ipopba - stock.adobe.com
EHR interoperability gaps threaten to limit AI's potential
In our second story in a series on AI in healthcare, healthcare leaders discuss the challenges in getting AI tools to work as intended amid EHR integration and interoperability challenges.
Achieving interoperability of health IT tools is an ongoing challenge for providers, and experts say that barriers to EHR integration and secure data sharing could hamper AI tools from delivering on their promises.
Following a New York City panel hosted by health AI company Heidi earlier in the summer, healthcare leaders spoke with Health IT and EHR, discussing the pros and cons of integrating AI tools into EHRs as well as how a lack of interoperability could be impacting that integration.
Johnathan Lyon, assistant director of behavioral health services at Hawse Health, a nonprofit federally qualified health center in rural West Virginia, highlighted the difficulties of integrating AI with EHRs. In fact, his team opted not to integrate Heidi, an AI scribe, with their EHR platform.
"Part of that was due to technological constraints, but Heidi's actually more reliable than our EHR," Lyon said. "[With] our EHR, we've had significantly more downtime than we've had downtime with Heidi, so it's easier just to run that and transcribe it later, or copy/paste to the EHR later."
However, Nancy Cibotti, M.D., regional chief medical information officer at Heidi Health and a partner at MD2,, a primary care practice in Boston, noted the advantages of integrating ambient AI with the EHR. Heidi was not integrated with the EHR when she first deployed the AI platform at Beth Israel Lahey Health, her former employer. Six months after BILH implemented it, the AI tool was launched within Epic, Cibotti said. This offered some significant benefits for clinicians.
"Less clicks, and you can also pull some data into Heidi from Epic," Cibotti said.
The ease of integrating AI tools into EHRs depends on the flexibility of the latter, she acknowledged.
"Some are more rigid than others," Cibotti said. "For example, at Heidi, our templates are highly customizable, but some of that can be lost if the EHR doesn't accommodate."
She also noted that sometimes integration with EHRs can limit the functionality of AI features.
"Integration isn't the be-all and end-all," Cibotti said. "I think it's more important to have the right functions and the right features in an AI."
So, providers must sometimes choose between integrating tools and having the right AI features, she suggested.
"Seamless AI integration is the goal, but it depends on the characteristics of the EHR," Cibotti added. "Integration doesn't always equal convenience."
Cibotti envisions the user interface of a clinical AI tool sitting on top of an EHR rather than within it, so clinicians would only need to interact with AI tools, rather than the health record, to do their job.
"The EHR is still important for data storage, but the AI is the clinician's tool in providing care," Cibotti said.
Lack of EHR interoperability may hinder AI
Interoperability standards affect AI integration, potentially hindering AI's potential to function at its highest level, Michael Attilio, M.D., chief medical information officer at Rome Health, said in an interview after the panel.
He cited continuation-of-care documents as an example of a standard format for sharing medical data. However, CCDs are limited in the amount of data they offer.
"AI thrives by having much more data than a CCD can provide," Attilio said. "That's why OpenAI now has a read-only connection with Epic. The more info it has, the more it can do."
Pramila Srinivasan, Ph.D., CEO of CharmHealth, previously told Health IT and EHR about how a lack of interoperability leads to providers missing key data from external systems on emergency visits, inpatient stays, procedures, medications and diagnostic results. Without being able to communicate with EHRs, AI models would similarly lack access to the necessary data to provide recommendations to clinicians.
Attilio described EHRs as "simply data that feeds the work that AI does." He'd like to see a switch from AI supporting EHRs to EHRs supporting AI.
However, getting AI companies and EHR vendors to work together on this issue is an ongoing challenge.
"It's a little bit tough now because organizations like Heidi have to figure out how to work with all these different companies that make all these different tools and all these different EHRs," Attilio said. "I think that someday we may start seeing the AI companies like Heidi develop their own records to try and create those data repositories that are going to help the AI do the best job the next time. So it's going to be interesting."
Brian T. Horowitz started covering health IT news in 2010 and the tech beat overall in 1996.
