Peer-reviewed publication in the Journal of the American Heart Association found the AI algorithm detects clinically significant murmurs comparable to expert cardiologists
Eko, a cardiopulmonary digital health company, today announced the peer-reviewed publication of a clinical study that found that the Eko artificial intelligence (AI) algorithm for detecting heart murmurs is accurate and reliable, with comparable performance to that of an expert cardiologist. The findings suggest utility of the FDA-cleared Eko AI algorithm as a frontline clinical tool to aid clinicians in screening for cardiac murmurs that may be caused by valvular heart disease.
Algorithm performance for detecting murmurs was found to have sensitivity of 90.0% and specificity of 91.4%, when excluding grade 1 murmurs which are difficult to hear. The NIH-sponsored, multisite study published in the Journal of the American Heart Association was the largest study on AI analysis of cardiac murmurs. Study investigators included collaborators from Northwestern Memorial Hospital, University of California San Francisco Medical Center, Los Alamitos Cardiology Clinic, and Mount Sinai Medical Center.
“When it comes to listening for heart murmurs, the standard of care involves a lot of subjectivity on the part of the listener,” said John Maidens, PhD, head of Data Science at Eko. “It takes expert clinicians many years to master the art of hearing and interpreting heart murmurs, and there is still a lot of variability. Our study demonstrated considerable variability even among our expert cardiologists.”
“We designed the study to see whether the algorithm could not only identify murmurs in an experimental setting, but whether it could help clinicians actually make difficult diagnostic decisions at the bedside. And that’s what the results show,” said Dr. John Chorba, lead author of the study and cardiologist at the University of California San Francisco. “I’m incredibly excited because this may just be the tip of the iceberg with the technology, and as it gets more widely used, we’ll get to see how much further it can take us in diagnosing heart disease.”
For moderate-to-severe aortic stenosis, the algorithm was found to have sensitivity of 93.2% and specificity of 86.0%. The algorithm significantly outperformed general practitioners listening for moderate-to-severe valvular heart disease, as a 2018 study showed general practitioners had sensitivity of 44% and specificity of 69%.
“We are thrilled by how Eko’s digital stethoscopes with Eko AI analysis can help any frontline provider discover meaningful, clinically important cardiac murmurs at the point of care,” said Dr. Adam Saltman, chief medical officer at Eko. “By detecting heart disease earlier, patients can be treated earlier. Costs are typically lower, outcomes are better, and they are more likely to retain a better quality of life. Appropriate patients can be referred to specialists for care when they need it. In addition, the high specificities mean that healthy patients will be much less likely to have unnecessary testing.”
“The stethoscope has changed remarkably little in its 200+ year history,” said Dr. James Thomas, director of the Center for Heart Valve Disease at Bluhm Cardiovascular Institute at Northwestern Medicine and senior author to the study. “Now for the first time, artificial intelligence is being applied to stethoscopic heart sounds to improve the auscultation skills of health care professionals. This technology has the promise to extend this expertise to all clinicians.”