3 Apr 2025 · 54m · Michael Krigsman, David Reich, David Bray, Teresa Carlson · 20 quoted facts
Useful AI in healthcare starts with a real problem and a measurable return, not a shiny tool. The panel argues that AI must fit clinicians' existing workflow, save time or money, or improve outcomes. It also needs open, usable data and sensible regulation. Without these, even strong technology adds work instead of removing it.
The central tension is between healthcare's fragmented, decentralized, fee-driven system and the push toward shared data and outcome-based care. Panelists say large language models can now make sense of messy, unstructured records without perfect standards. But privacy, patchwork regulation, data hoarding, bias and AI errors mean humans must stay in charge.
David Reich says a tool can pay off by avoiding regulatory penalties, shortening stays so beds can be refilled, or improving outcomes. This lets a hospital judge a pitch against a concrete benefit instead of novelty.
Reich says startups offering to work for free overlook the IT and clinical workflow work needed to integrate a tool. David Bray adds that electronic health records already created more human labor.
Reich's delirium model only became strongly predictive after natural language processing was applied to physician and nursing notes. He argues such tools can summarize huge charts, easing the lack of perfect data standards.
Teresa Carlson warns that a federal AI model plus fifty state models plus foreign rules could choke innovation. She wants healthcare handled through the FDA and industry experts, while Reich worries about being regulated out of existence.
Bray and Carlson argue that paying for outcomes requires seeing a patient's long-term history, such as pre-diabetes. That needs data access, consent and transparency for patients and clinicians, which Reich says is hard in a market where patients can go anywhere.
David Reich says to sort the idea into a bucket such as patient experience, workflow or outcomes, then ask what return it brings. That return can be money, avoided penalties, or shorter stays that free up beds.
Panelists say interoperability standards are decades old but still incomplete, vendors can use standards as a moat, and data holders have an incentive to hold on to it. David Bray notes that rules against information blocking exist but are not working as intended.
Reich quotes a Mount Sinai radiology chair saying AI won't replace radiologists, but radiologists who use AI will replace those who don't. The panel frames AI as augmenting clinicians, for example by summarizing long records.
Value-based care means paying for patients getting healthier rather than for each visit. Bray and Carlson say it requires understanding a patient's long-term data, which AI can help sift, though Reich says the US system makes this hard to measure.
Reich says tools built on biased data can mislead people, citing a study where radiologists were often led astray by deliberately wrong AI output. The panel also raises privacy, since Reich says no genome can be truly de-identified, and the risk of a regulatory patchwork.
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
David Bray worked on a bioterrorism preparedness program that tried to connect to hospital and state systems from the early 2000s to 2005.
“In a past life, uh, I was with the bioterrorism preparedness program. This was back in the early 2000s to 2005”
Teresa Carlson says CitiBlock, a General Catalyst-backed company serving underserved patients, is paid only when patients get better, under a value and outcomes-based model.
“they literally do not get paid until the patient gets better. So it's a value outcomes-based model”
David Reich cites the University of Pennsylvania as having used a nudge unit to switch prescribing toward less expensive drugs.
“University of Pennsylvania, uh, managed to switch, uh, prescribing, uh, behaviors to less expensive drugs.”
David Reich says the Consortium for Healthcare AI includes Mount Sinai and other major health systems.
“There's an organization called the Consortium for Healthcare AI. Mount Sinai is involved all of the major health systems.”
Teresa Carlson argues that the FDA should lead healthcare AI policy rather than creating broad new regulation.
“The FDA has been doing their job a long time. Let's let them actually work toward what the right policy is for their environment for healthcare.”
David Bray says the number of CPT claim codes grew from about 12,000 to about 72,000 over roughly twenty years.
“about twenty years ago there were 12,000 CPT codes for claims, now we're up to 72,000 CPT codes for claims.”
David Bray says the UK uses data trusts, and the US can use data cooperatives, to let communities control how sensitive data is used.
“In the UK, they call them data trust, here in the United States we call them data cooperatives.”
David Bray says a Birth to Three effort used existing contract law to create a data cooperative that includes community members and bars monetizing the data.
“we used existing contract law to create a data cooperative that includes members of that community”
David Reich says Mount Sinai began building a predictive algorithm for hospital delirium about four years earlier.
“we started working on a predictive algorithm for delirium in hospitals, which is a big problem.”
David Reich says the delirium model only became strongly predictive after natural language processing was applied to unstructured physician and nursing notes.
“as soon as we added natural language processing to the unstructured data in physician and nursing notes, the model became predictive”
David Reich quotes the chair of radiology at Mount Sinai as saying radiologists who use AI will replace those who do not.
“AI won't replace radiologists, but radiologists that use AI will replace radiologists that do not.”
David Reich says a Gartner survey found healthcare spends much less on cybersecurity than the banking industry.
“we look at how much we spend on cybersecurity versus the banking industry. And it's much less surprisingly”
Teresa Carlson says venture capital invested about $11 billion in healthcare AI in the year of the episode, within a health market of about four to five trillion dollars.
“venture capital is investing, I think, alone, about 11 billion this year in AI technology for healthcare.”
David Reich says European national health insurance systems ration care by deciding what drugs and services are covered, an approach he says would not suit US culture.
“they ration care by deciding what's in the national basket, what drugs will be provided, what service will be provided”
David Reich says Mount Sinai has 300,000 genomes in a project aiming for one million and that no genome can be truly de-identified.
“Mount Sinai has 300,000 genomes into 1 million genome collection project. There's no such thing as a de-identified genome.”
Teresa Carlson says Hippocratic AI, a General Catalyst investment, aims to close the healthcare worker shortage by helping staff scale.
“their mission is really to close the growing shortage of healthcare workers by helping them scale.”
Teresa Carlson says Hippocratic AI uses an AI agent that phones patients to ask follow-up questions, without prescribing or acting as a doctor.
“there's an agentic AI that calls the patient and says, ask them questions about their health.”
David Reich cites a radiology study in which deliberately wrong AI output misled residents about eighty percent of the time and experienced radiologists about fifty percent.
“residents were wrong eighty of the time. Even the experienced radiologists were wrong fifty of the time when the AI led them astray.”
David Bray says ultrasound for pregnant women met early resistance before becoming standard, as a model for patient adoption of healthcare technology.
“Initially, when ultrasounds and sonograms for pregnant women were introduced, uh, there was resistance.”
David Bray says laws rarely expire, leaving healthcare rules from the 1940s to 1960s that may no longer fit.
“there's things still on the books from the nineteen forties, nineteen fifties and nineteen sixties that may have made sense then”
The panel says AI pitches fail when they solve problems providers don't have or ignore integration cost, so return on investment must be judged from the provider's side.
“free isn't free because my friends in information technology are”
Carlson describes portfolio companies serving rural and underserved patients that get paid only when patients improve, as an example of aligned incentives.
“they literally do not get paid until the patient gets better”
Reich explains how a major health system screens ideas and counts penalty avoidance, outcomes and shorter stays as returns.
“ROI isn't always strictly financial because in our world, uh, we have penalties”
Carlson and Reich warn about overlapping rules and describe industry efforts to assure AI quality, while Bray calls for policy that keeps up.
“we also can't have a patchwork of a regulatory environment that doesn't enable this”
The panel debates why data stays locked up, rethinking the 'data is the new oil' idea, and how medical coding has ballooned in complexity.
“we need to unlearn the meme from about a decade ago that somehow data was the new oil”
Reich gives real examples, from a delirium predictor to chart summaries before surgery, showing how AI works around imperfect standards.
“as soon as we added natural language processing to the unstructured data in physician and nursing notes”
Carlson cites venture funding and sandboxes for trying tools, while Reich explains why measuring outcomes is hard in a free-choice US system.
“we advocate for sandboxes where we can allow healthcare institutions”
Reich raises genome privacy and the need for safe harbors, and Carlson describes an AI that follows up with patients.
“There's no such thing as a de-identified genome.”
Bray proposes data cooperatives, Reich warns about AI errors misleading experts, and each panelist gives closing advice.
“it's becoming quite the wild west out there”
We're exploring AI in healthcare today on CXO Talk number 874 with three experts shaping this future. Doctor David Bray is distinguished chair of the accelerator at the Stimson Center. Teresa Carlson is president of the General Catalyst Institute, and Doctor David Reich is chief clinical officer of Mount Sinai Health System and president of the Mount Sinai Hospital in New York. We're going to explore what's real, what's hype, and how AI impacts decisions, data, and patient outcomes.
So, thoughts on solutions and how AI can support the overall healthcare system?
The answer is to really analyze from the perspective of the providers, what is the true return on investment? I have many people in the startup world who are often approaching and presenting me with a problem that they can solve, but I don't actually have that problem. And so they're trying to convince me I do. And there are many problems I have that they don't really work with. And the other thing, of course, is that people in the early stages want to work with the health system and they say, well, we'll work with you for free.
Well, free isn't free because my friends in information technology are, you know, very, um, you know, upset with me is as a clinician. I come with them and I say, look at this bright, shiny object, and they try to explain to me the amount of effort that would go into integrating it into the IT workflow, let alone the clinical workflow. So I'll stop there for a moment.
I want to amplify what what David was saying. It's bringing back a lot of flashbacks. In a past life, uh, I was with the bioterrorism preparedness program. This was back in the early 2000s to 2005, and we were trying to plug into hospital systems, state systems and things like that to basically monitor. And as David so well put, the irony of electronic health records is they've actually created more human labor to complete. And and I think also that's what David was saying.
Any good solution in this space has to fit into the workflow that's already being done by clinicians. Now, there's a bit of a catch-22 because technology also changes the art of the possible. And so it's really you have to involve clinicians and then to, and and I'll give a nod to what Teresa's talking about. Um, we're now in a place because technology is changing what's possible so quickly, doesn't mean it's being adopted, but it's changing what's possible so quickly.
Policymakers almost have to iteratively learn also from what's even possible in the technology, both to remove barriers, but also to put in place incentives if we're going to move forward and think about third and fourth order effects almost in real time, which makes this a really hard space to find solutions. It's doable, but it's a very hard space.
David's so right, don't bring a solution to an area that there's no problem. And I think that should be the number one use case of the companies that are coming down. What problem are you actually solving? And there's got to be an ROI because you can't ask for more cost. Our job should be able to make our clinicians' lives easier, our physicians be able to service more of these patients faster, more effectively, and provide them the data that they need to get their job done more rapidly.
And for us, I would tell you, it's it's really, very much about ensuring that we're doing value-based care and outcomes-based care. And I'll just share one example. We have a couple of companies that we were with yesterday that we've invested in that are really thinking about patients that are, uh, really underserved populations. I'm from Kentucky, hence the accent. I'm from a very rural part of the United States, and one of our founders is focused on rural-based care.
And ensuring that we're not only just, uh, training and teaching, but we're actually using technology to force multiply the capabilities that are there. Because a lot of times you have the specialist, you don't a have the access to education or advocacy for these individuals. And they have individual and unique needs in their healthcare, and they are usually the last to get it sometimes. And so she's thinking about a problem, and there's a lot of America that is in rural America, not cities or urbans.
Uh, and then the second part is for, uh, a company that we have called CitiBlock that's taking care to the most underserved. And they're there's a Medicare Medicaid overlap with what they do, but they literally do not get paid until the patient gets better. So it's a value outcomes-based model, and both of these companies, Homeward and CitiBlock, are doing this model. That's very unusual. And when we talked to the legislators on Capitol Hill about it that yesterday, there's like, W- what?
I didn't even know this existed. So, you know, one of the physicians, uh, from CitiBlock, Toyin, who started this company, said she's a physician herself. And she said, Look, I could see a lot of patients during the day. I could give them a prescription and hope that they leave my office and go fill that prescription and they take it. But I can't ensure they're doing that, but I get paid for that visit regardless. And their model is much more about ensuring that there are outcomes from what they're actually prescribing and doing.
And that's what I call a shared responsibility model between that physician and the patient. So we're really looking for creative ways to ensure we're getting healthier outcomes, just not charging the system more.
David Reich, your thoughts on this kind of relationship between innovators and hospitals hospitals? I mean, you're president of of a major hospital, so I'm s I'm assuming you're dealing with these practical issues all the time.
One of the the key things that we're working on, Michael, is a a seamless pipeline for, uh, sort of digesting the the multiple groups that are coming to us with fantastic ideas. General Catalyst being, you know, one of the the top groups that we, um, we pay attention to. We regard anything that comes through an organization like General Catalyst as being highly, uh, something that's highly thought about. And, you know, it. It comes to us with a a degree of, um, uh, should we say it's it's curated in a way that we should pay close attention.
And as we do this, we, uh, try to break them down into the different buckets. You know, for example, is it something that would improve patient experience, something that would improve workflow, something that's focused on a patient outcome? And so that we can look at, uh, what our ROI is related to this particular idea. And ROI isn't always strictly financial because in our world, uh, we have penalties, uh, for failing to meet certain criteria. And so sometimes penalty avoidance is an ROI, sometimes better outcomes and shorter length of stay is an ROI because we can backfill, uh, with other patients.
So it all comes down to whether the idea integrates into, uh, sort of a category of ideas that make sense for us and is something where we can rationalize, uh, an ROI and sometimes partner. And since we're, uh, an organization that likes to think of ourselves as entrepreneurial, potentially, uh, work with, uh, startups and, uh, other groups, uh, and put in, in our case, what is often the sweat equity in helping develop and making a product commercially viable.
Check out cxo talk dot com, subscribe to our newsletter, join our community. David Bray, I'm hearing a core issue here is, uh, alignment between groups that have somewhat overlapping agendas and goals, but also distinct. Is that a correct way of looking at it? And how do we bridge that gap?
Yes, I think this would be the definition of what we would call a wicked problem, where you have not only different folks aiming for different goals that sometimes overlap, sometimes don't, but the very goals and how you're gonna pursue it are changing at the same time, too. That's why, you know, everyone's like, there's a magic wand, it'll magically fix healthcare and solve it by the next hour. I think we're all we're all trying to say what you really can do is if you get the stakeholders together.
And I think because things are moving so quickly and and things are shifting, do three things, which is think about, first and foremost, how does this improve the practice? And and and David Reich was very good at pointing out, thinking about ROI and ROIs, not just in terms of financial return, but giving back time to the clinicians, uh, healthier outcomes for the patient in terms of getting them, uh, released earlier, and then also penalty avoidance. And so practice first and foremost, and then on top of that, then thinking about the new program and having some curation.
So it's not just every shiny idea gets tried out because you just can't absorb that. The organization would be stressed and it actually might disrupt the organization more than help. And then finally, the policy layer, and that's where Teresa was saying, because we have policies, you know, our laws don't have expiration expiration dates usually, sometimes they do, but usually there's no expiration dates. And so there's things still on the books from the nineteen forties, nineteen fifties and nineteen sixties that may have made sense then, that don't necessarily fit now.
And I think what I would submit is one area in particular that I'm personally passionate about, coming from a background with the people center and Internet Coalition and what we're doing at the Stimson Center. Now, data, in my opinion, is a form of human voice, and I think we've got to think about, instead of data being something that's opaque to both the, um, patient, but also the clinician about where it came from, how it's informing and everything like that.
We've got to think about how it gives both more visibility and stakeholderism to the clinicians as well as the patient, because this is going to be helping to influence outcomes. And if you're in a rural state where maybe things aren't even digitized yet, they're not even accessible yet to help inform a clinician with your care. That's an unfair thing that we have to figure out how to address. But at the same time, we don't want patients and clinicians to lose the ability to choose when and where that data is used.
If we wanna take advantage of these most transformative technology companies that we hope will help David in his practice and our patients. The one thing I will share is that we also can't have a patchwork of a regulatory environment that doesn't enable this. And like, as a good example, I think, is healthcare. But if you think about, there could be a federal AI model and then fifty state models. And then if these companies work outside the US, they have another model that they have to adhere to.
That is not it. You're really, it's an extent, it's an existential threat to these really transformative technology. And we want the doctors and physician, the physicians and the nurses and the clinicians to have the best technology for what they're trying to achieve. But what we actually adhere to is we believe healthcare is a really good example where you can allow the industry to take control of how they want that AI to be enabled. And I'll just give one quick example.
The FDA has been doing their job a long time. Let's let them actually work toward what the right policy is for their environment for healthcare. And they're actually kind of already doing it. We have companies working in health with FDA. AI Doc was with the show. They work directly on the physician, uh, to look for, uh, capabilities to allow the physician, uh uh, physician to see if there's medical errors to help them with that. And we know that medical errors is one of the largest costs that they're moving fast, they need technology to support their efforts.
That if we can let those industries kind of operate, then we can take this AI capability industry by industry with the policies. And I think it will allow us to move a lot faster and let the experts who know that industry work on the right policies.
There's an organization called the Consortium for Healthcare AI. Mount Sinai is involved all of the major health systems. And one of the aspects that we're all concerned about is that we could be regulated out of existence before we exist. And so, and it isn't just at the federal level and the state level. We also have Joint Commission, other certifications, we have local health boards as well. And so what we've done often is tried to come together as a group of healthcare organizations and bring all of the three letter federal groups, at least as a start together.
With the concept that we might develop assurance labs such that as the AI tools are developed, that we have a way of assuring their quality and assuring that they, you know, fit into the, I would say, the the culture, as well as the all the different matrix elements of working in complex health systems. And I think the, you know, one of the ways we get around it right now is often a lot of the tools that we've implemented, especially ones we've developed internally for decision support.
We try to think of bringing the right team to the right patient at the right time. We don't actually make the diagnosis, but we try to identify the patients who most likely have that diagnosis and then bring the clinicians there. We'd like to move beyond that at some point, but that's going to take a little help. We need to have a real partnership between government, industry, and healthcare providers in order to be successful at that.
We've had John Halamka as a guest several times on CXO Talk, who's one of the founders of CHi. Looks like an interesting organization. We have some questions that are coming in from LinkedIn and Twitter, so why don't we jump there? And this is a question from Arsalon Khan, and I'll toss it out to the three of you, whoever wants to grab this. Has anyone mapped the entire set of healthcare workflows? And what about having AI recommend which workflows are archaic?
This is actually a needed space. Um, and there are different groups talking about, um, uh, how do you translate what is business process modeling in the healthcare domain into some way that can then be analyzed rigorously across different systems. Uh, there are different vendors that have tools for business process modeling in healthcare space, but they're each, in some respects, proprietary. And so if we can get to that and and actually have almost a universal way of expressing business process modeling in a healthcare space, we can actually do another thing too, which is we can go to the journals.
And this is where I'll give a nod to, uh, my other fellow colleague David to talk about. Sometimes the journals actually have peer-reviewed articles that recommend different ways of achieving outcomes that might actually be different. But because it's right now in text as opposed to business process models that can do comparison, you actually can't compare and say which of these leads to better efficacy for the patient. So this is actually an area where with AI, if we can actually get to some way of not being locked into proprietary vendor systems, but instead have a way of having interoperability to express business process modeling across both hospital systems and IT systems.
But then also translate it out of the text journals in terms of what is the peer-reviewed goal, then you could actually start to do some very interesting things and actually achieve that. However, first, we've got to figure out what's that place that encourages people to think more broadly than their own specific system, but think about holistically across systems of systems.
I'd like to address the, uh, the word archaic. One of the the jokes in medicine is that the way you learn is you, um, if you've done one case, you say, in my experience, uh, if you've done two cases, you say in my series, and if you say have done three cases, you say in case after case after case. So one of the problems that we have in medicine is this, um, uh, culture, uh, where experience, uh, uh, leads to, um, best practices, uh, that are not always evidence-based.
And as we develop the evidence base, it getting the physicians to change, getting the nurses to change, getting the transporters to change, everyone who's involved in care in a healthcare setting, uh, requires some sort of persuasion. Uh, the common term now is the nudge unit, uh, and the nudge unit is intended to identify and to gently or maybe s not so gently correct, uh, behaviors that are are headed in the wrong direction. Uh, often those are things that are are wasteful, either clinically or, you know, sometimes financially.
Like, uh, a perfect example was, uh, where University of Pennsylvania, uh, managed to switch, uh, prescribing, uh, behaviors to less expensive drugs. And we're very successful. And we see similar things at Mount Sinai, where we've, um, had some success. So I think going back to archaic, it regards, um, uh, the the most important thing I take into account is we have to think about behavioral change. And sometimes that'll be based upon behavioral economics, and so the behavioral economics, perhaps, are the m the answer that I would give to, um, starting to look at ways of overcoming archaic processes.
We have another question that has come in on a related topic, on the topic of interoperability. And this is on Twitter from Chris Peterson, who says, We're thirty plus years into healthcare interoperability standards, but we're still not there in many respects. And so how does this lack of standards affect the advancement of AI in healthcare?
We do hear a lot from our companies that it is because they are so creative. I mean, these entrepreneurs, I can't stress, they are trying to solve a problem, not just look at a problem and jump over it and step in it. I mean, they're really trying to solve a problem. But I would tell you that they do say interoperability is is a top challenge for sure. But I think what we're also seeing, Michael, are creative ways that they are trying to work around it and look for new ideas, even starting from scratch on some things, which I know nobody wants to do.
But they're like, Okay, we gotta solve it, not keep just keep talking about the problem, but look for creative solutions. And I will tell you AI, because the internet and these data sets, these large language models are getting so good that you're almost now beginning to create new systems of record without even knowing it's being created, right? So you're seeing these ideas and also now they're beginning to access systems that are there in a way that's creating new ideas to look at that data and look for solutions.
So while it would be we we talk about freeing the data a lot, and I think we need to free that data. Data needs to be free, uh, you know, our physician David needs to be able to take advantage of that in every aspect of his practice. He should never not have access to the information that he needs at his fingertips to treat a patient. On the more research side, our doctor David Bray, we want our researchers and our new our the entrepreneurs creating new technologies to have those capabilities that they can bring back to that clinical work.
So I would say, whatever we do in this country and around the world, we've got to make sure that we have open and accessible data. And just by the way, not just for the CL clinician, but for me, the patient. I want 100% access. I want to make decisions about my own data. I think you're seeing consumers get a lot smarter. So we need systems that are open and interoperable in order for the patient themselves to make the right decision about their care.
David Bray, let me ask you a a a question. Then, as Teresa was just describing, well, as David Reich and Teresa were just saying, interoperability is very important. And Teresa was just describing the need for, uh, flowing, the free flowing of data in order to support innovation. But on the other side, the economic incentives for healthcare information systems, in many cases, is to keep that data because he or she who holds the data holds the money.
So what about that?
I think we need to unlearn the meme from about a decade ago that somehow data was the new oil. Uh, oil, use it up, it's gone. Data, you use it, it's still there. And in fact, if you involve people associated with the data, whether it be a patient or a clinician, they'll find things wrong in the data. They'll fix it, they'll make it better, they'll create more data, they'll create better quality data. So in some respects, we need a different metaphor than data is a new oil.
Um, I also think we need to simplify the consent mechanisms, but also the transparency so a patient can see where is my data going? Cause right now you fill some forms, but they're paper-based forms usually, and you have no idea where it's gonna go. And then similarly, a clinician, as Teresa said, needs to be able to get access to that data, um, to to do what they need to do. And so there are technically policies that say you're not supposed to do information blocking.
Um, and and and this is one of the sort of tensions of the US. system, given that we have a free market, which is great, but we also then are trying to think about how we overcome individual self-interest to hoard. Um, the last thing I would say on the interoperability challenge, Michael, uh, I often use the phrase that standards are like toothbrushes. Everybody says they want to use one, they just don't want to use anybody else's.
And so you have to be careful because not all standards are created equal. Some standards are really vendors trying to build a moat, sadly, uh, it's it's standards that are propped up by them. But other standards groups really are about trying to level the playing field, whether it's for established companies, startups and the like. A And so usually I look for do they charge or not to use that standard. I think the last thing I would say is also technology is is advancing just what's even possible in medical medicine.
Um, and we've gone from about twenty years ago there were 12,000 CPT codes for claims, now we're up to 72,000 CPT codes for claims. It's getting where it's really difficult for a human to know which claim code to use, whether it's an experienced coder, let alone a patient trying to figure out when they actually are looking at their bill. So this may actually be something where we need to think about a new strategy, almost like what Teresa was saying.
What would it look like if you had your own personal chatbot to help you understand your bill if you got your bill and or preauthorization to navigate what is a very complicated standards-based process? Um, and how do you make it more accessible to everybody, um, not just someone who's a technological expert?
We have a question from LinkedIn from Ravi Karkara, who is the co-founder of the AI for Food Global Initiative, and David Reich, he says this. What AI skills will be needed for medical and nursing schools to teach in order to create this AI medical revolution?
We don't teach so many things in medical and nursing school other than perhaps the traditional uh uh things that emerge from anatomy, physiology, pathology. And we need to teach people about systems-based practice. And that it it isn't just AI. The whole point is technology is here. I would argue that healthcare is a technological industry that has failed to recognize it for decades. And so, uh, I couldn't agree more, uh, with the individual that asked the question that, uh, as we reform the curricula, it is about teaching people how to best integrate with technology, with, uh, the proviso that you have to teach also that, uh, AI and other technologies that it's not a threat.
It's actually a way to augment the way you deliver care. And, uh, I have a friend who's chair of radiology here at Mount Sinai, and he says that AI won't replace radiologists, but radiologists that use AI will replace radiologists that do not. So I'll I'll leave it at that.
I just want to remind everybody that you can ask your questions on Twitter, use the hashtag C X O Talk. If you're watching on LinkedIn, pop your question into the chat. And truly, when else will you have the chance to ask such an esteemed group of people pretty much whatever you want? So I hope you take advantage of it. And this is from, again, from Arslan Khan who comes back. And he says, Regarding healthcare data, do we need more healthcare data or less?
And what can the government do at the federal policy level to help this data and interoperability issue?
I've been in the world, I've been working in the public sector world for over twenty eight years, and I'm used to a lot of data, and so I've seen it just continue. So I so I don't know that we have a limit on the data anymore. And artificial intelligence is really a key solution for us to be able to enable and understand that data, uh, research on it, get immediacy from it. So I don't, I don't think data.
I think I would be personally okay. I think as much as we need, it's okay, cause we're gonna keep building on it. So I don't know that there's a limit. When you're working with folks like the intelligence community and defense, and you see over the years the amount of data in healthcare, I think we're okay. But, uh, to your point, what are the policies? I would tell you we need to educate a lot more our policymakers.
I do think they want to understand, but I don't know that we've done a good job actually explaining to them that if the data is not free flowing and there's not accessibility to it, what's the harm that it does? Why does it inhibit our ability to make America healthy again? If that is the goal, if that is this administration's goal, what does it do? So I see our job as an example at the Institute working with partners like this to go in with examples.
And and to your point, uh, David's point earlier, there are policies in place that are ma- supposed to make sure that the data is open and free flowing. But for some reason, we hear from a lot of our companies and a lot of big and small companies, it's just not working. And not just them, we hear from the physicians and the clinicians and the hospitals, it's just not working the way it's supposed to work. So one, we need to figure out why that is, and then we need to ensure that the policies that are in place already are being activated, activated.
And somebody is watching out for the patient to make sure that that data is open and available to them, the clinicians, and the hospital systems, and the vendors, the partners that actually have to have it to make America great again and healthy. Sorry, healthy. We're already great. I think America is already great. Let's make it healthy.
And Michael, if I can build on what Teresa said, because, um, I want to give a nod, General Catalyst, uh, the General Catalyst Institute put out a report. And what was interesting is they were talking about what Teresa mentioned earlier with some of the examples, shifting, uh, from right now, what is a service where you get paid a fee, but doesn't necessarily mean that the patient followed through with the healthcare advice, followed through to get actually get healthier.
Shifting from that to value-based care and outcome-based care. And I would submit this is where the data is so crucial, because if you're going to do value-based care and outcome-based care, you almost first have to understand the the sort of the long, the long tail of that patient and understand are they pre-diabetic? And therefore, if you get involved now, you can avoid some of the more costly things that they actually become diabetic. Um, but that's one of those things again, where you know, uh, it's both as what, uh, what David Reich was saying about how do you help fit into the existing workflow for clinicians?
Cause if you give them a forty or fifty year old's patient record, they don't have time to wade through it all, um, to make sure they catch everything. And so how can you give them confidence if they're working with an AI augmented tool? It's not replacing them, but augmenting them, confidence that the AI has brought the relevant things to help them inform that. And then when you teach future generations of doctors and nurses, how do you make it?
So it's not just about the individual clinician, but part of a larger system to make sure that handoff for all the people that are providing that value-based, outcome-based care are thinking about that patient centrality. And so that's a major revolution to try and fit into the next five to ten years. Um, that's already got a system that's already got strains. But, you know, that's what also makes this space very interesting. And that's why we need to have these conversations now about let's make sure we don't lose sight about empowering the patient, and then also making sure the clinicians have the tools at their hand to actually help with that.
When we talk about maximizing fiscal responsibility in US. healthcare, the only other point that I wanted to make here, if we want to cut the red tape out and do that, we don't want to make it more expensive to get to the data and be able to access it to get those outcomes. Because again, let's remember AI and technology over time should drive the costs way down, right? We should think about, I know people talk about it driving it, it should drive the cost down because you can do it faster, more efficient, get to the outcomes.
And then let's just not just less expensive, but then you get to the results faster. So that allows our clinicians to be able to make decis- the the decision making much faster. So I just wanted to say that because we we talk a lot, this this administration is very focused on cost cutting. So we can't make it more expensive, and that's why technology should be the enabler, both to move faster, get the resourcing there, and again, don't make getting to the data more expensive.
Let me give a couple of practical examples, if I could, that I think illustrate the points we're just making. About four years ago, we started, maybe even longer, we started working on a predictive algorithm for delirium in hospitals, which is a big problem. And the structured data, um, were not very helpful. Did not end up with a predictive model that was, you know, all that impressive. But as soon as we added natural language processing to the unstructured data in physician and nursing notes, the model became predictive, almost scary predictive in terms of how accurate it was.
And so I think that, uh, as someone who was working with SNOMED and ITSI do years ago, trying to establish standards and then washing watching the Motes, as, uh, David Bray described, being built by the different electronic health record providers. At first, I thought, Oh, this is just the end of standardization, we'll never have interoperability. But now I'm starting to see that the tools, especially large language models that are able to create some summarizations of massive data sets, be very important.
I walked into the OR this morning, and the patient had had five previous cardiac operations. I'm a cardiac anesthesiologist. Just imagine trying to summarize that data. It's challenging, but honestly, if a large language model had been available to me and had given me a five paragraph summary of the most important, uh, unstructured data in all of those records, it would have made my job much simpler. I would not have spent forty five minutes last night poring through electronic records.
So I think those are practical examples of how we've lost the battle in terms of having the data perfectly structured. But with interoperability and using newer tools, I think we can overcome those, uh, uh, sins of the past, we'll say.
What I find fascinating is that we even have to have a discussion where we remind ourselves that we have to put the patient first.
It's just there's so many different stakeholders in this system, and I do think it is, you know, again, because we separate our our our public sector from our private sector, which is a strength. But it does mean you have to think about, you know, how do you empower the clinicians, which are plural, which are working across different IT systems that are provided by different vendors? And then also the fact that the patient has the ability to choose, in some cases, where they go, and they may go across state lines.
And for some mysterious reason in the Constitution, because nobody mentioned, uh, who got to oversee healthcare, it defaults to being a state right, unless the federal government uses some preemptive clauses. And so I actually think, in some respects, uh, I liked what David Reich was saying, cause I also have those battle scars from the standards battles about twenty years ago, which sadly, there were people that were trying to to to create a standard identifier for everything in the known universe.
And I would submit that is never going to be doable. But now, with using unstructured data and using new approaches that can actually say, I don't have to have everything specifically specified. But I can help the clinicians with understanding whether it's through a summary of the text, but it could also be using computer vision to go through a thousand computer images and actually say these are the ones you should pay attention to. I think, if anything, it may be we're in a world that's drowning in data, but lacking in insights.
And if you can give the insights back to the clinicians and the patient, that's how we're going to get through this.
We have an interesting question from LinkedIn from, uh, Greg Walters around this set of issues. Before I take that question, I just want to mention to everybody who's watching, you guys are an amazing audience. You ask incredible questions, and we want you to join the CXO Talk community. So just subscribe to our newsletter, go to cxo talk dot com. So we can tell you about our upcoming shows, cause we have incredible shows coming up, really awesome ones.
So Greg Walters says this, it's, he says, that he believes AI can be the glue in healthcare, not only accessing old data, but creating and tracking new data and therefore being the being that connective tissue. And he wants to know what's the level of investment dollars and the commitment to the m uh, uh, regarding this mission of AI that the healthcare industry is currently putting towards these technologies? And what is the what is the engagement level as well?
We look at ourselves using, uh, standard surveys, I believe Gartner did one for us, where, for example, we look at how much we spend on cybersecurity versus the banking industry. And it's much less surprisingly, uh, and and especially with the challenges in healthcare, we're an industry with very low, uh, margin, and in fact, with, uh, downward pressures that we're going to see related to, uh, you know, a a desire on the part of the administration to reduce healthcare spending.
That we really wonder if we're going to be able to afford, uh, to stay up to, uh, the level of spending that other industries have related to information technology writ large. And so the, uh, answer, I think, has to be that we, um, always when we look at the ROI on a tool, have to see a way that it actually reduces some aspect of the expense. And so, for example, finding tools that are better able to engage patients as partners in their own care.
Truly personalizing medicine isn't just looking at genomic markers and biomarkers and environmental factors in deciding what's best for that patient. You also have to analyze something that we've never really looked at, which is the receptiveness of the individual or the family to actually being a partner in that care. So as we move from, you know, good old fashioned, you know, case after case after case that I referred to previously, uh, towards population health, towards personalized medicine, we have to really start to, uh, bring it down.
Not just to the level of the provider and the patient, but the patient and their desire and interest to interact with the provider. Making America healthy again is actually a partnership.
Hey, Michael, if you look at venture capital, venture capital is investing, I think, alone, about 11 billion this year in AI technology for healthcare. And and about a, you know, could be small in a four, 5 trillion dollar health market, but it's getting there and we believe in it. And I will tell you, our CEO, Hemant Nuneja, has been investing in healthcare, I mean, for so long, he's a believer, he's created companies. We we believe in this model so much of AI transformation in healthcare, we have a whole group just dedicated to this.
And the other thing I'll share is, um, we are seeing on the healthcare side, the the six CEOs and founders we just had up here yesterday, their business is growing at a really rapid rate, all focused on AI. And we're seeing the healthcare industry, the the payers and providers, really lean in because they see the opportunity. So I expect this, uh, the AI efforts in healthcare to grow, uh, exponentially. And I'll just give you one example that we've been talking to this administration about.
And we put in our white paper that we just put out a week ago, which we advocate for sandboxes where we can allow healthcare institutions and the US. federal government and states to see the value. So they try out, they realize how, you know, they need to understand it, and we're a big believer in experimentation because we don't want back to doctor David, uh, MD's point. We don't want h- people just selling him stuff. He should see the value in a technology before he adopts it, so he should have the opportunity to try it out in a sandbox.
Because he may say, to his credit, he's a smart guy, he works every day very hard with pay. He might say, I'm not sure, but I'd like to try it. So we need ways to allow for our, uh, clinicians, healthcare providers to try out tools and show the value quickly, and then they can acquire it and scale it. And I think in today's world of cloud and AI, it's not a license based world. You should be able to try things out and then scale it.
David Reich, how do you measure the clinical value, clinical impact of tools such as, uh, Teresa was just describing and and the outcomes, the impact on patients?
Well, it really comes back to the, you know, the the many ways that ROI can be calculated. Um, a simple one for a very complex tertiary or quaternary care hospital is, can I assure a better outcome, a lower mortality, a lower infection rate? Can I empty a bed more quickly and fill it with another patient that generates another DRG? Just at the simplest level, at the population health level, uh, something that we have not really been successful at, at least much less so on the East, uh, East Coast, a little bit better on the West Coast, is value-based care.
So if I had a panel of patients and they really had a narrow network and they really had to use our health system for their care, and I better manage, let's say, their diabetes, their hypertension, et cetera. You know, can I reduce their spend? Can I truly bend the healthcare cost curve? When you have a wide open sort of rodeo here in the United States, where if I'm not happy with Mount Sinai Health System, I can go to the health system a mile down, you know, Second Avenue or Third Avenue and go to a different one.
You know, we don't really have the ability to really model and measure at that level. So it's, you know, really, to answer your question a bit, Teresa and Michael, it's such a mess because we don't really control the, you know, the spend, and we don't have a good way of measuring the outcomes. In an environment where people can get their care wherever they choose, and that's part of what we value as Americans, is freedom of choice.
But if you go to Europe, then, you know, they've taken a different approach with national health insurance, and they've created, you know, sort of a captive market. And they can, for, you know, it's not a word that we like to hear, but they ration care by deciding what's in the national basket, what drugs will be provided, what service will be provided, and can apply standards that are not really acceptable in US. culture. So I think we have to also think about these cultural barriers to bending the healthcare cost curve.
Doctor Faith Maimon-Gould, who is CEO of the View Hospital in Doha, he says, he appreciates the valuable medical examples shared for AI applications during this discussion. However, we should not overlook the significant potential of AI in non-medical areas within healthcare. As we continue to face challenges around resource allocation and affordability, AI can play a transformative role in driving efficiency and improving access to care.
We know, for example, with this incoming administration, they want a focus on eating healthier, um, also trying to change your lifestyle, you know, go outside, enjoy the sun. Those are all things that you don't need a medical doctor or a nurse to both prescribe for you, you don't have to prescribe. And two, it's hard for them to follow up on it. But you could imagine if you did have an AI app that was actually about helping inform the patient so that they could actually do the behavioral changes that David Reich was talking about.
So much of health, especially if you want to sort of bend that cost curve he was talking about, they're things that traditionally in the United States we called public health. They were behavioral changes. But again, because the strength of the United States is we're highly decentralized, that's also the challenge of the United States is we are highly decentralized. You have freedom of choice. The question is, how do you bring, how do you bring the nexus of both choice and agency to the individual to let them pick what they want to do?
But also find a way to bring together all those different parts that are not in the immediate medical setting that can improve health outcomes. And I think that's where there's a long tail, both before you see a doctor and a clinician, but then also afterwards, too. That is that is huge opportunities for innovation in this space. You know, and recognizing, again, different countries have different choices. The US. strength is we're decentralized, but any solution has to take that into place.
This is from Philip C. on LinkedIn, who describes himself as a builder, investor, and advisor. And he says, Amen to sandboxes. Is there a realistic pathway to bring together user data isolated in healthcare systems and user data scattered across their consumer apps? It seems like that's a natural next step for greater contextualization of people's healthcare needs via gen AI and other AI techniques. So going back to what we were talking about, uh, for, uh, David Reich, you want to jump in to this one?
The how to how do we bring data together?
Well, I, I have a friend who taught me the expression that no one lies to their um, search uh, bar, uh, search engine. And, you know, and it's, uh, you know, when we stop laughing about it, there, you know, there's a certain truth to it. Because, you know, uh, you know, we, we, we really want to know certain things. And I think that, uh, you know, looking at, you know, consumer information, marrying it with medical information, it, it certainly is going to provide additional insights.
And so it's something that I would look forward to. The other thing that we really haven't addressed so much in this talk is is the concept of privacy. What is privacy? Did we give it all up once we started tracking ourselves with our phones? Well, of course, we gave up some aspects of it. And there's a balance in society between the respect for privacy and the need for things like national security. So where are these balances going to occur?
And as genomic information proliferates, Mount Sinai has 300,000 genomes into 1 million genome collection project. There's no such thing as a de-identified genome. So we have to bring together what can be de-identified and perhaps work on that. And I think we also have to create safe harbors as something that's in the national interest, where we can bring together data that have to be protected so that individuals' genetic information is not released. But to truly personalize medicine, to truly bring together, it's the information outside of the medical records, it's the information inside the medical records, and it has to be shared in a way that we are comfortable with as a society.
I'll just say one thing, Hippocratic AI. I'm gonna answer our gentleman from Doha and this new gentleman by saying, I agree that there's so many health applications out there that actually not not just the day-to-day clinical care that we're talking about, but help in in many other ways. And one of the solutions that I just love is one of our investments called Hippocratic AI, and their mission is really to close the growing shortage of healthcare workers by helping them scale.
And they don't do, they don't do prescribing, they don't do, but they do follow-up. And like they call down, there's an agentic AI that calls the patient and says, ask them questions about their health. Have you checked your blood pressure? Have you taken your medicine? Did you get your prescription filled? Have you eaten healthy today? Back on nutrition, where you could scale a nutritionist to a million people at a time. But then merging those two data sets, they're a great example of being able to open up the worldwide web of the internet of information to also some clinical details without going into prescribing or doing the work of a doctor.
But helping that doctor and nurse be better, have more time on their hands, and them getting that information about back about what is the patient doing, how, when I actually see them face to face or do a telemedicine call, what are the core things that I have to focus on now? So we're seeing so many great applications now that are truly helping, uh, our physicians and nurses scale their work effort. But also on the back end, it's more administrative work, it's helping educate and advocate.
But things, honestly, I'm sure doctor David would love to do, but doesn't have time to do. He's he just doesn't always have time to do that. So this is where AI is really gonna help us, I think, in our world of of care for patients.
David Bray, very quickly, question from Twitter How do you deal with ensuring ensuring that AI is used ethically in healthcare? There are competing agendas, including extracting value from patients to business. So how do we manage these competing agendas and where do the where do the ethics come into play? And very quickly, please.
Two things that I think we need to see more of, um, and and I've I've seen this already work in the UK, we can do it in the US. In the UK, they call them data trust, here in the United States we call them data cooperatives. I volunteer with an effort called Birth to Three that tries to make sure every infant in the United States gets the necessary physical, mental, and emotional care they deserve. Obviously, data about infants is very sensitive data.
And so what we did is we used existing contract law to create a data cooperative that includes members of that community, including representatives, and they have choice as to how the data is used, where. And we explicitly said at no point in time will this data ever be monetized, but they can actually then choose which systems are done. And we're increasingly we're we're pushing for the algorithms, if they're going to train, come train in situ with the data, as opposed to porting the data somewhere else.
This also solves the interoperability problem because it's not about interoperability of the data elsewhere, it's about can you bring the algorithms? Now, there's challenges with that, but you can do that. The last thing I would also say is we may need to have more boards that include members of the patient community being served by a clinical setting, be involved maybe every three months, every six months for just let's explain what we're using the AI for, and you can ask questions.
So, stakeholderism, both on the data and on the AI.
David Reich, very, very, very quickly. You mentioned the use of gen AI to summarize, uh, complex patient his histories. Question from Twitter Who should be held reliable if the AI recommendations are hallucinat- hallucinat- hallucinations, and how would you even know that they are hallucinations? Very quickly, please.
We have a real obligation as we develop the tools to assure that they are accurate. Uh, I often refer people to a study done by radiologists showing that when in then mistakes were intentionally entered into a predictive algorithm, that it led people astray and the residents were wrong eighty of the time. Even the experienced radiologists were wrong fifty of the time when the AI led them astray. So I think it is a prerequisite that the tools we develop have to be overseen, because when we build tools, we build them on data that are frankly biased in nature because of inequities within our healthcare system.
So that's part of the quality assurance process. We make a lot of errors in medicine and we're always struggling to try to make less and less and to improve things. The same thing applies to any technological tools that we have at our disposal.
As we finish up, I'm gonna ask each one of you, uh, a separate policy-related, advice-related question, and I'll ask you to answer, uh, quickly. Teresa, let me start with you. Given all of this that we've been discussing, what advice do you have for policymakers when it comes to technology, such as AI in healthcare?
My top advice to them is they have to learn they and and I'm seeing them open to this, but they really have to learn and their policies have to move faster. The a technology is moving so much faster than they're instituting policies, and so they they've gotta learn and understand, they've gotta be open to new ideas, they've gotta be really on top of this. But last, what we went to earlier is they need to be thinking about AI at the industry level.
Let the experts do their job, don't do broad policies that hurt the industry. Do something that allows the industry to understand, don't overregulate, but let them do, they're already doing it, let them just own it and do it well.
David Bray, what advice do you have for technology developers when it comes to AI in healthcare? Very quickly, please.
Learn from what happened with the introduction of, uh, ultrasound for sonograms, um, in the healthcare space. Initially, when ultrasounds and sonograms for pregnant women were introduced, uh, there was resistance. There was questions about the efficacy, it had to be proven, but also, how do you work it into workflow? But now nobody would question, in fact, if anything, it might actually be wrong to not do an ultrasound for a pregnant woman. So it's worth understanding that there is a journey, be patient with that journey, and figure out again, as David Reich was saying, how do you plug into that workflow, but also show quality?
David Reich, what advice do you have for healthcare leaders when it comes to AI and healthcare?
Around yourself with a group of forward-thinking people and people who are perhaps even part of the old guard, you need to have a brain trust, and that brain trust has to be there. It has to come together and provide the best advice because it's becoming quite the wild west out there. And we have to have a way of digesting and, uh, and really facilitating better healthcare within our organizations, reducing waste, cutting out cost, uh, and assuring quality.
And, uh, you know, technology can be a double-edged sword, so we have to be sure that, you know, we, uh, do the best that we possibly can. And the way to do that is to surround yourself with the best minds.
Makes sense? Surround yourself with the best people. And with that, a huge thank you to Doctor David Bray, distinguished chair of the accelerator at the Stimson Center, Teresa Carlson, president of the General Catalyst Institute, and Doctor David Reich, chief clinical officer of Mount Sinai Health System and president of the Mount Sinai Hospital in New York. And a huge thank you to all of you folks who watched and who asked such great questions. Check out cxo talk dot com, subscribe to our newsletter, join our community, and come back, come back next week.
We have two members of the House of Lords next week talking about AI and the ethical aspects and regulation. So join us. Thanks so much, everybody. I hope you have a great day and we'll talk to you soon.