by Chandler Watson
It was, of course, pouring that day – typical, if not cliche, of the Portland waterfront. Grappling with anxiety, I rubbed my list of questions between my fingers as the droning “slurrk? sheeek! slurrk? sheeek!” of the windshield wipers ushered in the start of another doused afternoon. Without warning, a well-kept, nondescript building passed on the right. Blue-grey windows peered under crimson archways at glass-covered buildings with neon-tube art and off-center architecture. No one in the car seemed to notice as the red-bricked monolith shot by. The GPS realized our mistake and we turned around. Soon, a sign quietly advertising the “3030 Building at South Waterfront” appeared. Twenty-one slots for engraving rested on its face, none of which bore any trace of the Oregon Health and Sciences University Center for Health Systems Effectiveness, or OHSU CHSE. It was already clear that the CHSE team did not make publicity a focus.
Indeed, the building itself appeared to face away from the rest of the city. Opposite the road, cascades of subdivided windows were latticed together, pointing along a quiet stone path from the front door toward the sea. Trees stripped of all foliage studded two dirt ridges on either side of the path, reaching purposefully with fractal branches into the afternoon sky. The lobby itself, though its roadside face was hidden by brick, framed itself with polished glass panels, opening to an immense brick corridor. Two saplings, both carefully wrapped in festive LED strings, stood by the entry, while an immense number of yellow-tinged light bulbs cast a sunny gleam from the ceiling onto a switchback staircase leading to a carpeted balcony. The charming hall I had entered was strangely quiet, not in the literal sense – there were the obligatory office noises, such as the stapling of documents and clacking of footsteps – but in the very nature of the place. The word unassuming seems to fit best.
After an eternity of door-clicks and rustled papers, a tall, well-dressed man exited a snowy-glass door, stepped onto the balcony, and called me to his office. I nervously jogged up the stairs to greet Dr. K. John McConnell, a brilliantly humble man, devoted father and husband, and trained appreciator of mathematical beauty. McConnell works as the head of OHSU CHSE with a primary focus on medical economics. His bright facial features are sharp and apparent, but held in a relaxed pose. Initially, his steely hair, deep focus, and confident stance – not to mention his mathematically sophisticated career – could portray an exclusively professional individual. Yet his instinctive grin, childlike curiosity, and serve-others-first attitude all suggest otherwise. At the top of the stairs, McConnell and I shared a solid handshake, after which he guided me to a cluster of rooms that looked like the offspring of Intel and a high-end apartment complex.
* * *
So what is CHSE? By leveraging the power of mathematical, economic, and statistical thought, CHSE aims to evaluate the effectiveness of medical practice and determine how to improve it. For instance, one of CHSE’s current projects is a health survey in South Dakota. With its large rural areas, many cultural divisions, and low hospital counts, South Dakota presents a challenge in delivering quality healthcare to all of the state’s residents. CHSE has responded by connecting with Native American tribes, families experiencing homelessness, refugees, and immigrants – groups typically more difficult to reach – in order to form a holistic view of how to deliver healthcare to all of South Dakota. They have also met with almost one hundred professionals, including epidemiologists and hospital leaders, and made two extensive on-site visits in order to understand the situation better. According to the Helmsley Charitable Trust, who provided funding for the project, this is the first research project of its kind that “assessed all healthcare needs… in rural areas,” deeming the information critical to “catalyze communication… among stakeholders, prioritize projects, implement effective services, and have an impact on rural healthcare.” Yet the members of CHSE aren’t the first ones to leverage mathematical thought like this.
Even though CHSE was conceived just four years ago, the idea it represents is centuries old (“Curriculum Vitae”). Back in 1654, two famous mathematicians, Blaise Pascal and Pierre de Fermat, began correspondence about a troubling problem. An Italian gambler known as Chevalier de Méré had proposed his version of a dice-throwing game to Pascal. In this game, two players would gamble over dice throws, but partway through the game, one player could stop and leave with his winnings. The game itself meant nothing to either mathematician, but the ideas behind how to win would form the basis of modern statistics. After a series of breakthroughs in their correspondence, Pascal and Fermat defined the core idea of expected value (“Fermat and Pascal on Probability”). And it means exactly what it sounds like – if you were to play the game over and over, it predicts how much you’d be expected to earn, on average.
Imagine a certain medical practice is considering a new medicine for anxiety treatment. Let’s say this medication works just as effectively as other leading brands in three out of five cases, but also saves an average of $100. Unfortunately, in two out of five cases, it proves ineffective, costing an additional $150 before switching to another medication. To calculate the expected value, we just multiply each gain or loss by how often it happens. A back-of-the-napkin calculation shows that three-fifths of $100 minus two-fifths of $150 is equal to negative $15 – your expected loss per patient. A statistician might then conclude that changing medicines would be a poor choice, and save the practice thousands of dollars in its first few hundred patients. Of course, most medical problems are much more complex, typically involving massive databases of patients and many more variables. In response, statisticians over the centuries have come up with even more clever techniques to deal with more complicated data.
* * *
Following McConnell into the warm CHSE office, I noted all of the features of a formal yet cozy office environment. Incandescent lighting provided a surreal feeling in the workplace as McConnell and I glided through. To my left stood a tiny kitchen with an assortment of china and plastic cups resting on a drying rack. Ahead lay a tiny, well-kept cube farm where sharply-dressed programmers, economists, and statisticians ground away at various problems. A single aisle cut through the middle of the office with a clear view of rain-soaked Moody Avenue. After offering me a drink, McConnell guided me to a pair of cubes where he introduced me to Stephanie Renfro, MS and Benjamin Chan, MS, two of the fourteen individuals in CHSE. Both are classified as “Research Assistants” on the group’s web page, but Renfro and Chan are constantly working hands-on with massive data sets, tailoring statistical models to fit them and reveal more information. McConnell made clear that the groups and individuals at CHSE tend to work very synchronously – whenever a new problem is presented, the team takes a divide-and-conquer approach, reconvening and discussing in meetings as needed. A brief chat later, we began walking down the hall past a set of four more statisticians in a spacious, glass-walled room with bookshelves at the perimeter. We stepped into a conference space across the hall and started to chat about McConnell’s work.
Even though he may not say it, John McConnell makes it very clear that he’s human. If you were to take a poll on what someone with a PhD in Management Science might be like, you’d probably end up with the stereotypical professorial statue. McConnell doesn’t fit this description in any sense – his tall, thin figure, contoured features, and deep brown eyes are often paired with a confident, affable grin. His deferential, relaxed style sharply juxtaposed the well-pressed suit and tie he wore that day. Despite his official designation, McConnell works and communicates at the same level as his team members, attending the same meetings, discussing the same problems, and providing the same leadership opportunities to all of his team. During one CHSE project, McConnell and Renfro were discussing how to analyze racial disparities in care. Renfro mentioned a technique called an f-test, or Analysis of Variances, which uses the degree of spread of different sets of data to determine how distinct they are (“Analysis of Variances between Groups”). McConnell’s immediate response was “that wouldn’t work,” but Renfro continued explaining. Soon, McConnell changed his mind. “She described it more, and I realized she was right,” McConnell said. It’s just this kind of self-assertive environment, created by McConnell and his team, that makes it fourteen people – not one – responsible for obtaining success in everything from changing lines of code to changing quality of treatment.
* * *
McConnell believes that OHSU, his home as a research professor since 2002, is truly exceptional. Unlike most medical schools, OHSU is founded on a functioning health system. As such, students can get a general education, then continue on to cutting-edge research in a specialized area that appeals to them. Most of this research is either lab research – “think labs and test tubes,” McConnell says – or clinical research, where treatments are tested on actual patients. But if research were a spectrum with lab research on the left and clinical studies in the middle, health systems research (HSR) would be far to the right. According to McConnell, the admirable goal of HSR is to “set up organizations and finance… to try to give the best healthcare to as many people as possible.” Before CHSE, many of the specialties at OHSU would have one or two HSR researchers, but their efforts were often disconnected from their peers and lacked the mathematical rigor a statistician or economist could provide. “The missing link was around quantitative analyses,” McConnell states. In 2011, he was asked by OHSU to create “cohesion” around HSR, so he and a couple of others, including Benjamin Chan, started plans for the group that would become CHSE.
* * *
Quite appropriately, McConnell’s foray into economics was an experimental trial. Choosing to pursue his degree at Stanford “was a bit of a random choice for me,” he recalls. His father happened to work in economics, and McConnell knew that he needed a practical degree. Nonetheless, as he studied more and more, he found an elegance in the applicability of economics. “Even general economics training helps you understand some of the things happening in the world,” whether it be Social Security or the Eurozone. Beyond that, however, is a mathematical rigor that is “constantly evolving” in his eyes. New statistical ideas are constantly generated by students reading work formalized centuries ago. Even though economics is typically seen as a business degree – a means to an end – McConnell sees the beauty in the way technical details can answer the big questions of our day. Statistics, in McConnell’s eyes, is the art of balancing practicality and purity. One fantastic example of this mathematical beauty is demonstrated by McConnell’s favorite mathematical tool: the Gaussian Mixture Model.
* * *
The Gaussian Mixture Model, or GMM, is a mouthful of a technique that changes how statisticians see their data. Traditionally, statisticians like to understand data by fitting special curves, known as distributions, to them. Just as a lockpick might listen to the tumblers on a vault, so a statistician uses individual data points to guess what kind of system – or distribution – is generating that data. One enigma of nature is that many random variables, be it average height in a class, fluctuations in economies, or error in a Mars rover sensor, tend to come from bell-shaped curves where average values – the top of the bell – show up more frequently than values on either side. Named after the famous mathematician Carl Friedrich Gauss, these bell curves are endearingly known as Gaussians. By observing a series of data points – let’s say, various incomes in a community – one can use mathematical methods to fit a Gaussian to the data. From there, one instantly knows the average income, how spread out incomes are, and even how likely certain ranges are to occur.
If there’s one thing that a mathematician has to learn, however, it’s that not all data is that straightforward. Let’s say that our income data came from a source with high economic diversity, and we suspect that there are not one but two groups of incomes. Here, the GMM solves the problem by adding together two Gaussians, or bell-shaped curves – one for each income group. Now, the distribution of incomes looks less like a bell and more like the back of a two-humped camel. We likely find that the GMM fits the data far better than a simple Gaussian (after all, there are two groups, not one) and we can now focus on the two groups in more depth (McConnell). “It sort of opens up new windows to thinking about data,” McConnell states. He draws the parallel to prenatal care, where most mothers keep good health and don’t need treatment, but for smaller, less healthy populations, a little treatment can make a big difference. Fitting a GMM to both groups allows one to know exactly how much of a difference one could make. In McConnell’s work, be it advising legislators on local spending practice or surveying quality of hundreds of hospitals, tools like the GMM can be the key to understanding the circumstances behind data in medical situations.
* * *
Although he feels strongly devoted to his home state, neither McConnell’s higher education nor his first job were in Oregon. Per his CV, after he earned his bachelor’s in Quantitative Economics at Stanford, McConnell journeyed up the coast to the University of Washington, where he earned his master’s in the same field. He then returned to his alma mater to earn his Masters of Science in “Economic Systems and Operations Research,” and finally earned his doctorate in Management Science at Stanford. After excelling through the collegiate ranks, McConnell kept local for two years at the Institute for Health Policy Studies at UCSF, foreshadowing his future career at CHSE. He would ascend the coast once more in 2002 to become a Research Assistant Professor at OHSU in both the Department of Emergency Medicine and the Department of Public Health and Preventative Medicine. He would soon advance to the position of Research Associate Professor, after which he became the director of the Center for Health Systems Effectiveness in 2011. Over this period, McConnell has contributed to over 50 peer-reviewed publications, and now actively serves on seven distinct medical committees. Certainly, McConnell has led an academically remarkable career.
Despite these accomplishments, McConnell maintains a down-to-earth, approachable modesty. When offered a list of possible interview questions to answer, McConnell replied, “let’s skip my background a little bit.” Humility appears to be an epidemic in the CHSE office. Indeed, having completed two massive research projects since 2011 – one evaluating the quality of care of over 600 cardiac care facilities, and the other extensively predicting the number of care providers Oregon would need until 2020 – along with publishing 27 peer reviewed articles to various journals, CHSE as a whole already has a very clear place in advising Oregon’s health systems and setting the standard for rigorous health systems studies. Yet everyone still seems to share McConnell’s infectious and unpretentious attitude. With far more talent than they portray, the fourteen health system scientists are devoting themselves to nine independent and eight collaborative projects, one of which has received considerable attention for its financial and social impact on the state of Oregon (“Current Research”).
According to McConnell, the biggest question that CHSE wants to answer is “did the CCO work?” CCOs, or Coordinated Care Organizations, are regional clusters of healthcare providers that share patients and resources to deliver care. By doing so, the hope is that overhead costs will decrease while quality of service will improve or remain the same. CHSE’s “Current Research” page states Oregon has recently begun a transition to this form of medical treatment, and in response, the Centers of Medicare and Medicaid Services, or CMS, has promised $1.9 billion to help. A series of goals were independently agreed upon to guide Oregon to a 2% cut in spending while keeping care up to standard. If these goals go unmet, however, Oregon could lose hundreds of millions of dollars. Quality of care, CMS financial assistance, and the future of Oregon’s hospitals lie in the balance. What’s left is nothing less than the ultimate story problem for CHSE.
In 2012, the CCO Evaluation Team, a subdivision of CHSE led by McConnell, published a paper that would forever change Oregon legislation regarding mental health (“CCO Evaluation Team”). McConnell explains that early on in the CCO system, commercial health plans would cover any physical injury until recovery. On the other hand, mental health patients were only covered for ten visits. This disparity caused much public upset – in response, a law was passed called the Mental Health Parity and Addiction Equity Act which, among other actions, removed the ten visit limit. It’s clear that this law would allow for much more treatment to a population that could potentially need it; at the same time, the question arose whether such a move was fiscally responsible.
Cue McConnell et al. Their study found that after the law was passed, mental health costs increased a small amount, but only as much as other CCO costs. On top of that, mental health copayments decreased, and it was found that a small yet significant group of individuals needed more than the ten days. “I think it was part of a larger conversation around [the] law,” McConnell states. His advice, based on the statistics from his study, helped shape further legislation regarding the mental health parity. “It feels like I’m one of many people pushing a giant boulder forward… it’s not like I publish something and it goes out and changes the world, but I think the CCO work is important,” McConnell says.
* * *
Later that week, on another drizzly morning, I would return to CHSE to go running. Chan, a witty, tall, sharp-featured statistician with an aura of unimposing yet infectious confidence, happens to be an ultramarathoner. Not too long ago, Chan managed to convince a few of his officemates to go on three or four mile runs, eventually creating a Friday mid-morning tradition. As I waited again in the familiar brick lobby, the chilly air blew through the door as men and women with heavy rain jackets and umbrellas strode inside. From the top of the stairs, I heard someone utter an incredulous “is it raining?” followed by a comic groan. While waiting for everyone else, Renfro, a bright-eyed, relaxed, yet driven statistician with an early passion for mathematics, discussed with me an amusing memory of her running career. After a good laugh, Chan and Renfro, along with Melinda Davis, PhD, CCRP and Beth Sommers, MPH, CPHQ from the associated Oregon Rural Practice-based Research Network, all laced up and headed for the waterfront.
“It’s a different kind of runner’s high,” Chan said, referring to his 50 mile runs. The kind you might feel three days later, when you look back and wonder how you did what you did. He understands his work – and science in general – similarly: you can spend 90% of your time setting up an experiment, and in the end, it can all fail. But the joy is in the success. No longer do 5K runs bring Chan joy, because he knows that even in the worst of condition, with the poorest of times, he could still somehow finish. According to him, in a 50 mile race, there are moments where you truly want nothing more than to give up – and you honestly don’t think you’ll make it. Those trying moments are, ironically, what make ultramarathoning enjoyable for Chan.
In an analogy only a data scientist could conjure, Chan relates his philosophy on delayed gratification to two statistical algorithms, one known as least-squares regression (LSR), which yields results almost immediately, and a more complex alternative called Bayesian inference which can take hours on end to compute. Comparing the two algorithms, Chan addressed the the short runtime of LSR with a straightforward “where’s the fun in that?” His work often requires him to run multi-hour marathons of computations using algorithms – such as Bayesian inference – which churn through thousands of rows of data, extracting only the most pertinent information. At one point, Chan needed to implement a complex statistical algorithm for a CHSE project and ended up with a program runtime of around four hours. McConnell suggested running the program over the weekend, but instead, Chan dug through the system and found that he wasn’t using all of the memory he could. Rewriting the program to run in parallel over multiple processors, he cut the runtime down to an unprecedented thirty minutes. McConnell then asked Chan to tell the CHSE group in a meeting what he’d done. “Well, I wrote a loop…” Chan said, jokingly downplaying hours of work. McConnell grinned recalling the moment – clearly, no one knows how to brag in CHSE.
It’s clear that Chan may love a good challenge, but he isn’t the only one. After McConnell and I finished talking in the conference room, McConnell invited Renfro in to discuss her background. “I loved math all through school, from a young age,” Renfro opened. It simply made sense to her – but Renfro certainly doesn’t restrict herself to one field. She refers to her graduate degree in Spanish as “just her fun thing,” and along with formal training frequently finds herself self-teaching new topics. “Most projects require a little bit of self-education,” she said, “but it builds.” And indeed it has – often, Renfro finds herself referring to work from previous projects to assist in current ones. When asked whether she enjoyed her work, Renfro proceeded by bisecting the set of problems she typically solves. Often, problems that appear mundane are, if not fun, important and result in valuable information. But “sometimes there are just fun exercises, just like good mental challenges, and I’ll go home pumped up about solving something that was just hard,” she states. When presented with a difficult challenge, Renfro not only solves it but thrives on it. “It’s nice when each project requires you to do something else a little bit trickier,” she says. “You don’t want to be doing the same thing over and over.”
Despite the group’s brilliant and hardworking nature, there is a surprising amount of fun that happens. “We don’t take ourselves too seriously,” McConnell states. When asked if the team ever did activities together, he responded, “this was my idea, and they tolerated it… we did one of those sixty-minutes-to-escape things.” Apparently, McConnell had led his team over into southeast Portland to an escape room without giving any hint of where they were going. The group split into two teams, each trying to escape first. Taking part in the seven on seven match, McConnell framed the activity as a good team-building and problem solving activity. Not to mention a lot of fun.
Then McConnell brought up a point which epitomized the spontaneity of the team. “There was an insider joke around llamas,” McConnell continued, “so we had someone bring in a llama for someone’s birthday.” Now CHSE gifts are always llama themed, and it still somehow gets a giggle out of the team. If the reader has any notion of pretension or pomposity about CHSE, it should have left a sentence ago. Appropriately, McConnell proceeded to describe what he looked for in an ideal teammate. In his words, they should pursue excellence – “do what excites them,” he adds – have little to no ego, and be curious. “Getting out of their way and letting them do work is the best part,” he says.
* * *
Regarding the future of CHSE, McConnell looks forward with a hopeful eye. He is hesitant to add more team members, though, as he always fears that the team might become too big and possibly “impersonal.” Looking back, he says that “when we started, it was… broad, like ‘we’re going to do something with health services.’” Soon, McConnell and his team focused on the Medicaid CCO change, devoting a large section of their work to it. But McConnell has larger goals: “I’d like us… to go beyond Oregon [and still] focus on Medicaid and payment reform – there’s a lot of issues around these Medicaid programs, [like] what’s the right way to set up payment.” McConnell recognizes this as an ambitious goal – but despite all the humility in CHSE combined, he knows he has the right team.
As I finished talking with McConnell and Renfro, I proceeded back down the switchback staircase, across the brick hall that so strongly welcomed me earlier. When the wash of excitement wore off, and the transcribing of interviews began, the humanity of CHSE really sank in. Fourteen men and women have, day after day, entered an office, done what they love, and pushed to improve all kinds of medical situations using the greatest, most elegant tools known to humankind – statistics, economics, mathematics, and written communication. As a child, I often thought of mathematical and scientific careers as fascinating but selfish – CHSE stands as the coup de grace to that worldview, proving that exceptional good can come from even the most human of us when we put our passions into our work.
“Analysis of Variance between Groups.” College of Saint Benedict and Saint John’s University. College of Saint Benedict and Saint John’s University, n.d. Web. 17 Nov. 2015. <http://www.physics.csbsju.edu/stats/anova.html>.
“Center for Health Systems Effectiveness.” Oregon Health & Science University. Oregon Health & Science University, n.d. Web. 22 Oct. 2015. <http://www.ohsu.edu/xd/research/centers-institutes/center-for-health-systems-effectiveness/>.
Chan, Benjamin. Personal interview. 30 Oct. 2015.
“Current Research | Center for Health Systems Effectiveness.” Oregon Health & Science University. Oregon Health & Science University, n.d. Web. 22 Oct. 2015. <http://www.ohsu.edu/xd/research/centers-institutes/center-for-health-systems-effectiveness/research/index.cfm>.
“Faculty and Staff | Center for Health Systems Effectiveness.” Oregon Health & Science University. Oregon Health & Science University, n.d. Web. 22 Oct. 2015. <http://www.ohsu.edu/xd/research/centers-institutes/center-for-health-systems-effectiveness/chse-faculty-staff.cfm>.
“Focus on South Dakota: A Picture of Health.” Helmsley Charitable Trust. Helmsley Charitable Trust, n.d. Web. 5 Dec. 2015. <http://helmsleytrust.org/publication/focus-south-dakota-picture-health>.
“Fermat and Pascal on Probability.” University of York. U of York, n.d. Web. 17 Nov. 2015. <http://www.york.ac.uk/depts/maths/histstat/pascal.pdf>.
McConnell, John. Personal interview. 26 Oct. 2015.
McConnell, Kenneth John. “Curriculum Vitae.” Oregon Health Research and Evaluation Collaborative. Oregon Health Research and Evaluation Collaborative, 22 Jan. 2013. Web. 4 Nov. 2015. <http://www.ohrec.org/sites/default/files/profile-cv/McConnell%20CV_0.pdf>.
Renfro, Stephanie. Personal interview. 26 Oct. 2015.
Many thanks to Neil Duffie, Dr. Wendy Watson, and Dr. Randy Watson for reviewing this paper, to Dr. K. John McConnell for being such a wonderful friend through this process, to Benjamin Chan and Stephanie Renfro for providing their fascinating views on CHSE, and to my precious little dog Elle, who provided emotional support throughout the work.