Michael Snyder, Director of the Center for Genomics and Personalized Medicine at Stanford University, is obsessed with health-tracking wearables. During our interview, he wore four wrist-worn devices – watches and bands – and two rings. He is convinced that these small but powerful gadgets will help drive the transition from today’s “sickcare” to true healthcare, which, as its name suggests, should start with proactively monitoring health rather than waiting for disease to develop. They can also provide the longitudinal health data we need for personalized medicine – and to feed data-hungry AI models.
However, Snyder’s scope of interests is much wider than that. He made waves with a study that reported two distinct periods of pronounced age-related change in people’s 40s and 60s. He also studies the deeply individual patterns of aging, which he calls “ageotypes,” and is developing the concept of intrinsic capacity as a quantitative measure of health to be used in longevity research and drug trials.
I just saw you wearing all those wearables, and I think that’s a great place to start: how do we work with all the data from these devices?
In my case, I’m collecting a lot of data, but that’s more on the research side. The average person will normally wear just one, possibly two of these things – not six or seven or eight like I do.
Wearables are powerful because they measure 24/7, as long as you keep them charged. That means they’re tracking your health, they know your healthy baseline, and you can look for shifts from it. We got involved when they first came out as fitness trackers and thought they could be powerful health monitors because they measured resting heart rate and a few other things. Now they measure even more – heart rate variability, which is super important for health monitoring, blood oxygen. There’s medical value in that information.
We have strong circadian patterns throughout the day. Heart rate goes up during the day, blood pressure shifts – but if you track all that, you can look for shifts from the norm.
We started putting wearables on our cohort and discovered pretty much right away that we could tell when someone was getting ill. In my case, it started with Lyme disease. I picked it up presymptomatically because my blood oxygen dropped.
Then we showed that we could detect respiratory viral infections. It really hit home when the pandemic arrived in 2020. We partnered with Fitbit and showed that you could tell when people were getting COVID in advance of symptoms.
It works particularly well for COVID because it has a long presymptomatic period. Influenza and some other respiratory viruses have more like a 36- to 48-hour incubation before symptoms appear. We can still pick those up before symptoms, but with COVID you have more time.
Especially with COVID, I guess that’s important in two ways: preventing transmission and starting treatment as early as possible, such as with Paxlovid.
Yes, definitely. Paxlovid is thought to be especially effective when you use it right away. We’re running studies where family members of transplant patients have devices. If a family member gets ill, they’re alerted and can self-isolate so the transplant patient doesn’t get infected, which could be very serious.
We’ve had family members get red alerts. They quickly test for COVID or influenza, and often we can tell it’s one or the other. Sometimes, we know they’re ill but aren’t quite sure with what. The treatment differs depending on what they have, as you point out, and they can self-isolate. They’ve been pretty grateful for the alerting system. I think that’s the first sort of clinically actionable demonstration.
You’ve been working with wearables for years, and both their capabilities and the accuracy of the measurements have been increasing. Are they really coming of age now? Can they meaningfully change how we do research? There’s a distinction to be made here between tracking an individual trajectory and getting population-level insights.
They’re definitely embedded in research now. Everybody uses them to track activity and basic physiological parameters. They’re not yet embedded in the clinic. Some concierge services are bringing in these data – but not most standard health plans – partly because the existing system hasn’t been set up for it and isn’t incentivized that way.
We tend to practice sick care rather than healthcare because the financial incentives aren’t aligned. We need to fix that. If we actually practice healthcare, people will routinely wear these devices because tracking is so integral to keeping people healthy.
Other groups have shown that you can pick up atrial fibrillation with a smartwatch, and it works reasonably well – in the 30-50 percent range, I think. None of these devices is perfect, although they keep getting better.
Interestingly, for our alerting system, the number one trigger of red alerts is workplace stress, not a respiratory viral infection. That makes sense because it’s mostly built around heart rate and related measures. Now that we have other data types, I’m quite confident we can distinguish respiratory infections from mental stress. That’s something we’re working on.
Knowing when people are mentally stressed is a big deal. If something is mentally or physically stressing you, you should take care of yourself. We don’t have good biomarkers for mental health. There’s a lot more to do, but I think wearables will be excellent indicators of both mental and physical stress.
We haven’t published this yet, but we had a case of someone who died of a heart attack, and his wife shared his data. He had an Apple Watch and an Oura Ring, and it’s pretty clear he had a step-function change about four and a half months before the event – resting heart rate and a lot of other parameters shifted.
We need real-time systems that pull in the data, track people, and alert them when something is off. It comes back to the car-dashboard analogy. If a light goes on, you may not know exactly what it is, and it may be a false alarm in terms of something serious. But usually if an alert goes off, something is happening, and then you can follow up.
On the population side, wearable companies are sitting on these huge mountains of data. Do you think it will ever be possible to anonymize and use it? Are there efforts in that direction?
I don’t think companies are incentivized to share their data, and I expect they generally won’t. Some are forced to share if they want to publish. So a lot of this is going to come from academic researchers.
Many of us share our smartwatch data, but continuous glucose monitoring is a good example of the problem. There are many CGM studies, but it’s very hard, if not impossible, to get the underlying data. With smartwatches, some big biobanks are now putting wearables on people. All of Us has Fitbits on many participants. UK Biobank has done more with ActiGraph, but they’re discussing wearables as well.
I’d like to think the cohort we’ve followed pioneered some of this too. It’s relatively small, but we did deep -omics profiling, put wearables on people early, and learned that these things are powerful.
One of the more important things we learned was with continuous glucose monitoring. CGMs were being used a lot for insulin-dependent type 1 and type 2 diabetics. When we got involved, they weren’t really being used in so-called normal people and prediabetics. We put them on those groups and discovered right away that a lot of people thought to be normal weren’t so normal. They were spiking pretty badly – sometimes as badly as diabetics.
That was actually one of my questions. I’ve worn a CGM a few times for a few weeks.
Very powerful, right? You’ll never eat the same again. I mean that in a good way. You see what spikes you, and that’s very personal. What spikes your glucose can be very different from what spikes mine. We’re all very different.
Do you think the question about the importance of glucose spikes is settled? Are they really that crucial for health?
I think it’s settled in certain ways. There are at least two obvious lines of evidence. Time in range is strongly related to diabetes, and many studies show that diabetes strongly affects health, especially cardiovascular disease. There are also studies showing that postprandial spikes – spikes after meals – are associated with cardiovascular disease independently of diabetes. So, in my mind, the data on spiking are pretty clear. There are nuances, of course. If you lift weights, for example, you can break down glycogen into glucose, so not every rise has the same meaning.
Constant monitoring obviously is amazing: it gives you important insights into your health and can flag many things early on. But what about overdiagnosis, overtreatment, false positives, or just the constant background anxiety you can get when you measure yourself all the time?
I think it’s an education issue. You’re right that some people get overly anxious, and you want to be careful about how information is returned. It’s up to the person to decide how they want that information, in my opinion, but most people are quite capable of handling it.
When people first get these devices, everybody gets very absorbed in them – what foods spike you and all that. Then you settle into a pattern where they’re fairly useful. They can alert you to some pretty serious health issues, so I think it’s better to know. I like the car analogy. Your car has lots of sensors. You wouldn’t dream of driving it without a dashboard.
I could even argue that things like whole-body MRI can reduce anxiety in some cases. I know someone who was very worried because their family had a history of ovarian cancer. They got a whole-body MRI and were very pleased that everything looked good.
The mismatch with whole-body MRI is that people assume, ‘If you have nodules, you may have cancer.’ That’s the wrong way to think about it. You want to know what nodules you have, but the real issue is whether any are growing. Everybody has nodules. We need longitudinal data.
I have nine nodules. I get whole-body MRIs every three months. That may be overkill – I’m trying to see how often we really should measure people. If you have an aggressive cancer, it can take off pretty quickly.
Yes, we’re believers in whole-body MRI. You’re going to have nodules, and the key is knowing where they are so that if you ever get cancer, have it operated on, and then get follow-up imaging, you know what your background looked like. Without that baseline, you may be in a difficult position.
In my opinion, everybody should get a whole-body MRI so they know their baseline. I think you should get a discount on your health plan if you do these sorts of things – whole-body MRI, wearables, genome sequencing – because you’ll be better able to manage your health.
Hopefully we’ll eventually see some involvement from insurance companies. But for that, we need hard evidence that continuous monitoring actually works. Do we have it now? Are we close?
For wearables, I’d argue there are plenty of cases where they’ve been useful, but have the proper trials the medical establishment wants to see been done? Not really. It would be nice to have them so we can show that this keeps people healthier and maybe saves lives in some cases. They’ll have to be large because we’re doing health tracking, not disease tracking, and that’s a big difference.
Let’s move to your ARPA-H project, which I think is the biggest recent news. It studies intrinsic capacity. What is intrinsic capacity, why is it important, and how could it affect the way we do longevity research?
Intrinsic capacity is sort of a wellness score – a functional health state, if you will. We have lots of measurements for disease, and that’s what’s embedded in our health system. We have ICD codes: if you have a disease, it can be coded, with reimbursement mechanisms for the diagnostic tests and therapeutics associated with it. We don’t have comparable measures for wellness. That’s where intrinsic capacity comes in.
It’s built around functional areas. The WHO came up with five categories: cognition, locomotion, psychological well-being (things like depression and anxiety), sensory function (such as hearing and vision), and vitality. Vitality is kind of a giant bucket and probably should be broken into subtypes because it involves heart aging, blood, and many other things.
There are validated tests already, many of them surveys, and other measurements are coming from wearables. You can measure gait, heart rate, heart rate variability. Grip strength is a good one. Locomotion in general relates to mobility and strength. We’re funded to do two things. One is to build an intrinsic-capacity score that gets FDA approved – that’s the mission – and, if possible, divide it into subdomains. I like that because we’re big on something called ageotypes, which we can come back to. The idea is to have a quantitative score for someone’s health, not their disease.
That could also be a big deal for drug trials. Imagine you have a good drug – the GLP-1s, for instance, are now thought to have many important health benefits. How do you know whether someone’s health is actually improving? Or, you discover a new drug and think, ‘This is the solution.’ How do you prove it? You need a quantitative measurement that tracks improvement. That’s what intrinsic capacity is about.
The other part of this ARPA-H PROSPR grant is to build a home test. We think the final score will probably combine blood, wearables, and surveys. The idea is to make something simple that you can do frequently to see your health state. We’re supposed to get the home test down to around $100.
We’ve invented microsampling in the lab, where you collect small drops of blood, mail them in, and we can measure thousands of analytes from that tiny sample.
Is it fair to say that you basically did what Theranos tried to do?
In a sense, yes, but they could have done what we did, and they didn’t. They tried to reproduce a lot of conventional clinical tests. Some of our measurements are clinical-grade, but we don’t try to do every standard clinical assay – we don’t do LDL, for example – because that’s not how the technology is set up. What we do are scientifically validated measurements that are in the literature and are valuable markers.
You collect these small blood samples – from a fingertip, or there are methods that collect them from the upper arm – mail them in, and we make all these measurements. We spun out a company called Iollo that does this. You send in the sample, they make about 650 measurements, combine your data with information from the literature, and use AI to make very specific recommendations.
Most people improve their markers. We think this is the future: health tracking through wearables, facial and voice recognition, and biochemical measurements you can make at home. Instead of going to a physician every two years and waiting until you’re sick, you could do this routinely while you’re healthy.
If it’s easy, people will do it often. Going to a doctor’s office is inconvenient – you have to take time off work. It’s a pain. The key is making this easy and convenient so people get measured often and keep themselves healthy. That’s the mission.
We’re not doing every biochemical measurement you’d do in a physician’s office. We do quite a few, but some are surrogates, and they may even be better for certain purposes.
Physiology measured in a physician’s office can also be quite off. There’s white-coat syndrome: people get nervous, their heart rate and blood pressure may be high. If you pull someone’s resting heart rate from a smartwatch first thing in the morning, that’s often a much better reflection of what’s going on.
That’s a good point. When I go to the doctor, my blood pressure is always high for no apparent reason. It’s also one measurement a year or a few months, so unless it’s extremely off, it doesn’t reveal very much.
Exactly. What do you do with a measurement you know is flawed? Same with heart rate. Longitudinal measurements can be useful for infectious disease, mental health, and, we believe, heart issues and other things as well.
Another area we’re getting involved in is supplements. Supplements have a bad reputation, somewhat deservedly. You walk into CVS and there’s a whole row of them, and for most, the data aren’t very strong.
The data around foods containing many of those compounds are often better. Diets rich in antioxidant-containing foods, for example, are generally associated with better health outcomes, fewer events, and lower all-cause mortality.
But, critics will say, ‘You haven’t shown that the same thing works as a supplement.’ And that’s fair. We need more studies. There are some supplements, such as vitamin D in particular settings, where there’s evidence of benefit, but broadly, we need much better data.
So we launched a website called MySuppleHub. It’s a community-driven supplement encyclopedia. You can look up supplements you use or are curious about, get information about them, and share your experiences.
That actually sounds like it could be a game changer.
I hope so. We’ll see if it works. We just launched it and several thousand people have already signed up. We’d love to get millions. Then we want to take the supplements that are most widely used or look most interesting and run studies around them to see how they really affect health. I think that could be super cool.
I can see how an intrinsic-capacity score works for health monitoring and early detection. But, if we’re talking about aging biomarkers for use in aging research, what are its advantages over something like an epigenetic clock, which may be less explainable but perhaps easier to measure?
I think the epigenetic clocks from Steve Horvath and others are the prototype for all of this. They work. The data are pretty strong that they’re associated with all-cause mortality and other outcomes when the clock is accelerated. The limitation is that they don’t give you as much actionable information. What do you do with a methylation clock per se? There are methylation markers that are surrogates for particular things, but the overall number doesn’t necessarily tell you what to act on.
Through our deep profiling – metabolomics, proteomics, transcriptomics, and other measurements – we track people over time and see how they change. Everybody changes differently. Some are cardiovascular agers, some metabolic agers, some show more oxidative-stress aging. Some are all of the above; you can have combinations of things going off. Back to the car analogy: your car ages as a whole, but certain parts may age faster.
The entire car doesn’t break at the same time.
Exactly. You want to know what the weak link is. We call these aging patterns ageotypes. I like that name because it covers organ-specific aging – heart age, kidney age – but also more systemic things like oxidative stress and inflammation. So we can track how you’re aging.
Iollo, the company I mentioned, uses microsampling to measure your metabolic profile. They estimate biological age and your ageotype. You can see things like heart age and oxidative stress. Then AI can see what’s off and make very specific recommendations – not just ‘exercise more’ or ‘eat better,’ but specific dietary and lifestyle changes. People who follow the recommendations improve their markers about 95% of the time.
I think this is the future. Between wearables and microsampling for biochemical measurements, we’ll be able to measure people much more often, track their trajectories, and follow how they progress.
That brings up an important concept. The healthcare system focuses on population averages: are you inside the normal range or outside it? We think the individual trajectory is much more important. You can sit at the low end of normal as your healthy baseline, then double a value – a liver enzyme, for example – and still technically be within the normal range. Your physician may say nothing. But if a marker suddenly doubles, something may be off.
We’ve seen this in our research. In one case, a person reached out after a liver marker shifted and said, ‘Mike, what’s going on here?’ I said, ‘I don’t know – why don’t you get another measurement?’ He did, and the next measurement was outside the normal range. Under the traditional system, he might not have gone back until a routine checkup two years later, if at all. Who knows what damage could have occurred by then?
So, we think the individual trajectory is much more important than comparing one measurement with a population reference range.
That’s a paradigm shift that could eventually require redoing our entire healthcare system.
It is a paradigm shift, but I don’t think it’s that hard, technologically. You can have algorithms tracking you. We’re all going to have agents tracking our health in the future. There’s going to be a lot of information around you, and the system can alert you when things are off.
What about the incentives in the US healthcare system? Are they likely to help or impede this transition, especially compared with other countries?
The US is at a huge disadvantage because the financial incentives aren’t aligned. Our health system is fragmented. The average time someone stays in a health plan is about 18 months. Why would an insurer put a lot of money into prevention if, 18 months from now, you’re probably going to move to another plan?
Other countries often have single-payer systems. Once you get your genome sequenced, for example, that information stays in the system and can be used over time to help manage your health.
I think health plans should give people incentives for having a smartwatch, getting their genome sequenced, getting checkups – things that help people keep themselves healthy. Ideally, those people will have less chronic disease and cost the healthcare system less, although you could argue that maybe you’re just delaying some costs until the last year of life.
Do you think we’re too obsessed as a society with keeping our health data private?
Way too obsessed, in my view, because almost nothing is private anymore. There are cameras on every street corner. We all use credit cards, which generate a lot of personal data, and nobody panics about that because nobody wants to walk around with bags of cash.
With health data, what people are really worried about is abuse. I believe that in a wealthy society there should be some minimum level of healthcare, and you shouldn’t be discriminated against because of your health information. If you can protect people from misuse, everybody should be able to benefit from health tracking. To me, the privacy issue is overemphasized relative to the potential value.
Healthcare organizations have vast troves of data that we can barely touch because of all kinds of restrictions.
And they haven’t figured out the analytics. That will change with AI. The physician of the future is going to be an AI agent. In the immediate future, there’ll still be a human in the loop, and humans will remain very important for a while. Down the road, we’ll see.
But, we all need AI agents because there’s simply too much information. An agent can pull together all the information collected about you and make recommendations. And you can interface with it 24/7, which is kind of nice.
If people are shown clear benefits from sharing their data, you think they’ll become more open to it?
I hope so, because not sharing is a disaster. We have eight billion people on the planet. Even detailed information from just 0.1% of them is about eight million people. That would be an enormous amount of data. We need detailed data if we want to track all the elements of health.
Can we touch on your famous paper about the transitions around ages 44 and 60 and nonlinear aging? It became very widely discussed and probably misunderstood in some places. It also wasn’t a huge study, so I’d like you to explain what it actually showed.
It was a small number of people, but they were densely tracked, and that’s the key. The main point is that aging is nonlinear – certain things change more at certain times.
Some of what happens in the 60s was already well known before our study. Your immune system declines, you lose muscle mass, and so on. But we saw other things too. Oxidative stress changes throughout life and tends to increase sharply as you hit your 60s. We also saw a lot of changes in the 40s.
Some of the statistical methods we used have since been questioned, and those criticisms are correct – we did make a mistake there. But the broader conclusion that aging is nonlinear is correct, and we still see waves of change in the 40s and 60s.
Then the question is what underlies those changes. For the wave in the 40s, we think lifestyle is probably part of it. In your teens and 20s, you’re often very active – at least I was. In your 30s, you’re developing your career, you may have a family, and however hard you try, you’re probably not quite as active. I think some of that catches up with you.
We also shift our preferences as we go through life. But, if you look at people who live long, healthy lives, there are some basic ingredients. They’re very active. They tend to avoid ultra-processed foods. They generally have good social and community networks, and many have strong family networks. That’s understudied and underappreciated. My prediction – also not studied nearly enough – is that they probably have good sleep patterns too. Most people don’t sleep enough. Sleep is my weak point, by the way. I’m pretty good on the other things, but not sleep. I’ve been working on it.
It’s hard to maintain all these things throughout your lifespan. In fact, we train people improperly from the start. We put kids in school where they sit all day, and prolonged uninterrupted sitting is bad. Studies show that getting people to move every half hour is beneficial. That’s hard in many settings, but at least moving once an hour would help. We need to ingrain healthy habits earlier, and maybe then we can make aging a little more linear.
Circling back to wearables, if young people start adopting them en masse, maybe that will help move them toward the idea that they need to adopt a healthy lifestyle earlier, while they still feel fine, or to continue being active in their 30s, just like you said.
Exactly. Don’t wait and try to fix a broken system. Keep people healthy rather than fixing something after it breaks.
One more question about that study. It included a little over a hundred participants, if I remember correctly. Did you see people who didn’t show that two-wave pattern or didn’t show it nearly as clearly? There’s probably something to be learned from such outliers.
That’s a good question. We’ve tended to look more at people aging unusually rapidly than at the straight-liners you’re talking about. In our earlier work on ageotypes, we see people with very different aging patterns. Some are metabolic agers, for example, and some are quite obese.
But your question about people with relatively straight trajectories is a great one. We should go back and look more carefully. We’ve enlarged the study somewhat and are almost finished collecting about 12 years of data. What’s powerful is that the dataset is both dense and long.
But the follow-up in the original study was pretty short.
It was – around three and a half years for that analysis. Now, for some data types, we have much longer follow-up. That means we can correlate very specific lifestyle patterns with aging phenotypes and biochemical and physiological changes. I think that’s going to make a huge difference.
We run all kinds of studies – fiber supplementation, for example – and people in the cohort also go on and off things like statins or GLP-1 drugs as part of their normal lives. A lot of that is embedded in the longitudinal data. The number of people is still small, but the information is extremely dense. It may not generalize to a million people, but it can give us strong hints.
We already know that GLP-1 drugs have many effects, but I predict we’ll find some new ones that aren’t as well known because of how densely we sample and follow people. We’ve seen the same thing when people become sedentary or, on the flip side, start exercising – dramatic changes.
Correlating those changes with lifestyle will be very valuable. At the end of the day, we want actionable information that lets people improve their ageotype and their metabolic and other health phenotypes.
I want to end with AI. Like many researchers in aging and longevity, you seem to see AI as essential for deciphering the extraordinary complexity of aging – especially once we start collecting huge amounts of wearable, biochemical, and other data on each person. But AI has also become very controversial in society. How do you think about its advance? Should people in our field be ambassadors for the beneficial side of AI?
Some people feel threatened about job security and things like that, I guess. But to me, AI is the future. It’s going to be integrated into our lives and we’ll use it. Smartphones are integrated into our lives now. There are downsides – as a society, we probably have too much screen time – but it’s still an enormously useful tool.
Information is incredibly valuable for managing health. Medicine and health are information sciences, and there’s more information than any human can handle. We need AI agents that can collect the information about you, pull it together, and help determine what’s best for your health.
The caution is that AI builds on existing information. You need data to make these recommendations. AI doesn’t inherently know where the blank spots are. It can try to project into them, but that’s not the same as actually having the information. We need to fill those gaps – first so everyone is represented and can benefit, and second so we can give the best medical and health advice possible.
AI is already capable of a lot more than many people realize. In some circumstances, it clearly outperforms physicians.
I don’t think that message is out there enough. What people keep hearing is, ‘AI will help us develop new drugs.’ That’s important, but it’s not the whole story.
Not at all. The health-management side is already emerging. For certain tasks, like diagnosis from images, AI can be much more powerful than a human. We’re going to use things like retinal scans for health in the future, and much of that will be AI-driven.
