A CHOC team has been selected to be part of a national study that will use natural language processing (NLP) models to extract patient information from clinical notes in electronic health records (EHRs) in an effort to develop scalable methodologies for using machine learning and NLP to fill gaps in knowledge in pharmacoepidemiology studies.
As a use case, this analysis involves borrowing select methods from an existing study that examined the potential relationship between an asthma medication and mental health effects with a key difference, i.e., integrating more complete information using NLP on clinical notes from multiple health systems in addition to EHR and claims data.
Funded by the U.S. Food and Drug Administration through the Sentinel Innovation Center, the project is the latest example of how AI may be used to enhance healthcare studies using real-world data.
Two-year project
CHOC is the only pediatric healthcare system participating in the two-year, multi-center project. Its primary role is providing mental health and pediatric pulmonary expertise while secondarily providing feedback on the AI portions of the work, says psychiatrist Dr. Michael Chu, one of six CHOC associates involved in this project.
EHRs contain terabytes of rich unstructured data but manual analysis is not efficient in making sense of the information. That’s where NLP comes in. It’s a tool that gives computers the ability to understand large volumes of text and spoken words rapidly – even in real time.
The FDA’s Sentinel Innovation Center selected Cerner Enviza (an Oracle company) to lead the project. They are collaborating with John Snow Labs to develop the NLP methodology tool using the structured and unstructured Oracle EHR Real-World Data and commercial claims data.
Looking at the asthma drug montelukast and its possibility of mental health side effects, the two-year project will demonstrate how the use of machine learning and NLP technology with unstructured data may help fill gaps in knowledge.
Making sense
By developing an NLP technique to utilize multicentered unstructured EHR data from doctor’s notes that is transportable to other EHR systems, this broad-based team hopes to advance population health studies and patient safety.
“Most computer scientists aren’t clinicians,” notes Dr. Chu. “That’s where CHOC comes in. For example, if a note in an EHR indicates a patient has hallucinated, is that the same as having delusions? The AI tool will group or separate words that CHOC will help it make sense of.”
Along with CHOC, National Jewish Health and Kaiser Permanente Washington Health Research Institute are providing clinical expertise and consulting on the project.
Beginning stages
Dr. Chu, who sees patients on the medical floors of CHOC as one of the hospital’s two consultation liaison psychiatrists, cares for children and adolescents with a wide variety of mental health conditions. He collaborates with patients, their families, and their health teams on healing children who may be struggling with mental health needs while in the hospital.
Dr. Chu is working on the project with colleagues Dr. Heather Huszti, chair of the division of pediatric psychology; Dr. Hoang “Wayne” Nguyen, director of psychiatry; Dr. Olga Guijon, division chief, primary care; and Louis Ehwerhemuepha, PhD, director of Research Computational and Data Science (Computational Research) at CHOC.
The FDA’s Sentinel Innovation Center hopes the project will provide a better understanding of the promise of NLP on extraction of real-world evidence around pharmacological effects of medications on large populations.
“This study’s use case involves looking at a specific asthma drug,” Dr. Chu says, “but if this works out, we could extrapolate this method and apply it to other things, and that could be huge.”
The project is at the beginning stage with results expected in late 2024.
Learn about pediatric research and clinical trials at CHOC
