Permanente Federation CIO Sees AI Impacting Care System Navigation

A recent conference hosted by the Centers for Medicare & Medicaid Services focused on reducing administrative burden on clinicians. Panelist Brian Hoberman, M.D., CIO and executive vice president at the Permanente Federation, said one of the next steps up from the EHR era is going to be using AI to help patients navigate the care delivery system. 

The Permanente Federation  is the national leadership and consulting organization for the eight Permanente Medical Groups (PMGs), which, together with the Kaiser Foundation Health Plans and Kaiser Foundation Hospitals, comprise Kaiser Permanente. As CIO for the Federation, Hoberman has accountability for IT that includes operational performance, technology integration, and innovation, according to his online bio. His leadership role involves collaboration with national, regional, and medical center leaders and staff throughout all Kaiser Permanente entities, which includes technology strategy, governance, and investments, as well as development and design, support for clinical content, risk management, security, training, communications, support, analytics, and vendor relations.

Hoberman mentioned that he started working in Kaiser Permanente in San Francisco before they had electronic medical records. He led the rollout of EHRs in KP’s hospitals in Northern California, and has been involved in the EHR evolution ever since. 

“We went from a world that I started off in which was paper records, which is the ultimate in information fragmentation, and we put all that information in a single medical record,” he began. “Kaiser Permanente is a very big, integrated organization, so we had these large integrated medical records that, of course, interacted enormously with other organizations, some of which at that time were still on paper. As time passed, almost everybody had an electronic medical record, and we developed interoperability, which was also promoted by CMS. But it didn't solve the problem, because navigating all that information, even if it's digital, is very challenging. When I think about burden reduction, I think about navigation, and when I think about AI, I think about having a tool that can process this monumental amount of information and lead to the best possible navigation. 

Some of this is just summarizing what's in the record, he said. Some of it is just great convenience for patients such as translation. As an aside, Hoberman noted that there are still a lot of regulations around translation that require humans in the loop. “That's a burden. Certainly human needs to be in the loop for many things, but do they need to be in the loop for translating discharge instructions?”

Health systems end up with all this information they can use to risk-stratify the situation the patient finds themselves in, Hoberman noted. “Are there ways of using computing to try to simplify that from the perspective of the patient who has a new complaint or has a complex illness? Or the provider who works within an organization that has 350,000 employees?” No individual is going to know everything about all the resources of the organization, about what the very next step might be, he added.  “I think that the next level up, and it is a gigantic level up from the EMR era, which was itself a huge level up from the paper era, is going to be using AI to help navigate within care delivery itself.”

He asked the audience to imagine a world where a delivery system is offering a conversational AI triage. “As a doctor, one of the questions that I get the most from my friends and neighbors is: such and such just happened to so and so; should I take them to the ER? This doesn’t exist today, but imagine that a Kaiser Permanente member goes into our AI, where we know who they are, because we have access to their electronic records. We know the difference between a healthy person and a person who's got chronic disease burden. We also know who had surgery last week, who says, ‘I'm having knee pain.’ Now, in today's world, we handle that in a variety of ways, but in a world where the AI is curated by the delivery system, and it knows the delivery system’s capabilities, and it knows you, could we decrease some of the touches that today really are just extra steps for the patient and get them straight to where they need to be definitively faster? We think the answer to that is yes.


“Do we think we can trust it? We think we can,” Hoberman added. “It’s going to require a delivery system to do that more than a tech company, because the delivery system has access to information that the tech company isn't necessarily going to have, even if the patients have put all kinds of information about themselves into the tech company’s platform. Just as people trust their doctors today, we want the AI to be an extension of who those doctors are in the future — as long as it's deeply curated by docs and the delivery system trusts it by constantly scrutinizing it, making sure that it's doing what we think it's going to do. That also requires a super-tight connection between the informatics side, the operations side, and the platform side of the delivery system, where we really understand how the technology is evolving, because small changes can have a substantial impact on care delivery.”

Implementing ambient scribes

The Permanente Medical Group began working with ambient technologies early on when they were just coming on the scene and were not yet integrated with the EHRs. The positive reaction from clinicians was immediate. “Over the years of introducing new technologies and enhancements, the greatest praise we ever got was no complaints,” Hoberman said. “Hardly anybody ever applauded what we did, even though they basically all agree that they can't live without it. When we put in ambient everywhere, we were getting pats on the back like we finally did something that gave them time back.”

But he recalled a question coming back from the finance-minded people in the organization: Can you see more patients now that you get the time back? “Fortunately, the answer that we ultimately all agreed on was no, let's use the time saved to do better patient care, because nobody's going to just leave early,” he said. “They are all committed to trying to do the right thing all the time. Take that burden off their back and watch what happens. And in fact, what happened was higher-level patient satisfaction, lower-level provider burnout, and generally higher quality care, because we didn't have an exhausted, burned-out workforce. This was something that was really enabled through the application of AI.”

Building trust

Hoberman also touched on the unresolved question of how you build trust in AI. “When we put in ambient, we launched three different teams who looked at it in three different ways, and we monitored every single transcript,” he recalled. “We got feedback from the docs. We did this at the level of individual specialties because we didn't think that we could necessarily say if it works for family medicine, it's going to work for pediatric neurology. We gave a massive amount of feedback to the vendor, and it evolved.”

The other thing they did was put in place ongoing monitoring. If the vendor changes their model and sometimes very tiny changes, it can lead to significant performance changes. “We provide that feedback to the vendor right away,” he said. 

Another component of it was understanding the way the workforce and patients feel about AI. “What we learned pretty quickly was that most patients were pretty trusting about ambient, because, for one thing, their doc said they wanted to use it, but for another thing, because of the principle of the human in the loop. They knew that this was something that was going to help the doctor do what the doctor needed to do, and the doctor was literally with them while this was being done, so that that helped with the patients.”

Hoberman closed by saying he’s excited by the fact that in the next 12 months Kaiser Permanente is going to be doing a lot of research and development on how AI can enhance patient and provider delivery system navigation.