What an AI program found in an ECG

Quiros visited the ED in February 2025 saying he had been coughing up blood and finding it difficult to breathe for the past four days. His heart was beating fast, and he wasn't getting much air to his lungs, but an X-ray of Quiros' chest didn't find any abnormalities.

Quiros received an ECG that was abnormal but didn't show anything that would lead to a clear diagnosis, instead indicating that he might have coronary heart disease, something rare in a person of his age. Then, the doctors in the ED learned that Quiros had been exposed to wildfire smoke on a recent visit to California, so they sent him home with asthma medication and an inhaler.

Later, as part of a clinical trial, researchers used an AI program called EchoNext to analyze ECGs from patients across the NewYork-Presbyterian medical system and determine whether it could detect signs of heart damage that humans might miss.

EchoNext was designed by Pierre Elias, medical director of AI and a cardiologist at NewYork-Presbyterian and Columbia University Irving Medical Center, and his colleagues. Elias said that EchoNext reads an ECG less than 10 minutes after it's performed, and they analyze almost 500,000 ECGs each year.

EchoNext analyzed Quiros' ECG and found evidence of possible severe heart damage. The team at the hospital called Quiros back a week later for an echocardiogram, which shows the beating heart. They found that Quiros' heart was beating so slowly that just 10% of its blood was pumped out with each contraction. In addition, his mitral valve was leaking blood back into his heart.

After investigating the cause of Quiros' problems using genetic testing, doctors discovered Quiros had a rare genetic disorder associated with sudden death. To save his life, doctors performed a heart transplant. Quiros' case study was published in June in Nature Medicine.

EchoNext can flag possible poor pumping action, damage to heart valves, abnormally thick walls of the heart, and high blood pressure in the part of the lungs connecting the heart — all things that can indicate serious damage.

Zack said EchoNext was chosen as the first AI program available in OpenEvidence because it was recently approved by FDA, meaning it was sufficiently validated to be marketed and advertised. Even after AI models are approved, "it typically will take a long time before they are used outside of Boston or San Francisco," where major medical centers are, Zack said.

Elias said it's been a struggle at times to explain to doctors that AI can help find patterns in data and images that humans miss, which includes things that, upon closer examination, can seem obvious.

"The reality is that who is sick and who is not is not black and white," Elias said, adding that even if an ECG is abnormal, "many ECGs are abnormal. If we ordered echoes for every single abnormal ECG, we'd probably bankrupt health care."

"To say an otherwise healthy 45-year-old who is coming in with mild shortness of breath had horrendous heart failure based off the ECG would be shocking," Elias added.