Thoughts
Start Here: Core Ideas
This site explores how information, technology, and crisis are reshaping healthcare and other institutions.
If you're new here, these essays lay out the central themes:
Disruption for Doctors, Part 1: What Is Disruption?
A clear explanation of Clayton Christensen’s disruption theory, applied to medicine.Disruption for Doctors, Part 2: Examples in Healthcare
How Walmart Health, AI diagnostics, and automation are already changing the system.Disruption for Doctors, Part 3: The Rise of Self-Care
The biggest disruptions bypass doctors entirely — and move care into consumers’ hands.
You can browse all posts below.
Hidden Connections: What John Muir Can Teach Us About Apple’s New Hypertension Notifications
The naturalist John Muir saw how everything in nature is connected — and today AI is showing us the same truth inside the body. From Apple Watch studies on atrial fibrillation to new hypertension alerts, hidden links in long-collected data are transforming how we understand health.
“When we try to pick out anything by itself, we find it hitched to everything else in the universe.”
John Muir, circa 1902
When the naturalist John Muir penned that quote, he was writing about the natural world after working as a shepherd in California’s Sierra Nevada, immersed in the mutual dependence of mountains, forests, rivers, and animals.
More than a century later, his insight is proving just as true for the wilderness inside of us. We’re discovering that nothing in our physiology exists in isolation. The eye, the heart, the skin, the voice — each carries hidden connections to other systems. And increasingly, it is AI that is uncovering those links.
An Early Example: Apple Heart Study
In 2017, Apple and Stanford launched the Apple Heart Study, enrolling more than 400,000 participants. Researchers asked a simple question: could the Apple Watch’s pulse sensor — a photoplethysmograph (PPG) LED originally designed just to measure heart rate — be used to detect something far more serious?
The answer was yes. By applying machine learning, Apple showed that irregular pulse patterns captured by PPG could flag atrial fibrillation (AFib), a condition often silent until it causes a stroke. The connection between a consumer watch’s pulse sensor and a dangerous cardiac arrhythmia was a hidden connection — one not previously recognized or used at scale. And gaining the valuable additional information didn’t involve any new hardware at all.
More Hidden Connections Revealed by AI
Since the Apple study, machine learning has surfaced hidden links across medicine, and not just with watches and phones and other mobile devices:
Retinal images — long used to check eye health — can reveal sex, race, smoking status, blood pressure, and even cardiovascular risk.
Gait and voice patterns — captured casually by smartphones or microphones — can detect the earliest signs of neurological disease without a visit to a neurologist
Electrocardiograms (ECGs) — the same tracings used for decades to diagnose heart rhythm — now allow AI to screen non-invasively for anemia, electrolyte imbalance, and left ventricular dysfunction
CT scans — often ordered for one purpose — can be re-analyzed at minimal cost to estimate bone density or coronary calcium, each tied to long-term risk — without needing additional scans (or radiation)
The information was always there. What’s changed is that AI gives us the ability to see it.
The Apple Watch and Hypertension
Perhaps the end of the beginning
This is particularly true for Apple’s new hypertension notifications, because they don’t use any kind of new sensor in the Apple Watch. The same green LEDs still measure blood oxygen levels. The electrodes still capture heart rhythm. The accelerometer still tracks movement. What changed was Apple’s recognition that these existing signals already carried the imprint of something new: hypertension.
For years, millions of Apple Watch users have been streaming this data. But it’s only now — after training AI on the right relationships — that the watch can flag blood pressure risk with confidence.
Another hidden connection uncovered.
Beyond the Watch
We can expect that this pattern will repeat itself everywhere: applying AI to old data streams to produce new insights. Combined, of course, with new data streams — very likely including non-invasive blood glucose monitoring in the near future.
Credit card purchases — will allow us to understand not just spending, but nutrition and mental health
Browsing histories — can understand cognition and promote education
Medical images — will be repurposed by AI to surface cardiovascular, skeletal, or metabolic risk factors unrelated to the original reason for the scan
Muir’s observation was ecological, but it applies just as well here. Health is not a set of isolated numbers. It is a network of hidden links — and we are only beginning to trace them.
Closing Thought
Apple’s hypertension notifications are not the end of something — they are the beginning. They can help us imagine how every stream of data we collect may reveal more hidden connections between body systems than we ever imagined. And the future of health will be built on our ability to recognize those links connecting what we already measure to what we most need to know.
Forget the EHR — Your Health Data’s On Your Phone
The overwhelming majority of health-relevant data —movement, behavior, speech, sleep — is now generated outside the clinical setting. As a result, health innovation is increasingly shifting toward consumer devices and tech platforms that actually hold the data — not the EHR or the healthcare system.
In 2020, I asked a provocative question:
“What does it mean for the future of health when Apple makes more money from AirPods than the top ten EHR vendors combined?”
At the time, it might have seemed like an unfair comparison—enterprise software versus global consumer electronics. But five years later, it’s clear that this contrast was not only valid, but also predictive of a broader shift: health data—and thus health innovation—has moved out of the clinic and into the consumer’s pocket.
The Real Center of Health Data Is Not the EHR
“We have 50,000 times more consumer health data than EHR data. So where is AI health innovation going to happen?”
The Electronic Health Record (EHR) remains the data centerpiece of traditional medicine. But in terms of data volume, it’s no longer the main source of information about our health. And it’s not even close.
U.S. EHR Data Volume
Each U.S. patient generates an estimated 80 megabytes of EHR data annually, including clinical notes, lab results, and imaging. Multiply that by roughly 250 million Americans interacting with the healthcare system annually, and you get:
→ 20 petabytes per year of U.S. EHR data
U.S. Consumer Health Data Volume - Calculated by Total World Data
There is no question that the real volume, and the real growth, in health-related data is happening outside the exam room, and outside the EHR.
Every day, we generate health-related data through wearables, smartphones, browsers, keyboards, credit cards, sleep monitors, voice assistants, and fitness apps. Not to mention smartwatches, rings, and earbuds.
One way to estimate this is by looking at the total data produced worldwide:
Global data generation reached approximately 147 zettabytes in 2023. If we very conservatively estimate that only 5% of this is health-related consumer data (i.e. somehow related to, or useful for, our health), we get approximately 7.4 ZB/year globally.
Adjusting for the U.S. — which has about 4-5% of world population but likely 10-15% share of digital activity — we can conservatively estimate that the U.S. generates roughly 1 zettabyte (1,000,000 petabytes) of consumer health data annually..
→ That’s 50,000 times the volume of U.S. EHR data
Another way is to look at individual data generated.
For example, everyone with a smartphone — which Pew estimates to be 91% of the roughly 230 million US adults, or 209,000,000 people. Even if we just look at the accelerometer data for those people, leaving out all the other health-related information on their phones (like speech patterns, word choices, grocery bills, etc) continuous low-frequency sampling (1–2 Hz) on an iPhone can generate 1 GB of raw movement data annually. Which is 209 pettabytes for the adult population.
→ That’s still ten times the volume of U.S. EHR data
So even our low-end estimate shows that we’ve got 90%+ of our health data sitting on our phones, not in the EHR. And it’s not hard to believe that if we wanted to predict, for example, cardiovascular risk, using either the average person’s EHR data (which is likely minimal) versus that person’s continuous data on how fast, how far, and how often they move . . . it’s the phone data that will produce a better prediction.
Although as far as I know, no one is yet doing that. Sigh.
Lark health coaching
Innovation Follows Data
If the EHR holds such a tiny part of U.S. health-relevant data, where should we expect most of the next breakthroughs in diagnostics, prediction, and prevention to occur?
Not in the hospital. Not in the clinic. Not in the EHR.
Innovation is flowing to where the data is—in consumer-facing platforms and tools. We’re already seeing this play out in:
AI Symptom Checkers: Platforms like K Health and ChatGPT-based tools provide users with AI-driven assessments of their symptoms, offering preliminary insights before consulting healthcare professionals.
Wearable-Based Sleep Optimization and Stress Tracking: Devices such as the Oura Ring and WHOOP monitor sleep patterns and stress levels, providing users with actionable insights to improve rest and manage stress effectively.
Early Illness Detection: Wearables like the Oura Ring have introduced features such as Symptom Radar, which can detect early signs of respiratory illnesses by analyzing physiological data, enabling users to take precautionary measures promptly.
Fitness and Glucose Feedback Loops: Dexcom G7 offers continuous glucose monitoring that now connects directly to Apple Watch, allowing users to track their glucose levels in real-time. Similarly, Levels and Ultrahuman provide platforms that integrate glucose data with lifestyle factors to optimize metabolic health.
Personalized Coaching and Risk Scoring: Apps like Lark Health utilize AI to offer personalized health coaching, helping users manage conditions such as diabetes and hypertension through tailored advice and monitoring.
Mood Inference and Mental Health Support: Ellipsis Health has developed voice-based technology that analyzes speech patterns to assess mental health, providing insights into conditions like depression and anxiety.
These advances are possible not because of structured ICD codes—but because of rich, continuous consumer data at massive scale.
Conclusion: The Real Medical Record Is Distributed
Ultrahuman metabolic tracking
In 2025, the EHR remains essential—but it’s nothing like central. The richest, most frequent, most predictive health data lives outside the clinic.
And that means the future of health won’t be driven by Epic, or Cerner, or Allscripts. It won’t even be driven by United Healthcare, or Kaiser, or Mass General.
It will be driven by whoever best understands—and leverages—the flood of health-relevant data being generated by consumers, not patients. And that is likely to be the consumer tech companies, both giants like Google and tiny startups or that haven’t even left the garage yet.