Renton, Wash.-based Providence has developed a technology that deflects MyChart messages before they get to clinicians.
The 52-hospital system uses conversational artificial intelligence to answer patient inquiries and direct them to the right place before they even become formal messages.
"The in-basket piece of it, like basically understanding a message before it gets created and intercepting it appropriately, is not being done anywhere else," Providence Chief Strategy and Digital Officer Sara Vaezy told Becker's at Becker's 12th Annual CEO + CFO Roundtable on Nov. 11 in Chicago. "The reason for that is that it's actually really hard for outside companies to access those APIs [application programming interfaces] from Epic, for instance, to be able to intercept the message at all."
The so-called "conversational navigation platform" has 150,000 active users per month, with plans to grow that number to 1 million. Patients typically encounter it now through Grace, Providence's AI chatbot, but the health system plans to expand it to MyChart itself.
Providence pursued the technology after surveying clinicians about their top pain points. In-basket management was "like No. 1, 2 and 3," Ms. Vaezy said.
"There's lots of ways to tackle navigation, there's call center navigation, there's all sorts of stuff out there," she said. "But nobody could tackle the in-basket the way we could, because these are our patients, and this is our technical environment, and we know exactly what our patients are asking for, and we can navigate them to the proper workflows and do it in a way that doesn't overburden our clinicians, in a different way."
The "intent recognition engine" understands about 85-90% of prompts. The AI helps patients navigate MyChart, schedule appointments and manage their medications online.
So far, the platform has deflected 30% of administrative messages in MyChart. For messages that still need to go to clinicians, Providence also has an in-basket management tool, ProvARIA, that triages and drafts responses to patient communications.
The AI navigation technology evolved from a COVID-19 screening chatbot that could connect patients to virtual visits to, with the advent of large language models, what it is today.
"We can actually understand more of what people are telling us," Ms. Vaezy said. "Traditional, simple machine learning — if it was a multipart question or need — couldn't do it. If you put it in a way that was obtuse, it wouldn't understand. These LLMs can get pretty accurate. It's a pretty powerful tool."