Who Owns Your Health Data? LLMs, Data Access & Venture Capital with Dr. Timothy Martens (Part 2)

15 years ago, getting your medical records meant visiting the hospital, signing forms, and walking out with a photocopied stack of 5,000 pages. Today you can import the same data into an LLM and ask it what's wrong with you. That shift — and everything in between — is what Part 2 of this conversation is about.

In this episode, IDAAHub Podcast host Deepti continues her conversation with Dr. Timothy Martens — congenital heart surgeon at Northwell Health, Director of Data Strategy & Innovation at Cohen Children's Medical Center, PhD in Biomedical Engineering, and General Partner at Picap Fund — picking up at the question of patient data ownership and utility.

Dr. Martens explains what patients can realistically do with their health data today: how LLMs are replacing the photocopied records stack with an instant, queryable medical history, why EMRs are caught between federal mandates to open their data and the disruption that follows when they do, and what it means to build a true "living health record" that updates dynamically alongside a patient's care journey.

The second half of the episode traces a 25-year arc of healthcare data evolution that Dr. Martens experienced directly — from manually copying blood pressure readings row-by-row into Excel spreadsheets, to building relational databases and using HL7 for data transfer, to the emergence of data lakes and Tableau dashboards inside systems like Epic, and finally to today's AI-native stack. His core observation: the tools have changed dramatically, but the fundamental challenges of normalization, deduplication, and fuzzy matching across siloed systems remain structurally the same.

🧠 What you'll learn in Part 2:
→ What patients can realistically do with their health data right now — and where it still falls short
→ How LLMs are turning a 5,000-page records request into an instant, queryable health profile
→ Why EMRs are now in a difficult spot: penalized if they don't open up data, disrupted when they do
→ The 25-year arc of healthcare data: from copying blood pressures row-by-row into Excel, to HL7, to data lakes, Tableau, and AI — and why the core problems of normalization and deduplication never actually went away
→ How Dr. Martens built Mark Ventures as a venture studio, evolved it into an investing syndicate, and joined Picap Fund as General Partner
→ Why having a clinician on a VC investment committee changes which bets get made — and which ones don't

The episode closes with Dr. Martens' transition from clinician to investor: founding Mark Ventures as a venture studio to build and advise healthcare technology companies, evolving it into an investing syndicate as founder demand for capital outpaced demand for advice, and ultimately joining Phycap Fund as General Partner — where his clinical domain expertise directly shapes the fund's investment decisions.

Guest: Dr. Timothy Martens — Congenital Heart Surgeon, Northwell Health | Director of Data Strategy & Innovation, Cohen Children's Medical Center | General Partner, Phycap Fund
Host: Deepti — Founder, IDAAHub | IDAAHub Podcast: AI in Finance & Healthcare
[Part 1 available in the previous episode]

Transcript

SPEAKER_00 0:12

Can we back to data? You talked about with the EMR opening up their data to patients, you know. So patients will get access to this data or they own it. Even if they do, what are they gonna do with that? I mean, a lot of us are as patients, we don't even know what we do or aware of how that can be used or should be used for that matter.

SPEAKER_01 0:42

Yeah, so it, you know, technological adoption takes time. But um, you know, 15 years ago, if you wanted your data, you went to the hospital's medical records department, put in a request, signed some forms, and then maybe they handed you a photocopied stack of 5,000 pages, maybe they gave you a flash drive or a CD with you know four gigabytes of literally scan stuff on it. Um and then it was on you to ingest and read and figure out, and maybe you were super motivated and organized, and you you kept your operator reports and your lab values, and you threw out all the like day-to-day nursing notes that like didn't have a lot of uh you know granular high-yield information in them, or the cut and paste and copy forward pages of regurgitated inaccurate text, like you had to filter that. This is again another perfect use case for LLMs, and it's what OpenAI and Chat GPT and countless number of companies are doing. It's it's very easy now to import. Um, you don't need the paper copy, you don't need to scan it anymore. You can use one of these QHINs, you can take the flash drive and dump the you know the all of the PDF, and then the LLMs can go to town and they can filter through that and say, okay, based on what's in here, here's the diagnoses that are confirmed, here's the lab values that go with it. They can start to put together a pretty comprehensive history that you can now query against, ask questions to, you know, and obviously you have to everything's predicated upon the models were generalist and not necessarily perfectly trained for this, but they definitely get better with time. The hallucinations are down. Uh, you know, you're seeing models now being offered that were specifically meant to look at medical records rather than the garden variety, like I ingested Facebook, you know, and was trained on the internet stuff. Um, so the the tools are there. Um you know it puts the it puts the EMRs in a bad spot because they get penalized if they don't open up and make the data accessible. Um but uh as they make it accessible because there are all these other streams now, your your wearables, your whatever, and there are tools, you know, instead of instead of having to go to my chart to find uh, you know, or go to medical records to print out 400 pages of PDF, now you can go to my chart, you can tell the LLM, hey, you have my history. I I just got you know, I just got a whole bunch of blood drawn. What numbers do you want? And then pull just that out of my chart, put it in. Like it's um it's much, much easier to sift through the massive amount of information that's in a data dump and find the pertinent things. Um then once you've done that process once, like as long as you have like an audit trailer and you know what you did, you don't have to do it every time.

SPEAKER_02 3:42

Yeah, yeah.

SPEAKER_01 3:43

You can kind of create uh you know a living health record. Um dynamic and follows you.

SPEAKER_00 3:51

I I I know actually, patients have started doing this. I uh yeah, a lot of them. It's not just in this case. I've also heard of patients who have been um trying to use insurance claims, putting on Chat GPT, and there was recent news on trying to talk to their insurance as well. So a lot of these things are changing. Yeah. So let's get to your um venture now, uh, Warp Ventures and FICAP fund, uh, your role at 5CAP fund uh and you're a founder at Warp Ventures. Can you tell a little bit about that?

SPEAKER_01 4:30

Yeah, so I've you know, like I said, I've I've always had an interest in technology and how it can make things more efficient. And one of the most inefficient things I I've seen throughout my career is you know uh something as simple as clinical research. Uh, you know, you would have the EMR, and this is 20 years ago now, but you would have the EMR full of all sorts of data. And then if you wanted to study the effect of pick something hypertension, high blood pressure on you know mechanical circulatory support, you would have an Excel sheet with a list of all the patients that had mechanical circulatory support, and then you would literally open up the charts with IRB approval and all the you know research consents, and then manually go in there, try and find okay, like let me let me list blood pressures for the last 10 days and then put them into an Excel sheet and you know, like row, like row by row, column by column, cell by cell, manually transpose data. And so one of the earliest interests I had was was going back to data and data integration. Like this stuff exists, you know, as bits, as what ones and zeros in a computer system. Like there's got to be a way to find it and move it uh without losing context to make things more accurate, more repeatable, more efficient. So um, you know, early on that meant building relational databases and then using HL7 to try and move data around, and there you have to again solve the same problems normalization and deduplication and fuzzy matching and all this stuff. Um obviously the techniques are much, much more sophisticated now. So things evolved from let's manually copy to let's see if we can automate data transfer with you know SQL Server and relational databases and again all of this interoperability stuff. Um and then, you know, so 20, 25 years ago was still very manual. 15 years ago, you started to see analytics tools and dashboards pop up, whether it was within Epic or you know, uh data fed into data lakes and then Tableau and whatever, you would find reporting packages and software that would try and take the raw data and make it discernible. Maybe it's a population level. Here's 10,000 patients, and you can sort by diagnosis, or you can sort by age or character, whatever characteristic. Um so having worked within multiple hospital systems, Columbia University as a researcher, you know, Will Melinda as an attending children's hospital Los Angeles at the end of my training, the the problems and the tools were always way behind uh compared to what was happening outside of the health system with consumer technology, with you know uh um and so when I uh moved back here and started working for Northwell, um I thought kind of a little bit naively, if we if I had the dev team that I've worked with inside the hospital on the outside, you know, I could build tools, train models, whatever, much more efficiently without the red tape and the you know security committees and and all of the friction that slows down development within the system. Um and so I formed Bark Ventures initially as a venture studio, and the intent was to build and advise companies that were building. Um and we had a couple projects we worked on, uh, but the unintended consequence of that was um companies that were building were looking for money, and you know, you know, they didn't just want advice, they wanted financial support. And having worked on the biotech side for startups in the stem cell therapy space, I you know, I had gone through the FDA process for complex biologics for devices for catheters, so I had a pretty good handle on that. The newer rules, software as a medical device, were really an attempt to use the older device pathways for software, which didn't work well, but at least the mechanisms, the committees, and some of the you know, some of the submission process was familiar. And then having again built and trained and done these things, it wasn't wasn't too hard to understand and pick apart some of the companies. And so the venture studio transition from a focus on build and advise to a focus on advise and invest, and it's morphed into a investing syndicate. The fundamental concept is the same for the syndicate as it is for PICAP fund. PiCap Fund is a traditional venture capital fund that I'm on the investment committee for and a general partner for. But the premise is if you're building solutions for healthcare, the earlier you involve clinicians and involve your end users, the more successful you're gonna be. Because uh we see companies all the time that have technical teams that built the technical solution because they convinced themselves there was a clinical problem. And if they had talked to practitioners that were dealing with that clinical problem, they would have realized that either their market size was way off or their solution didn't fit an existing workflow and wasn't getting adopted. Like it doesn't matter. There's hundreds of ways that you can build a perfect technical solution that no one wants to use. And so whether it's you know, through the syndicator or through FICAP fund, we bring that lens to bear. Like, what are the operational challenges, the distribution challenges, the workflow challenges that are unique to medicine? And you know, again, going back to the beginning, like it's a very antiquated, not always efficient, you know, process. And so if you don't understand how CapEx works at a hospital and like what steps you need to go through to get your $150,000 device into the OR, you can underestimate how hard that is or how long it takes. Um same thing with the FDA, like we see plenty of companies that are like, oh, there's a predicate because this this there's there's a stent in the cardiovascular space that goes in coronaries already. So we're we'll just say we're at stent also. Well, if you're not the same material, if you're drug eluding versus not, like there's other factors that um without uh a certain amount of clinical knowledge or depth, you can you know raise a million dollars, give yourself 12 months of runaway, and then realize you're not a 510K, you're a de novo, you need a PMA, and you're gonna take five years to get there. That million dollars is not gonna get you anywhere near done. There's you know, all of those sorts of things go into it.

SPEAKER_00 11:22

So yeah. Um, I know you also mentioned in the beginning the investment into startups have got harder now. Um, especially this year, probably you were talking about, I guess. Um, anything specific you have with regard, like why do you I I I thought last year was great for AI startups, basically. So is this year harder?

SPEAKER_01 11:49

AI in general, yeah. There's a lot of froth, there's a lot of excitement. There's probably what green, you know, Greenspan called irrational exuberance or whatever. Like the you know, the people know that the tools are gonna get adopted and they're gonna change pretty much every every aspect of every vertical that they touch. Um, but they're it's too soon to know who's gonna own a space. And so um, you know, as the the different verticals mature, you're gonna you have the usual market dynamics. Someone's gonna take 60%, the next guy's gonna take 30, and the last one's gonna get the 10% leftover. So if you have 10 people building in the space, they're not all gonna survive. Are they gonna be acquired or are they gonna wither on the vine? That depends. Um, but that's just like the the general challenges. But because whoever is in that 60% is, you know, and some if it's a huge space or a huge market for 30%, like there's substantial upside there. So a lot of investment is going in there, you know, hoping to pick the horse that wins the derby. Um health tech is a little bit, at least from my perspective, a little bit different. It's harder to get the explosive growth. There's relatively fewer unicorns. The the power law and the J curve, I think it's much flatter. It takes much longer. Uh a lot of growth can be you know much closer to linear than exponential because the cell cycles are long, because of the regulatory challenges for a lot of reasons. So I think the mistake that was made early on was the same kind of irrational froth that went into some of these other spaces was applied to the medical space for digital therapeutics, for a lot of different things, without the realization that despite the technology being transformative, the you know the adoption curve wasn't going to allow that same, you know, you know, a million ARR to 10 million AR to 100 million ARR in you know six-month intervals. Like you just don't see that in healthcare the way you can with other technology.

SPEAKER_00 14:02

Um that's true.

SPEAKER_01 14:05

So people will pull back, and it doesn't mean that you can't raise money, you just the the focus had to be adjusted. There were some pretty high profile companies that went belly up because they raised too much too soon and couldn't grow up, you know, couldn't grow to meet the the metrics. Um, so I would say people are much more cautious there, you know, you can gain investment, but you have to show ROI, you have to show that you can move outcomes, you know, it's not just we've found a problem, but it's okay, but how did you solve it?

SPEAKER_00 14:41

Right, yeah, yeah, yeah. Yeah, um you're right, because AI has also made it easier to build now. So you will see a lot of similar uh applications out there now.

SPEAKER_01 14:56

Which so you I mean the technology has gotten commoditized, the build process has has gotten shorter. So, yes, you you're gonna see, you know, instead of two or three people building in a space, you're gonna see 10 groups. But the gate for who succeeds hasn't changed. You know, you the the technology becomes table stakes, finding you know a problem that's real, like you have to do that. But once you've done that, that's not enough. Like, okay, you just sat down at the table, you still have to solve distribution, the regulatory environment, you know, your go-to-market, changing, changing clinician behavior, driving adoption, like all of those things still need to happen. You might have 12 people trying to do them instead of three, but at the end of the day, like it's only going to be the one or two that win. So everyone else goes away.

unknown 15:48

Yeah, yeah.

SPEAKER_01 15:49

Now, the good thing is because it's cheaper to build and faster to build, those 10 people didn't each burn, you know, five million dollars of capital to get to that point. You know, maybe they did it in a garage, maybe you know they spent a couple hundred grand and then gave up. So yeah, yeah, yeah.

SPEAKER_00 16:06

I mean, there's good and the bad side to this now, right?

SPEAKER_01 16:10

Right. And this this is why FICAP exists and why Bar Ventures exists. If you aren't familiar with healthcare and you aren't familiar with clinical workflows and all the things that drive adoption, and then you look at that bucket of 10 companies, how do you decide who's gonna win? Like, how do you pick apart which one? But if you do understand things, you I think you have a better chance. It's by no means guaranteed, but maybe five of the ten could be eliminated on very common recurring modes of failure that once you learn how to spot them, okay. Um, but without that clinical insight or experience, you're not gonna spot those patterns. Maybe you outsource and you hire someone to do it for you, but it's much better to do that in-house, I think.

SPEAKER_00 16:58

Right. Yeah, there's actually a lot to talk about this startup world and uh investment. Um but before that, I I just wanted to talk also about what Northville did in terms of uh the recent pediatric heart transplant um program that opened in Long Island. I heard that was the first one to open up in Long Island. I I guess it it was done last year and a year ago. About a year ago. Um so what what does this mean for the people around there? Is is that uh getting access faster in in that region or was that not possible before?

SPEAKER_01 17:44

Or right, so um uh unfortunately the wait list times, the you know, how long it takes to get in in organ once we've uh met and decided that a patient is a good candidate for heart transplant and they go through the listing and evaluation process, um the time to get an organ doesn't change. It's still way too long because there's not enough donors and you know for a variety of reasons. Um but if you lived on Long Island, um, you know, almost half of the pediatric patients that need a heart transplant also need some form of mechanical circuitory support, uh artificial pump to keep them alive while they wait for the transplant, because it can be six, nine, even twelve months on the wait list, and without those pumps, they would die waiting. Um the overwhelming majority of patients that do get heart heart transplant are inpatient and have been inpatient for months. So if you're a family with three children, one of whom needs a heart transplant, before we opened our program, your options you had to basically live in New York City. Or send your child to New York City and then they would be alone in the hospital for six, nine, twelve months while you continue trying to put food on the table, take care of your other kids. So um what we've provided is access to a a very needed service to the residents of Long Island so that they don't have to, you know, completely upend their lives and families um while they wait uh you know for the chance of of getting a transplant.

SPEAKER_00 19:30

This is amazing. I I guess it's really helpful for people around there to get early access, right? Yes.

SPEAKER_01 19:43

Yeah, it's huge.

SPEAKER_00 19:46

So we actually I know we're getting I would love to talk about a lot of these things. Um I'm enjoying the discussion, but uh I guess we're getting close. So uh I am amazed by how you have connected technology, biology, and as an investor, everything together. Um just from a closing perspective, uh, from a big picture. Um you mentioned about some ideas about how medicine or healthcare can be fixed. Any any any of that you want to talk about?

SPEAKER_01 20:21

Yeah, I think there's a lot of uh fear. Um you know fear, doubt. Um, you know, there's there's a lot of reasons for it. You know, the the political divide and the kind of tribalism and everything that's happened over the past 15 years definitely hasn't helped the you know the the perceived lack of science uh and justification for some of the policy decisions that are made. Um also the you know the lack of transparency on the other side, you know, like there's a profound mistrust, I think, right now, of everything, you know, not just healthcare of politics, of the governmental bodies, and you know, a lot of it is justified, but it's created an environment now where you have a rapidly developing technology, and you've all you know heard and seen the it's the end of the world, AI is gonna take your job, the Terminator robots are gonna come and wipe us out, like all sorts of doom and gloom. Um and you know, fear is much easier to sell, to popularize, to get headlines than rationality or you know, goodwill. But I I honestly do believe that um there's a huge role for the responsible use of AI. Uh I I think first and foremost it's going to be patient empowerment, you know, and finding out how to cut through that distrust to Provide insights, to provide guidance that's personalized, that is delivered in a format that people are willing to receive, um, but also proves itself to be valuable. You know, it's uh it's gonna take time, but I I think if it's done correctly, it will drive adoption and it will you know make a lot of aspects of healthcare and of personal care and and um better, you know, and that's really what you want to see. So finding a way to utilize the technology humanely, rationally, um, that's the goal. And I think it can cut through a lot of the background noise and ideology that um doesn't help anybody.

SPEAKER_02 22:51

Right.

SPEAKER_01 22:52

So however, however we get there, that's that's kind of my my goal is is drive that positive change with technology without losing our humanity, um, in a way that's gonna increase longevity and increase quality of life. I mean that's that's what medicine set out to do.

SPEAKER_00 23:12

Right. Oh that's I mean, it's uh amazing how personalized medicine uh care can make a difference, especially with what we are doing right now. And like you said, uh in healthcare or in medicine, no two patients are the same, also, right? Yeah. Thank you so much. This I've learned a lot in this episode, and um thank you everyone for watching. Thank you.

SPEAKER_02 23:43

Thank you.