Primary care was meant to be the front door to the healthcare system, but in some ways never set up for success to begin with. We need a new operating system for primary care—one with a different, deeper understanding of the patient, the context of their world around them, and the processes we have in place to figure out who sees a doctor and when, to use the system most efficiently.
In this episode of the a16z Podcast, we talk about what the primary care of the future should actually look like; what kind of data about patients we should be collecting, from where, and to tell us what; how you ask the right questions of that data, to use the resources of our healthcare system most efficiently and for the best care; and what the PCP of the future might look like. Joining us for the conversation are General Partner Julie Yoo, physician entrepreneur Ivor Horn, a primary care pediatrician for more than 20 years, and Jeff Kaditz, CEO and founder of Q.bio, a platform that identifies and monitors each individual’s biggest health risks.
Hanne: We’ve been seeing COVID and the coronavirus put enormous pressure on the entire healthcare system. So, let’s talk about what the effect of that has had on primary care. Where have we seen primary care succeed in this moment? Or has it? Or where have we seen it fail? What are we learning about the cracks in primary care from this particular moment?
Ivor: We all remember the primary care of older times when it was our doctor in our community, and that doctor knew about that community, and had the trust of the community. And one of the fundamental things and foundations of that primary care of was that experience with trust and being able to share information with that provider.
I think some of the things that have been helpful about primary care is the fact that there is that level of trust. Yet, that’s also where things broke down, because people ran to the place and the space where there were limited resources, and overwhelmed that area, and there weren’t the opportunities taken to use other mechanisms such as telemedicine or telephones, to communicate with people and to do that triaging that needed to be happening rather than people being exposed, even in the doctor’s office.
Julie: Yeah, it is the, what we call low acuity sort of entry point for care. whether it’s a sniffly nose, or a rash or, you know, something very basic. a patient can get a very quick evaluation and not have to necessarily see a higher-end specialist, or go to a hospital or some other more sort of expensive and more complex to the care setting and, you know, essentially get their needs taken care of in the most cost-effective way possible. Primary care was really meant to be the front door to the healthcare system. The unfortunate irony of the current situation is primary care was already at almost a crisis level with regards to access. You know, your ability to actually get an appointment with a primary care doctor despite the fact that that is actually the most appropriate entry point, was sometimes months, right?
Jeff: There’s just a very fundamental economic fact, which is the most scarce resource we have in healthcare is doctor’s time. Doctors are extremely expensive to make, and not to mention the fact that the ratio of GPs per capita globally is going down. So if their time isn’t used effectively, that’s the most wasteful thing we could do in healthcare. This whole flattening the curve—just in general, primary care should be about flattening the curve. Really it’s a failure to be able to quickly and accurately route resources based on need and priority. If flattening the curve was really about not overwhelming resources, how do you prioritize those resources? Well, people who need care sooner should get it first.
What this is exposing is our ability to potentially effectively triage and segment risk in a population quickly so that we can prioritize who needs and gets attention. What we really need to figure out is how do you make it easy for a doctor to know on a continuous basis who’s at the highest risk, who do they need to spend time with, in order to really focus their care. Because if we can pick out the one person who needs to see a doctor in any given year out of 10, that means the doctor could effectively care for 10 times as many people.
Ivor: The other thing is all of the people that are around the doctor that also provide support to patients that we haven’t actually utilized effectively, whether it’s the nurse, or the front office staff person, or especially community health workers who know the context in which people live, to actually do some of that early stage understanding of who really needs to see the doctor and how you can communicate with him on a more regular basis such that when they do need to see the doctor, they actually are coming in, but that time is of use and used appropriately and well.
Hanne: So at the moment, this sort of triage thing is done in, like, the most inefficient way, where people are literally left in a giant vacuum of trying to get, you know, in a telephone queue and describe some vague symptoms that one person may describe in, like, a completely different way. You’re talking about a different kind of both support and information gathering for that type of triaging. So, let’s talk about what that could look like.
Jeff: Traditionally in medicine, you measure something if you want to diagnose something. I think that we have to move away from that notion. We should think of measuring information as health monitoring, not looking for illness. That’s how we’re going to get to much more sensitive diagnosis, is thinking about when we see patterns, you know, or accelerations of changes across multiple variables. But to embrace that, we have to stop thinking of screening for disease versus monitoring health.
I think the way to think about it is a spectrum. There’s low-fidelity, high-frequency data, right? And then there’s high-fidelity, low-frequency data. And there’s lots of information between. When actually information needs to be gathered from a person that requires a physical visit, does an actual doctor need to be there, or can that information gathered very effectively so it’s available when the doctor actually has a conversation, whether it’s in-person or remote? In theory, no doctor should meet with the person unless they required intervention. And if the system was really optimal, that’s what would happen.
Hanne: Can you give an example of what that looks like?
Jeff: Well, I think it’s different levels of triage. I think, in theory, you could be monitoring somebody at home, and based on changes in risk, say, “We think you need to get a lipid panel done,” right? And then based on that lipid panel, say, “We’re going to notify this doctor that you should schedule a time to talk to them,” and automatically connect them in the next week.
But you can also imagine an intelligent scheduling system that went into this, that would actually prioritize a doctor’s schedule based on need, right? It’s tragic if a person is going in for just a general checkup to say how they’re doing, and like an 18-year-old, healthy person with no health risks takes time from a person who is having, like, severe chest pain, you know, and has a lot of indicators that says they really should talk to a doctor.
We think there’s just fundamentally a missing layer to primary care, which is this automatic data collection layer which automatically determines what is the right set of things to monitor about an individual, and then can alert an individual and a doctor when a doctor’s time is required to intervene and have a discussion.
Ivor: What’s really important for when we’re thinking about the tools, recognizing that primary care has to be able to not understand that information in the silos, but along and across the care continuum, and how do providers begin to connect that data and prioritize that information in how they support and provide care. People are not entering into the healthcare system at one place. They may be entering into the healthcare system at an urgent care clinic, or via telemedicine, or via a sub-specialist for that matter.
Julie: Yeah, and I think you’re highlighting that it’s not just the information chasm that leads to all these challenges, it’s also the logistics challenge as well. And we think a lot about movement of healthcare into the home, and the fact like you have to go to your doctor to even determine that you need a certain lab test, and then you have to wait for the lab test to be done, to come back again to your doctor to actually interpret those results and then get your care plan.
You hear all the time about patients deteriorating in that window of time when they’re waiting for those things to happen, when had you done that test upfront before they came in for their first visit, you may have been able to act on that sooner.
And you see the same thing on the flip side, where after you discharge patients from let’s say, a hospital or other acute care setting let’s say you’re a heart failure patient, you know, generally speaking you’ll want to set that patient up with check-ins after they leave the hospital. But many of them end up actually getting readmitted into the hospital because they don’t get the care that they need.
Hanne: What is it that’s so hard about just flipping that one simple thing? Like, why would that be? What is it about the system and the way it’s set up that would make it so hard to just flip that?
Jeff: There’s a general problem that we’re talking about, which is overload, right? Like, that’s why flipping the switch is hard, is because there’s a new class of clinical decision support tools that needs to be there. Otherwise, you’re actually creating more work for a doctor. Like if you measure 1,000 things about every person and a doctor’s expected look through those things, that’s not reasonable. So, you need to have intelligent tools that can actually highlight the key things.
Julie: Because it like flips the whole paradigm on its head, right? Because, like, the current system is that the patient has to determine whether or not he or she needs to go see a doctor versus shouldn’t it be the doctor who actually knows when to reach out to you.
Ivor: But one of the things that we also need to consider is the context of that data. Understanding the context and the environment in which people live and what that data means in the context of their life. You may have someone who has a cardiac condition and has a cardiac treatment, and not having the context of the fact that there’s no one in their home, there’s no one to actually acknowledge to them that they’re having a change in their status to say, “You’re not breathing correctly, you need to call in,” if we do or do not have that data following them in that short period of time, it matters in how we triage that data and how we bring that data forward to the provider.
We have the capacity to bring information and data forward to providers in a way that prioritizes that, not just based on what the labs test shows and what the trend of the lab is, but also some of those social factors and those behavioral factors in context, is this person not moving as much as they typically would How do we take that into consideration in that dashboard that a provider gets?
We all know that there’s bias in data. We know that people have not collected race, ethnicity, or language preference data, and how we interpret that data, right? And what comes up in that algorithm or what comes forward in that clinical decision support tool. And it’s really important for us to not run away from those biases and ignore them or say they don’t exist, but run to it. Identify it. Correct it. Make the changes that we need to make, ask the questions that we need to be asking, so that as we’re moving forward, we’re actually improving things and making them better, that we’re including the communities that are impacted by these biases as we’re building, and while we’re building, and getting their input along the way to make sure that what we create is for everyone, and creating more equity as opposed to more inequities in care.
Jeff: That’s a huge part, I think, of it. the context is so important to determine whether or not a measurement or trend is significant. We’ve spent a ton of time figuring out how we weigh the significance of the measurements based on genetics, lifestyle, medical history. I think the right way to think about it honestly, is you can call it an OS, or even an analytics platform for the body, again, where the goal of the system is to monitor what’s changing so by the time a doctor sees a person, they actually understand and have all this in context, and the tools to understand where this person lives, how is this person like other people where they live, other problems people have had in that area.
Julie: One of the paths to overcoming these challenges that you’re describing is actually to go, you know, to think beyond the electronic health record, because I think so much of the bias that does exist today is that we’re relying on these highly structured, very sporadic, right? You know, Jeff, you said earlier, the low-frequency, high-fidelity data points. Like, that’s pretty much solely what we depend on today in traditional medicine and traditional primary care. Whereas, like, the vast majority of insights that probably determine, you know, both your current state as well as what your progress is going to look like over the course of time, comes from everything else, like all the social determinants and behavioral and demographic-related information that Ivor is describing.
Part of the challenge of why we have so much bias and why it’s hard to overcome that is that we haven’t collected that data historically. just the notion of, like, longitudinal data between physician encounters, that is completely unaccounted for in traditional medical record systems.
I mean, even when you look at these, you know, chat bots that are popping up everywhere to help us triage whether or not we need to go see someone for COVID-related issues, none of those questions are being asked. And so I think that’s one of the huge opportunities here, is to really open up the aperture on the nature of data that’s been collected.
Jeff: I mean, if you think about EMRs, they’re designed to administer and bill, right? Most of the information we have in EHRs are biased towards sick people, they’re biased towards people who have access to care, and when we talk about, like, a healthcare system that gets better, unless we can decouple measuring the human body from care decisions, which are opinions, at the end of the day, they’re predictions like saying, “This person should do this,” we will never actually close that feedback loop, because we can’t look back retrospectively and say, “Okay, knowing what we know now, would we have come to a different opinion?” If you’re just capturing the opinion not the inputs of the opinion, you can actually go back and learn.
One of the interesting things that you’re talking about, Julie, is if you take a step back in thinking about almost a person that goes out and interacts with their environment as a sensor. I actually see the future of healthcare being able to prevent things like Flint Michigan if you were actually monitoring the population and the clinicians had access to information, you’d see a change in population health as soon as those water pipes were switched, not two years later when it was damaging kids’ neurological systems.
Ivor: Understanding all of those social determinants of health, one of the things that we’ve learned as part of this process is the context in which people live, learn, work, play, pray, can’t be bucketed into just housing or just food and security. It has to do with a context of the number of people in your home, the needs of those people in your home, what your job is, and the requirements of your job, and the limitations of what you can and cannot do for your job. All of those things impact on the data that needs to come forward.
When we talk about social determinants of health, we often talk about the negative consequences of social determinants of health. Yet, we don’t often talk about the fact that people may have a community and a social network that impacts on their ability to get support, that we didn’t understand or that we didn’t tap into.
We didn’t think about the level of resilience that a person has, and what are the things that influence a person to actually do more in terms of their exercise or the way that they’re eating that should come into play with that provider being able to give more effective and more useful guidance to that person when they come in, when they’ve been triaged accordingly.
Hanne: So, other levers you can pull besides a prescription, besides a diagnostic test, besides an office visit, but communities and support.
Ivor: Exactly. And some of those things can be done via telemedicine. we often think about it in this one-on-one video perspective. But, one, there’s a lot that you see in a telemedicine visit that’s around a person, that gives you context.
The other is the simple use of a telephone conversation, and using that as a tool for checking in and that being an important factor in making sure that we’re creating more longitudinal data. The value of longitudinal data is so important that we don’t take into consideration, because we piecemeal it togetheras you said, in those low-frequency, high-fidelity EMR-type visits. But we have that sort of those more frequent steps that we get that actually broaden our understanding of a patient in ways that we never could do before.
Jeff: I actually think the key to personalized medicine is really in the ability to figure out what are the most important things to track about each individual, based on their risks. based on this person’s genetics, medical history risks, what is the subset that actually needs to be monitored about this person and the frequency?
And all this telemetry is just connected. That first order triage or the collection of data should almost happen passively without a doctor having to worry about that the right things are getting measured, so when the time comes and a person is, let’s say, they have to be rushed to the ER or they start to have symptoms, a doctor has all the context that they need, right now if you get rushed to a doctor, the doctor starts with almost nothing in the ER, right? And it becomes an information gathering journey before any decision can be made.
Hanne: Right. I hear such a sort of tsunami of, like, new types of data available that can be incredibly valuable, aren’t being used the way they should. And also, like, major shifts in the entire orientation of the system. what is the sort of management process and pipes that need to be built to make this vision closer to reality?
Julie: Today, we only measure the things that are diagnostic in nature. And part of the reason why is that those are the things that get reimbursed, right? And so I think, like, that’s a huge part of the answer to this question is, how do we not just create the pipes, but how do we actually make the cost effectiveness argument that measuring that data actually has enough clinical utility that makes sense to pay for?
Part of why we’re in this challenging spot is the fact that we were reliant on a system that only paid for individual tasks, therefore, it didn’t make sense from a payer perspective to reimburse for a million things to be done. It only made sense to reimburse for the things that, you know, really mattered and really move the needle.
Whereas in the value-based care world, they are able to innovate in unique ways to take advantage of new data sources to engage with patients in ways that wouldn’t even fall into the definition of clinical medicine, you know, 10 years ago, but are now absolutely the direction that primary care, in particular, is headed. And we see that in light of programs like the primary care’s direct contracting program with CMS, you know, more and more ACOs, you know, getting traction with even commercial payers, etc.
Ivor: You’ve got to realize that really a little over one in nine people actually have health literacy enough to understand how to manage their healthcare, and manage the healthcare system when you start talking about health literacy. So the ability to communicate and translate that information into a way that people can effectively provide and support themselves in their care journey, because the majority of their care journey will happen outside of the four walls of any healthcare system.
And any information that we can get that allows them to do that effectively means that they’re going to have better outcomes, means that they’re going to have better quality of life, and means that they’re going to have better quality care.
And so understanding those fundamentals of how we use data across that care journey is really important. As a primary care provider, sort of the onslaught of information that we have from wearables, from our mobile phones that tell us how people are moving can be overwhelming if it’s given all in one place and not with any context or with any prioritization. I think that’s the journey that we’re on when we start looking at, it’s important for us to get this data and it’s important for us to understand this data in context of what we do, and there’s the data for the primary care provider, and there’s the data for the person.
Julie: And I think that highlights the fact that patients are not actually an end user. That’s of consideration when it comes to traditional clinical tools. I was a patient of a specific hospital when I lived in Boston, and it turned out when I was admitted for labor, for delivery, I had multiple records in their systems, based on different instances where I had different needs, and we’re describing primary care and the responsibility of this notion of a PCP knowing everything about me, when that can be, number one, extremely overwhelming to know for every single part of my healthcare journey, which may have very different needs. If I’m pregnant and going through a maternity journey versus if I get sick with COVID or anything else, you know, the type of information and the type of judgment that’s necessary in each of those instances is very different.
How do you sort of appropriately balance the horizontal view and the longitudinal journey of a given individual with the notion of, you know, the bundles of care, the unbundling of primary care across the different mini journeys that we all have as patients. The type of data that again, I need for journey one versus journey two can be very different.
If the cost of measuring everything is low enough such that I can collect all that information, perhaps that’s the best way to go. But how do I then, you know, sort of appropriately overlay the right semantics and the right context, as Ivor was saying, for that particular instance of care need?
Jeff: There’s a lot of times where doctors are forced to, and when time is of the essence, to make decisions based on partial information to be safe. And I think that if they had the context of a person’s entire history and what’s changed, there’s a lot of things that they might associate with an immediate symptom that are actually normal for that person.
You know, we’re all used to like tools like Shazam now, but trying to figure out what’s wrong with a person based on a single measurement, or even a set of measurements at a point in time, is a lot like trying to identify a song based on a single note in that song, right? It’s just not possible. A lot of songs share the same notes. You need to hear a sequence of notes for it to actually be, you know, be a song. And just similarly, I think you need a sequence of measurements to actually understand the story that’s going on in a person’s physiology, and that can explain where they are.
Hanne: You need to hear the whole song to know what it’s saying.
Ivor: Jeff, I love your Shazam analogy. One of the things that I think is really interesting about Shazam is that if there’s a song in there that hasn’t been played enough, you can play that song and Shazam won’t pick it up. I think that’s the same thing that’s true with data. And whether we’re collecting data in all the people that we need to be collecting data from, because if we don’t have that information, we’re not gonna be able to recognize that song.
I think we need to make sure that we’re including folks so that we can recognize that song in everyone as we’re making these transformations in health care. But I think it’s a really awesome opportunity that we run to, instead of running from.
The other piece is around, how do we, when we give people information, their ability to make those changes is also impacted around the environment, and the priorities, and the access that they have, whether it’s to the ability to exercise or healthy foods, or what their job requires for them to do, or the ability to move around in their neighborhood safely. And so I think us thinking about that in the context of how we can impact and help people on all levels once we have the data, is really important.
Jeff: Yeah, I totally agree. This information is so valuable for us just optimizing, you know, our society. that’s, I think, ultimately how we get to a healthcare system that actually gets better, or every generation is healthier than the last because we understand better how to care for each other. What we’ve started to see is that when you give people information, feedback, they can very quickly and intuitively correlate changes in their behavior to improvements in their health, or decrease risks.
But they don’t have that feedback right now.
Julie: It also begs the question of what is the primary care provider’s skill set, what does that skill set need to be in the future, right? I mean, we’re almost appending the very definition of what is a PCP; it’s no longer just about, like, interpreting the test results or, you know, doing your basic workup but really, it’s about like, how do you ask the right questions of the data.
It’s almost like the wave of data science that occurred in general engineering and computer science, where, you know, the skill set became less about, like, how do I write really good code, but more about now that we have so much data, how do you best interpret that data and write tools to that? It’s almost like you can imagine another credentialed, you know, provider type that has to exist to make all of this work.
And then what happens to the traditional physician archetype of the person who’s doing the real clinical interpretation is that does that continue to exist but in a way that only has to focus on the sort of the things that get escalated to that human who actually requires some judgment to be able to look holistically at that patient in that context with all the information, etc.? And then you have a sort of a separate, you know, class or tier of folks who are standard in a clinical practice that are these data ists essentially, that support that physician.
Jeff: If we do that, we’ve failed to build the right tools. Like, technology should not require people to get a data science degree, like doctors should. These tools should liberate a doctor to actually make just decisions.
I mean, going to the library and using the Dewey Decimal System wasn’t gonna work for the internet, how long did it take to learn to use Google? I think actually, that the clinical decision support to the future liberates a doctor just to ask questions, and the system will give answers. It will be, you know, Doctor Ivor will say, “Tell me about Jeff’s respiratory system,” and the system will just summarize that the tools might require data scientists to build, but there should not be cognitive burden on a doctor to actually use those tools, any more than I should have to have a degree in statistics to be able to search the internet.
Yes, it will absolutely optimize what we do and help us to do things better and faster and more effectively so that providers are not burnt out by the overwhelming information that they get. And, there has to be an integration for the opportunity to let that human-to-human interaction inform the information that’s in front of them.
Ivor: Our ability to gather and collect data now is phenomenal. And it’s wrought with biases that we have to recognize and understand. And those biases impacting in the decision support for a provider are significant in the outcomes for a patient. There needs to be more understanding of how to analyze data by providers. The lack of ability to understand how data can be used transformed to tell whatever story we want it to tell is becoming quite apparent to us right now, And the ability to understand how to not just look at a lab result and say, “Okay, it’s within the normal range or it’s not within the normal range,” is no longer gonna be acceptable.
Hanne: So, primary care five, 10 years down the road, does that just mean it’s all around us all the time like there is no primary care, it’s just everywhere care What does that shift look like at the farthest end of the spectrum?
Julie: Yeah, I think the couple of dimensions that change are, you know, one, the notion of resource constraint that we started with. You know, I think that will look completely different in the future when we are able to tap into the nationwide or even global network of PCPs through virtual care, through telehealth in a way that is reimbursed, in a way that, you know, the licensure sort of burdens and things of that sort of are taken off the table.
So the notion of like, I have to rely on the supply within a five-mile radius of my home, you know, such that I can get the care I need goes out the window. I think that’s one thing. And then I think the other thing is, you know, flipping the paradigm from one in which, like, we as the consumers and the patients are the ones who have the burden today of figuring out whether or not we need to get care, to one in which the system because we can be proactive about identifying signal in that data that says, “Actually, Julie, you’re the one who needs to come in now,” versus “Hanne, you’re fine, and you can stay home for the next six months.” I think that whole paradigm will flip such that we wait for the doctor to tell us what we need versus us having to put ourselves in the queue to figure out whether or not we even need to come in.
Jeff: I think that primary care doctors just, the role if anything, is are amplified. They’re the QB of your health, the quarterback, and they’re the director, like they’re calling the plays. They just have a lot more data at their disposal, and tools that help them understand what the most important parts of that data is so they can ignore noise.
Ivor: The primary care provider may be the quarterback, but what the coaches look like are very different. The coaches may be community health workers, they may be family members, they’re definitely going to be, the patient themselves, they’re gonna be the head coach. And then you’re also gonna have other resources like wearables and smartphones that are part of your defense and part of your offense, that are also playing as part of the team and recognizing that it’s a team sport.
Hanne: That’s awesome. Thanks for joining us on “a16z Podcast.” And thanks, especially to all the primary care docs being all our quarterbacks right now.
Julie Yoo is a general partner on the Bio + Health team at Andreessen Horowitz, focused on transforming how we access, pay for, and experience healthcare.
Hanne Winarsky is the Head of Writer Acquisition & Development at Substack.
The a16z Podcast discusses the most important ideas within technology with the people building it. Each episode aims to put listeners ahead of the curve, covering topics like AI, energy, genomics, space, and more.