The Road to Wellville
How Tepper School alumni and faculty are reshaping health care
Access to health care has been a prime issue for voters and politicians in the last few election cycles. New innovations are reforming the industry toward care that is more individualized than ever before.
Take an industry that constitutes almost one-fifth of the U.S. gross domestic product, add a generous dose of rapid-fire technological advancements, a complicated relationship with public policy, and a high-risk, high-reward value proposition, and you have what could be the perfect incubator for business innovation: health care.
It touches virtually every person in the nation, and its sheer size and complexity have proven daunting for some of the most astute thought leaders society has to offer. Health care is an 800-pound gorilla we all hope to tame, or at least harness, with the end result of helping people live longer, more comfortable lives. It means using data analysis to find cures for rare diseases that once were overlooked by the system. It also means finding solutions to really big problems, including such behemoth targets as organ transplants, cancer, and cystic fibrosis. As health care continues to grow — it is expected to hit 20 percent of the GDP within the next five years — it has become a favorite target for some of the most robust research and business applications in recent memory.
At the forefront of this research are Tepper School of Business faculty who study health care problems across a variety of disciplines. And from diagnosing diseases at the earliest possible stages to finding the best possible drug therapy for an individual patient, Tepper School alumni are making significant strides in advancing the field.
Tepper’s Quantitative Approach
“You’re looking for solutions or ideas that are robust enough that [they] can continue to be useful, even as the landscape changes,” says Alan Scheller-Wolf, Richard M. Cyert Professor of Operations Management, Senior Associate Dean for Faculty and Research. “This is a field where you can see there is a large potential to have a direct, immediate impact on people’s lives — often people in significant need.”
The Tepper School, which specializes in using quantitative analysis to untangle and clarify dense, intricate issues, is uniquely suited to train the minds that are disrupting health care and adapting to the shifting forces that are shaping it.
“We are not the physicians, but we are giving them data-driven methods, including machine learning algorithms, for making decisions to make their lives better — which means they make their patients’ lives better,” explains Sridhar Tayur, Ford Distinguished Research Chair, Professor of Operations Management.
The stakes couldn’t be higher. Even as costs climb, the U.S. population continues to age, leading to a greater need for more complex care. Meanwhile, an increasingly savvy public is demanding more precise, personalized care and greater accessibility — all forces that are shaping the market and driving demand for better solutions in both the public and private sectors.
“It’s a huge challenge. It’s a huge amount of dollars,” says C. Talbot Heppenstall Jr. (MSIA 1985), President of UPMC Enterprises and Executive Vice President and Treasurer of UPMC. Those two factors combined are why tech giants such as Apple, Microsoft, and Amazon are eager to invest in health care, he notes. Likewise, UPMC has partnered with Carnegie Mellon University and the University of Pittsburgh to form the Pittsburgh Health Data Alliance, which is focused on harnessing big data to better predict, prevent, and treat disease.
“I’m 58, and the health care industry will not run out of problems in my career,” he adds.
Scarce Resources, Better Distribution
For Tayur, health care has been fertile ground for optimization research for years. He co-authored a handbook to help guide research in health care operations, including a comprehensive explanation of the industry’s organizational structure as well as specialty topics.
It also led him to an entrepreneurial venture, OrganJet, designed to help alleviate a pressing logistical problem: matching transplant patients with donor organs across geographical regions. A current project with the United Network for Organ Sharing creates data-driven methods for optimal allocation of the scarce livers that are available for transplants to the most suitable patients.
In the United States, approximately 115,000 people are currently on a waiting list for a lifesaving organ transplant, according to the American Transplant Foundation. Only about 30,000 transplants occur each year. It is widely known within transplant circles that the time a person waits for an available organ varies widely depending on where they live; wait lists on the East and West coasts can be double those of the Midwest.
Tayur, who previously had no background in transplants but had created algorithms that optimized scheduling for time-shared private jets, thought he could apply the same principle to assist transplant recipients and patients. The result was a new company, OrganJet, which provides both an information service and air travel to connect waiting patients with matching organs.
After that, Tayur began exploring new projects. Understanding both supply and demand and the tradeoffs between efficiency and fairness, as well as providing objective guidance on data, allows doctors to do their jobs better, he realized.
To date, through a combination of its advisory services and transportation options, OrganJet has helped more than 300 patients who were waiting for a transplant, according to Tayur. And the impetus for the company’s formation has inspired him to branch out by exploring other health care optimization problems.
One of his current projects includes developing data-driven methods for the United Network for Organ Sharing to use in determining the optimal distribution of livers for transplants. In patients who have liver cancer, Tayur and his collaborators at Carnegie Mellon and Massachusetts General Hospital created a scoring scheme that calculates information about tumor growth and other factors to determine which patients should be prioritized for transplant.
Tayur created a scoring scheme that factors in whether a patient has a complicating condition, the size of the patient, the size of the liver, and other information and then helps prioritize the matches.
In a similar line of research, Scheller-Wolf has studied ways to optimize the distribution of blood platelets. The idea is to create a policy model that distributes the freshest platelets to those in greatest need, while blood that is older but still viable can go to patients who will still benefit, reducing waste and improving outcomes.
In addition to his work in optimization, Tayur serves on the scientific advisory board of Mitra Biotech, a Boston-based startup backed by venture capitalist James Swartz (MSIA 1966). The company is developing tools to help personalize cancer therapy by better defining both the tumor and its microenvironment.
Oncologists send a live sample of a malignant tumor along with the micro-environment around it to the Mitra lab, which tests various drug combinations on the sample. Algorithms then analyze a huge unstructured data set, including information on genetics, how the tumor changes, cytochemistry, and other factors, before calculating a score.
If the score is over 25, the drug combination is likely to help the patient, with a prediction accuracy of about 87 percent. If it scores under 25, it is highly unlikely to help. Mitra sends the scores back to the oncologist, helping determine which treatments to give the patient.
The analysis results in “incredible cost savings, but more importantly, you’re not wasting time” by administering the wrong treatment, Tayur says.
“This is an example of what CMU folks do: We do data-driven analytics,” he says. “I tell people, ‘I don’t know how to do transplants, but we can help you decide who to transplant.’ Additionally, we are able to increase the number of transplants by reducing organ discards.”
From Netflix to Nanoparticles
Andrew Li, Assistant Professor of Operations Research, applies his expertise in optimization and machine learning to cancer as well as Alzheimer’s. He worked on new liquid biopsies, which are blood tests that can detect diseases in their earliest stages by measuring thousands of proteins all at once. The technique involves injecting the patient’s blood sample with nanoparticles that stick to the proteins associated with a tumor, or with Alzheimer’s.
By measuring the amount and type of proteins stuck to the nanoparticle, the test yields what Li calls a sort of fingerprint of the disease, which artificial intelligence can then analyze to predict the diagnosis.
Mathematically, the analysis is the same as systems that companies such as Netflix or Amazon use to recommend movies or products to customers. It gathers data based on a user’s behavior: clicking on a particular item, watching a trailer, putting a product in a virtual shopping cart. Known as tensor recovery, the process builds a matrix of the data it collects to make its predictions. Data from a single person is not very accurate, but the accuracy improves dramatically when data is collected from many people and machine learning refines it.
“The general work I like to do is mathematical, but it also has impact,” says Li, who also has worked in e-commerce applications. “I think that was the appeal of getting to apply these same techniques. It was a totally different setting, and possibly more impactful.”
As technology improves in its ability to process massive quantities of data, the predictions become more precise; five years ago, the measurements from the nanoparticles in the liquid biopsies would not have been possible.
But the challenge in applying intelligent data analysis to health care now lies in collecting enough data online to make the analysis possible, Li notes. Privacy laws restrict how much of that information can be shared, posing another problem that business may seek to solve. Currently, the companies competing in the liquid biopsy space are competing in the area of data collection — and Li predicts that will be true for the next five to 10 years.
Narrowing the Targets
While Sujal Shah (MBA 2004) thinks the term “personalized medicine” is overused, he does believe that health care has become more customized in general.
Shah serves as President and Chief Executive Officer of CymaBay Therapeutics, a biopharma company that develops drugs for liver and other chronic diseases. A better understanding of the biology and genetics of the patient’s condition does allow for more targeted therapies, he notes.
“I do believe there is a significant opportunity across drug development for better understanding of how various patients are going to respond to your drug, both in efficacy and safety,” he says. And artificial intelligence helps the pharmaceutical industry understand the differences in patient response.
For example, if patients are measured for 1,000 parameters, AI can help discern which of those factors contributed to a bad reaction, and the patients can be screened for those factors to prevent similar responses in the future, he says.
Caroline Loewy (MSIA 1990), who serves on the board of directors for CymaBay and several other biopharma companies, says AI can also help drive the development of treatments for rare disorders. Genetic sequencing can help narrow down the causes of these conditions, and big data helps in particular to isolate a mutation from a sequence that includes billions of nucleotides.
“It’s still, in many cases, more of an art than a science,” Loewy says, but the process is more efficient than it used to be.
Loewy’s son, James, is among those who carry a rare genetic disease — in his case, a mutation on the KCNQ2 gene, which causes recurrent seizures and severe cognitive impairment. When he was first diagnosed, there were only 10 or so people worldwide who were known to have the same condition. Today, there are about 700.
Developing drugs for rare diseases is costly, but Loewy points out that companies might be able to run more efficient clinical trials or repurpose existing drugs for these indications.
“For a more targeted drug for a smaller population to be commercially viable, there has to be enough value and a clear change in the course of the disease,” Loewy says.
Better Teams, Better Outcomes
Vertex Pharmaceuticals targets cystic fibrosis and currently manufactures drugs that treat the cause for about half of the people who have the disease. Another 40 percent could benefit from a drug the company plans to file with the U.S. Food and Drug Administration later this year, according to Amit Sachdev (BSIM 1990), the company’s Executive Vice President and Chief Regulatory Officer, and the remaining 10 percent would require genetic therapy. The company is also interested in pursuing a treatment for sickle cell disease, a genetic disorder that Sachdev considers part of the early frontier for gene editing.
“When you think about drug discovery, it’s all about solving problems,” says Sachdev, and a deep understanding of business, particularly through a quantitative lens, is helpful in achieving success. “I think that Carnegie Mellon trains some of the best and brightest.”
Problem solving also can extend beyond machine learning. Linda Argote, the David M. Kirr and Barbara A. Kirr Professor of Organizational Behavior and Theory, has studied the optimal makeup of teams in hospitals for creating the best outcomes. She focuses on transactive memory, which is how well each team member understands who is good at what tasks.
“I think hospitals are interesting from a societal as well as a research point of view,” says Argote, because the stakes are so high — both for patients, who don’t want to be there any longer than they have to be, and for hospitals, which want to use their facilities efficiently and to their best effect.
In one of her current projects, with Drs. Jeremy Kahn and Matthew Rosengart of UPMC and Tepper Ph.D. students Jerry Guo and Ki-Won Haan, Argote is studying interactions in teams working at trauma units in hospitals that treat the most injured people. The research team examines whether trauma teams that have a well-developed “transactive memory,” or knowledge of who knows what, perform better than teams lacking such knowledge. Knowledge of who knows what enables teams to assign tasks to their most qualified members, to trust each other’s expertise, and to coordinate that expertise effectively.
Performance is assessed by a variety of clinical outcomes, including how long it takes for the patient to get to the next stage of care, such as a surgical or intensive care unit, and how long patients spend in that next phase as well as their entire length of stay at the hospital. Time is a critical factor in determining the patient’s clinical outcomes. The initial data indicate that having a stronger transactive memory resulted in patients’ having a shorter length of stay. Later, the research will examine what promotes better transactive memory, including the composition of the team, their experience working together, and their experience at their tasks.
Ultimately, Argote would like to help hospitals determine how to optimize teams to improve patient care outcomes.
“Emergency care is not an individual sport; it’s a team sport,” she says. She describes the goal of the research as “developing guidelines for how teams can work together effectively.”
Scheller-Wolf similarly has studied factors impacting quality of care, including the number of individual caregivers the patient sees. The higher the number of new caregivers, the more often the patient has to explain their needs, which can create inconsistency of care at hospitals and long-term care facilities.
“A huge part of the problem is that people call off, and then you get any random staff member as a replacement in there — or the facility calls an outside pool, and it’s automatically a new person,” says Scheller-Wolf.
One of his students, Vince Slaugh (Ph.D. 2015), came up with the idea of creating a voluntary on-call pool that draws from in-house staff already familiar with the patient. In another project, Scheller-Wolf found mathematical models that optimize nurse phone lines that people can call with questions about non-urgent medical concerns. The idea is that such services can offload demands for emergency services.
Overall, the health care industry continues to grow more complex, Scheller-Wolf notes.
“With the advent of medical advances, with the advent of more information, the problems just become more complicated,” while “money continues to be tight,” he says.
A Road to Single Payer?
Heppenstall, the UPMC Executive Vice President, contends that the crux of the problem lies in the way health care is financed in the United States — which also contributes significantly to the dearth of data that Li points out.
Most of the 6,000 or so hospitals in the U.S. are nonprofit, and since they compete locally, they don’t want to share data, he says. And the current tax structure, which puts employers in the middle of the funding conversation, is another problem that is driving inefficiencies.
“You don’t have to look too far to see the U.S. health care system is broken,” he says.
Katherine Kohatsu (E 1997, MBA 2005), a principal with PricewaterhouseCoopers who has focused exclusively on health care for the past eight years, says efforts to reduce costs and administrative burdens typically draw a lot of pushback.
“There’s that tension in the system in terms of who gets the dollars,” she says. “At a macro level, I think we’re all trying to do better, but at the individual level, who bears the burden of cost reductions? Also: How do we manage costs but at the same time get better outcomes?”
For Heppenstall, the answer is simple: Public policy surrounding health care will change, because it is unsustainable in its current form.
“We’re on a path that we can’t afford,” he notes, and the U.S. population demographics explain why.
Longer lifespans and the aging of the baby boomer population are both contributing to more people entering the Medicare system, while fewer people are left behind them to pay for their care, particularly as it becomes more complicated in old age.
Kohatsu cited a regional plan’s Medicare program, which has been segmenting its care teams for people from 65 to 85 and then for people who are 85 and older.
Regarding a single-payer system, she suggested that states may take steps toward single procurement systems, in which the government negotiates what prices they’ll pay for prescription medications for their Medicaid beneficiaries — a model that is being tested in California. Louisiana is going even further and investigating a subscription model for unlimited access to hepatitis C drugs.
“I think there are a lot of levers there,” she says, “That may be the impetus by which you go to single payer.”
Kohatsu notes that when she graduated from the Tepper School, she spent time consulting with other industries before she made the jump into health care. Like many fellow alumni — regardless of what aspect they work in — she says she stays because she wants to help advance the field.
“I feel like I can sit back and think about this big picture that I’m trying to move and solve, and I feel like I’ve had a small, tiny part in it,” she says. —
by Niki Kapsambelis