January 23, 2025: Aneesh Chopra, Chief Strategy Officer at Arcadia, tackles the promises and pitfalls of data-driven healthcare transformation. How do economic incentives shape the use of technology in healthcare? With the adoption of AI and large language models, are we nearing a future where personalized, evidence-based care becomes the norm—or are we still bogged down by regulatory and cultural barriers? Aneesh explores the "good, bad, and ugly" of data-driven healthcare, the impact of AI on clinical decision-making, and the ethics of transparency in patient communication.
Key Points:
Donate: Alex’s Lemonade Stand: Foundation for Childhood Cancer
This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.
[:(Intro) we're very much in the spirit of let's rising tide lift all industry boats. Not like my vendor machine will beat your vendor machine. But we need the country to succeed.
My name is Bill Russell. I'm a former CIO for a 16 hospital system and creator of This Week Health, where we are dedicated to transforming healthcare one connection at a time. Our keynote show is designed to share conference level value with you every week.
Now, let's jump right into the episode.
bably goes back to what? Like:nd it was, and consumer led, [:Yeah. And so that that's what we connected on.
And I was like, look, Data can transform healthcare. And you're like yes, it can. And we're going back and forth and we're creating a cheering crowd together. But here we are. We're a decade later. A little over a decade later. I want to touch base. I guess I came in from outside the industry.
This has moved a lot slower than I could have even imagined. I thought, oh my gosh, we have all the tools, the technologies there. And, you And quite frankly, even over the last two years, we had the advent of a whole new set of tools. Like we should be jamming at this point. Are we jamming or is this just the pace things move?
anila folders and maybe some [:And we'll talk about the FHIR API. So the good is a very clear win in terms of, you can't have data driven without data. The bad is we're not as data driven as we should be, which is that to the extent that we have emphasized over the last decade Supply side regulation. There's a little bit of a if you build it will they come mentality.
Did app developers come and make use of the data that we made public? Or did we find ourselves focused on more of a compliance? Did we have to do it? What did we do? That question, Bill, gets a little bit beyond . technology has a lot more to do with economics and incentives. And so if I'm in fee for service medicine, it's an incentive.
's a It doesn't give you the [:I can't really blame, The medical community for not using quote unquote the data because it was designed for the supply side to demonstrate they can produce it. It wasn't really optimized for the consumption and use. Now I assume the market would have solved for that. I'll go back to
this is going to be too strong of a statement, but I'm disappointed in the consumer. The consumer didn't show up to this game. Like, we expected them to go, Yes, I'm going to get my data, but just didn't show.
Yet, I want to come back to the yet part because ONC just did a pretty powerful report on patients with cancer.
al to extract the data. But, [:Call the doctor's office. What does this mean? What do I do? How do we deal with it? And so the market of applications that would naturally help guide me through it, that hasn't quite happened. And I think the ugly, to get, metaphorical or metaphysical or philosophical, the use cases that work, Bill, are the use cases that have a clear economic return, risk adjustment.
y incentive is to accurately [:But relative to the under diagnosed fee for service system that doesn't even know what you have, doesn't record your ongoing clinical conditions, it looks like you see this sort of up coding dynamic. And MedPAC has called this a 70 billion dollar, gap in terms of the use of data. Where there's economic incentive in MA relative to the lack of the use of data in fee for service.
And I look at that ugly. Some would look at that and say that's a distortion in the market. It's an economic highest and best use of the data today. And what I think will happen in the not too distant future, especially with the advent of LLMs in our pocket, we may see a leveling effect where there'll be more regular and routine use.
djustment and HEDIS. If it's [:Should you do bariatric surgery consult first? Should you try nutrition counseling? Should you do GLPs with nutrition counseling? What combination of interventions can help someone with obesity manage overall health? That kind of data driven learning health system, I see it. I see it. I want to bring it to life, Bill.
I'm going to dedicate a good chunk of my time to that future. But at the moment, it hasn't lived up to the hype, to your point. Good, bad, and ugly.
Good, bad, and ugly. It's interesting to me, as we look at the, we now have a new set of tools. And I remember back in the day, one of the things that we heard was, what's the patient going to do with the data?
advisory board. I said, all [:I'm going to respond to you. And I started responding to him and I'm like, okay, how close is it? He goes, that's that's really good. That's really good. I know, I understand the limitations of it, and I understand it can hallucinate and whatnot, but we're closer now than ever that I can take my data, put it into a tool like that, and start interrogating it and say, what does this cancer diagnosis mean for me?
And literally in my inbox this morning on the day we're taping, a brand new paper pre print, the title I'll read, Superhuman Performance of a Large Language Model on the Reasoning Tasks of a Physician. This is the O1 model. You look at the possibilities. And you see where this movie is going.
and we're getting to be this [:Man, talk about reducing the variation in clinical care delivery. If you look at every one of the papers in LLMs, what gives me so much excitement at this moment in time is they lift up the bottom performers, right? So the variation is hey, I've seen a lot of these. If your ER doc is amongst the best who's triaged it, like Halamka, who we both love, is like the global expert on mushroom related illness, right?
formers sometimes mistakenly [:So while there's a lifting at the bottom, that's a clear win. There's a worry about the super awesome might downgrade to just like default, kas sunstein nudge style default.
I read that document. Do you really think John Halamka is gonna read it and go? Yeah I'm gonna agree with the LLA.
No, he's not,
no. I, and he's super human, so we love John. But in terms of the broad statement, if you're busy and you're trying to get through your day, and there's a decision support tool that's helping you make a differential diagnosis, isn't it human nature to accept and move on and get to the next assignment?
I don't know. It's a challenge. This is where this whole cultural change, liability, LLM liable if it nudged me down the wrong path, or am I ultimately responsible for double checking all the math and the work? And if so, where's the productivity gain? It's fascinating.
Talking to CIOs, there's a [:One is we talk about transparency, these LLM models and that whole concept that you just rattled off Hey, I'm going to check its work. We're not going to check its work. We're moving too fast in the care system to check its work. What does transparency actually mean in a large language model?
And do we have the resources to really check those things? I know Dukes. Spending a lot of time and some others are spending a lot of time on this. And there's value in it because, we're talking about LLMs. There's a whole. another set of tools that are AI related tools that we do want to know what the math is.
We saw this with sepsis and some other models that came out that were not vetted. One of the things that's interesting, I'd love to hear your thoughts on this is we took a path when you and I first met, I was taking a path of, we've got to build these tools. We've got to build this stuff.
s. And recently we had an AI [:And I was asking them, about the implementation of AI. And I said first of all Epic just rolled out a hundred tools. I don't have to know AI at all. I just have to configure it and then fine tune it and then roll it out. And I said how prevalent is that? I said I got it from Epic.
I got it from my imaging provider. I get it from ServiceNow. I got, they're all dropping. AI into their tools and they're like I don't know if I'm going to have to build this out. I think those players are going to just bring it to me and say, configure it.
Yeah. You're raising an extraordinarily important question, which is there will be a marketplace of capability.
engineering we have not yet [:There's a lot of room. You may not be a builder in the sense that you have to physically go and train a model, but you're also not going to just swipe the credit card and configure. There's a lot in between. And let's just take the example of the New York Times story. of I think the headline was something like, that message from your doctor may have been written by AI.
And it was the story of the, I think Epic has the inbox assistant tool, very ubiquitous at this point, allowed doctors to save time, pajama time, responding to messages by providing an initial draft, right? Pretty safe. It's a draft, it's physician facing. UC San Diego, who signed on to a group that I've helped organize called HealthcareAICommitments.
ht have had script problems. [:Smith. So they chose to watermark, and the article basically was like kudos to UC San Diego for having watermarked. Another signatory UNC Health was also named in the article and they chose not to watermark. They did a governance process and their judgment was we don't need to watermark.
Because there's a human in the loop. So it actually is human feedback.
It's it's the functional equivalent of I used a macro in Excel. Do I have to say I used a macro in developing this spreadsheet? Ah, I don't know. That may be a bit far. My point is. It's not a configuration technical thing.
ysician users, patients, and [:I don't have the exact survey quote in front of me, but something to that effect. There's a lot of work. And even if you ask the question, there's a lot of debate right now there's a scaling law, throw more compute, more NVIDIA chips, and we just get more and more brilliant models.
There's some perception that we're hitting a sort of a scaling limit, and that now it looks like there's other attributes, the algorithms and the software that make all this work, like a reasoning engine. The role of agents to take actions on the systems, where you could have the same model that hasn't really been updated, but you can do all these additional things to make it productive.
t just flipping switches and [:There's a lot of product management requires thinking, focus, time, and energy. And I don't know if that's the CIO who's going to end up leading that effort. I don't know if it's a chief AI officer or chief medical officer or someone who's doing care delivery reform. Man, but someone's gonna have to think about this in a more holistic way from ethics, reliability, trust, a whole range of subjects.
this year in:Test for making it possible. [:
I'm on record as saying I believe that Judy Faulkner should have been the CIO of the year for the last 20 years. because she has done most of the work for CIOs to make them look good, to roll out new capabilities, and make sure that the EHR implementation went well.
think too many CIOs are sitting on that. they're going to take care of me, they're going to make sure. Yes, what you describe happens at just about every health system. A group gets together and says, all right, we need to form governance around this. We need to identify what models we are taking forward.
How are we going to engage with our community? Are we gonna do a watermark? Are we not gonna do a watermark? And yeah, Longhurst and the stuff they've done at UCSD that's great stuff. And in Stanford and Peffer they follow suit and did the watermark.
what was the but I could see [:Oh Longhurst did publish that paper.
The button has two choices Start over or edit. There's no send.
Oh, interesting.
So Longhurst and I'm sure others have as well, but I love the transparency of Longhurst. He shows you what he does. By the way, the community we've organized shares that in a portal so that the health systems can work with each other and the health plans.
So there's no best practices yet, Bill. We're all learning, but there's an open commitment to collaborate to say, we tried this. Is this reasonable? It's a little bit. You're trusting to say, how do you do this? Something as simple as an AI inventory, use case inventory, you would think.
That's what they were
ople were saying, look, man, [:And I'm like, I need the library. And so that question, HHS just published its AI library. The VA did the same.
That'd be a fun project, Bill. What are the Indian Health Service AI use cases? And what's the VA? And what's CMS? And let's just consolidate into a master inventory of use cases. And then people can say, okay, I want to run deeper here and try this there, and then ask questions about, which one's going to get us the biggest productivity bang for the buck.
We desperately need productivity reform. This is the one sector of the economy where we added all this digitization. Back to your data driven comment, your first question, productivity went down in healthcare. In every other industry, you add technology, productivity goes up. Man, some combination of culture, economic incentives, complexity of the market, productivity went down with digitization.
That was a bummer.
an interesting four years. I [:Yeah. And one of the things you impressed upon me is that this whole idea of freeing the data and making the data an asset that can work for healthcare is a bipartisan view of how this was going. I remember having that conversation.
Yes.
Do you still? This is.
Absolutely. Okay.
In fact, I am in this area. I'm excited. Let's just say we're excited for three reasons, right? Excitement number one, I think we're all excited about Doge. Okay. Let's be clear. Anybody that wants to bring change if you believe there's opportunity to weed out some historical regulations that don't seem to make sense you want to champion that we're going to have a focused effort with presidential support to dig into this.
Now, there are a lot of cynics, eh, what's this going to do? How's it going to help? I'm hopeful we're going to see some clearing out the underbrush.
My, my [:Yeah, you're
just like, what?
So these sorts of things some of that is comedic, but there's also productivity barriers, right? There's a lot of things we can't do because of X, Y, and Z. And we need to have FAQs, guidance, like, why don't we clean this stuff out? By the way, we had this as an Obama administration priority too. I partnered with Cass Sunstein, who was our head of regulations, back to your economics degree, Office of Information and Regulatory Affairs.
Cass wrote the book on how to do this in a more cost effective way, so we launched it. And we had backing and support, but it wasn't at the scale of the way the Doge is being structured. It's like front and center. Anyway, so category one, I'm actually excited about Doge, number one. Number two, for whatever it's worth, RFK Jr.
or the other, we're going to [:And so I'm excited about a transparency agenda. But again, that was day one Obama. Day one Obama is I want a more open and participatory and collaborative government. Day one! My job was created and introduced to the world in a day one memo on open government, right? I believe RFK Jr. will bring that philosophy on steroids into HHS.
not just for the employers, [:There may be a window here where this team goes a little bit harder on economic models to bring down the cost curve. It may be more consumer price transparency. It may be more capitated. Risk programs, it may be embrace MA as the engine to bring some of that capability to market. Maybe different than what a traditional democratic administration would have prioritized, but I think you're going to see more emphasis on the economic model to drive value in the system.
d by the Medicare Innovation [:That I'm giving you my best case here.
Best case. I, there's concern, it's let's just take Medicaid redetermination and all that stuff. Let's take you live in Virginia, so there's the if certain things don't get funded, I think it's written into the law that essentially couple hundred thousand people or even more than that are just going to drop off the Medicaid rolls and that's going to be a burden to health care providers because they're still going to come in and we're still going to care for them.
so there is what I'm hearing from the CIO ranks and the meetings is concerned that the reimbursement levels are going to go significantly lower. We're already at one or two percent, operating. And if you go significantly lower, you pull more people off of those roles, that's going to tip the thing upside down.
let's just take a minute and [:count on you to be the positive one. You seem to be able to find those things.
If you don't have something to look forward to, what do you do with your day? Oh man, this sucks.
No, we move forward. And we work towards President Obama said very clearly that Washington fights within the 40 yard lines. It seems like, oh, the world's coming to an end and wow, the world's going to be amazing overnight. That's me. We're here, okay? The ACA is a Republican idea from Massachusetts, right?
system said, hey, I got this [:And he said, you know what? Let me first find out how many of them actually qualified for our own charity care policy and how many qualified for Medicaid. And of the engagements that they, again, email campaign, they didn't just communicate with a few, they communicated to everybody. And for, I don't know, let's just say it's a third engaged.
Of the third that engaged, half qualified for Medicaid, and we're just sitting there on the bad debt rolls. Half. No change in law, just we missed it administratively. Half qualified for charity care. If we have a charity care policy, wouldn't we want to administer it? We have rules, let's make sure that we have.
rds in the ACA called Section:How do we share data, either B2B or agency to agency, which is one view. Or we came up with an architecture that said, no, it should all be B to C to B. If I apply for tax credits, I should be able to keep with me all those certified data elements of my last year's tax stubs. my immigration stubs and so forth.
Get me all the proof and then I should be able to bring that with me vetted so that I can apply for other things. If I walked into the Average Hospital's Patient Financial Counseling Service, that technology works today for insurance brokers selling ACA plans, web brokers. , is there a single health patient financial counselor at a health system that can do that kind of level of automation?
Why?
Yeah, for those who are listening on the podcast, I shook my head no. Why doesn't it exist? I the answer I'm going to give is so lame and it's just going to be busy. There's so many priorities. There's so many things to work on.
But you just [:And I've just explained to you the power is in the hands of the CIO to work with financial services to say, wait a minute. I want every single person screened for Medicaid. Not just those that have got, if you do the math on it, Bill, hospitals spend so much money after the fact hiring lawyers and teams to go.
I remember in Virginia, a decade, 15 years ago, when I was in state government, I had a state hospital tell me it could be a thousand dollars per patient to hire someone to go chase them after discharge, get them to fill out the paperwork, get them enrolled, and it's still at a positive ROI because it would move from bad debt to insured.
s house to fill out forms? A [:CMS, four years later, has still not issued the final rule because, quote, the technology doesn't work. Are you kidding me, Bill? We can't generate a good faith, Jiffy Lube gets me a good faith estimate, and we, woe is me, it's so hard to get a good faith estimate in the US. Give me a break. Focus. Is it a priority?
o be a balanced budget act of:So we have the plumbing and the infrastructure so we can make our way through. And I'm bullish that between Doge and the RFK, there's going to be a receptive audience on ways we can better collaborate in this sort of thing.
I want to talk about the United Healthcare thing, but before we get there, it's interesting you bring up
billing transparency in some organizations that I've dealt with did do the work and there is a tool or it shows up in their portal and it's here's the estimate. And I'm like, okay they did it. And then others, just, if I remember the bare minimum you had to do was to take a spreadsheet or something.
same thing. Because one gets [:And so there's no way for the average person to do that, but other health systems did do that work. And I think because they saw it as, either part of their mission or part of their service to the community. I don't, I'm not sure they saw it as an economic driver of any kind.
It, in my view, good faith estimates, at least for scheduled services, should reduce consumer related bad debt.
Yeah, I can see that. You'll
know in advance, and you'll have an understanding of how to make it work. If you ask the average CIO today, or the medical revenue cycle, a lot of that bad debt are from insured patients, where the consumer exposure is high.
Yeah, UnitedHealthcare CEO. it is tragic.
And definitely not what we want to drive change, but do you think this will have a significant impact on anything? Do you think, Health systems and providers will start to look at this and say, okay, we've got to do something different. I don't know.
Know Brian Thompson [:We were going to solve prior authorization automation anyway. Congress is on the verge of passing bills. We've got the prior auth rule. We got AI infused, interpretations of evidence based guidelines. It is already on the path of improvement by reducing friction, not by turning it off. The sadness that this guy is celebrated in parts of our country, that just tells you we've got an underlying cultural problem, a portion of our neighbors think this is somehow worthy of celebrating.
am saddened that was [:I wear the hat around that says This Week Health and so people will stop me and ask me the question and I'll be like, there's a challenge here and it's a significant challenge and it's the opaqueness of this thing and then all of a sudden it hits you.
Oh, my claim has been denied or whatever and all of a sudden you're now one of the many and it's still the leading cause of bankruptcy is medical you Bankruptcy, medical debt bankruptcy in the United States. And so it's getting to the point where we all know somebody who has been impacted in that way.
And we're nervous that could be us because we're not really sure how it works. And I was talking to a payer executive, actually a former CEO of one of the Blue Crosses. And he said, the reality is we have a communication problem because our policies clearly state.
This, and this, [:Put it this way.
Can I delegate it to an app of my choice?
That's a good question.
Under terms of service, it's an unclear question. Is it really? Yeah, Medicare Advantage plans are required to be transparent on the rules. But when you go to the website of where those rules are stored, you'll see language that says effectively a human can read this, but no third party app developer can use it or incorporate it into their products.
was, and presumably would be [:Hey, the doctor just gave you a good faith estimate. Hit the spell check, make sure given the dynamics of the plan you have and the economics and all the rest,
describing my number one use case for large language models. I throw information at it and I, are you allowed to?
Let me give you the case. My use case was there's an AI tool that you can essentially make a deep fake of yourself. For good reason, you would do that, right? And so I'm looking at this thing. I'm going, man. Before I give up my name, image, and likeness, I want to make sure that I'm not signing away the keys to the kingdom.
r words they want it to say, [:That's the use case. I what is the limitation? Why are they not letting us do that?
First of all, we have to ask ourselves basic questions. Why are these regulations rules paywalled. And if they're paywalled, how do we make it accessible to those who are allowed to use them in a manner that's productivity generating?
So I don't think it takes a lot of work to get people to sit around the room to say, look, what we want is adherence to evidence based medicine. What we don't want is waste, fraud, and abuse. Check. What we have are EHRs that speak a certain language, and I have claims data coming to me through the Provider Access API 1.
dence based criteria to say, [:our orders and guidelines and next steps, then would that not move our whole industry towards evidence based that can lower costs and improve outcomes? And to me, that's the ideal use of AI. It's to generate evidence in the real world. So write the next rule in real time, make it computable and accessible to anybody that has a longitudinal health record so you can understand where you are against it, and then evaluate all the clinical alternatives if there's a problem.
, and prior authorization on [:If we have a common view of that's computational I think we can make a lot of progress.
This is a little bit of a jump here, but there was the big dust up, over the past year, where Epic actually shut off access to certain systems who were using data in a way that was deemed not in the best interest of the health system.
And one of them, I think one of them was called like IntegraTort or something like
that. Oh, that, man, I'm grateful for Epic for calling out BS on that. That is wrong. Yeah. So just, Bill, let's remind our listeners here. What did I say when you asked me about data driven? Good, bad, and ugly. In the ugly bucket, the highest economic use of data is risk adjustment in the Medicare Advantage program.
bution to risk adjustment in [:Hey, Bill, you're my physician neighbor. I'm your physician neighbor. In the spirit of goodwill, we're going to make it freely available so that if you see a patient, I'll see a patient, and we're all going to collectively benefit. So we just want the system to work better. Not a really economic model. It's really a neighborly principle.
patients that might have the [:When Epic, called them out on this behavior. Hallelujah, Epic. Thank you. Because the ugly part of data sharing is if it's only available through this one pipe of my neighborly trust, but it has huge economic value and I'm venture backed, I might feel a lot of pressure to squint my eyes to say I think of this as legal, but it's not.
And they did have a few use cases named that IntegraTorque made it laughable, but there were also, effectively, payers looking to source data for purposes of risk adjustment. I'm in value based care. We need it for value based care. Is that treatment? Is that operations? What are the economic terms?
erately need to go from this [:Microsoft for your listeners, we need Tefca for operations, and we need to trial it. Mickey at ONC has been ASTP has been calling it the Tefca to get 10 payers and providers one by one, point to point. Let's get a few folks to work on the use case of operations. And even before we have this national legal structure, let's demonstrate it's viable in a much more controlled experiment and scale what works.
I'm giving you a very long winded answer to your question. No, I love that you took it
there. Tefca for operations. it's so interesting because people are like, man, this has taken a long time. But, we're at the point now where the frameworks are in place.
Yep. The mechanisms are in place. . The data is probably in the best state that it's ever been. And we have, all the pipes, all the contracts and negotiations. We're ready. It's we're ready.
he business model. So if you [:In value based care contracts where I'm already paid for quality, under total cost of care accountability, call that the umbrella, for those of you who are nerds, HCP LAN, Category 3B and above. Let's make that an extension of the neighborly. So we'll do neighborly treatment, plus value based care accountability.
greement we could get done in:What's next for you?
nce value based care through [:So we had a lot of shared customers and a very clear alignment of ethics and philosophy. So we came together over the summer. So Care Journey merged with Arcadia to form one. chassis for VBC data infrastructure enablement, and I'm grateful to serve as Chief Strategy Officer. I do spend a reasonable amount of time volunteering on public private partnerships, from Sync for Social Needs, to Project Clarity on Price Transparency, to a whole range of Veteran Interoperability Pledge, and a whole range of other use cases, the Healthcare AI Commitments Project.
I love that. And so we're very much in the spirit of let's rising tide lift all industry boats. Not like my vendor machine will beat your vendor machine. But we need the country to succeed. So I love what I'm doing. It's a privilege and I'm grateful for it. And I'm lucky to be a part of a bigger chassis this year than I was last year.
to catch up with you. And I [:You're educating them. Your hands are on many wheels indirectly, my friend.
That's why we're having so much fun today.
Let's say my hands on the shoulder of the captain who has their hand on the wheel, and I'm encouraging them in a direction. Hey, I appreciate it. It's great catching up with you and hopefully you have a great holiday. You too. Godspeak, my friend. Thank you.
Thanks for listening to this week's keynote. If you found value, share it with a peer. It's a great chance to discuss and in some cases start a mentoring relationship. One way you can support the show is to subscribe and leave us a rating. it if you could do that. Thanks for listening. That's all for now..