Preparing AI-Powered Virtual Assistants to Improve EHR Usability

Vanderbilt University Medical Center is exploring the potential for Epic EHR-integrated virtual assistants to boost EHR usability.

Virtual assistants may help to improve EHR usability with some fine-tuning.

By Kate Monica - 

Problems with EHR usability continue to counter the potential benefits of EHR systems and leave physicians, nurses, administrators, and patients by and large feeling dissatisfied and frustrated with the technology.
While patient EHRs contain a wealth of potentially actionable clinical information providers and researchers can use to improve clinical decision-making, care delivery, and patient health outcomes, administrative burden associated with documenting and retrieving health data still weighs heavily on providers.
EHR-integrated, voice-activated virtual assistants powered by artificial intelligence (AI) may help to resolve some of the industry’s longstanding problems with EHR usability. Virtual assistants enable providers to dictate physician notes, request specific data, order prescriptions, and view lab results without touching their keyboards. 
But is the technology effective, or will it only cause more frustration for providers? That’s the question providers at Vanderbilt University Medical Center (VUMC) hope to answer.
VUMC Assistant Professor of Biomedical Informatics Yaa Kumah-Crystal, MD, is currently working with Nuance to develop an EHR-integrated virtual assistant capable of summarizing patient health information and dictating that information to providers in response to verbal queries.
READ MORE: EHR Usability, Workflow Strategies for Reducing Physician Burnout
Kumah-Crystal uses a prototype of the Nuance virtual assistant integrated into VUMC’s Epic EHR system.
“I've been testing it, and I actually use it on a regular basis to review some of the content in the patient's chart before I walk into the room to see them,” Kumah-Crystal told “We have another provider who's testing it as well, and in the next month, we're going to be doing some studies with the rest of our clinicians.”
While new and emerging virtual assistants will be used to complete a variety of tasks within provider EHR systems, VUMC is testing a product primarily designed to ease patient management and health data access for providers.
“The main purpose of the voice assistant is to help summarize the information that's hard to get out of the EHR,” Kumah-Crystal explained. “We’re going to have providers there with the scheduled patients that they're going to see within the next week. We’re going to review the patient's information in the chart.”
VUMC providers who participate in upcoming studies exploring the efficiency of voice recognition software and virtual assistants will use Nuance technology to find information about patient’s vital signs, medication lists, problem lists, and general health status. Kumah-Crystal and her team will then compare individual provider’s experience using virtual assistants with each provider’s experience retrieving information manually. The VUMC team will do a timed task analysis to determine whether providers are able to access and review health data more quickly using either method.
READ MORE: Boosting EHR Usability, Safety Throughout the EHR Life Cycle
“We’ll also want to see their perception of feasibility, usability, and how it performs compared to the regular EHR,” said Kumah-Crystal.
So far, VUMC researchers have found evidence to support claims that virtual assistants can boost clinical efficiency. In one early study of medication ordering, researchers at VUMC found using virtual assistants was significantly faster than using a mobile application.
“If someone is on the go and they’re trying to be efficient, it would make more sense to ask for prescriptions to be written rather than going to the medical interface to type it out,” said Kumah-Crystal.
But efficiency isn’t the only factor that determines whether a provider will actually use a new technology. Perception and usability also affect clinician buy-in and satisfaction.


According to VUMC research, providers generally only preferred virtual assistants when they understood commands accurately the first time.
READ MORE: Different EHR Implementation Strategies May Affect EHR Usability
“We found a correlation between how many times the person has to make a request and the perception of the ease of use,” explained Kumah-Crystal. “When it understands them the first time, people are just absolutely delighted. When they don't understand you, the frustration just bubbles up right away.”
“That was one of the really interesting findings from the study — the fact that it has to be at a certain threshold for it to be considered as good as accomplishing it the other way, and better,” she added.
In addition to the frustration of repeating commands multiple times, some providers have expressed a lack of interest in adopting virtual assistant technology because it can only help with simple tasks in its current iteration.
“We've been working with Mercy Hospital to circulate a survey about how providers used its voice technology in their natural lives — not even directly related to the EHR — but how they used Siri and Alexa and all those other things,” said Kumah-Crystal.
“The main reason people say they wouldn't use voice technology is because it doesn't understand them well or it doesn't do the complex things they want them to do,” she continued.
While voice recognition software and virtual assistants may be useful for retrieving information from patient EHRs, there are questions surrounding whether the technology is ready to record provider dictation.
2018 JAMA study found clinical documentation generated through voice recognition software is often error-prone, with physician notes generated through Dragon Medical and Nuance’s eScription showing a 7.4 percent error rate. Clinician review can help to reduce dictation errors and boost the accuracy of voice recognition-generated notes.
Some providers are also concerned about potential problems with patient privacy.
“We’re working on and researching the best ways to communicate certain information due to concerns about speaking some information out loud and having it spoken back,” Kumah-Crystal said.
Nuance partnered with VUMC to address these and other provider concerns as soon as possible. VUMC Lead Developer Daniel Western Albert, MD, prioritizes getting new solutions into users’ hands quickly.
“One of the things that we always do when we build a new solution like this is try to iterate quickly, so deliver initial functionality — what we call the minimum viable product,” said Albert. “Let them actually use it, and give us some feedback to improve it.”
VUMC is currently in the process of building a use case for utilizing virtual assistants to prep providers for patient visits.
“You're gathering information to build a picture in your mind about who the patient is and what the next steps and their care will be,” said Kumah-Crystal.
The tool could be especially effective for prepping providers during their morning commutes. Kumah-Crystal hopes to further develop commute summaries through virtual assistants after getting more active participation with the existing workflow.
“The concept of a commute summary where you know that you have a schedule of 10 patients for the day, you have a 30-minute commute to work, and you can say, ‘Tell me about my patients,’ or, ‘What's today look like?’” said Kumah-Crystal. “And then it summarized for you just patients on your list while you're on your commute.”
“It saves me time as a physician, and I'm oriented by the time I walk in,” she added.
While concepts like the commute summary hold potential, Kumah-Crystal recognizes that some providers will be hesitant to integrate virtual assistants and voice recognition software into their workflows.
“They're right to be reluctant to completely adopt something that's not tried and true and tested,” she said. “And that's exactly what we're doing.”