Chima Ndumele: Reinventing Medicaid

Harlan Krumholz: Welcome to Health & Veritas. I’m Harlan Krumholz.

Howard Forman: And I’m Howie Forman. We’re physicians and professors at Yale University, we’re trying to get closer to the truth about health and healthcare. We are excited to welcome Professor Chima Ndumele today. But first, we always check in on current or hot topics in health and healthcare. So Harlan, what do you got?

Harlan Krumholz: Well, it’s another day, what I want to talk about? Artificial intelligence! Artificial intelligence, Howie. And I’m going to go back to the foundational model. So first, let me just break this down. I know people have heard me talk about this all the time, but I realize that there’s some people listening who still are not sure what the heck are these things. So again, these foundational models are a type of artificial intelligence that really are a base or a foundation for various applications. Think of it like a brain of a super-smart assistant that can understand, learn, help with a wide range of tasks and has been trained on the world’s knowledge. So we call them “foundational” because they can be used for this or that, and they have a lot of core capabilities. They can understand language. I tell you that the best technology doesn’t force us to conform to it by making us take classes and training us how to use it.

It’s ready to go, more like your phone, where it’s just intuitive. You’re not having to take class, that’s what this is. Plus it knows almost every language that’s out there. It’s got a whole range of skills besides text. It can understand and process images, videos in various types of data, and it continuously learns. So in the world that we’re in, the reason it’s so impressive is its speed and accuracy, its breadth of knowledge, and pretty much its consistency. We talk about things called latency, which is how long does it take between the time you ask it to do something and the time it gets back to you, and I think what really took my breath away when I started to see these in the beginning was the speed with which it was able to do this. Imagine you’re querying it, that’s going up in the cloud somewhere, it’s being processed and gets back to you in microseconds. It’s amazing.

So today I wanted to dive into some groundbreaking advancements that these new Med-Gemini models that are being developed by Google DeepMind. And they put out an abstract, an archive. Archive is one of sort of the very prominent preprint servers. So this means that they put up an academic article, and this one’s a whopper, it’s a huge deep detailed piece, before peer review. So they’re able to share with the community, while a medical journal and others are sort of getting through it and providing feedback and giving information. But this one has a who’s who, the people at Google on it, and Med-Gemini is a family of really highly capable multimodal models, meaning they can handle a whole bunch of different kinds of data.

They’re specializing in medicine. They build on, as you may have heard, this word Gemini. Gemini is the name of these AI models that are being put out by Google, and they are known for their general capabilities in handling various types of data, and Med-Gemini—so remember, we’re in this sort of arms race. There’s Anthropic, is a thing that puts out this one type of thing. Microsoft’s putting out ChatGPT, there’s a bunch of—but Google, of course, is a major player. Med-Gemini takes it a step forward by integrating web search capabilities and customizing novel medical modalities to make it a powerful tool for clinical reasoning and data analysis. And this thing they showed, for example, how it took the test, the USMLE test, which a lot of physicians are taking to demonstrate competency and scored a 91%, Howie. Our mothers would be proud of us if we got 91% on these exams.

But more than that, that it could handle text images, videos, and do things like the New England Journal of Medicine image challenges and medical video questioning. It could also handle a processing understanding of long and complex medical documents so that it could summarize and organize leading to enhanced diagnostic accuracy. It could handle complex data for better patient record analysis for maybe better patient care, supporting clinicians. Anyway, I’m just pointing people to this publication. Some people might want to take a look at it, but I’m just saying it’s heralding this new era. And then today I was reading something as you know, ChatGPT “o” came out, which is a marked advance over ChatGPT-4—it’s ChatGPT-4o, and in its ability to act as a chatbot and almost, again with no latency, be able to have these sort of conversations, much like the movie Her, you may remember.

And of course there’s this whole thing with Scarlett Johansson because OpenAI was using something close to her voice when she said, “Don’t use my voice.” She was the voice of the AI in that movie. And when that came out 10 years ago, people thought, “God, that’s just Crazyville. When is that ever going to happen?” But now, you could see it happening, and they’re announcing ChatGPT-5, which will come out sometime soon, is just going to blow away the current era. So I’m only just saying that these things that I’m watching around us both fill me with excitement and a little bit of trepidation because the world’s moving so quickly.

Yet when we go into the hospital, we’re seeing largely workflows that were built 20, 30, 40, 50 years ago, even though there’s still our computers everywhere. And I just have this feeling that we’re on this tipping point, and yet it’s going to be important that we mitigate any unintended adverse consequences and adopt these things wisely because there’s so much at stake. But Howie, you can hear my voice. I just remain so excited about the possibilities with these technologies.

Howard Forman: I agree, and I think that we have to be ready for the fact that patients, particularly educated patients, are going to come into the office now far more advanced than their understanding of their own condition than the physician will be, at least initially. And you’re going to have to meet the moment. If physicians aren’t able to meet the moment, then for a lot of conditions, there’s going to be a great deal of frustration because patients are not going to want to show up to the office having done their homework to find a physician who scratches their head and starts to look at their own computer. So we’ve got to be ready for a cultural change within the practice of medicine.

Harlan Krumholz: Well, I’ll make a last point here. This is going to really democratize things because all the patients going to have access to these tools. So imagine it takes six months to see a cardiologist for a routine visit, well, imagine if people could just sit down with a chatbot and get a lot of information. If we had these technologies, we could use this to triage, oh my goodness, this patient needs to be pushed ahead because this is what’s going on with them. And in our interview with a patient could then organize the data in ways that say, well, we can get this started while we’re waiting for you. It can really extend the workforce, and I think it could make a huge difference. And what about low-resourced areas in the world where they don’t have specialists? This could be game-changing. Okay, hey, let’s get onto our interview anyway, I just wanted to share.

Howard Forman: Dr. Chima Ndumele works at the Yale School of Public Health as a professor of public health and health policy. He’s also appointed to the Institute for Social and Policy Studies at Yale. His research focus lies in equitable and accessible healthcare for populations that have been historically underserved. He specifically looks at the impact of healthcare safety nets on outcomes and how local policy and health insurance changes impact Medicaid and the care received by people dealing with chronic health issues.

In addition to his teaching positions at Yale, Dr. Ndumele also sits on the board of the Connecticut Advisory Board for Transparency on Medicaid Costs and Quality. He’s also one of the first graduating members of the Aetna Academy of Health Diversity Scholars Program. Dr. Ndumele completed a master’s of public health at Tufts and received his PhD at Brown and has been here with, I think, for over 10 years right now. And so I want to start off by, I love to hear what motivated you to get involved in health policy writ large and then specifically Medicaid. And I wonder if you could just start off and tell us where’d you grow up and what got you interested in this?

Chima Ndumele: Yeah, so my parents came over to America from Nigeria in the mid-’70s, and by the mid-’80s, they owned a series of home health agency staffing businesses. A function of that, I suggest, you might think that I’m somewhat predisposed to think about healthcare. It turns out that in that business, many people don’t know this, but one of the things, home health agencies and some of the specialized care for the elderly is actually cared for and paid for by Medicaid. So my mother would often complain about the reimbursement for Medicaid, the late payment for Medicaid and the like.

And I remember thinking to myself growing up, why would anyone ever take this insurance? So over time, as I got more interested and more knowledgeable about health and started thinking about meaningful ways to address health equity, I naturally gravitated towards thinking about the Medicaid system, a place where we could do more than document health inequities but think more carefully about how to actually solve them.

Harlan Krumholz: Chima, I’ve been admiring your work for so long. You are one of the world’s experts in medical insurance and particularly around Medicaid. And I wanted to just reflect with you about the enormity of this issue about Medicaid. I think it became clear to me when I saw—now this is a while back, but it said a third of the states spend more than a third of their budgets on Medicaid. And then when I reflect in Connecticut, our state of about, I don’t know, 3 million, almost a third of the population, almost a third, are being covered by Medicaid.

Yet I see an absence of real innovative thinking. I mean, that doesn’t mean there aren’t talented people who every day are looking at Medicaid, but I sort of think this is our chance to really redesign. It’s a vulnerable population in great need. There’s a lot at stake. I had the benefit of talking to the governor this week because he was at a Yale event, and I said, “You know, we’ve got to double down on Medicaid and pointing to you as really one of the thought leaders in the nation, if you were in a position where you could do something that would transform Medicaid, what would you do?”

Chima Ndumele: Yeah, let me underscore some of the points that you made first, because I think they’re really important. On average, states are spending about 30% of their budgets on Medicaid, and that has increased dramatically over the last two decades, crowding out spending on transportation and public health preparedness and education and all types of things that are arguably—quite frankly, not even arguably—more important for health and more important for the thriving of the population than healthcare spending and Medicaid itself. We’ve got to get it under control.

There are a lot of things to do. I think the first thing that we should do is stop looking for our keys under the lamp. The reality is that most people who are on Medicaid aren’t just on Medicaid. Forty-five percent of individuals are on Medicaid, are on at least two other assistance programs. Ninety percent of individuals who are on SNAP are also on Medicaid. The reality is that people don’t live in the silos that we put them in for the ease of administration. We need to start thinking about whole people. We need to start thinking about the whole safety net. We just finished some work now, and I’ll give you one of the top lines of numbers here. People with diabetes are 91% more likely to get kicked off of SNAP than people with no chronic conditions. That just doesn’t make any sense.

And that’s not because anyone’s trying to be mean, or the reality is that we haven’t linked that data to think carefully enough about how our policies are affecting people’s whole lives. So there are many places to go within Medicaid, but first we have to start thinking about Medicaid recipients as people rather than just numbers on the spreadsheet.

Howard Forman: And just for our listeners, SNAP is, I think, Supplemental Nutritional Assistance Program. It’s a way of providing food or payments for food for poor individuals, right?

Chima Ndumele: Yes, that’s right.

Howard Forman: So two areas that you’ve worked on are of interest to me. I think everything you’ve worked on is of interest to me, but two areas in particular. One, you’ve worked with Betsy Bradley and Lauren Taylor about thinking about social determinants of health and how they interact, and you’ve sort of addressed that a little bit already. And the second is, you’ve done some work on a Medicaid managed care, which does not look the same as Medicare Advantage or managed Medicare.

Medicaid is particularly hard to study because every state is different. And so you have a state like North Carolina that’s making massive investments in social determinants of health, at least as a pilot project. And you have other states that really just use it as a payment scheme. Can you give us some ideas about what innovations look most promising to you right now, and do you have any hints that you might offer for the state of Connecticut or any other state about where they might go?

Chima Ndumele: Yeah, this is a big focus of my energy these days. The most charitable version of the story about Medicaid that we tell ourselves is that they serve as, that states are supposed to serve as laboratories for each other. States are supposed to innovate. That innovation is supposed to diffuse through the nation and then supposed to scale. The reality, however, is that most states don’t do a tremendously good job of creating credible evidence. And even if they did create credible evidence, there really is no mechanism for things to scale.

So part of what we’re doing, Harlan, you mentioned the governor here of Connecticut. Part of what the state of Connecticut is allowing us to do is to set up a lab to think more carefully about how to generate credible evidence within Medicaid running RCTs [randomized controlled trials]. We don’t know how to do this yet. There are several, and that ranges from the big things, like should we expand Medicaid in southern states? To the small things. How do we run the long-term care program in ways that are most efficient? Who needs to be institutionalized? Who can get care through home and community-based services?

So I think as a macro point here, we have to do a better job of creating evidence that is useful for people, whether they are in Oregon, New Jersey, or Indiana. But in terms of the things that we think are going to work, this push towards social determinants and smart investment in social determinants I think has real potential for the future. But let me push a little further because that’s, I think, in two ways. There’s both the question of do we give money to these programs? But there’s also the question of how we structure and design the programs themselves. Should we give out food stamps once a month or twice a month? How generous should we make the ancillary programs? All of these questions have huge implications for health and the way that we deliver services. And quite frankly, for a very long time, our approach to that has been to shoot first and to ask questions later, and I think we can do better.

Harlan Krumholz: One concern I have about the way these programs are structured is the payment, going back to your mother and what she was complaining about, and it creates this sort of terribly corrosive system, which means that people almost want to avoid taking care of people with Medicaid because the way that the pricing structure and the reimbursement works, how are we going to turn that around?

Because the states aren’t eager to raise the reimbursement rates. And of course, if it’s going to be capitated coverage, if they’re going to cover populations, then they can create incentives. But this is something that’s just always bothered me. You’d like to be able to say we value everyone the same, but when we’re going to put people under insurance coverage where hospitals and health systems are actually seeking to avoid providing services because they lose money on those individuals when they provide services, we’ve got a big problem.

Chima Ndumele: Harlan. It’s really difficult to think about how to craft a safety net that’s fair to everyone, but state-sanctioned disparities aren’t the answer, right?

Harlan Krumholz: That’s exactly right. That’s exactly right.

Chima Ndumele: Exactly right. On average, providers or doctors get about 70 cents on the dollar for a Medicaid patient relative to what they would get for the same service for a Medicare patient. Everybody understands how business and numbers work. Of course they’re going to avoid Medicaid patients. On average, you get about 60 to 70% of doctors annually, let’s say, that they’re willing to accept new Medicaid patients. If you start off the system with a third of providers that aren’t willing to accept your care, you’re in trouble. By the way, it’s worth noting that these are the people who often need care the most. Medicaid is, by definition, defined by disadvantage. These are the people that we should be trying our hardest to get into care and accessing all types of care in as timely a fashion as possible. Harlan and I did a study maybe two years ago now, and I’ll give you the quick upshot from it.

It essentially says that if you were to take a census of the people that were on the Medicaid program now and to look a decade in the future, about 53% of them would still be there a decade later. This is of a 90-million-person program. We have people that are permanently engaged in this insurance that is clearly not optimal for them. Now, is Medicaid better than having nothing? Absolutely. But quite frankly, we need to do one of two things.

If we want to turn the Medicaid program into a true safety net, we need to surround people with the full array of services that it can actually help them get off. Or we need to engage with the fact that Medicaid is going to be a kind of long-term program for a really large group of the population. And at that point, we need to invest in it in the way that we invest in all other long-term programs and the way that we invest in Medicare and the way that we invest in commercial insurance. We really have a striking opportunity to shape midlife morbidity, to shape outcomes for Medicaid populations in a way that we haven’t before. And it feels like we’re missing it.

Howard Forman: Is there appropriate funding to do the work that you mentioned earlier, being able to actually look state by state, either to compare states or look within states about the effects of programs? Is the funding available do that type of work? Are we under-investing in the health services research around Medicaid that would give us the answers we need?

Chima Ndumele: It’s a good question. I don’t think the challenge is necessarily the funding, at least on the research side. I think the challenge is, as you mentioned, every state Medicaid program is a little different, therefore the quality of the data is a little different. The availability of the data is a little different. It is just really in the last five years or so that federally we got data that we could trust from all states. So I suspect that we’re going to see an influx of new Medicaid researchers. I suspect we’re going to see an influx of new insights that really do drive our conversations moving forward. But traditionally, it just hasn’t been there. The data hasn’t been there to really make the right kind of insights to drive the right policy.

Howard Forman: And just to follow up on that, did the imperative to get better data occur because legislative change or regulatory change occurred? How does that even happen? The program is coming up on 60 years. How does this happen?

Chima Ndumele: Yeah, it happens when the federal government mandates it. And there is this open question about how we deal with federalism. How do we give the states the autonomy to make decisions, to innovate and to do things that matter for their populations and priorities while making sure that we’re not doing things that are harming people, both within states and then federally? So I think the federal government over the last half decade or so has started taking this job a little bit more seriously in articulating the need for us to get credible evidence, both within and across states, and then facilitating the data sources and the funding to help researchers do so.

Harlan Krumholz: I know you’ve been watching with interest what’s been going on in North Carolina, for example, where there are these broad-based investments. We’ve talked about it on the program. Is it possible for these programs that are using such large amounts of resources from the states to start investing more broadly in things like housing and actually even economic development, recognizing that these investments may ultimately reduce the amount of healthcare utilization that’s needed?

Because to me, this is what this gets back to, your original comment about seeing the whole person. If we can conceptualize this, not as just paying for services but actually promoting population health and then trying to net-net, bring these healthcare costs under control. I mean, you take asthma, for example, which could be precipitated by environmental factors within certain parts of a city because the exposures are different than they might be in other places and so forth. Climate change; heat’s different in different parts of the city. Vulnerable populations tend to be a couple degrees warmer because they lack foliage. That’s going to in some way mitigate the climate. Do you see nationally this kind of discussion taking place where you might be hopeful that these kind of investments can take place or people just still splitting everything?

Chima Ndumele: So yeah, I do think that there are several states now—in some ways there are the states that you would expect—but there are several states that are starting to think about how to invest in these social determinants of health in a more thoughtful way. I think what we know is that social determinants of health matter. I don’t think we know how to invest in them. I don’t think we know who should be the people investing in them. Is it the hospitals? Should it be the plans? Should it be community health centers? I think we are still very much in a testing phase to figure out how to get from knowing something matters to figuring out exactly how we should do something about it. But that being said, we are light-years ahead of where we were 10 years ago in terms of, again, as we noted, recognizing the whole person and having states starting to invest in them.

I think “food is medicine” has taken over across the country. I believe there are more than 10 states with current waivers to the federal government to allow them to pay for food. I think there’s similar movement on the housing front. I think we should be excited about the fact that we are moving forward and recognizing that Medicaid shouldn’t just be an offshoot of the Medicare program, that these are individuals with real specific and different needs. As a matter of fact, if the Medicaid program looks just like the Medicare program, there’s a good chance we failed, right? The Medicaid program has people with very different needs, and thus we need to structure it in very different ways.

Howard Forman: We could go on with a lot more questions, and I have more to ask you, but I think we’re going to wrap up here. I’d love to have you back. I still want to hear your thoughts on the answer to the question for your parents about why do people even take Medicaid, which you touched on but more. It is fantastic to have you here. We are so lucky to have you. I literally remember the day that Paul Cleary told me that we had gotten you here and how excited he was, and it’s just an honor and privilege to have you.

Harlan Krumholz: Let’s have his parents on next time.

Howard Forman: That’s right. Talk about home health. That’s true.

Chima Ndumele: Yeah, that’ll be fun.

Howard Forman: Thank you very much for joining us today.

Harlan Krumholz: It’s great to see you. Great to see you.

Chima Ndumele: Thank you so much for having me, guys.

Harlan Krumholz: Hey, that was a terrific interview. I so enjoy him. But now I’m getting to the part that I really love on the podcast, which is to hear what’s on your mind this week, Howie. So what have you got?

Howard Forman: I appreciate that, Harlan. This is not the first time we’ve even talked about the issue of race in medical decision-making and how it can lead to varying patient management. But we’ve not talked about lung function measurements, which by the way is a very common tool that pulmonologists and even internists may use in the office. And these measurements, when factored into equations that are race-based, can have enormous effects on patients. And to have that enormous effect based on the color of your skin or the continent of your origin should be concerning to people. And in a recent study published in The New England Journal of Medicine, authors from the U.S. and UK looked at the effects of proposed and mostly instituted changes in how we judge lung function in patients using 2012 equation that included race as a factor. And then through a 2022 update that was race-neutral.

And first, they confirmed what we sort of expect at this point, that by removing race from the equation, more Black patients were diagnosed with new or more severe lung disease than with the prior measure, while white patients were less likely to be so diagnosed. So it changes for both, in a way. Disability payments for veterans would go up by $1 billion per year for Black individuals and down by $500 million a year for white individuals. When looking at longer-term outcomes, these changes appear to have very little effect. So doing this reclassification did not change outcomes so much, but it did change the process of care and who was allowed to have care. Many of the changes may also be seen as negative for Black individuals. So while you’re more likely to move up the waiting list for a lung transplant, you become less likely to have surgery for lung cancer.

And while you receive extra payments for veterans’ disability, you are now less likely to be a firefighter. So this is not to say that these changes are not medically appropriate and just, merely to say that there are winners and losers in this reassignment. So pulmonologists and thoracic surgeons are not changing these measures in order to right some societal wrong. They changed these measures because the earlier measures were based on the false idea that race is coded biologically. There is no doubt that our race or ethnicity or ancestral origins do increase or decrease the likelihood of certain genes, most of which are not even fully understood. And we may yet adjust formulas such as these for what those genes portend. But we should be careful to not bias ourselves to think that one group of individuals is just prone to having lower lung or lower renal function and essentially dooming them to worse care or treatment by society due to organized medicine’s desire to explain differences purely on the basis of race.

Harlan Krumholz: You know, Howie, it’s a challenging area for a couple reasons. Race certainly is a social construct. There are some genes that, like you said, track with your ancestry, and so there’s something there. But in general, it’s dominant social influences that really make a difference in terms of outcomes and risk. I struggled with this around readmission, which is a different issue, but one in which people said to us that, well, Black people have a higher risk of readmission. So these risk models for hospitals should take into account the race of the individual, saying that if you care for a lot of Black patients, then you should be expected to have a higher readmission rate because you have Black people in your hospital. And I was arguing that—

Howard Forman: You’re doing it wrong.

Harlan Krumholz: Why should I say my readmission rate should expect it to be high? Shouldn’t I be interrupting that social problem rather than accommodating?

Howard Forman: And just for our listeners, Harlan, you took a lot of flack for that, a huge amount of criticism for that. And it was a principled viewpoint that I think is being held up at this point.

Harlan Krumholz: It is. I will say, there’s a different thing on the payment side. If it turns out that hospitals, that one of the reasons is because Black people end up going to lower-resourced hospitals who have worse performance. Then if you’ve got a payment plan that actually holds back payments to those hospitals, you could make the problem worse. So there’s a whole thing about that, but it’s separate from trying to hide the issue about this social construct being a driver of outcomes. Anyway, I’m glad you brought it up today, it’s a really important topic, and we should continue talking about it. But the thing that’s most reprehensible is, again, when we were saying things like we think Black people have thicker skin, so they need higher radiation.

Howard Forman: Yeah, it’s the same thing.

Harlan Krumholz: We need to increase the radiation for the X rays—

Howard Forman: We’re basically saying, we’re basically saying, we expect you to have worse lungs, so you’re really not so bad for a Black person. It’s just horrible when you say it out loud.

Harlan Krumholz: It’s astonishing to see that these things still are existing within our system. And it’s good to see that people are addressing it.

You’ve been listening to Health & Veritas with Harlan Krumholz and Howie Forman.

Howard Forman: So how’d we do? To give us your feedback or to keep the conversation going, email us at health.veritas@yale.edu or follow us on LinkedIn, Threads, Twitter, or wherever you can find us on social media.

Harlan Krumholz: And we really want to hear your feedback, questions, any experiences with these topics. Rate us, that helps other people find us. It also teaches us something, whether you have something constructive to sell, so we can do better. Or if you like us, of course, we always like to hear that. And we appreciate anything our listeners do to give us feedback.

Howard Forman: We really do. And if you have questions about the MBA for Executives program at the Yale School of Management, reach out via email for more information or check out our website at som.yale.edu/emba. And a final reminder to our listeners. Next week, on May 30th, we’ll be doing a live podcast at the Yale Innovation Summit. Again, May 30th. The summit itself runs from the evening of the 28th until, I believe, late in the day on the 30th.

Harlan Krumholz: And all gaffes are going to be included.

Howard Forman: Exactly. And we have a great guest lined up for that. Links are in the show notes today. Come to the summit and see us interview some of the greatest health and technology innovators.

Harlan Krumholz: Yeah, Josh Gebelle’s put together a really great event, and—

Howard Forman: Apparently over two thousand people are signed up now.

Harlan Krumholz: Look, New Haven is the next “it place” for—

Howard Forman: I think that’s true.

Harlan Krumholz: …as a hub for innovation. Health & Veritas is produced with Yale School of Management, Yale School of Public Health, thanks to our researchers, Ines Gilles and Sophia Stumpf, and to our producer, Miranda Shafer. Terrific people, we just love them.

Howard Forman: Yeah, they’re awesome.

Harlan Krumholz: Talk to you soon, Howie.

Howard Forman: Thanks very much, Harlan. I’ll talk to you soon.