Nuance. Health. Digital. Language. Innovation. Humanity. Data.
Below is the transcript of the complete Interview I did with the Health Data Consortium out of Washington DC. Their tagline is unleashing the power of open health data. The title of their post: The Nuances of Language, Data, and Innovation. I have taken the liberty of intercalating some other ideas/posts within the interview.
We are pleased to share this recent interview with Dr. Shlain. In this first part, Dr. Shlain discusses the nuances of language and how they shape the way we think about health care.
On your Twitter account, your tagline is “Purveyor of subtleties in the science of medicine. Internist. Questions the obvious and looks for patterns and common sense. Always trying to simplify.”
Could you talk about your quest for subtleties and perhaps suggest a few that have emerged in your work that readers might find surprising?
I see patients every day and given the tectonic changes in medicine, I also co-exist in the world of data and innovation – and what I care about most is the impact on both doctors and patients. Ultimately, the only thing that matters is the health and well-being of a person whose health is in jeopardy. We are all fortunate to be at the intersection of the spectacular changes occurring in medicine. I have the ability to rapid a/b test different products and processes to see if and how they work; and I’m not afraid to fail or be wrong; these are merely glimpses into your deficits. But when you do something over and over again and run into the same inefficiencies and likelihood for errors each time, you start to see these patterns and realize many aspects of the practice of medicine are steeped in nuance. Unless you really understand these contours, you’re coming up with a shortcut or heuristic, which often leads to a less optimal outcome. Some of the subtleties that I have come to appreciate, many of them live in the world of language. Medicine is both an art and a science. Sadly, the world of technology focuses on the science, often relegating the ‘art’ to disdain. The doctor patient relationship is the atomic unit in medicine. The output from this doctor-patient interaction leads to a chain reaction of money, risk, data, outcomes and if we are not communicating properly, we have a multitude of ‘butterfly effects’ that can occur.
The language that we use often dishonors the reality of what healthcare is. Specifically, one example is the word “readmission.” What is a readmission? It’s a number a hospital uses to define the number of people who return to the hospital a certain number of days after a procedure – so it’s a linguistic concept made by the hospital. The truth is, a readmission is really a person who is ill that had to go back to the hospital after they just left, even though they didn’t want to go back. Really what it represents is a treatment failure. Either they were pushed out of the hospital too soon, there was poor follow-up or insufficient follow through. By using the term “readmission,” we dehumanize it into a mere statistic, a word in a policy statement. But if you called a readmission a “treatment failure,” it would change the way people and our policymakers think about what a readmission really is: unnecessary suffering. We often use a form of dumbed-down language, enabling a tacit dishonesty with ourselves about what certain words mean – in this case, a suffering person. Hospitals, policy makers and doctors would take it far more seriously if they had to talk about treatment failures instead of readmissions.
The more lazy and general we become, the more chaos enters into the world of policy. Language is important, and these are some of the nuances that I’m constantly fascinated with. Hammurabi would be disappointed by the dilution of his invention.
A great book that I’ve recently read, is The Secret Life of Pronouns by Dr. James Pennebaker. A great read.
What does this say about the need for better definitions and more precision in our use of language? How important is it that we bring real clarity to these definitions?
Specifically, one of the issues is the words we use. Words are the building blocks of our understanding of the world. When two people are communicating with each other – the words I use to describe something might be interpreted by someone else another way, and they may not be what I intended.
For example, if I ask you what the word “provider” means, your answer will be different than every other person I ask the same question. So if everyone has a different answer, maybe we shouldn’t use that word. I have written many pithy blog posts on LinkedIn trying to expose these flaws in our lexicon.
See Cracking Open Medical Jargon
Case #1 Patient Engagement
Case #2 Patient Portal
Case #3 Reimbursement
Case #4 Provider
Case #5 EOB (Explanation of Benefits)
If you talk about a provider as a blanket term, a provider could be a hospital system that makes billions of dollars, or a provider could be a single doctor in Plano, TX with one nurse. When you talk about providers broadly, how can you put those two – a large hospital system that’s in 8 states and a single doctor into the same box? I break up the world of provider into human providers and non-human providers. The non-human providers are generally at-scale: hospitals, radiology facilities, LabCorp – big businesses. Then there’s human providers. My preference is that we call them clinicians, call them doctors, nurses, and healthcare professionals. We have to respect the fact that doctors didn’t spend 10 years of their lives in school to be called a provider. I went to medical school at Georgetown, and I didn’t get a medical “provider” degree – I got a medical doctor degree. Why do we not honor that? We’d rather say “provider” than be specific. We’re lazy, and laziness leads to bad language and bad language leads to bad policy and bad policy hurts people.
Another is the word “quality.” Everyone always talks about quality of cost and outcomes. These are another set of words that mean different things to every person you ask. Simplicity matters. It’s hard to be simple – it’s hard to make things that are complex simple to understand. I feel that my role in the healthcare discussion is as a translator.
So many people at the 30,000 foot view have no idea what goes on at the ground level and conversely, many of the people at the ‘problem surface’ of medicine, those of us in the trenches, have no connectivity or understanding of what the those people in the clouds are thinking. There is often a major disconnect in the policy to bedside translation. The policy is always focused on money, outcomes and special interests for entire populations; while a doctor can really only see one patient at a time – attempting to, in the spirit of their healing art, ameliorate suffering.
The atomic unit in healthcare is the doctor-patient relationship. Everything that emanates from decisions made in these interactions kick off lots of data and generates trillions of dollars in transactions. The sad bit is that everyone thinks they can dis-intermediate or ‘disrupt’ the doctor patient relationship into some form of digital transaction. While there is significant room for communication efficiencies, if we lose sight of what it means to be sick, afraid, anxious and human, we lose. Words are often critical glue that binds humans together in times of need; empathy is not digital – at least not yet.In our use of language, if we’re not being specific and precise, we create confusion and more problems. Sadly, I believe there are a lot of actors who prefer and proffer confusion in the healthcare arena to propagate their agendas using linguistic opacity. After all, which players in the system want the status quo gravy train of billions of dollars to stop?
In The Health Care Blog, you wrote about something you call the Data Response Curve.
Could you elaborate?
One of the central tenets we learn in pharmacology during medical school is the dose-response curve. If I give a patient ibuprofen at a very low dose, it won’t have any effect. If I give it at a higher dose, we get some pain relieving effects, and if I give too much, there’s toxicity. When I give a patient a dose, I need to evaluate if and when the next dose should be given and what the dose is for their age, weight and other comorbidities.
Like medications, with data, too little may not be enough to make a good decision and too much can be toxic and lead to analysis paralysis. For example, my father was a surgeon and had to make decisions on trauma patients sometimes with a minimum amount of data to make the correct decisions. Giving a physician like him too much data of dubious quality may hinder his ability. There are many studies regarding how data does not understand the concept of ‘benefit of the doubt’ and if we rely too much on data as the panacea, we are at risk of losing touch with the very nature of our own cognitive abilities. We need to find the balance between human and machine cognition.
The Data Response Curve is about asking questions like “What is the right amount of data to make the best clinical decisions?” and “What is the littlest amount of data you need to make correct decisions?” to figure out how to most effectively utilize data as an important tool in a clinician’s toolbox.
See Data Response Curve post.
A theme that emerges in your work is that we are too focused on “more data” as opposed to “better data.” Could you talk to the distinction and why we might need to shift our focus? How should we think about “big data” in that context?
The only data that ever matters to an individual is the “little data” – the data about them, their lives and the tangible bits that affect them. Big data should inform the little data and vice-versa. A critical aspect to appreciate data sets is the law of large numbers and Gaussian distributions better known as the bell curve. Generally speaking, we’ll look at the big data and say, as best we can tell, you fit underneath this bell curve, so you should do this because we know it works. And in a world of Obama’s precision medicine efforts, a one-size-fits-all problem solving idea falls short. It really depends on how many standard deviations from normal you’re willing to get comfortable with…and what you’re really getting comfortable with is a 20% of it not working. Once we get into the realm probabilities, it gets very confusing, very fast. Having just read two books, The Rational Animal and the Improbability Principle, it becomes clear very fast that when humans need to make probability decisions based on data, they have a very difficult time comprehending the subtleties of mathematics. Furthermore, what the data does not know is what matters to patients, only what is the matter with them. Often what matters to people is difficult to quantify and represents an incarnation of the existential and the subjective.
See post: Who is your sugar Data
For example, I had a patient who had really bad heart disease for many years, and then he suddenly developed lymphoma. He asked me what the best course of action was. I told him that all the data, big and small, says you have to start chemo right now, and we can extend your life maybe 6 months to a year. He was a very pleasant 70 year old grandfather who said to me, “Dr. Shlain, I have planned to go to Mexico months ago with my family over Christmas. I really want to go to Mexico, and I want to sit on the beach, drink tequila and margaritas, and watch my grandkids play in the ocean.” The data would say: “No, you can’t do that. You have to start chemo right now.” But I said to him, “Why don’t you go to Mexico and we’ll delay chemo for a couple of weeks.” What mattered to him trumped what was the matter with him. He went to Mexico with his family, sent me pictures, and had a great time.
His heart disease was manageable and problematic his entire life yet the cancer was the bigger issue and the proximal threat. He eventually came back from his trip, and a week later he suddenly died of a heart attack – before starting the side effect prone chemotherapy. There’s no data in the world big or small
that could’ve predicted that. He got what he wanted, and it had nothing to do with data. It had to do with what matters, what’s meaningful to him.
Everyone is hyped up and excited about the promise of big data and algorithms as though they are going to magically solve all problems. There is definitely a place for these technologies, and in order to truly be authentic in our desire to take care of people better through our use of data, we have to acknowledge their shortcomings. Big data is a blunt instrument, and doctors need to determine what it means specifically for each patient.
Tell us what drives your intense interest in understanding the role of data in improving health and health care?
Firstly, I think we don’t appreciate that there are more than one type of data. I believe there are at least four different types of data that have distinct properties in their own discrete ways.
- The first type is “fixed” data – such as date of birth or town and hospital you were born in, or dates of immunizations. These pieces of data are never going to change because they’re fixed facts.
- The second type of data is “parameter data” – data that looks like it’s fixed but actually can and does change. For example, hair color, eye color, zip code. This data needs to be looked at differently than fixed data because if we’re basing on decisions on what is historical but might be changing, that can affect the care of patients.
- The third type of data is dynamic data or “Delta Data” which changes all the time – thyroid levels, blood pressure, weight, and temperature. Dynamic data are changing so often that we really need to appreciate them in respect to trends more than the data elements themselves.
- The fourth type of data is “U-data,” and it’s non-quantifiable and subjective. U-data represents the state of ones mood, cognition and mental presence. For example, if a patient comes to my office and is in the middle of a divorce, they are likely distracted or distraught and do not have the capacity to be listening and participating in their care. U-data has no home or any meaningful instantiation in an EHR due to its qualitative nature. You can lead a horse to water; and like patients, they need to be receptive, aware, participatory and trust that decisions before them feel appropriate and oriented towards a good outcome.
Sadly, data that is collected sits in a database. To me, a database is a “wait-a-base” because the data just sits there and waits for some person to find it. It doesn’t have any agency to talk with other data to solve problems.
See Post Medical Data Handoffs and Fumbles for more on Data.
The benefit I have of practicing medicine each day is that I have a birds eye view on how patients interact with the health care system. In fact, it’s not really an ecosystem like other industries, it’s really a combination of a freak-o-system and an egosytem; which makes it hard for innovation to take hold quickly.
Your company HealthLoop was recently funded by VCs to the tune of $10MM.
Tell us your motivation for founding the company and what you hope and investors hope to accomplish?
HealthLoop was born out of innovation by irritation. There was a woman who came to my office with cough, fever, and shortness of breath – she had pneumonia. I sent her home with an antibiotic and my cellphone number, and told her to call me if she didn’t improve in a few days. Seven days later, she was in the emergency room going into respiratory distress and being sent to the ICU and put on a respirator. I thought to myself, “Why didn’t she call me?” After all, I gave her my cellphone and made it as easy as possible for her to contact me. Then – my “Aha!” moment was “shame on me, why didn’t I call her and follow up. After all, she was sick, had pneumonia, and I should have been checking in on her.”
Typically, when doctors send someone out of the office, they assume no news is good news – if you don’t hear from a patient, you assume all is well. But the truth is no news is no news. News is, in fact, data… so can we say no data is good data?
I looked around to figure out a solution that checks in on people after they’ve left the doctor’s office – a product that could proactively and asynchronously check in on patients, which could surface exception-based alerts when a patient was not recovering as expected. I looked at every CRM product, and asked all my techie friends in Silicon Valley. The product didn’t exist… so I endeavored to solve the problem of doctor-patient communication between visits, after visits and before visits. Where was the feedback loop? Where was the news?
So I started this very simple program where patients who came in would get an email checking in with them on their status, and would ask if the patient is the same, better, or worse. A “loop” was round of check-ins with a patient until they were on the right track to recovery. If the patient answered they were “better,” the algorithm yielded a +1; “worse” yielded a -1. But if a patient said “the same,” then it was a -0.5 because, for example if a patient has pneumonia, that’s a negative trend as the patient should be getting better if they’re on antibiotics. This was my insight. As soon as these patients started saying “better”, the algorithm would ask them less frequently, and would stop once the trend was on track and close the loop. For the trends that weren’t on track, then I could reach out to them by telephone. It’s a proactive mechanism to reach out and touch base with patients on any given day.
It turned out, patients loved to be checked in on. Patients zero in on the sense that they feel cared for by their doctor – “Is there a genuine sense that this clinician cares about and for me?”
Somewhat by accident, I created the digital extension of this clinician empathy through a pre-scripted email. They’re not automatic, their automagic. Doctors are trained on how to diagnose and how to treat, but there is sparse literature on how to follow-up with a patient. In fact, the most common follow up for most problems is: come back in a week. With HealthLoop, we can check in with people every single day and provide them with relevant and contextual information about where they are in their progression and we can ask them a host of specific questions, in multiple-choice format about how they are going. Furthermore, we treat engagement like a symptom itself. That is, if a patient isn’t engaging with their medical team, perhaps something is wrong. The amazing thing is that we have over 80% engagement on all check-ins. #patientlove is our Twitter hashtag for all the great things patients are saying.
The good news about HealthLoop is that this is something I started, and we’re about to make some big announcements this year. It’s been very organic growth.
Do you have any advice for aspiring entrepreneurs looking to breakthrough into health care?
With HealthLoop, it was an inside-out solution. I’m a doctor, and I solved a problem for myself and other doctors – on the inside of the health care system. The advice I would give to entrepreneurs is to make sure you talk to clinicians before you build something and that, if they agree it’s a good idea, they’ll also test it for you. Many smart, talented entrepreneurs start with ‘I see a problem, it happened to me (or my mother) and I’m going to fix it.’ The problem here is that they don’t appreciate how complex and intertwined health care is. John Muir once said, “When you tug on a single thing in nature, you realize it’s connected with everything else”. The same is true for health IT. Many of the outsiders trying the outside-in approach run into the issue that, when they try solve one problem in health care, that creates four more.
See The Economist Innovation Forum panel on Health Innovation
Also, with a health care system that is overwhelmed by meaningful use, ACOs, penalties and regulations, doctors don’t want to juggle something new if it’s one more thing to do. It’s not that your product isn’t great, but changing health care is like trying to change a tire on a moving car. Unless if you have a champion internally, it will be hard to get anything going. If you choose to go into an incubator or accelerator program, make sure they have a proven track record of getting products into a clinical practice in some meaningful way. However, once you get in and you break into a health system, your product is automatically legacy, and once you’re legacy, people don’t want to get rid of you because they’ve already made the investment in the switch. Nothing goes viral in health care. Things go super slowly, and you have to have a realistic time horizon. Things don’t go viral in health care like they do in the rest of the internet, they go bacterial. Slow and steady.
In consumer technology when we look at data and tech companies, they say it’s a $2.6 trillion industry and people want a slice of that. Consumer tech deals with fun and gadgets, but in health care, we’re dealing with life and death. Health care is a people business in need of technology, not a technology business in need of people. That simple statement really implies that there’s a lot of people building health tech that they need people to interact with it, rather than seeing where there is a real need and figuring out how can we leverage technology and make people’s lives better. Health care is a “get rich slow” scheme, but hopefully you’re doing it to change the world and having a meaningful impact and not just to make money.