From data to display, Pitt is giving the EKG an upgrade

Researchers at Pitt recently developed a machine learning algorithm to interpret electrocardiogram (EKG) data that outperforms medical professionals when it comes to diagnosing and classifying heart attacks, according to their publication in Nature Medicine.

Now Christian Martin-Gill, professor of emergency medicine in the School of Medicine, and a team of researchers are working with medical specialists in a variety of fields to develop a user-friendly platform to access the algorithm's findings whether the user is in the field, the dispatch center or the cardiology wing of a hospital.

Today, a paramedic relying on an EKG during an emergency transmits the data to an emergency physician. Along the way, the data is interpreted through a basic algorithm that provides a limited interpretation. It shows up on the physician's screen as a version of the familiar peaks and valleys waveform that describes the electrical activity of the patient's heart.

The addition of a new algorithm wouldn't change much from the paramedic's point of view, "but the physicians would have access to a dashboard in real-time instead of only an image of the EKG," said Martin-Gill, who is also chief of the Division of Emergency Medical Services.

While the readout a physician sees now provides a few key features of the data, "What we've been able to develop provides enhanced feature detection and substantially more information about that EKG."

Over the next four years, supported by an NIH grant, Martin-Gill and colleagues - including former School of Nursing faculty member Salah Al-Zaiti - will continue to work with medical professionals working in the field to craft an interface that can help them make the right calls, particularly in cases where other clues might not show up.

Per Martin-Gill: "They've all been very engaged and have provided very helpful feedback from each of their individual perspectives," from the paramedics responding to emergencies, to emergency physicians providing guidance with very little information, to cardiologists downstream who can make better decisions about which patients may benefit from earlier interventions.

Perhaps as important, the features highlighted by the dashboard will not only help this group of medical professionals determine who has likely had a heart attack, but also, who likely has not had one.

That's because a lack of clarity can lead to a barrage of tests, an intense assessment of personal history and days under observation in a hospital. The additional information the dashboard provides may not only save someone's life who's had a heart attack, it can save time, money and the stress of uncertainty for someone who hasn't.

New medical technologies are often described in medical journals, where their potential to save lives is just that, potential. They can't save lives if they don't make the transition from paper to clinic.

"There are many, many publications where individuals are creating these kinds of computer models that can help predict this or that," Martin-Gill said. But often, research isn't translated to real-world applications.

The research team has developed a three-phase process to overcome this problem, helping to ensure that what began as an idea will eventually be found in hospitals and used to help real patients.

The initial phase was developing the algorithm the researchers are using, which continues to be perfected over time.

Currently, the team is in the second phase: designing an interface for medical professionals that's informative, easy to interpret, and useful in clinical decision-making. They'll soon be seeking additional funding for phase three: testing the new algorithm and interface in a clinical environment. Then they can move to implementation in the real world.

"That translation piece is where more work needs to happen so that we can get all of the real-world value from these technologies and algorithms that are out there, published and in the academic world.

"We need to figure out ways that we can translate that knowledge into clinical use," Martin-Gill said. "The exciting thing about our work is that we're currently creating a real interface that will make those types of algorithms usable to a clinician, and that can actually be implemented and potentially impact patient care."