Jason Bahn and How AI and Machine learning are enabling early disease detection.
Jason Bahn, co-founder and chief medical officer of Prognos Health, joins Harry to talk about how machine learning is being used to dig into multi-sourced clinical diagnostic data to improve health by predicting disease early.
Harry Glorikian: Welcome to the Moneyball Medicine podcast, I’m your host Harry Glorikian. This series is all about the data-driven transformation of the healthcare and life sciences landscape. Each episode we dive deep through one-on-one interviews with leaders in the new cost-conscious value-based healthcare economy. We look at the challenges and opportunities they’re facing and their predictions for the years to come.
I’d like to welcome our next guest Dr. Jason Bahn, who’s the co-founder and chief medical officer of Prognos. He’s a family physician and serves as chief medical officer of Prognos. He’s regarded as a national expert in the applications of technology and medicine, a topic on which he speaks regularly at institutions and conferences, such as health 2.0, M-health, E-health collaborative and health data palooza.
He’s also done extensive strategy consulting with different companies including pharmaceutical companies and others. Welcome to the show Jason a pleasure to have you here.
Dr. Jason Bahn: Thank you.
Harry Glorikian: So, tell me a little bit about Prognos, maybe a little bit about its history and what you guys are really doing.
Dr, Jason Bahn: Sure, so we have an incredibly ambitious vision which is to eradicate disease. And you might look at that and say, alright well that seems pretty incredible how are you gonna get there. And that is you know it’s sort of like our 20-year vision out and we’ve got a mission which is to you know find and predict disease even earlier than it is today. So, when we started the company, probably seven or eight years ago you know it was -, we were trying to figure out how we could take data that was available in the system and use it for improving the lives of patients.
So, we looked at different data sets that were out there, we looked at you know claims data and prescription data. And what we really knew, I would really know from my years of practice was that sitting there in my office seeing you know 30 patients a day and ordering lots of blood tests on them and lab tests on them, and then going back and seeing. You know going back to my bat back to my desk and sitting there and looking at all of the test results that had been ordered from the prior days. I would sit there and I would make more relevant clinical decisions based on the lab test results that I was seeing that I did during the whole day of seeing patients.
So, we knew that you know this concept of lab data was really important. And in fact there’s some studies out there, they show that more than 70% of all clinical decisions are based on lab test results. It’s even more in areas like oncology and rare disease or it can be up to a hundred percent. So, we figured that we would start to work with lab data, and that was not easy because the lab system is fragmented. There’s something like five or six thousand labs in the United States alone but many of them, but there’s a you know that the top probably thousand handle most of the lab testing in the US, aside from acute care settings.
And so we started partnering with labs like LabCorp and Quest and working with them and helping them with their data, which was sort of the second problem is once you can start to aggregate all of that data, you have to clean and standardize all of that data. Clinical data is not easy to work with; it typically has a lot of unstructured aspects to it. So, we spend a lot of time just figuring out how to collect it and organize it, more than anything else. Once we were able to do that, we saw that there’s a tremendous amount of value in it, both in the pharmaceutical space and in the payer space where we offer products today.
So, while we aggregate and collect all of this data and standardize and clean it, we actually turn it into products. And those products are what services are for clients and the idea was that you know early on, we were able to find patients that had particular diseases and we were pretty good at doing that. And then you know the sort of the next phase that we moved into was to predict which patients were going to develop X Y or Z.
It could be which patients were going to fail a particular therapy, predict which patients were going to go on to a particular therapy, which patients were needed to be tested more regularly or didn’t need to be tested or we’re missing tests in order to clinch a diagnosis. So, as we move into the prediction phase which has been over the last couple years, where we’ve really beefed up our computational expertise and AI and engineering. We’ve kind of understood more about how to predict these events. And then the idea is once you can predict something, you have to figure out what are the bright points to intervene.
And once you can intervene before something happens, then you’ve potentially moved towards the potential of eradicating that disease. So, that’s sort of how we have our vision and how we’ve been moving towards it for the last couple of years.
Harry Glorikian: What have been some of the challenges that you’ve faced along the way to implement or make this a reality?
Dr. Jason Bahn: Yeah, there’s a lot of challenges paired in this space. One I talked about was the fragmentation of the data, the data set itself. Two was obviously organizing and making that data fit for purpose, but there are a lot of other things that were challenges, right. So, in the sort of the standard healthcare data world, there are claims data and prescription data. And both of those are fairly commoditized there are a couple of you know players who have organized and brought all of that data together.
So, it was pretty well known but you know clinical data is different. And you know one of the things that we faced was it was new and just working with labs was new and labs were you know, it wasn’t their business to be in data, their businesses running tests. And so anything new is you know foreign and guilty until proven innocent. So, we had a lot of work to do with just the labs in order to get them comfortable with the fact that we had all the safeguards in place for managing data.
All you know that we were compliant with HIPAA and that you know that they weren’t going to risk by providing all this, and that we were providing them something of value with what we were doing. So, that was a big hurdle, just in accessing this data and that took a long time just to get there. And then the other side of it is proving the value, right. So, everybody’s used to using something on the far end of things right, as pharmaceutical companies are used to using claims and prescriptions, and so are payers.
So, convincing them that lab data and this clinical data was good enough or better than what they were currently using either to augment what they were currently using or replace it altogether. So, that was another challenge and anytime you’re new to the market with something it’s always a challenge. So, that’s probably the biggest.
Harry Glorikian: So, where have you seen something that you weren’t expecting, and where do you see this having the biggest impacts?
Dr. Jason Bahn: Sure, so one of the one of the things that we didn’t expect was in the diversity of care that the patients receive. You know there are clinical guidelines that are published and doctors are supposed to follow clinical guidelines and patients are supposed to present themselves all the time and they’re supposed to do. So, that they can, the doctors can follow the clinical guidelines. But you know the real world is different and doctors practice differently and patients aren’t always as accessible as you would hope them to be, and even when they have major illnesses they don’t present themselves as often as they should for care.
So, one of the things that we looked at, one of the diseases that we looked at was CML or chronic myelogenous leukaemia, which is a type of blood cancer. And what we found was that this is actually a great disease for lab testing. Because there’s been a sort of a new, you know a test that’s been around for a little while that looks at a molecular marker and you’re able to track the course of the disease over time in the blood, which is you know that’s like the holy grail for cancer.
And so basically there’s this blood test and once you use it to diagnose the disease and then you use it to track the course of the disease. And so you’re supposed to continue to test you know once a quarter until the person is in remission.
So, that’s four tests a year and therapy has changed based on the results of the tests. If you are driving the tests you know they’re driving the presence of the mutation down, then you know you can stay on your therapy if it’s changing then it changes. And so what we found was that number one, patients on average we’re being tested like 1.5 times per year instead of four times per year.
And those patients who were tested more frequently were having better outcomes, and so what we were able to, we actually did the work with the professor who came up with helped write the guidelines.
And he was just as shocked as we were that this testing frequency was so low. And what we kind of found was that look, you test people more then there are better outcomes and you drive them more towards remission. And I think that was sort of a shocking thing to find, not that it occurs but that it occurred at such a high rate and was such a discrepancy from the clinical guidelines. And you could certainly make many arguments on that, you could say that you should be educating providers about the importance of testing and not under testing, but making sure you’re doing the appropriate amounts of testing.
You could educate patients with the disease on the importance of going to your providers and getting tested regularly. You can give heads up to payers on patients who aren’t getting tested as frequently, as they should and getting them higher and more engaged with the patient’s, so that the outcomes are better and the costs are lower. And you can work with pharmacy companies on educating providers, educating patients and figuring out how to even figure out how to pay for some of this testing, which you know I didn’t occur.
You’d almost want it all to happen, you do want it all to happen it’s you know ideally that’s what would happen. But all we did was find all the correlations and then pass out the information to folks and hope that they can power some of their resources towards.
Harry Glorikian: So, when you guys are looking at the data Sciences side, you know sure that in the beginning it was much more simple analytics. You know actually probably the majority of your time was cleaning and organizing just to get it useful, but now it’s now that it’s sort of in a better State let’s say or in a much more usable State. What are the challenges around you knowing hiring the right people, you know when you decide what sort of data analytics do you use? How complicated is it you know? Do you settle on a platform where you’re constantly evolving to keep up with this constant set of change?
Dr. Jason Bahn: Yeah, so you know there are a number of different questions in there. One is just finding the right talent, that is not easy. It does make it easier that we have a very large unique and interesting data set. And it makes it very interesting that we are in the healthcare space. So, the people that we tend to find are those who want a new challenge with lots of data and want to make a meaningful contribution to the world. So, we know, we have recruited people out of the medical space and usually those data scientists are the ones who were recruited by a hospital system or by a company that had great science, but no data.
And so they were kind of tired of not really doing anything and just you know kind of theorizing all the time. And then we recruited folks out of industries like Ad Tech where their mantra was you know the right Ad to the right person at the right time, where we were saying things like the right drug to the right person at the right time, and that’s very enticing to people. So, you know we made a hire a couple years ago and found our chief data scientist who came out of the Ad Tech space. And he’s been great, he’s a mathematician, peer scientist and loves the theory and you know and it keeps us all on our toes and pushes us to great new things.
He’s also recruited an amazing top-tier group of AI data scientists that help us do what we do. And you know they are, the way they work is it’s almost like they’re playing with toys, and there’s a new shiny one and they go and grab it. And that’s great because that’s the way you want to approach this space. You know something changes every week, there’s a new platform that comes out every week or month. And it may actually be better than the one you used before.
You know I think actually this week, we’re presenting alongside Amazon at the Jupiter con for being one of the biggest utilizers in the healthcare space of sage makers, which is their new platform that’s AI base that’s kind of going up against Google and Microsoft and others. So, you know we’re experimenting with new technologies all the time, and you know AI technology is really a commodity at this point. You can, as long as you have the data and you have it formatted in a way that it can be absorbed into a system, then you can use just about any application out there. And oftentimes multiple applications in order to get the right answer.
Harry Glorikian: So, you know we were chatting on the way up here and you mentioned one of the people that’s benefiting from the data that you’re giving them and how they are shifting from sort of looking at the world from an actuarial perspective to actually predicting. Can you walk the people you know someone through that, how that evolved over time?
Dr. Jason Bahn: Sure, so we know that with clinical data, especially lab data you can infer a tremendous amount of information about an individual. How sick are they? What comorbid conditions they have, whether they have a disease or not but also where they are in their disease state? So, from a payer’s perspective someone with diabetes is interesting, but someone with diabetes that’s poorly controlled who also has high blood pressure, high cholesterol is much more interesting from a number of different perspectives. One, because they’re more likely to get sicker and two because in the world of insurance especially government you know either Medicare Advantage or the ACA population or Medicaid, the payer actually gets reimbursed more to care for that individual.
So, one of the products that we have out there is, both a sort of an identification and predictive product for payers. And what we do is we help them identify in a population of patients either that is existing for them or a new population that’s coming in for them, where their risk is. And which patients are going to get sicker over the next 12 months and which ones are going to cost them more money or have more disease burden. And they’re using that for two things, one is to direct resources towards those patients. So, that they can either impact their cost before it gets out of control and improve their health.
And second in the reimbursement from the government, because the more complex a patient is and the sooner you know about that complexity, the more money you can recover in order to take care of that patient. So, that’s one and then the other is really around where you know predicting costs and traditional methods are in actuarial tables, right. Looking at demographics, slopes of an individual, where they live, what their age range is, what their occupation is sometimes, zip codes and other things. And then using that and sort of the law of large numbers and predicting what sort of bucket of cost that they’ll fall into over the coming 12 months.
And what we’ve discovered is that, if you take lab tests and cost and look at that retrospectively and build, and let the Machine sort of go at that, that matrix of data you can then, based on lab data alone you can predict the future 12 months of cost or disease burden that’s coming down from that patient population. And then what would you do with that, well you could do anything from directing resources towards it to correctly predicting your cost for that group of that population of patients and pleasing your investors as well as your bottom-line.
Harry Glorikian: So, where do you see this capability going in the future? Do you add other data sources to it that really changed the paradigm, like instead of just looking at lab tests you take on wearable data, so you monitor people in between. Where do you see the future going and what you guys have built and where would you like to see it be?
Dr Jason Bahn: Yeah, that’s a great question. I mean, I think the Holy Grail is as much data on as many people as you can get. Health is so multifactorial, there are so many permutations of why a person goes down a particular path. That I think unless you have millions and millions of patients with millions of data points, we’re really not going to understand what and why. Even the predictions that we’re making now are inaccurate, because of that. We’re more accurate than the old ways, but we’re still inaccurate because of all the different things that can go into a person’s health.
And to your point, I think adding more data sets is a great way of improving that, right. There is wearable data, there’s that sort of whole healthcare data layer of information that’s collected on patients either through passive or active sensors. The challenge with that is they’re not mainstream yet, I mean the people who are using fit bits are the ones who probably don’t eat it as much they’re generally healthy and walking and other things. So, it’ll be great when you know you know Apple kind of gets a critical mass of users using their systems, more people are using fit bit’s and so on and so forth.
But that’s a layer, I think, so seeing like the spending behaviors of people is really impressive. I mean you know location information at any given time, you know if you could imagine you see someone hop from McDonald’s to McDonald’s on a day-to-day basis. And then correlate that with the amount of money they’re spending there and then with their lab test results and their claims history, I mean that’s incredibly powerful. And then I think adding in genetics data to that once we sort of know what to do with full genome sequencing, is a really powerful addition to the set.
So, you know, continuing to add data and add data, and ideally the cost of computing all of that and continues to drop, so that it doesn’t become prohibitive. Because that is right now even so sort of an issue, it is sort of the cost of crunching all this data. And then where does it go, I think ultimately it goes to the individual. I believe that you know strongly that health care reform in the last ten years has been about empowering individuals, educating them, driving cost down and improving care and improving access. But I do believe that empowering people with the information will help drive change.
And we do that now through Pharma and payers, but ultimately I think you know you drive it to providers and give them the information that they need and ultimately down to patients. And give them the information in a format they can consume and with recommendations that make sense. And I think that’s when you really start to drive like the disease curve and the cost curve down.
Harry Glorikian: So, what do you see is the next set of hurdles, either for you guys or for companies like you to sort of move the needle on the, what you guys are trying to do?
Dr. Jason Bahn: Yeah, I think you know the sort of the four-letter word and in the industry is interoperability. There is a lot of data available on a lot of people out there from a healthcare standpoint. Unfortunately, it’s in a lot of silos and those silos are, they’re not just from a technical standpoint but they’re from a process standpoint and just a business standpoint. If you can imagine if you’re an electronic health record company, who makes your money on you knowing your subscriptions to your EHR, and the server that sits in the doctor’s office. And the doctor says you know, hey I want to port my data out of this to somewhere else. What’s my incentive to do that as an EHR?
I’m just gonna lose that doctor as a client, because they’re gonna go to the next EHR that’s just as easy. So, you know and it’s certainly not like a judgment on these folks, these are all businesses, but there’s no incentive to share data. So, the government’s been working on that for years and has really yet been able to, as have yet been able to incentivize all this or create a standard that the industry can use. So, while the Holy Grail is collecting all of the data on all of the people. I still see we’re a long way off from that, so we’re just kind of attacking it in a piecemeal way.
Harry Glorikian: So, that begs the question of like, Apple making EHR portable on its platform. It’s not everything in the record, but is that a bridge are they disrupting the interoperability of these systems and sort of almost usurping, the players not wanting to play.
Dr. Jason Bahn: Yeah I hope so, now they’re still using the standard which is called fire and so all the EHRs need to play nicely with fire, which they may or may not. And in all honesty the most valuable piece of information, the most immediately valuable piece of information coming out of the EHR as biometrics. It’s like blood pressure, height, weight and some basic social information. So, that I think the Apple software will be able to pull relatively straightforward, and you’ll still have some issues with it. But it’s a lot easier than saying trying to pull out a note and deciding, whether I met Michigan or myocardial infarction.
So, there’s a lot of work that has to go into that, but I think there’s some immediate wins that can come out of apples play here and make some data available to the system which has not really been at scale available.
Harry Glorikian: Yeah, and I mean that I think the next version is that the user will be able to share that data with an app that they could want to share with.
Dr. Jason Bahn: Yeah.
Harry Glorikian: So, if Prognos had an app available that could interact, that would be another way to have a data ingestion engine with a standardized set of data.
Dr. Jason Bahn: Yeah and I presume that Apple will probably allow anonymized versions of that data so that consumers will be able to consent and non-anonymize versions of their data to go out without someone having an app that’ll be available. And I think that’s actually, that’s how we function. Our registry which you know has 18 billion records on 180 million patients, is actually all de-identified. But that’s our training set, that’s where we build all of our algorithms from. And then once you have an algorithm built you know, it’s all these little fragments and pieces of patient journeys.
Once you have a person come in as an individual, you then map them to whatever point along the journey that they are and then you can give them their individualized ideas, so that is, right. So, if Apple would make or if anyone makes de-identified data available more broadly, then you can use that to create those algorithms and then create the app that an individual would then get mapped against and it would tell them what their health is or what they’re you know predicting what their future looks like.
Harry Glorikian: So, it sounds like a very promising future. I know the majority of your clients are you know insurers and pharmaceutical companies, but it sounds like you guys are slowly moving towards you know eventually getting to the patient.
Dr. Jason Bahn: Yeah, I mean ultimately that’s I do believe that that’s where the biggest impact will be made. And interacting with patients is very different than interacting with Pharma or healthcare systems. You have to have a very tailored approach to that. And you know as a physician I definitely know how to work with patients and how to get them to do what you’re trying to get them to do in order to improve their health. But it’s tough, I mean there are entire companies out there that are focused solely on how do you get a patient to engage with their own health. And that’s not an area of expertise for us, but we will certainly piggyback on that and power whatever we can to help them figure out how to get those patients engaged.
Harry Glorikian: So, it sounds like you know maybe a partnership within Amazon or a Google, that has a tremendous level of data on the consumer and what drives different behaviors might make sense for an organization like yours.
Dr. Jason Bahn: Yeah absolutely, I think that does make a lot of sense and looking and you know those guys do have great understanding of consumer behavior. So, what if you were to add in a layer of medical information or clinical data on top of that, what could you understand or predict or influence. I think what you want to do at the end of the day is, how do you get that person to not walk into McDonald’s?
Harry Glorikian: Great. Is there anything else that I didn’t ask that you think would be critical for people to hear about you, the company or where you think the space is going?
Dr. Jason Bahn: Look these are exciting times and it is the AI and healthcare space is evolving, so quickly. I do get a little concerned when I see you know these small startups with no real business models. Because we do need businesses that are able to sustain themselves and you know have models that allow them to you know companies accompany them, generate revenue and that will last. I’ve been in the health 2, 0 and health tech space for a long time now, and I just see too many companies start and fail.
So, you know spending time on a really good business model, figuring out how to take incremental steps towards solving problems instead of like trying to just leapfrog, now if you’re, you know if you’ve just sold your startup and you’ve got money to burn. And that’s great, go solve the biggest problems but not everybody has that opportunity. So, it’s really about you know we would all love for the healthcare system to free all its data and for everything to flow and work perfectly together.
But it doesn’t work that way, and I think you know the more companies that are taking bite-sized chunks out of it, towards moving us all towards that solution and then cooperating and collaborating between them. I think that’s sort of the way that the thing the industry will go, will succeed.
Harry Glorikian: Yeah, I think of it like when I know one person makes a scale the other one makes the blood pressure cuff in it. But they have API’s that allow, so one app to sort of aggregate that data into one place. And so there seems to be a free flow of data if you’re on the, keep yourself healthy side. It’s once you cross over into the well you are sort of sick or in the healthcare system that everything gets locked down.
Dr. Jason Bahn: Locked down and also, unfortunately the people who are sick and unhealthy are not the ones who are big consumers of the wearables and the scales and all of this. And that’s really the challenge, when we get good at passive data collection, I think we will have a kind of a breakthrough on data on the on the sick population, not just the well population.
Harry Glorikian: Well on that note, I want to thank you very much for joining today. Hope everybody listening enjoyed it and look forward to continuing the conversation in the future.
Dr. Jason Bahn: Yeah great, thanks for the opportunity.
Harry Glorikian: And that’s it for this episode. I hope you enjoyed the insights and discussion. For more information, please feel free to go to www.gloryccamp.com. Hope you join us next time, until then farewell.
How to rate MoneyBall Medicine on iTunes with an iPhone, iPad, or iPod touch:
- Launch the “Podcasts” app on your device. If you can’t find this app, swipe all the way to the left on your home screen until you’re on the Search page. Tap the search field at the top and type in “Podcasts.” Apple’s Podcasts app should show up in the search results.
- Tap the Podcasts app icon, and after it opens, tap the Search field at the top, or the little magnifying glass icon in the lower right corner.
- Type MoneyBall Medicine into the search field and press the Search button.
- In the search results, click on the MoneyBall Medicine logo.
- On the next page, scroll down until you see the Ratings & Reviews section. Below that you’ll see five purple stars.
- Tap the stars to rate the show.
- Scroll down a little farther. You’ll see a purple link saying “Write a Review.”
- On the next screen, you’ll see the stars again. You can tap them to leave a rating, if you haven’t already.
- In the Title field, type a summary for your review.
- In the Review field, type your review.
- When you’re finished, click Send.
- That’s it, you’re done. Thanks!