with an image of a heart over it

Edinburgh is the coordinating site for Health Data Research UK Scotland, a recent £5 million award from the MRC that aims to establish Scotland as a world-leading site for health care data science and research. Scotland has an internationally recognised wealth of linked health-related data assets, covering 5.4 million people. We have access to genomic, biomarker and clinical imaging data through disease specific registries for coronary heart disease, diabetes mellitus and renal disease linked through a national network of safe havens and accessible through the analytics platform at the National Farr Institute.

Personalised medicine

In patients with myocardial infarction, treatment with anti-platelet or anti-coagulant therapies may prevent future cardiac events or precipitate bleeding, with the balance between benefit and harm varying widely between individuals and populations. Current risk prediction models are very inaccurate and based on selected clinical trial populations that have substantial differences in their risk-benefit profile from real-world populations. In this project we are harnessing routine data from electronic health records and machine learning methods to develop a clinical-decision support tool that can be used to improve the targeting of treatment in myocardial infarction. We will train and test our model in our HighSTEACS trial population dataset: a cohort of 53,000 patients from 10 hospitals in Scotland in whom the index diagnosis, major adverse cardiac events (MACE) and bleeding events were adjudicated by panel. Computationally we will: i) address competing risks and the interaction between comorbidities and outcomes, using longitudinal data to track dynamically risk over time; ii) harness neural networks to bypass incomplete data sets; and iii) develop Bayesian approaches to address confounding in the assessment of treatment effects on these outcomes. In combination, these approaches will help us to develop a probabilistic clinical-decision support tool, that estimates the risk of MACE and bleeding for an individual patient.

Diagnostic support tools

The aim of this project is to develop and test a decision support tool that will provide individual probabilities of the diagnosis myocardial infarction for use in patients with acute chest pain in the Emergency Department.

Principal Investigator, Co-Investigators

Our research is supported by a BHF-Turing Cardiovascular Data Science Award and is jointly lead by Professor Nicholas Mills (cardiology), and Dr Catalina Vallejos (data science) with additional co-investigators at the BHF Centre for Cardiovascular Science (Professor David Newby, Dr Catherine Stables) and the Alan Turing Institute (Dr Ioanna Manolopoulou, Dr Chris Russell). We have also received support from the MRC-funded Doctoral Training Programme (DTP) in Precision Medicine and Health Data Research UK award. This funding supports Dimitris Doudesis who is reading for a PhD under the supervision of Dr Athanasios Tsanas (Centre for Medical Informatics, Usher Institute) and Dr David A McAllister (Institute of Health & Wellbeing, University of Glasgow).