A new tool in development by the University of Pittsburgh and UPMC uses artificial intelligence to analyze EKG data, identifying heart attacks more quickly and more accurately than current methods. Researchers published their results of the tool's performance in late June in the journal Nature Medicine.
"The classic adage for heart attacks is that time is muscle," said Christian Martin-Gill, chief of the Emergency Medical Services division at UPMC and a co-author of the study. "If we can identify these partial blockages or heart attacks that we might otherwise miss for hours or days, then we have an opportunity to intervene more aggressively."
Researchers from Pitt and UPMC have been working to develop this tool for more than a decade, and published their research after testing it on real-world patients in Pittsburgh and North Carolina. Study co-author Ervin Sejdić developed the model at Pitt and has since moved to a position at the University of Toronto.
A standard EKG takes data from multiple electrodes placed on the chest and synthesizes that data into a set of wavy lines on graph paper that clinicians are trained to read. The new tool from Pitt is more accurate in part because it can look at far more data — all the readings from each electrode — than the synthesized version provides.
Using machine learning, researchers developed the model by feeding it all of that data, along with outcomes, and allowing it to figure out its own system of analysis.
"With machine learning, you do not give rules to the computer, you give the data," said Salah Al-Zaiti, lead author of the study and associate professor in the Pitt School of Nursing. "Eventually, the machine learning model learns the rules and how to reach from EKG to a certain conclusion. It may start learning rules that we do not know about and start identifying new insights."
Researchers were surprised to find, through real-world tests, that the model had not just matched current science in accuracy of identifying heart attacks, but was actually more accurate.
Before beginning treatments or further diagnostics, doctors want to know they are treating a heart attack and not a condition such as reflux or pneumonia that may have similar symptoms.
"Part of the challenge is that we haven't had a way to identify these early enough to know what we should do to prevent some of that heart damage," said Martin-Gill.
Widespread use of the tool is still years away, but the research team is currently developing a user interface for the model, which it plans to deploy in partnership with the city of Pittsburgh's Bureau of Emergency Medical Services.
The cloud-based model under development will be able to integrate with hospital command centers, said Al-Zaiti, providing real-time guidance on medical decisions.
"This information can help guide EMS medical decisions such as initiating certain treatments in the field or alerting hospitals that a high-risk patient is incoming," Martin-Gill said. "On the flip side, it's also exciting that it can help identify low-risk patients who don't need to go to a hospital with a specialized cardiac facility, which could improve prehospital triage."
©2023 the Pittsburgh Post-Gazette. Distributed by Tribune Content Agency, LLC.