Fueled by the recent availability of chatbot interfaces like Chat-GPT, generative AI has become a hot topic across many industries, and healthcare is no exception. Several EHR vendors are tapping the technology to streamline administrative workflows with the hope of giving clinicians more time to deliver patient-centered care.
Widespread EHR adoption has revolutionized healthcare in several ways, including improved care coordination and clinical decision support.
However, EHR use has contributed to a growing trend of clinician burnout. A 2019 study found that forty percent of physician burnout is attributable to EHRs.
Generative AI could help automate time-consuming EHR workflows.
As defined by the Government Accountability Office (GAO), generative AI is "a technology that can create content, including text, images, audio, or video, when prompted by a user."
Generative AI systems learn patterns and relationships from massive amounts of data, allowing them to generate new content that may be similar, but not identical, to training data. The technology processes and produces content using machine learning algorithms and statistical models.
Recently, EHRIntelligence caught up with four EHR vendors to discuss how they are integrating generative AI into their platforms to support clinical efficiency.
Generative AI is best suited for automating tasks that are administrative and repetitive, according to Phil Lindemann, vice president of data and analytics at Epic. Take responding to patient portal messages, for example.
"The clinician or support team essentially has to take all of the data points that they've got in their head and turn that into a narrative human response," Lindemann told EHRIntelligence in an interview. "Generative AI can draft a response that the clinician can then review, make changes as necessary, and then send to the patient."
A group of early adopter health systems are currently testing the new Epic feature, which aims to speed up message response time and allow doctors to spend more time with patients. However, the functionality is just one of 60 generative AI projects Epic is working on.
Another example is a tool to summarize new information since the last time a provider saw a patient. Instead of searching a patient's health record to understand what happened since their last visit, the tool synthesizes information to provide physicians with a synopsis.
The next phase of Epic's generative AI integration will focus on helping patients better understand their health records, Lindemann said.
For instance, clinicians usually document diagnostic test results in a mode intended for doctor-to-doctor communication. Generative AI could help explain test results to patients in words they understand.
"We think that the ability to be a universal translator and upcode and downcode to an education level or a reading level is incredibly powerful," Lindemann said.
What's more, generative AI can translate information into different languages.
"Health systems spend a lot of time trying to make patient education and different things available in certain languages, but they'll never have every language possible," Lindemann said. "This technology can take human language, translate it at any reading level in any language, and have it understandable."
Lindemann emphasized that while there are several beneficial use cases for generative AI, the technology will not solve all of healthcare's challenges.
"We see it as a translation tool," he said. "It's not a panacea, but there's going to be really valuable use cases, and the sooner the community can agree on that, the more useful the technology's going to be."
Oracle Health is beta-testing a generative AI-based chatbot for healthcare professionals that aims to automate administrative workflows.
The EHR vendor built the tool upon the Oracle Digital Assistant, a platform for creating conversational interfaces or chatbots.
"We see some excellent opportunities to help remove or lessen the administrative burden that our providers are facing today, so we're applying all the learnings that we've had over the past decade or so in AI," said Suhas Uliyar, senior vice president of product management for healthcare AI and intelligent automation at Oracle.
The Clinical Digital Assistant tool has several core capabilities, including summarization of patient information, which allows clinicians to review patient charts using voice. For instance, a provider could ask the system, "What does my day look like?" or "Tell me a little bit about my next patient," and the tool surfaces relevant information.
The technology also supports automated note generation by listening to patient-provider conversations. Once a visit is over, the system automates clinical notes for provider review.
Suhas noted that during this mode, providers are able to "interrupt" the digital assistant and ask it to pull up relevant patient information. For example, while the doctor is talking to the patient, they can ask the tool to pull an x-ray or the patient's last three A1C levels.
"The physician doesn't need to look away from the patient when they're doing this," Uliyar said. "They can ask the digital assistant, and it can surface the contextual information as they're having this conversation with the patient. Having that eye-to-eye contact with the patient goes a long way to make them feel like they're being heard and that someone's there to take care of them."
The tool can also propose next actions within the EHR based on the encounter discussion. For example, if a provider mentions prescribing medication or scheduling labs during the encounter, the EHR will suggest these actions. Uliyar emphasized that the physician must review and sign off on all orders before the system sends them out.
The clinical digital assistant entered beta one in October. Oracle Health plans to make the solution generally available in the second quarter of the 2024 calendar year.
EHR vendor eClinicalWorks is also leveraging generative AI-based technology to help automate clinical documentation. In October 2023, eClinicalWorks partnered with health IT vendor sunoh.ai to integrate ambient listening technology within the EHR.
"Language models have a huge potential in impacting almost every workflow," said Girish Navani, CEO of eClinicalWorks. "Whether it's reading information and summarizing it or creating the right type of contextual response, language models can help reduce cognitive load. We believe that generative AI has the potential of being a personal assistant for every doctor, and that's what we're working on."
The EHR integration uses ambient listening to generate draft clinical documentation. Providers must accept the note or make any necessary edits before the system documents the information.
"It could save hours," Navani said. "You capture the essence of the entire conversation without touching a keyboard. It is transformational in how it works and how well it presents the information back to the provider."
Sunoh.ai also provides tailored highlights and relevant next steps for the provider to review and approve based on their conversation with the patient. For instance, the integration captures lab, imaging, and medication orders, referrals, and follow-up appointment details.
Early adopters are currently using the EHR integration, and the offering will be generally available to customers in early 2024.
EHR vendor MEDITECH is collaborating with Google on a generative AI project to streamline clinical documentation, with an initial focus on hospital discharge summaries.
While critical for care coordination, discharge summaries are time-consuming for clinicians, especially for patients with longer stays, according to Helen Waters, executive vice president and COO of MEDITECH.
"Providers are asked to go in and review previous notes and results and try to bring that all together," Waters said. "Generative AI can help auto-populate the discharge note by bringing in the discrete information that would be most relevant to substantiate that narrative and enable time savings for those clinicians."
In essence, the system generates a draft version of the discharge summary. Before documentation in the EHR, clinicians must approve the note and make any necessary edits.
By allowing clinicians to expedite discharges, the feature seeks to address clinician burnout and workforce shortages currently plaguing the industry.
MEDITECH is currently identifying clients interested in testing the generative AI use case for hospital discharge summaries. Waters said the EHR vendor plans to work closely with healthcare organizations to deploy the solution.
Customers testing the tool will work with real patient data to do validation testing against the EHR before the feature is generally available.
According to Waters, ensuring a measured and thoughtful approach to implementation in clinical workflows will be key to driving the adoption of generative AI tools.
"We'll take it to customers to make sure there are no hallucinations or inaccuracies in what's being summarized," she said. "Getting comfortable with the technologies and confidence levels in the technology is our first step. If we go out of the gate too soon with something and physicians don't immediately trust or think it's adding value, we will have lost time and effort to a promising technology."