Wearables Will Save Lives, But We Need a New Relationship With Biometrics First

One year ago today, one of my best friends died of throat cancer at 44. To almost everyone who knew him, it came out of nowhere.

It didn't. It had been developing for a long time while nobody, including him, measured anything that would have revealed it. He said he felt off for months. His doctors told him he was too young and too healthy for it to be serious, and didn't test him until his symptoms were bad enough that everyone could see something was wrong.

I've spent twenty years working on how organisations measure what matters and how badly they get it wrong. We make the same mistake with our own bodies, at far higher stakes. A symptom has always meant the moment you feel something wrong, which is why medicine starts looking so late.

People don't get sick suddenly, they get diagnosed suddenly

The suddenness is an artifact of measurement, not biology. Throat cancers develop over months to years and atherosclerosis over decades, while insulin resistance precedes a diabetes diagnosis by years in which the body is sending signals nobody reads.

Your own perception fails first. One bad day gets filed under a bad night's sleep. That becomes a bad week you put down to stress or age. Eventually it's just how you feel, and the memory of what a good day felt like is gone. You never decide to accept a worse baseline. You lose the comparison, quietly, and stop being a reliable witness to your own health.

Three signals that were there the whole time

Hindsight makes everything look obvious, and no wearable on the market diagnoses throat cancer. But it's worth being specific about what a continuous record of him would have held.

• His voice. A change in the voice is the most common early sign of throat cancer, and far too gradual for anyone to hear in themselves. Millions of us now wear devices that record our voice all day. They transcribe every word we say and pay no attention to how we sound.
• His sleep. Something growing in the airway disturbs breathing at night well before it's noticeable in the day. Sleep and breathing tracking is now standard on Oura, WHOOP and Pixel Watch. Months of steadily worsening sleep is a pattern you can point at.
• His resting heart rate. A resting heart rate that creeps upward over months doesn't tell you what's wrong, only that something is. It's the simplest number these devices collect, and the newest ones are finally accurate enough that a small drift reads as signal instead of noise.

The one that would actually have found it is close. Researchers in Shanghai have built a wearable ultrasound patch that images tissue continuously with AI reading the results in real time, accurate enough to stand in for an invasive catheter. It was built to monitor pressure around the heart, and pointing it at the neck is an engineering problem rather than a scientific one. A patch that watches a lymph node is plausibly two to three years out.

The first three would not have printed a diagnosis, but together they would have given him what he never had: evidence that his own normal had moved, months before anyone was willing to believe him. The fourth would have shown someone the mass.

A system built on triage cannot look for what isn't presenting

The second failure wasn't personal. Every public health system I've lived under is underfunded, oversubscribed and organised around triage. That's arithmetic rather than incompetence. There isn't capacity to work up every tired 44-year-old, so tests get rationed by risk profile, and risk profiles are built from population statistics.

The physician who saw a young man with no family history was reasoning correctly from base rates. In that population, those symptoms overwhelmingly mean nothing. That reasoning is right most of the time, which is what makes it so hard to argue with in the room. It was right about the population and wrong about him.

Both failures share one structure: he was compared against a population instead of against himself. Neither he nor his doctor had a record of his own normal, because none existed.

Steps and heart rate are the vanity metrics of health

Step counts and resting heart rate are the health equivalent of pageviews and daily active users. Cheap to collect, visibly responsive to effort, wrapped in a satisfying daily ritual of closing rings. They're also nearly disconnected from the processes that kill people in their forties. You can hit ten thousand steps every day for a decade with deteriorating metabolic function, rising blood pressure and a tumour growing, and your device will congratulate you every evening.

Those metrics are built to be reassuring, which is what makes them useless as warnings. The industry optimised for numbers that keep people wearing the device over numbers that would tell them something they need to know, because the first kind sells hardware and the second occasionally ruins your morning.

What the new generation actually reads

The devices arriving this year read function rather than activity: each answers a question your annual physical never can, because the answer doesn't exist in a single measurement. It only exists in a trend.

They watch what moves first. Google's Pixel Watch 5 tracks insulin resistance, a first for a consumer wearable. By the time a standard glucose test comes back diabetic, the process behind it has usually been running for years. Insulin resistance moves first, so watching it puts you ahead of the diagnosis while there's still time to change the outcome. Dexcom's Stelo now sells over the counter for $98 with no prescription. Continuous glucose data used to require being ill enough to qualify for it.

They're accurate enough to trust a small change. Trend detection lives or dies on precision, because the drift that matters is a few beats over months. If a device's margin of error is wider than the change you're looking for, no amount of patience will reveal it. The latest research has narrowed that margin by roughly a factor of four.

They fill the gaps between appointments. A smart bra now screens breast tissue in half an hour at home, without radiation or a clinic visit. Between annual mammograms there is no monitoring at all, and disease doesn't schedule itself around the appointment calendar. WHOOP came at the same problem from the other side, pairing blood panels that include a multi-cancer test with AI analysis and dropping the requirement to own one.

None of this diagnoses anything. Each system learns what is normal for one person and flags departures for a human to interpret. Detecting deviation across longitudinal data is what this technology is good at; diagnostic reasoning is what it's bad at, and the products worth trusting know the difference.

What's already coming

Your data meets your medical record. Microsoft's Copilot Health draws on more than 50 devices and 50,000 US providers, reading wearable data and clinical records together in plain language, and OpenAI has connected Apple Health and medical records inside ChatGPT. Your trend line stops being a private curiosity the moment it can sit next to your bloodwork.

The rules get written. Cedars-Sinai has published a framework for how hospitals should handle alerts generated by AI from wearable data, covering what gets validated and who carries liability when the machine is wrong. Until rules like these are normal, a clinician who takes your data seriously is doing you a favour rather than following a protocol.

Further out, in the five to ten year window, sits the thing that would end deaths like my friend's. Every wearable today reads physics: motion, pressure, temperature, the way light bounces off blood. The leap is reading chemistry continuously. We already have proof it works, because a glucose monitor sits a filament thinner than a hair in the fluid between your cells and reports one molecule every five minutes. Nothing about that approach is specific to sugar. The same filament, or a patch of microneedles too short to feel, could report inflammatory markers, hormones, and eventually the fragments of tumour DNA that circulate long before a lump is large enough to find.

The Galleri test already detects signals from many cancers in a single vial of blood, but you have to book it and almost nobody does. Move that onto something you wear and cancer stops being found by an appointment you happened to make. It gets found by a line on a chart that changed in March.

What I built with my own data

I wear a Garmin. On its own it gave me what everyone else's wearable gives them: a great many numbers and no synthesis.

So I built a health brain. My Garmin data flows into a GitHub repository that updates daily, and I connected Claude to it. What I have now is something I can ask questions of. What changed in my sleep during the weeks my recovery scores dropped. Whether my resting heart rate moves with travel, or alcohol, or how hard I've been training.

Having the record changed what I did next. I got an executive scan, a sleep study and a full blood panel, not because I felt unwell but because daily data is only worth what you can anchor it against.

What the record makes possible is individualised experiments. My resting heart rate is 32, which is low enough to take seriously. A very low resting heart rate can mean a well-trained heart, or something wrong with the way the heart is firing, and population averages cannot tell you which you are. I was worried it was the second.

So I designed a test. Working with Claude against my own history, I worked out what would separate the two explanations: whether my heart could climb properly under load, hold a high rate, and settle quickly. I ran those alongside a proper cardiac exam. It climbed cleanly, held and recovered fast, which is what a well-trained heart does and a failing one can't.

That's the boring outcome, and for most people it usually will be. Suppose instead those experiments had shown a heart that couldn't climb the way it should, or the scan had found an enlarged chamber. Catching that at 44 rather than 45 is the difference between managing a condition and being treated for an emergency.

Where to start

Most people buy the device first and work out what to do with it afterwards, which is how it ends up measuring nothing that matters. Invert that.

Name the concern before you name the device. Heart disease in your family, a metabolic risk you suspect, sleep you know is bad. Decide what you want visibility on, then buy the thing that gives it to you.

Learn to read the numbers. If you read one thing, make it Peter Attia's Outlive, the clearest guide to which markers predict how you'll actually die and which are noise. Andrew Huberman's podcast is a reasonable second source, though he attracts fair criticism on specifics. Take the premise rather than the regimens: your physiology is something you measure continuously instead of investigating once it breaks.

Start the record before you need it. A baseline is worth nothing in the week you first feel unwell and worth a great deal after six months of showing what your body looks like when it's well. It can only be built in advance.

Watch function, not activity. How your blood sugar responds to meals, where your blood pressure is heading, which direction your resting heart rate has moved over a year. These shift before a diagnosis does. Step counts never will.

Adopt a system or build one. Copilot Health and the ChatGPT health integration will be enough for most people. What matters is that something holds your history and can answer questions across it.

Change what you bring to your doctor, not who diagnoses you. Your clinician remains the expert, and using AI to diagnose yourself is a genuinely bad idea. What a personal record changes is everything around the appointment: you notice a shift within weeks rather than months, you can describe what changed and when instead of offering a feeling, and you have a better sense of which specialist you need. "I've been tired" is easy to dismiss. Eleven weeks of a resting heart rate above your own normal is a different conversation.

The hard part was never the sensor

In twenty years of helping technology companies work out why good products fail to get adopted, I have never seen an opportunity to affect people's lives on this scale. The engineering will arrive roughly on schedule. Three things stand between it and any real reduction in preventable death, and none is a sensor.

The first is form factor. Whatever collects this has to disappear into a life somebody is already living. A ring is effortless and limited in what it reads; a patch reads far more and asks more of you. There is a serious argument among researchers that the richest signal most of us produce daily is the one we flush without looking, and that a smart toilet would out-diagnose anything on your wrist. The winning product will collect the most valuable data while demanding the least discipline, and those pull in opposite directions.

The second is behaviour, and it's harder. The people who most need early detection are least likely to go looking for it. Men in particular put off appointments, talk themselves out of symptoms and treat medical caution as a character flaw, which is part of why my friend's cancer got as far as it did. A device that detects a problem accomplishes nothing if the person wearing it won't act on it.

Third is a new relationship with our own health. There are more ways to improve your outcomes than at any point in my lifetime, and they open up the moment you decide they're worth the money and the attention. We'll spend without thinking on new clothes or an expensive cocktail, then balk at the cost of a test that would tell us something real. Nobody needs Bryan Johnson's level of obsession here. The shift is smaller than that: recognising what ten more good years is worth, to you and to the people who would rather not lose you early. Sleep and cortisol are the plainest place to start, because they sit underneath everything else you're chasing.

If you're building an AI wearable startup and need an expert in overcoming barriers to adoption and deepening client relationships, I want to help. You can find me at ph1.ca or reach me at [email protected].

My friend ran out of time a year ago today. I hope you and the people you love make the choices that add years to your lives.