Editor choice


AI identifies 5 types of Heart Failure to Predict risk and Treatment

Heart failure affects millions of people around the world. The disease can be caused by many factors that require different methods of treatment. Now the researchers have trained several machine learning models using a large dataset to identify 5 subtypes of heart failure.

Traditionally, different types of heart failure are classified according to the left ventricular ejection fraction (LVEF) - the amount of blood that the left ventricle of the heart pushes out with each contraction. However, a 2018 Swedish study using machine learning showed that LVEF does not predict survival in heart failure.

Researchers from University College London used four machine learning models to develop a system for determining subtypes of heart failure. They studied the anonymous data of electronic medical records of more than 300,000 patients who had been diagnosed with heart failure for 20 years.

"We sought to improve the classification of heart failure in order to better understand the probable course of the disease and convey this to patients," says Amitava Banerjee, lead author of the study. - Currently, it is difficult to predict how the disease will develop in individual patients. In some people, the condition has been stable for many years, while in others it is rapidly deteriorating."

To avoid the bias that can occur when using a single machine learning model, the researchers used four models to divide heart failure cases into groups. After training on data segments, the models identified five subtypes based on 87 out of 635 possible factors, including age, symptoms, the presence of other diseases, medications the patient took, health parameters such as blood pressure, and test results. Subtypes were tested on a separate dataset.

Five subtypes were grouped according to specific characteristics. Young people with a low level of risk factors were considered to be "early starters". The "late start" included older people who took few medications and suffered from cardiovascular diseases. The subtype "associated with atrial fibrillation" included people with irregular heart rhythm or with heart valve disease. The subtype "metabolic" included overweight people with an average level of risk factors, but with a low level of cardiovascular diseases. The subtype "cardiometabolic" included overweight people taking a large number of prescribed medications, with a high level of risk factors and cardiovascular diseases.

The researchers found that the risk of death within a year after diagnosis differed between subtypes. A year later, the risk of mortality from all causes was highest in the subgroup associated with atrial fibrillation (61%), followed by subgroups with late onset (46%), cardiometabolic (37%), early onset (20%) and metabolic (11%). According to the researchers, the results of the study can be used to improve the treatment of heart failure.

"A clearer distinction between types of heart failure could lead to more targeted treatment and help us take a different look at potential treatments," Banerjee said.

Researchers have developed an application based on a machine learning approach that doctors can use to determine which subtype a person belongs to.

The study was published in the journal The Lancet Digital Health.

Share with friends:

Write and read comments can only authorized users