Researchers at University College London have used machine learning to identify and validate five subtypes of heart failure. The researchers examined anonymised data from over 300,000 people aged 30 or older diagnosed with heart failure in the UK over 20 years. Once the most robust subtypes had been selected, they were validated using a separate dataset. The subtypes were determined using 87 (of a total of 635) variables covering age, symptoms, medical conditions, drugs taken, and the results of tests and evaluations. The subtypes identified were early onset, late onset, atrial fibrillation related, metabolic, and cardiometabolic. The death rates of these subtypes at one year after diagnosis were 20%, 46%, 61%, 11%, and 37%, respectively. The researchers developed an app for routine clinical use that could significantly improve future risk predictions, inform patient discussions, and help clinicians to tailor treatments.
Heart failure is a complex syndrome with multiple underlying causes, and heart failure subtypes are not completely understood. Each type of heart failure appears to have its own distinct characteristics, with some patients remaining stable for many years, while others deteriorate quickly. However, predicting the likely course of heart failure for individual patients is challenging for clinicians.
Improving the classification of heart failure subtypes using machine learning could help clinicians to better understand the most likely outcome for individual patients and communicate this to them. This improved understanding could also lead to more targeted treatments and innovative therapies.
In this new study, published in The Lancet Digital Health, the researchers sought to classify heart failure subtypes and improve predictions of risk for individual patients. They used machine learning and AI tools to analyse over 300,000 anonymised patient records from the UK. The resulting five subtypes identified by the researchers helped to better predict patient outcomes, and the app developed by the team will enable the clinical use of this knowledge.
The subtypes may also help to more accurately evaluate the effectiveness and cost-effectiveness of different treatments. In addition, the more detailed distinctions between types of heart failure may lead to more targeted treatment plans and help to think in a different way about potential therapies that could benefit patients.
The development of machine learning tools to define subtypes of heart failure is a promising step towards more personalised and targeted treatments for this complex and heterogeneous condition. With further research and clinical trials, it may be possible to use these tools in routine care, improving the quality of information provided to patients and their treatment outcomes.
