VUMC pilot to use voice commands to retrieve EHR data

What if clinicians could use voice commands to retrieve information from electronic health records in the same effortless way that consumers use virtual assistants like Amazon’s Alexa or Apple’s Siri?

It’s a future vision for medicine in which natural language recognition systems could revolutionize how physicians interact with EHRs by making the platforms as responsive as voice-controlled virtual assistants in the commercial market.

Vanderbilt University Medical Center will take a major step next month to make this vision a reality with the launch of a pilot to demonstrate a prototype of such a system, which leverages Nuance’s AI-based Dragon Medical Virtual Assistant technology integrated with VUMC’s Epic EHR.

Also See: Epic, Nuance collaborate to add AI-enabled virtual assistance

The Vanderbilt EHR Voice Assistant is the brainchild of Yaa Kumah-Crystal, MD, assistant professor of biomedical informatics and pediatric endocrinology at VUMC, who is leading the two-month pilot with 10 pediatric endocrinologists to put the system through its paces.

“We’re trying to reimagine how we’re using the EHR so that it actually helps serve providers rather than being a burdensome, onerous platform,” says Kumah-Crystal. “It just makes sense to be able to ask for the information you need and get the information back.”

Crystal_Yaa-Kumah-CROP.jpg
Dr. Yaa Kumah-Crystal for first installment of a new editorial series called Next Up that is a Q&A style format with some of VUMC's thinkers and doers.(John Russell/Vanderbilt University)

Kumah-Crystal contends that clinicians having the ability to efficiently access information from the EHR that they can query by voice is more effective than physicians being tied to their computer keyboards and screens, which she says detracts from the doctor-patient relationship.

At the same, she says there are times when a voice query doesn’t make sense for some clinical workflows when “it might be inappropriate” to talk out loud about sensitive medical information. In those cases, the system responds with text on the screen and does not articulate them out loud.

Kumah-Crystal notes that a major litmus test for the Vanderbilt EHR Voice Assistant will be the system’s response time—being able to respond to clinician queries within 200 milliseconds is the standard for voice assistant technology. She says that response latencies above that threshold would be unacceptable.

“We have a proof-of-concept model that we’re going to be using starting August 23 when we go live with the pilot in order to get feedback regarding what works and what doesn’t work,” adds Kumah-Crystal. “Structured data such as labs and medications—anything that is a discrete element—is easy to query. Free text in clinical notes that’s harder to summarize. We’re working on it but that’s second tier in terms of what we’re trying to accomplish.”

Initially, the Vanderbilt EHR Voice Assistant prototype is focused on pediatric endocrinology. However, Kumah-Crystal says ultimately the goal is to scale the system to serve as a platform for other medical specialties.

For reprint and licensing requests for this article, click here.