Latest / HumanifAI HealthTech / Making Healthcare Data Actionable With AI w/ Dr. Arif Nathoo, CEO, Komodo Health
Transcript
- 0:00SPEAKER_00: Everyone knows that the biggest headache in healthcare is that most of the time the right hand doesn't know what the left hand is doing.
- 0:08SPEAKER_00: It is so frustrating.
- 0:10SPEAKER_00: It leaves patients and providers trapped in a never-ending game of broken telephone.
- 0:16SPEAKER_00: Today's guest set out to solve this breakdown and created a solution that acts as a universal translator, taking data from 60 completely different daily refreshed data pipelines and weaving them into a single cohesive map of over 300 million patient journeys.
- 0:40SPEAKER_00: Welcome back to Humanify Health Tech, where we explore how AI is rewriting the rules of healthcare and putting people back at the center where they belong.
- 0:49SPEAKER_00: I'm Brian Chi, Senior Director of Content at NetHealth, and your host.
- 0:54SPEAKER_00: Today I'm joined by Dr.
- 0:55SPEAKER_00: Arif Natu, co-founder and CEO of Komodo Health, a healthcare intelligence company that uses its proprietary AI engine to track patients' real world health journeys.
- 1:07SPEAKER_00: Arif holds a medical degree and a master's in public policy from Harvard and worked as a leader in the medical affairs practice at McKinsey and Company.
- 1:16SPEAKER_00: Arif, thank you for joining us today.
- 1:18SPEAKER_02: Hey, thank you so much for having me, Brian.
- 1:21SPEAKER_00: Yeah, it's a pleasure to meet you.
- 1:22SPEAKER_00: Um, and I want to talk a little bit about Harvard to start off with, but a little bit in a little bit of a different way.
- 1:29SPEAKER_00: So you hold degrees from Harvard in molecular biology, medicine, and then after completing medical school at Harvard, you then decided, no, I'm gonna go into consulting at McKinsey for a while.
- 1:44SPEAKER_00: Then you went back and you added a master's in public policy from Harvard.
- 1:49SPEAKER_00: So here's a question.
- 1:50SPEAKER_00: What did you see at the time to make you decide that that was the right pivot?
- 1:56SPEAKER_01: Such a great question.
- 1:58SPEAKER_02: Um, I had this distinct memory of being in my third year of medical school.
- 2:02SPEAKER_02: So in traditional, you know, pedagogical medical education, you spend a couple of years in the classroom and then you you get the exposure of being in the hospital for the first time really in your third year.
- 2:14SPEAKER_02: And um in the hospital, you know, I this was in the like early 2000s, right?
- 2:19SPEAKER_02: So we were just seeing the rise of EHRs and not, you know, really in the cloud yet, but just really basic uh kind of EHRs that pretty much every hospital was building and inventing themselves.
- 2:31SPEAKER_02: And I remember this distinct uh like experience where the medication list of one of my patients was completely wrong and pulling in the wrong place.
- 2:41SPEAKER_02: And like we almost made an error that could have led to that uh patient having a horrible outcome.
- 2:49SPEAKER_02: And it just dawned on me in that moment how important that transition from like, you know, the old world to what was happening with the digitization of everything had profound impacts on how patients, you know, would would basically consume care.
- 3:08SPEAKER_02: And at the same time, we were talking, and this is 2000, about the consumer-driven healthcare revolution, right?
- 3:15SPEAKER_02: So imagine this 25 years ago, we're saying, okay, consumers are gonna be in power, digitization is happening now.
- 3:21SPEAKER_02: In five years, we're gonna have a completely remade healthcare system.
- 3:25SPEAKER_02: So for me as someone who loves technology, loves data, I was like, oh, this is this is incredible.
- 3:31SPEAKER_02: I want to be part of that.
- 3:32SPEAKER_02: And it was a really hard thing for me to say, look, everything I've done to date, you know, is um set me on this path to be a provider, but I've got this really fascinating itch with the way all of these changes were gonna fundamentally transform healthcare.
- 3:48SPEAKER_02: I just have to be part of that game.
- 3:50SPEAKER_02: And it led me down the path of going into consulting, it led me the path of, you know, really uh taking that time to do a degree in public policy.
- 3:59SPEAKER_02: Uh, but my my goal was to ultimately um make a difference in the way that these systems were gonna truly revolutionize and change how healthcare was driven.
- 4:10SPEAKER_02: So that was where the passion came from.
- 4:12SPEAKER_02: And it it wasn't hard to say goodbye to something that I loved because uh I was truly excited about uh where things were going and my role uh potentially in that.
- 4:22SPEAKER_00: So EHRs were gonna change the world back then, but they kind of did in good and bad ways, right?
- 4:27SPEAKER_00: We were you like, did you like see see not only that opportunity, that digitization opportunity, but also the opportunity of what EHRs weren't getting right, right?
- 4:39SPEAKER_00: That kind of calm chaos that was part of that whole story.
- 4:45SPEAKER_02: Yeah, it's incredible.
- 4:46SPEAKER_02: Um I I I wouldn't say I saw it.
- 4:49SPEAKER_02: Like I experienced it over the course of time like everyone else.
- 4:53SPEAKER_02: I I don't think anyone back in like 2001, 2002 would have imagined the EHR landscape, you know, 25 years later.
- 5:00SPEAKER_02: It's definitely one possibility we would have ended up where we're at.
- 5:04SPEAKER_02: But I I don't know that uh the vision for what EHRs do, you know, beyond being a container for data, was very clear.
- 5:13SPEAKER_02: You know, how central would it be to the entire care delivery model?
- 5:17SPEAKER_02: Um, how much interoperability was going to bite us over the course of the uh you know prior two decades.
- 5:23SPEAKER_02: And I think that's really where we learned a lot through that process.
- 5:27SPEAKER_02: And I clearly didn't predict any of that.
- 5:29SPEAKER_02: Um, but what I did find was just opportunities left and right to improve the way that information was being exchanged across providers.
- 5:38SPEAKER_02: And that's really where the thesis of Komodo came in.
- 5:41SPEAKER_02: And frankly, where we see continue 25 years later from those days, see the opportunity, um, which is that uh, you know, in a in a world where every EHR is a walled garden trying to drive as much traffic through what it owns, it is incredibly hard to share information.
- 5:59SPEAKER_02: And no matter how much how many laws we pass on information blocking, uh, the behaviors of rational actors in a profit-driven system will not change just simply because there's regulation that encourages the right kind of behavior.
- 6:13SPEAKER_02: And so uh we think companies like uh like ours are really designed to allow us to start stitching together and seeing what's happening.
- 6:21SPEAKER_02: And to the exact point you made, how frustrating it is for a patient who goes and seeks care at multiple places, not even being able to connect it.
- 6:28SPEAKER_02: So so much of the observations I had in that in that time, you know, unfortunately have persisted today, but that's allowed um Komodo to really play a role in trying to trying to mend those fences.
- 6:40SPEAKER_00: Okay, so let's talk about this.
- 6:42SPEAKER_00: I'm gonna, I'm gonna, I'm gonna, I'm gonna, I'm gonna try to understand Komodo Health, right, in like a sentence, which is basically setting myself up for failure, but that's okay.
- 6:51SPEAKER_00: So so as I understand it, Komodo Health acts like a giant search engine and and probably a predictive map for the entire American healthcare system.
- 7:00SPEAKER_00: You're essentially connecting the dots of patient medical data, right?
- 7:04SPEAKER_00: Just in order to see a bigger picture and to be able to map what that looks like moving forward into the future.
- 7:10SPEAKER_00: Um is that right?
- 7:12SPEAKER_01: That is a great that was a great start.
- 7:14SPEAKER_01: You did you did wonderful.
- 7:15SPEAKER_01: Thank you.
- 7:16SPEAKER_00: Okay, awesome, awesome.
- 7:17SPEAKER_00: But it's only a start, I know, because then there's a lot of stuff after that map, right?
- 7:21SPEAKER_00: There's there's a lot of tracking, there's a lot of, and then that's where the data insights come into play, right?
- 7:26SPEAKER_00: And through it through through your proprietary AI, you you actually find those insights and you're you're actually helping to guide providers and payers in terms of um in terms of where to go.
- 7:38SPEAKER_00: Is that is that is that how it works?
- 7:39SPEAKER_00: Is that is that absolutely the idea of it?
- 7:42SPEAKER_02: You you kept on going and it just got better and better.
- 7:45SPEAKER_02: I Brian, that's exactly one of the things we realized was that um stitching together the observations is just really the first step in the process.
- 7:55SPEAKER_02: And we developed what we call our healthcare map, which is a de-identified view of patients across their entire life cycle.
- 8:03SPEAKER_02: So the average patient goes to many different providers, shows up at many different pharmacies, changes these things year over year, and that's because their insurance changes over and their behavior changes with it.
- 8:16SPEAKER_02: And as a result of that, capturing those histories plus maybe what they consume outside of the formal healthcare system, all of that together is the tapestry by which every company, whether you're a pharmaceutical company, a payer, a provider, whether you're a regulator or working in public policy at a state or county level, has to look at in order to take care of your population.
- 8:44SPEAKER_02: But the information is just the first part of it.
- 8:48SPEAKER_02: Like getting that linked together was hard work, and building the pipelines to do that has been a decade plus in the making.
- 8:55SPEAKER_02: Now, on top of that, though, we always talk about being data rich but insight poor or analytic poor, and that's really where the challenge we see happening is that there aren't enough folks that are uh statistically trained in the world to spend uh to answer every question that gets asked in healthcare.
- 9:13SPEAKER_02: And this is really where the AI revolution has enabled us to not only apply like predictive algorithms to really identify patterns that uh will tell us a lot about what happens to populations, but really be able to run that uh generatively and continuously as models have really accelerated over the last few years.
- 9:34SPEAKER_02: And so we're just um at the tip of the iceberg of being able to use, in my opinion, real-world evidence to shape uh provider and payer and life sciences behaviors at a scale that we've never seen before.
- 9:47SPEAKER_02: And this is really uh just the beginning of what the future looks like, which is that our ability to continuously understand what's happening in real world practice, not just you know what gets published on a controlled study, that's really where uh the world is going and what we're in very good position to be able to not only see but act on.
- 10:07SPEAKER_02: And I think that's really where uh we see the greatest opportunity for the system to improve.
- 10:12SPEAKER_00: Okay, so so tell me that sounds great, but what does it do for patients?
- 10:17SPEAKER_00: Like ultimately, like one of the things that we talk about at NetHealth is um the idea that our our our ROI is based on better outcomes, right?
- 10:27SPEAKER_00: The stuff, the technology that we do has to end up in better outcomes for patients.
- 10:32SPEAKER_00: That's the ultimate.
- 10:32SPEAKER_00: That's our shining star, our north star.
- 10:35SPEAKER_00: So let me ask you that question: how does all of this amazing technology, the solution, actually help somebody get better?
- 10:43SPEAKER_02: So we know that every enterprise healthcare company is out there to live their mission.
- 10:50SPEAKER_02: And their mission is to take care of their population and based on how they're structured, turn a profit or take care of their employees or do what they need to do to survive.
- 11:01SPEAKER_02: Our mission is to pair those companies up with really deep understandings of where patient unmet need is, and when you hit that sweet spot, patients really benefit.
- 11:14SPEAKER_02: So I'll give you an example.
- 11:16SPEAKER_02: Uh, years ago, we uh supported a company in diagnosing a rare disease called TTR amyloidosis.
- 11:26SPEAKER_02: This is like seven or eight years ago now, and uh set actually uh enabled the company to send test kits out to doctors that looked like they had patients with you know specific phenotype.
- 11:37SPEAKER_02: And through that process, tons, hundreds of patients got diagnosed within the first few weeks.
- 11:43SPEAKER_02: And that allowed a lot of patients to get on life-saving therapy.
- 11:46SPEAKER_02: And this has happened in so many conditions where uh oftentimes a provider will see one or two of a certain kind of adult onset rare disease in their life.
- 11:57SPEAKER_02: And what if you could actually use this tapestry to help you understand where those patterns are and devote, you know, the right kind of resources to get them moving?
- 12:06SPEAKER_02: So just uh for us, it's been about finding those intersections where there is a need and there is a highly motivated institution, uh, and we see that uh driving this incredible benefit to patients.
- 12:20SPEAKER_02: And I think over time, and again, I joked about how the consumer-driven healthcare revolution started in like 2000, you know, like there's a uh there's a professor at Harvard's business school, Regina Hertzlinger, that was spending a lot of time talking about that.
- 12:33SPEAKER_02: And you look at, you know, like the world today, and you could argue that's it's nowhere near being consumer-driven.
- 12:39SPEAKER_02: We're seeing a lot more uh folks using chat and claw to actually take um their healthcare into their own hands, which is incredible.
- 12:46SPEAKER_02: And I believe the next several years, with aided on the fact that we have access to incredible tooling, will fundamentally change how providers have to work with patients.
- 12:55SPEAKER_02: So we're at the again, I think we're now finally gonna start seeing that revolution the way that we had imagined it 25 years ago.
- 13:02SPEAKER_02: But what I believe is that uh the institutions that are like highly motivated in this world are providers.
- 13:11SPEAKER_02: You know, they're gonna lose patients in mass as patients seek care through AI payers as patients realize they can get cheaper medication and access outside of the payer.
- 13:22SPEAKER_02: We're gonna see a fundamentally reshifting of how these folks think about providing care.
- 13:27SPEAKER_02: And I think that is a huge opportunity for them to learn and grow from how patients are consuming care.
- 13:33SPEAKER_02: And that's really what we see in our healthcare map, and that's how we provide them better care than they can probably get today.
- 13:40SPEAKER_00: So is this some this this thinking, is this behind your collaboration with NCQA?
- 13:45SPEAKER_00: You know, the kind of understanding real-world patient outcomes, that sort of thing.
- 13:50SPEAKER_00: Is that is that where you're headed there?
- 13:52SPEAKER_02: Absolutely.
- 13:53SPEAKER_02: I mean, credentialing is so important for institutions because it really levels the playing field on a whole bunch of metrics that matter to outcomes.
- 14:01SPEAKER_02: And I think historically, NCQA has been using a lot of claims data to set uh HETIS measures, uh, and the conversations we're having with them expand the range of kinds of data they can use.
- 14:12SPEAKER_02: I mean, should we be using lab data and outcomes around that?
- 14:15SPEAKER_02: Should we tracking more sophisticated uh outcomes for uh populations?
- 14:20SPEAKER_02: Should we be thinking about risk in a totally different way?
- 14:23SPEAKER_02: Um, the NCQA represents really an incredible organization and opportunity to think about generating measures for the year that we're in now.
- 14:34SPEAKER_02: And the speed at which they're able to build, test, and validate a new measure has compressed like massively.
- 14:41SPEAKER_02: And that is really going to open the door to a much more sophisticated framework by which all actors that are risk-bearing in the system have to understand their risk and play by.
- 14:53SPEAKER_02: And I think uh that is a huge benefit for patients.
- 14:56SPEAKER_02: I know the patient doesn't feel it like, okay, I'm asking Chad a question and I'm getting a clear answer.
- 15:01SPEAKER_02: They don't feel it there.
- 15:03SPEAKER_02: They feel it in, hey, you're gonna get a better test, get to an outcome you need, and get therapy faster.
- 15:09SPEAKER_02: And that's gonna create like better life outcomes for you because the payer and the provider are no longer getting in your way to that outcome because the payer is highly motivated to ensure that you get that outcome.
- 15:21SPEAKER_02: And the more that we can do that and drive the behavior of the institutions that take care of us, the better off the patient is, even if they're not the ones that feel like you know they're getting the answer right away.
- 15:31SPEAKER_02: And I think that's really why our company has to exist, because a lot of these companies are not using the data they need in order to make better decisions that lead to better patient outcomes.
- 15:41SPEAKER_02: So bringing um uh uh groups like the NCQA on really allow us to drive the right glue in the system to provide the right incentives for people to behave in the right way.
- 15:53SPEAKER_02: So we we're really excited about this.
- 15:54SPEAKER_00: So that one thing you said there about how you know what your company does is sort of key to those better outcomes, right?
- 16:00SPEAKER_00: Crucial to that, it makes me think about our our shift now to value-based care, to the value-based care model, right?
- 16:08SPEAKER_00: And and one of the things that kept recurring to me when I was looking into Komodo was it seems very aligned in terms of accelerating that sort of long-term care journey outlook, right?
- 16:19SPEAKER_00: Where there's some risk, but ultimately the quality of that is the total care of the patient.
- 16:26SPEAKER_00: Is that where strategically, is that what you guys are thinking as far as how you're building Komodo and the things that you're doing, that sort of idea of value-based care?
- 16:36SPEAKER_02: Yeah, absolutely.
- 16:37SPEAKER_02: I mean, so much of what we like every institution optimizes around is a very narrow part of the journey that they're effectively trying to get paid on and optimize their payment rate against it.
- 16:50SPEAKER_02: I do think that as we start thinking about journeys and episodes and uh longer uh view, and we work with uh this incredible company that's deriving the next set of groupers and sort of wonderful nonprofit called Paces, I have a like deep belief that like as we start thinking about episodes differently, as we characterize them differently, as incentives are driven in the right way, they've more and more closely align to the reality of the patient.
- 17:16SPEAKER_02: Right now, I believe that there's still a lot of optimization that's not necessarily in the patient's best interest because somebody gets paid in a way that doesn't quite align with what the patient's trying to achieve.
- 17:30SPEAKER_02: Um, and we've seen that, you know, really as GLP ones explode, that this is not so much about improving outcomes as it is about managing cost.
- 17:38SPEAKER_02: And so it's all just pushing us into like a place where the patient's not necessarily getting the best care, even though it's being rationalized as, hey, this is better, we're achieving the outcomes we want.
- 17:49SPEAKER_02: Um, you're not.
- 17:50SPEAKER_02: Like patients are not as happy and like, you know, have a lot of other problems when we don't address obesity as aggressively as we should be as a society.
- 17:58SPEAKER_02: And so there's just a great opportunity for us to like really move to a value-based care system, but one that actually thinks about episodes of care and thinks about like the duration of patient uh needs in a much more expansive way.
- 18:11SPEAKER_02: And I think look, AI is making that a lot easier for people to conceive and and and and do the analytics on.
- 18:17SPEAKER_02: You can now procedurally run you know millions of different uh combinations, like far easier than you ever could before.
- 18:25SPEAKER_02: And so, you know, you you can actually do this type of work.
- 18:28SPEAKER_02: Um, and I think that's gonna like lead to all boats rising, but we need to see our credentialing organizations, we need to see policymakers really pushing our private sector to do more of that work.
- 18:39SPEAKER_02: And I think that's why data is a great equalizer in that conversation.
- 18:44SPEAKER_00: Yeah, that's for sure, right?
- 18:45SPEAKER_00: And then the other thing that I think is a equalizer, maybe, or an advantage, is the native part of the AI, right?
- 18:52SPEAKER_00: The proprietary part of the AI.
- 18:54SPEAKER_00: One of the things that I see you consistently mentioning is that, right?
- 18:58SPEAKER_00: And and then it and then it makes me think though, well, everybody's kind of saying that right now.
- 19:03SPEAKER_00: Everybody's saying native AI, proprietary AI, AI driven, AI powered, AI this, AI that.
- 19:09SPEAKER_00: What's the difference, right?
- 19:11SPEAKER_00: What is a why is native AI or or proprietary AI not a buzzword?
- 19:16SPEAKER_00: What is it, what is it, what does it actually mean in terms of value?
- 19:19SPEAKER_02: Yeah, it's so funny.
- 19:20SPEAKER_02: I I I don't know that there's any uh I think when we put the word proprietary in front of it, it means or maybe implies that, hey, you're not using a frontier model that anyone else can have access to, which is why I'm allowed to charge you a lot more for it.
- 19:32SPEAKER_02: Um I think the the thing that we are going to um get to, so there's two worlds.
- 19:38SPEAKER_02: Like people either believed in the like kind of closed weight frontier models are really going to dominate the next several years.
- 19:44SPEAKER_02: There's an equal kind of voice that, okay, open weight models that are getting smarter and smarter, really going to be the way that we, you know, cost arbitrage the use of AI down.
- 19:53SPEAKER_02: And I really think if you actually get, you know, that's the really the wrong fight.
- 19:57SPEAKER_02: And you know, I think at the end of the day, the right fight is are you driving better outcomes for patients or for your customers?
- 20:04SPEAKER_02: Can you prove that?
- 20:06SPEAKER_02: And they should put those outcomes up against what every other system they can they can use, whether it's a system they build off of the best frontier models, whether it's another company of space that's got their own proprietary AI.
- 20:19SPEAKER_02: At the end of the day, outcomes are really what matters.
- 20:22SPEAKER_02: And with Marmid, it's something that we've spent a lot of time.
- 20:25SPEAKER_02: Marmid is our AI platform that allows companies to really take this incredible like plethora of data and drive really actionable work off of it.
- 20:37SPEAKER_02: We preach outcomes.
- 20:38SPEAKER_02: Uh, we, you know, you will not invest in this unless you're driving, you know, you're finding patients faster, you're getting them on therapy.
- 20:45SPEAKER_02: If you're a life sciences company, or if you're a payer, you're finding needs, unmet needs that need to be addressed so you can improve your performance against your measures, or if you're a provider where you can see patients leaving your system because they're not satisfied with the care.
- 21:00SPEAKER_02: And if you kept them in there, you would do better.
- 21:02SPEAKER_02: And so every opportunity that you're able to find that drives a better outcome for you, you're willing to pay for.
- 21:08SPEAKER_02: Right now, you pay a lot of consultants, you pay a lot of advisors to give you those answers.
- 21:14SPEAKER_02: But what if you could empower your own people with the same level of quality, the thinking, drive to a better outcome faster?
- 21:20SPEAKER_02: Could you change your business?
- 21:22SPEAKER_02: And that's really the thesis that we adopted with Marmid.
- 21:25SPEAKER_02: And we're just so excited to take to market because it really resonates with our customers.
- 21:30SPEAKER_02: We're not talking about proprietary AI.
- 21:32SPEAKER_02: We're asking, like, are you able to get the better outcomes with Marmot than you are with anything else?
- 21:37SPEAKER_00: So, what guardrails are you putting in place there?
- 21:39SPEAKER_00: Because, you know, I mean, it it sounds in our conversation, this is super exciting, this is amazing, but these are real actual human lives.
- 21:48SPEAKER_00: They're not just data points or whatever, right?
- 21:50SPEAKER_00: So so you you're dealing with humans, you're you're studying humans, de identified, right?
- 21:56SPEAKER_00: What are those guardrails?
- 21:57SPEAKER_00: What are you what are you putting in place to to Protect and care for those humans?
- 22:02SPEAKER_02: That's a is a great, great question.
- 22:05SPEAKER_02: So the thesis behind uh de-identified data is that uh we do not re-identify patients.
- 22:13SPEAKER_02: And that is a the fundamental thesis why you know the government created HIPAA and the rule set that allows for de-identified analytics.
- 22:21SPEAKER_02: And so we constantly not only certify our data with an external third parties, I constantly say, like, okay, if I bring these two data sets together, do I increase the probability of statistical re-identification?
- 22:33SPEAKER_02: We always maintain that privacy.
- 22:35SPEAKER_02: So our data is not used in any fashion to re-identify a patient.
- 22:40SPEAKER_02: That's the first guardrail.
- 22:41SPEAKER_02: And it is the sanctity with which we're you know given data to use for this purpose.
- 22:47SPEAKER_02: But we ensure that both the usage of the data and the privacy guardrails around the data are number one in the way that we operate as a business.
- 22:56SPEAKER_02: And uh our our privacy and compliance team really focuses on that.
- 23:00SPEAKER_02: Number two, AI has an uncanny ability to personalize the journeys of patients and make them sound like real people.
- 23:07SPEAKER_02: And I think there is a kind of grossness to the way you ask an AI to look at healthcare data and the way they personify uh actions and things in the data, and it really creeps people out.
- 23:19SPEAKER_02: So we spend actually a lot of time uh like figuring out what the right way is for AI to like actually engage this kind of data, work with it, like censor data in the right way.
- 23:32SPEAKER_02: And so uh we've built a ton of guardrails around uh allowing our users to specify how they want data to be analyzed, what ways they want it to be presented.
- 23:43SPEAKER_02: And that's all very selectable.
- 23:45SPEAKER_02: So it's uh yes, you can impute an individual preference, but organizations can place a lot more control over how they want data used.
- 23:52SPEAKER_02: And that's the second kind of guardrail is just really creating those rules and guardrails around the way that data is used.
- 24:01SPEAKER_02: And then finally, um, what's really important to us is verifiability of the work that's being done.
- 24:07SPEAKER_02: So Marmitt doesn't actually just run through and give you some answer, it generates code that can be run on the data, and every piece of code that it generates, maybe three or four hundred pieces of code over the course of a conversation, will is all fully auditable, and you could run that directly yourself on the data and see what it returns.
- 24:28SPEAKER_02: So you have this like incredible control over that.
- 24:31SPEAKER_02: Because if imagine if you're preparing a study, you're gonna submit it to the FDA, they're gonna ask Did you use AI for this?
- 24:36SPEAKER_02: You know, and the reality is that an AI can write a piece of code, but you own the outcome, you own the submission.
- 24:44SPEAKER_02: And so every piece of code that's written, you have to be able to run it, you have to be able to get verification on it.
- 24:50SPEAKER_02: And so we give our teams the ability to do that.
- 24:53SPEAKER_02: And that is game-changing for a lot of them because now all of a sudden they have a framework by which they can see exactly what's being written, they can change it, they can modulate it, they can adjust it with or without an AI, and they own it.
- 25:05SPEAKER_02: And I think that ownership is really important to doing work that you have to stand behind as a as a business.
- 25:11SPEAKER_00: So let me talk about, let me press in on the ownership piece there, uh, especially considering the whole hugging face uh scenario and this current call to slow down AI development, right?
- 25:22SPEAKER_00: You know, AI agents have become more and more powerful to the point where we don't even if you tell them not to do something like re-identification, they may do re-identification in order to accomplish the task, right?
- 25:38SPEAKER_00: Maybe a bit of an impossible question for you to answer, but how does how do your guardrails ensure that that's not going to happen, even if you're telling the AI not to do those things?
- 25:49SPEAKER_02: It's a great question.
- 25:51SPEAKER_02: Um so there are two sort of general approaches.
- 25:55SPEAKER_02: Like one, you can sort of limit the AI's access to the internet.
- 25:59SPEAKER_02: Um, because imagine if you're just taking public information and you're like grabbing as much public information you can and really trying to whittle down based off of you know what somebody may post on social versus what they find online, to what this de-identified identity is.
- 26:17SPEAKER_02: So there's that whole like, do I give them access to the internet or not?
- 26:20SPEAKER_02: Um, and there are uh definitely cases where we enable or whitelist only certain sites that it can use.
- 26:27SPEAKER_02: And there are some clients who are like, we don't want that, and there are other clients who are very open to the usage there, provided that you set appropriate guardrails exposed, like after the collection of information.
- 26:39SPEAKER_02: The second uh approach is to actually um really test the system, meaning like you test your own data uh aggressively against uh the best that uh the market has to offer.
- 26:53SPEAKER_02: So if you're gonna use a mythos or fable class like AI to do the work, you let it run amok on the data to test whether in a protected environment it can actually re-identify any of your patients.
- 27:06SPEAKER_02: And so we have taken the uh approach of both.
- 27:09SPEAKER_02: We have scrutinized our data in the new world because every person that sits on any piece of data has to vouch for the way it's used and how it's used.
- 27:18SPEAKER_02: And so you do the testing.
- 27:19SPEAKER_02: Like you act as if you have to be better than any threat actor.
- 27:23SPEAKER_02: And so you kind of keep your senses up, and then you also put all the guardrails in place for what it can do.
- 27:29SPEAKER_02: Hugging face represents a, you know, the problem that a lot of frontier companies have, which is that they are way ahead of the market.
- 27:36SPEAKER_02: Like if you look at what has been released to the market, they're at least a year ahead, in at least a year ahead in terms of what they're sitting on.
- 27:43SPEAKER_02: And when they get scared, um, it says a lot about what the next year is going to look like because they are so far ahead internally from what they released to market.
- 27:52SPEAKER_02: So I think this call to action is actually late.
- 27:55SPEAKER_02: Uh, I think everyone's fear was like that, this is what we should have been thinking about and pushing harder on a year ago, but there's a lot of monetization in a couple companies that have to go public that's driven the behavior of like the utilization and these, you know, like kind of the growth of these businesses.
- 28:11SPEAKER_02: But I think we are past that point of of uh of where an AI can actually be an incredible threat actor.
- 28:19SPEAKER_02: I think it's far worse than uh you know we're we're we're led to believe right now, which is why I think we're in a in a very dangerous time where I think most companies are incredibly naive to the way that they're using AI and are setting themselves up for um these types of failures in the near future.
- 28:37SPEAKER_00: How does how does healthcare react to this?
- 28:38SPEAKER_00: I mean, I mean you're you're you're it's a best of best and worst of both worlds, right?
- 28:43SPEAKER_00: Because on one hand, healthcare needs AI to advance so many incredible things, like value-based care, like healing, like like cures for diseases and and and fixing the process that's so broken.
- 28:56SPEAKER_00: But on the other hand, there's this danger and and this this what you're just saying about how it's a lot more serious than we think.
- 29:04SPEAKER_00: What how do we react in the healthcare space?
- 29:07SPEAKER_02: Well, what's amazing is that, and I you know, I feel like this is a wonderful time, you don't need a like the top end of the frontier model to impact a huge amount of change in healthcare.
- 29:18SPEAKER_02: And I think what we have not done is tap into the incredible power of frontier models from even a year ago to driving that change.
- 29:27SPEAKER_02: You don't need the you know, the the top of the line to impact health.
- 29:32SPEAKER_02: And and I think this is the truth of it.
- 29:34SPEAKER_02: We can do so much with where frontiers are were like a year ago that we're not even uh scratching the surface of the change that has to happen.
- 29:44SPEAKER_02: So I'm an optimist on this front, which is to say, even if we were to shut down all like frontier work for the next year, and we were just to limit ourselves to exactly what we have now, healthcare has a decade plus of optimization they can do with just what the tools they have now.
- 29:58SPEAKER_02: That is how far behind the system is to where AI is.
- 30:01SPEAKER_02: And that's not just people running skills on their machine and feeling proud that they can slightly automate a process.
- 30:08SPEAKER_02: That's actually creating fully agentic loops that drive better outcomes for patients if you were to completely reimagine the system.
- 30:15SPEAKER_02: And so the next breed of companies that are coming up don't require like, you know, fable astral level like performance in order to actually impact patients at a huge scale.
- 30:25SPEAKER_02: So I'm a huge optimist in the belief that like we've already got the tools.
- 30:29SPEAKER_02: Now it's actually time to go out and make it happen.
- 30:32SPEAKER_02: And I think that is the uh beautiful opportunity that healthcare is faced with.
- 30:37SPEAKER_02: So, you know, as worried as I am about where Frontiers are going, there's so much opportunity uh just in the and how good things are today.
- 30:45SPEAKER_00: Let's talk about the other part of the AI thing, right?
- 30:48SPEAKER_00: That the before all this really got into the national consciousness, right?
- 30:53SPEAKER_00: This whole idea of accuracy with AI and healthcare, right?
- 30:56SPEAKER_00: That's the thing that everybody was worried about.
- 30:58SPEAKER_00: Two weeks ago, we were all talking about wow, AI, is AI more accurate than a clinician?
- 31:03SPEAKER_00: What you know, who's gonna diagnose me, right?
- 31:06SPEAKER_00: So, and you talked about this, you've talked about this being in defensible intelligence, right?
- 31:11SPEAKER_00: So what is how is how does Komodo look at this idea of plausible answers not being good enough and that AI needs to make it make a make a decision or an answer that is truly defensible?
- 31:25SPEAKER_02: Yeah, it's a great question.
- 31:27SPEAKER_02: Um, I I would say like in our I'll give you an example in our space, right?
- 31:30SPEAKER_02: So we help um healthcare analysts uh do do work.
- 31:35SPEAKER_02: And if you think about an uh like any healthcare analytic question, like so I had somebody ask me, it's like, oh, is the allergy season in Texas better, you know, worse this year than last year?
- 31:45SPEAKER_02: So we just ask a simple question.
- 31:47SPEAKER_02: You're fascinated by like a you know something from a public health point of view.
- 31:51SPEAKER_02: You ask 10 analysts how to do that analysis, they'll give you, they'll give you like 15 or 20 different methods, right?
- 31:56SPEAKER_02: So there's so much art, and there's no one method that I think dominates.
- 32:00SPEAKER_02: And so much of it's dependent on the data set, the coding that you choose to use, like is it allergic rhinitis?
- 32:05SPEAKER_02: Do you define it, you know, through EHR features, through claims features?
- 32:09SPEAKER_02: How do you think about the denominators?
- 32:10SPEAKER_02: Like, you know, how do you normalize?
- 32:12SPEAKER_02: How do you even, you know, maybe you know where Texas is, but how do you normalize for like the different county coverage levels in the state?
- 32:17SPEAKER_02: Like these are the kind of things that, and then how do you define, you know, one season over another?
- 32:22SPEAKER_02: I mean, really simple things, but every analyst has a different way of thinking about the problem.
- 32:27SPEAKER_02: And if you go and look, you know, read all the literature, you'll see incredible diversity in the way different researchers even approach these problems.
- 32:34SPEAKER_02: There's no one standard.
- 32:35SPEAKER_02: And the beautiful thing is that AI allows a researcher to be conscious of all the variables that maybe they're not conscious of.
- 32:43SPEAKER_02: And I think that is a beautiful thing.
- 32:45SPEAKER_02: Like a lot, a lot of times our biases used to be checked, like maybe when we presented some research to journal club or to our peers and they say, Hey, did you think about this?
- 32:54SPEAKER_02: Like, oh yeah, I didn't think about this.
- 32:55SPEAKER_02: AIs have become this incredible thought partner to put everything on the table, to help you think through all the issues.
- 33:00SPEAKER_02: At the end of the day, though, you have to choose a method.
- 33:03SPEAKER_02: Now, I can leave it to the AI to choose a method.
- 33:06SPEAKER_02: It'll choose a fine method.
- 33:07SPEAKER_02: It'll choose the most like probable method, uh, unless I tell it not to.
- 33:12SPEAKER_02: Uh, but you can kind of guide it around any road that you want.
- 33:15SPEAKER_02: Ultimately, though, you have to make the call.
- 33:18SPEAKER_02: And I think um empowerment of humans to make judgments based on everything someone can offer you is a beautiful way for us to think about AI.
- 33:29SPEAKER_02: We think about this in education kids.
- 33:31SPEAKER_02: Okay, they're just gonna go and use AI to write their essays.
- 33:33SPEAKER_02: But what if they were like it was their thought partner?
- 33:35SPEAKER_02: You know, like why did you choose this word or do what the teacher does?
- 33:39SPEAKER_02: How would you imagine this differently?
- 33:41SPEAKER_02: You know, I think this is the way that we, you know, I honestly feel like AI shines.
- 33:46SPEAKER_02: They shine when they are counselors to people's own ingenuity.
- 33:51SPEAKER_02: And for those of us who use AI at like really like incredible level, like I use it a lot.
- 33:57SPEAKER_02: It like I see the thought partnership evolution in the models over the last year.
- 34:02SPEAKER_02: And that's the thing that gets me the most excited because I think every human wants to be able to talk about things that you know they don't understand.
- 34:09SPEAKER_02: Uh, I learned how to cook last month and I just had conversations with AI around like all the things that I'm insecure about.
- 34:15SPEAKER_02: And, you know, oh, is this you know too hot or is this, you know, is this browning the right way?
- 34:19SPEAKER_02: Things you can't just Google.
- 34:20SPEAKER_02: And it's made me a better cook, you know, like so it's just funny because I feel like when you think about AI as not a decision maker to resolve ambiguity for you, but as a like way to help you think through your own ambiguity, the power of doing research, I think, accelerates mightily.
- 34:40SPEAKER_02: Uh, I just believe that we are so afraid of abuse that we don't we don't use AIs in the right way.
- 34:47SPEAKER_02: And I think that's really meaning that like we wanted to just solve our problems rather than helping us think through our problems.
- 34:53SPEAKER_02: And I think that's really uh an opportunity for us to all grow.
- 34:57SPEAKER_00: Yeah, for for about two weeks, I was the smartest dad in the entire world.
- 35:01SPEAKER_00: And then they caught up with me.
- 35:03SPEAKER_00: But I know two weeks were gold, man.
- 35:05SPEAKER_00: And it's all laid on.
- 35:06SPEAKER_01: I knew everything.
- 35:09SPEAKER_01: I love it.
- 35:11SPEAKER_00: One of the one of the things that we we talked to um Dr.
- 35:14SPEAKER_00: David Rue from Microsoft about this a little bit, and it was really interesting because he was saying what you were saying, which is basically uh AI shines a light on where you need to focus, right?
- 35:25SPEAKER_00: That's that's sort of the thing.
- 35:26SPEAKER_00: And and so I want to uh my question to you is your AI engine, where is it shining its light on financial leaks, on care inefficiencies?
- 35:36SPEAKER_00: Where what what are the things that you see most frequently in those buckets that um um that that are that are important and pressing?
- 35:47SPEAKER_02: Oh, it's a great question.
- 35:48SPEAKER_02: We I mean, what's amazing about Marmitt is that it's been adopted by so many different folks in the healthcare system.
- 35:54SPEAKER_02: We've got payers on it, providers on it, life sciences companies, financial services firms that are trying to understand like the future of different companies and how they'll report earnings.
- 36:05SPEAKER_02: I think the a beautiful thing about Marmid is that it does cover this diversity of use cases.
- 36:10SPEAKER_02: But the one that I see the most is really understanding like unmet needs of patients.
- 36:16SPEAKER_02: Like we have companies that are like producing uh lab tests that don't even know what types of doctors are using what tests and and how that correlates to where patients actually are.
- 36:26SPEAKER_02: And when they see a behavior, they're like, wow, folks are not testing this kind of symptom pattern.
- 36:32SPEAKER_02: It's so stark in the data.
- 36:33SPEAKER_02: Like, if we could just drive this behavior, we could see a thousand more patients, ten thousand more patients tested.
- 36:39SPEAKER_02: And uh, that tends to be the place where I see a lot of inquiry, and it's it's a beautiful thing because we are seeing people use it to do exactly uh that which is shine a light on things that they don't know and they're not aware of.
- 36:54SPEAKER_02: Um that's a that's like I'd say hands down the number one use case, but they apply it in different areas, you know, like whether it's a device or whether it's a drug intervention or whether it's just a diagnosis, uh, that tends to be the number one use case.
- 37:07SPEAKER_02: And then beneath that though, there's just uh hundreds of ways people are using it to augment their work, to perform studies, to you know, really challenge their own thinking.
- 37:17SPEAKER_02: Uh so we we've just been impressed with uh just how creative folks are and and and pushing the AI to kind of move and and help them get better.
- 37:27SPEAKER_00: Okay, wild wild shot here, but I I gotta ask this financial question, right?
- 37:31SPEAKER_00: Kind of a financial question.
- 37:32SPEAKER_00: Like like payers, right?
- 37:34SPEAKER_00: Like CMS, they'll drop a new ruling or open up comments for a new ruling, and everybody will freak out because it's gonna change the business model for some large part of healthcare, right?
- 37:45SPEAKER_00: Like RPM and RTM, for example, most recently, right?
- 37:49SPEAKER_00: And everybody's like, well, what what's gonna happen now, right?
- 37:52SPEAKER_00: This is all gonna change, right?
- 37:54SPEAKER_00: So that relationship between the payer and the provider, how does having sort of this concept of a single data set, the having Komodo's sort of ground truth thing there, how does that help improve the relationship between these two these two functions?
- 38:11SPEAKER_02: It's a great question.
- 38:12SPEAKER_02: I mean, as long as they're still uh splitting the dollar, they're gonna constantly negotiate for themselves.
- 38:17SPEAKER_02: I mean, we, you know, obviously with the first things to change were how call centers were run.
- 38:22SPEAKER_02: And so then it's just like AI started fueling them three, four years ago, and you know, now it's just the war of like bots to kind of argue the cases for coverage.
- 38:31SPEAKER_02: I feel like the challenge is that as long as a dollar is shared, everyone's gonna be seeking as much arbitrage as possible to give them an advantage in that negotiation.
- 38:41SPEAKER_02: What I do see happening though, and I do think is really important, is that data provides both parties with the ability to like look, you know, to present their analytics rationally so that somebody else will listen.
- 38:54SPEAKER_02: And I think that is one of the most powerful uh things that we can do is to say, look, here is the representation of you know this intervention or this disease in your population, population you know well, and it's not 100%, but it's a large enough sample that we can derive some observations from, and here's what we see in it.
- 39:14SPEAKER_02: And that conversation tends to be very good.
- 39:16SPEAKER_02: It's also aided really well with the Nintelligent AI.
- 39:19SPEAKER_02: And that's why like the ability to like study the data and patterns and argue for it and consider what the issues are and the way you're presenting the data really enables people to have a much more rational discussion.
- 39:31SPEAKER_02: So we've seen the quality of discussions, at least between payers and providers, rising.
- 39:35SPEAKER_02: Um, if not, you know, for the fact that they are competing effectively on who gets the dollar.
- 39:41SPEAKER_02: I do think that the level of discourse is rising and the ability to tie it back to patients is there.
- 39:46SPEAKER_02: But I mean, Brian, as you know, this is not going to happen without more regulation that's more systematic and not to your point just one-off rule changes that tweak on the margin.
- 39:56SPEAKER_02: So we'll see how that uh plays out over the next few years.
- 39:59SPEAKER_00: Awesome.
- 40:00SPEAKER_00: Awesome.
- 40:00SPEAKER_00: Okay, so only a couple more questions before we wrap up here.
- 40:03SPEAKER_00: But I want to talk a little bit about your personal philosophy.
- 40:06SPEAKER_00: I uh I heard you say on a podcast, um, this was great, this was great.
- 40:10SPEAKER_00: Um, something along the lines of, and I'm paraphrasing here, so so so forgive me.
- 40:15SPEAKER_00: Failing forward is really great, but you know, it's just better to be right.
- 40:21SPEAKER_00: And so it is so much better to be right, right?
- 40:25SPEAKER_00: The for the first time or or or or whatever.
- 40:28SPEAKER_00: But it dawned on me when I was listening to that that you really were really right twice in your journey, right?
- 40:34SPEAKER_00: When you left that residency because you became enamored with the opportunity um that you saw on the on the tech side and and the larger systemic side, and then when you co-founded Komodo.
- 40:46SPEAKER_00: So my question is really simple.
- 40:49SPEAKER_00: How?
- 40:50SPEAKER_00: How, right?
- 40:51SPEAKER_00: If you're a founder, if you're an entrepreneur, you're getting out there, what how do you do that?
- 40:57SPEAKER_02: Uh so uh I think the right word is courage.
- 41:02SPEAKER_02: Um, I was uh at McKinsey for many years as a consultant.
- 41:08SPEAKER_02: Um, I had a really strong like academic pedigree.
- 41:13SPEAKER_02: And um I even after six years in McKinsey didn't really understand how the world worked.
- 41:18SPEAKER_02: And uh I realized that I'd been in sort of the kind of academic professional bubble for too long.
- 41:26SPEAKER_02: But leaving is really hard.
- 41:28SPEAKER_02: You know, you you you basically like you exit an organization like that to start your own thing, and you're you know, you're left whittling a spear in the Serengeti for food.
- 41:39SPEAKER_02: Like you have no resources behind you, you have nothing.
- 41:42SPEAKER_02: I mean, frankly, your education matters not.
- 41:45SPEAKER_02: Uh you know, it's it gives you nothing.
- 41:48SPEAKER_02: And what you have is your census and you have your own personal courage.
- 41:52SPEAKER_02: I I think the hardest thing for me to to have done at that time was to leave what would have been a secure job uh or secure set of opportunities if I had gone to industry to to really just start something on my own.
- 42:06SPEAKER_02: And um the the how is like it's not easy, you know.
- 42:10SPEAKER_02: I went through many crises of identity, of self, uh like times where I you know question myself very deeply, like every human does.
- 42:18SPEAKER_02: Like, I mean I don't think we're all any any of us are immune to self-criticism or doubt, and uh just being able to to work your way through it, to, to, to silence the the voices that tell you no is it's hard, you know, and it's it's the I think the courage to kind of do something different.
- 42:35SPEAKER_02: Um and whether it's better, I don't know if I'm a right, Brian.
- 42:39SPEAKER_02: You know, one of the things I always say is that like you know, like when we look at the world in normative terms like right and wrong or good and bad, it's like we end up putting too much pressure on decision.
- 42:50SPEAKER_02: No, something that may be right today may be wrong tomorrow.
- 42:52SPEAKER_02: And I I look at the life I've led and I, you know, there's certain things I'm really proud of, and there's certain things I say I still have work to do as a human.
- 42:59SPEAKER_02: And I I will always own that because I think that's what makes us better.
- 43:04SPEAKER_02: Um, the context of that conversation was on starting companies, you know.
- 43:08SPEAKER_02: I think the courage to start companies is hard, but you know, we're all we're most we're way more often wrong than we're right.
- 43:14SPEAKER_02: How do you call a spade a spade?
- 43:16SPEAKER_02: How do you pivot to the next thing?
- 43:17SPEAKER_02: Well, you know, that's that's hard, but I always do think it's better to be right, but I also think it's okay if you're wrong.
- 43:23SPEAKER_00: So and you know, learning learning how to be right also means failing too and learning.
- 43:29SPEAKER_02: Absolutely.
- 43:29SPEAKER_02: Yeah, absolutely.
- 43:30SPEAKER_02: I think surviving this business for 12 years has taught me so much about just what it takes to go through good and bad.
- 43:37SPEAKER_02: Um, and I don't think any founder who is still running their company after a decade, you know, has not experienced the worst uh uh that a that a business has to go through short of dying.
- 43:49SPEAKER_02: Uh and and I have, you know, like I've gone through all my uh lumps and I've also had great moments in the business and I uh and I love what I do and I continue to do it every day because it's the act of doing it that that brings me the most joy.
- 44:03SPEAKER_00: Or thank you for this.
- 44:04SPEAKER_00: This has been such a great conversation.
- 44:06SPEAKER_00: We started with this game of a broken telephone, and you and you kind of took us through this journey uh to show how the healthcare system can actually create a clearer map.
- 44:16SPEAKER_00: And that's that's just that just means better health for humans, right?
- 44:20SPEAKER_00: So thank you so much for your time.
- 44:23SPEAKER_00: Um we're gonna we're gonna we're gonna wrap up now.
- 44:26SPEAKER_00: So um I'll I'll let you go.
- 44:27SPEAKER_00: Thanks again, though, Arif.
- 44:29SPEAKER_01: Thank you, Brian.
- 44:30SPEAKER_00: Thank you.
- 44:34SPEAKER_00: All right, recap time.
- 44:36SPEAKER_00: Here are the top three things I took away from my conversation with Arf today.
- 44:41SPEAKER_00: First, solutions for better outcomes require a broad approach.
- 44:45SPEAKER_00: As a physician who walked away from residency to look at the macro business of healthcare, Arf showed that you can only solve the systemic healthcare challenges we face when you look at it through the lens of clinical science, public policy, and scalable technology.
- 45:02SPEAKER_00: Second, enterprise AI demands defensible intelligence.
- 45:07SPEAKER_00: In high-stakes industry where tech mistakes carry real-world clinical and economic consequences, we need native architectures engineered specifically for healthcare data structures that show their work, audit their data, and ensure absolute precision.
- 45:24SPEAKER_00: Finally, predictive analytics is the absolute engine for value-based care.
- 45:30SPEAKER_00: By mapping the continuous long-term health journeys of millions of patients, we turn that economic risk into a predictable clinical strategy that lets providers lower the total cost of care, close critical gaps, and focus on keeping humans healthy.
- 45:45SPEAKER_00: And that should do it for today's episode of Manify Health Tech.
- 45:49SPEAKER_00: As always, thanks for tuning in.
- 45:51SPEAKER_00: If you're joining us on Spotify or YouTube, we'd love to hear your thoughts in the comments.
- 45:56SPEAKER_00: And don't forget to leave us a rating while you're there.
- 45:59SPEAKER_00: Once again, I'm Broman Chi from NetHealth, and I'll see you next time.