How healthcare leaders are thinking about using AI

I spent Tuesday at the Modern Healthcare Digital Innovation Summit. It was a chance to learn from health system, payer, and digital health leaders in the middle of figuring out how to use AI to improve the US healthcare system.

Sharing some of the common themes of the day:

1 - Tech Innovation v. Ecosystem transformation.?

A large part of the conversation was figuring out how transformational AI will be. Even if it lives up to its technological potential, it will be deployed within a certain business environment.

As Microsoft Health Senior VP Dr. James Weinstein said, “The system produces the results it’s designed to produce.”

We’ve had the technology, for instance, to break down data silos between different health systems for some time. Yet sharing data between health systems (especially those with different EMRs) remains challenging because the business case to build those connections isn’t there.

No doubt AI will make many process improvements by lowering the cost of retrieving information. But there will be an upper limit within the existing incentive structure.

2 - AI is only as good as the data it is supported by

Sara Vaezy , Chief Strategy and Digital Officer at Providence, shared some incredible work they’ve done helping physicians manage patient messages. They receive 700K patient messages to clinicians a month, across 25K physicians (not all of whom are patient facing, so those messages are further concentrated among a smaller subset of physicians).

Before a patient message goes through to a physician, an LLM reads the message to understand the intent of the message. It then responds with a possible resolution to the message. If the resolution isn’t adequate, the patient can push that message on. This has reduced incoming messages to providers by 23%.

For the messages that are pushed through, they then bring together all of the relevant information that a physician would need to respond to that specific message to accelerate the response. The goal is to reduce physician “pajama” time and they’ve seen tremendous anecdotal support from physicians.

Sara emphasized that both the pre-message resolution and providing the context for physicians to respond only work if the LLM is connected to the right data sources. It needs to be connected not just to the patient’s medical record, but past patient interactions, past patient questions, the scheduling system, the billing system, etc.

That, truly, is the work that allows Providence to take full advantage of the LLMs. To get this level of connectivity they had to build their own system; the in-house EMR wasn’t sufficient.

Note - this is the kind of work that we at 3Pillar do. We can build new LLM tools, but you need the foundational work done first. You can learn more here: https://www.3pillarglobal.com/offerings/data-ai/enterprise-generative-ai/

3 - Getting through the technology complexity paradox.

Technology of course can simplify processes, but it also enables greater complexity, as Glen Tullman , CEO of Transcarent, pointed out.

Take medical billing. There are 600,000 CPT codes. Each employer self-insured plan has its own set of benefit rules. Plan design is increasingly complex, with 150 page reports.

This system would not exist without technology. And that wouldn’t necessarily be a good thing - with all this added support we can ideally create better, more personalized insurance products.

The challenge is when that complexity isn’t simplified down for the member, as most of us have experienced personally. There are many designers trying to build workflows, but no product workflow will account for / be able to explain the 150 pages of medical policy.

Optimizing this is where AI can plan such a key role. Understanding the intent of what a person wants to know, synthesizing all the data quickly and invisibly, and then providing a direct, clear answer.

4 - Smaller takeaways

Admin before Clinical for AI

A common theme articulated by Rohit Chandra , Chief Digital Officer at Cleveland Clinic, was that providers are cautiously approaching how to incorporate AI into clinical pathways. Ambient listening and clinical notes documentation is clearly where people are moving fastest on the clinical side.

However, they are pushing as hard as possible to incorporate AI on the admin side. Explaining benefits, streamlining admin work flows, more efficiently staffing clinicians — health systems can’t move fast enough here and that’s where the immediate opportunities are.

Bringing relevant insurance info to the point of care

Peter Serio , Chief Customer Officer at Florida Blue, is laying the blueprint for how payers can improve the member experience. Similar to Providence, they are focused on ensuring that patients get the same answer regardless of which channel a member interacts with - mail, phone, web, app, etc. That means connecting underlying data systems, allowing AI to pull intelligence from there.

Pete and team are also working to bring plans’ rules to the point of care to inform care plan design. For example, currently a physician will set a care plan with a patient and then submit a prior auth. It takes days to manually review that prior auth, and if it is denied, time has been wasted for the patient in addition to creating frustration. Instead, they are working to push the rules / logic engine to the point of care so that a physician can see if a given service will be covered and share that with the patient in real time as they plan for next steps.

Set expectations around AI chatbots upfront.

Alya Sulaiman , a partner at McDermott Will & Emery LLP, stressed the importance of leveling with patients / members going into a chatbot experience. It’s okay that the chatbot can’t solve every problem. Be upfront about what data it has access to and the type of problems it can solve well. Create the escalation path to a human for areas it can’t solve something.

Hospital at Home, powered by Best Buy

Deborah Di Sanzo closed out the day with an exciting talk about how Best Buy is enabling the shift to Hospital at Home. The work they’re doing with Geisenger, for instance, was nicely captured in this video that she shared.?As someone who spent 5 nights at a hospital for monitoring with my youngest when he had RSV this winter, I can't wait.

Last thoughts

The conversation left me bullish on the work 3Pillar does with our clients. The leaders at this conference don’t need help identifying and evaluating how to improve care. They and their orgs need help doing the work – designing and building new workflows, updating applications, and creating a data architecture that pulls out the right insights.