VA building AI-powered knowledge graph of veteran health
- By Stephanie Kanowitz
- Dec 19, 2016
A partnership between an artificial intelligence firm and the Department of Veterans Affairs seeks to improve health care for veterans while enabling the company to test its technology.
Through the five-year collaboration, which was finalized in October and announced at the end of last month, Flow Health will use machine learning techniques to take aggregated and anonymized VA data and create a knowledge graph that health practitioners can use to better treat veterans.
“Imagine having artificial intelligence that is clinically validated that can help individualize medical decision-making,” Flow Health CEO Alan Meshkin said. The technology can facilitate more personalized care planning by using all of the data resources that the VA has, he said. It can look at past events, “learning from that continuously to help make doctors better doctors.”
The company is tapping into VA’s electronic health records system, the Veterans Information Systems and Technology Architecture, to draw on more than 30 petabytes of longitudinal clinical data from VA records on 22 million veterans over 20-plus years. Longitudinal data looks repeatedly at the same data points and variables over time and can include not just structured data, but also unstructured information like clinical notes.
“A lot of the work is ingesting the clinical notes. That’s probably the largest source of meaningful unstructured data,” Meshkin said. “It’s using natural language processing to really understand … the physicians’ observations when seeing a veteran.”
The company will integrate the data from separate VA sources and build a single cloud-based data repository of veterans’ health information. After scrubbing any identifying information to comply with Health Insurance Portability and Accountability Act privacy requirements, Flow Health will seek out relationships in the graph – classifying and clustering the data -- using algorithms as well as machine and deep learning methods. For example, when a certain set of symptoms appears with frequency in a patient, Flow Health’s platform can start to recognize the pattern and make predictions about what they mean for diagnoses down the road, Meshkin said.
“What’s really unique about the VA dataset versus a lot of other private-sector health systems is that they’ve seen these veterans before being diagnosed with, say, lung cancer, so we’re able to find those patterns and early symptoms,” he said. “I think that’s going to be one of the really powerful aspects of our data analytics work here.”
The graph will make possible what humans can’t, he added. “It’s something that would be impossible to expect a doctor to digest all the history of every patient who’s ever been seen,” Meshkin said. “It’s basically the understanding of the aggregate relationships.”
Once the graph is made, it will be available to VA workers via an application programming interfaces integrated into their workflow. “The idea is that tight integration back to the [electronic medical records] so they don’t have to start using another interface, which is always very challenging,” Meshkin said.
VA is increasingly using data to research health.
A recent study by Dr. Jennifer Garvin with the VA in Salt Lake City, Utah, tested a natural language processing prototype that searches through unstructured information in case notes for data related to chronic heart failure.
The VA’s Million Veteran Program aims to build one of the world’s largest medical databases for disease research by collecting blood and health information on 1 million veteran volunteers. Although Flow Health’s work is separate from MVP, Meshkin said there could be collaboration later on.
Flow Health is set to begin operationalizing its plan in January and expects no changes as a result of the upcoming change in administrations as President-Elect Trump takes office.
“Everyone is very excited -- both sides of the aisle,” Meshkin said. “Everyone wants to improve care for the veterans, and modern technology and approaches that are used outside of the federal government, outside of health care, are definitely the answer.”
Stephanie Kanowitz is a freelance writer based in northern Virginia.