Full Width Image Background
Text

Expert advice on practical innovation

Optum Forum 2019: Emerging tech session soundbites

Video Component

Kerrie Holley:                     How do we take ideas and do practical innovation? Well, it starts by having a landscape that we look at in terms of emerging technologies. Let's take a look at blockchain for example. So to understand blockchain, most people understand blockchain from the reality of cryptocurrency. So we've all heard about cryptocurrency. The technology underneath it is blockchain. Now, blockchain is actually very easy to understand. It's really just a collection of blocks, immutable blocks, meaning I can't change them.

                                                Each block represents a ledger. And anytime I want to add to the ledger, I chain a new block. I have private or public blockchains. How does that apply to healthcare? One of the biggest problems we have in the healthcare system is maintaining provider directories. So maybe Optum has the correct address, correct phone numbers, maybe Humana has a different, and maybe Humana is more accurate. If we work together, we can actually create a directory that public, that's accurate in the consortium.

                                                And that's what we've done. We've actually created with Quest with Humana and others and Optum, we've actually used blockchain to solve this problem, which creates friction, creates pain for providers, for members. So that's an example of using blockchain. Really just a public database that we can use globally.

                                                Another technology that we're using is augmented and virtual reality. Think about behavioral health. When you think about behavioral health, there's a lot we can do in terms of, for example, phobias. Instead of taking someone to the Grand Canyon, we can actually in the therapist's office, use augmented reality to recreate that fear, and the therapist doing their magic can actually create a better outcome.

                                                So another example, ambient computing. If you were on stage or listened on stage job this morning, you heard about the chronic conditions aging at home, in home care. One of the opportunities we have with a lot of these technologies as we use them holistically is to be able to actually make a difference in home care. We can use ambient computing, the ability for devices and things that are intelligent to sense and have contextual awareness.

                                                For example, I know that someone has fallen or I know someone has not gone to the kitchen in five days. So I can begin to use technology to see behavior and to perhaps take corrective and preventive action. Voice analytics, think about nurses and we have something called a nurse line. People call in, they leave voicemail messages, and think about yourselves when you read a voicemail and it takes time. What if you had a hundred voicemails to go through? Wouldn't it be easier if we could take all those voicemails, give you a written summary, but still not requiring you to read them all? Instead we pinpointed what the caller was actually asking about, and then surgically help you understand what the answer is.

                                                How do we do that? We do that by analyzing voice. We do that by natural language processing. So again, another example. So we're going to cover more of these smart speakers, things that we're doing with things like Alexa to be able to help members do adherence in terms of their medication. So a lot of technologies in a play here that we are spending time in terms of learning. The picture that you see in front of you now is a historical picture.

                                                It shows the arrival of the personal computer and it shows the on the far end, the arrival of artificial intelligence. Even though we know artificial intelligence has been here for a number of years. We'll come back to that story in a moment. There's a couple of messages here. We talk about these kinds of technologies as general purpose technologies. When you think about electricity, electricity is a general purpose technology. It fundamentally has changed the way you and I work, live and play.

                                                That's what a general purpose technology does. It has a fundamental impact on economies, on nations and on business and on business models. A general purpose technology is enormous and the impact that has on societies. We've seen that with the computer itself. We've seen that with the internet.

                                                The other thing that is also showing in this picture is the modality. You and I at one time spent a lot of our modality in terms of keyboards and we're seeing the use of the web, but today we're seeing more and more use of voice as a modality. But the key message I want to point out to you as a conversation I had with some city planners in Chicago some weeks back. And one of the points I made to them is would you ever think of designing the city without electricity? And the answer of course is no.

                                                Well, why would you think about designing a city without artificial intelligence? Artificial intelligence is the most over-hyped and under-utilized technology of our generation. It is also a general purpose technology. I realized the statement I just said is a bit of a paradox, but I point that out because we use the word generically when we talk about artificial intelligence, but artificial intelligence is much like a Russian nesting doll.

                                                When I start peeling back the layers, I began to see the various technologies, whether we're talking about machine learning, whether we're talking about deep learning, whether we're talking about natural language processing, natural language understanding, and the list goes on. There's a ton of technologies that comprise what we describe as artificial intelligence, but we're going to come back and go through that in a little bit more detail for you a little bit later in our conversation.

                                                So for now, the message I want you to walk away with is that artificial intelligence is a game changer. And I know there's a lot of doubt in individual's minds, but think about if the year was 1950 when the computer itself was just becoming popular. There weren't a lot of applications being applied, but we're about in that same window of time. If we look at the opportunity for artificial intelligence to make a difference in healthcare, we have a huge highway ahead of us in terms of opportunity.

                                                Whether you're a health plan, whether you're a provider, whether you're a payer, it doesn't matter. The opportunities to actually make people healthier, to actually make the system work better, to actually reduce administrative costs, to actually reduce and make operations more efficient, to create awesome customer experience. This is the opportunity we have with this general purpose technology that I'm describing as artificial intelligence.

Text

How to solve today's business challenges with emerging technology

Look at the landscape of emerging technologies with Kerrie Holley to learn how general-purpose technologies will change health care.

Horizontal Rule
Video Component

Jess Lewis:          I get really energized and excited whenever I hear Kerry or others speak about some of these emerging technologies and capabilities. It's just fascinating to think about where we're going to be in five to 10 years within healthcare using some of these. But I want to talk a little bit about how we can apply those even today. You're going to hear some examples about how we're taking more of a practical approach to starting to dip our toe in some of these artificial intelligence capabilities and emerging technologies.

                                So we've taken a very disciplined and structured approach to applying these. And as you see on the screen, it's really a three stage process. Kerry and his team and others are really looking at how to stay abreast of the emerging technologies and capabilities and looking at what's happening in other industries. It always seems to me that healthcare is a little bit lagging in terms of applying those capabilities. But we're really trying to stay abreast of how those are being used and where the maturity level is with some of those capabilities that he spoke of earlier.

                                Working with you and with our operations folks and our product folks, we then can really think about how to transform and change some of the processes that exist within either healthcare administration or the care management, the medical management, and even the delivery of care. So that's kind of the next step in this process. I'm going to go into a little bit more detail in the next few slides. But as you can see, this is really a cross discipline approach.

                                We run what we call immersion sessions from time to time where we're bringing operations, product, customers and technologists together around what could change. Given this capability, could we operate our provider network differently? Could we change the way we're interacting with patients with the way we're interacting with customer service? So a lot of that is resulting in ideas that we go through a proof of concept and a minimal viable product effort.

                                And then the last stage is really important, and that's really about how you scale this. We all know that there are legacy systems. There're legacy processes. There's regulations. There's a lot of scaled production criteria that we've got to take into account as we move some of these capabilities into a full scale development. And you're going to hear today about some of the efforts that have been going on related to prior auth in that realm.

                                So let me talk a little bit more about some of the work we're doing in the first step of this process. Technology is changing rapidly. It's tough to keep up. I live in the technology world and I can't keep up. I think Kerry does a much better job than I. So he and his team are constantly monitoring and checking to see what's coming.

                                We really think about this in a horizons framework. Some of these capabilities are here and now, zero to two years. Let's apply them in the business that we're running today. Some of them are a little bit further out, maybe two to five years. And even some of them are five plus years. He talked about ambient computing. He talked about so many other technologies that maybe aren't ready for prime time. But let's get our hands dirty and let's figure out what capabilities those can bring to the table so that when we are at scale or when we do have those capabilities ready to apply, we've got a little bit more experience and start to apply them into our current business.

                                Many of you seen the hype cycle on the slide here. One thing that we're trying to do is test and iterate early with these early stage capabilities so that we're not getting such a peak of high expectations. We really have a little bit better idea of this is what it can and can't do at this time. We're definitely trying to avoid that trough of disillusionment where we've set the expectations so high and we think that it's going to solve this process in just a completely different way. So we're trying to iterate and test and touched base with current operations early rather than waiting and hitting that trough of disillusionment.

                                When we get to the point where Kerry and his team and others have worked on a technology that's now ready for deployment, something like, I'll take the voice area. We've done a lot with voice transcription and natural language processing, and are now really starting to apply that. You'll see that even today in the example. How do we then take that and start to move into the second phase of this practical innovation approach.

                                This is, again, a cross disciplinary opportunity. It's where I really get a energized because now we're bringing the groups together and really looking at how can we change the process that we're currently employing. What are those ideas? The backlog of ideas and the backlog of concepts coming out of that is really what we use to drive the development. Some of those ideas aren't going to pan out. We're going to iterate around those. We're going to spend the first few months really looking at what is the solution really need to entail?

                                It might be a combination of people and process and technology. That technology might be emerging technology and existing technology. So there's a lot to consider there. Very importantly, what we try to do is set up some hypothesis around when we get this accomplished, we'll see these metrics impacted, whether it's improved engagement, whether it's lower medical expense, whether it's efficiency in the operations. Then we'll test and iterate and incubate that.

                                And really that incubate stage is all about reducing risk when we get to the point where we're scaling this. We've now tested this to the point where we've got a high degree of confidence that this works in a production environment. You're going to hear an example of that today. That paves the way for that integration phase. Then we get into a lot of the scaling and the work of really embedding this within the at scale business that we're all running.

                                Again, if you think about your business, our business together, there's healthcare administration, there's clinical management, there's the delivery of healthcare. One of the examples that we're going to talk about today is a process that we all hold near and dear to our heart. It's called prior authorization. Prior authorization really breaks down into two components. We're going to talk a little bit about the administrative aspect of that, their administrative review of prior authorization. Kerry's going to describe a little bit of what we're doing in that arena using deep learning. Then we're going to hand it over to William and Robbie to talk about the clinical review. If you think about those two processes, they warrant different approaches. I think you're going to hear some of that today.

Text

An approach to consider for practical innovation

Listen as Jess Lewis describes a cross-discipline approach that Optum takes to iterate around new ideas and capabilities.

Horizontal Rule
Video Component

Kerrie Holley:                     Let's talk about what we're doing with deep learning because we're doing a lot of things in the company. We're going to talk about just one process, which is the administrative side of prior authorization. Now prior authorization is a big process, and I'm only going to be talking about a small part of it because we started small. This is our process today. It's a big process. Again, I keep emphasizing, I'm only talking about the administrative side. Today it takes three to 10 days, even on the administrative side, for us to give the member, for us to give the provider a go. This has been approved three to 10 days, that's ridiculous. Why does it take three to 10 days on an administrative review. What we're doing with artificial intelligence and specifically deep learning, we're taking what has been a three to 10 day process. What do you guess? What do you think we're reducing it to, anyone? Is it one day? Is it hours? Is it seconds? Anyone got a guess?

Speaker 2:                           Six hours.

Kerrie Holley:                     Six hours, I hear.

Speaker 3:                           30 minutes.

Kerrie Holley:                     30 minutes. Take milliseconds. Milliseconds, and we're doing this at scale. We're talking about millions of procedures. That's the punch line. No delay, no abrasion. Eliminate the administrative burden, use artificial intelligence. Now I've told you artificial intelligence is a broad term. I'm going to be specific, again, because we're using a specific technology called deep learning. Let's give you another lesson in deep learning. All of us have taken tests before and when you take a test, wouldn't it be nice to have the answers? That's what we do with deep learning. We have something called a training set, which is akin to having the answers ahead of time. It's like the answer I just gave about muffins versus cats. If I label all of these images, if I have a million distinct images, and I label them cat, muffin, cat, muffin, cat, muffin, I can then now take a new data set, which is called my validation set, maybe, and I can begin to train the model.

                                                Let's back up a little bit. We start data, okay. In the case of prior authorization, we're talking about data like age group, CPT codes, diagnostic codes, and others. We start with data. We then do something called feature engineering, which is really just a simple process of what are the actual data that matters? What are the variables that matter? By the way, the other interesting thing that we can do with feature engineering, and I didn't have this conversation with you yet, but deep learning is a subset of machine learning. They're often used interchangeably, but they're different. Deep learning is what I describe, it starves for data and we can do more.

                                                Let me give another example just to drive the whole point. Let's take a flashlight, for example. Let's say you had a flashlight that was a machine learning flashlight. With audible cues, that flashlight when the word dark is said, would turn on and it would do it reliably. It would even turn on if you said, "Dark" in 20 different languages. Let's say we had a flashlight that was machine learning/deep learning. Now my flashlight is powered with deep learning. What's different about my flashlight? Now my flashlight, because I fed it with so much data, now I walk into a room and maybe it turns on automatically because it's got the sensors or maybe it turns on because I said, "I can't find the light switch." Suddenly, all sorts of things that you and I say as a human, "I can't find the light switch, I can't see," no word of dark, and suddenly the deep learning, which is different, it's learning on its own.

                                                In a nutshell, machine learning is the use of data to create outcomes with computer programs versus writing procedural code like if, then, else. That's what machine learning is. Deep learning is machine learning on steroids. Because we're deep learning, we can now have the computer teach itself. It can learn through data, it can learn patterns. It can see things that you and I can't see. This is why a radiologist will still be in business, but now we can arm them with tools where they can look at millions, not that they need to look at that many, but they can look at hundreds of scans at a time. Let's go back to this notion here. When we look at a deep learning model, we basically train it, and we are able to get a reliable prediction.

                                                In this particular example with the administrative side of prior authorization, there's a few points to know here. One, which I've said already, milliseconds. We've taken a three to 10 day process and gotten it down to milliseconds. That's awesome. Secondly, we know that many of us think of AI as a black box, but you and I work with black boxes all the time whether it's our television, whether it's this device I'm holding in my hand. What we need to do in order to make AI usable is we need to explain it, and that's where we have this explainable models. Not only did we build the model, we opened it up, and we allow the clinician to actually see how we made those decisions. That's what I mean by the explainable model. With that, you can see this is at scale. I hope you walk away with a pretty good understanding of what AI and deep learning is.

Text

Practical innovation applied to prior authorization's administrative review process

Learn how a deep learning pilot has reduced prior auth from days to milliseconds, minimizing administrative burden.

Horizontal Rule
Video Component

William Krenz:                   For those who work in the administrative side, some percentage of items in prior authorization go to clinical review, right? And we've got nurses in positions going through complex set of data to make a decision, and I'm going to show you a video in a second. But I'd ask you to go to this place. Assume you've got a loved one you're supporting through that process where you're awaiting authorization for a PET scan or a knee surgery, right? It's really important we get it right. This first piece might get us an answer quickly on the administrative side. Ravi and I, have been working now on how do we speed up the clinical component? Let me show you the machine that we built.

Speaker 2:                           Think of the complex constantly changing rules that nurses and medical directors need to keep up with every day. Think of all the data they need to navigate just to make a decision on a prior authorization request. The process can be confusing, tedious, time consuming, and prone to error. Machine-Assisted Prior Authorization eliminates all that, ensuring patients get the right care at the right time.

                                                This solution expedites the approval of appropriate medical procedures, increases clinician productivity and helps them practice at the top of their license. The technology is based on an artificial intelligence platform that has at its core, natural language processing tools and expert systems that mimic human reasoning. Vetted clinical criteria for almost every medical situation have been entered into the platform. Patient information including past claims, lab results, prescriptions and clinical notes are gathered in real time when needed.

                                                The result is a simple portal which makes a determination based on clinical data, whether medical necessity criteria have been met or not met. The clinician has the final say on whether a prior authorization request should be approved or denied. Machine-Assisted Prior Authorization helps decision makers make the right decisions faster. It ensures patients get the services they need and unnecessary procedures are avoided.

                                                Over time, the pain points of prior authorizations disappear and the trusted bonds between payers, providers, and patients strengthen. The power of data. The power of Machine-Assisted Prior Authorization. Together, making the health system work better for everyone.

William Krenz:                   There we have it and I'm excited to tell you it's not just a pilot. When we were planning this forum, we were kind of in that stage, we're now live in three markets or three states that operate in. And you know what? I'll tell you as we were talking about the administrative component, imagine yourself, in your mind right now a different role. You're the nurse, and I know everyone here is a nurse and has done this function, but you're looking at a 40 page fax and it's a little blurry, right? And you're digging through claims data, and then you have these kind of decision tree logic branches that you've got to go through that CMS provides.

                                                And let's say you're in a given state, it's a local coverage determination, right? And then over your shoulder is a supervisor who's saying, "Do three, four, five, six of those an hour, let's go, let's go." Right? And at the other end of the call is a human being who is somebody's grandmother who's awaiting this approval, right? It's not the best use of time.

                                                We've taken a lot of the complex datasets, and Ravi will explain it in a minute. And put it in this machine that quickly reads all the clinical data and notes, the claims, pharmacy data, right? And I would describe it if I could use the fun Japanese term, Pachinko machine, right? And it comes out with the answer because it's all branching binary logic, right? And with that, then puts it in front of a clinician who can then make a determination.

                                                We did not do this with the machine independently, right? And if we need to elevate it from a nurse to a provider, the provider then has a much simpler path to look at the information and confirm the decision. And for those in the room that working in the industry, like I've had the pleasure for a number of years, you can start to see the path we can go forward with this. Prior auth is just the start, it's not where I'd like to spend my time, but I know it's necessary for a lot of good reasons and there's more to it.

                                                I will just finish with this and give the mic to Ravi, to describe how he built it. The the thing you have to understand what's so complicated about this, remember that nurse? They're reading through this fax and the term and/or, or, might appear somewhere in those 40 pages and that one word will determine whether or not something should be approved or not.

                                                And yet we really get frustrated when our quality and our consistency isn't high. We've created a job that's almost impossible to do from my perspective. And I've watched after this thing for years and years and years. I really appreciate Ravi, our partnership and all we've done together. We've made the lives of our employees a lot better and are on the path to making a lot better for our patients too. Why don't you tell us how you built it?

Ravi:                                      Certainly. Thank you William. William, has so eloquently outlined the problem. What we have done is to basically build a system that is for the clinicians. The clinician, the medical director or the nurse has the final say on what happens and it is built by the clinicians. They own the clinical criteria. The first step in this process is really to have this clinical review board that is chaired by the medical director, and we have the nurses and we have a clinical analyst. They review all the reference documentation, basically the criteria that CMS basically says in it.

                                                The National Coverage Determination documents, the LCDs, Milliman Care Guidelines, whatever they use, they review all that and come to a consensus. It is the medical directors and the nurses who interpret those criteria and they come to a consensus and document those criteria. Once it is documented, then we encode that into this expert system.

                                                Just think of this. All of the knowledge that the medical directors have acquired for their training and their education, their experience, all of that is now quantified and made available for everyone to use. The second thing, consistency. If you have three doctors looking at two records, they probably come up with four different opinions, right? What this does is to make it objective, it makes sure that all the reviews are done consistently.

                                                Everyone basically follows the same guidelines but it doesn't take away the human process. Ultimately, it's the medical director who has the discretion in it to say, "Yes, as per the rules, these criteria are not met but I'm going to use my discretion. There are extenuating circumstances and I'm going to go ahead and approve it." It's completely within their hands to do it.

                                                You can't teach a machine compassion, so we aren't trying to do that. MAP will follow the rules as specified by CMS and applied exactly and say these are the criteria that are met and it will give them the results, and the physician and the nurse can override it. They can say, "Yes, this makes sense but I'm going to override it."

                                                How does it do it? With MAP, what happens is, the first thing that happens, a product request comes in. And that could be let's say a Joe coming in for say, knee surgery. As soon as a product request comes in, via phone, fax or email, whatever, right? MAP basically picks it up. MAP picks it up and then it starts going through the same process that the nurse does. What does the nurse do?

                                                They say, "Okay, we have Joe, let's start assembling the health profile for Joe. Let's look at the past claims. Let's look at all the other historical prior output and so on." That is what MAP starts doing. It starts going to the claim repository, gets all the claims, gets the lab results, get prescription claims and so on. It grabs all of that. It takes about two years worth of medical history for Joe and starts assembling it.

                                                What providers also do is basically send records, progress notes from their EHRs, and that usually comes in the form of a fax. What MAP does is to grab that progress note, the radiological reports, pathology reports and so on, run it through OCR process, scans them and then run that through a natural language processing engine. What that does is to basically take those faxes that William was talking about, that 40 page fax, unreadable. It converts them into a structured format, it converts them to XML S. It converts words like, "The patient has dementia," into an ICD-10 code.

                                                All of that gets converted and that's what you see there. And this is information about Joe. At the end of all this, what you have is two years worth of medical history for Joe and which is comprised of claims, labs, so on, and also other information that is gleaned from the progress notes and so on. Now that we have all this, then the evidence engine comes into play.

                                                The evidence engine says, "Joe is coming in for a knee surgery." What are the pieces of data that are relevant for knee surgery? It knows that because it is actually working the rules, so it knows, "Okay, does, does Joe have arthritis? Is Joe complaining of leg pain? Has he tried a conservative therapy like say take cortisone injection, so on?" It looks for all of those pieces of evidence and it will discard other things that may not make any sense in him. Perhaps, Joe has amnesia, right? That may not be relevant for knee surgeon, so it discards those pieces of evidence.

                                                It curates all of this, packages it and then sends it to the expert system. The expert system has all the rules that the clinical review board has come up with. It works through all of those rules, figures out which criteria are met, and then comes back with a recommendation. This entire process from start to finish when the prior auth request comes in, to when MAP has actually figured out that whole determination takes about a minute. I know it's probably 10,000 times more than what Carrie has done with 10 milliseconds, but I think there's a different process here.

                                                Finally, when the nurse actually looks at the display, the nurse is the final arbiter, the nurse and the MD. What they see is this kind of an interface. They will see the same product request and MAP will tell them this procedure. It will recommend whether the procedure should be approved or denied based on the criteria that are met or not met. But this is really the beauty of the system. It's not just about saying "Yay, go ahead," or, "No", it's why it should be done. It is a clinical rationale for that decision.

                                                The clinical rationale, and as Carrie was explaining, it is the explainability of the results that is critical. And that explanation needs to be in a language that the clinicians understand. It needs to be based on clinical criteria that CMS or MCG comes up with, and that's what MAP does. It says, "These are the criteria that came from CMS." And which criteria are met or not met, and all the evidence that goes to basically meet those criteria. It lists all of that.

                                                The nurse or the MD has the ability to click on each one of those to validate what was the piece of evidence and then it goes down into the details of which claim breaker or which lab result actually contributed to that evidence and so on. But this is the other cool thing. I mean, we know how we can tabulate and collect information from planes and show that, right? But then again, going back to the 40 page facts, what do we do with that?

                                                A lot of this evidence also comes from those progress notes. What MAP does is to basically pick up the key pieces of evidence from those progress notes using this NLP technology. And it says, "You know what I found on page number 20, the patient complained about arthritis." It'll say which page number and where is the sentence where it found that particular piece of evidence. And then it provides a link, so the nurse can click on it and then immediately it'll pop up this page, page number 20.

                                                The nurse doesn't have to go through the 40 pages of fax. They can just look at the specific page and the specific context and where it occurred. This provides a huge benefit for the nurses in terms of productivity, in terms of quality, and ensuring that we are really spending the time, the nurses and the MDs are really spending the time not doing grant work, not doing clinical work, collecting the information, but in really exercising their clinical ability to look at exceptional cases.

William Krenz:                   Thank you Ravi, for the fabulous work helping us build the tool. What I'd comment on for those in the room, a ton of energy has gone into adoption. We do a lot with fancy technology, it's wonderful, but getting the buy-in all started with the clinical review board and then a very careful, thoughtful engagement with nurses. It's quite a different change for them. There's lot of upside, but it also dawns on them that this changes their life in a fundamental way and as their time of work goes down, what's that going to mean? Right? It's going to mean less capacity needed for this work.

                                                That's true in all areas. It's kind of why Ravi and I, I think are a good team. I spend my energy on that, figuring out how we guide through that. It is showing significant productivity improvement, no great surprise, right? It is twice as fast or three times as fast. I'm confident of that. It's also getting at, and this is really important, consistency, right? Which is not only what our regulators want for those in the room who might oversee processes that are highly regulated like this.

                                                Defending decisions is difficult. This helps a lot. And it also has the implication, and I mean this sincerely, as someone who manages a delegated function for risk, right? We pay claims. It does say, "Hey, these are not procedures that are warranted." And it helps us not follow through with those. And I think that's the right thing.

Text

Practical innovation applied to prior authorization's clinical review process

Machine-assisted prior authorization increases productivity and consistency, and reduces costs in clinical reviews. See how.

Horizontal Rule
Video Component

Speaker 1:           Just a couple thoughts.

                                First, I think you've seen the application of emerging technology, AI, natural language processing, deep learning, to a real world problem here. It's now. But it's the convergence of truly deep technical resources with deep clinical resources and the operations folks that brings that all together. And that's not an easy thing to do. I think that's the example, why we picked that example is because it shows that cross disciplinary work and how you get, at the end of the day, some of these practical innovations across the board.

                                Three things that kind of come to mind. First, William really talked about it there at the end, focus on the people, the process, and the technology. We could have, and we did, demonstrate what the AI component could do in going out and finding that evidence, and in going out and approving the administrative process. But the real work was the adoption, the clinical buy-in. A lot of this is a people, a process, and a technology component.

                                Second is, we continue to take a very structured approach. If you think about it, the administrative side of things, approving in network or benefit coverage is something that we can take deep learning and machine learning and really get after in a very automated fashion. But when you talk about clinical review, where you're going to have to defend some of that decision, that's a different approach. So we take that disciplined approach to which technology makes the most sense and how that process feeds into it.

                                And then finally, there's an integration with the emerging technology and existing technology. This is where we're working with our systems that, whether it's the claims, or the billing, or the-

Speaker 2:           Built a few years ago.

Speaker 1:           Exactly. Just a few years ago.

                                So there's a lot of effort to put this into the integrated scaled business as well. But that's part of this process.

                                We talked today about the administrative, prior authorization process. We talked at the beginning about some of the big changes that we can make in the clinical management. If you think about some of the technologies that we've talked about today and addressing, what if we have genetic information, all of the claims' data and the individual health record information that we can gather on an individual, and arm the physician with that at the point that they're making a decision around which medication and which care regimen makes sense. But you got to do it in their workflow. And they've got to have the ultimate say.

                                You hear a lot of people talk about AI replacing jobs. I think this is just going to make our jobs much more effective, much more efficient, much more consistent. So I'm excited. I hope you're excited. We're anxious to hear about what problems and opportunities you're trying to get after. And we hope you can join us on our journey.

                                Thanks a lot.

Text

It's your turn now

Get the key takeaways to consider, as shared by our Optum experts, before you embark on your practical innovation journey.

Horizontal Rule
Text

Technology transforming health care

Explore more videos to see how technology and analytics are making health care work. 

WATCH VIDEOS

Text

Get the latest tech insights and deep dives.

Color Block

Text

 

Join us

Help us solve the most wicked problems in health care through research, development and emerging technologies.

See open positions

Text
Text
Text