About
Former U.S. Army Airborne Ranger turned Longevity Physician.
I bring the…
I bring the…
Articles by Nicholas
Activity
940 followers
-
Nicholas Cohen, MD shared thisI tested 7 of the newest consumer body-composition scales against DEXA. The results surprised me. This was an independent study with no industry funding or manufacturer involvement. I could not find another head-to-head comparison of these current models, so this gave us a chance to generate some genuinely new data. I partnered with Patrick at DexaSlim: https://dexaslim.com/. We recruited 19 volunteers, including men and women ages 22 to 66, with DEXA body-fat percentages ranging from 15.7% to 40.3%. Each participant completed a DEXA scan and then, during the same visit, completed measurements on the consumer scales. We used a predefined analysis, retained all readings in the dataset, excluded only prespecified flagged measurements from the primary analysis, and ran sensitivity analyses and paired statistical comparisons. The full participant-level data and statistical results are in the images. The biggest surprise was the $23 Etekcity. It was the simplest device we tested and the only one without hand electrodes, yet it finished just behind the top-ranked $479 Tanita RD-545. On average, Tanita was 10.0% off from DEXA and Etekcity was 13.6% off. Among participants measured by both devices, the difference in absolute error was small and not statistically significant. At the other end, the $599 Withings Body Scan was 26.4% off from DEXA, nearly twice the error of Etekcity. Hume was 42.1% off, more than 3 times the error of Etekcity and more than 4 times the error of Tanita. And all 7 scales underreported body-fat percentage relative to DEXA on average. Higher price, more hardware, and heavier marketing did not guarantee better agreement with DEXA. This is a small independent study, not the final word. But the results were consistent enough across our analyses to give me much more confidence in which devices performed well and which did not. And this is only half the story. Next week, I’ll share the second part of the study, where I looked at how these same scales performed at measuring muscle. So if you’re thinking about buying one, stay tuned. #Longevity #BodyComposition #DEXA #HealthTech #PreventiveMedicine #DigitalHealth #EvidenceBasedMedicine
-
Nicholas Cohen, MD shared thisSomewhere in middle age, the pounds start to accumulate. Not all at once — a little each year. You notice the waistband, the photos, the number on the scale drifting the wrong way. And what makes it worse than the weight itself is the feeling that comes with it: frustration, and a quiet sense that there's nothing you can really do about it. I felt that too. So I ran an experiment on myself. I want to be upfront — this isn't something I recommend to patients. But it worked well enough that I'm curious what people here think of the logic. It takes standard weight-loss advice and adds one twist. Start with the basics: → Weigh yourself daily. → Aim for one pound of loss per week. → Eat three meals a day. No snacks. So far, nothing new. But here's the question almost no plan answers: what do you do when the scale won't budge? That's where most people stall out. The number sits there, motivation drains, and "try harder" is the only advice on offer. Here's the twist I tried on myself: on the days I was behind my one-pound-a-week pace, I skipped breakfast. That's it. One clear move, on exactly the days I needed it. I call it conditional fasting. It's an n-of-1 experiment, not a study — but the logic lines up with what behavioral research actually shows about weight loss: → Daily weigh-ins only help if they trigger an action. Self-weighing drives weight loss by prompting corrective behavior, not by the measurement itself. A number with no response attached does nothing. (JAMA, Obesity Management in Adults, 2023; Brockmann et al., Obesity, 2020 — https://lnkd.in/e3ruvGNQ) → Skipping breakfast on off-pace days creates an intermittent, self-correcting caloric deficit — the deficit does the work, and it switches on only when you're behind. (Bonnet et al., Obesity, 2020 — https://lnkd.in/eTKgPk3J) → Having a reliable move for the stalled scale builds self-efficacy — that sense of "I've got this" is one of the most consistent predictors of who keeps weight off. (Comșa et al., Behaviour Research and Therapy, 2020 — https://lnkd.in/e2_xyxmw; Nezami et al., Health Psychology, 2016 — https://lnkd.in/eTQTw-DH) One caveat: skipping meals isn't for everyone — a history of disordered eating or a tendency toward low blood sugar are both reasons to be cautious. Give it a try and let me know what you think. #WeightLoss #BehaviorChange #Habits #Health #longevity #diet
-
Nicholas Cohen, MD shared thisWhat I Tell My Patients About Eggs Eggs are one of the most debated foods in nutrition. Patients ask me about them constantly, and the conversation usually surfaces three arguments. Let me walk through each — then share what I actually recommend. --- "It's not the eggs — it's the bacon and butter Americans eat with them." This has real merit. In Asian populations, egg consumption is associated with slightly lower cardiovascular risk. In the US, it's consistently higher. American eggs travel with sausage, cheese, and buttered toast. Asian eggs travel with vegetables, rice, and fish. So is it just the company eggs keep? Probably not entirely. In Zhong et al. (https://lnkd.in/guJZq493), a pooled analysis of 6 US cohorts (n=29,615), adjusting for dietary cholesterol made the egg-mortality link non-significant — suggesting cholesterol in the yolk is mediating the risk, not just the surrounding diet. The NIH-AARP cohort (n=521,120) confirmed this (https://lnkd.in/gGprt5sj). Could residual confounding explain some of this? Yes. But the signal tracks with cholesterol content, not just dietary pattern. --- "Isn't saturated fat the real problem? Not dietary cholesterol?" Saturated fat is the bigger driver of LDL. But "bigger" doesn't mean "only." Dietary cholesterol independently raises LDL, especially in "hyperresponders." A single yolk contains ~186 mg of cholesterol — one of the most concentrated sources in the American diet. If you have elevated cholesterol, that matters. Reducing saturated fat and dietary cholesterol aren't competing strategies. They're additive. --- "I thought eggs were fine now. Didn't the guidelines change?" This traces to the 2015 Dietary Guidelines, which removed the 300 mg/day cholesterol cap. Headlines declared cholesterol doesn't matter. But the report stated dietary cholesterol should still be minimized — the cap was removed for insufficient dose-response data, not because cholesterol was exonerated. The AHA's 2021 guidance reaffirmed that dietary cholesterol raises LDL and should be limited. The "eggs are fine" narrative was a media creation, not a scientific consensus. --- So what does the data show? Each additional half egg per day was associated with a 7–8% increase in relative risk of death in US populations. The relationship was linear. There was no "safe threshold" below which the risk disappeared. My recommendation: Keep the whites. Skip the yolks. Egg whites are excellent, low-cost protein without the cholesterol — one of the simplest swaps with real evidence behind it. 🥚 Whites in. Yolks out. #CardiovascularHealth #Nutrition #EvidenceBasedMedicine #Cholesterol #HeartHealth #PreventiveMedicine
-
Nicholas Cohen, MD shared thisWhich body-composition scale should you trust? There are dozens of consumer scales that claim to measure body-fat percentage and muscle mass. They use different electrodes, frequencies and proprietary algorithms. They produce polished reports with precise-looking numbers. But there is almost no objective way for a consumer to know which scale is actually more accurate. That matters because body-fat percentage and muscle mass are meaningful measures of metabolic health, physical function and healthy aging. And historically, some consumer scales have been very wrong. In prior validation studies, certain devices underestimated body-fat percentage by more than 10 percentage points. Lean- and muscle-mass estimates have also shown poor agreement with reference methods and have been particularly unreliable for determining whether someone is actually gaining or losing muscle over time. Pasted text.txt The confusing part is that these scales can still appear highly precise. Step on the same device twice and it may give you nearly the same result. But a measurement can be consistent and still be wrong. Much of the published research evaluated an earlier generation of devices. Today’s scales use newer algorithms, multiple frequencies, hand and foot electrodes and segmental measurements. I could not find an independent study comparing the major scales people are buying today head-to-head. So I decided to do one. I purchased eight popular body-composition scales from: Withings Hume Renpho Tanita InBody Omron Etekcity GE Their prices ranged from $19 to $620. The question is whether any of them—and which one—provides a result that meaningfully agrees with DEXA. I partnered with DexaSlim in Midtown Manhattan to find out. We will enroll people already arriving for a DEXA body-composition scan, have them step on each scale, and compare their body-fat and lean-mass results directly with DEXA. And, refreshingly, there is no commercial sponsor behind the study. I purchased every scale myself. I do not own stock in any of these companies, and none of the manufacturers provided equipment, funding or input into the study. I have no financial interest in which scale performs best. The goal is simple: Which consumer body-composition scale is actually closest to DEXA—and which one should you trust? I’ll share the results in an upcoming post. Which body-composition scale do you use—and how accurate do you think it actually is? #BodyComposition #DEXA #MuscleMass #BodyFat #Longevity #Healthspan #ConsumerHealth #IndependentResearch
-
Nicholas Cohen, MD shared thisHow Much Protein Do You Actually Need? You've probably heard you should eat "1 gram of protein per pound of body weight." It's catchy. It's easy to remember. And it's wrong. This number comes from bodybuilding culture, not from science. It overshoots the research by about 40% — and the excess doesn't build more muscle. It gets burned as fuel, converted to fat, or flushed out as urea. Worse, it may cause harm. Too much protein doesn't help. The "1 gram per pound" rule pushes most people to roughly 2.2 g/kg/day. At that level, the evidence raises real concerns. A prospective cohort of over 9,000 adults found the highest protein quartile was associated with a 3.5-fold higher rate of kidney hyperfiltration and a 1.3-fold higher rate of rapid kidney function decline. (Nephrology Dialysis Transplantation, 2020) Sustained high protein intake has been associated with intraglomerular hypertension, glomerular injury, and proteinuria. (JASN, 2020) A meta-analysis of 11 cohort studies found higher animal protein intake was associated with a 9% increased risk of cardiovascular mortality. (European Journal of Epidemiology, 2020) And for all that risk? Zero additional muscle. Surplus amino acids are simply deaminated in the liver and either oxidized for energy or converted to glucose or fat. (New England Journal of Medicine, 2024) Too little protein doesn't work either. In the largest meta-analysis of protein and lean body mass, adults under 65 consuming 1.2–1.59 g/kg/day during resistance training had a small, non-significant effect on lean mass (SMD = 0.15, p = 0.07). At 1.6 g/kg/day or above, the effect nearly doubled and became clearly significant (SMD = 0.30, p < 0.01). (Journal of Cachexia, Sarcopenia and Muscle, 2022) This matters because muscle is not cosmetic — it is a longevity organ. Low skeletal muscle mass is associated with a 57% increased risk of dying from any cause. (PLoS One, 2023) Muscle protects against falls, metabolic disease, and functional decline. Underfueling it has real consequences. So what's the right number? Two of the largest analyses ever conducted — over 100 randomized controlled trials combined, spanning thousands of participants — tested every protein intake level and asked: at what point do you stop gaining more muscle? The answer: 1.6 grams of protein per kilogram of ideal body weight per day. That's the ceiling. Above it, additional protein produced zero further gains in lean body mass. The dose-response curve flattens completely. (British Journal of Sports Medicine, 2018; Journal of Cachexia, Sarcopenia and Muscle, 2022) Below it, gains are suboptimal. At it, your muscles have everything they need. One number. Backed by over 100 trials. Not 1 gram per pound. Not "as much as possible." Just 1.6. For those of you who counsel on this — what do you recommend? #ClinicalNutrition #PreventiveMedicine #LongevityMedicine #Sarcopenia #PhysiciansOfLinkedIn
-
Nicholas Cohen, MD posted thisHow Much Protein Do You Actually Need? You've probably heard you should eat "1 gram of protein per pound of body weight." It's catchy. It's easy to remember. And it's wrong. This number comes from bodybuilding culture, not from science. It overshoots the research by about 40% — and the excess doesn't build more muscle. It gets burned as fuel, converted to fat, or flushed out as urea. Worse, it may cause harm. Too much protein doesn't help. The "1 gram per pound" rule pushes most people to roughly 2.2 g/kg/day. At that level, the evidence raises real concerns. A prospective cohort of over 9,000 adults found the highest protein quartile was associated with a 3.5-fold higher rate of kidney hyperfiltration and a 1.3-fold higher rate of rapid kidney function decline. (Nephrology Dialysis Transplantation, 2020) Sustained high protein intake has been associated with intraglomerular hypertension, glomerular injury, and proteinuria. (JASN, 2020) A meta-analysis of 11 cohort studies found higher animal protein intake was associated with a 9% increased risk of cardiovascular mortality. (European Journal of Epidemiology, 2020) And for all that risk? Zero additional muscle. Surplus amino acids are simply deaminated in the liver and either oxidized for energy or converted to glucose or fat. (New England Journal of Medicine, 2024) Too little protein doesn't work either. In the largest meta-analysis of protein and lean body mass, adults under 65 consuming 1.2–1.59 g/kg/day during resistance training had a small, non-significant effect on lean mass (SMD = 0.15, p = 0.07). At 1.6 g/kg/day or above, the effect nearly doubled and became clearly significant (SMD = 0.30, p < 0.01). (Journal of Cachexia, Sarcopenia and Muscle, 2022) This matters because muscle is not cosmetic — it is a longevity organ. Low skeletal muscle mass is associated with a 57% increased risk of dying from any cause. (PLoS One, 2023) Muscle protects against falls, metabolic disease, and functional decline. Underfueling it has real consequences. So what's the right number? Two of the largest analyses ever conducted — over 100 randomized controlled trials combined, spanning thousands of participants — tested every protein intake level and asked: at what point do you stop gaining more muscle? The answer: 1.6 grams of protein per kilogram of ideal body weight per day. That's the ceiling. Above it, additional protein produced zero further gains in lean body mass. The dose-response curve flattens completely. (British Journal of Sports Medicine, 2018; Journal of Cachexia, Sarcopenia and Muscle, 2022) Below it, gains are suboptimal. At it, your muscles have everything they need. One number. Backed by over 100 trials. Not 1 gram per pound. Not "as much as possible." Just 1.6. For those of you who counsel on this — what do you recommend? #ClinicalNutrition #PreventiveMedicine #LongevityMedicine #Sarcopenia #PhysiciansOfLinkedIn
-
Nicholas Cohen, MD shared thisYou wore an N95 last week when the sky turned orange. PM2.5 hit 87. This morning you rode the train, where the platform averages 139. You didn't wear it. THE SUBWAY IS MORE DANGEROUS THAN THE WILDFIRE NYC subway platforms average 139 µg/m³ — roughly 14× street-level air. Inside the cars, 99. The wildfire that made the news hit 87, and lasted a day. The commute is 250 days a year. A 2025 randomized trial put 80 healthy adults on a subway platform two hours a day for five days against an office control. Lung function fell — FEV1, FEV1/FVC, peak flow. Airway inflammation rose. The antioxidant enzyme GPX1 dropped. Healthy 24-year-olds (Part Fibre Toxicol 2025). THE KITCHEN IS MORE DANGEROUS STILL Pan-frying or air-frying dinner drives kitchen PM2.5 to 150–2,000+ µg/m³. The high end is more than twenty times the wildfire day. It happens every night, in an enclosed room, and the haze afterward carries as much exposure as the cooking. Cook and go to bed without ventilation and you breathe it for hours. THIS ISN'T JUST LUNG IRRITATION Within hours of a PM2.5 rise: systolic blood pressure climbs, heart rate variability falls, flow-mediated dilation drops, CRP and IL-6 rise, and coagulation shifts prothrombotic. The AHA links these to acute coronary syndromes, stroke, heart failure, and arrhythmia (Circulation 2020; JACC 2021). WHERE THIS ARGUMENT IS WEAKEST The subway trial ran at 193 µg/m³ with people seated on a platform for two hours — not a 20-minute commute. The one NYC-based study found no cardiopulmonary effect beyond reduced HRV and more symptoms (Part Fibre Toxicol 2024, NYC cohort). And no trial has shown that a mask or a filter prevents a cardiac event. Note where the gap sits. Short-term PM2.5 rises are associated with ~1–2% higher risk of MI, stroke, and heart failure per 10 µg/m³ (JACC 2020). The hazard is well characterized. What's missing is the intervention arm. WHAT SURVIVES THE CAVEATS Two exposures, measured, daily, and larger than the one everybody masked for. RUN A HEPA FILTER WHILE YOU COOK. Keep it running afterward — the decay phase matters as much. The single highest-yield intervention for most NYC residents, with the strongest randomized support of anything here (JACC 2015; Circulation 2017). WEAR AN N95 ON THE SUBWAY. Not outdoors: below roughly 75 µg/m³ the work of breathing through the mask cancels the filtration benefit (Sci Total Environ 2023), and Manhattan's air is around 8. The mask goes on at the turnstile, not at the front door — and a hot platform in July is a real cost, which is why the threshold matters. The orange sky was one day. Dinner and the commute are most of them. We masked for the one exposure we could see. What's the case for not masking for the ones we can't? #AirQuality #Longevity #PreventiveMedicine #PublicHealth #NYC #EvidenceBasedMedicine
-
Nicholas Cohen, MD shared thisHealthy people take metformin to slow aging. The biology says it should work. No trial has shown that it does. It's the most talked-about drug in longevity medicine — originally a diabetes therapy, now taken off-label as an anti-aging bet. THE BULL CASE It hits nearly every hallmark-of-aging pathway — activates AMPK, inhibits mTOR, lowers inflammation, suppresses senescence. It extends lifespan in worms, flies, and some mouse strains. And a widely cited observational study found metformin-treated diabetics outlived even non-diabetics. So what's the problem? NO TRIAL HAS SHOWN BENEFIT IN HEALTHY PEOPLE — AND THE ADJACENT EVIDENCE IS UNFAVORABLE → DPP/DPPOS (21-yr follow-up): no benefit in all-cause, cancer, or CV mortality (Lee, Diabetes Care 2021); none in frailty (Hazuda, J Gerontol A 2021); none in cognition (Luchsinger, Diabetes Care 2017). → MASTERS (65+): metformin blunted muscle gains from resistance training (Walton, Aging Cell 2019). → MET-PREVENT (frail, ~80): no gain in walk speed; 34% hospitalized vs. 8% on placebo; poorly tolerated (Witham, Lancet Healthy Longevity 2025). A frail population — but a real safety signal. THE OBSERVATIONAL DATA IS WEAKER THAN IT LOOKS The "diabetics outlived non-diabetics" finding is almost certainly bias — immortal time bias (users had to survive long enough to get the prescription) and confounding by indication (metformin goes to healthier diabetics). With rigorous matching including discordant twin pairs, one Danish study found higher mortality in treated diabetics vs. non-diabetic controls (IRR 1.52, 1.37–1.68; Keys, Int J Epidemiol 2022). THE ANIMAL DATA ISN'T A SLAM DUNK The NIA ITP found metformin alone did NOT significantly extend lifespan by log-rank test (Strong, Aging Cell 2016). A 2024 Gehan-test reanalysis found a modest benefit in males only (Jiang, GeroScience 2024). Metformin plus rapamycin did extend lifespan — a different story. IT MAY BLUNT EXERCISE BENEFITS A 2026 meta-analysis found metformin attenuated VO₂peak gains by ~1 mL/kg/min vs. exercise alone — roughly a 40% cut to the training response — with no offsetting improvement in glucose, insulin, or HbA1c (Etayo-Urtasun, EClinicalMedicine 2026). TAME — designed to be the definitive test in older non-diabetic adults — still has incomplete enrollment as of mid-2026. THE BOTTOM LINE The biology is interesting. The clinical evidence in healthy people isn't there — and there are real trade-offs for people who exercise. The interventions with decades of evidence haven't changed: exercise, diet, sleep, not smoking, social connection. Before reaching for the drug, ask the real question: are you even the person it was studied in? That data comes from people with metabolic disease. If you're metabolically healthy, it isn't you. Would you take metformin for longevity? #Longevity #Metformin #AgingResearch #EvidenceBasedMedicine #AntiAging #Healthspan
-
Nicholas Cohen, MD shared thisYour fasting glucose is 90. Your A1c is 5.5. Your doctor says the metabolic panel looks perfect — see you next year. That “perfect” can be years out of date. And the one number that would tell you isn’t on the panel. Here’s the trap. Fasting glucose and A1c are lagging indicators. When insulin resistance sets in, the pancreas compensates by pumping out more insulin, and that extra insulin drags glucose back down into the normal range. Your glucose reads normal not because nothing is wrong — but because your body is working harder and harder to keep it looking that way. By the time glucose or A1c finally drifts up, that compensation is already breaking down. That isn’t early disease. It’s the late stage of something that started years earlier. What moves first is fasting insulin — captured by HOMA-IR, a simple calculation from fasting glucose and fasting insulin. In non-diabetic adults, higher HOMA-IR tracks with higher all-cause mortality, and the association is strongest in people at a normal weight — exactly the people a normal glucose reassures (Ausk et al., Diabetes Care 2010). Risk sat lowest at the bottom of the range and climbed from there (Zhang et al., Bioscience Reports 2017). HOMA-IR isn’t a flawless test — fasting insulin assays vary and there’s no single agreed cutoff. But a fasting insulin costs a few dollars, and it sees what a normal glucose can’t. If you’re serious about metabolic health, this is the number I’d want years before the standard panel says a word. #InsulinResistance #PreventiveMedicine #MetabolicHealth #Longevity
Nicholas Cohen, MD replied to a comment
10h
Laird Doman Exactly. That was one of the biggest surprises in our study too. Our $23 Etekcity scale was the second-most accurate of the 7 we tested for body-fat percentage, while several $300–$600 devices performed substantially worse.
The caveat is that agreement in one individual doesn’t necessarily tell you how the device performs across a population, which is why we tested all 7 scales in the same 19 people against DEXA. But your result is another good example of why price and apparent sophistication aren’t reliable proxies for accuracy.
Nicholas Cohen, MD replied to a comment
13h
Jacob Kelly MD, MHS Thanks, Jacob. I’m really looking forward to seeing your N=2 results. I’d be especially interested in how the wearable estimates track against your formal CPET values over time. That kind of real-world comparison is exactly what I think is missing from a lot of the wearable VO₂ max discussion.
-
Nicholas Cohen, MD liked thisNicholas Cohen, MD liked thisThe AI Conversation Isn't Just About Capability. It's About Responsibility. The people building these systems aren't only solving technical problems. They're making decisions that could shape how we work, learn, and create for decades. That's why this moment calls for more than innovation. It calls for thoughtful leadership. The goal shouldn't be to build AI as fast as possible. It should be to build it in ways that expand human potential while earning public trust. That's the challenge worth solving. #Leadership #ArtificialIntelligence #Innovation
-
Nicholas Cohen, MD liked thisNicholas Cohen, MD liked this🎯 A patient says: "My watch says VO₂max 48. Your test said 41." Here's why — and how to answer. No smartwatch measures oxygen. Two families: Family 2 — never watches you exercise. Polar Fitness Test, Fitbit (no GPS), Whoop. Resting HR + HRV + demographics → population lookup. Family 1 — watches you run. Garmin, Apple, Coros, Suunto. This one matters. 🔍 HOW FAMILY 1 WORKS 1️⃣ Logs HR against GPS pace. 6:00/km → HR 140. 5:00/km → HR 160. 2️⃣ Filters junk. Firstbeat tests HR–speed correlation within each segment and discards downhills, stops, cardiac drift. 3️⃣ Extends the line to assumed HRmax → theoretical max speed. 4️⃣ Converts speed to oxygen cost. Pace and pulse, extrapolated. That's it. 🏆 WHICH BRAND IS BETTER? Not taste — validation. 🥇 Garmin. Only algorithm with published methodology AND repeated independent validation. Three labs, three models, six years apart: MAPE 7.05%, 7.3%, 7.2–7.9%. That convergence is the argument. 🥈 Apple. Independently validated but weaker. S9/Ultra 2 vs COSMED: underestimated 6.07 mL/kg/min, MAPE 13.3%, LoA −6 to +18. ❓ Coros, Suunto, Samsung. No independent peer-reviewed validation. Similar method ≠ similar accuracy. 📊 THE ERROR HIDES IN THE AVERAGE Pooled bias: −0.09 mL/kg/min. Looks superb. It isn't — limits of agreement run −9.92 to +9.74. And the error is directional. Forerunner 245: ▪️ Moderately trained → MAPE 2.8–4.1% ▪️ Highly trained → underestimated 6.3 mL/kg/min, ICC 0.34–0.41 Regression toward the population mean. Validity isn't a property of the watch — it's a property of where your patient sits relative to its training data. Clinical populations aren't at a tail. They're off the distribution. ⚠️ THE HRmax PROBLEM Step 3 rests entirely on assumed HRmax. Firstbeat's own paper: 15 bpm error → 7–9% VO₂max error. Most devices use 220−age. SD ≈ 10–12 bpm. Baked in before the sensor does anything — and worst in beta blockade, chronotropic incompetence, post-anthracycline cardiotoxicity. For scale: CPET carries ±5%. Submaximal equations, 3.7–4.5 mL/kg/min SEE. A watch isn't competing with CPET. It's competing with a step test. 🩺 WHEN THEY BRING YOU THE WATCH 1️⃣ Don't dismiss it. They're tracking their health. 2️⃣ Name what it measured: pace and pulse, never oxygen. 3️⃣ Reframe: the number is unreliable, the trend isn't. Same device, same conditions, watch the direction. 4️⃣ Explain what CPET added — VE/VCO₂, VT1/VT2, O₂ pulse, breathing reserve. The watch answers how much. CPET answers why. "Why" is what brought them to us. ❓ How do you handle it when wearable data contradicts the test? #CPET #VO2max #Wearables #Whoop #Garmin #Apple #Coros #Suunto #Samsung #Firstbeat
-
Nicholas Cohen, MD liked thisNicholas Cohen, MD liked thisApple relaunches VO₂ max algorithm and just made the most important longevity metric on your wrist meaningfully more accurate. With the watchOS 26 update line (Sep 2025 → 26.4 in March 2026), Apple quietly rewrote how Apple Watch calculates VO₂ max and shipped something most fitness wearables don't: the ability to retroactively update your historical readings with the new algorithm. Here's why this matters VO₂ max, the maximum amount of oxygen your body can use during exercise, is one of the strongest predictors of all-cause mortality we have. As Dr. Peter Attia puts it: "Going from 'low' (bottom 25th percentile) to 'below average' is a 50% reduction in mortality. Going from 'low' to 'above average'? About 60–70%." There is only very few other lifestyle intervention that moves the mortality needle that hard. And here's the silent problem: Apple Watch had a documented gap to lab-grade VO₂ max for years. → Apple's own published validation paper claims accuracy within 1.2–1.4 mL/kg/min (~4%). → Independent PLOS One peer-reviewed study (Lambe et al., 2025, Apple Watch Series 9 + Ultra 2 vs. indirect calorimetry) found a mean underestimation of 6.07 mL/kg/min, SD 6.22, MAPE 13.31%. For a generally fit user, the independent gap is the difference between "average" and "excellent" between thinking you have work to do and thinking you're already there. What changed in watchOS 26: → New estimation algorithm, trained on more data, more contexts → User-reported lab comparisons after the update show the gap often shrinking to ~1 mL/kg/min (anecdotal, not yet replicated in peer-reviewed work) → A new Health-app feature lets you recompute your historical VO₂ max readings so your trend line actually reflects reality → watchOS 26.4 (March 2026) further refined the model This is the part that's underrated: Most wearables update silently and silently change the truth of your past data. Apple let users opt in to recalculating their history. That's a small UX choice with a big trust implication: your training trend should not get rewritten by a software update without you knowing. What's still imperfect (because honesty matters): - Available only on Apple Watch Series 3 or later, with classifications shown only for users 20+, and a supported range of 14–65 mL/kg/min (per Apple Support 108790). - Estimate only updates during Outdoor Walk, Outdoor Run, or Hiking workouts. Indoor cycling, treadmill runs, gym sessions, still don't count. - The algorithm factors in age, sex, weight, height, and medications affecting heart rate. A stale Health profile silently degrades the number, tall or heavy users with outdated data can be underestimated by up to 20%. - Wrist fit, sensor cleanliness, and tattoos materially affect the heart-rate signal the estimate is built on. - Some users reported drops of 3–5 points overnight after the .4 update.
-
Nicholas Cohen, MD liked thisNicholas Cohen, MD liked thisI got home from walking my dog last night and was shocked to see that my watch thinks I’ve fallen off a cliff. It wasn’t a fall alert. It’s my VO2 max, aka the metric it treats as the valuation of my cardiovascular worth. Over the last 3 months, it’s dropped eight points. Eight! That's not a subtle dip. It’s a 16% drop. So what’s changed in my routine since spring? Nothing. Same dog on the same walk taking the same route. Same guy waking up and doing the same workouts, eating the same foods. So what did change? It's July, so our evening walk now takes place in a sauna. The watch doesn't account for that though. It knows my heart rate and my pace. And in this heat (94 tomorrow), my heart is working overtime just to keep me cool, pushing blood to my skin, sweating, spending effort it didn’t have to when it was 50 outside. Same walk, higher heart rate. The algorithm turns on the check engine light. But the engin’s fine. The engine is just in Boston in July. It’s the same reason we never trust a single-view X-ray. One view can look pristine while the second view shows the consolidation. But now instead of finding a teaching xray, the lesson that I want every intern to learn as they start work this month is probably sitting on their wrist. A number can be perfectly accurate and completely wrong if you don't know the context when it was measured. It's entirely about perspective and context changes everything. Eight points. Ask me again in October. Tucker, for the record, is thriving and couldn't seem to care less what temperature it is. Every walk is the best walk of his life.
Experience & Education
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content