On this episode of Highmark’s Health Care Reinvented, learn how Highmark uses massive data sets to improve patient experience and outcomes while simultaneously managing data bias and the utmost security.
We welcome Ian Blunt, VP of Advanced Analytics at Highmark Health, where his group form a center of expertise for all things data science and sophisticated analytical techniques.
Leading a team of 100-plus data scientists, researchers and engineers, Blunt and his team are at the forefront of developing new capabilities for novel use cases, building predictive analytics and machine learning models, next best action selection, behavioral engineering, evaluation of interventions, analytics enablement, natural language processing, coordinating Highmark’s analytic product portfolio and transitioning analytics to the cloud.
At the end of the day, Blunt says all of the work comes down to directly improving the patient experience and proving health outcomes. Listen and learn how the human experience is expressed through the data!
All right, Audrey, we're gonna get just a little geeky today on healthcare reinvented because we have someone who is no stranger to the healthcare reinvented podcast rodeo, we have Ian Blunt hanging out with us today. Very excited because this guy, when it comes to all things, analytics, we want to talk to you. Because he always I think last time we talked to moderate our brains were like, ooh, like this.
And now it's great to do a deep dive. So we're really excited. So do you want to? So Ian, why don't you just tell people a little bit about your background, and why you're such an expert on the topic that we're going to be talking about?
Sure. I'm very excited to be here talking to you guys about it. So my name is Ian Blunt. I'm Vice President of advanced analytics for Highmark Health. So we form a center of expertise for all things data science and sophisticated analytic techniques within the organization, Highmark Health integrated health care organization. So we've got a health insurance arm, we've got a provider arm in the form of Allegheny Health Network, and we've got a few diversified businesses as well and my team kind of sit at the parent company and help each of the business units with the various use cases they've got to apply. And that's that I think techniques that every now and again, we did get to this really interesting thing, or blended health, where we're using information from both sides in a regulated way to improve the experience for all of our members and patients make it sound so simple, but healthcare is really complicated. That's one of the things that I think makes it such a rich territory for analytics. It's also been an industry that's kind of slow to adopt some of the analytic techniques, often for good reason, because of the magnitude of the decisions being made and security concerns around the data privacy, and what have you. But I think when you look at how high marks really trying to transform healthcare, it all comes down to I live in health strategy, which really is about making healthcare simple, making it personalized, and making it more proactive. And there's advanced analytics applications throughout that kind of mission statement.
So what are the big things that you're working on? So like, what keeps you up at night, so that, you know, sort of set the table for us so we can understand because sometimes people are listening to this? And they're like, huh, how does that, you know, how does that really translate into meaningful experience and meaningful outcomes?
There's so much going on. It's really exciting, again, kind of a factor of healthcare having been traditionally slow to adopt these techniques, and also the huge range of use cases that Highmark Health covers. So some really good examples, I think about how we can improve the efficiency but certain services work with so got a really nice technique that's operational within chemotherapy infusion facilities with within Allegheny Health Network. So traditionally, if you have cancer, new on chemotherapy drugs, you'll go to the infusion center, you'll sit in an infusion chair for variable length of time. And you'll do that several times over the course of your treatment. And traditionally about was just booked based on where the local administrators thought they could see a slot, what we are able to do is, first of all, build a model that predicts how long each infusion is actually going to take. So personalize it around the patient and the treatment that they're having. And then apply some optimization techniques. So we can optimize for schedule. That means we can fit each potential appointment in where there's a good slot for it. And one of the things we found is we were able to increase the utilization of each chair from about 55% of its time being occupied to bet 90% of its time being occupied, which is great, because these are saving for the provider, which means we can lower the cost of health care overall. And also, these are really valuable resources, potentially life saving. So we want to maximize the access to it as much as we can. There's also other fringe benefits. So when you're not, don't have that schedule optimized, some sessions over run, they just take longer than expected, which means the next patient, you know, getting frustrated there, whoever's, you know, often people using the service need a ride, so whoever's come to pick them up, is now hanging around, no need to be there and no idea what's going on. And also, when the whole system kind of comes a regular, you put a lot of pressure on your pharmacy that mix in these chemotherapy drugs, which of course, if they're under pressure, that increases the likelihood someone could make a mistake. So there's some potentially very serious outcomes there. So by smoothing that all out with providing a much more efficient use of the service which is lower cost with providing better outcomes, better experience for patients and providers, as well. So it's it's a real win for that kind of application of advanced analytic techniques.
That's an interesting use case and thinking. I mean, it seems like so niche, you look at all the things that goes into healthcare, but how important is that, I mean, the idea that you can get more people into the chairs, so that way, they're able to get the treatment in a faster way. And then the ripple effects, not like the safety of like a pharmacist mixing something and not being rushed to mix up things so that you're making sure they're getting the right, the right treatments. That to me is is amazing having problems come to you to say we need to solve this, obviously, there were some issues around that. So they came to your team. And they're like, Wow, we need to solve this. How can we do it? So how does that process work?
Yeah, so I'd say it's about 7030, the business coming to us with things they'd like to do, and 30% of us heading out with ideas solved. So I think also, the cool thing about the way we've implemented that chemotherapy infusion example is even though chemotherapy infusions niche, and we're going to scale to all of the infusion centers, and then that's kind of a limit there, you see that pattern repeated many times. So for example, operating rooms, again, variable length, important resource problems, if it was schedules, that's the sort of clog up and then dialysis appointments, even MRI. So there's lots of places where those same basic principles can be repeated, to achieve similar even better outcomes. And then, of course, the challenge we've got is well, how do we manage it all, there's far more demand for these services for analytic techniques than we can meet. So it's about a process of running backlogs of requests, prioritizing them based on value. Again, one of the things I really enjoy about working in healthcare is we talked about value in terms of stakeholder quintuple aim. So it's not just dollars and cents, although lowering costs is an important part of the going to play. But you've got how can we bet get better outcomes for the population, and you've got the clinical and patient experience very important factors. And just recently, the IHI who defined quintuple aim added increase in health equity to that as well. So that kind of changes nicely with a lot of the sort of community based organizations that Highmark works about works with, and how we think about equity and social needs alongside traditional healthcare needs.
What about the issues like around security? Right? Like, can you talk about that, there's so much data, people get so worried and intimidated. And yet they really don't have all the facts in terms of, of how that data lives and how that data is protected? Yeah.
Healthcare data, medical industry data is I think, most fiercely protected by the public. I've seen literature out there where people do surveys, and they are more interested in protecting privacy around health care than they are, for example, criminal records, or their tax records. So critically, critically important that we treat those data securely. And with respect, it's a huge privilege to get to work with this information. And part of executing on that privilege really is concepts within healthcare industry, we call minimum necessary. So we should only be sharing the minimum necessary data that we need to go and execute. And in my line of work, that typically means we don't need to see names and addresses, we need to see some of the demographic factors, we need to see the health history, but I don't need to know who that piece of information is associated with, I can do my job just as well, without knowing that. So that's another way that we keep it secure. And then the other challenge we've got is thinking about exchanging information within healthcare. So it's actually relatively a security people shout at me, it's relatively simple to create a completely walled garden where just nothing can get in, nothing can get out. And one of the things we need to do as a modern healthcare companies share information. So share with provider partners share with vendors provide healthcare solutions shared with all sorts of people. So it really is about being able to have those systems that facilitate rapid ability to move information to where it needs to be but also keep everything secure, to make sure that that information isn't going to go anywhere that it shouldn't be going and we've got a well matured data governance infrastructure that supports us in doing that and it comes all the more important is interoperability. Rose is a trend within the healthcare industry. We're not Talk in file transfers anymore. Now we've got people accessing through API's and other means we need to be able to adapt quickly, as needed. But also make sure we're doing it in a very measured, controlled and secure way.
I was really curious to talk more about like how you're able to eliminate bias. And this as well, too, that's a lot that I've been hearing is obviously keeping it secure. But then, as you're gathering this data to make sure that you're not getting bias out of this, you can actually make decisions and things based upon real clean data, I guess, for lack of a better term.
Yeah. So that's a fascinating subject. And I think one that in general, is going to come into the sort of AI and analytics industry, it's already come in. So if we think about data security is what we can and can't do. There's a lot of law there. And then there's also the question of what we should and shouldn't do. And that's where I think data ethics really comes into play. Highmark really wants to establish itself as a thought leader on within the industry, it's very important to us personally, that the Insight we're generating is being used in an ethical way. And that covers lots of different applications, particularly around model bias, which has been a very hot topic for the last couple of years. And we've put a number of systems and processes in place to help guide our thinking around that, and also working with a number of partners as well to inform what we do. So to be very kind of like, high level on it, when we're ever we're considering sort of an AI based use case. The first thing we do is we think about the use case itself. So what before we do any modeling tall, even thinking about the data? What are the business asking us to do? You know, how might the data or how might that action, introduce some sort of bias? Is that even the right thing to be doing, and we can drive a conversation within the business on that? Then the next one, once we've accepted the use case, and we're working on it typically involves some level of predictive model. Within a predictive model, you can find a list of influential features, basically, what data elements drive in the model outputs? And depending on which algorithm you're using, sometimes you have to use simple explainers to get at that, but it gives us a really good sense of what's going on there. And then we check that back with our business colleagues. So we literally go down the list, is there anything problematic in that, that we're worried about? Is there anything that's proxy and something that might be problematic, and then the next step is really important. So it's not the sense that if it's in there, and problematic, we automatically take it out, we need to think about the impact it's having. There are some times when actually it's really important. For example, prevalence of diabetes is different depending on people's ethnicity. So there's a really good argument for having ethnicity based information within a diabetes model. And again, it's about the actual output, what happens differently in the real world because of what the models say. And so we think through those various use cases, take information out rebuild the model if we feel we need to. And then we do a series before it's elevated to production, we do assess what's called back tests. So essentially, we look at model performance over a range of different categories and get a good sense of, you know, is it performing particularly well on one category, but particularly poorly on another, and we use scores in that as well. So it's kind of typical check on what the model is performing. And then the last thing we do is we monitor the model's performance over time. So we're checking for drift so we can understand and be alerted to if anything, when we launched it production was unbiased, has now seems to become biased, for whatever reason, and we can dig in to that. So that's, that's our system, we go through that. We're very proud of it. I've got to say, we've not had too many cases where anyone's had to raise their hand and say, I think something's not quite right here, which is great. But I think it's important that we're always vigilant and always developing around some of this stuff as well. And I think one of the important things come in, just in general, like the trends in the latest legislation around AI ethics, and what have you, I think are going in a really good direction. We used to train people used to try to regulate specific use cases. And I think AI is going to become so pervasive, that doesn't really make any sense. What we need to think about is harms. So what's the harm of something being biased if I'm Netflix recommending a movie, versus what's the harm of saying biased if I'm offering a health care solution, but it is going to have a demonstrable impact on people's lives and the degree of scrutiny that you're applying to each application needs to flex based on Basically the risks that you're incurring by operating in that space.
I know what I'm thinking. I'm wondering, what advice would you give to people who are in college or thinking about trying to have an impact, right? A lot of everyone's looking for? How do they have an impact? How do they make a difference? And, you know, people might not understand that, that this is another opportunity right there in terms of the work that you're doing, what guidance, what advice would you give people if they want to follow that kind of trajectory of career opportunities and in analytics.
Oh, I would definitely say pick something you're passionate about. Because that, that kind of gives you that whole side. So I was talking a little bit about the data ethics piece, I mean, a lot of being able to spot what what might be problematic in the use case, or in inferential variable comes from having data scientists that are really engaged in healthcare. They're interested in the policy, their strategy, as well as the technical solutions. And there's kind of two models in general of doing data science. So one is, you kind of keep all your data scientists locked up in a dark room somewhere. And you've got this analytic consultant figure that goes between them and the business gathering requirements. It's fine, it's very transactional. But what we think is one that's not very satisfying for the data scientists. And two, it doesn't really give the optimal solution. So the data scientist will excellently solve the problem that set them, but we really miss them being in the room, around the table with the business, so they can actually help the business iterate on what the real problem we're trying to solve here is, and at Highmark, we definitely practice that partnership model, seeing our data scientists at the table, understanding the strategy helps the business understand the data in the analytics as well. So they trust it more. And ultimately, I think gets us much better solutions. And, I think, really phenomenal implementation rates. So one thing you see quoted quite a lot is data science projects, only like 30 to 40% of them make it to production. At Highmark, that number is much closer to 80, or 90. So we're really getting value out of this stuff. And to me, it's all about that partnership model. Now, that does create some challenges for us, because now we're looking out for not only data scientists with top notch technical skills, but we also need people that are interested in healthcare and strategy and can do the communication piece as well. But I'm proud to say like we've hired that throughout. And we've got a really fantastic team here, data scientists and research that physicians and analysts and data engineers and software engineers, strategy analysts, and product owners, who can all operate in that way. And I think that's one of the reasons we're so successful in just bringing this huge number of use cases to bear to transform healthcare.
How many are in the team? Seems like you kind of quite the cruel obviously, that makes this happen. It's kind of exciting to lead a team like this.
Yeah, we do. So we sit within the larger enterprise and data organized hate enterprise and data analytics organization. We specialize, as I say, in the sort of advanced analytics enter the market. So your data science sophisticated techniques, there's about 100 of us within the advanced analytics group. And that covers everything from big data analytics, machine learning and AI and natural language processing, but also out into what we in healthcare called core ideas, more commonly in industry, next best action, basically working out what someone's needs are, and then what the best way to meet that need is and then pushing that message into a workflow system so it can be actioned straightaway. We also do think about how we package that call to action as well. So using nudge techniques, a be testing to make sure when that call to action is out there for the next best action, it's more likely to be acted upon. And then we also do analytic evaluation analytics. So one of the things really important to us, is once we've triggered an action, we understand the benefit that both Highmark and the member and patient got out of that interaction. And we create this learning system so that the idea is tracked, does more of what we know works, and less of what doesn't. And we can also use our evaluation techniques to understand well, how might we improve our interventions to so we can get the results that we need or continuously improve the successful one. And on top of that, we're also managing kind of holding the ring for all of our analytics partners with transition from on prem technology into Google Cloud.
What about Ian Blunt the man what is in do for fun? And what do you think about living on this side of the pond?
Oh, so that's been a great big adventure. We moved over here in 2015. My wife grew up in Pittsburgh, which is how I landed here. Pittsburgh has always come back. That seems to be the rule. And it's been wonderful. We, even before we have kids, we agreed that we wanted them to both grow up spending time in both sides of the Atlantic. So it's been wonderful. Pittsburgh's a great city. It really is so much here to do so much. So easy to get around the Yeah, obviously, America is going through some interesting times. And I'm glad to be here. And part of seeing those changes go.
I tell you what it's for so glad that you're here in Pittsburgh doing what you're doing, because this is the type of work that's really setting Pittsburgh apart. And to have that coming through. Highmark Health is just amazing. And every time we talk to you, Adrienne, I was like, Oh, my God, man, he gets to have a lot of fun, but he's got some serious responsibility on his shoulders.
Oh, it's an awesome job. But to your point, like, the install of the states is watching Pittsburgh and healthcare. There's a lots of thoughts, but obviously, the US healthcare system has a fair amount of challenges. And the solution, one of the places it's likely to be born is in Pittsburgh. So Pittsburgh really has a national focus within the healthcare industry on what's going on here right now. So it's a super exciting place to be.
We couldn't agree more, that's for sure. Yeah, it's so much fun talking to you. We're gonna have to have you back. Again, because we want it to be continually changes. There's always something new to update us with and so we really appreciate your time.
Anytime glad to thank you.
I love having these conversations because I get to hang out with you and for half an hour and learn about these things anyplace else. So too much fun. This has been Jonathan Kersting and this is Audrey Russo. Thank you so much. We're from the Pittsburgh tech Council. We love bringing you healthcare reinvented with iMovie
Transcribed by https://otter.ai