Working with a big data sets to create better outcomes
Recommended Book:
Best Way to Connect with Sanji:
Company Website:
Removing Waste and Creating Agility with Artificial Intelligence with Sanji Fernando, SVP Artificial Intelligence & Analytics Platforms at Optum 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. Our automated transcription algorithms works with many of the popular audio file formats.
Welcome to the Outcomes Rocket podcast, where we inspire collaborative thinking, improved outcomes and business success with today’s most successful and inspiring health care leaders and influencers. And now your host, Saul Marquez.
Saul Marquez:
Welcome back to the Outcomes Rocket, Saul Marquez is here and today I have the privilege of hosting Sanji Fernando. He’s a Senior Vice President at Optum, where he leads the Artificial Intelligence and Analytics Platforms team. He’s responsible for developing platforms that support the design and development of leading edge AI models and analytic tools for the enterprise. Previously, Sanji was Vice President at OptumLabs and led the OptumLabs Center for Applied Data Science, which is called CADS. The CADS team applied breakthroughs in AI and machine learning to solve complex health care challenges for a UnitedHealth Groups and by developing and deploying software product concepts. CADS pioneered using deep learning to streamline administrative processes in a revenue cycle management and develop graph analytics tools to support provider network design. Among other innovations that they’ve done, he’s been at them labs since 2014 from Nokia, where he created Nokia’s first data science team. So definitely no rookie to data science and AI. Before that, he spent those 9 years there and has done an incredible amount of work as well as starting his own business. So it’s a true privilege to have Sanji on the podcast here and excited to cover his insights in healthcare and AI. So, Sanji, thanks so much for joining us today.
Sanji Fernando:
Thanks so much Saul very very, I’m looking forward to our conversations.
Saul Marquez:
Likewise. So, Sanji, what inspires your work in health care? Because at the beginning, you were in health care, right? You’re a technology guy that gravitated to health care.
Sanji Fernando:
Yeah. You know, I was in telecom and I worked in internet companies way back when. And I really had a thing in technology. I studied computer science in college. And what brought him back to health care maybe was growing up in the health care family. Both my parents were physicians. I had a great exposure to them. They were a primary care physicians in Connecticut and saw their day to day work set of as they touched. And while I didn’t pursue that as a career, I understood the almost subconsciously that it was something that was very satisfying. And now I’ve got a wife and three kids and you know, health care facilities all. So my hope is that by understanding and pursuing some pretty advanced technology over my career, I could bring that back to health care a little bit and see if we can make some significant changes in how how we use technology to enable health care in the US and even abroad.
Saul Marquez:
I think that’s great Sanji and so it’s that that story, you know, your parents grew up as physicians and when you took that left turn, Sanji, what did they think? I mean, did they say our son’s not going to be a doctor or were they happy that you became an IT guy?
Sanji Fernando:
I think they were. They were always supportive. I mean, you know, from the age of like 8 or 9, they were spending way too much money buying a computer and gaming consoles. And I think they knew that that was something I was passionate about. And there were times where you want to follow your parents suggestion. Being a physician is an incredibly honorable profession, incredibly stable profession, and, you know, I’ve made it in that effort to get the organic chemistry in college. And after that, I said, you know, I think I wanted to be really a little bit more excited about.
Saul Marquez:
Yeah, na, I can appreciate that. So it sounds like you were kind of drawn in by by technology at a very young age, gave it a swing, but said, I’m going to follow my passion, mom and dad.
Sanji Fernando:
Yeah. And they were supportive all the way.
Saul Marquez:
That’s awesome.
Sanji Fernando:
An important lesson for itself.
Saul Marquez:
That’s the way I especially as parents now, I mean. So, hey, kudos to you mom and dad, your Sanji is doing pretty good now. What makes what you guys do there at Optum different and how is it adding value to the ecosystem of the business and of your customers?
Sanji Fernando:
Yeah. My focus at Optum, it’s such a multifaceted business. We touched so many portions of the healthcare system, from patients to providers to governments to employers to pharmaceutical companies. And that gives us this is great opportunity to take a look maybe more holistically at how we can really drive and deliver better outcomes and change in the healthcare system. Specifically, my focus area, we’ve got a wealth of data across our enterprise over two hundred and twenty million, the identified patient lives data that gives us this great perspective on how we can take some of these new often referred to as artificial intelligence, say at the heart of artificial intelligence, machine learning approaches where we can take the machine learning approaches that by definition allow us to learn from the data itself rather than explicitly state exactly what to do next. And that’s a real important breakthrough for us, especially in health care, because health care is so complex and something we try to create a structured set of imperatives or rules or heuristics. There will always be edge cases and exceptions and it becomes very difficult to construct and stimulating from that. And so what’s really what’s fun for us is that we have in our my team in the realm of the professional, several of them have this opportunity to work with this vast array of data in the hope that we will enable better outcome than really allow people to work at their full license.
Saul Marquez:
Love that, Sanji. And so as we think about machine learning in AI in health care, that these big data sets. What would you say is the is the opportunity? What could we do more of? Or what can we do differently to get more value for outcomes and business?
Sanji Fernando:
Yeah. You know, we’re very excited about the potential for any AI. And I think if you read a lot of the press today, a lot of the really amazing work happening at academic institutions, at leading technology companies. A lot of it focuses on potentially using this data to support better clinical decision making. And no doubt that will be a big portion of this story in the next 5 to 10 years. Right now, today, we have a great opportunity to look at some of the more mundane parts of our healthcare system, parts that are coming in changes, changing administrative processes or services like claims adjudication, revenue cycle management, risk judgment, and medical problems. These are important parts of our process and the need for us to deliver appropriately and well reimbursed health care. I think its complex and it’s these areas that we’ve looked to see how we could buy some of the latest and greatest breakthroughs in our artificial intelligence to try to automate and reduce some of that complexity. Free people up to work on the most complex cases and try to automate and try to reduce the amount of manual work on things. There are a little bit more repetitive and that was in good shape.
Saul Marquez:
Yeah, it’s it’s about freeing people up to do more things of higher value.
Sanji Fernando:
That’s exactly right. Our theory is that if we can allow people to spend the time on the most complex questions, there would be a niche. It is a clinical that’s going to be the best use of their capabilities and their skills. And the more we can leverage these approaches to prioritize their work, to automate key steps, automate approvals or agree on reimbursement levels that will free both payers and providers up and hopefully allow them to apply that to what are really problems that do need people to read and understand and solve something that machine learning approaches won’t be able to do for us.
Saul Marquez:
That, I think it’s it’s it’s that’s a great opportunity, Sanji, and so much time and effort is being spent on these these tasks that really doesn’t contribute to overall outcomes. How do you believe you guys at Optum are are using this technology differently than what’s out there or even better than how it’s being used?
Sanji Fernando:
We’re at the very front end of our journey, but we’re really excited about some of the products and services that we’ve brought to market that we think are helping us take those first steps in healthcare. One of them is Optum360case adviser product on your product, which automates the review matter.
Saul Marquez:
Which one is that Sanji? I missed that.
Sanji Fernando:
Optum360case adviser.
Saul Marquez:
OK. OK. Interesting, huh?
Sanji Fernando:
Yeah. And it it allows us to automate a review of medical charts so we can help providers decide which charts are the most complex and require a third party physician review. In the past we put that burden on case managers or folks working on the course to decide how to route that back to their party position for an independent outstation. Now we can take all the charts, review them all, and then determine which of the ones that really do require a physician to read and review. It just optimizes both the workflow for the folks on the floor in those hospital systems, as well as focuses our third-party physicians on other break cases to focus on. That would have been very hard to do if we tried to construct a complex set of rules that govern that. What we did was that we trained a deep learning model to identify those charts based on what our physicians had already reviewed in the past. So we can learn from those decisions and by training the model on past decisions. We not only can get a great deal of accuracy of this chart should be reviewed. We don’t miss most many of them. And more importantly, our criteria change as guidelines change as practice changes. We can continue to learn from those decisions and the model will essentially update itself to reflect that change.
Saul Marquez:
I think that’s super interesting.
Sanji Fernando:
Yeah. And you know, we’re excited. One of the things that coming from a tech background. Some I hear this with a tech, I think say, hey, you know, that that’s all they could do very quickly and easily using amazing tools from Google and Facebook like tensorflow and PI Torch. And that’s definitely the case. But the one thing I’ve learned in healthcare is that we do have an important responsibility to make sure we’re getting things right. Whether they be administrative or clinical, we’re constantly working with important patient data. Some of that identifiable these are important decisions that affect reimbursed and how that gets reflected in your overall health personally. And so oftentimes we just need to be very deliberate on how we test and evaluates. So I think that’s something that coming from outside of a healthcare program for many years we have gained a greater appreciation is that these changes may feel like they’re not moving as quickly as we’d like to. Got a lot on the line here. This is not about getting a catalog selection or signing a photo. Right. an Instagram post Right.. A level of democracy at home, although here we have some very important decisions that we want to ensure everyday Right.. And so it requires a lot of thoughtfulness and companies to do that. But I’m excited because now that we’ve demonstrated this capability in some of our products, we’re hoping to scale that up quickly and carefully across the enterprise.
Saul Marquez:
Yeah, and I think you’re bringing up some great point, Sanji, that that healthcare complexity and the spillover effects when you’re talking about people’s lives and outcomes. So is the intention to have these systems be used to grow the Optum business internally and can help be more efficient and better at what is being done? Or is the intention also to maybe eventually offer these systems and technologies to your customers so they could also benefit from them?
Sanji Fernando:
It’s both. I think because we serve so many portions of the healthcare system, we oftentimes off on behalf of and coordinate entities, other the payers and providers so we can enable our internal processes. But also like with case advisor and other tools where we’re embedding them in our products and services that customers use every day. So it’s a little bit of both.
Saul Marquez:
I love it. I love that you’re able to improve the business efficiency internally and then also help your stakeholders get better as well.
Yes. Now. I think that every day we’re working towards pursuing new applications of this technology. Same point as you’re probably super well aware of. Breakthroughs are happening around machine learning and artificial intelligence coming out almost every day. And so we’re excited to take advantage of all this great investing, all this great work, both in academic institutions and back companies to see how we can apply this new and novel ways across our business.
Saul Marquez:
Know it’s a must. And for those listening, thinking about how they can start applying some of these technologies that don’t have a data science team or don’t really know where to begin, I mean, what kind of advice would you give them?
Sanji Fernando:
We’re always there to help. I’ve got to federal from either myself or partners across Optum as we embed these tools into our products and services. And hopefully more people will be able to take advantage of it. They said if you’re getting started, it can feel overwhelming. There are a couple of things I would think about with regard to where to begin. We’ve built some tools are sound not. There is a small candidates that help us think about how to solve a problem and take advantage of AI when we solve it in some aspects that we focus on. Are you getting your chart? Right.. What do you want to accomplish? How’s it coming today? How do you want to improve it? What problem are you solve? Other important questions is to understand what data you have. What is information that you’re going to use to train a machine learning model? And with that data is a very important question. The lack of a better structure oversimplifies all of it. The question is, if you have a piece of information like a name or a chart or a lab report, do you have the answer you want? Do you have a record of what someone did or what you’d like to be the right answer? If so, if all if a model is presented with that information, that’s a really important question. And one way we entered in Optum, we realized we lost right answers. Awesome in daytime’s terms, it’s called labels. Labels represent might represent what a medical professional did when they tried to rework the home to get to the appropriate reimbursement level. It might be the decision of a physician to authorize or approve of a procedure or or a reimbursement. These are all sort of the right answers. We can learn from. So when you look at your results, you should find out what the data you have and what sort of answers you have. That’s not to say that if you don’t have those answers, you can’t build machine learning models. But it does a little bit harder. It moves into a realm of what we call unsupervised learning that can be more difficult to predict or difficult to maintain and support. And while it’s not a show stopper, that’s an important question for us. And finally, the one thing I would also think about is how much data you have received. You really have oftentimes we think about very big stops. Now you have millions of records, millions of pieces of information. But for the problem you’re solving. You have to ask yourself how much we need you to answer that question. This comes up quite a bit. I mentioned earlier that we have over 43 million patient loads of data. When you start whittling down, say, conditions or geography’s data can get very small very quickly. And so they’re having a large set of data is really important. Again, not a show stopper, but it makes things a lot easier.
Saul Marquez:
Now, there’s some great highlights here, Sanji, for anybody wanting to head in this direction, you provided a great framework. What are you most proud of in business accomplishments and in your career as it relates AI and health care?
Sanji Fernando:
I’m really excited that we spent about 3 years at OptumLabs an R&D center to launch, that being the podcast, really trying to understand some of the breakthroughs in AI like deep learning and how it might fit in. We then went through a real thoughtful process of testing and failing and testing different ideas and concepts and how we can apply these new technologies that ultimately resulted in products like Alpha 360 advice and what I’m really excited now was that because of those early successes, because of those early failures, we’re not at the point where we have enough information and understanding to say this is something we’re ready to scale. And the challenge we’ve been place to place on that’s right now is how do we scale artist intelligence solutions across her health care business? How do we take on bigger problems, bigger ideas and enable more of a business with this technology? So that’s what I’m really excited about. And that’s what gets me through the morning a little bit. Yes, it came up in the morning and it’s still on fearing the efforts and investments and failures. But really, I don’t see that we’re moving towards exit stage operating margin.
Saul Marquez:
I mean, it really shows the commitment of Optum to take this technology and leverage it to make make business and outcomes better 3 full years of just sort of sandbox figuring things out. And now you guys walked out of that with some really good ideas and they’re starting to put it into action.
Sanji Fernando:
Yeah, I really take offense. I always had a big convention commitment to innovation. I’m almost in here all over the leadership. And so but, you know, with that commitment requires patience. And I’m happy to say, I think at least for the AI, that patience has really paid off.
Saul Marquez:
I love it. It’s a great view into the culture and the innovative character of you and the leaders there. What would you say is is one of the biggest setbacks you you experienced during those three years? And what was the key learning?
Sanji Fernando:
If I think about it, we failed a lot. And overall, I don’t think any one set back defined what we did want to be out of. When you are working with more innovative, more realistic, more speculative ideas, technology, you have to be run for failure. We tried tons of stuff. Some that were typical of it didn’t pose as you’re talking. I’m just reminded we when we first started trying to interpret a medical text, we used a great library that Facebook trained on the English language to understand sentiment. From all we know about the stuff, we’re going to work on our medical tests. And it’s totally factually it didn’t completely fail. It didn’t perform as well as we had hoped. And when we looked at the top itself, software was so different from what have Texas trained on was that we had to train our own model to take into account such a vast variation in that medical short hand with fear and in my medical notes. And so from those I don’t see it so fast. But almost like you have to have your DNA, the ability to structure work like experiments and like a sound system as seem they fail, machine failure until it is proven successful. And that’s just a little bit of attitude and culture to be OK with, but not everything’s going to work out. But he’s going through a systematic almost a method that’s almost rooted in the scientific method where you’re systematically trying new things. Hopefully we have the okay to show that that hypothesis really just not going to work out. That’s the biggest takeaway from my talking an awful lot.
Saul Marquez:
Yeah. You know, Sanji, and that’s the word that kept coming up in my mind when you were sharing that as scientists. It’s a scientific method, right? Yes. Failure is is in the fabric of the duty.
Sanji Fernando:
Yes. And I will say that that wasn’t readily apparent for me, at least coming into control 5 years ago and really rethinking how to think about success and failure, that if you treat those very first steps as experiments, evaluate the results. I think overall, over a longer period of time, you’ll get greater success because then you’ll have better information to know how to proceed and how to proceed correctly. So that was a big learning for me over the last 5 years.
Saul Marquez:
I think it’s great and a great framework to put around this effort of integrating AI and deep learning machine learning into into these very complex and very niched efforts. I’m just talking to a guest not too long ago that shared about how they wanted to come up with a CRM for pharma companies to engage patients, and they shared that they their initial attempt was to build upon an existing CRM and how that failed miserably. But it really taught them a lot about the importance of customization. But then being able to scale and and now they’re able to appreciate all the nuances, kind of like you are right with trying to use this Facebook algorithm, having to retrain it because of all the shorthand. I mean, it’s really, really neat to hear your experience in that particular experiment.
Sanji Fernando:
Yes. And it sounds like they they have the same one, too. I think we’re all learning a lot about how to think about innovation differently. You know, credit where it’s due. I think a lot of this thinking has been pioneered with startups in Silicon Valley, but it’s very uncomfortable in to large organizations like ours, others as we try to engage at the same pace as some of companies.
Saul Marquez:
For sure. What would you say, Sanji, is is one of your most proud accomplishments thus far?
Sanji Fernando:
I didn’t predict setting a bad case of Isaac just to be something that I take being from what we could tell, we were one of the first about the first to market in the new cycle management product that you keep learning. I think we establish the value and benefit to the business. Well, I think some very smart people see that promise today. I given how much they invest the Mocha Moms. Until you actually deliver a product when the market is always in a question like, is this really relevant? Are we are we calling the Right. problem or are we in sort of quote-unquote a? And so that’s what we’re really excited about. I think that a lot of us can now think about great investments, think about different products and services that can be able to buy and how to expand. But it doesn’t happen without that precise.
Saul Marquez:
Love it. It’s a great, great call out. And what would you say is the thing that you’re most excited about today?
Sanji Fernando:
I think there’s a lot to be excited about. But I you think about the future of artificial intelligence and higher can change health care. I think I’m really excited about the possibility of getting to a guy that can help with clinical decision support. We are not there yet, but we might be there. You know, even this morning there, there’s so many people working towards overcoming what is essentially the biggest challenge with some of these breaks. He was like this one, which is getting to real interpret ability sometimes because of black box problems. If you really want to be able to recommend a course of clinical care for someone, you can’t explain why. And that was that sort of answering the question of why the recommendations being made is he’s also broke too broadly on set up a purchase, talk, cause or methods to try to understand the causality for a problem or data. And when I look out for I see amazing investments and breakthroughs happening at companies are mentioned like Google and Facebook, academic institutions like MIT and Stanford for researchers and data scientists and engineers trying to figure out how to get to two qualities. Once we get there, I think we will be more confident possible. I think physicians and clinicians and nurse practitioners and all those folks who are really the most important part of our healthcare system, people who are actually caring for you, to me, they’ll be more confident that the recommendations we’re making are understandable, are comprehensive and safe and trustworthy. That requires a lot of work for all of us to explore and continue to be we’re not there yet. But given all the work happening across the country on pumping that we’ll see some breakthroughs and overtime with lot of common marker testing and validation, we might be able to apply that to a clinical challenge.
Saul Marquez:
Love it. Love that view. That horizon viewing into the future, it’s definitely promising. Be able to lift the hood on that outcome of the decision and understand causality. So if you could have lunch or coffee or tea with anybody, who would it be?
Sanji Fernando:
Yeah. You know, one of the folks that I’ve been a little bit inspired by and I love a watch with is having Charlie Munger, who had you know a lot about. He is Warren Buffet’s sort of best friend, partner in Berkshire Hathaway. And he’s got an amazing perspective on business investment where he looks at, he’s constantly testing yourself to understand what is it saying that that describes a complex problem of his. And he’s got a book that at Princeton is full of his speeches and thinking about how to think about problem. It’s called Charlie’s Almanack.
Saul Marquez:
And I actually have.
Sanji Fernando:
Oh yeah, yeah, yeah, sure.
Saul Marquez:
Oh, yeah, yeah. I’m looking at it. I’m like, oh, my God, I got to pull that thing down. I haven’t looked at it in a while, but please. So why should we take a look at it, Sanji? Like, tell us a little bit more. I was really intrigued when I learned about Charlie, the man behind the man. I mean, that’s such an interesting guy.
Sanji Fernando:
Yeah. You know, I’m positive. I’ll take it. He studies all search areas of academic study and theory and practice, looking for different ways, different frameworks to solve problems for the overall size, the economics. What is the framework that defines a two sided marketplace, like we say, versus what is the framework that drives behavioral economics and some of the trends of patterning? And what’s amazing is not all these different ways, other problems from many different diverse disciplines. And you’ll you’ll never know what probably in front of you. But if you have the sort of set of frameworks to think about the problem and draw on these frameworks that have been established, that seems like a good way to get started solving the complex problems that can arise about reading about him in different mental models and how to approach problems.
Saul Marquez:
That’s pretty cool.
Sanji Fernando:
And I’m thinking that’s the difference there.
Saul Marquez:
Yeah, that’s pretty cool. So that being the book you recommend or would it be a different one?
Sanji Fernando:
Well, I highly recommend that book. I think important Charlie is out in a push. That is not so for me. Yeah. Energy is a shortcut. I think is a blog called Farnan Street. Near the Post is a real big name, Charlie Munger. And he’s got a little bit of a crib sheet like 100 mental. That is sort of his vision of an element of really, you don’t want to have it, you don’t know how to. Maybe you can find Charlie’s Almanar. I know it’s been relatively defined in publications, but maybe also great place to start is interested in mental multimodal models.
Saul Marquez:
Yeah, I’ve heard of that. You know, a friend of mine is in listens to it is a fan of, I think, a shame. Right. Farnam. Yeah. Yeah. Farnam Street. So they go, folks, there’s a a digital version of Poor Charlie’s Almanac. And then also try looking it up might be out of print, but you might get lucky and find a copy out there for resale. Sanji, great, great recommendations. And that’s a first two, by the way, for no one’s recommended that one before. So what’s the best advice you ever received?
Sanji Fernando:
I can’t begin to think about so much. They’ve also gotten over the years. I think one thing that in my life right now that’s important to me is really thinking about time and time management. Our whole hospital, we look it’s really important to have time to think about where had problems. And you’ve got to prioritize on your count during your life, especially given the nature of the types of things that they would be doing. I’d like to do the case finding one that sometimes we can fill out and it’s a 9 to 5 meeting that two things to do. But if you don’t prioritize that time to think about hard problems, apply some thoughtfulness and time to plan. That’s so important. My goal in my career and while I’m trying to bounce back with all the communities that’s supporting lots of folks and also sending great parties and family and make a show about is often I think management is a great quote. I guess I think friend Warren Buffett, maybe he said no amount of money will buy you another second.
Saul Marquez:
Oh, yeah, yeah. Yeah, I love that. That is great advice and thinking through those hard problems, you really do have to take a step back and so appreciate that call to action for the listener, Sanji, before we conclude, by the way, this has been fun. I really have enjoyed them both getting and knowing you better Sanji in learning about your philosophy around AI and applying it to health care. But I’d love to have you share a closing thought with the listeners. And then where can they continue the conversation with you or follow the work you’re doing now?
Sanji Fernando:
Yes. One pulls got a HILES is endless series as folks who want to make different health care. I know there’s a lot of hype around AI and I’ll be a lot of failed policies. And I think we’ve all seen those kinds of periods of hype and I don’t know who at the top of it. But at some point we’ll be at the downside of that AI have gone through waves of success and failure, and I have no doubt history may repeat itself. But I think if we could get through some of those cycles, I think we can make meaningful changes and pitching policies. I think that’s on all of us, too, to understand these masses, understand what’s possible. I hope someday that we’re all experts in our intelligence machine learning confusion. It’s too. And I say that because they think about point, we’ll be making a material impact. But it’s not going to be with failures. It can be without down cycles. And hopefully we’ll see through that and can get to a place where we made a material impact elsewhere. And no, anyone wants to keep in touch with me, I’m on twitter make them. And my email add is Sanji.fernando@optum.com.
Saul Marquez:
Outstanding. Sanji, thanks again. This is really been a true pleasure and I’m excited to see what you guys do with the work. You work so hard at OptumLabs to figure out and the work with the Atom 360 Case Advisor and beyond. So I really appreciate the time you spent with us today.
Sanji Fernando:
Thank you. It’s great with you for to catch you up too.
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.
