New AI Tool Detects Widely Underdiagnosed Heart Condition

A new Yale-developed artificial intelligence (AI) platform can detect individuals at risk for amyloid cardiomyopathy, an underdiagnosed cardiac condition, using data from a quick and accessible heart test.

Amyloid cardiomyopathy is a disease caused by the buildup of misfolded proteins on the heart muscle. Left untreated, it can be deadly. But many patients only receive a diagnosis in its later stages after the onset of dangerous complications such as heart failure.

Now, Yale scientists have built an AI platform that doctors can access from their smartphones. It screens for the condition using images from electrocardiograms (ECG), a non-invasive test that measures the heart’s electrical activity. The platform—described recently in JAMA—can help doctors detect the disease more quickly and intervene before extensive heart damage occurs.

This work builds upon Yale School of Medicine’s mission to harness AI for improving diagnosis and treatment of diseases and more effectively identifying at-risk patients.

“We are able to leverage just an image from a very simple ECG into a screening test for cardiac amyloid,” says Rohan Khera, MD, director of the Cardiovascular Data Science (CarDS) Lab at Yale School of Medicine and the study’s principal investigator. “For a disease that’s massively underdiagnosed, identifying those at risk is very critical.”

Amyloid cardiomyopathy develops when misshapen proteins form clumps and accumulate on the heart where they displace healthy cardiac muscle. There are two main types of the disease based on which protein is misfolded. The new study focused on amyloid cardiomyopathy caused by misfolded transthyretin. Researchers evaluated the ability of AI to detect its buildup.

Transthyretin is made in the liver and can become misfolded due to genetics or age. The accumulation of the protein on the heart makes the organ stiffer and impairs the electrical system that regulates heartbeat. Without intervention, the disease can eventually lead to heart failure or death. The average life expectancy for untreated transthyretin amyloid cardiomyopathy is five years.

“It’s as aggressive as some of the most aggressive cancers,” Khera says.

Unfortunately, many cases of amyloid cardiomyopathy aren’t detected until after complications arise. Part of the reason the disease is challenging to diagnose is because its symptoms can overlap with other cardiovascular conditions.

ECGs are a commonly used, non-invasive cardiac test in which electrodes measure the electrical signals of the heart. Each year, doctors around the world conduct hundreds of millions of these tests. While providers currently do not use these tests to detect individuals at risk of amyloid cardiomyopathy, the researchers wondered if they could use AI to help detect subtle patterns in ECG data to better identify vulnerable patients.

The CarDS Lab has been at the forefront of AI innovation in cardiology. It is one of the largest multidisciplinary AI labs in the world, and the scientists have built tools to detect many cardiovascular diseases. “Amyloid cardiomyopathy is one of our final frontiers because it is so underdiagnosed,” Khera says.

Researchers in Khera’s lab first trained an AI model to interpret ECG results using data from thousands of de-identified patients. Then, using the few hundred patients in the dataset already diagnosed with amyloid cardiomyopathy, they built on their initial platform to create a model that could recognize patterns in ECGs associated with the disease.

“You can take a photo of an ECG like you would take a photo of a bank check,” Khera says. “Our model can pick up from that photo whether the ECG came from somebody at risk for cardiac amyloid or not.”

In the new study, the team tested the model’s ability to detect cases of transthyretin amyloid cardiomyopathy on eight distinct cohorts of patients across the United States and Europe. Their findings revealed that the platform could successfully identify individuals with this disease subtype.

“Our tool can really narrow down the funnel for who should be further evaluated for cardiac amyloid,” adds Philip Croon, MMed, associate research scientist at Yale School of Medicine and the first author of the study.

Yale School of Medicine is committed to investing in the future of precision care and personalized medicine. The study highlights how available healthcare data can be leveraged effectively by AI to target care to the individuals who need it the most.

Cardiovascular diseases are the leading cause of death worldwide, and their prevalence is only growing due to increasing rates of risk factors such as obesity and diabetes. By diagnosing previously overlooked conditions earlier, doctors have the opportunity to change the disease trajectory before complications arise, potentially saving lives.

While in the process of developing their platform, Khera’s team received the U.S. Food and Drug Administration’s (FDA) Device Designation through its Breakthrough Devices Program. Devices in this program undergo an expedited FDA review process in order to provide more timely access to life-saving technologies. The tool is currently under FDA review.

In the meantime, scientists are now studying the implementation of AI platforms, such as this one developed in the CarDS Lab, at 13 health centers across the country. The observational study, named the TRACE-AI Network Study, will evaluate how the multimodal tools are able to detect transthyretin amyloid cardiomyopathy at a large scale.

“We’ve been able to solve a critical bottleneck in deploying therapies by identifying more at-risk people,” Khera says. “This is one of our biggest accomplishments—making care accessible by finding people who need treatment the most.”