John Santelli, CIO of UnitedHealth Group® and EVP of Optum® Technology, sat down with Zeeshan Syed, founder and CEO of HEALTH[at]SCALE®, a machine intelligence startup creating and deploying precision delivery technologies. The two leaders discussed the growing opportunity for artificial intelligence (AI) in health care and unique considerations that make its successful application different from other industries.
Question: What differentiates health care from other domains where big data is used?
John Santelli: The universe of data we work with in health care is incredibly complex. Information is captured in insurance claims, electronic medical records, lab results, pharmacy databases — sometimes even information an individual agrees to share, such as their ZIP code or data from a fitness tracker, can offer useful insights into the type of care he or she should receive.
The supply is vast, yet a limited amount of data is relevant for making predictions for patients. In addition, the data is often fragmented, noisy or siloed. Some of it is in the form of images or text notes or biomarkers.
Zeeshan Syed: There are several important differences between health care and other domains where big data is used. For example, the enormous variety of data, challenges of noise, missingness and reporting biases, longitudinal and dynamically evolving information, and treatment and provider effects. But there are two differences that stand out.
The first is that, unlike many other domains, in health care, “small data” is a bigger problem than big data. Even with large populations, once we filter for relevance and specificity to a particular disease condition or provider, we end up with small sample sizes. This limits the ability of techniques that might work in other domains to deliver value for health care.
The second difference is that our problems exist on an entirely different scale from other domains. In health care, the stakes are very real and very high. If I’m recommending films on a streaming platform, I can give five bad suggestions and it doesn’t really matter. If I’m generating ads, I can show 100 ads and if someone clicks on, say, 10 of them, I’ve still done a good job. But with health care, we’re dealing with people’s lives. We have to achieve a new level of accuracy that you don’t need in many other domains. As important as it is to get things right, it is equally important to get nothing wrong. The distinctive needs of health care make the playbook for traditional analytics and conventional artificial intelligence and machine learning obsolete.
Question: Why is health care so well suited for cutting-edge artificial intelligence projects?
Syed: There is a tremendous opportunity to address the inefficiencies that plague health care today through the right kinds of AI designed from the ground up for these challenges. At HEALTH[at]SCALE, we’ve devoted ourselves to continuously pushing the envelope on health care-specialized AI and machine learning (ML) as a means for sustainable and affordable progress for patients, health plans and care providers. We use our models and algorithms to target multiple opportunities along the care continuum to improve the management of complex populations across complex care networks and uncover actionable insights at a scale that is impossible for a person to do individually.
Santelli: We spoke a minute ago about the complex data that stakeholders across health care generate. Zeeshan mentioned “complex populations” and “complex care networks.” There’s a theme there — complexity — that today’s AI is very well suited for. By designing AI applications that address specific challenges in health care, we’re able to boil down a lot of information into something actionable. There’s no shortage of opportunities for improvement in the delivery of or administration of care, and there’s no shortage of highly skilled people who want to make a difference. The right forms of AI enable those professionals to focus their energy where it will make the most impact, either by flagging data that need closer inspection or reducing the administrative burden by automating repetitive tasks.
Question: Do you need health care experience to successfully manage a health care AI project?
Syed: As we discussed earlier, health care is a very different application domain for AI than other areas. The margins for errors are small, and you absolutely cannot afford to get decisions wrong. Making predictions requires dealing with a lot of very heterogeneous and complex information — the data elements John described earlier — ranging from traditional health plan claims and data in the EMR, to less structured but highly valuable clinical notes, imaging and lab results, and social and behavioral data. We have patients, treatments, disease and providers that are all evolving over time. And any solution needs to be delivered with an understanding of the unique operational constraints that have to deal with providing care, financial incentives and many other such factors. Based on all this, we cannot simply repurpose AI from across other application domains and expect success. It’s a very different ballgame.
Santelli: I agree. Unlocking the true potential of AI in health care requires a deeper appreciation of the unique challenges and needs within our industry — so we can solve the real problems. Fundamental innovations in both process and technology are required for most applications, which require this business context. It also impacts the mix of talent on our teams. There is strong evidence, from early work by Optum and HEALTH[at]SCALE, that bringing our health care-specific experience and resources together with our deep AI expertise and experience can achieve improvements in outcomes and costs for all stakeholders — patients, care providers and payers. The mix of health care clinical and business experience working alongside IT, analytics and data science professionals is generally more important than the underlying technology.
Syed: Creating the right team is easier said than done. There is a growing gap between the need and availability of data science talent. Working with data and applying AI is a lot like doing surgery — you must have the right tools specialized for different parts of a surgery, but you also need skilled individuals who can use these tools, who can effectively train, evaluate and deploy them. AI is not self-contained; it needs the right people and processes around it. Unfortunately, the need for data science talent greatly exceeds the ability of top academic programs to train individuals. This has not been helped by many top faculty across the country leaving to join industry, which has made it even more important for industry to provide apprenticeship models and other resources to develop their talent on the job.
Unlocking the true potential of AI in health care requires a deeper appreciation of the unique challenges and needs within our industry — so we can solve the real problems.– John Santelli, CIO of UnitedHealth Group, EVP Optum Technology
Question: With the shortage of data science talent, how are you attracting these professionals to work in health care?
Syed: From the time I was a freshman in college, I wanted to do something useful with my life and drive positive societal impact — and many of the students around me felt the same way. AI and ML are a fabulous outlet for this. The ability to interpret data and build models is a highly desirable skill in today’s world, and I think there’s no better place to sharpen that skill set and make a true positive impact to society than in health care. There will always be new levels of performance we can achieve — each percentage point we add to our prediction accuracy means more real people living better lives and less waste bogging down our system. Health care offers career opportunities that are intellectually stimulating in terms of both solving difficult problems and doing so with novel, forward-looking technologies. That’s a very powerful combination.
I’m extremely proud of the team we have at HEALTH[at]SCALE, with an extraordinarily talented and committed group of individuals working together to shape the future of health care. For HEALTH[at]SCALE specifically, we also have the benefit of working with Optum, a company that is very established and has deep experience. The complexity in health care — the different forms and quality of data, the intricate connections between symptoms and diagnoses, and a quickly evolving regulatory environment — makes it important to find opportunities at organizations that live and breathe health care every day. With our predictive model for Medicare Advantage (MA), Optum and HEALTH[at]SCALE have already launched what may be the largest deployment at scale of machine learning in health care. We’re helping to find the right care setting for over 4 million MA members. Looking at their specific needs, we can direct them to the right provider at the right facility. It’s been out there for over a year.
The distinctive needs of health care make the playbook for traditional analytics and conventional artificial intelligence and machine learning obsolete.– Zeeshan Syed, CEO, HEALTH[at]SCALE
Santelli: Optum has been building vast stores of longitudinal and real-time patient data for some time now. In addition to using this information to provide better care for our own patient populations and those of our partner health plans, health systems and employers, we de-identify it for research purposes. Our database is comprised of clinical, claims and employer benefits data spanning nearly 240 million de-identified lives. That’s a rich data source for data scientists in the sense that they can train their models on real assets rather than simply “toy data” that are synthetically produced and may contain biases or miss crucial connections. The results are not theoretical — our work is put into practice to make concrete advancements in medical practice and the administration of care. Because of our scale, that can have a real societal impact.
We know we need to remain at the forefront of leveraging data and analytics to improve health care, so Optum invests over $3.5 billion annually in research and development. The solutions and services that our teams offer are all infused with OptumIQTM — the combination of curated data, leading analytics and applied expertise — because we know that’s the key to reduced friction and better care. We’ve developed our own internal data-science education programs so that our people continue to hone their technical skills while never losing sight of our foundation in health care. And we’re always looking for partnerships with true luminaries whose goals align with ours.
Question: What opportunities/applications of AI are data science professionals at Optum and HEALTH[at]SCALE exploring?
Syed: When applying AI, it’s important to have an actionable use case. Projects should identify specific actions and interventions that can achieve improvements in outcomes, costs, access, satisfaction and/or equity. For example, we do not need AI to tell us that patients who are physiologically frail, have lots of comorbidities, are advanced in age and have undergone frequent hospitalizations are at increased risk. What we need is help understanding how these patients can be managed effectively and how to intervene with patients who may be moderate or low risk to prevent their progression to high risk.
Santelli: Optum already has significant experience with AI, and our internal teams are developing new forms to benefit all corners of the health system. It’s the type of fluency you can only achieve by sitting at the epicenter of health care. Through our work across population health management, health care operations and even care delivery itself, we know there are even more ways that AI can help us and society at large.
With so much opportunity out there, we recognize the value in enhancing our own capabilities by tapping into additional academic and business talent through HEALTH[at]SCALE. The collaboration enables us to work with a company whose top talent is engaged in advanced machine learning. They appreciate all the subtleties and challenges of doing this work in health care and share our commitment to help people live healthier lives and make the system work better for everyone. In doing this, Optum is applying its size and scale to nurture new talent while taking the industry forward.
Question: What are some of the other projects you’re currently working on?
Santelli: In addition to the project to navigate MA members to the right provider, we’re also using predictive algorithms to look at Medicaid and MA members who have multiple chronic conditions to get clues about when they might be headed to the emergency room or to the hospital.
Syed: That’s right. And we’re not just predicting when they will go to the hospital — we’re predicting why. Then we take that data and make it actionable. We can do an outreach and take care of the health plan’s member ahead of the adverse event happening. We’ve put it in the hands of care managers, who can access it in seconds.
There’s a lot of discussion and evangelism right now about the promise of AI, but in a sense that’s yesterday’s news. We’re out there putting it to work now, and as we look to continue growing our team and capabilities, we want aspiring data scientists and machine learning engineers to know there’s a place where their skill set can make a real difference in somebody’s life. Not in some far-off future, but today. And it’s important we bring this talent together with clinical and business expertise to drive maximum impact as we continue to innovate and design new AI applications to improve health care.
Taking part in the conversation
CIO UnitedHealth Group, EVP Optum Technology
John leads Optum Technology, which serves the broad customer base of Optum and Minnetonka, Minnesota-based UnitedHealthcare®. John also serves as UnitedHealth Group’s chief information officer, where he is responsible for technology strategy and delivery across the enterprise. John joined UnitedHealth Group in 1986 and has extensive experience in software engineering and application development. He has served as CIO since 2007 and is credited for leading enterprise architecture and foundational technology capabilities within UnitedHealth Group.
Founder and CEO, HEALTH[at]SCALE
Zeeshan is the founder and CEO of HEALTH[at]SCALE, a machine intelligence startup working with leading health plans, provider systems and self-insured employers to predictively optimize complex care decisions. He previously was a clinical associate professor at Stanford Medicine and an associate professor with tenure in computer science at the University of Michigan. Zeeshan was also part of the team that launched the Google[X] Life Sciences initiative (now Verily). Zeeshan is a recipient of multiple awards for contributions to health care machine learning, including a CAREER award from the National Science Foundation. He holds a PhD from MIT EECS and Harvard Medical School in computer science and biomedical engineering, and MEng and SB degrees in EECS from MIT.