The model identified risk patterns in ECGs that human experts did not.
Photo/Thumb have been modified. Courtesy of spainter_vfx - stock.adobe.com.
Using more than 2 million electrocardiograms (ECGs) from more than three decades of archived medical records, an artificial intelligence (AI) model was able to pinpoint patients at an increased risk of developing a potentially dangerous irregular heartbeat or of dying within the next year.
A deep learning AI model predicted that within the top 1% of high-risk patients, one out of every three was diagnosed with atrial fibrillation (A-fib) within a year. The AI model accurately predicted the risk of death in patients who physicians deemed had a normal ECG, researchers reported.
“This is exciting and provides more evidence that we are on the verge of a revolution in medicine where computers will be working alongside physicians to improve patient care,” said senior author Brandon Fornwalt, M.D., Ph.D., associate professor and chair of the imaging science and innovation department at Geisinger.
Researchers at Geisinger completed two studies using AI and ECG data to predict the risk of two future health events — A-fib and mortality.
The research team believed that a deep learning model could predict A-fib before it develops. The researchers used 1.1 million ECGs that did not indicate the presence of A-fib in more than 237,000 patients.
Highly specialized computer hardware trained the AI model to analyze 30,000 data points for each ECG.
Along with one out of three of the top 1% of high-risk patients being diagnosed with A-fib within a year, those predicted to develop the condition at one year had a 45% higher hazard rate in developing A-fib over a 25-year follow-up than the other patients.
“Being able to understand who is at risk for having irregular heartbeats or atrial fibrillation then helps us understand who may be at risk of also having a stroke and then treating these individuals and preventing both atrial fibrillation and perhaps a stroke down the road. Having these techniques at our fingertips and having more precise techniques to uncover potential atrial fibrillation now or in the future, is absolutely tremendous.” - Jennifer Hall, Ph.D., chief of the American Heart Association Institute for Precision Cardiovascular Medicine.
Researchers analyzed the results of 1.77 million ECGs and other records from nearly 400,000 patients to identify those most likely to die of any cause within a year.
The data were used to compare machine learning models that either directly analyzed raw ECG signals or needed human-derived measures and commonly diagnosed disease patterns.
The AI model that directly analyzed raw data was better at predicting one-year risk of death, the researchers found. The model detected risk patterns that cardiologists were unable to recognize, making them wrongly identify the ECG as normal.
“This is the most important finding of the study,” said Fornwalt. “This could completely alter the way we interpret ECGs in the future.”
The research will be presented at the American Heart Association’s Scientific Sessions 2019 from Nov. 16 to 18 in Philadelphia.
Get the best insights in digital health directly to your inbox.