Speak data, p.8
Speak Data, page 8
It’s an important question. A large portion of the data we collect is objective biological data. But another large portion comes from formally designed patient-reported outcomes. And we definitely need both. But right now in our clinic, we often rely more on the patient-reported outcomes than we do on biological data because many of the complex conditions we’re seeing are poorly characterized. Meaning, we don’t know what the “right” biology should look like.
For instance, look at human performance in athletes. When I’m working with an elite female athlete and I look at the existing literature on how to optimize female athlete performance, the textbooks look like they were written in the Dark Ages. They literally say things like, “When a woman is on her period, she shouldn’t run because she’s going to be irrational and her performance will dip.” And it’s just like—oh my god! So some of the work that we’re doing is daily hormone monitoring to understand how hormones of menstruation and ovulation affect performance output or muscle strength, because work of this nature is rare, and we have few guidelines. In the short term, if you’re a responsive performance professional, you listen to what your athlete is saying. And if an athlete says, “I’m not lifting my best today,” we can say, “Yes, that’s because your progesterone is peaking and you’re going to have issues with resistance training.” But right now, we can only really listen to what they’re telling us and guide appropriately. If you don’t listen, you harm. Listening to our patients and deriving as much structured qualitative data as we can is super important.
What is your definition of data, if you have one? Or how would you explain what data is?
I would explain data as any piece of information—whether it comes from personal communications, questionnaires, or biological testing—that you can use to form or reinforce an opinion or guide an act.
It makes me think: We had a conversation just the other day with the US Food and Drug Administration for our brain-computer interface trial. We’re studying whether people with severe paralysis can receive a minimally invasive implant to allow them to communicate and perform computer-based tasks at home. The majority of the patients who are enrolled in the trial have end-stage ALS. They’re completely locked in with their illness. They’re ventilated. They have minimal eye movement. And yet the majority of them actually have pretty good quality of life on the quality-of-life scale because they’ve got their family around them. They’ve got their needs met. They’ve cognitively and psychologically adjusted to their illness. And so they’re kind of just living their lives and they still have joy. One of the guys loves tasting scotch. They put a little bit of scotch on a sponge and sponge it around his mouth and he loves that. His quality of life is sky-high.
The main questionnaire that the FDA was interested in to show patient benefit was a quality-of-life scale. We advised the FDA that we felt it would be an ableist approach to just assume that these folks currently have terrible quality of life. If you implant them, yes, their quality of life is going to soar—but that doesn’t mean they don’t have a good quality of life now. So the FDA listened, and is looking to develop a new questionnaire that will evaluate these things more holistically. We always need to be recalibrating the ways in which we collect data from patients as we learn about how we have marginalized and biased ourselves against the patient population. We need to update the questions we’re asking and our understanding of what’s important.
Is it a catch-22, though? You keep collecting data to be able to then collect better data, but then those new data are not always comparable to the data you collected before. So you don’t have long-term, solid, comparable data. But maybe that’s just part of the process of innovating? We’d never do anything disruptive if we only did incremental little things. This applies for all types of data collection, right?
Absolutely. I’m not against incremental discovery or incremental innovation. I just don’t think it’s the only thing we should be funding, because while we’re waiting on that incremental discovery, we’re missing disruptive discoveries that could really improve quality of life for patients.
In my lab, you need to be comfortable with discomfort, because we’re never doing the same thing for too long. We’re always changing. It maybe sounds silly, but it’s something we need to tell people in the academic and clinical worlds, because most academics and most clinicians want the algorithm, the protocol, the step-by-step. We’re always churning over, working out what works and what doesn’t, and collecting new data.
Bon Ku
Bon Ku is a physician, public health researcher, and design strategist working to reimagine what health and health care can be. The coauthor, with Ellen Lupton, of Health Design Thinking: Creating Products and Services for Better Health, Bon is a leading proponent of a more humane, patient-centered health care system. Today a program manager for Resilient Systems at the Advanced Research Projects Agency for Health, he previously led Thomas Jefferson University’s Health Design Lab, a uniquely interdisciplinary think tank exploring newly inclusive paradigms for the patient experience. In this conversation, Bon discusses the role of empathy in medicine; what doctors can learn from architects; and whether data is truly destiny.
You recently made a career change. How’s it going?
Yes, I left a career in academia. I previously ran something called the Health Design Lab at Thomas Jefferson University, but I’ve just joined a new government agency called the Advanced Research Projects Agency for Health, or ARPA-H. It’s modeled after DARPA, the Defense Advanced Research Projects Agency, which developed everything from stealth technology to Siri on your iPhone. DARPA makes investments in new technologies, and because they’re willing to fail on some of them, it’s an effective model to explore innovation. We at ARPA-H are applying that same methodology to health care. We’re guided by the principle of prototyping quickly, and thinking of future solutions that don’t currently exist but could really expand our imaginations. I think my background in design suits me well for the work we’re doing.
It sounds very, very exciting. It’s hard for people to think about health and design together, but you’ve used this as a prototype for how hospitals can work better, for instance.
I’m interested in the intersection of design and health care. As a physician, I’m trained in the scientific method. But humans are messy, and sometimes the scientific method alone cannot provide the best solutions. I was attracted to design for its ability to apply a creative mindset, and to couple that with a scientific method when thinking about human health. I argue to physicians and other clinicians that the skill set of creativity is fundamentally important to healing people. But many physicians don’t think that way. We think of those with a “creative mindset” as designers or maybe architects or musicians. But applying design principles can open up our imaginations to explore future possibilities that otherwise would not have been available to us.
What would be an example?
One of my favorite research collaborations was with an architecture firm called KieranTimberlake. They’re based in Philadelphia, and they have a research division. I was interested in them because we have this problem with overcrowding in emergency departments. As you can imagine, emergency rooms are chaotic environments. They’re open 24-7. Overcrowding is rampant. Patients might wait three, six, nine, twelve hours to see a doctor. There are constant interruptions. Even violence can occur.
I wondered what we could do to design a better emergency department. So I reached out to KieranTimberlake to see if we could apply their mapping methods to a hospital setting to understand how patients, clinicians, and nurses interact in time and space. I have no training in this at all! But they do.
We started off with a research prototype. We had research assistants use pen and paper to track a physician’s movements in time and space over an eight-hour shift throughout the emergency department floor plan. Eventually that led us to publish research exploring questions like: Does the specific layout of the emergency department impact the volume of interactions among staff? We would never have gotten there if it weren’t for this new type of research and collaboration that is normally not done in health care. It was a fascinating exploration into a whole new research area, and it happened because of our willingness to go beyond our domain expertise.
You also have a major focus on the patient experience.
What I appreciate about seeing patients in the emergency department and taking care of them is that they’re like health data personified in a human. When I look at public health studies and especially mortality statistics, it’s abstracted. You know, just numbers and tables. But when I practice in the emergency department and see a Philadelphian who was shot and have to take care of them, that data becomes real. It’s manifested in humans. So even though I’m dealing with a sample size of one, my interaction with a human at bedside gives me a lot more empathy than when I look at datasets.
I think if I were just a data scientist without interacting with humans, I would not have as much passion to scale interventions to help underserved populations in our country. A lot of the work I have done over my career is with underserved populations because I see them every time I work a clinical shift. When I treat a person with unstable housing, someone with substance use disorder, someone who is a victim of gun violence—if I were just a researcher working with those datasets, I don’t think I would be as motivated to spend a lifetime working with these populations and dedicate my career to them. Some people are a lot more empathetic. But I’m not that empathetic of a guy. But if I see the human who is represented in that data-set, that to me is motivating.
Is data destiny? Or, said another way: How much can you predict about a patient by their data—their biometrics, their presenting symptoms, even the zip code they live in?
It depends. For instance, one thing I’m working on now is improving access to high-quality care for rural Americans. There are about sixty million people living in rural areas, and in almost every disease category, they have worse health outcomes. They die at an earlier age. They have to travel long distances to get care. There’s many disparities there. My work as a clinician and, you know, seeing inequities show up in the emergency department has shown me that certain populations have worse outcomes just because of their zip code. Generally, if you live in a poor zip code, you’re going to have worse health outcomes.
That has inspired me to do stuff at scale through ARPA-H because we have the ability to impact millions of lives. Again, I don’t geek out about datasets. When you just look at numbers and read about statistics, it’s easy to be very detached from them, right? But because I see the end results of inequities in our system manifested in patients showing up in the emergency room, that motivates me to design interventions at scale.
As you said before, humans are messy and nuanced. Where have you seen “traditional” medicine overlook something important because of a purely quantitative, purely analytical, approach to data?
I’ve done some research on super-utilizers, which are patients who use hospitals excessively, like more than ten ER visits a year. In particular, there was one patient who had made almost a hundred ER visits in a year. This patient was an outlier. He had end-stage renal disease. If you just look at his emergency department utilization stats, you would label him as an abuser of the system. Based on the data, it’s easy to make that conclusion because no one should be going to the emergency department that much.
Well, I got to know this patient. I actually made a home visit, which is a little bit weird to do as an emergency room doctor. But it allowed me to understand his environment and dig into the reasons why he showed up in the emergency department so much. He had an unstable housing situation. He had children, which was surprising to me. I’d assumed that he wasn’t a father. But he actually had people who cared for him. He lived with his aunt at that time. And he had a lot of fear and anxiety. I think that’s what drove him to go to the hospital—he felt that the hospital was the safest place for him. That helped remove some of my bias against this patient. It helped to humanize him.
I try to do that with every patient encounter because it’s so easy, working in a stressful environment, to dehumanize patients. Being able to understand who he was as a human, to understand his fears and anxieties, gave me more empathy for him.
It doesn’t help that the field of data visualization, in our opinion, is still very focused on communicating quantitative information. What you’re talking about, the stuff that might be better at building empathy, is often left out. This is particularly true in the medical field.
In terms of data visualization, I think we’re at version 1.0, and we’ve been stuck in a very outdated form. The data visualizations that I have seen in medical journals have not changed in over a hundred years. It’s tables, bar charts, pie charts. That’s why I am fascinated by ways that we might get to a version 2.0 of this and start to humanize the data.
I think one powerful example was when The New York Times published data visualizations of numbers of deaths during the pandemic. In fact, I have on my wall right here a clipping of one of those front-page articles as a reminder of the deaths caused by COVID-19. Those representations of the data triggered emotions for me in a way that what I read in a medical journal cannot. So I think there are ways. I think the work that you do is a powerful example of that—meaning, how we can represent data visually to trigger an emotion. Often, data is very scientific, and it’s easy to divorce yourself from these sometimes terrible statistics we see.
COVID-19 changed our understanding of data. How did it affect your thinking?
I have a lot of thoughts on this. One is that we saw the rampant misrepresentation and misinterpretation of data. Yes, it was great that the general public got familiar with public health data, but most of the general public has not taken a class in epidemiology. I appreciated that it’s very easy to misrepresent data. But when we think about new ways to represent data, we also need to think about how important it is to do so in a rigorous, scientific way and to educate the public about it. It’s so easy to put wrong information out there on the internet and people think it’s true. If we’re putting more data out there, how do we also give people the tools to validate that data? We also need to think about how we collect the data, which can have its own bias. Which data do we collect? Many people assume that data is neutral, but it’s not. We all bring our biases into which data we collect, and especially thinking about machine learning and AI algorithms and generative AI, I think this is going to be the greatest challenge of our time. How can we design an equitable and fair algorithm? That’s going to be determined by the ethics of how we collect data.
Moving forward, how should we define data?
Oh my gosh, this is a hard question. I don’t know! I’m going to riff here as a physician and a researcher. Data are a representation of patients—but not a complete representation. It allows us to quickly make generalizations about larger populations. I guess that’s one of the benefits of being able to abstract information from patients in an incomplete way. There are benefits to that, to generate therapies and treat patients at a population level.
When a patient receives their lab test results and looks at those numbers, something strange can happen. For example, maybe a patient has always been on the very low end of a certain range of white blood cell count. They take a new test, and the result says their cell count is elevated, but still within the acceptable range. And the doctor says, “Well, you’re still within the range. There’s nothing to worry about.” But the patient knows that this is an outlier for them! They are a unique individual with a journey and a history that is not being taken into account. It can be a very stiff and rigid way of measuring.
One hundred percent. In medicine, we label data as normal or abnormal. It’s very binary. But as you know, data is continuous. Also, in medicine, we often only get spot checks of data. So instead of measuring your white blood cell count once, what if we were to measure it over a week, over a month? That would give a far more accurate representation than a spot check. Blood pressure is the same way. Patients will take their blood pressure at home and say, “Oh, it’s so high. It was like 160.” Then they show up in the ER, and it’s 120. That’s why you shouldn’t base anything on “normal” or “abnormal,” or on one measurement. Checking blood pressure every day for a month, recording the data, will lead to a better assessment.
It happens every day. I’m not blaming the patient. I’m blaming American medicine because that’s how we have operated in the past. The tools we have for collecting data are very rudimentary, right? We have a blood pressure cuff. We check it. Thankfully, as technology advances, we’ll be able to take continuous measurements of vital signs, and continuous measurements about other diseases.
But to play devil’s advocate: Is there such a thing as too much data?
This is also something I think about all the time. I don’t think we could label too much data as bad or good. That’s too simplistic. Before we start collecting data, we have to ask ourselves, what are the repercussions? And is that data going to be actionable for that patient? We see over diagnosis and overtreatment all the time. Almost every patient with a chronic disease has been through something like that. We’re going to be seeing a lot more of that as we are able to extract ever more data from our human bodies through advanced technologies.
That’s a whole other conversation.
Yes, but I think one of the most important questions we have to address now is: Who owns the data? Does a patient own their health data? Because right now, data are not owned by patients. Data are not easily portable. Data are owned by hospitals, by insurers, by pharmaceutical and device companies. That data is powerful. How can we enable patients to have more control of their data—decide where it goes, and who takes a look at it?
