Optum's Srinivas Sridhara provides an update on the latest in AI and health care.
Artificial intelligence has long been on health care's radar, and the industry is seeing historic investment this year. At the same time, health care is wrestling with challenges, such as negating unintentional bias. Joining the podcast to explain the latest AI developments in health care is Srinivas Sridhara, VP of Enterprise Data Strategy and Data Science at Optum.
Ira Apfel00:06
Welcome to UnitedHealth Group's Weekly Dose Podcast. We'll get you up to speed on the latest trends shaping the future of health care. I'm your host Ira Apfel. This week we're talking about artificial intelligence in health care. AI has long been on health care's radar. Big tech firms like Google are experimenting with AI. startups are joining the scene. And even mental health companies are trying to leverage AI for their patients. At the same time, the industry is wrestling with challenges. They include the ethics of AI in medicine, and how companies address unintentional bias, not to mention how the FDA regulates the technology that is constantly changing. Meanwhile, the sector is seeing historic investment in companies aiming to improve clinical care through algorithms. Joining us today to explain the latest AI developments in health care is Srinivas Sridhara. Srinivas is vice president of enterprise data strategy and data science at Optum. So, Srinivas, welcome to the podcast.
Srinivas Sridhara 01:09
Thank you for having me.
Ira Apfel01:11
Sure. So way back in November 2020, we had a colleague of yours on Steve Griffiths and he talked about AI and back then he said, the top three applications. Were health care executives, were looking to use AI were wearables, accelerating research, and faster and more accurate diagnosis and reimbursement. Are we still seeing that now? Or have things changed in the ensuing seven or eight months?
Srinivas Sridhara 01:37
Yeah, I mean, I think these appear to continue to be some of the top priorities that folks focus on, I like to bucket sort of domains into a few areas where I think people are investing time and energy today. One is on things that are operational tasks, you know, so a little bit you mentioned it around, say, reimbursements, or accurate diagnoses. But this is expanded over time to sort of think about risk capture or claims processing prior authorizations, fraud detection, all of these things we do routinely, and in an operational state in health care, and they are costly activities. And so there's real opportunity for AI. And I think increasingly, there's a focus in these areas. The second bucket is saying clinical applications. So you mentioned diagnoses, or even when we think about things like wearables, it's all around trying to understand why you know, somebody is going to get sicker in this moment, or have an increased risk of an illness or utilization, or can we help them by intervening sooner, or provide other support services. And so there's a lot of things in that space as well, that you see, and less than, than that, in some ways, you know, so maybe on actuarial methods, and so forth. So I think, especially in the last year and a half or so, there's been quite an acceleration of interest and understanding of some of the various use cases in AI. And folks are starting to really both invest and start setting some meaningful targets around them.
Ira Apfel03:19
So you mentioned targets. And, of course, we live in a results-oriented world. And I feel like sometimes that seems to be something that kind of holds people back when they when they think about AI. You're not seeing the huge results or big numbers. And there seems to be sometimes in the popular culture over promising and under delivering, or at least, you know, very high expectations for AI and what it can do. Are we seeing results right now? Is it still we're still kind of ramping it up in these three buckets? But like you said, and we're just kind of beginning to really kind of dive deep?
Srinivas Sridhara 04:00
Yeah, this is a great question. And really important as we try to help both our internal businesses and our external clients is understanding what's real. Gartner puts out this, what they call their hype cycle on many different areas. And AI has a very real hype cycle right around what is real today. And what I would qualify as real today, or if you want it to automate some document, review and clap, you know, classification around those people can and can and are achieving real benefit from those similarly, say, doing better predictions of risk for admissions, or, you know, costs. There's some real value that some folks are starting to generate out of those. And so I think it revealed importantly that we have poor interoperability and poor ability To really gather the data needed across health care, and all of that, and when you think of AI, foundational to doing anything in AI is having the right data and the right scale of data. And that's the gap. A second thing is that, because of the way that health care structured and the need for AI to have loopbacks, of, you know, information, you know, we're not a very customer consumer centric business, it's really, we're solving billing problems or provider and payer problems oftentimes, and that interaction is missing. And so we don't actually understand enough about the people or the members or the providers or others who are interacting with health care often. And that creates a gap in our knowledge and our data, and then Thus, the AI systems we can build.
Ira Apfel05:52
So what is Optum focusing on right now regarding AI in health care?
Srinivas Sridhara 05:58
So I think we have a multi-pronged strategy. So there are things where if I relate back to some of the main buckets that I shared, you know, these are three big areas, I think, a common theme that you see about, for example, these engaging members or clinical applications, I think of, we're answering often three sort of questions and sort of who needs some sort of service or, you know, offer management, you know, these sorts of things, a targeting application, if you will, you know, how do we engage them? So is this channel of communication or through what sorts of interventions? Will we drive optimal engagement? And then what do we do for them, sort of what recommendations that we have on a service, or program or intervention that they need? And so we're often trying to answer those three questions with our AI and machine learning solutions. And I think it's safe to assume that in terms of leveraging AI, for the enterprise for UnitedHealth Group and Optum, and UHC, we're focusing on those three buckets as well, correct? Absolutely. And I think one important thing that cuts across all of these is, you know, increasingly, we've had a lot of interest in making sure these solutions are not sort of accelerating, exists existing health equity issues, and health care, you know, and sort of trying to make sure we are doing our best to provide to limit bias in our algorithms and so forth. So that's a cross cutting need across these applications for us.
Ira Apfel07:40
Yeah, you just anticipated my next question, which is, of course, unconscious bias is a huge issue in AI, just throughout the entire industry, not just AI in health care, but everywhere. And I'm wondering, how is Optum working to prevent that? Because it seems like a really pernicious challenge and problem, and it seems to, you know, always be rearing its ugly head.
Srinivas Sridhara 08:06
It is. Absolutely. And I think, you know, one of the things when I think about this is that we have to start with the basic question, which is more that health care, health care fundamentally is inequitable. And, you know, the issues of say access or, you know, benefit in health care interventions, and how we think about it. Fundamentally, health care has these challenges. And when we build AI systems and applications, we are leveraging data that is built on that inequitable system. And so necessarily algorithms. And it's not just about AI, you could be building just basic descriptive analytics that are making decisions, you will, unfortunately have a risk of perpetuating that bias. And so that's, that's what is I think, things that we need to understand there are algorithmic solutions that we can work on. There are sort of data or sampling kind of things that we can do to avoid such bias. But ultimately, we need to pair these with programs and interventions that really address some foundational issues in how our society enables access to health care how we deliver it, and what incentives there are around it. So with that context, I think, when we look at what are we trying to do? One is we need to provide tools to be able to just understand where are we perpetuating such bias. So if we think in our IQ studio tools that we are building in, in Optum, we are trying to build tools that will help teams assess their existing capabilities and algorithms to understand what is the bias that they that is being persistent. So that they can then have the longer and harder discussion of what to do about it. Some of it will be around the data or algorithms, but a lot of it will be around the programs and interventions. So there's a set of work around that. A second set is we fundamentally need to raise awareness and understanding across not only our analytics teams, but our business and clinical teams about what are the risks that we face in perpetuating bias and health care through these algorithms. And so we've been developing training programs around, you know, Ai, fairness, bias, these topics to help people. And of course, we're responding to regulatory requests and action on them. And across all of these, then it's sort of that's that whole series is around addressing what we have today, and how we better manage it. And then as we build new capabilities, what we're doing is ensuring that the combination of training, these tools, and new capabilities we build are doing the best to avoid bias, or at the very least make apparent the bias that is being created or generated, so they can be managed programmatically.
Ira Apfel11:16
Which AI? project, are you most excited about? what's the one project where, where you really think this is, this is a game changer, or at the very least, it's just like so, so wildly creative and innovative. It just kind of gets you, you know, giddy thinking about it.
Srinivas Sridhara 11:37
Yeah, I would put it into two big buckets, maybe rather than give you one, maybe two, and I think, and they address core challenges in health care, right? One in a very practical way. Health care is challenging the, you know, in that it is so expensive. And we all struggle from whether it's our company, or if we think about, you know, health care at large at a policy level, the cost of health care is detrimental to the system as a whole. So I think there's a lot of applications that we're building that are around understanding how can we automate decisions and automate the processes so that we can reduce costs of health care? So whether that's in helping people not have extended reviews and prior authorization, or in responses on their imaging, decisions and feedback? can we better understand what are people's health care risks? And can we more precisely provide clinical recommendations for them? So this is where I think there's a real opportunity to leverage the specific benefits of AI, which is in handling the varied data that exists in the scale of data that that generates, and provide the notion of precision medicine. And I, I see that with caution, because I think precision medicine often people assume a lot more as possible immediately. But I think that is the opportunity that we're running at is to be able to better target clinical interventions for people for clinical programs, to yield earlier interventions for them and more precise interventions for them.
Ira Apfel13:34
Well, Srinivas, thank you so much for being the podcast. I'm sure in six months or so, things will change and we'll come back to you to get another update.
Srinivas Sridhara 13:44
Wonderful. Thank you for having me on.
Ira Apfel13:46
Thank you. That's it for this episode of UnitedHealth Group Weekly Dose podcast. Thanks for listening and have a great rest of your week.