Lumeris’ David Carmouche, M.D., on the Potential for Agentic AI in Primary Care

The career of David Carmouche, M.D., has included leadership roles at Ochsner Health, Blue Cross Blue Shield of Louisiana, and Walmart Health. Last fall he took a new position as executive vice president and chief clinical transformation officer at value-based care company Lumeris. In an interview with Healthcare Innovation, Carmouche spoke about his new role and the potential for Lumeris’ new AI-based solution to optimize patient outcomes while lightening the workload and improving the satisfaction of physicians.

Carmouche is responsible for accelerating physician and patient adoption of Lumeris’ clinical solutions across its network of provider partners, which is comprised of more than 11,000 physicians. 

Healthcare Innovation: After your roles with organizations like Ochsner and Walmart Health, what drew you to Lumeris? 

Carmouche: I knew that the next wave of innovation and transformation in healthcare was going to be a technology-based one, because Open AI had already burst on the scene. It didn't take long to understand what the potential of that was going to be in healthcare. And I knew enough about Lumeris’ data environment to imagine that Lumeris would be a potential player in leveraging large language models.

I asked them to use the title chief clinical transformation officer, and it's really to be the clinical mind in the company — to think about how we deploy these novel technologies to support providers and the patients they serve.

HCI: There are other companies such as Aledade and Privia working with primary care practices on value-based care transformation. What are some things that are unique about Lumeris’ approach? 

Carmouche: I don't know that there are other organizations that largely or exclusively partner with health systems. A lot of those companies work with small or independent providers who need a lot of support. Our history has been to partner with health systems. Those are much more complicated partnerships. Those are very political, layered, multi-stakeholder organizations, but they've aggregated half of the primary care capacity in the country. A large proportion of Americans get a substantial portion of their healthcare in health systems. So I thought if you're going to materially impact healthcare delivering the country, you can't work around the edges of that. You have to actually go where that is. And it’s something that Lumeris has learned how to do effectively over a decade.

There are other differentiators with the business model. One is that we own a Medicare Advantage plan, so we understand health plans — specifically in the Medicare space — and have used that effectively as a learning lab. 

The other thing is that we have bundled our services from a business model standpoint into performance on value-based care contracts. So we get paid when the health system succeeds. We lose money when the health system fails. Some companies have fixed fees or act as consultants. Some take risk, and we're firmly in the aligned risk space, which I think is unique. So basically, put your money where your mouth is. It also creates the notion of Lumis being an operating company, not just a consultant, and I think that changes how you show up.

HCI: I understand that Lumeris just introduced something called Tom, which is described as an AI-powered primary care-as-a-service solution. What does the phrase “primary care-as-a-service solution” mean to you?

Carmouche: First of all, it means that there are lots of services that can be delivered in primary care, some of which are delivered today inefficiently, and that could be done more efficiently with automation. There are some services that should happen but that just don't happen because the care team doesn’t have the time or capacity to do that. Primary care as a service is using a data foundation to power multiple AI agents that can take on those services and do that in an organized way based on the knowledge of what an individual patient needs next on their health journey. 

So if you have this data infrastructure that is inclusive of the electronic health record, potentially claims, data sets, maybe consumer data items that we've purchased, social determinants of health insights, etc., we can determine, in many cases, the next best thing that should happen for someone to improve their health, and in some cases, that can involve an automated outreach to a patient. It may be educational, it may be a reminder, it may be a nudge, it may be a prompt. We're not taking humans out of the loop. Humans are still required to make diagnoses. We may suggest information or summarize information for a physician that might inform a diagnosis, but ultimately, a clinician is doing that. A clinician still is required to prescribe a medication, but there's a lot that happens around that that we think AI can help with as a 365 24/7 always-on partner for a primary care doctor.

HCI: This was named in honor of Dr. Tom Doerr, who was a Lumeris founder. Did you ever get a chance to meet him?

Carmouche: I didn't. Sadly, he died five years ago, suddenly and unexpectedly. However, I feel like I know him because his widow allowed us to synthesize his voice. She gave us access to some recordings and voice clips, so the initial demos we've used with customers are in his voice. I have read his writings. From all accounts, he was an outstanding clinician, but he was also very analytical, and he really believed that ultimately, it was going to require technology for primary care to deliver the highest value that it could. He wrote about that over a decade ago. So I feel like I have a sense of what he was thinking about.

HCI: Does the Tom tool have to be custom-integrated into the EHR of the provider system? Doesn’t Lumeris deal with folks on lots of different EHRs?

Carmouche: Yes, I think it's like 119 different EHRs we work with now. We don't have deep integrations with 119 EHRs, but with the big ones like Epic, etc., we do. I believe that the more we can integrate Tom into the EHR workflow, the more likely clinicians are to adopt it. There are parts of what Tom will do that don't need to get integrated into the EHR. There could be tasks and alerts to other care team members that probably could come into a different application in the office. But the idea that Tom could be interacting with patients outside of the view of the doctor is a new dynamic, right? It happens today with humans— for instance, a care manager might call a patient and interact with them, and somewhere in the EHR, there may be a tab where that is documented. Early on with AI, to build trust, I think we're going to have to surface to the clinician summaries of what Tom has done. I think they're going to want to see that. Whether that's deeply embedded in the EHR somewhere, or whether it's a desktop application that kind of slides over into view and just has a brief summary, I think each organization will have a different view of it. We're prepared, where necessary, to integrate to the degree that the customer will want it or demand it.

HCI: Is Lumeris also looking at buying or developing ambient AI to help generate clinician notes. 

Carmouche: We think of Tom as a platform so we won't build everything. There’ll be other best-in-class applications that Tom can leverage. So, for example, you mentioned this notion of ambient voice scribe, which is kind of the current wave. First of all, that is really good for Tom. Because until that happened, I think if you ask most clinicians whether technology has made healthcare better or worse for them, most would think of their EHR, and largely they think that has not made their life better. But the ambient voice scribe technology has opened clinicians’ eyes to the notion that some technology actually can solve problems and make their life better. 

So the fact that that that wave of implementation has been under way for a couple of years makes it easier to have a conversation around Tom. Also, I would imagine in a world where ambient listening technology is available, and the conversation between the physician and the patients is being transcribed in real time, that becomes a substrate for Tom to do some analysis, right? So Tom might be reading that transcription, and may offer an input to the clinician that might be useful. 

HCI: Have you piloted this with a health system yet?

Carmouche: We are piloting it. Some of our current customers are supported, at least in their value-based care lives, by Tom already, because the technology stack that powers Tom is the technology stack that we've perfected over the last decade. So the underpinning of Tom existed before conversational AI was available through LLMs. We were still automating things like text outreach or e-mails or EMR notifications. Applications using deep conversational AI are the ones that are now being piloted with partners.

HCI: Once your health systems partners have used Tom for a while, It will be interesting to get their impressions of how it's impacting their workflow and patient engagement.

Carmouche: We are ecstatic to learn with them and to report back on that. This is moving so fast. I can tell you that the things we are experimenting with today are dramatically better than three months ago. The pace of improvement and underlying technologies — I've never experienced anything like it. We won't get everything right on the first shot, but with the way that technology is moving and with the partners’ interest in this particular problem, this is going to be a fun problem to work on together, and we look forward to updating you at some point in the future. Frankly, I'd love our partners to tell the story.