Innovator CEO profile: HEALTH[at]SCALE’s Zeeshan Syed

Syed discusses the company’s ‘precision care delivery’ model, demonstrating industry-leading results in reducing costs of care, identifying patients at risk for adverse events, and more.

Health Evolution | February 23, 2021

When Zeeshan Syed and partners founded HEALTH[at]SCALE Technologies in 2015, the mission was to advance ‘precision care delivery’ and move away from population-level cookie-cutter approaches for managing patients and toward more sophisticated individualized care delivery focusing on one patient at a time.

Health Evolution interviewed Syed about what inspired him to start the company, health data’s unique machine learning needs, building a world-class team with expertise in machine learning and health care to solve some of health care’s most difficult problems and more.

What is the HEALTH[at]SCALE origin story? Or what inspired you to start the company?
Health care is personal to me. About six or seven months after I had finished my Bachelors and Masters in Computer Science at MIT, I received a call in the middle of the night from my uncle informing me that my father had just suffered a major heart attack. As a diabetic, my father had a silent heart attack without any of the stereotypical heart attack symptoms. He did not recognize that he was having a heart attack at the time. It was another day or so before he went in to see a doctor and by that time unfortunately a lot of damage had been done. It was frustrating to me at the time and continues to bother me to this day that we have millions of patients who have heart attacks each year, but cannot get ahead of these health events, manage or prevent them, and at the very least, educate high-risk patients about symptoms for both stereotypical and silent heart attacks so that they can get the right, timely care. I channeled that frustration by going back to MIT’s School of Engineering and Harvard Medical School and getting a PhD through a joint program between the two institutions researching ways in which machine intelligence can help in places where our delivery system is failing. Upon graduating, I joined the University of Michigan’s Computer Science department as a faculty member in AI and Machine Learning and ran a successful lab funded by the National Science Foundation, National Institutes of Health and the American Heart Association. After getting tenure, I joined Stanford Medicine as a faculty member in the Stanford University School of Medicine and a director at Stanford Health Care. A lot of the inspiration for HEALTH[at]SCALE came from seeing first-hand the inefficiencies in our health care delivery system, spending time with a group of like-minded people, and understanding that we needed technology to solve some of the hardest problems in health care delivery. Ultimately, working with other colleagues who felt as strongly about re-inventing health care as I did, we decided to found the company to use what we’d learned about machine learning in health care to have a greater impact on patient outcomes.

There are many problems machine learning can be used to address. What are you focusing on first?
The fundamental problem we are solving is that the existing health care system is designed to treat an “average” patient, who in many cases may not even exist. Care delivery at present is imprecise and impersonal, and lacks a deep and clinically nuanced understanding of the patient’s individual needs. This lack of personalization and relevance has made existing care delivery a blunt instrument leading to worse outcomes, lower satisfaction, decreased access and higher costs. For example, the journal Nature had a commentary back in 2015 highlighting how the top ten highest grossing drugs in the U.S. at the time failed to benefit between 75% to 96% of the patients who are on them. Our own recent studies show that roughly 80% of patients who seek specialist care in the U.S. may be sub-optimally matched to providers within their local geography. Then there are results from other studies, such as the Camden Coalition in 2020, finding that hotspotting approaches targeting outreach to high past utilizers do not improve readmission rates. Existing care delivery, whether you look at the choice of treatments, providers, or outreach, is imprecise; and therefore ineffective.

HEALTH[at]SCALE’s mission is to disrupt this status quo and re-invent care delivery as a precise and personalized process that is deeply aware of each patient’s unique health characteristics and needs. There has been a lot of work done previously on precision medicine, where the goal has been to develop new treatments customized for individual patients. We are now taking that further, by getting away from the development of new treatments and toward the question that we believe is even more important — how do we take the treatments and resources we have available now, and deliver them in the most effective way customized for individual patients. If we can truly understand each patient in a richly contextualized way, with the ability to forecast how different care decisions might impact their outcomes, there is a substantial opportunity to move the needle on outcomes, costs, experience and access.

In addition to personalization, the other problem we are keenly focused on is optimizing long-term outcomes rather than short-term costs. There has been a growing interest recently in value-based care and cost transparency, which in itself is a good thing, but many of these efforts obsess over what happens in the first day or few days of a healthcare encounter with very little to no insight into what happens over the next months and years. In many ways, this is a classic penny wise, pound foolish approach failing to appreciate that near-term decisions, for example, picking the cheapest option for getting a surgery, imaging or other clinical service, can end up being highly sub-optimal in the long run. Our work at HEALTH[at]SCALE looks instead to understand how care decisions might impact longitudinal outcomes for individual patients, and use this information to guide care delivery that ultimately achieves meaningful and lasting improvements in outcomes, costs, experience and access.

When we started the company in 2015, it was very much focused on building a solution to enable precision care delivery and for us that means moving away from population-level guidelines and “day one” costs to manage patients with a much more clinically nuanced understanding of how care can be unique, predictive, outcomes-based, and longitudinal. That’s what we are doing today, working with plans, employers and providers that are taking on risk to apply these capabilities and substantially improve outcomes for the populations they serve.

As we look towards precision care delivery, which requires a deeply clinical nuanced understanding of the patient, one of the biggest challenges is how to solve the “small data” problems to develop an understanding of an individual patient with an individual provider for a specific condition at a particular point intime.

Zeeshan Syed , HEALTH[at]SCALE

Machine learning is recognized as a game changer in other industries. Why hasn’t it had a similar and broader impact in health care?
Machine learning is being used as we speak by several other industries. From stock trading and detecting credit card fraud to targeting ads and recommending movies to detecting network intrusion and powering self-driving cars, there has been growing adoption and impact of machine learning in other industries. In health care, machine learning is yet to have the same broad impact, and I think that comes down to a few things.

One is that health care is an entirely different challenge than these other domains. It requires a different set of machine learning technologies than the ones that might be usable across industries for other applications. Both the nature and the amount of data available in health care differs from other domains. Building successful machine learning systems in health care requires dealing with an extremely diverse variety of data, all the way from ‘omics’ to claims to images to notes to time-series. In addition, the amount of data we have in health care, for an individual patient or provider is often severely limiting. More than the ‘big data’ problems that are the focus of other industries, health care requires us to solve the ‘small data’ problems where robust inference is needed for individual providers who do tens of a specific surgery each year, or specialty drugs that are infrequently used, or to model outcomes in populations with specific combinations of clinical characteristics. These challenges make it harder to drive impact in health care by simply re-purposing approaches from other domains, and create the need for specialized machine learning technologies that are primed for the kinds of datasets available for health care use cases.

The other reason, in my opinion, for why machine learning has not yet had the same broad impact in health care as in other industries is talent. Machine learning is an advanced scientific discipline, and while other industries have embraced this concept and gone hard after recruiting PhDs and Masters students from top colleges and universities, health care is far behind. HEALTH[at]SCALE is unique in that we have brought leading machine learning experts together, including top faculty and students from top machine learning departments around the world, to take on some of the hardest machine learning challenges facing health care. But we are probably the only health care company that has managed to successfully compete against the big tech companies to recruit top talent. Others have struggled with this, and especially if you look at the large payers and providers out there, much of their machine learning is still being done in-house by teams that do not have formal training on what is a highly nuanced and advanced discipline. For machine learning to have broad impact in health care requires the right talent, and with some exceptions like HEALTH[at]SCALE, which has managed to productionize machine learning at scale for millions in live production deployments, much of this impact is gated on talent.

These challenges aside, machine learning is an amazingly powerful technology that allows us to move away from subjective decision making toward much more robust data-driven approaches that can provide personalized insights at an unprecedented scale. Humans struggle with data that are high dimensional. So anytime we’re reasoning about hundreds, thousands, or tens of thousands of parameters that are simultaneously changing, such as those that are related to an individual’s health over time, that is the sort of situation in which humans struggle to make decisions. When dealing with data from millions of patients, collected over long periods of time, it is even harder for humans to make decisions that optimize outcomes. Machine learning has the potential to allow us to manage these kinds of challenges and understand what is happening over time, across large numbers of patients, and bring that knowledge to bear on the problems that need to be solved. Then there is the whole automation side that allows you to automate decisions that generally are prone to human error.

You discussed precision earlier and that often conjures ‘precision medicine’ or even ‘precision health’ but HEALTH[at]SCALE describes itself as enabling ‘precision care delivery.’ What should CEOs understand about the concept?
Precision care delivery, at the high level focuses on moving away from the one-size-fits-all, cookie-cutter processes for delivering care, which are grounded the implicit assumption that all patients are alike, and that what works well or poorly for one patient will work well or poorly for everyone. With precision care delivery, we are focused on understanding each patient and identifying the next best action customized and specific to each individual’s unique health conditions and needs. Some of the problems we focus on include determining when an individual needs care, which provider should provide that care, where they should provide that care, and what constitutes low vs. high value care for that individual. In each case, our focus is on advancing care delivery so that it understands the diversity within large populations rather than ignoring it.

While precision medicine focuses on developing new treatments customized for each individual, precision care delivery focuses on using treatments and available resources in a way that is customized for each individual. Both are essential, in that we need better treatments, but we also need to optimize the delivery of both new and existing treatments so that their full potential can be unlocked to improve outcomes, costs, access and experience. While precision medicine has been a focus for many years, HEALTH[at]SCALE is establishing precision care delivery as a critical resource for payers, employers and providers, and prompting a re-imagining of how care should be delivered.

In our work on precision navigation, for example, we are moving away from the existing paradigm where patients consider providers based on process-based metrics, reputation rankings, high volumes, and other metrics that incorrectly assume that provider performance is the same across all patients. Instead, precision navigation looks at providers in a hyper-personalized, outcomes-based way, providing each individual with a personalized rating of providers in the geography where they want to receive care. So, for example, when I have to find a provider to do a hip replacement for me, I should be looking at someone who is most likely to achieve the best outcomes for me individually given my unique calculus, not for a provider that might be rated as five stars because of process-based metrics or reputation rankings, which are not personalized and not consistently associated with outcomes improvement.

Similarly, in other work we do on precision interception, we are moving beyond the existing paradigm of targeting outreach to patients who have high past utilization or costs: an approach that has recently been shown to not significantly improve outcomes. Instead, precision interception gets at the question of understanding which patients in a population are likely to be emerging risk or rising risk, and what their specific risks are, so that this can inform targeted and individualized outreach to get in front of adverse health events, rather than trying to claw back irreversible health deterioration after the fact. This type of solution requires a personalized understanding of each patient and which outcome that person might be at risk of in the future. It’s much more predictive and outcomes-based and respects the uniqueness of individuals.

Why is that manner of personalized care delivery so difficult to implement?
The question of what is the right next best action, whether it is an outreach, the choice of provider, or even a treatment, is a very hard problem. Analytics and conventional artificial intelligence and machine learning are incapable of dealing with that challenge. As we look towards precision care delivery, which requires a deeply clinical nuanced understanding of the patient, one of the biggest challenges is how to solve the “small data” problems in health care where you are trying to develop an understanding of an individual patient with an individual provider for a specific condition at a particular point in time. That requires very fine grained inference, where the datasets of relevant encounters for similar patients, getting similar treatments, from similar providers, for similar conditions might be small for conventional approaches. That has been a major challenge for existing analytics technologies. From my perspective, one of the things that makes precision hard, and has historically made it hard to achieve, is advances in machine intelligence to deal with these sorts of issues. That is where having a strong research background, especially the ability to advance machine learning to specialize for these challenges, is so important. We have been fortunate at HEALTH[at]SCALE in that we have been able to deal with that problem by hiring a team that includes some of the most widely respected faculty members and researchers in this space.

How does your professional, academic and research background, which includes time at Google, the University of Michigan, Harvard, MIT and Stanford, inform your strategy as CEO of HEALTH[at]SCALE?
The first thing I’ll say is that I want to credit the amazing team that is involved in all of our decision making as a company — from John Guttag, our CTO and a former chair of MIT’s EECS department and a distinguished professor in machine learning there, to Mohammed Saeed, our CMO, who has an MD from Harvard and a PhD from MIT and is on the faculty at the University of Michigan Medical School, to Zahoor Elahi, our COO, who previously spent eight years as SVP of Technology at UnitedHealth Group. Our team has health care and machine learning experience that is both broad and has depth in different areas, with a lot of specialized expertise that positions us well for our mission. For me personally, I have benefitted from being a tenured a faculty member in engineering at the University of Michigan, a faculty member in medicine at Stanford, a hospital administrator at Stanford Health Care, a software engineer at Google and Oracle, and most recently an entrepreneur at HEALTH[at]SCALE. Those experiences, and most importantly, having been a consumer of health care and a care provider to my parents and my father in particular, has been invaluable in informing our strategy at HEALTH[at]SCALE.

Obviously, 2020 was a tumultuous year so looking ahead what are you anticipating for 2021?
Health care has obviously come front and center and I think we will continue to deal with many of the spillover challenges from 2020 this next year. We postponed a lot of elective care and will need to think about how to prioritize that in a way that understands which patients most urgently need attention and which can safely wait to receive deferred care, which goes back to the notion of precision to determine the patients likely to have better or worse outcomes based on the scheduling of their elective care. Finding providers also becomes a bigger challenge because we have a system that is slammed with patients whose care was suspended and now need to resume, in many cases, with new providers. So guiding patients towards the right providers for them, which balances network load, will become even more important. As we deal with COVID-19 variants and other pandemics, having a precision model that helps identify the most vulnerable populations to selectively flatten the curve will remain important. Also, there is already a lot of COVID-related fraud, waste and abuse in testing, which will have to be addressed. And finally, as we have more and more people vaccinated and being treated for COVID it will become increasingly important to understand how these treatments are affecting patients long-term, and especially how different individuals may be affected differently by these treatments. It is my sense that as we look at the future, all of these things will continue to be major considerations in 2021 and beyond.

What career accomplishment as the founder and CEO are you most proud of?
I am most proud of our team and the impact we have delivered. We have brought together a world-class team with deep expertise in machine learning, software engineering, medicine and health care; and they are motivated and passionate about driving positive societal impact. At a time where machine learning talent is heavily competitive, we have been able to assemble a team that is, I would say, the best and bring them to focus their skills and attention on health care. I am also very pleased with the fact that we have taken a discipline and been able to demonstrate use cases that are very forward looking, such as how health care should be delivered with precision and focus on outcomes. I think it is a big win for us to re-imagine health care as something that is truly smart, integrated and outcomes-based. And we have developed highly differentiated technology that has shown real impact and results. For example, we are able to show that the providers who were rated highly by our machine intelligence substantially improve outcomes and costs; in a peer-reviewed study we found a 35% reduction in emergency department visits and a $3,300 reduction in total cost of care for patients receiving hip replacements. We are similarly able to identify patients who are likely to exacerbate in their health conditions with a precision of over 90 percent, which is industry leading. And we have been able to identify substantial amounts of fraud, abuse and waste in the system. So I feel very proud of the fact that we have been able to move machine learning toward deployments where we are now being used by populations in production settings — and we are able to deliver real impact to drive real results.

For more information about HEALTH[at]SCALE, complete the form below:

About the Author

Health Evolution, Staff Writer