A team of researchers have used datamining and machine learning techniques to find subtle changes in electrical activity in the heart that can be used to predict potentially fatal heart attacks.
Researchers from the University of Michigan, MIT, Harvard Medical School and Brigham Women’s Hospital in Boston sifted through 24-hour electrocardiograms (which measure the electrical activity in the heart) from 4,557 heart attack patients to find errant patterns that until now had been dismissed as noise or were undetectable.They discovered several of these subtle markers of heart damage that could help doctors identify which heart attack patients are at a high risk of dying soon. Electrocardiograms (ECGs) are already used to monitor heart attack patients, but doctors tend to look at the data in snapshots rather than analyze the lengthy recordings.
The team developed ways to scan huge volumes of data to find slight abnormalities — computational biomarkers — that indicate defects in the heart muscle and nervous system. These included looking for subtle variability in the shape of apparently normal-looking heartbeats over time; specific sequences of changes in heart rate; and a comparison of a patient’s long-term ECG signal with those of other patients with similar histories.
They found that looking for these particular biomarkers in addition to using the traditional assessment tools helped to predict 50 percent more deaths. The best thing is that the data is already routinely collected, so implementing the system would not be costly.
Around a million Americans have heart attacks each year, with more than a quarter of those in groups who survive the initial attack dying within a year. Current techniques miss around 70 percent of the patients that are at high risk of complications, according to Zeeshan Syed, assistant professor at the University of Michigan Department of Electrical Engineering and Computer Science.