, a new study shows.
The results suggest collecting relevant sounds through a stethoscope (auscultation) using AI-powered technology is an important primary care tool to detect VHD, study author Moshe A. Rancier, MD, medical director, Massachusetts General Brigham Community Physicians, Lawrence, Massachusetts, said in an interview.
“Incorporating this AI-assisted device into the primary care exam will help identify patients at risk for VHD earlier and eventually decrease costs in our healthcare system,” he said, because timely detection could avoid emergency room visits and surgeries.
The findings were presented at the annual scientific sessions of the American Heart Association.
VHD Common
Clinically significant VHD, indicating structural damage to heart valves, affects 1 in 10 adults older than 65 years. Patients may be asymptomatic or present to their PCP with an unspecific symptom like fatigue or malaise.
If VHD is undiagnosed and left untreated, patients could develop more severe symptoms, even be at risk for death, and their quality of life is significantly affected, said Dr. Rancier.
Cardiac auscultation, the current point-of-care clinical standard, has relatively low sensitivity for detecting VHD, leaving most patients undiagnosed.
The deep learning–based AI tool uses sound data to detect cardiac murmurs associated with clinically significant VHD. The device used in the study (Eko; Eko Health) is approved by the US Food and Drug Administration and is on the market.
The tool identifies background sounds that might affect the evaluation. “If there’s any noise or breath sounds, it tells me this is not a good heart sound, and asks me to record again,” said Dr. Rancier.
A doctor using the AI-assisted stethoscope carries out the auscultation exam with the sound data captured by a smartphone or tablet and sent to the AI server. “I get an answer in a second as to if there’s a murmur or not,” said Dr. Rancier.
Not only that, but the tool can determine if it’s a systolic or diastolic murmur, he added.
Real-World Population
The study enrolled a “real-world” population of 368 patients, median age 70 years, 61% female, 70% White, and 18% Hispanic without a prior VHD diagnosis or history of murmur, from three primary care clinics in Queens, New York, and Lawrence and Haverhill, Massachusetts.
About 79% of the cohort had hypertension, 68% had dyslipidemia, and 38% had diabetes, “which aligns with the population in the US,” said Dr. Rancier.
Each study participant had a regular exam carried out by Dr. Rancier using a traditional stethoscope to detect murmurs and an exam by a technician with a digital stethoscope that collected phonocardiogram (PCG) data for analysis by AI.
In addition, each patient received an echocardiogram 1-2 weeks later to confirm whether clinically significant VHD was present. An expert panel of cardiologists also reviewed the patient’s PCG recordings to confirm the presence of audible murmurs.
Dr. Rancier and the expert panel were blinded to AI and echocardiogram results.
Researchers calculated performance metrics for both PCP auscultation and the AI in detecting audible VHD.
The study showed that AI improved sensitivity to detect audible VHD by over twofold compared with PCP auscultation (94.1% vs 41.2%), with limited impact on specificity (84.5% vs 95.5%).
Dr. Rancier stressed the importance of sensitivity because clinicians tend to under-detect murmurs. “You don’t want to miss those patients because the consequences of undiagnosed VHD are dire.”
The AI tool identified 22 patients with moderate or greater VHD who were previously undiagnosed, whereas PCPs identified eight previously undiagnosed patients with VHD.
Dr. Rancier sees this tool being used beyond primary care, perhaps by emergency room personnel.
The authors plan to follow study participants and assess outcomes at for 6-12 months. They also aim to include more patients to increase the study’s power.