Digital Alzheimer's Detection

Yeah, totally. I've been in and around this Alzheimer's early-detection research space for a while. The first project I did was back in grade seven — I'm 22 now, so that puts it about 10 years ago.

When I was a kid, my grandma had Alzheimer's. She was living out in rural India at the time, and slowly, over the phone, she started repeating herself and those types of things. So we brought her over to the States to try and seek treatment, see some doctors. And one of the main roadblocks that everyone cited for her was just that the condition was diagnosed super late. So I wanted to take a stab at solving that problem. Obviously, as a kid, nothing really fortuitous came of it for a while.

I spent a couple of years tackling it myself, and then a couple of years doing research at a local university through high school — really trying to understand: if I could look at my grandma and tell that she has Alzheimer's or a form of dementia, maybe we could train a computer to do that identification much earlier. Some interesting results and publications came out of it.

Then I went to college — went to Yale, studied computer science and economics — and had this shift where I saw some of those shiny things, the jobs in finance that people get jumped into, and ended up jumping over to go work in that space. What I found was that the machine-learning skills I'd developed building these predictive algorithms for Alzheimer's were super lucrative. Some of these hedge fund and bank folks were willing to pay me a lot of money to work on those problems for them. So I did that for a couple of years. It was exciting, it was fun, but I always felt like something was missing.

So as I was coming up to graduate, I started thinking about whether I really wanted to take this corporate job, or whether there was something I could do that would be more fulfilling and more impactful. I circled back to the research I'd done back in the day and looked at the state of the art around Alzheimer's detection — and I was pretty shocked to find out no one had actually effectively commercialized the tool. So I decided: why not take a stab at solving that problem? We've been building Craniometrix — a commercial set of Alzheimer's diagnostics and treatment tools — for about a year now.

Yeah, absolutely. Back in the day I'd done some pretty primitive research. The idea was: you have people who do get diagnosed early with Alzheimer's or other dementias — it doesn't happen too often, but they do. You get a PET scan, an MRI, one of these deep, invasive, super-expensive procedures. It doesn't work for most people. So what I'd done was take folks who had gone through those procedures, alongside folks who hadn't been diagnosed early, and try to understand: at face value you can't really tell who has impairment, but maybe if we put them through a series of computerized tests, we can measure cognitive processing and find something interesting. Back then we were using fairly simple patient statistics.

So the first thing I did when I decided to start Craniometrix was go back, take the data we'd collected, and build better models — bring in the machine-learning and deeper predictive skills I'd picked up over the past few years and ask: can we build more accurate predictions off the pool of data we already had? And we could. It's possible to approach accuracy rates of 90%, 95%, compared to what doctors are getting today after pen-and-paper tests — roughly 30% or so. That was the first proof point: we can do this, and there's an interesting opportunity here.

The next step was to think about what that looks like from a business perspective. How do you build this into a product people want? You could put a tool inside a doctor's office, or into your annual physical. But what we found was that all of those bases were covered — there are a lot of players spending time on building accurate diagnostics and bundling them into a physician's existing workflow. What was lacking was the question of how you actually get people to use these things. The average utilization for tools that are supposed to be used in an annual checkup is less than 30%. Doctors are overworked; they don't have time to give people an extra 10-minute screen. So we found that building a direct-to-consumer product was the right way to go.

At that point I was trying to decide whether this was a viable career to jump into, or whether I should take the job offer I'd lined up. I spent a couple of months pitching and refining the idea and trying to raise money. Nothing worked for a few months — we got a couple of offers here and there, but nothing super attractive. And so by the time July came around, when that job offer was supposed to start, I ended up just saying "screw it" and jumping all in anyway. Which, to your point about pivotal moments, was probably the biggest jump I've ever made — and it ended up paying off. We got into Y Combinator that October. I think what folks really liked was that idea of: how do you actually get people to use tools like this, and why don't other platforms exist in this D2C space? I raised a pre-seed round around that YC acceptance, went through YC that winter — January through March — and just closed the seed a couple of months ago.

Totally. It comes down to a couple of things. One is that we really want to be the earliest first check for these conditions. We're not looking to test people above 70. The target for an early screen is: can you test people anywhere from 45 to their early 60s? That's a very, very early step. And that's an age cohort that does have a smartphone, does use a lot of these services.

That's also where you can have the most impact — because if you're 75 and you don't already show symptoms of cognitive impairment, then even if you have early-stage impairment in your brain, the impact is going to be very small. These things take a long time to play out; you're looking at catching them anywhere from 15 to 25 years before you see visible changes. So that early stage is just where we look to meet people. My parents are 60 and are well-versed on their phones. We did some work with digitizing pens and so on to make it easier for an older cohort, but what we settled on is: we focus on this younger cohort. Most of them are already comfortable with technology.

And two, it's kind of the right call anyway. We're building a product we want people to use for the next 40, 50, 100 years — a frontline detection people keep using in perpetuity. So maybe you have a 10-year lag where 65-year-olds today aren't well-versed and won't take the test even though they should. But five years from now, that's no longer a problem.

Yeah, totally. The testing suite is pretty simple. One of my favorite examples is a digitized version of the clock drawing test. You're probably familiar with it, but for everyone else: you draw a clock on a piece of paper, and a doctor looks at it and asks — did Nikhil draw all the numbers, are the hands pointing to the right numbers — and comes to some conclusion about whether I have cognitive impairment. As it turns out, it's really difficult to do that accurately at an early stage. Accuracy rates for catching cognitive impairment from a clock drawing test before visible symptoms appear are sub-30%, which makes sense — if you couldn't notice it in a conversation with me, how are you going to notice it based on this clock?

So what we do, on the clock drawing test and every other assessment we've built, is focus not on the final product but on what we can learn from the way you completed the task. If you did draw all the numbers — what's different about how you drew the 12, three, six and nine versus the other numbers you added later? What can we learn from how many times you picked up your pen while drawing? Those features end up being really predictive.

And what we're building towards is people taking these tests on a recurrent basis — quarterly or biannually — so we can understand how you change. It's helpful to know how you compare to your demographic group when you take the test once, but it's far more impactful to know how you're changing, and how your rate of change differs from other people's. We start with that ongoing diagnostic, and then our goal is to lever it into actually providing care along the continuum.

Absolutely. And that idea is starting to permeate a lot of medicine. We've done testing on how we might use our tools to catch conditions like PTSD, and research on how you could use this on football fields to identify when someone is safe to go back out or is at risk for a concussion. There are a lot of interesting ways to do this next level of abstraction — understanding how people use technology. What's really interesting is how you turn these into passive listening devices down the line. If we can predict based on how you're typing into a word processor, then realistically we can also predict based on how you're typing text messages. It opens up a whole host of ethical complications, but it's an interesting idea.

Absolutely. Something we were toying with early on. There are a host of reasons we're focusing on Alzheimer's — one of the biggest being my personal connection to it and the prevalence in my family. But early on we experimented: what if we applied the same idea to other predictives? If I took a five-second video of myself in the mirror saying something every day, it becomes really interesting — you can predict colds a couple of days before they happen. There's a lot we should be doing more of.

It's a tough question to answer, because companies like Apple are well positioned to really understand how you passively listen and what the best ways to do it are. But based on what we've done — it's things like your keyboard on the phone. And once you're playing a game, you can pick up a lot of the same signals we get off the clock: how quickly are you responding to new stimuli, how are you moving the blocks, and so on.

We call it gamification, though the way we've built these clocks, it's not really a game — no one's doing it for fun. We started with the idea that the best thing to do is take the existing medical standard, build the same thing but make it work a lot better, and build adoption in a direct-to-consumer capacity so people understand why these things work. That gives you a foothold that lets you slowly pivot to passive listening over time. We're probably never going to get to the point where it's entirely passive — you install it and let it run in the background. But we can get to the point where you put this behind your Wordle, or behind your New York Times crossword. That's super feasible and easy to set up, and we get a lot of rich information from it. So hopefully, as we keep building, we can shift consumers from these active clock-drawing things to where it's passive — just plug it behind the game, and you do what you were going to do anyway, and we learn something useful about your cognitive capacity from it.

It's a tough question. We're in the early stages of treating conditions like this. For a long time, lifestyle changes — diet, exercise — have been shown to have some positive effect on slowing the progression. To your point, for a long time, maybe the next 50 years, we're not going to be able to reverse these conditions, to reverse damage to the brain. What we can do now is slow the rate of damage.

For us, one of the main reasons for catching this early is that today, as you go through these conditions, you have your fall-risk person, your primary-care person, all these different folks helping different parts of the journey. That's a super painful process to go through — so why not centralize all the services and build a one-stop-shop care ecosystem around the condition? If you catch this early, you're inevitably going to need a lot of services, whether prevention or treatment, and we can optimize how those are delivered — both for you and for your family caregivers.

It's a good question — we've had multiple pharma companies reach out on that front. What's interesting for us is how you do it in the most altruistic way. From a pharma company's perspective, the incentive alignment is: we identify who's at risk and triage them to that pharma company, so they maintain price discrimination and all of that. From our perspective, that's not as attractive — our goal is to democratize the experience. What's super interesting is that if we can become this first lever, this first net that's catching folks, then as multiple companies come out with these drugs we get the ability to build a competitive marketplace. We have some say in how these things are priced, and the ability to triage people towards the right drug instead of just whichever one they find. We're very cautious about that when working with pharma, but we've had a lot of interest, some traction, and we're going through the early stages of some partnership trials.

Back to what I mentioned earlier — the point about what you do if you can't necessarily treat or cure these things — it's becoming that one-stop-shop care platform for Alzheimer's and other dementias. Becoming the place people go when a caregiver has a question, when a patient needs to take these recurrent cognitive assessments, or wants to engage with digital therapeutics, which is another area we're seeing really interesting traction on across the research landscape. We want to be the one-stop shop for all of that — provision of pharmaceuticals, synchronous consultations, symptom management, pathway planning.

Every piece of the Alzheimer's journey — diagnosis, prevention, treatment — is going to change a ton over the next 10, 20, 30, 50 years. What we want to do is position ourselves as that central consumer window into the space, the provider of care directly to consumers. That way we can evolve with whatever direction the industry goes. If a groundbreaking drug comes out in six months, how do we just follow consumers towards that drug? If a groundbreaking digital therapeutic comes out, it becomes that instead. In 10 years, we want to be the place where someone starts to develop Alzheimer's or has a concern about dementia — the first place they come, and the only place they have to go, through the entire life cycle of the disease.

It's a really good question — one I'm deeply passionate about outside the lens of Alzheimer's too. A couple of things. One of the things that's crazy — that I hear a lot when we talk to people who have a parent going through Alzheimer's — is that the reason they don't pick up on it for a while is that the person who has the condition usually knows.

They know they're in the early stages of what's happening — starting to forget things. Something emblematic of the condition across the board is that parents start shortening phone calls. Instead of talking to you for 10 minutes, they talk for eight, or seven, or five — shorter and shorter, because they know that if they stay on the phone too long they'll start repeating themselves. They don't want their kids to know, don't want to look weaker. Everyone does it, and it happens in a crazy number of cases — I'd say close to 30 to 40% of the cases we talk to, people report that in hindsight. And that happens when that phone call is the only point of interaction you have with an elder family member.

If you look at multi-generational households — across Asia, or even countries in Scandinavia; Norway is doing a much better job than much of the West at multi-generational living — where you have much more frequent touchpoints, you have the ability to catch conditions like this earlier anyway, because you're around those folks. In one way, that's what we can try to replace: if we can catch it even earlier than you could by talking to a relative every day, we can solve that problem digitally.

But the other thing is those mental exercises — keeping yourself stimulated as a way to slow the progression of certain symptoms. Some of the most prevalent symptoms, like short-term memory loss, can be strongly affected by that engagement. And we've unfortunately built a society where older people are relatively siloed — people alone in their homes essentially 24 hours a day, maybe going out once or twice a week. You don't see that as much in India, or in places where you live in multi-generational households, where your oldest generation is taking care of the little kids, everyone works together, everyone stays engaged.

As to how we fix that — it's a difficult question. I've personally angel-invested in a couple of startups doing interesting work in the US around multi-generational housing, building it in a community aspect rather than family-specific — how do we start to replicate multi-generational living? It's tough, particularly in the way the US is laid out. It's super easy in a lot of Europe or Asia to walk outside your door and see other people and engage in community; not as easy when you live a 15-minute car ride from the nearest park. You lose a lot of that random interaction. The research isn't super clear on how to quantify the effects on the prevalence of Alzheimer's, but intuitively it feels like it's there — and at the very least it would help from the perspective of symptom management.

Well, you hit one of them on the head — we get a lot of just, "what do you do?" You catch this early, and it's, what do you do with that information? Our usual take is the same as what I told you: there are a lot of small things you could do, but regardless of how much you can cure or treat this condition, it takes a lot of work, resources and money to get through. So if there's just optimization you can do from a care perspective, there's a massive opportunity — and people usually like that a lot.

Altruistically, diagnostics are really helpful for the idea that we can help people earlier and accelerate clinical research. But diagnostics are also something like a lead-gen function for care. We spend close to $300 billion a year in the United States on Alzheimer's care. So the question is: who's going to win that space, which today is super fragmented and has no direct-to-consumer provider? Catching this condition earlier for people, being that first leg in, is a great way to do that. Because even if I tell you, "we caught this, and there's nothing you can really do for the next few years" — that's unfortunate, but when symptoms start to present and you need nursing care, consultations, digital therapeutics, the person you turn to is that first player you've had a relationship with for multiple years.

So it's funny — it's the same thing people hate and love. They hate the idea that you can't really do anything immediately, that there's no clear-cut science on what to do. But they love that you can use this diagnostic leverage as a way to win that care landscape. It sucks from a real-life perspective, but it's super interesting from a business perspective. We don't know a lot about Alzheimer's at all, and we're going to learn a lot more about what treatments look like — so it's going to be a fantastic position to be in, three, five, 10, 15 years from now, to have been an established DTC provider of care that can wedge in any of these other solutions.

YC is a fantastic program. I don't know if I love the Harvard-of-startups name — I went to Yale, so we have a little thing with the big H. But it was a great program. Some of the things people really liked about us: early on we were able to demonstrate this wedge — this idea that you should be able to build digitized diagnostics for Alzheimer's and dementias. Coming into applying that September, we'd been able to show that people actually want this — people willing to sign up and get on a waitlist.

I think what YC liked was a couple of things. One, this is a massive problem in a massive space. For a program like YC — and, honestly, for me as I think about angel-investing into really early companies — it doesn't really matter exactly what you're doing right now. It doesn't matter what exact problem you're trying to solve within Alzheimer's, because it's a massive enough market with a ton of problems that need solving. If you have some experience in the space, you'll be able to solve one of those effectively. And that, plus my personal connection to the space. Although, to caveat, I'll never really know what they liked or didn't.

On takeaways, two things. One is the people you go through the program with — a fantastic cohort of companies all solving interesting problems in slightly different ways. You get this bundling of knowledge that doesn't happen in a lot of other places. Our technical lift is super light compared to some of the back-end, quantum-computing companies going through YC — so when you have a question on something like that, you call one of those people up and, instead of spending a week finding a solution, you get it in 10 minutes. Similarly on the medical side, there are a lot of founder-physicians going through YC who have experience with the insurance landscape — something I knew nothing about, and which is relatively difficult to learn online because the resources are scattered. You call one of these people up and get a lot of context very quickly.

The other big takeaway they teach you is: go really, really quickly. At the start of YC they say it in the first talk, and I thought, "oh yeah, whatever, we all know that intuitively." But YC does a really good job of forcing you to do it, building the structure that makes you set goals and try to achieve them. Most importantly, they tell you to set goals you're only 50% sure you're going to hit — which I'd never done before. There's an intrinsic shyness about setting goals that are too lofty, because no one wants to fail by their own standard. So YC forces you to set these two-week goals and try to meet them. Honestly, I don't think we hit our two-week goals every two weeks — but by the end of the program you're way further than you thought you'd be, just from trying to hit those heavy targets. And when you do it surrounded by everyone else doing it too, everyone leans in and goes all in. You get an energy you don't get many other places.

It's funny — related to what you said about me speaking really quickly. I get that all the time. And it's interesting in pitching investors: I got a lot of feedback early on, "oh, you talk too fast." I tried to fix it. But what I found is that if you look at almost every one of the investors in our $6 million seed round, one of the things they said they loved was exactly that.

If you're excited about the problem you're solving, you naturally speak a little quicker, and that excitement comes through. That was part of what everyone liked. I used to try to slow myself down, but what I found is that investors or people who don't care are going to zone out whether you're talking quickly or not — and people who do care will try really hard to be part of it. You can transfer a lot of that excitement to other folks.

That said, one thing that's been super useful is that I like to write as many things as I can. For anything important — and this is something I hated doing through school and college, but now do all the time — instead of putting together a PowerPoint for a presentation to the team, we write things. We write a two-to-three-page thesis on some perspective: what should we build, what should come next, who should we hire, what type of person should this be — even for specific people. Sometimes when we're interviewing, I'll have a perspective on somebody and our CTO will have a different perspective, and we'll both write a two-to-three-page paper on what we think, and read through them together. There's something really valuable in laying out an argument the way you have to when you write it down — which isn't required in the same way in a PowerPoint. When you write, the only thing that can shine is your argument, and that sometimes gets obfuscated in a PowerPoint or when you're speaking. So that's one of the new habits from all this startup stuff that's been super helpful.

Super interesting. To me it's like — if I look back on any of the interesting speakers I've heard, I don't know that I could tell you many salient points from what they said, but I could tell you how I felt afterwards, the vibe I was left with. That's a huge part of public speaking, probably the most important part — how you make people feel. One of the ways that comes through is the speed, the excitement, whatever it is. It's definitely slept on. When founders come and ask for advice on fundraising now, that's one of the things I say: honestly, if people talk fast, keep doing it. And if they talk slow, I tell them to kick it up a notch to get more of that excitement in there.

Yeah, good question. One of the things that's permeated my life in general — I was always the kind of kid doing homework on the bus on the way to school, trying to write in final answers while the teacher was picking it up. Because I was always behind on things, always kind of slow, and somehow it always ended up working out. In hindsight everything works out, but I always did best when I felt there was a fire under me. That was a really bad habit growing up. But it's interesting in the startup space, where it actually works pretty well — because there's always fire under you. There's always a customer asking for something else, always things burning. So it lends itself well to that; there are always things to get done.

In medicine, people have established views on how they think the system should work. We get a lot of investors who say, "this would be really interesting for us to entertain once you've built insurance partnerships or gone down that line" — and we really don't want to do that at all. People also say, "you should go around integrating this with physician practice, become the tool that sits inside offices." People give a lot of advice that makes sense through the lens of how medical tooling was done 10 years ago — how those things were built, how people achieved scale.

But now we're in a really interesting space in medicine, definitely accelerated by COVID, where you have the opportunity to build pure consumer plays — tools that go straight to people at home and, at least to start for us, are really cash-pay. That works well because, one, no one else is doing it. There are a lot of people building for purely value-based-care infrastructure — that's interesting, and we obviously want to do that eventually. But what makes way more sense is: if you build a really interesting consumer opportunity first, you can build it to a scale where you collect all the data while doing it, which makes the insurance companies more willing to sign onto a value-based payment plan. Whereas most other folks are raising money and spending it on a lot of trials and research and data collection — which is fine, but it's slower and more expensive, and there's no reason to take that route.

So we get a lot of people who want us to go the more traditional route, and some people who really love the differentiated perspective we're taking — all the investors we brought on board, for example. That's the piece of advice I've not taken, or thought was pretty bad. It makes sense for some people, some skill sets.

But one thing I think is important for founders in general: there's a difference between what the best way objectively to do something is, and the way you will be best at doing something — and people conflate those too often. Objectively, a great way to build a medical tool is to do the research, chase the insurer route, get approved, go through all that. Say that's the gold standard. It doesn't actually matter what the gold standard is if it's not what you're going to be good at building. The question for me was: am I going to be better than everyone else building here at navigating insurer landscapes, or doing large-scale academic research? I'm not — I'm not going to be better than people who've done that for 20, 30 years and are building products in that space. What I could be better at is: can we build a better consumer experience? Can we build something people actually want to use? So that was a big inflection point — a lot of people give you the advice they think is objectively best, and you have to filter out a lot of it and ask: what are you actually going to be best at?