Getting ready for the AI era of virtual healthcare

Create vetting processes and guardrails

With AI evolving at a rapid pace, healthcare provider organizations must have internal guidance in place to ensure safe, accurate and responsible AI use in virtual care settings.

Madelyn Knowles, senior research manager at Rock Health Advisory, underscored the importance of governance when integrating AI into virtual care.

"Governance has certainly been a part of the conversation for some time now," she said. "So, organizations managing the liability of AI integration, especially when we think about some of those clinical decision support or more clinical applications, as opposed to the lower stakes admin type work."

Managing liability may include approaches like human-in-the-loop, where a human user oversees AI's outputs, as well as triggers to alert staff when a model might have drifted or is malfunctioning.

Joshua Jones, M.D., medical director-telepsychiatry at Providence Virtual Care and Digital Health, echoed Knowles, adding that AI is not yet ready "for primetime as a diagnostic tool." Thus, having a human in the loop, especially for AI integrated into telemental healthcare, is vital.

"You still have to have a very skilled clinician on the other end interpreting this stuff," he said. "You can't do away with that. I think the other thing that we, as the human operators on the end of it, have to be vigilant about is that we can't get lazy. We can't abrogate this responsibility to the chatbot."

In addition to individual vigilance around AI use, healthcare organizations must establish systemwide governance frameworks to ensure a standardized approach to AI adoption.

Raj Patel, M.D., vice president of digital patient experience at Mass General Brigham, shared that an AI steering committee can be immensely helpful in ensuring standardized AI governance across an organization.

At Mass General, the steering committee comprises senior leaders who provide centralized oversight for AI, including policy development related to model review, monitoring and risk assessments.

Establish patient and provider trust in AI

Although AI tools can enhance patient and clinician experience, trust is vital to their adoption and utilization in virtual healthcare.

Caroline Goldzweig, M.D., chief medical officer of Cedars-Sinai Medical Network, noted that AI tools must be integrated into the patient-provider relationship, rather than be used to replace it.

"I personally believe that there's something important about the human connection, and I don't think you would want to erode that," she said. "You want patients to trust in their physicians and their healthcare organization…So, one of the big conundrums is going to be, how do the patient, the AI and the doctor, or other clinician, work together? That's going to be a brand-new experience."

Trust becomes cemented when patients and providers are confident that the AI tool will work as intended. Thus, Goldzweig emphasized the importance of vetting and verifying the accuracy of health AI.

Additionally, transparency is crucial for gaining clinician trust in AI.

"As a clinician, you have to have an understanding of why a recommendation is being presented to you, why this decision support [tool] is telling you something," Goldzweig said. "I think that helps a lot."

Providence's Jones agreed, noting that there needs to be transparency regarding how data will be used to power AI and the data privacy protections in place.

"I think you have to have that dialed in before you can really open up the sandbox of how you're going to use [AI tools] with PHI [protected health information]," he said.

Incorporate change management into provider training for AI

Encouraging AI adoption within virtual care programs rests on more than trust. Effective and ongoing training is crucial to this endeavor.

According to Michael Dalton, CEO of virtual-first provider Ovatient, change management needs to be incorporated into AI training programs.

"It's the same thing that we've seen from a virtual care perspective -- just because you went to medical school or you were trained as an MD, it doesn't mean that then you somehow can just instantly provide phenomenal virtual care," he said. "Same thing when it comes to AI. So, I think there's a level of training that is going to be needed, a level of reset."

Understanding their concerns and pain points regarding AI implementation and effectively addressing those during training is crucial for gaining clinician buy-in.

Buy-in is not only important for encouraging AI use, but leaders also need clinician feedback to determine if a tool is worth its cost.

"[Health system leaders] need to stay close to their providers," Dalton said. "They need to be able to utilize their feedback, but also then have the hard conversation with them about, okay, so where and how is this creating an ROI?"

Develop a vendor selection strategy

With the flood of AI tools on the market, virtual care leaders must be strategic about which tools they select and vendors they choose to partner with.

Rock Health's Knowles recommended that leaders "really get under the hood" of any solutions they are considering implementing.

Leaders need to pay attention not only to the clinical outcomes of the solution, but also to the measurements that vendors are capturing to demonstrate validated efficacy and impact, she said.

Mass General's Patel noted that leaders should have a strategy in place for identifying vendors who fit within the organization's governance frameworks. These vendors should be ready to provide performance and risk of bias data in a standardized format to support explainability.

"I think what we're trying to do broadly, AI or not, is choose partners that complement our core platform, don't duplicate functionality and provide a broader set of capabilities with one, two, three, four, five, 10 vendors, rather than having kind of a fragmented ecosystem of 30, 40 vendors," he said.

Implement data processes & IT infrastructure that support AI

As AI-enabled virtual care gains momentum, healthcare organizations must ensure that their data processes and IT infrastructure can effectively support these services.

According to Mass General's Patel, accessible and usable data is essential for AI integration into virtual care. The data that fuels AI needs to be relatively rich and built for AI ingestion and interpretation, he said.

Not only that, but organizations must break down their data silos, bringing together disparate clinical, operational and cost data, to achieve successful AI integration.

"What you need to have available is a kind of lake, a data lake, with an integration layer above it in order to enable these technologies," he said.

Fortunately for providers, vendors have, by and large, embraced AI, developing solutions that already include the necessary data integration and infrastructure.

Oleg Bestsennyy, partner with the healthcare practice at McKinsey, pointed out that, similar to how virtual visits became integrated into EHRs, the IT and data infrastructure to support AI agents and AI scribes in the virtual care setting are starting to be integrated into existing care platforms.

"I think we're on the precipice of such things happening in the next year or two," he said. "And then [AI-based virtual care tools] will probably be adopted in the same way that the current adoption of things like AI scribes or ambient listening scribes is happening, which is that companies will just basically put them as a layer on top of the existing workflows that exist in EMRs."

With the rapid advancement of AI, it is inevitable that virtual care, as we know it, will undergo significant changes. Healthcare organizations that meet the moment with carefully considered governance, workflow and IT frameworks will find it easier to successfully integrate the technology into their virtual care settings.

Anuja Vaidya has covered the healthcare industry since 2012. She currently covers the virtual healthcare landscape, including telehealth, remote patient monitoring and digital therapeutics.