A new study shows that using a computer science technique to help determine the risk of death among heart attack patients yields more accurate results, and could save lives.
Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), the University of Michigan, Brigham and Women’s Hospital in Boston and Harvard Medical School developed the new technique. It searches for subtle indicators of risk hidden in a patient’s electrocardiograms (EKG), which measure and display electrical activity in the heart. Current methods of assessing the risk of death in patients who have suffered a heart attack generally succeed in predicting only a small percentage of subsequent fatalities.
The study was conducted by CSAIL researchers John Guttag, the Dugald C. Jackson Professor of Computer Science and Engineering; Collin Stultz, the W.M. Keck Career Development Associate Professor of Biomedical Engineering; Zeeshan Syed of the University of Michigan; and Benjamin M. Scirica of Brigham and Women’s Hospital and Harvard Medical School. Their work was described in a paper published in the Sept. 28 edition of Science Transitional Medicine.
The researchers developed three new risk factors that can be identified by computers sifting through hours of a patient’s EKG recordings, a task too time-consuming for doctors and nurses to perform routinely.
“The basic idea is that there are a lot of interesting phenomena that are hard to observe looking at little snippets of data, but that are visible when you look at large chunks of data,” Guttag says. “It lets you sift through massive amounts of data and draw conclusions.”
Guttag and his colleagues examined how these new indicators might be combined with current risk-assessment methods such as echocardiograms and the TIMI risk score, which is based on a patient’s medical history. While both of these existing techniques provide useful information, the researchers sought to enhance current risk-assessment strategies.
“Today’s methods for determining which heart-attack victims need the most aggressive treatments can identify some groups of patients at a high risk of complications. But they miss most of the deaths — up to 70 percent of them,” says Syed, an assistant professor of electrical engineering and computer science at Michigan.
Using data-mining and machine-learning techniques, the researchers sifted through 24-hour continuous EKGs from 4,557 heart-attack patients enrolled in a large clinical trial led by the Brigham and Women’s Hospital/Harvard Medical School TIMI Study Group.
Believing that current risk-assessment techniques miss potentially important information, the scientists invented three new metrics, called computational biomarkers, to assist in determining the risk of death. The biomarkers —morphologic variability, heart-rate motifs and symbolic mismatch — signal heart defects and abnormalities that occur repeatedly over long periods of time. Such information is often overlooked during routine examination of a patient’s EKG recordings, which typically focus on much smaller segments of data.
Morphologic variability measures the electrical stability of the heart, a potential clue to heart-attack risk. Heart-rate motifs are patterns that could indicate heart problems. Symbolic mismatch compares a patient’s EKG history against those of both sick and healthy patients, looking for similarities and differences that could give clues to the health of the heart.
The study found a strong correlation between the three biomarkers and cardiovascular death over the two-year period following a heart attack. For example, they found that those with at least one of the abnormalities were two to three times more likely to die within 12 months of a heart attack. By adding all three computational biomarkers to doctors’ current assessment tools, the researchers say they could predict 50 percent more deaths than current techniques, with fewer false positives than predictions based on data retrieved from echocardiograms alone.
Analysis of computational biomarkers could be implemented into current medical practices without imposing any additional burden or cost, the researchers say. Information required to identify the three biomarkers is already gathered through EKG monitoring conducted during routine examinations of patients who have suffered heart attacks.
“It imposes no extra labor, no extra financial burden, it just takes data that they gather anyway and analyzes it in a different way,” Guttag says. “Therefore, it’s a lot easier to envision how it could get done than if you’d asked [clinicians] to perform some tests that they don’t routinely perform. It’s not adding any burden to the caregivers or to the patient.”
Using the biomarkers not only improved patient risk classification, but the MIT researchers also hope it will prove helpful in selecting treatment methods for patients following a heart attack.
Mark Haigney, director of cardiology and a professor of medicine and pharmacology at the F. Edward Hébert School of Medicine at the Uniformed Services University, says the study should provide assistance in identifying patients who are at greatest risk of death.
“The study demonstrates that time-series analysis … provides independent and significant prognostic information when added to conventional biomarkers,” Haigney says, adding that the population studied was representative of a large demographic among heart-attack survivors: those with well-preserved left-ventricle function. “We currently have little reliable data to help us manage these subjects,” Haigney says.
Going forward, the MIT researchers hope to apply their work with computational biomarkers to other areas of medicine where data-mining and machine-learning techniques could be used to sort through large amounts of information for clinically useful indicators.
“Our findings demonstrate that low-cost, non-invasive and easily obtained biometric data, such as the EKG, can be used to identify patients at high risk of adverse outcomes,” Stultz says. “We hope that this work ignites a new paradigm in the field of cardiovascular risk stratification.”
This research was funded by the National Science Foundation, the Center for Integration of Medicine and Innovative Technology, Quanta Computer and the Harvard-MIT Division of Health Sciences and Technology. Ongoing research is supported by the American Heart Association.