Large language model-based system identifies cardiac event data in cancer patients’ electronic health records

Patients with breast and lung cancer are at increased risk of cardiotoxicity, or heart-related damage caused by cancer treatments, because of the proximity of the heart, lungs and breasts. Cardiotoxicity increases the risk of heart attacks, heart failure and other cardiac conditions. Looking for evidence of cardiac disease in these patients requires reading through hundreds of patients’ health records, which is often too time-consuming to be practical.

Thomas Jefferson University researchers have created an algorithm for large language models (LLMs) that can successfully extract cardiac event data from patients’ electronic health records (EHRs) in a fraction of the time it takes humans to comb through EHRs. The paper was published in the International Journal of Radiation Oncology, Biology, Physics.

“It took physicians, residents and students a long time to collect patients’ data manually,” says Wenchao Cao, Ph.D., the study’s first author. “We hope our study demonstrates that this process can be taken over by the large language model to save time.”

Researchers searched for evidence of cardiotoxicity in patient EHRs using open-source LLMs, an advanced type of artificial intelligence. The prompting framework they created discovered cardiac event data in patient EHRs 71% to 85.5% of the time. Prompts identified keywords and false positives, so “ruled out a heart attack” wasn’t flagged as a cardiac event.

“You would be surprised to see how many different ways there are to describe a negative situation, like ‘patient denied’ a certain disease,” Cao says.

Humans and LLMs scoured the EHRs of 411 breast and lung cancer patients. Humans took about two hours to review each EHR, while the LLMs took 20 to 42 seconds.

Further refinement of the LLM framework may advance cardiotoxicity research.

“Hopefully we can identify patients who are more likely to develop cardiotoxicity,” Cao says. “This could inform a more individualized radiation therapy plan to improve patient outcomes and reduce side effects.”

Jefferson oncology resident Nilanjan Halder, MD; medical students in Jefferson’s summer oncology program—Isis Lloyd, Moorin Khan, Michael Dichmann and Patrick Faherty—and Jefferson undergraduate Femi Adejolu contributed to the research.

Who’s behind this story?

Sadie Harley

BSc Life Sciences & Ecology. Microbiology lab background with pharmaceutical news experience in oil, gas, and renewable industries.

Full profile →

Advertisements

Andrew Zinin

Master’s in physics with research experience. Long-time science news enthusiast. Plays key role in Science X’s editorial success.

Full profile →