Utilizing technologies and resources to create value in healthcare
Recommended Book:
Black Holes and Warped Spacetime
Best Way to Contact Chris:
Christopher.Steel@iqvia.com
Mentioned Link:
Domain Expertise is Critical to Derive A.I.’s Power with Christopher Steel, Senior Director, Artificial Intelligence and Machine Learning at IQVIA was automatically transcribed by Sonix with the latest audio-to-text algorithms. This transcript may contain errors. Sonix is the best way to convert your audio to text in 2019.
Welcome to the Outcomes Rocket podcast, where we inspire collaborative thinking, improved outcomes, and business success with today’s most successful and inspiring healthcare leaders and influencers. And now your host, Saul Marquez.
Saul Marquez:
Welcome back to the podcast. Today, I have the privilege of hosting Chris Steele. He’s a Senior Director of Artificial Intelligence and Machine Learning at IQVIA. His passion is providing technology, vision and strategy to the business and leveraging A.I. machine learning, big data, cloud, virtualization, and DevOps to quickly and effectively implement solutions that deliver business value. We’re in the days of value for service, although a lot of things are still fee for service. Value for value… fee for value is something that a lot of providers are focused on and even companies. And bringing these insights and value is what Chris does best. He’s recognized for bringing insight, enthusiasm and quantifiable results to business opportunities with a make it happen attitude to transform and scale organizations. In his past life, he’s he’s done roles such as running his own consulting firm. He is the CEO and Founder of FortMoon Consulting, co author of Core Security Patterns in 2005 and a little known, but he was actually the Chief Architect of Pay that govt. So definitely a dynamic leader in our space and really, really excited to to jump into his experience, perspectives and and the work that he’s currently doing and at IQVIA. So, Chris, without further ado, I just want to give you a warm welcome, my friend.
Chris Steel:
Thank you Saul. I’m really excited to be here today.
Saul Marquez:
Yeah. Excited that you’re here. So did I miss anything in the intro that you want to share with the listeners?
Chris Steel:
No, I think that covered it pretty well.
Saul Marquez:
Good. Good. Well, you know, obviously, as well as you could summarize your deep experience and career, couple bullet points that we had there. But, you know, I’m curious, what is it that got you into the medical sector?
Chris Steel:
That’s an interesting question. As you noted in my background, I had a wide variety of jobs in different sectors. About five years ago, I was lucky enough to work with a company that software AG had acquired that was building out a machine learning platform. I got to know the beta very well and really got me reinterested in AI machine learning. I had looked at it back in the 90’s, really hadn’t been able to do much with it back then, but saw how much has come along and really got learning. It took a bunch of online courses that were recommended to me that a lot of reading and playing around and was interested to start to put my skills to use. And when you look at machine learning and artificial intelligence, there’s just so much opportunity there. And I wanted to do something meaningful with that. I think five years ago, a lot of it was being done in retail advertising. You think of some of the big tech giants, how they really got the efforts launched with the Netflix challenge on Google, etc. But I wanted to do something a little bit more profound with it, something that would impact people’s lives. And coincidentally, it happened that a friend of mine that I’d worked with a couple times before my past ended up at a IQVIA. He had saw the updates on my LinkedIn page and reached out to me. So it was good timing all around.
Saul Marquez:
Well, that’s awesome. How it worked out. You went for the for the meaningful for out and and here you are. And it’s amazing how how fast time flies. But you’ve been in the business for a bit now. What would you say is a hot topic that needs to be on health leaders agendas today? And how are you and the folks over there approaching it?
Chris Steel:
Yeah, I guess it’s no surprise that I believe artificial intelligence and machine learning are the big hot topics in the medical field today. If you talk to anybody, they’re either working on it, looking at it or worried about it. And at IQVIA, we’ve actually been doing it for quite some time now. So I think we’ve only as an industry started to scratch the surface in what can be done there and how it can be done. I think one of the things that a lot of people are struggling with is how to bring value using A.I. and machine learning. There is a ton of things that can be done, but how do we use it to really increase our value? And at IQVIA, what we’re doing is we’re making sure that before we jump into a lot of these different endeavors, that we have proper business cases. In other words, what are the results for patients, for our clients? What kind of value are we bringing to the table through AI machine learning? Because we know there’s a lot we can do, but we have limited resources. So we want to spend those resources the most efficient way possible. I’m open and taking a step back and understanding where the big bang is today. What can the algorithms achieve realistically? We don’t want to be pioneering too much into new areas where we don’t see value. So I think that’s a very big topic and a lot of different healthcare providers are looking at.
Saul Marquez:
Yeah. You know, Chris, I love that you went there because value and value creation is definitely forefront on the minds of a lot of health leaders today. And A.I. is one of those pathways that people are taking to get there. Give us an example of what you guys have done and what value you’ve created through the use of A.I. to add value?
Chris Steel:
So when I first joined we there had been a slide that was put together that showed that we had around 30 some different AI amount related projects around the company. My first task was to go out and do a thorough inventory. So back about six months after I joined, I completed that inventory and it turned out that we had at that point eighty three AI amount related projects. In the meantime, we’ve now grown to well over a hundred. So it’s hard to really capture all of the different things we’re doing.
Saul Marquez:
Oh yeah.
Chris Steel:
It’s spans the gamut of early disease detection, old timer’s prediction, ton of stuff on the commercial side. So bringing value to our clients through things like next best customer looking at how do we make our data more efficient. So in the past, we spent a lot of time imputing the data. So when you have data sets with missing data or unclaimed data, we had statisticians going through and cleaning up. We now are building machine learning models that can do that. We’ve done a lot of stuff in terms of pioneering type work. We have a researcher that looked at using a general to adversarial network again to take for rare disease detection. One of the major issues is there’s not a lot of data. So we built this scan to go and look and create these synthetic data sets that allow you to run machine learning models on it to do the prediction. So building up the actual training sites using artificial machine learning in the first place. A lot of really good stuff going on.
Saul Marquez:
Now for sure. A lot of projects. And and so guys focus on data science and sort of hone in on human science. Can you talk a little bit about that distinction and and help the listeners understand sort of the vision there that you guys have?
Chris Steel:
Yeah, absolutely. Human data science. That is our tagline. And I think the distinction is, is that it is, it is patient. It’s client focused. It’s putting humans in the loop as the recipients of it. So I think on a lot of organizations, you have researchers that are somewhat displaced and a lot of what they’re doing, they can’t directly translate back into the impact on humans. But the human data science is really starting from the beginning with understanding how we’re going to impact people, how we’re going to improve outcomes and using that as the focus of our efforts so that we don’t end up burning down a lot of rat holes that don’t lead to to something significant that’s going to benefit people that we we spend the due diligence to, as I said before, analyzed the business case to make sure that what we’re doing does have a positive impact.
Saul Marquez:
I think that’s a great call. Know on a love that you guys lead with the human because data science alone without the deep knowledge of domain expertise, really won’t yield much value. So I think it’s awesome that you guys are your core focused on that. So as you guys have added value, you work with providers, payers, life sciences, really all of the stakeholders in health care. So I find it fascinating that a lot of these folks are taking advantage of the platform and the things that you guys offer. But as you guys have evolved, what’s been something that has been a setback, that has created some major learnings and made you guys better?
Chris Steel:
Great question. I think it’s important to start with the fact that over the past several years, we’ve really done a good job of adopting the wean methodology. So if you’re not familiar with wean, very similar to Agile, but really embrace the fail fast motion, so rather than attempt large projects and go forward to to start small, really concentrate on MVP – minimum viable product. Flush out the riskiest areas first. So the biggest unknowns attack those and be ready to fail. Be ready to say, OK, we’re not ready to undertake this shot. The technology is not there or the expertise is not there. So quite often we do fail, but we fail in this and manageable way that doesn’t impact our clients or our external customers. Now, with that said, in terms of failures, I’ve seen my own, obviously. One of those was a recent prototype that I’ve been working with on our with our H.R. department. So H.R. had come to me and they really they’ve been sort of stagnant for the past five years in terms of their analytics and how to how to be able to recognize and pre-empt attrition, etc.. So they were wondering if a memo could play a part in that. And I got involved and we wanted to I wanted to get something done quickly for them. And so I went out and I talked with different data engineers, tried to understand, you know, what type of H.R. related data we had out there, and then pulled that in to see what kind of analysis we could do. Unfortunately, a lot of that data, of course, is privacy related. So we had the option. We could work with our privacy analytics team and get all of that be identified, et cetera. But I made that decision to just sort of move forward without any of that type of data. And really what happened was because we we chose to ignore that particular type of data. We ended up with a very slim data set, would not a lot of different features to go off. So even though I built several different models, we played around with that a lot. We just we weren’t getting predictions that were really worthwhile or usable. And so that was a failure there. But it’s OK because that was a quick non billable type project. The lesson learned is you can’t you’re not going to be able to build relevant models without the Right. data. And now we know that. Now we realize and they’re going back and they’re assessing what we want to do from here. Do we want to commit to actually building this out? They now understand the process better. They understand what’s required of the data that they’d be providing to really give truly leverage a good model. So I think in that regard, it was a really good experience. But the lesson learned for me is you can’t skip on the data without the right amount of data on the right quality of data, you’re not going to be able to build worthwhile models.
Saul Marquez:
Yeah, I think that’s a great call. Chris, and thank you for sharing that. So, you know, it makes me wonder and something to note for everybody listening, when you think about the promise of A.I. and machine learning to your business, to your patients, you know, there’s there’s really the the exterior factors delivering service to their own care. And then there’s the internal factors like the example that Chris just mentioned and improving your own processes and way of doing business. What percentage of the work that you guys do, Chris, is care and and service oriented vs. back of the house types of process improvement?
Chris Steel:
Well, I think that’s a really good question. IQVIA, as you know, is a 10 billion company. We have a lot of different silos. I haven’t began to scratch the surface of all the different groups out there. I think ideally the mix would probably be somewhere around 70, 30. So I think that we want to improve, you know, that that operational ratio where we’re always looking to reduce the OpEx, etc. And we’ve made significant investments over the past couple years to really build up it internally. First of all, we have one of the largest data lakes in the world today. All right. Yeah. So we think we’re a top three and that that includes all different industries. So that’s it’s going up against Google and everyone. And yeah, it is. But if you think about how long we’ve been collecting and selling data and really working with it, it makes it makes perfect sense. So in the past where there’s a lot of that data has been siloed. We’ve now been spending the money to put together, you know, this very large data lake that gives us access to all the different data in a structured way so that the right people with the right access are able to easily get access to that data.
Saul Marquez:
Yeah, that’s interesting. And that’s fine. I know you guys are huge and I just didn’t understand kind of the breakdown, but that gives us, I think, a big idea. I appreciate that. That side, rather, that we just tack. But tell us about one of your proudest moments, Chris in healthcare.
Chris Steel:
Probably my proudest moment was last summer. I was asked to bring on an intern. And at the last minute it sort of ended up being two interns. And I hadn’t quite really thought about what it is I wanted to do with them. And so it was a busy time of the year. They sort of ended up on my doorstep last May. And I needed to sort of quickly come up with a roadmap of what it was that we wanted to accomplish as a team, what they could do to increase or their personal growth, etc. So I had a little earlier in the year had started playing with this notion that maybe we can use one of these generative adversarial networks, as Ganz that I mentioned before. Yes. To actually create synthetic datasets out of structured data. So whereas at the time the vast majority of people out there who use was all around creating images, Right. each in the data to paint pictures and do things with images. And I had this idea that, hey, if they can learn to understand the relationships and images, why couldn’t it do the same for structured data? So if you take an example like a diabetes data, you’ve got your blood pressure, your BMI, all these different things. If you believe that there are some underlying equation amongst those those features that tells us what the outcome is, whether or not somebody has diabetes or not. Then you have to think that maybe the machine can inherently learn that relationship. And if it can do that, then what it can do is it can create synthetic data, data that is statistically similar to your original dataset, but different. So the generator in this case say these gams are built from you know, there’s two different models, ones a generator and the other is a discriminator. And the generator, it never sees the original data. So it just starts throwing stuff out. The discriminator, starts looking at it and trying to decide, is this real or fake? And then you get in this battle until the generator learns to generate data to the degree that it fools the discriminator. So now that we have that ability, if we use that unstructured data, we can create these synthetic data sets that don’t have privacy concerns. Right. They don’t require imputation. They’re clean to begin with. And also, to a degree, it gets you around proprietary licensing. So if you think of all of our clients, all the data we have, that’s license. So it’s proprietary that we’re not allowed to sell externally. How much of that could we could we use to create synthetic datasets that we could then turn around and use for other machine learning models? Now, obviously, that’s that one piece is something that we’ve been struggling with legally. It goes beyond the legal boundaries into things like client relationships. We can’t just go out and create a synthetic data, a super proprietary one, and not not tell our customers and not work with us, because that would be unethical. But it does at least give us an avenue. And then a lot of cases what we’re finding, people want us to do that. So ask anybody who does machine learning, who brings in vendors, etc. that want them to do prototypes. One of the hardest things with these vendors are always asking for data. They give us some of your real world data and we’ll show you, you know, what we can do. Well, that’s a very difficult proposition for a lot of companies, especially ours. Their ability to create these synthetic datasets allows us to go and give vendors and the open source community different competitions, etc. Give that data out there, which is a huge boon. So to get back to your original question, this was all sort of a concept in my head. I said these two them turns down on that task and was surprised when they came back with a working prototype in under a month. And. Yes, from there we. We were able to go on and work out a lot of the kinks. Make it a lot more generic, basically more or less production wise. And ended up getting a patent on it. Who’s just accepted recently? So overall, yeah. It was it was a really proud moment. I was really, really impressed with those guys. The machine learning modeller was he has just completed his sophomore year at Duke. So he was very young. And my my other intern was a statistician out of Purdue, but both of them were great. And it was really it was really proud to be able to do something innovative that can really have business impact and get it done over the course of a summer with just two interns.
Saul Marquez:
That’s awesome. And it’s a testament to Chris that if you give people amazing things to do, they’ll do amazing things.
Chris Steel:
Absolutely. I couldn’t agree with that more.
Saul Marquez:
So really, the promise of having a synthetic data set like this is that it’s accessible and you get around the worries of licensing and things like that. Right.. So it makes it makes it easier to innovate.
Chris Steel:
Right. One of the keys to machine learning is having the relevant data. I think I’ve made that point a couple times already today,.
Saul Marquez:
But I’m glad you’re hitting that cow bell though, because it matters.
Chris Steel:
It does. So having the ability to replicate and clean that data and then be able to provide it to the community as a whole to further research in healthcare, that’s a big step because we we saw and we believe that a lot of the next real breakthroughs are going to come from at home data scientists, people playing around in cattle competitions, etc. Now that the work they do is going to be breakthrough that we’re going to be able to utilize to significantly improve health outcomes for everyone.
Saul Marquez:
Fascinating. Very cool. And for those of you that don’t know Kaggle, Kaggle is a site and where the brains of A.I. and deep learning machine learning go and they just compete, right. and they write code. And based off of the work that you’ve done and the quality you’ve delivered, you get stack ranked. Right. Chris.
Chris Steel:
Yes, correct.
Saul Marquez:
It’s fascinating. I learned about Kaggle probably about five years ago. So if you’re looking for the stars, you know, these folks don’t need a resume. And oftentimes companies will go straight to Kaggle. Kaggle is the resumé, Right. Chris,.
Chris Steel:
That is the resume. I hate to say, how many people are really top scientists that we’ve hired recently just based on their work in Kaggle alone? It is. You don’t need a resume. It speaks to exactly how good you are, how motivated you are, and how focused on real world outcomes you are.
Saul Marquez:
It’s impressive. It’s so cool that you guys are doing that. And yes, folks, if you haven’t learned about it, check it out. They’re looking for talent. This is how you get it. It’s not gonna be cheap. They’re high on the list. FYI. So tell us a little bit about an exciting project you’re working on today.
Chris Steel:
There’s quite a quite a few that are going on. I already mentioned some of the H.R. prototyping work that I’m doing. We continue to accelerate the synthetic data generation stuff. So we’ve been working on that across the company. There’s just so much really good stuff going on. I find it hard to start. I guess maybe what I’ll do is I’ll sort of defer and talk about a side project that I’ve been working on at night looking to to use that generative adversarial networks to create music. So I think if you’ve been on therapy, know what again is you know how how good it is at doing things with images, looking at using that music. How can we take a playlist and be able to train the model on that and get it to create music that fits that genre that the particular band, etc.? One of the things I think that’s really interesting is in the next five years or so, look for look for these synthetic bands coming out. So look for, say, your your favorite rock artist. That’s that’s now dead. Buddy Holly or somebody look for new songs coming out that sound like him. It’s a style. It’s his voice. It’s the same kind of music is that. That’s…
Saul Marquez:
So synthetic bands are a thing?
Chris Steel:
I wouldn’t say a thing right now. They’re becoming some at the video conference a couple weeks ago, I was talking with two guys that had done a lot. I’d listen to their generated music. It was a punk band. But while I couldn’t tell the difference, I honestly I wouldn’t have been able to tell the real from the fake there. So I think. Maybe it’s not a thing today, but it will be soon.
Saul Marquez:
Huh? That is so fascinating, man. And in many synthetic bands. Amazing. I mean, if you think about it, if you’re able to produce synthetic biology, synthetic data sets in, you’re in your case, why wouldn’t you be able to create synthetic music?
Chris Steel:
Right. Right. Yeah. I mean, there’s there’s synthetic videos out there. You can modify a video in real time. You can automatically take people out of pictures and insert different people into them. In some regards, it’s a bit scary because.
Saul Marquez:
It is.
Chris Steel:
You know, with the whole issue around fake news today.
Saul Marquez:
I know.
Chris Steel:
It’s really going to become a problem in the near future.
Saul Marquez:
Something to think about. So getting close to the end here, Chris, this has been fun. I’ve got a lightning round for you. I got a couple questions and I’ll ask you. Lightning round style and then we’ll follow that by a book that you recommend to the listeners. You ready?
Chris Steel:
Sure.
Saul Marquez:
All right. What’s the best way to improve health care outcomes with data?
Chris Steel:
You need three things. You need to have the data itself. So large amounts of it. You need the domain expertise. So the people that understand the problem domain, you can’t take generic data scientist and have them work on health care outcomes. And then you need the analytics and the platform itself. You need a platform that’s going to be scaled up, provide you with the bleeding edge technology. You need to both analyze that data.
Saul Marquez:
What’s the biggest mistake or pitfall to avoid?
Chris Steel:
I think if you task one of the large tech players in the company, they would probably tell you that the biggest mistake is to not have the domain expertise to try to solve health care outcomes without having the people that live and breathe that on a day to day basis.
Saul Marquez:
Love it. That data science so that the appropriate human, right..
Chris Steel:
Correct.
Saul Marquez:
How do you stay relevant as an organization despite constant change?
Chris Steel:
I think there’s a lot of different things you can do. First of all, you need to focus on emerging technologies. You need to stay ahead of the curve and understand what’s happening, not just in our industry, but in all the other industries as well. Have people that are very dedicated to looking at those things and to be able to sort of separate the wheat from the chaff, know what to go after, and then spend money, spend the research dollars to make sure that you’re not going to get blindsided. So you think about quantum computing, nanotechnology, all these things that could revolutionize the health care space. You have to constantly keep an eye on that, but also not allow yourself to get sucked into betting on a lot of big banks and having them fail. You need to have good people that really can understand which are these technologies, which are these new developments is going to work and which ones are years down the road still.
Saul Marquez:
As it relates to A.I. and health care. What’s one area that drives your work?
Chris Steel:
I wouldn’t say there’s one area I would go back to what we’ve talked about before, the human data science part and applying the A.I. and M.L. to real world problems, things that have tangible benefits for patients, for providers, for payers. So really in my line of work, I really focus on the business value that A.I. and M.L. out brings.
Saul Marquez:
These next two are a little more on a personal note, Chris. Number one is what is your number one health habit?
Chris Steel:
My number one health habit is trying to stay active. I’ve got four kids, two of which are still at home, and I spend a lot of time coaching them through sports, etc., staying active in the community. I had been jogging for a while, but it ended up breaking my leg a year or two ago. So I’ve been off that. Now just finding other ways to keep myself fit.
Saul Marquez:
Love it. And what is your number one success? Had it?
Chris Steel:
I think my number one success is really getting up early and preparing for the day. So being ready to having that sort of sweet time in the morning between 6:30 – 7:30 where I’m really more focused on prioritizing what needs to happen during the day and preparing for the meetings, et cetera. I think it’s very easy, especially in this fast paced world. We live in today to get sort of caught up in reaction mode. And I found that to be successful. If you really need to get out of that mode and prepare yourself to be able to prioritize and not to be afraid to cancel or reschedule meetings because you don’t feel that you’re going to get the correct amount of benefit out of it, it’s just because of all the other things going on that day that are a higher priority.
Saul Marquez:
Great. Great tip, Chris. And what book would you recommend to the listeners?
Chris Steel:
That’s a really hard question. I read a lot between technical books, books for leisure. It’s hard to say. I think I guess I’d have to go back to a book I read way back when I was barely a teenager called Black Holes and the Warped Space Time was by a guy named William Kaufman III. And it really had an impact on me. It’s why I became a physics major. And I’ve ever since really been interested in cosmology, astronomy, etc. Just fascinating work. Unfortunately, I never had the really high degree of math skills that would enable me to go into that field to become a professional physicist. But I do to this day continued to read all the books and keep track of the research going on.
Saul Marquez:
Awesome. Very interesting. Black holes, I think. How did it work? It can be a whole different… Chris, this has been fun before, he concluded. I love if you could just leave us with a closing thought, and then the best place of the listeners can learn more about you, your work, and the things that you’re up to.
Chris Steel:
Yeah, I think, again, just to reiterate what I’ve been saying, I think that human data science is really the important thing to maintain focus on the outcomes, to understand the value that you bring to others. If you use that as your main driver, it makes a lot of things easier and you become a lot more successful because you’re focused on the task at hand. You understand exactly how it is, the impact that it’s going to have. And with that, like I said, you need three things. You need data, you need domain expertise and you need the technology to be able to do that. So with that, I guess I don’t have a whole lot more to add. If you want to find out more, you can go to iqvia.com. And the Web site has a lot of a lot of really good stuff under the about tab, there’s a whole whole section on human data science and the different people that are involved and some really good news cases, etc. And of course, you can always email me at Christopher.Steel@iqvia.com.
Saul Marquez:
And that is STEEL will provide a link on the show notes so that you guys could learn more outcomesrocket.health type in IQVIA or type in Christopher Steel and the search bar. You’ll see this interview. Chris, this has been a ton of fun, man. Thanks for spending time with us.
Chris Steel:
Great. Well, thanks for having me, Saul. I really enjoyed it.
Thanks for listening to the Outcomes Rocket podcast. Be sure to visit us on the web at www.outcomesrocket.com for the show notes, resources, inspiration and so much more.
Sonix uses cutting-edge artificial intelligence to convert your mp3 files to text.
Thousands of researchers and podcasters use Sonix to automatically transcribe their audio files (*.mp3). Easily convert your mp3 file to text or docx to make your media content more accessible to listeners.
If you are looking for a great way to convert your mp3 to text, try Sonix today.
