Allon Bloch is the co-founder and CEO of K Health. Allon’s problem is this: Can you use AI to make seeing a doctor easier and more helpful?
Today, thousands of patients a month are treated through K Health. The company has an AI-based patient interface and it employs about 150 doctors. And K Health has plans to expand beyond primary care -- and they just raised another 50 million dollars to help them get there.
Pushkin.
Going to the doctor sucks, and not just because you're sick, but because it's hard to make an appointment. Then you finally get an appointment, and you go in and you fill out all these forms. Then you wait for whatever an hour. Then you go in and see the doctor and she asks you all the questions you just answered on those forms.
And by the way, it's not great for the doctor either.
She has to see a million patients and do all this paperwork and doesn't have time to keep up with what's happening in the medical literature with all the new studies and recommendations.
And everybody knows this is true.
Nobody's figured out how to fix it yet, but it is a problem that people are.
Trying to solve. I'm Jacob Goldstein and this.
Is What's Your Problem, the show where I talk to people who are trying to make technol logical progress. My guest today is a lawn block. Alan is a serial entrepreneur.
He was the co CEO of Wix, the site for building websites. He founded a company called Room that buys and sells used cars, and then in twenty sixteen Alan founded a company called k Health.
Today, thousands of patients a month are treated through k Health.
The company has this AI based patient interface and they employ around one hundred.
And fifty doctors through an affiliate, and k help has big growth plans. They just raised another fifty million.
Dollars to keep growing.
The problem Alan and k help are trying to solve is this, can you use data and technology to make going to the doctor better? Better for patients and for doctors.
We built essentially a machine that can engage with you. It's a predictive mention that is trying to figure out the next best question to ask Jacob. Jacob comes in and he's got a headache, and and and you know, if we don't know anything about you, we asked for your age and your gender, and we have a series of conversations given what we know, you know, you know, what is the next best question to us to get to the differential diagnosis?
I will say, uh, this is in fact some some part of your product as it exists now is is what you're describing. And I know this because I tried it out before this interview, and as it happens, true story, I have COVID right now. Despite my chipper demeanor, I feel like asks. And so yesterday I loved Indo K Health And the one thing I didn't tell it, I didn't I have had.
A positive COVID test. But I thought it would be too easy to tell it that I had a positive COVID test. But I said I have. I'm feeling tired? Was my ask you?
You log in and ask you first, like you know you're presenting complain, asked like why are you here today? What seems to be the trouble? And I started with tired. And it has a specific list of possible answers. Right, it's not like it's not like chat GPT, it's not free form, right, it's very it's clearly categorized, and it puts your answer into keating. And so once I said I was tired, it asked me a lot of other things that said, when did you start a fatigue? Right, it called it fatigue, and then it's how long have you been feeling fatigue? Lots of questions about fatigue any, did your fatigue start after eating, after sun exposure?
Are you more or less fatigued at the time of day?
And then it asked me other symptoms that I said, I had a fever and It asked me lots of specific questions including have you been in close contact with anyone who has COVID? Lots of other symptoms, you know, loss of smell, no malaise, yes, lots of general questions, and then it said the results I'm about to show you are not a diagnosis or medical advice, which I get that. And then and then it gave me an interesting It gave me a probabilistic answer, right, which was quite interesting to me. The sort of final screen, and I got a screen of I'll just read it to you now.
It says what your symptoms may mean, and then it says, here's how eleven thousand, eight hundred and eighty one people with cases like yours were diagnosed sixty percent COVID, nineteen forty percent upper respiratory infection. And then it said think you needed doctor, says it get better from home today, chat with a clinician now, and that's we can talk about that part later, but let's just talk about I'm really interested in that, the specific quantitative nature the eleven thousand, eight hundred and eighty one people with cases like yours, and then the probabilistic outcome. That's really interesting to me because it's sort of like what would.
Happen if you go to the doctor, but also sort of different. So tell me about that.
So first of all, that makes me ready happy and we should have started there. I'm so happy you used it. And by the way, coming in with the fatigue case is not trivial. You go to your primary care or good care and say I have fatigue. Okay, there's many many things that can lead to fatigue. It can be temporary, can be long term term. It could it could it can be related to a serious health condition, or it can be something transitory. So that's great. You came there and you acted all down. Apologies, but you didn't say I have COVID door I think I have COVID dours effected. You gave it was a broad based thing. Fatigue can also be related to cancer. It can be related to thyroid issues, insomnia. So so what what we're trying to do just to tile the pieces together. It builds a dynamic cluster of people like you, of this, of this, of these people that have the same gender, age, medical history and symptoms that are relevant to what you're having, and it's trying to figure out what it has it gives you a statistical output because it's a probabilistic nature of what you might have. And it's and as as as we say very clearly we don't diagnose you, but compare it to doctor Google or any other system out there, you won't get that rigger. And every one of the questions we ask in our clinical AI is a is a very specific question. Is in medicine, this is not a hallucinating let me just try and guess what you have. It's not chat ept, it's not trying to predict some set of sentences. It's trying to predict your diagnosis. We generate a diagnosis in treatment, right, but because you know this is not regulated, we tell people this is what you might have, and if you want to, you can press about it and be in front of a doctor twenty for seven and deal if not.
In fact, it says it is not a diagnosis. You're saying it's a diagnosis.
Trying to make Jesus to be clear, this is not a diagnosis. Is trying to mimic a diagnosis. What we're trying to generate is trying to mimic a really high quality physician by having this medical conversation, this medical chat and having showing the differential because maybe if you didn't have a positive laptist, maybe it's maybe it's an opera respiratory infection, maybe it's something else going on here. And our goal was not to say, Okay, this is what you have and it's impossible that it's something else, because that's not how medicine works. The goal is to consider the diagnosis. But that's a tricky part of medicine. There is a science and art here about how to do it, and we're trying to mimic that capability. But in it I think it's quite magical that you can engage with a machine, have an understanding what you might have and which is not some kind of guess, work on doctor Google, and then press a button and was in a matter of minutes and be in front of a doctor and get medical advice and get prescription, get anything else, anything else you need twenty four to seven.
So to build this thing, this machine that's trying to come up with, you know, something like a differential diagnosis, obviously you need a ton of data. And I know there's this key moment early in the life of the company when you got access to this like big beautiful data set. Tell me about that, Okay, So.
People pointed me all over the world to try and find data. At the time people spoke to me about Estonia, Denmark. Any conversation I try to have with a with a large hospital group in America at the time was completely stonewalled. I can get anywhere because of hip et cetera. And we ended up licensing a data set from an HMO in Israel called Maccabi. And I'll explained specifically what we have there. At the time, we didn't realize just how what do you call it? Big and beautiful? The data set is Maccabi is structured similar to Kaiser Permanenta. So it's an insurance company that also has its own captive providers, but they also have their own drugstore, their own lab, a single EMR, their own hospital.
If providers means doctors who work for the company. Just to be right, okay. So it's a complete integrated health system, right.
People who are in this system, they get their prescriptions there, they go see the doctor there, they go to the hospital there. So this Macabi, this Israel, is going to have all of the data for the patients there.
And then one more big step. People don't switch HMOs in Israel. So we essentially got access to two and a quarter million people for a data set over twenty years, and we essentially were a fly on the wall for all these people. Did they go to a doctor, did they get a prescription? We know also if they picked it up, we know, if they had labs, we know. We could put together a time series of all of that, and of course postilizations as procedures, surgery's outcomes. So just to give you a sense, we got access to about four hundred million medical charts, over a billion labs, half a billion prescriptions, two million hospitalizations.
And I presume it's anonymized.
We built an anonymizer.
So you have this very large, very robust, very complete, longitudinal data set.
Great, what do you do with it?
We wanted to focus on these things are both what you go to your doctor, your primary care physician, and to urgent care. My head hurts, my stomach hurts, I've got a rash, I've got an upper rest with our infection, I've got a fever. All these different things.
And some kind of problem that just started. It's not a chronic disease.
It's not a physical it's something that like, oh, my head just started hurting, I better go to the doctor.
It's that, yes, And now it could be my head started hurting and I've got a history of migrains a synoscientists, and I think I know what it is, or it might be something completely out of the blue.
What's one thing you tried at k Health over the course of the company that didn't work.
We tried in the beginning as we were building through as we're learning healthcare, we're trying to send people. When we built the information, we said, okay, here's the information and we'll connect you to physical doctors in person in Manhattan and Brooklyn because we're basing the arc and then work at all.
Why not? So why not?
Because people wanted immediate solutions of their problems right now, that's a scientist or UTI or migraine. And you know the whole beauty was it twenty four to seven service?
So was that when you realized that you had to hire doctors.
We needed to hire doctors and build our own solution, own the resolution.
Which is a huge that's a huge leap right there. You're going from being a software company to being like a much more complicated, much more expensive to build company at that moment.
Right, Yeah, but you we didn't go in here, timid. We went in here. We want to make real changes. We don't want to be some yet another cute digital health company that does cute things and sells usself. We want to build a big, independent company, and we want to build a company that will make a difference in care delivery in people's lives, starting was primary care. So yeah, you have to you have to pick your battles.
Yeah.
Well, so let me ask since that came up, I mean, so, I know you're just raised another fifty million dollars, which congratulations. Hard to do at this moment from what I've heard for a lot of companies, and you've raised what three hundred million or so so far? Is there some Well, do you have a sense of when you'll be profitable?
I think we'll be profitable in the second half of next year.
Great, So let's let's talk more more fully about how your business works at this point, right, I have I tried out this sort of consumer facing first step, I didn't click through to talk with a clinician, but what like, how does the business work now?
So our business is wet. We have both a direct to consumer relationships and we work with payers, and we work with large health system providers.
And payers means insurance companies. Payers is what people are in healthcare call insurance companies.
So we essentially offer our services and we get paid as a provider to take care of patients. So if it's direct to consumer, we have an offer for consumers and you can just sign up and pay out of pocket. We announced that a.
Few weeks ago that we're.
Working with Cedar Sinai. We're working with other health systems.
Theesar sign up a big hospital system in Los Angeles.
Yeah, the largest hospital system in California. We see thousands of patients every day, which would be the size of a large health system with multiple hospitals and multiple art patient centers.
So large, and how often do you end up referring people to see a doctor in the physical world, or get a test in the physical world, to do something in real life, to do something in meat space.
So we refer people to in person physician about eight percent of the time. That could be to the er, it could be to specialty care or to an person primary care. Okay, so I was going to say that, Just to be clear, there is many times a physical interaction. For example, our doctors will send a patient to do ELEB. Yeah, but ninety two percent of the time we can resolve it without a bricks and marker doctor reviewing that lab. A portion of the time we need to we need to refer the patient to an in person visit in primary care, or to specialty care or to er. But again, the vast majority of stuff we can do online. So it completely changes the paradigm on access because if it's ten pm, what are your choices?
In a minute, Alan talks about the problems that k Health is still trying to solve. It's interesting to think about to compare k help Health versus a traditional doctor. Say good, you know, seventieth percentile a good but not amazing traditional doctor.
And I'm curious. I would imagine there are some settings where KA health is better and some settings where a traditional doctor might be better. Some things that are just easier to detect in person, say, and I'm curious, like, are there particular kinds of cases that worry you? Are there particular kinds of patients who you think KA health may not be well well suited for.
I think the same things that doctors will struggle with online are things that they will struggle with in person. Patients that are not compliant with their medications or with their care protocols. Patients that have are just very fragile, they're very old, they have you know, they have pneumonia, but they also to have multiple other background diseases. Remember that when you go into a primary care doctor or clinic and you walk out you're ass it's very difficult for the doctor to interact with you and follow up with you. So there's multiple advantages actually to build a system online. People can come back very easily to our platform.
I mean, I've asked, is there somewhere where KA health is worse? And you've given me somewhere where KA health is better. But I am curious within the universe where you were treating people, are there places where for whatever then maybe there's a smell, maybe there's some you know, truly right, like there are things that can be perceived in the physical world.
It's a genuine question.
No, No, I understand the question. I think the areas that we're less strong and now is areas that we've seen for your patients. It's stuff a little bit different from what your intuition is. Your intuition was similar to mine, you know, six seven years ago. But it's stuff around or simpedic huh, back pain, extremity, medicine, you know. Uh, the trauma. We don't do trauma, you know, so so stuff around.
That trauma trauma like did I break my wrist when I fell that?
Or you've got a pain in your in your in your foot for a few days. Maybe it's planter fasciaetus if you've ever had it, it's pretty painful and appears out of the blue. But you know, maybe maybe you broke a bone or maybe you've got you know, you know.
Is the reason is the reason you don't do that, because the doctor needs to touch the patient in short to make that diagnosi.
I think we can build systems for orthopedics as well. Uh, it's just not it hasn't been an error focus about for us, and and over time, I think the whole online versus in person is going to be mute. It's kind of like going to the ATM to get out cash versus venmo. Both are utilities of of getting cash or transferring cash. You will use both interchangeably in different reasons for different settings. I think that's how medicine's gonna go to. If you can resolve your problem right now from home in a high quality, high confidence environment, and I can follow up with you wherever you are. Maybe you need a prescription, maybe not. It's typically better than going in in person. Sometimes you need to go in person for various things. Medicine is really tricky. It's really tough to I'm still at all of how doctors make decisions because they don't always have all the facts, but yet they need to act. They need to decide should I send the patient to the R do I tell them to rest at home? I would just jacb I'm jumping into something to me is ready burning and important for us. We built k to give people access to high quality medicine. I think we have uneven quality medicine. Yes, there's Mayo Clinic and Cleveland Clinic and top academic institutions that act as lost results or solving complex care Cedar sin I other institutions, But most of the time people need to manage their diabetes and hyppertension, or even better avoid getting diabetes talked to or high pretension. Many people who have insurance and have money do not manage their diabetes and hard conditions. Well, it's not one or two percent, it's double digit percent. Why Because it's tough, it's it's expensive, it's messy that you know who wants to deal with it? Can we build care delivery systems that are dramatically better, that are much more data driven and are tailored to you admin that take into account prevention. Everybody talks about prevention that are preventative systems are similar to for the most part, I'm not talking about oncology here to the seventies and the sixties. That's pretty awful.
So let's this is interesting.
Let's talk about it at a at a slightly more specific level, Like you're making a compelling kind of abstract, big picture society wide point. Tell me something specific.
About what you're doing to do a better verse of helping patients manage their diabetes or high blood pressure.
I'll give you one specific example. We took maoclinic data and we're able to look at mayoclinic patients in an anonymized manner over long periods of time, and look at high pertension patients and figure out the best medication or combination of medicines to stabilize high blood pressure given their current blood pressure, gender, age, comorbidity, and ethnicity, and be able to tell a doctor in that same statistical split here is potentially the best way to give this medicine, and then math medicine.
And high blood pressure is presumably a good place to do that, because there are lots and lots of good, cheap, old high blood pressure medicines, but which ones to use or which combination to use for which.
Patient is hard to figure out. Presumably that's why that's a good sort of target problem.
I would venture to say most doctors do not in the in their head that algorithm of gender, age, medical history, ethnicity, culpabidity, and you know, the current blood pressure. So this is really helpful to our doctors. Eight percent of the time they change the patient's regimen or change change their mind around what they want to do. And it's based on male clinical mail clinics standard of care. So that's that's really powerful. Every day that you stabilize in better, you know, to reduce the risk of strokes and heart attacks.
What's something you haven't figured out yet that you're still working on. What's what's the sort of frontier problem, a frontier problem that you're working on.
Hmm. The stuff that we're working on integrating right now is tying what we're doing to a non hallucinating large language model. So how do you take what we're doing and we we help the patient understand what they might have, We put them in front of a doctor and they get diagnosed and tree. But let's assume the patient is a diabetic patient who also wants to lose ten pounds and doesn't like fish and wants to do a certain diet, and it needs to be tailored to what we do. How do I enable the patient in a free form, free text engage with the system that's non hallucinating, that is also tied directly to what we're doing. I think that's a direction we go.
So plainly, what you have right now is very structured. I can't just come in and like say a million things about myself. I can have a presenting complaint and then your system will ask me some questions, and there's a sort of set universe of categories of conditions whatever.
And plainly, large language models like chet GPT, which is very different from your system, offer a much more wide open playing field. A wide open means of interaction between a human and machine, with the notable downside, the profound downside in this case, that they are not reliable sources of information.
Right, So what if I understand you correctly, what you want is the reliability of the system you have built.
With the freedom of a large language model like chat chip.
Yeah. So fundamentally, nobody wants to go to a doctor that hallucinates ten percent of the time and you don't know when the doctor's almost think that's kind of meaningless. It's going back to the same doctor Google WebMD problem. We care about the clinical context and that is really important. I think that's where I's going, and I think LLLM could be a huge ally here. But fundamentally, when you're asking people specific questions, when you're asking questions that are medical and clinical by nature, there's no guesswork here. It might need to be modified and change over time, but you're asking very specific questions. So for me, a big part of what we're answer is integrating that.
We'll be back in a minute with the lightning round. You were the co founder and CEO of Room, a used car retail.
Retail, and oh yeah, I.
Definitely have a couple of used car questions for you. One, what's the most surprising thing you learned about the used car business?
How predictable it is? Huh, how predictable car pricing at a certain period. But I was lucky to work there pre COVID, where you know, the supply and demand and completely changed. That was a big surprise to me. It's much more data driven and predictive than people realized.
Okay, what's one tip for somebody buying or selling a used car?
So most people are buying a car also trading in. They're selling a car, okay, And about eighty percent of the time people are taking a loan okay. And most people come in and they do reams of research and coming with their armed guard to a car dealer, and they focus just on the buy. In the meantime, they get a shitty price on the trade and they get a and they get a you know, a bad financing deal on their car, and that tax on you know many South of daughters, so the cost, but they got a great deal on the bi in their minds.
I read that you bought a house on Ridyon Avenue, which was interesting to me because I used to.
Live on Ridian Avenue in South Beach. Uh, And I couldn't think, like, oh, what's that? We both have lived on Bridan Avenue. Question, I don't know. Do you go to that Publis? How's South Beach strading you?
I'm here right now. I spend most of the time, okay, most of the time in New York, but I'm here. I'm here right now.
It's a very hot time to be on Ridian Avenue.
From my experienced, I haven't here this time of the year. I have a pool, and I'm okay, and I'm close to the beach, and I love Miami and I love Miami Beach. So it's a lot of fun. It's never at an moment here.
Well, so you grew up in Israel, you lived in New York.
Right, New York little known fact is on the beach, and you have a house now in Miami Beach. You live a lot of the time in Miami Beach. Of all those places, where are the beach snacks? The best best beach snacks?
Oh, Israel, by far, Israel is much more street food and snacks.
Yeah, so, I mean, what is falafela beach snack? I don't think of falafel as a beach snack. What's the beach snack or is it falafel?
Yeah, well you can have falafel and almost anywhere in Israel, so you could you can also get it on the beach. But people people in Israel a eating all the time everywhere, so you know, they also eat on the beach, and it's very family oriented and the beaches are crowded and people are happy.
Alon Block is the co founder and CEO of k Health. Today's show was produced by Gabriel Hunter Chang and Edith Russlo. It was edited by Sarah Nicks and engineered by Amanda k Wah. You can email us at problem at Pushkin dot fm. You can find me on Twitter at Jacob Goldstein. I'm Jacob Goldstein and we'll be back next week with another episode of What's Your Pop?
Don't Myself