‘Make the right thing the easy thing’ — A vision for AI to improve primary care

If patients are the most-discussed issue in health care, artificial intelligence (AI) may be a close second. Could the rapidly advancing technology be the key to sustaining primary care at a time when it’s most needed by patients — and a time when it’s in dire need of physicians?

David Carmouche, MD, is an internal medicine physician who has gained experience across health care technology, management and payment models. In Louisiana, he worked with Ochsner Health and Blue Cross Blue Shield. He served as Walmart’s senior vice president of healthcare delivery, leading that company’s Walmart Health centers.

Carmouche now is executive vice president and chief medical and commercial officer for Lumeris, a health care technology company. At the end of March, the U.S. Department of Health and Human Services announced he would be one of five physicians on the new 18-member Healthcare Advisory Committee, a panel to lead improvements in patient care and modernization of the nation’s health care system. Recommendations are not binding, but the group is expected to have the ears HHS Secretary Robert F. Kennedy, Jr., and Medicare Administrator Mehmet Oz, MD, MBA.

Carmouche spoke with Medical Economics to discuss the latest developments, including what is OK and what is not, for physicians learning about AI technology.

This transcript has been edited for length and clarity.

Medical Economics: How would you describe the current state of AI and its best uses for primary care?

David Carmouche, MD: To answer that, or at least understand my answer to that, maybe 30 seconds of my perspective, and that is that we're kind of at a tipping point in this country where the needs for primary care are greater than they've ever been. With the demographics of the aging population, the burden of chronic disease, there's never been a greater need for primary care, and it's at a time where the workforce for primary care is disappearing. If you think about it, over the next decade, there’s going to be 85,000 to 90,000 too few primary care physicians. And if the population needs more primary care, something has got to give. And historically the way we've solved for that is with team-based care, maybe adding nontraditional or nonphysician roles to the care team, nurse practitioner, PA, care coordinator, social worker. All of those human resources are expensive and they don't scale very well, and today, there's 100 million adult Americans that don't seem to have access to longitudinal primary care. So there's a problem that it's hard to see the solution being a human capital solution only.

It's within that context that I think the way that AI and technology should be supportive. First, to extend that human workforce to allow them, to enable them to take care of more patients. Frankly, if you look at panel sizes of physicians in primary care over the last decade, they've shrunk. Largely they've shrunk because we've added so many tasks to the primary care portfolio that they need time to do that, and so that it comes at the expense of access. And so we need to change that so that humans can take care of more.

And then the second is, we just need to make it easier to do a really good job of being a primary care physician. That means removing administrative burden, but it also means removing cognitive load and things that consume energy of the provider that could be made easier with technology. That's kind of how I see it. Across all of my jobs, the thing that's always kind of stayed in mind as a design principle: When it comes to physician-invasive change, whether that's through technology or otherwise, is, just make the right thing the easy thing. So make it easier to be a primary care physician and use technology in that way. So, not to replace the really important human component to primary care, but leverage that technology to make that job actually easier than it is today.

Medical Economics: You have experience with value-based care, and your company also works with technology in the shift from fee-for-service payment models to value-based models of health care. At the national level, could you talk about Medicare's current approach to value-based care?

David Carmouche, MD: I think at the end of the day, Medicare probably has two main flavors of value-based care. They have lots of CMMI (Centers for Medicare & Medicaid Innovation) programs, but the two main ones have been the accountable care organization (ACO) movements, the MSSP (Medicare Shared Savings Program) or ACO REACH, now ACO LEAD (Long Term Enhanced ACO Design), and the second has been Medicare Advantage, or the commercial privatization of Medicare. In those environments, Medicare Advantage providers frequently are looking to create value-based care payment economics for their network. So it's a little bit more of an indirect way of getting at it. In both cases, the goal is to get primary care doctors to be accountable for both economic outcomes and clinical outcomes and I think that's a laudable goal. It is the right way to align the needs of clinicians and patients and purchasers, whether that's the federal government or an employer. The concept of value-based care is a great one, I’ve spent a lot of my career in It. The challenge with it has largely been how these programs have been structured. And in some ways the focus or success has been through things like coding and documentation or risk adjustment or curating a network using actuaries to kind of say, if we took these providers out and left these in, the performance of the ACO would be greater. Those are things that are not really getting at the heart of making people live healthier lives and reducing the total cost of care.

The challenge for value-based care and the reason I think value-based care hasn't become the prevalent and predominant model, is that, number one, there's a cost to participate in it. There's a complexity, there's an investment in technology and informatics and analytics that has to be overcome through the economics of the program. And then secondly, most doctors would say, I'm being focused on things that aren't really the most important things. If I can have time to focus on helping my patient stay healthy and stay out of the hospital and stay out of the emergency department and take their medications and control their chronic diseases — those are the things that I want to focus on. But I'm being, I'm being managed to care gaps and AWVs (annual wellness visits) and risk and codes. We have the right idea, and I think the execution has largely just missed the mark a bit. If you would ask most primary care physicians, they'd say that's their challenge with value-based care.

Medical Economics: How do you anticipate AI programs will affect value-based care for primary care physicians and patients in coming months and years?

David Carmouche, MD: I'm very bullish about the role that AI can play. If you think about it, if you're fortunate to have an outstanding primary care physician, you probably spend two to maybe four, 20- to 30-minute periods of time with them every year — and that's if you have a really good one who's accessible to you. And yet, primary care is the net result of how you live your life every day: the foods you choose to eat, whether you choose to exercise, the habits you have as it relates to alcohol or smoking or vaping or other activities, whether or not you fill your medications, whether or not you monitor your own blood pressure or your blood sugar. All of these things are really important to your health. And for most of us, our primary care physicians, even the best ones, really their support for us stops the minute we walk out of their door, and we're kind of left to live our own lives between visits, and then periodically we come in for check-ins. Those may involve laboratory data, vital signs, and then some time with the clinician. That's a construct of a world where the delivery mechanism is largely a human one and the constraint, therefore, is how many minutes do the humans who are conducting primary care have available to them? And so the number of touches per patient, the number of patients, are limited by that human constraint.

AI offers the opportunity to totally unlock that. The way we've thought about it is that if AI starts with an understanding of your own data as a patient — so if the EHR data is available to AI lab, pharmacy data, HIE (health information exchange) data, claims data, potentially if that's available, consumer data. If that data is curated to give a digital view of you as an individual, and then that data and where you are in your health care journey is used as the starting point for AI to interact with you on behalf of your clinician, that's an amazing advance. It allows your clinician to oversee a care model that follows you outside of the exam room and is monitoring and sensing and supporting you. And then, frankly, alerting the clinician when you're off track, so that maybe instead of waiting a random six months to come back in, maybe you are someone who needs to come back in four months because you're going off track. On the other hand, maybe you're someone who's doing great — you're fine, like, you're compliant with your meds, your weight’s good, your blood pressures are great. And you wouldn't necessarily need to randomly come into my office just to sit down for me to confirm that, because I would already know that.

So I think it starts to unleash this notion of health care being reactive and episodic and siloed, to being continuous and connected and proactive. And I think I think in their physicians get to spend their human time on the patients who need it the most. And patients get the value of the support of their clinician, wherever they are, at home, at work, on vacation, and it's just part of their life. I think that's the promise of this technology, if used wisely in primary care.

Medical Economics: We know technology, over time, tends to become more reasonably priced. Things can be expensive, both from a software and hardware perspective. Can you talk about cost for the physicians who may be using some of the programs? We know that smartphones are ubiquitous in our nation, of course, but there may be some other technology that's needed at the patient end. Where does that come from? Who pays for it?

David Carmouche, MD: Great questions. Let's start at the patient level. I mean, I'm wearing an Oura ring. You may have an Apple Watch. I mean, people have devices. Those are not cheap, you know. And so there is a little bit of have and have-nots today from a technology standpoint. I do think that technology companies like ours need to build flexible architecture so that patients, over time, can bring their own devices to bear. I don't think that any product can be built where it mandates for you to be a user of the product, you have to buy a specific technology. But the reality of it is, there's very rich data in wearables or remote monitoring. The challenge generally has been that providers don't have the bandwidth or time to sift through the data and make good use of it. So it's kind of an interesting phenomenon and if you're a driven patient — you know, I'm very interested in my sleep. So Oura provides a feedback loop to me every morning on how I slept that night. And there's embedded within the application tips and tricks for better sleep. I don't need a clinician. I can self-manage myself on that journey. And so it's a consumer tool, I choose whether it was important, I was willing to pay for it, and therefore I get value out of it.

But for things like hypertension or diabetes or heart failure, where there's a monitoring capability that, frankly, to create value, needs to be connected to a provider who is able to prescribe medications frequently as part of the management — that loop is of interest to me, and I think there's two problems to solve. One is on the provider side: You've got to make that information very easy for me to digest and to consider in my treatment. So AI is a great tool for analyzing large data sets, normalizing it, summarizing it, and presenting it to a clinician. So that's there on the device side, AI allows for kind of high- or low-tech versions of remote monitoring. High-tech would be connected device, maybe Bluetooth enabled, or cellular device that ports information directly into an EHR (electronic health record), and so there's a data integration activity, and that's expensive to do. The device may have an expense to it, and that's unclear who should fund that and how that should get paid for.

There's a low-tech version of it, though, which is that you can buy an over-the-counter Omron blood pressure cuff for $39 and very high quality. And AI can just literally ask you to read me your last 10 blood pressures. There's no deep integration, it's just literally AI communicating with you through text or voice to say, give me your blood pressures, like, over the last week, can you just read them all to me? Thank you, and I store them in memory and at the time summarize them and aggregate them for clinicians.

So it's a complicated question. I think over time, devices will get less expensive on average. I think product builders will build flexible architectures that will kind of be, bring your own device, and we'll figure out how to take inputs in. The most important thing is, how do we connect those devices to a clinician because digital health tools, there's many of them out there. Most of them have failed to improve health outcomes because they're not closed loop back to a clinician who has the ability to prescribe or intensify medical therapy when your reading suggests that that's needed.

Medical Economics: With any new technology, there are early adopters, there's always skeptics, and primary care physicians are some of the busiest ones out there, maybe they just haven't had time to really make a deep dive into AI yet. What advice would you give to a doctor who still is just dipping their toe in the pool, so to speak?

David Carmouche, MD: I think that's OK. Health care has benefited from slow adoption at times. I mean, there's a safety reality. This is a new technology, and to the degree we turn it loose on either clinicians or patients, we should be thoughtful and careful, and it's OK that this is going to come in phases. I think early adopters are necessary to test and help validate these tools. And I think laggards — it sounds so negative, you know — sometimes just pragmatic and are just needing to see that evidence before they invest too much time. So I think that dynamic is OK.

What I don't think is OK is to be a clinician in today's world and to just bury your head in the sand around AI and pretend that this is just going to be another failed promise technology that's going to make my world better. Doctors, even the most skeptical ones, would benefit from reading about AI, reading in the medical literature, what they can. It's OK to be skeptical, but I think being skeptical without information and without understanding and learning is challenging. It's not hard to learn about AI. You can read medical journals and start to see now lots around AI. You could also go deep. You could go to Microsoft or Google or AWS and go online and have free access to AI primers and start to understand that. I think many physicians are using an AI tool, Open Evidence, and so they're starting to be familiar. Many are being ambient listening technology that gets them away from being on the keyboard in the exam room and starts to transcribe them. So I think, I think clinicians are seeing AI show up in their world, hopefully in ways that are positive.

It's a different thing when we start placing AI in between a physician and a (patient), that we need to be thoughtful and careful and use good judgment and require proof points. So, be interested, be curious. Avail yourself of an opportunity to learn more, even if you draw a line between learning and implementation and practice. As physicians, we're meant to be lifelong learners. This is a really exciting time in health care, and we should all, as physicians, shape how this technology is going to be deployed.