Digital health means AI and machine learning based on platforms that aggregate patient data. In this episode, two leading healthcare innovators share an insider’s perspective.
Digital health means AI and machine learning based on platforms that aggregate patient data. This healthcare transformation raises challenging issues such as ethics in healthcare, transparency in AI algorithms, and equality of access to healthcare. In this episode, two leading healthcare innovators share an insider’s perspective.
John Halamka, M.D., is the president of Mayo Clinic Platform. In his role at Mayo Clinic, Dr. Halamka oversees the future direction of the Mayo Clinic Platform that will help establish Mayo Clinic as a global leader in digital healthcare. Prior to the Mayo Clinic, Dr. Halamka served as the executive director of the Health Technology Exploration Center for Beth Israel Lahey Health in Massachusetts. Previously, he was chief information officer at Beth Israel Deaconess Medical Center for more than 20 years.
Daniel Kraft, M.D. is a Stanford and Harvard trained physician-scientist, inventor, entrepreneur, and innovator and is serving as the Chair of the XPRIZE Pandemic Alliance Task Force. With over 25 years of experience in clinical practice, biomedical research, and healthcare innovation, Kraft has chaired the Medicine for Singularity University since its inception in 2008, and is founder and chair of Exponential Medicine, a program that explores convergent, rapidly developing technologies and their potential in biomedicine and healthcare.
- About Dr. John Halamka and Dr. Daniel Kraft
- Data platforms in healthcare and clinical medicine
- Health data analytics and the Mayo Clinic Platform
- Digital healthcare and clinical workflows
- Digital medicine and the healthcare ecosystem
- Digital transformation in the healthcare industry and equal access to care
- Healthcare platforms and patient experience
- Cybersecurity and healthcare data
- Medical ethics and digital healthcare
- What is the role of clinical patient registries?
- Fighting COVID in 2020
This transcript was lightly edited.
Michael Krigsman: We're speaking with two of the most innovative healthcare leaders in the world. Dr. John Halamka is president of the Mayo Platform. Dr. Daniel Kraft is the founder of the Exponential Medicine Conference and a professor at Singularity University.
Dr. John Halamka: I've been at Mayo Clinic since January of 2020, and I was charged with running platform businesses globally. Of course, what that means is connecting producers and consumers of information using algorithms and mobile technology and creating value, new businesses, and new ways for Mayo to extend its reach worldwide.
Michael Krigsman: Daniel, tell us what you're working on right now.
Dr. Daniel Kraft: The focus this year has been COVID. I've been chairing the XPRIZE Pandemic Alliance Taskforce. The alliance is made up of 100 organizations from NGOs and academics to big startups and small ones, trying to connect the dots and accelerate solutions for this pandemic and to prevent future ones, and integrate some prizes like a rapid COVID testing prize and a new PPE mask challenge, so trying to connect the dots and accelerate things to help things out in the current day as well as catalyze the future of health and medicine.
Michael Krigsman: John, you're head of the platform at Mayo Clinic, and so when we talk about platforms, what do we mean and why is this important? Then, Daniel, please weigh in.
Dr. John Halamka: Let's start with, say, radiation oncology or radiotherapy, what is called auto-contouring. If a cancer patient needs radiation, a linear accelerator needs to be programmed by a physicist and an expert radiation oncologist. It takes 6+ hours of human time to review the films of the tumor and then program the linear accelerator.
What if one developed a cloud-hosted mechanism to ingest images of tumors—say head and neck tumors—AI algorithms that would be able to review those and, in literally near real-time, recommended the safest, lowest dose, most effective mechanism of delivering the radiation therapy to the patient and then auto-programmed a linear accelerator thousands of miles away without a radiation oncologist or a physicist nearby? Well, there's a platform that is connecting incoming data and algorithms, delivering something of value back and, ultimately, improving patient care. That is one example of what Mayo has worked on this year and will have in prototype by the end of 2020.
Broadly, platforms are connecting producers and consumers and building value.
Dr. Daniel Kraft: That sounds fantastic, but how do the radiation oncologists feel about that since often the challenge of platforms are the misaligned incentives because they're changing someone's lunch or moving someone's cheese and that's often harder than the technology?
Dr. John Halamka: What if, given your training in radiation oncology, instead of doing one case a day, you can do ten? What if instead of working in proximity to one town or one region, your scope becomes global?
The way I think about platforms is it's enabling us to deliver a higher quality of specialty services to more people in more geographies, democratizing access to expertise. At the same time, obviously, there is a business in doing that that creates value.
Dr. Daniel Kraft: I love the fact, Mayo and your work has often been global but, like politics, medicine is often very local. We've trained both in Boston. That's the way you do it in MGH versus the Mayo or the Stanford way. Now we can kind of crowdsource learnings, whether it's from China or Europe, into these platforms. Is interoperability still a challenge?
Dr. John Halamka: The answer is it depends. Isn't that a great answer?
DICOM, radiology images, yeah, okay, fine. It's a non-standard standard. Philips, Agfa, Siemens, and GE, they're slightly different in the metadata, but the core images are actually pretty comparable.
There is, actually, for radiation therapy, DICOM-RT. Believe it or not, every LINAC in the world is programmed using the same set of APIs. Hard to imagine.
Let's take something else: EKGs. Have you ever written for—pick your EKG manufacturer—a parser and display tool for EKGs? It's like a Ouija board. Every manufacturer does something closed, proprietary, and unusual.
In certain domains – the U.S. CDI, FHIR, HL7 V2, DICOM – it's good. Telemetry and consumer devices, not quite there yet.
Dr. Daniel Kraft: I'm still not quite there. I just had an imaging study back in the spring at Stanford. The only way I could get my cardiac study was on a CD-ROM. I don't even own a CD-ROM player anymore, so there are still some of those dots to connect.
Dr. John Halamka: That would be true.
Michael Krigsman: What kinds of platforms are necessary in order to provide the kind of results, John, that you were describing?
Dr. John Halamka: Why don't I start off with just a high-level look at what Mayo has built, because it illustrates the kind of componentry you need? Mayo has 154 years of patient data.
You say, "Wait a minute? How could that be?" Well, remember the medical record was actually invented at Mayo Clinic.
Back before TCP/IP, we had pneumatic tubes, conveyor belts, and chutes. You have paper. You have photographs. You have audio recordings – all this stuff.
We had to create a platform by which we could take all kinds of medical data, deidentify it, aggregate it, and then put it in an encrypted container for various algorithms, developers, innovators, and partners to work against. That was our clinical data analytics platform, deidentification storage.
Next, we wanted to be able to deliver care at a distance for serious and complex care, so we built a mechanism by which cloud-hosted components can be used for outpatient monitoring, dashboarding, taking care plans and rules, and helping us take care out of bricks and mortar that's serious and complex, not just ambulatory visits. Then finally, new ways of ingesting the kinds of data that Daniel and I were just talking about and building mechanisms of doing data transforms, so you can normalize a whole lot of these new data sets. We've built all three layers of components in the cloud.
Dr. Daniel Kraft: The key part to follow that seems to be, whatever platform you're on – there are some really amazing ones – all need to work into the workflow of the clinician who doesn't want to have to log into ten different apps or ten different platforms. Is there a Mayo experience that's synthesizing that since you're an integrated platform? What advice would you have for others who are trying to integrate? I always like the example, from data to actionable information and that actionable information then works at the bedside or on your mobile device.
Dr. John Halamka: Daniel, of course, and I have worked together since 1994, so he knows the hard questions to ask, which is the back-end, the AI, the data, the normalization. That's not the hardest part. The hardest part is the workflow.
You start asking yourself, "Well, okay. It depends a bit on your use case." EHRs have FHIR CDS hooks. Okay, there's a means by which inside the workflow with an EHR, in some places, can call out to a cloud service and get a response. Okay, that works.
All other SMART on FHIR apps, apps that we would layer on top of the EHR, they don't really give you a fractured experience because it's like an app on your phone, in a sense. It feels like it's part of the EHR itself.
Oh, and then there are sidecars. Now, these are not the greatest. I am in the middle of doing something in Epic or Cerner or whatever and then my Windows taskbar pops up and here is an alert or a reminder. Okay, it works. It's kind of in the workflow. It's a little clunky.
Then you start to ask, "Oh, well, maybe it isn't the clinician in the workflow that's necessarily the actor. Maybe it's, oh, a care manager and they're running a separate application." Then, "Oh, it's okay because the EHR isn't the center of the universe. Maybe it's a patient and they're getting an alert, a reminder on their wearable or their mobile device."
I think Daniel's question is so key. Ask who are you trying to change behavior with and then what is the vehicle by which you integrate into that workflow?
Michael Krigsman: Where does technology start? When you talk about workflow and changing behaviors, that has little to do with technology and research itself, right? That has to do with just the propensity, the difficulty of humans changing.
Dr. Daniel Kraft: I think, as clinicians, often the oil of clinical care is data, whether it's vital signs or labs. Now we're having more of it come at us.
There was a cartoon that was done in Exponential Medicine where there was the patient with the doc and the doc is like, "I see the problem. You're generating too much data."
Back to that sort of workflow. How do you synthesize that now that we have everything from our digitome, our sociome, our metabolome, our genome, and then make that actionable? Ideally, on these new platforms, learn from the clinician experience around the world, not just the Mayo way or the Stanford way, and synthesize the data into its actionable components.
No one wants to see the raw EKG data, blood pressure, or other elements, but what does it mean in context and even normalized to that individual? There are lots of layers to it, but it's an exciting time. We're starting to see the dots connect.
Dr. John Halamka: What if we agree there's going to be a Twitter stream connected to your bathroom scale so that, every morning, your friends, family, and colleagues see your weight? Oh, well, that's a daily social network for behavioral reinforcement. Oh, maybe that'll work, right?
You have to start thinking of who is going to get what value from what you are doing and use that to change behavior. No question, that's hard.
Dr. Daniel Kraft: Part of the challenge is, a lot of these things—whether it's your scale, your Fitbit app, or your EMR interface—are sort of one size fits all. I always like to use the example of, we don't have precision medicine but we need precision digital health interfaces that match the age, the culture, the language, and education.
I've got a connected scale. It's not connected to Twitter, but that's my most helpful thing. If I see I'm up a pound or two, I will cut out the popcorn that night.
To match someone's personality type in both incentives – you know, carrots and sticks – I think, as part of that, including for the individual clinician, nurse, or pharmacist. We tend to always build the system for the average.
Since I'm a pilot, I always love the example of the cockpit. After WWII, the Airforce tried to build the best new cockpit and they designed it for the average pilot. Of course, no pilot is average, and so pilots kept crashing the planes. So, something about the design element needs to come in as well.
Dr. John Halamka: Right. and what motivates you? We talk about social cues or pressure. "Oh, that may motivate you."
What about economics? What if you agree that every time you walk by a pharmacy and you're willing to share your geolocation and there's a coupon that pops up for a medication or product you might need as you're in proximity? Oh, well, that might motivate some people.
If you're a care traffic controller for your family, being able to get your children or your parents the care they need more easily. Oh, that might be a motivator. These are the much more interesting sociological parts of the platform we have to consider.
Michael Krigsman: John, as you're looking to build this data aggregation platform. What are the dimensions that are primary to you? Where do these workflow elements fit in? What are the data sources? Basically, I'm asking, what are you doing and where are you prioritizing?
Dr. John Halamka: At the Mayo Clinic, the primary value is the patient always comes first. That actually guides work by every part of the organization.
We've started to say, "Okay, well, if that's the notion, which is improving wellness, preventing disease, reducing suffering, and really making something that is going to improve the patient experience, what kinds of things would you do? We've started to think, oh, okay, how about cancer risk prediction? How about the ability to do more rapid diagnosis based on voice or other kinds of telemetry, or being able to do early diagnostic testing through AI algorithms that aren't traditionally used in this particular purpose?
We've started to look at those use cases. Then, underneath the use cases, what data elements do you need and how are you going to get those data elements and curate them? So much of what I've had to do is, yes, build the platform, some of the technical components, but then layer on top of that use cases that bring value, that improve the patient experience.
Michael Krigsman: Obviously, that then has its tentacles through the workflows that Daniel was describing earlier.
Dr. John Halamka: The ecosystem that we live in, it's patients, providers, payers, pharma. It's all kinds of new industry players and trying to connect all those dots and ensure that you have a coherent whole.
Daniel, did you ever work in the early days of Google Health or work on such things like HealthVault?
Dr. Daniel Kraft: I tried them out. The problem was, they were just a vault and there wasn't any real insight or other elements you would get from it.
I think now they're sort of evolving to provide personalized recommendations or insights. Even my Fitbit will look at my sleep and compare me to others my age and sex. But, yeah, they weren't particularly useful.
Dr. John Halamka: That is exactly the point I was going to make, which is, back in the day, we built these components and those components turned out not to be adopted because there was nothing end-to-end that brought value to you. That's the lesson learned from the past; we have to apply to the future.
Dr. Daniel Kraft: One thing that came to mind, John, with these new platforms, including all the big data and AI, some of the great work in Mayo taking a 12 lead ECG and doing all sorts of predictalytics on what does that really mean. Now others with ICU-based software that can predict sepsis or other things early.
There's also then the challenge of, there's no one number to look at. It might be a synthesized risk score. Then are physicians, in your experience, being resistant to that?
They don't understand what's underneath the black box. We don't even understand, often, what the machine learning is pulling from. As we get more data and sometimes there are magical insights, almost like the picture of the retina from DeepMind that can predict heart attack and stroke, how do we address the challenge of medical education and using that in smart ways when it's often a bit murky about where it comes from?
Dr. John Halamka: This is a great, great point. There isn't a Consumer Reports for AI algorithms, but there needs to be. Which is, I don't mind so much the black box issue as long as you say, "Here's the scientific paper. Here's the validation that was done on the algorithm with a data set that was different from the original data set used to develop the algorithm. Here is the population where it's going to be helpful to you or the ranges of utility." As long as I understood that I'm good.
To Daniel's point, we have an algorithm that can look at your 12 lead ECG and predict your ejection fraction with an AUC of 0.9. It's pretty good, in other words. Oh, okay. Well, if I look at an ECG and it just says on it, "Oh, this patient has a low ejection fraction. You might want to consider follow-up," it's not replacing me. It's augmenting me in a helpful way where, like, "Oh, I would have never thought of that. Of course, let's get this person an echo.
You just need a little bit of transparency for each of these algorithms because there are bad algorithms or biased algorithms we wouldn't want to use.
Dr. Daniel Kraft: The big challenge is if you're only doing European Caucasians for your genomics or your algorithms for your EKG to echo. There is a lot of bias that can be done. I'm not saying bad data, but not diverse enough data.
I love the All of Us trial out of NIH, which enables each of us to be a data donor, sharing your genomics, medical record, et cetera. They're starting to share that back. I'm not sure if you see that's it's useful yet, but it's sort of the Framingham Study on steroids that will hopefully make these algorithms much more equitable.
Dr. John Halamka: Well, and there's no question that the only way to develop these really equitable algorithms is by federating our learning because I could tell you, "Oh, we've developed this algorithm based on millions of patient lives in Rochester, Minnesota; Scottsdale, Arizona; and Jacksonville, Florida." Great. Does that work in Georgia? How about Southern California? Maybe yes. Maybe no. But if you brought in, oh, Cedars-Sinai and you brought in Emory and you federate, you learn, and train, you're likely to get a much better generalizable algorithm.
Michael Krigsman: We have a very important question from Arsalan Khan on Twitter who says, "The idea that every doctor should have access to all the data needed to make an informed decision is great, of course, but it's highly dependent upon the budget of the doctor and the organization they work for. Less budget means less access to these platforms and to this data." What about that?
Dr. John Halamka: So much of what I've been trying to work on is not only building these things and looking at the use cases, but asking questions about ethics, disparities of care, and equity. Of course, that applies to patients, but also providers.
You ask yourself, "How do you make this stuff generalizable?" Maybe it's not the best analogy, but Airbnb is a pretty sophisticated technology platform connecting producers and consumers but available to all at low cost with great utility. Many of these products that I am thinking about are things like, oh, you can get them on any website or you could get them on any phone. It wouldn't be a huge barrier to distribute a lot of these things.
Dr. Daniel Kraft: I think it brings up the idea we talk about of the social determinants of health, which is super key, but also the digital determinants of health. Does your patient have access to a smart tablet or even not even high-speed Internet. There are lots of parts, particularly in rural America, that don't even have low bandwidth access.
My perspective is a lot of these technologies democratize. This is my 11-year-old iPhone 2, which, 12 years ago, was pretty amazing. Now it feels slow and clunky. But even the power of this in the bottom billions' hands or the lower socioeconomic status can be hugely impactful. Now that you can connect the dots in platforms like an iOS HealthKit, and now there's one called CommonHealth for Android, I think it starts to help the equity piece as well.
Dr. John Halamka: There's no question there's a digital divide, however you want to look at it—whether it's geographic, whether it's educational, literacy, or all the rest—is quite real. We need to ensure that we are meeting our customers at their level of technological comfort.
The quick example, which Daniel has heard me give before, is I had invited a number of very talented engineers to a Medicaid clinic. They walked over to a homeless gentleman they said, "What's your favorite wearable?" His answer was, "Socks."
Assuming that you're going to have Apple Watch Series 6 or an iPhone 12 as a prerequisite is not going to work.
Dr. Daniel Kraft: There are sensors in socks, by the way, now for picking up diabetic foot ulcers, but there's also a digital divide, I think, amongst clinicians. The folks who are graduating from medical school now grew up on all these technologies. John and I were sort of at the cusp.
Actually, when we met when I was a medical student at Stanford and he was an ER-resident at UCLA-Harbor, I did a rotation there. We were the only two geeks with this Hewlett-Packard 200LX pocket computer where we could take notes and actually share clinical data. He gave me his whole database. It was amazing, but we were the only two using it in the entire probably state.
How do we start to educate clinicians who might be over the age of 40 to use some of these tools? I've been thinking about this because I'm always asked, "What wearable or app might be best for my patient to use?"
I've been starting—it's still a very nascent version 1—a platform called digital.health. It's a website where there's a bit of a digital health formulary, which could eventually enable you to match the digital tools and solutions for your patients. That's an example of a platform. I think we need the educational level for the clinicians as well to use them and match them to the right patient who may not have Internet access or has 5G.
Dr. John Halamka: This gets back to this concept I had about Consumer Reports, if you will, which is, let's look at the user experience or the validity of a product or service because right now a whole lot of our clinicians aren't sure what to buy or what to use. One hopes we get to the point where a lot of these technologies are just passive.
Wouldn't it be wonderful—I'm of course not advertising any product or service here—if I walk over to Alexa or Siri or whatever ambient listening device you have of the day and just speak and, oh, a diagnosis comes back? "There's a 67% chance that you have a neuromuscular disease based on the content and cadence of your speech."
I didn't have to do anything other than just talk. That's the kind of thing I think we're going to hopefully see more of from these platforms.
Michael Krigsman: John, to what extent are you focused on these non-technical issues? When you talk about patient experience and user experience, I separate that from the hardcore dimensions of the types of data sources. How do you divide up between those two?
Dr. John Halamka: Mayo Clinic has two organizations: the platform organization, which I run, and the Center for Digital Health, run by Rita Khan. Does Rita come out of the tech industry? No, she comes from Target, UnitedHealthcare, places that have looked at how does one create a product for a consumer. All of this usability and patient experience is parked in an organization with really deep expertise in that realm.
Michael Krigsman: Daniel, as you look forward, how do we take these innovations and make them accessible to clinicians? I think this goes back to the workflow issue that you raised earlier.
Dr. Daniel Kraft: Well, number one, it's potentially aligning the incentives or understanding the misaligned elements. If you prescribe that connected blood pressure cuff to help manage hypertension, then that data could come back to you. Are you rewarded for that as a clinician in some for or does it not just become a barrier?
I love the Consumer Health Report element. How do you provide the ability? When you're seeing a patient in the clinic who has got hypertension, depressed, is a smoker, and diabetes, here's a set of not just drugs but other wrappers and platforms that they can use and not have them overwhelmed with ten apps but prescribe a bundle of solutions. It might be their connected blood pressure cuff and an app to help them do smoking sensation and something for medicine adherence. It's integrated and not scattered amongst ten elements.
Then for the clinician, when you do prescribe that, hopefully, that's covered by a payor, in the future for some of these, whether it's a wearable or ambient sensing or voice for diagnostics. Then in the workflow, that comes back to the clinician in a way that's actionable and not overwhelming.
Imagine you're a primary care doc with 2,000 patients. You don't have that reactive mindset where you wait for them to show up in the ER with a heart attack, stroke, or late-stage cancer. You're seeing a dashboard that might indicate from their sleep data that their resting heart rate went from 50s to 70s and something is going on; you might need to call them. Or their blood pressures are out of range based on where you dialed them in. It's complex, but I think we're starting to get to that realm.
Some of the payors are starting to think about that digital formulary and rewarding those elements in a cohesive sense. The challenge is, we have thousands of healthcare systems in the United States, let alone the world, so it's not a one-size-fits-all.
Dr. John Halamka: Well, and COVID has had a material impact on all the things that Daniel has just described because in, say January, maybe 3%, 4% of our visits were virtual. By April, 90%. Okay, now November, December, maybe it's going to be 20% - something like that.
The point being is we've gone from 3% or 4% to 20% in one year. That has implied that we needed to get familiar with connected devices in the home: virtual visits, products that are commercially available on Amazon for patients to help navigate healthcare at a distance workflow. I think, in many ways, these last ten months have probably accelerated our industry five years.
Dr. Daniel Kraft: My favorite phrase is stolen from Regina Dugan who used to run DARPA. Sputnik sparked the space age and COVID is sparking potentially the health age and accelerating some of these elements.
Part of the challenge—and, John, I love your thought on this—is we still tend to digitize things. We digitized the medical record, which kind of sucks, or the fax machine. I think, hopefully, COVID as a catalyst is going to make these smarter systems, not just your pulse oximeters connected through your virtual visit but that you do the chatbot visit first that can really do the triage and the 20 questions. Then upscale those uses or giving some resiliency to the system because you can get burnout on bedside care as well as Web-side care. A lot of the elements can be almost driven by an AI or chatbot whether the clinicians like it or not.
Dr. John Halamka: Well, and I completely agree that as we move to more virtual visits as a new normal that it isn't just taking what used to be a bricks and mortar visit and making it digital. That's not helpful. It's asking questions like, "Oh, I woke up this morning and, you know, I'm feeling kind of tired." Not really, Michael.
"You know this cup of oolong tea? I can't quite taste it anymore." I go to a virtual visit and say, "Lost my sense of taste and smell. Feel tired." Immediately, then I am taken to a workflow that's going to help me get my COVID test.
"Oh, how should I get that COVID test? Should I go drive to Quest or LabCorp?" No! It is going to be delivered to my home, either as a swap that I would self-administer or a lateral flow assay like a pregnancy test, I just run myself.
These are the kinds of things, end-to-end, that are new, healthcare at a distance, requirements accelerated.
Dr. Daniel Kraft: What's going to come out of this, even on the rapid testing, so we're doing this rapid COVID testing enterprise. There's lateral flow. There's home-based PCR. There are already some of the lab-type things that'll plug into your mobile devices.
Hopefully, these will be platforms we'll continue to use that you just change the cartridge and it can check for the flu or other viral illnesses. Whether we're going to from hospital to home or hospital to hospital, we're going to have a whole set of not just virtual visit tools, but the idea of the medical trichord or other sort of elements that almost every home is going to have, that will dramatically shift how we do proactive prevention, diagnostics, and therapy.
The trick is to get it paid for, the workflow, interoperability, and the licensing issues. Then the chatbots will be federated and learn so that if it's John or Daniel answering the questions and I've got abdominal pain, it already knows that I got my appendix out or what my med list is, and so it's smarter about the triage and the flow.
Dr. John Halamka: There's one other thing that Daniel has invented, Michael, that you need to know about. We've talked about digital diagnostics but what about therapeutics? Oh, that's harder.
I think this is just brilliant. Hewlett-Packard gives you the printer and sells you expensive ink. Well, what you need is a medication printer. It's basically free, but then you have cartridges so that, "Oh, your doctor has just diagnosed you with this or that. In fact, that digital therapeutic is now delivered at a distance because the pill you need is printed in your home." I bet you have something to show us, Daniel.
Dr. Daniel Kraft: I was not trying to plug this but the idea I've been thinking about since I trained in pediatric and adult medicine and, in peds, we weigh every kid and dose by dose. But when you're an adult, all of a sudden, you get the same dose of everything or use a pill-splitter. The challenge of pill adjustment and combination and the idea of, essentially, in our digital manufacturing is 3D printing a personalized pill. That might be my aspirin, my statin, my beta-blocker, my Synthroid.
Some things might need to change day-to-day like my Lasix dose or Coumadin, make me measure from my home lab. Then, eventually, you would have a little home printer. These are just sugar pills, but this could be a little home printer that would adjust your meds for folks going polypharmacy, even just a couple of meds, as needed on a continual basis. That's called the telemedicine. There's a TED Talk about that if you want to see more details.
There are still challenges about regulatory. Is that automated compounding? How do you build the sliding scales? How do you build the AI to know how to best adjust your blood pressure meds, is a simple example, without the doc being on the phone and faxing your numbers back and forth.
Michael Krigsman: We have a question first about cybersecurity. Let's talk about that and then let's shift to some of the ethical dimensions because we have a question relating to that. This is from Khwaja Shaik, and he says, "Cybersecurity risks are huge for connected devices." He says, "It's time to adopt micro-segmentation and zero trust architecture principles to accelerate secure healthcare outcomes." As we're sharing all of this data, how do we make sure that it's secure and not used against us?
Dr. John Halamka: Well, I'll start off simply by saying multilayer defense. As we know, antivirus works some of the time. Firewalls work some of the time. AI to detect anomalies in data flows or port use, that works some of the time.
I'm seeing more and more companies evolve to the question that are offering new approaches to look for anomalous behavior. They've trained what looks normal, "Oh, this looks abnormal. Raise a red flag." There's no question that these techniques we've heard about, such as micro-segmentation of networks or zero trust architectures, are part of the solution.
Dr. Daniel Kraft: Isn't the solution just supposed to be blockchain? I learned and forget what that means many times. What's your take on that since that seems to be key?
Now we're seeing, even in the setting of COVID, hospitals being hacked and held ransom with ransomware. What's your take on blockchain since it's a bit of a buzzword too?
Dr. John Halamka: Well, as you know, I'm editor-in-chief of Blockchain in Healthcare. Why did I choose to be editor-in-chief of that? To tell everyone what blockchain doesn't do. [Laughter]
Blockchain is a mechanism that really does help us when we need to have nonrepudiation. Let's imagine, in a time of COVID, a Pfizer vaccine comes out and Daniel gets two doses and now – I'm making this up, Michael – he wants to go to a rock concert and the rock concert says, "As long as you bring proof of vaccination, you'll go in the door."
Okay, well, he pulls out his phone. He's a pretty smart guy. He creates a phony QR code. Well, the blockchain notion is, what if Pfizer or the administerer of the vaccine does a secure transaction to a blockchain backend and the QR code is generated from the blockchain backend? Oh, well, that pretty much says, "He couldn't have faked it." Where there's need for proof, blockchain at least tells you an audit trail can be believed.
Dr. Daniel Kraft: On that topic, which I think is super important now that we're hopefully going to have cheaper, faster testing at home, school, or work. You want to prove your negative to get to that concert, or that you've been vaccinated and you've got antibodies, let's say.
There's a platform that came out of part of our enterprise called CommonPass from CommonHealth and it's literally a digital yellow card. They're tying it into airlines and sporting arenas. I think that the CLEAR folks have something similar as well.
We're going to need those sorts of technologies to connect the dots on our vaccine status as well as to do the sort of Stage 4 follow-ups. Are you having side effects? How long did your protection last? This is still a big, global experiment in the COVID era.
Dr. John Halamka: The good news – just to this topic, Michael – is that the industry has agreed on one common standard to represent proof of vaccination—the health card standard that Josh Mandel created as a FHIR implementation guide—which is just one of the standards that are being used by the CommonPass or the various Apple kinds of healthcare type products.
Michael Krigsman: We have another question from Twitter. Personal health technologies are one thing for mentally "healthy" people. What about those suffering from dementia or other mental illness? I will add that one of the things you're describing with healthcare at home with sensors at home is, to some degree, you are offloading the burden of collecting data and tracking data from the clinician onto this distributed workforce, namely patients. How do we manage that? I think that leads us directly into some of the ethical issues.
Dr. John Halamka: My mom is almost 80. I said, "Hey, mom. Do you know that the information blocking and interoperability rule now allow you to use a FHIR R4 API to download your healthcare data and be the steward?" She said, "I have not a clue what you just said and I'm more than happy to let you do it, but I don't want to do it."
I could hypothesize that we are going to have a whole new kind of healthcare worker that will be the health coach, the care manager, this care traffic controller kind of person who does assist those without, say, digital comfort, getting these digital workflows implemented.
Dr. Daniel Kraft: The digital workflows are getting simpler. We mentioned earlier, voice. You don't need to train your 80-year-old mom to use Alexa. Just say, "Hey, did I take my meds?" or "Help. I've fallen and I can't get up," or to track voice or the work out of MIT with wi-fi that can seamlessly pick up behaviors and even sleep patterns and vital signs.
We're kind of moving to this ambient sensing world without wearables. That leads to privacy issues and who owns the data. When it picks up mental health changes from your voice, your behaviors, or your sleep, who gets notified?
I love the idea of the health traffic controller. Some of those are going to become more and more automated and up-level to the sun or the clinician only at the right sort of gating.
Dr. John Halamka: This is an Arthur C. Clarke rule that my mom, because I happen to have created platforms in her home, walks over to a device, just talks to it, and it just works. I didn't actually even train her and she's going and giving commands to these devices. It's like, "My God! This is actually magic."
Michael Krigsman: What are the primary ethical issues when it comes to the unique aspects of healthcare data such as the type that you're collecting and aggregating?
Dr. John Halamka: There are so many ethical issues. I serve on a number of national and international data ethics advisory boards. What data can be used from whom, for what purpose? If value results from an algorithm or something derived from that data, should the person who contributed it get a micropayment? What consent is needed if data that is de-identified and aggregated is used for algorithm development? These are all unanswered societal questions.
Dr. Daniel Kraft: Well, I like to think of it from the perspective that we all should be incentivized to become not just blood donors, organ donors, but data donors and get something back. My overused example is Google Maps and Waze. We're sharing our private data, our speed and location, and it builds the traffic map. We don't get paid to report that there's an accident or slowdown. It helps inform and sometimes it's gamified with points on Waze.
In general, we all benefit from donating, still hopefully in safe, anonymized ways. In fact, there's now a platform called Stuff That Works out of Israel where, if you've got a condition from psoriasis to migraines to cancer, you can donate your data and find others like you. Then it'll give you synthesis of information back. Hopefully, it'll be like, if you're a data donor, you'll get the benefits from that at big levels or small. We align that piece.
Also, maybe we need to reinvent or reshape HIPAA laws, which were predigital, so that they're not fear-mongering in terms of enabling folks to share data when they want to, in clinical settings or outside.
Dr. John Halamka: A couple of elements to that. Mayo is in the process of rewriting its general consent to give five kinds of choice. Here's an interesting challenge, Daniel. You can't just give infinite numbers of choices because then you're just not going to be able to enforce it or respect the patient's will, but you could probably give about five and say, "Oh, you say you don't want to contribute your data to research or product development. Okay." It's very discrete, right?
We're finishing up that. We'll be holding a conference in March, bringing together a number of international experts to look at the consent and get opinions.
Everything that I have done is done with de-identified data that has been certified by a third party expert as deidentified. Meaning that it is pretty hard to reidentify. The definition we use for deidentified is the data is sufficiently ambiguous such that it cannot be distinguished among ten different patients.
We've done this restructured data but, just last month, we released about seven million unstructured patients' records, notes, reports, and that kind of thing. Devilishly difficult to deidentify text but we have done certification with third-party experts that we've succeeded.
Dr. Daniel Kraft: It's interesting, I think, when you start to deidentify that and make it open. You can run data challenges on those and maybe identify biomarkers of disease or pathways of care that were not visible.
Our friend Atul Butte at UCSF has access to all the prescribing data amongst all the UCs. They prescribe very differently for, let's say, diabetes, et cetera. If you can get insight into that, you might find the ones that are really most evidence-based or most health economic impactful. I think the more we can federate and anonymize and share that data, let people mine it and un-silo – Mayo connects to Stanford to Harvard to Geisinger to Kaiser to NHS – a lot of things can accelerate.
Michael Krigsman: Khwaja Shaik makes the important point on this ethical issue. He says that ethnicity of data also matters. Solutions built on U.S. geographic data, for example, may not work for India. That's another layer to this ethical question.
Dr. John Halamka: This whole question that we've been talking about is equitable care, disparities of care. One aspect of that is bias in algorithms, and how do you even detect bias?
One of the projects that I've been working on as part of the Mayo Clinic platform, but also Mayo Clinic in general, is bringing together a number of experts to be able to say, "Can we develop the sensors that understand when an algorithm is biased? Or if it is biased, is it useful in some subsets versus others?" You'll see a whole lot of activity in papers coming out trying to answer such questions.
Dr. Daniel Kraft: Speaking of India, on not just the bias side but on, let's say, the genomic side now that it's getting cheaper to do full sequencing, there are groups trying to make sure you get folks from Asia and not just Europe. Then inform the algorithms when we're trying to do pharmacogenetics or polygenic risk scores that it's based on the population you come from.
Michael Krigsman: We have another question from Twitter. You can see I really like taking the question from Twitter and LinkedIn. This is from Irma Raste. Irma says, "What is the role of clinical patient registries in all this?"
Dr. John Halamka: Well, so why are registries important? We've talked about a lot of elements of a platform. A registry often takes much deeper data than would be stored in an electronic health record or an administrative billing transaction. It's used because it's disease-specific.
I will tell you; I've worked on a lot of cancer registries and so much of the cancer data, which is staging data, biomarker, genomics data, you're having to pull today out of text and that's not always easy.
I guess the last example I'll give you is in the time of COVID, especially as we've tried to understand efficacy of various treatments. Registries have provided us with data that we would have never gathered in a traditional way. Registries are really an important adjunct data source.
Dr. Daniel Kraft: Now, even the idea of the patient as a scientist, particularly with lung COVID, there are some good examples of patients themselves, who had all these strange neurologic and other symptoms, building their own sort of registries and crowdsourcing those that are going to be super helpful for understanding the long-term sequelae and how to treat them.
Michael Krigsman: Now, you've mentioned COVID and maybe that's a topic that we should talk about before we leave. Everybody reads the news. We know what's going on. Personalize it for us. What's actually going on right now with regard to COVID? Is there a problem?
Dr. Daniel Kraft: I'll just say one short thing if someone hasn't read The Atlantic article out by Ed Yong today about the impact on COVID on front-line clinicians who have survived a couple of waves and are now completely stretched thin, in the Midwest especially. We're speaking now in November of 2020. You'd think some of these things like even PPE that were an issue back in March/April weren't still an issue today. I think we're in a dark space. We hope the vaccine is going to come, but we're in a pretty challenging element right now and I think we still need to band together and do smart science and public health if we're going to pull out of this.
Dr. John Halamka: Since March of 2020, I have co-led the National COVID-19 Healthcare Coalition of 1,200 companies for situational awareness, for understanding PPE supply and demand, for looking at contact tracing, testing efficacy, cures, and vaccines. What can I say as of November 2020 is our current issue? We are in a moment – Daniel used the term dark – where we have unchecked, exponential growth of COVID in this country.
Yesterday, we passed 160,000 unique cases. It depends what model you believe: University of Washington, IHME, MDR. There's a whole variety. The models are currently suggesting, as of November 2020, that we will have one million dead by February. That is a wake-up call.
Michael Krigsman: It is pretty extraordinary to watch the numbers, those graphs on the news every day. It wasn't too long ago, a week, where the number of cases was 100,000. Now we're up to 160,000, as you said. Regardless of what the government may do at this point, short of immediate and very dramatic behavior change across the country, is there anything that can stop the continued exponential growth of this disease right now?
Dr. Daniel Kraft: Wear a damn mask, socially distance, don't get COVID fatigue in that sense. Stay vigilant for your personal self, your friends, your family, and your community. That can certainly help. We are in this exponential phase. A doubling—1, 2, 4, 8, 16, 32, 64—it moves quickly and people tend to think linearly. I think we all need to redouble our basic elements.
Hopefully, get our rapid COVID tests at scale that we need, even if they're not perfect. A test is going to be zero percent sensitivity and specificity if it's not taken. That's an element that needs to accelerate, and then supporting our folks on the frontlines. Hopefully, with the incoming Administration, getting real science and policy put into place.
Dr. John Halamka: Wear a mask everywhere, every time, no exceptions. No gatherings greater than ten people ever for any purpose. Social distance. Don't travel. This holiday season, don't travel.
These are just the very basics and I know so many of these messages may have been politicized. All you have to do is look at the data. It's the only way we're going to start to bend the curve.
Michael Krigsman: We've been given all of this advice since almost the beginning of this year. It hasn't worked so far, and so my question really is, is exponential growth for the next couple of months virtually inevitable to continue?
Dr. John Halamka: I hope not. I think Daniel and I both recognize it's not going to be one size fitting all. There's going to be regional variation.
Yes, there are certainly cures. Yes, there are vaccines coming. The challenge is, we're in this for the long-haul.
Even when the vaccines are fully deployed, there are 7.7 billion people on this planet. Depending on whether it's a single dose or double dose vaccine, let's say on average we're going to need 12 billion doses of vaccine. Just go look at the supply chain to deliver 12 billion doses of vaccine. It will take two to three years.
I really hope that we just don't lose this war through attrition and refusal to change. That as people recognize that the exponential growth means the death of loved ones, they will recognize that this is not a political issue. It's a survival issue.
Dr. Daniel Kraft: On top of the sort of pandemic, of course, the info-demic has exacerbated this, whether it's just lots of information and hard to grok, all the way to, on purpose, mal-information. We already had the issue with anti-vaxxers. We have a huge issue that's already here with vaccine resistance. Folks, even clinicians saying, "I'm not going to take it for the first six months."
Going forward, we need to re-establish trust across folks. We're not perfect. Anthony Fauci has changed positions as data and information has emerged. To get back to a setting where the info-demic doesn't make the pandemic twice as worse.
I would close it with a stronger suit. Maybe, John, you can riff on this just briefly is that the silver lining of COVID is that there's an amazing amount of innovation, whether it's 3D printing ventilators or new vaccines, new collaborations, new datasets, a lot of energy from virtual care to testing platforms that will hopefully transform healthcare in positive ways and will, big picture, hopefully, save more lives than COVID will take in the future. There may be some silver linings out of this.
Dr. John Halamka: What has happened in the time of COVID is competitors have come together for collaborations that I have never seen in my career. This means that we're moving forward on gathering data, distributing PPE, looking at novel testing, looking at novel therapies, disseminating evidence, and working together as a unified coalition. Sure, we're in a dark place, but the number of parties working together to make it better is many and it spans government, academia, and industry.
Michael Krigsman: All right. On that inspiring note, thank you, everybody, for watching. Thanks to our amazing guests. Be sure to subscribe to our YouTube channel. Hit the subscribe button at the top of our website and we'll send you excellent emails.
Thanks so much, everybody. I hope you have a great day.
Next week, we're speaking with Eric Yuan. He's the CEO of Zoom. Check it out.
Wear a mask, everybody. Bye-bye.
Published Date: Nov 13, 2020
Author: Michael Krigsman
Episode ID: 679