Join the conversation
.png)
Get ahead of the curve and learn from leading care providers and companies such as Cityblock Health, One Medical, Oak Street Health and more
Awell Co-Founder and CEO Thomas Vande Casteele hosted Dr. Darshak Sanghavi (Chief Medical Officer, Babylon) to talk about ways to shift from a 1:1 to a 1:many care model. Watch the full conversation here.
Our conversation was filled with a ton of golden nuggets, but here are our favourites:
If you are already familiar with the terms CareOps and care flow, skip this section and feel free to go straight to the key takeaways.
CareOps is a set of practices and tools to build, operate and improve software-enabled care flows. It applies principles from agile software development, quality improvement and design thinking to healthcare processes.
CareOps brings people from clinical, operations, product, data and compliance teams together, overseen by strong clinical leadership. Ultimately, CareOps increases a provider organization’s ability to deliver higher quality care at lower cost and drive improvement cycles more frequently than its peers.
For more context on CareOps, read What is CareOps and why do we need it?
Clinical operations or clinops is a function that helps making sure care is getting delivered.
CareOps is a cross-functional practice to design, implement and optimize software-enabled care flows that power clinical operations.
The central concept of CareOps is the “care flow”. At different care providers different terms are in use for this term, such as care program, care pathway, care plan, patient flow, patient journey, care journey, clinical protocol, care process, (clinical) service line, care process model, clinical workflow or even digital therapeutic.
In essence they’re all sequences of activities completed by a care team and/or patient to maintain or achieve a desired health status for that patient.
Care flows are often defined at the medical condition / population level: a care flow for “Anxiety”, “Type 2 diabetes”, “Total joint replacement with obesity for 65+”, “Menopause”, “Sexual health for LGBTQ+”, “Discharge after surgery”, etc. and can be patient facing only, care team facing only or include activities for care team and patient.
We’ve used “care flow“ as an umbrella term in the panel discussion.
Thomas Vande Casteele: The title of today's hangout is evolving from a 1:1 to 1:many care model. We'd like to hear from you what that means for you, and for Babylon?
Dr. Darshak: Traditionally clinical care was always delivered face to face with one individual, clinician, physician, nurse practitioner, advanced practice clinician who has a specialized body of knowledge working with that patient. In many ways, that was really how we thought about healthcare for many, many years. And while that's a wonderful approach it's also is problematic in many ways. Over the years, we've become increasingly cognizant of their challenges around diffusion of quality care broadly, and also the lack of availability of that type of one-to-one intensive resource. So, while we think of the one to many is something which is really dependent on technology that really started 30 or 40 years ago, when we think about much more broadly, just socializing the way knowledge is dispersed among the clinical community. Now, the challenge over the past few years, and what I have spent a lot of time on here is can technology actually make that better in some way? And can we actually then take advantage of that?
I think there's a couple of ways we think about it at Babylon, and you said to be concrete, I will mention one. The first is that the one-to-one is predicated on having attention from somebody for a period of time. So a one-to-many doesn't mean that one clinician is taking care of 50 patients all at once, while they're all on different screens. But rather, it's surfacing so that when that attention, which is the most precious resource we have, is the attention of somebody, it's actually fully focused and used in the most efficient way possible. So, what that means for us, for example, is digital triage for clinicians when people start off. So, can we collect information and baseline pre visit flows ahead of time? We spent a lot of time on that. That information then will go to the clinician. In addition to that, digital triage so that when there's simple questions, for example, medication refills, laboratory review, and so forth that can be done through self-serve. What I'm getting at is that one too many for me means that they're cued, the patients are organized, the information is there in a seamless way possible, so that when the clinician sees the patient, they have everything they need, and can actually work more efficiently. Maybe I'll start with that. Those are the process metrics we typically also track very carefully at Babylon grid.
Thomas Vande Casteele: Most people understand that transitioning from one-to-one to one-to-many is not as easy as developing an app and putting that app in the hands of the patients. From your experience at Babylon, what are the things that made the transition hard, but which might not be obvious for the audience, who are already convinced that the one-to-many might not be seeing those pitfalls?
Dr. Darshak: Let's say we have a centralized repository of, say 30 clinicians that are available to see patients, and that's the group that's going to take care of a much broader population. Now, in the US, those clinicians are licensed only at the state level. If I want to see or take care of a patient who lives literally a few hours’ drive away in the state of Connecticut, I cannot do that legally. That is a major problem where in fact, in the US, you actually need a license in every single state in order to practice nationally. So, those types of barriers that are more structural or policy related. That's just one but there are many structural barriers like that. In particular, insurance types, where pharmacies are located. So, there's an enormous amount of technical thinking that's we don't run that architecture. That's bucket one.
Bucket two would be things that are scheduling related. Simply put, just queuing and actually creating a national network, and one where we actually can-do scheduling across time zones. We have 30 clinicians. How do they know what hours they're working? What if one of them gets sick? How do you reschedule? Enormous amount of technical knowledge and architecture has to go into that.
And then, the final piece of it in terms of the one-to-many is that when people are triaged, even in that situation, we can't actually always predict how complex their needs may be at that time. In other words, somebody may need a five-minute visit or a 10-minute visit, suppose they just have a simple problem that can be addressed. And then, others when you start talking to me, you realize, wow, you see them on video, they've got a bunch of kids in the background, they have a scar on their face, you realize they might have interpersonal violence and other issues, the challenges of sort of addressing and metering that one-to-many, is often just unpredictable. And obviously, then as a result, our schedules get messed up, people aren't happy, they may not get the care.
Thomas Vande Casteele: At Babylon, at what points in the care journey, do you do triage? Is it kind of like a continuous triage that helps to make these priority calls? Or where does that exist? How is that working?
Dr. Darshak: It's done in a couple of ways. For those of you familiar with Babylon, when you log on to the app, there's a number of options. You can have digital advice given to our triage platform, or you can make an appointment or send an asynchronous text message. So, those are the ways in which communication is initiated. We also have outbound calling better triggered as well. For the digital triage, you can say I have a headache and then we'll go through the interview with you and then make a recommendation on the level of care that you need. In some situations, it could be self-care, and then that ends at that point. A significant minority of people just after self-care will then just take that advice. On the other hand, it could say, well, you have a headache, it might be due to sinus infection or something like that, would you like to make a non-urgent appointment? So, then it allows you to then make an appointment and then in some situations say, your headache sounds like you might be having an intracranial hemorrhage or something like that, or meningitis. Don't even talk to us. We recommend you call for an emergency. That's the first.
The second is then people can actually just text in or messaging what their particular needs are. In that situation, then that's looked at by a care manager on the other end, and a prioritization decision is made, and then they'll prioritize, do they need an appointment? Can this be handled asynchronously? Those are the kind of triage.
Thomas Vande Casteele: So a significant minority of people will follow up on the advice given after self-care flow, let's say. Can you explain that a bit more? Why would that be? Because it seems to me that in a one-to-many model, increasing that percentage of people who will indeed follow-up on that advice is actually a great way forward, right? It’s the key solution?
Dr. Darshak: I apologize, I don't think I was being very clear. In those situations, when the triage, the AI, it's a Bayesian belief network is sort of the approach we use for that, it will sometimes triage them to say, “We think you have a low acuity issue.” So, what I meant is, let's say, 100 people go through that flow, a significant minority, maybe 10-20- 30% of them will be directed towards self-care, so they can avoid an appointment and just self-serve – that's what I was referring to – as opposed to others would say, we're not sure let's make an appointment to see you. The idea is to reduce that so rather than having 100 people have an appointment, we have 50 people making an appointment if they go through.
Rik Renard: Which of the experiments you did to transition from a 1:1 to a 1:many care mode failed?
Dr. Darshak: One of our thoughts was that people when they had the self-monitor number of their own conditions because we urge wellness in addition to acute care. So, we have a product in our app called My Health, which is essentially like Apple Health. It allows people to monitor on their own – weight, blood pressure, glucose, and a number of others. We have a few dozen items they can monitor. We believe that this would be a very popular feature, particularly among individuals who hadn't had access to digital health before. In it, we started that market in the US and then went to the UK.
What we found is that, while we spent actually a fair amount of resources building out a beautiful app experience where people can track that, the actual uptake was actually quite low for that particular set of features. And what we found is that people in general don't just want to track their weight or track their pulse or their blood pressure, but rather, they need the connection with the clinician first to understand the trajectory of their health, and only then sort of work with them with remote monitoring solutions of some type.
It turns out the number of people that are really kind of into tracking all their metrics, and I think it's actually a really, really small number. I didn't mention this before, we take full risk on about 300,000 members, so we can't just appeal to people in a consumer directed way but we have to create a product when patients are assigned to us that a large majority of them will actually use. That's been a big challenge for us.
Rik Renard: How do you currently build engagement in a digital platform that is built for scaling and patients are not always willing to, for example, fill in these 10-minute free consultation assessments?
Dr. Darshak: So, as many people are aware, Babylon began in the UK as digitally enabled primary care in the NHS. It over the years built up a whole approach to delivering Comprehensive Primary Care. Now, the thing in the UK is that the NHS just gives a primary care capitated fee per year. In other words, for every member that chooses Babylon, we were getting paid a fixed amount of money to handle all their primary care for a year. The thing that many Americans are shocked by is the amount of money that it is. It is about roughly give or take 150 US dollars per person per year. Now, what that means is that if we create value and see patients and reduce downstream hospitalizations, all of that, we don't get the benefit of that. In fact, there's almost, in the NHS, a disincentive to do a lot of care. That's why waits are so long because again, the benefit doesn't accrue to the actual provider. So, there's no sense of value-based care. It's one of the reasons Babylon moved the model to the United States.
The first market in the US, we took a full risk contract in the state of Missouri. 20,000, roughly, patients were attributed to Babylon, in Medicaid. For some of your listeners, Medicaid in the US is typically health insurance for low-income individuals. And also, some states are much more restrictive in who they allow to have it compared to others, Missouri is one of the more restrictive ones. We inherited 20,000 patients, a lot of them children and young women. That was the age and then they just gave us a list. And you can imagine that organization, who are they going to give us their risk on. These are the patients that are the most challenging to reach. They don't have phone numbers. They don't have addresses. Yet we are financially responsible.
Sometimes when you first enter the market, you have to take on a challenging population. In a very real way, the challenge we're trying to surmount is how do we get 20,000 people that have not engaged in care, sign-up with a company they've never heard of with a name like Babylon, and actually come to us for their primary care. That was where the problem started. We actually spent an enormous amount of time and effort. We had people we hired to knock on doors, we advertised on billboards, we tried to hire companies to find these patients. We've recorded this publicly; it took us about two years or so to get about a third or so of the patients to download the app and register.
That was our learning curve. Then when we went to our new markets, we actually were able to make that half that amount of time because we got much smarter. But then we realize that engagement is really challenging in places like Medicare, Medicaid, because there's no way to drive members to use a digital first platform.
So, just this month, we have now launched a commercial plan on the exchanges. That is the best analogy for the NHS in the sense that people have to come to us for their digital first care. And that's been incredible. Like, they're engaged, they're motivated, and we're finding that even more rapidly, we're getting their engagement, we just launched on January 1st, so, it's been quite a journey, engagement continues to be a real challenge for us, not only for us, but for any digital first organizations trying to do value-based care.
Thomas Vande Casteele: Could you tell us during those two years where you activated, part of those 20,000 initial patients, what is a learning that could be relevant for others who are dealing with the same issues?
Dr. Darshak: In retrospect it is pretty basic, but it takes us a while to understand this. The simple thing is, you have to give people something they want rather than outreach with something we want. So, a simple example is let's say some of our contracts, there's some risk that's based on our quality scores and quality outcomes, like Hedis metrics in the US, Quaff metrics in the UK and so forth. And so, one can do an outreach broadly on a digital platform and say, hey, you should have your mammogram or your Pap smear or please get your to COVID vaccination. We found that that did not result in a lot of uptakes.
Again, that's not surprising, right? These are individuals often, I mentioned the Medicaid population, they have a lot of other things to worry about, everything from transportation to income in the home, how are you getting your kids to school, that is not front and center.
On the other hand, what we learned after a few months was we did an outbound campaign on getting people mental health care. Actually, fully half of our US visits currently are behavioral or mental health. We have a collaborative care model and so forth. We found that it had enormous traction, because many, many, many people have wanted behavioral health care assistance, both for themselves and for their families. And so, again, that was one learning where we need to simply say, well, what is it that you want? What can we help you with? And then they'll reply. I mentioned behavioral health. The other really powerful engagement tool we had was whenever people called in, then I mentioned I worked on the Accountable Health Communities, it's really for social needs, and then saying, “Look, thank you for letting us know, we're going to help you with this one particular need you've outlined.” That also is a really powerful engagement tool. Probably even more important than, for example, hey, you can monitor your blood pressure on our app in a cool way.
The last piece, I'll say is that is the learning I talked about Medicaid, we're really interesting things, we have Medicare populations as well, what motivates them is very different. So, what we need to think about is a hyperlocal strategy, maybe not hyperlocal, but at least state based or line of business based, and we're finding that the things people want vary enormously, we just need to then understand and meet them where they're at.
Thomas Vande Casteele: So, there’s multiple geographies, there's Medicare, Medicaid, how on earth are you identifying the individual needs of these individual patients? Because that seems to be really one of the core keys to getting that engagement. And everybody's after that engagement, so how are you tackling that at Babylon to get to that individual need?
Dr. Darshak: At Babylon, our core model is to have some way of collecting data on our patients, analyze that data, and then reach back or offer insights to them that encourages their engagement. So, everything depends on "Can we collect the data at the right place, right time, and analyze it in a way to surface those needs?". That model broadly, to me is reproducible in the sense that, regardless of where you live, if we create the right models and risk stratification and insights, it doesn't matter whether you're a Medicare patient in New Mexico, or a Medicaid patient in Iowa, or a commercial patient in Georgia, that sort of internal brain should be applicable around that.
Now, that's the theory. The challenge is that getting that data proves to be enormously complicated, particularly in the US. So, even just getting claims data in some of our contracts has been incredibly difficult. Again, not a surprise to a lot of people who know US healthcare. When we don't have claims data, how do we stratify them? So, I think, in answer to your question, what we do is we have to then be really scrappy. In Iowa, for example, we have almost 80,000 patients that have been attributed to us in Medicaid. It’s an expansion state. It's a wide variety of needs. We've actually had to do on the ground events, where we send our community teams in and talk to people and say, what are the kinds of things that are going on here? Because, again, in an ideal world, we'd have all the data we hear from them.
That's where we might hear, for example, last spring, as people are aware, there was a big baby formula shortage in the United States, that that was something that's not going to show up in claims or anything. But when you're on the ground doing that outbound, you realize many, many people are having a lot of trouble getting hold of baby formula. So, we did a big outbound campaign to engage people to say, well, we can help you get baby formula as you probably some of you might be aware of that. In that, for example, in some of our populations in Missouri, where we did that, we didn't have a lot of data, but that's where we found that one of the biggest needs was going to be prenatal and postnatal care. Many people are pretty far removed, only about 30% of patients, after even delivering a baby in the hospital, would come back for any care at all, incredible. But they wanted that there's that need. So, that's sort of how we had to do it actually, the truth is fairly manually in each of these areas.
Thomas Vande Casteele: It seems to me that what you're saying is that in a value-based care world, you've built this triage system, you're investing in analytics to better understand the needs of your patients, and to introduce needs. But there's no reimbursement code for that, right?
Dr. Darshak: That is exactly the approach for why we chose value-based care because there's a couple of ways one can go about building comprehensive digital primary care.
Let's sort of look at the alternate universe Babylon would have to operate in if we didn't have value-based care contracts. Many people have realized, the patient acquisition cost marketing, all of that work, if you could break into physician networks, get payers to reimburse that is such an enormous tax on an organization in order to do all that rather than developing the core models. So, it’s a consumer first approach, really, and then it also focuses even more on what I talked about, which is what it is that patients want, it is very rare for patients that are doing Consumer Direct program, to say, you know what? I want to manage my diabetes better, or my hypertension better, or my cancer prevention. Almost always consumer focused products, where it's fee for service are things that are going to be urgent care oriented, sexual performance, like all the companies, you see exploding, why? Because that's what people want.
Value-based care, one of the reasons I joined the company, is that the onus for that broader sort of work on health falls on us, and it liberates us from billing for every single little code that we're being charged here. Because coding, rev cycle management, all that just imagine the amount of resources you have to develop in order to actually do all that. So, it liberated us from that to take our limited resources and deploy them much more on putting out the care models. Now, I'll say the price we pay is that engagement status knowledge. Now, because of the app, we have 20,000 patients. How many of them are really coming to us when they have a question? How do we build that trust? That's the problem we choose to focus on, rather than how we acquire customers for future service? I hope that answers the question a little bit. We're really committed to a value-based care model broadly.
Rik Renard: The problem that we currently have with all the data is that it creates a big influx of a ton of new data and then this makes a problem for the data that patients need to look at a much bigger problem. My question to you is, how do you take all this data from, for example, the triaging questionnaires and how do you make it actionable, so your care teams can actually use the data? Is it with huddles? Is it with technology? How do you tackle this at Babylon?
Dr. Darshak: When I was at Optum labs, we had claims on 190 million individuals for about 20 years. And we had about 95 million full-text EHR records as well. I say that because this notion of data as a resource, I think increasingly in the world we're in, data itself is a commodity. You can buy it and get a hold of it. Having data is not what's going to distinguish us. In other words, what I'm arguing is we want to move from just knowledge to wisdom. The resource which is the hardest one to find is people who ask the very simple questions of the data to drive those insights. And this is, I feel very strongly, we talk a lot about AI and machine learning and random forest models and all that, and sometimes you literally just need to count things. I'm gonna emphasize that it doesn't take an enormous amount of computing power and very sophisticated resources to start to draw those insights out. So, I'll just start with that. And then, I would challenge both our own organizations and others to say, let's move away from the shiny objects and just talk about AI to say, what is the actual problem you're trying to solve?
So, having said that, I'll give a couple of examples of the really simple data we use. One of the things I mentioned, we found that mental health is an enormous contributor. And we find that when you address mental health, the total cost of care is also improved. Very simply, what we put in as part of our enrollment process in the initial engagement, we do a PHQ-9 and Gad 7, or PHQ-2, actually – even just two questions now, and Gad 7. Again, just simply gathering that data, that gives us a target and go get a list of individuals who have significant needs, and we push out resources to those individuals. That's what our caretakers and care managers will use as a go get, for example.
The second one, again, very, very simple, I want to emphasize this is we'd simply also just look at our ADT feed, like, who's been admitted to the hospital. Those are going to immediately have a 20% readmission risk, but most importantly, gives us a sense of what the community resources are like, and those individuals are highly motivated. The amount of work to get that ADT feed from where it is, to us. It's an enormous technical lift. We spent a lot of time on that as well. That also goes to our care teams of localized care teams, and then they'll go through that list.
And then finally, what we are trying to do now is, now once we have that outcome data, we're using that to then build our models to say, can we predict who's going to be highest cost and highest spent? We actually have a bunch of models that have been in development right now but then we're thinking about implementing them and most importantly, can we find out who's at high risk? But then what's the actionable insight that can come out of it? Our applied science teams working on that? Yeah, it's quite a process!
Thomas Vande Casteele: In a one-to-many model where you have self-care, where people are doing their own triage, when you're engaging all these people in different modalities and channels, how do you define what is the best next step for every patient?
Dr. Darshak: The sort of reproducible framework I've used, not only in Babylon, but also at United, and even when I was in the federal government and before, the first piece is having a broad-based sort of clinical understanding of the drivers of health in your population. Some of you are familiar with IHME. They do that internationally and nationally, the global burden of disease and national burden of disease. We take that approach and we localize that for each of our markets.
We develop reports in Missouri and everywhere else, and we simply just look at where the costs are going, and where the needs of the patients, based on claims and other data. What we find is, as you can imagine, significant local disparity. What we found in Iowa, in Missouri, the number one diagnosis driving a lot of our cost was oppositional defiant disorder, which is basically children who have a lot of challenges with self-regulation. That would not have come out in any other way. I'll say, then in another state, we found it was neonatal intensive care, and then the first few months after discharge from the hospital.
So, we just start with that framework, which is where are the problems? And that's not something that will come out any other way. We take the next step, and what I'm emphasizing is there's no real AI so to speak, or data here other than in the analytic work, then what's needed as a clinical insight.
In other words, for children with oppositional defiant disorder, or another big driver was asthma, what is the thing that we should be doing that we are not doing? I think this sort of gets to your CareOps approach. You need that insight, right?
For oppositional defiant disorder, it could be, like, parents need some education, or they need some tools to approach or they need an eight-week parenting course on how to handle a really difficult child who has really complex needs. So, we build those journeys in response to that.
Another really good example, clinically, we know that almost all asthma admissions in a well-managed population should essentially be zero. The major issue is a lack of understanding of controller medications, and appropriate use of controllers in action plans. That insight comes clinically but then we developed a journey, and a set of clinical protocols that are both self-directed on the app, as well as administered by our care team to do that one-to-many for asthma. But that's our approach, is it each market, I'll just summarize by saying we try to understand the major drivers of costs. We then direct our digital care teams and our digital tools to then develop a service that then addresses that specific need.
And then, we think, how do we increasingly digitize it? We probably start off pretty labor intensive, so we try to learn iteratively, and we just kind of throw people at it, then we learn well, oh, this could be automated, we can do this by Brave, we can do this by app alerts, or we can do this through automated reminders, we can do this with care assistants, and that's the virtuous cycle we try to push on.
TV: How often do you typically iterate on that process at Babylon? Let's say that my V1 is labor intensive. It’s live today. First patients today. How many times will you have iterated on that by the end of the year?
Dr. Darshak: About a year ago we started a program in which we identified 12 specific conditions that we wanted to focus on informed in part by the epidemiologic and sort of claims analysis work we did. It was things that most people can relate to – cardiovascular, pulmonary, endocrine, and we identified the disease areas.
For example: for type two diabetes, we made a list of the 30 things that we should be doing for somebody who is at risk of type two diabetes who has a referral to diabetes prevention program, initial screening, hemoglobin A1c, nutritional counseling. We made that whole list of all of that in an ideal setting, what would that care model look like? And what we did then was we said, okay, what are the two things we should be doing now? And so, the first one we just started was, we just need to identify an average population, and we said, if they have prediabetes, we'll refer them to a diabetes prevention program.
In other words, we didn't do the 50 things on that list, we just did two things. But we did that across all 10 condition areas. So, the learning, we typically will then iterate on a every two-month basis or so. What we found, for example, was, rather than building an entire complete program in diabetes, we did this widely. We found, for example, that in COPD, there really wasn't that much sort of uptake. And even our population wasn't really that engaged. And so, building out a whole program wouldn't make a lot of sense.
On the other hand, we did get some uptake in our, as I said, mental health, and then particularly in pregnancy care, transitional care management, a few others, because enrollments were coming in. We then would add out and build things and pick them off that list and build that out. That's sort of our approach. We're trying to build where our members are telling us they have their needs, if that makes sense.
Rik Renard: As we all know, making sure clinicians follow standardized protocols is painful. It's a common argument that they use is that they are creative thinking individuals and they're not robots and they don't want to follow standardized protocols. Have you found success in getting standardization among these providers on the ground and if so, any learnings that you can share with the audience on how to get adoption when it comes to standardization?
Dr. Darshak: I think my sense is it's not the clinicians that don't like being told what to do, it's that the direction they get is super confusing, and really just not clear and the tools aren't there. I'll give an example of smoking cessation. With that, there are probably eight or nine different pharmaceutical agents on the market for nicotine replacement and craving for reappropriate and others. The challenge is that there's no clarity on what should be your first line therapy on that list of eight. We don't tell people. The guidelines just say, “One of these.” The same is true in type two diabetes. You look at the antihypertensive, there's Metformin. But then second choice, it's unclear in the American Diabetes Association and others. It's very, very challenging.
In our view, what we do is we believe that clinicians want to do the right thing, and they want to practice evidence-based care. The problem is we haven't developed the simplified tools to help them do that.
What we've done is, I mean, simply put our back end EHR that we use Babylon, we've done a lot of technical, we use Athena. So, what we did simply put, after we had that meeting, we then are going to take the templates, and then we just create a simple Athena protocol. For this, we've agreed as an organization, this will be our first line therapy for nicotine replacement, for example. We again simplify that we have a whole knowledge base so that when people have to refer to it, they know where it is. I'm much more optimistic. I don't think clinicians are that independent and feel like they ought to be super different. But in these areas, we can have some consensus.
Thomas Vande Casteele: Are you then able to close the feedback loop and to say, okay, for this cohort of patients, that those were the patients where that first line therapy was adapted, how can we now compare that to another cohort to really start informing that continuous loop of getting back to your processes and updates?
Dr. Darshak: It's a great question. I would say that, in some areas, honestly, we have to just be satisfied with the process measures because the outcomes, the numbers may not be enough. Let's take statin therapy, right? There's many, many different statins for cholesterol control. Obviously, we're on them because we want to prevent stroke, and heart attack. But for every 100 people we treat perfectly with statins, the number needed to treat is about 50. So, we'd have to treat 1000s of people to detect a slight difference in heart attack and stroke rates. Instead, this is the whole concept between having a process and other measures is that the outcomes are, and you don't want to get a bad signal, so that's sort of how our approach is read. Yes, ideally, we would look at how many people stopped smoking, or how many people didn't get lung cancer. Instead, we have to satisfy ourselves by saying, okay, of the 100 people who we at least have a code for smoking cessation, how many of them do we get a medication on? And then, if we're lucky, we'll get a warm connection to see, do they even fill the prescription? That's sort of where we're at this point, to be honest.
