Machine learning system identifies new biomarkers to predict strokes

Its implementation is easy because it only requires electrocardiograms which are usually available in most hospitals

23 DECEMBER 2021

Atrial fibrillation is the most common cardiac arrhythmia and can be associated with various life-threatening complications, such as stroke, heart failure, and dementia. Approximately 33 million people worldwide suffer from heart rhythm problems, with the likelihood of being affected increasing with age. It is three to four times more prevalent in patients over 80 than in those aged between 60 to 70.

In Spain, it is present in 17% of the population over 80 years of age, making it a major concern for the health system. But the number of patients may be higher because it is known that one-third of people suffering the condition do not report it to their doctor.


Current methods to detect arrhythmia

The electrocardiogram (ECG) is the main technique for diagnosing heart disease and has been used for more than a century. The Dutch doctor Willem Einthoven received the Nobel Prize in Medicine in 1924 for the invention of the ECG, which over time has proven to be an inexpensive and accessible technique.

In recent years, new quantitative methods for detecting ECG biomarkers have been developed to identify hundreds of measurements using highly sophisticated software. Big data techniques have also been developed to analyse clinical data in cardiology.


Artificial intelligence that detects fibrillation

Researchers at the Hospital Universitario de La Princesa and its Research Institute have developed a new system of ECG markers to predict the risk of atrial fibrillation. By using machine learning on apparently normal electrocardiograms, hitherto unknown signs have been identified.

The research analyzed 566 parameters of 329,670 ECGs performed on 132,772 patients at the Hospital Universitario de la Princesa and its specialty centre over almost a decade (from 2010 to 2019). New ECG biomarkers for the detection of fibrillary arrhythmia were obtained from the study. Subsequently, the Atrial Fibrillation Automatic Assessment (AFAA) risk score was developed.

Photo: Guillermo J. Ortega on the left and Jesús Jiménez Borreguero on the right

The research team – led by cardiologist Jesús Jiménez Borreguero and physicist Guillermo J. Ortega – published the results in the journal Heart as an original article. The study – for which they have already applied for a patent – has project funding from the Carlos III Institute’s Health Research Funds. Doctors Ancor Sanz and Alberto Cecconi are also co-inventors and participated in the patent application. The cardiologists Alberto Vera, Fernando Alfonso, and Juan Miguel Camarasaltas, the latter from the IT service, also collaborated.