Heart failure affects millions of people worldwide, but it can be caused by multiple factors and requires different treatments. Now, researchers have trained multiple machine learning models using a large population-based dataset to identify her five subtypes of heart failure. This may inform treatment, patient education, and prediction of future risk factors.
“Heart failure” is an umbrella term that describes the condition in which the heart is unable to perform adequately to meet the body’s blood and oxygen needs. This can be caused by several underlying factors that affect treatment of the condition. Risk factors for heart failure include coronary artery disease and heart attack, diabetes, high blood pressure, overweight and obesity, and heart valve disease.
Different types of heart failure have traditionally been classified based on the left ventricular ejection fraction (LVEF), which is the amount of blood that the heart’s left ventricle ejects each time it contracts. However, a 2018 Swedish machine learning study found that LVEF did not predict heart failure survival.
Now, researchers at University College London are using four machine learning models to better inform treatment and determine subtypes of heart failure that may determine future risk. developed a framework for
Researchers examined anonymized electronic health record data of more than 300,000 UK patients diagnosed with heart failure over a 20-year period. Data were taken from his two large primary care datasets representative of the UK population.
“We sought to improve the classification of heart failure with the aim of better understanding and communicating the expected course of heart failure to patients,” said Amitabha Banerjee, lead author of the study. said Mr. “At this time, it is difficult to predict how the disease will progress for individual patients. Some remain stable for years, while others deteriorate quickly.”
To avoid the potential bias of using one machine learning model, the researchers used four models to classify heart failure cases into groups. After being trained using segments of the data, the model will be able to identify information including age, symptoms, presence or absence of other conditions, medications the patient is taking, health parameters such as blood pressure, and test results such as: We identified five subtypes based on 87 of the 635 factors identified. as renal function. Subtypes were validated using another dataset.
Five subtypes were clustered according to specific characteristics. ‘Early onset’ included young people with a low proportion of risk factors. “Late onset” patients were older, female, took fewer prescriptions, and had cardiovascular disease. “Atrial fibrillation-related” included people with atrial fibrillation, a condition in which the heart beats irregularly, and heart valve disease. The ‘metabolic’ subtype included overweight individuals with moderate risk factor rates but low rates of cardiovascular disease. Also, ‘cardiometabolic’ included overweight people with higher rates of risk factors and cardiovascular disease and taking heavy prescription medications.
The researchers found that the risk of dying within the first year after diagnosis varied by subtype. All-cause mortality at 1 year was highest in the atrial fibrillation-related subgroup (61%), followed by late onset (46%), cardiometabolic (37%), early onset (20%), and metabolic. (11%).
The researchers say the results of this study can be used to improve the treatment of heart failure.
“Better differentiation between types of heart failure could lead to more targeted treatments and could help us think differently about potential treatments,” Banerjee said. rice field.
Researchers have developed an app that doctors can use to determine which subtype a person belongs to, based on a machine learning approach. This can be used to guide patient education and improve prediction of future risks.
“The next step is whether this way of classifying heart failure makes a real difference for patients, improves risk prediction and the quality of information provided by clinicians, and changes patient care. It’s about seeing what’s going on,” Banerjee said. “We also need to know if it is cost-effective. The app we designed needs to be evaluated in clinical trials or further studies, but it could be useful in routine care.”
The study was published in a journal lancet digital health.
Source: University College London
