Ryan Williamson, MD

A 45-year-old executive sat across from me with a physical from four months earlier. Every number on it was flagged "within normal limits." His physician told him he was healthy. He wasn't. He was six to eight years into a disease process nobody had looked for. Here is what "normal" actually means. A reference range is built by measuring a few thousand people who happened to walk into a lab, then drawing a line around the middle 95% of them. That's it. It is a description of a population — and in the United States, that population is metabolically sick. Roughly 9 in 10 American adults have at least one marker of metabolic dysfunction. So when your result comes back "normal," what you've actually been told is: *you resemble the average person in a country where the average person is unwell.* Normal is a bell curve. It was never a target. What I order instead, and why: ApoB, not just LDL-C. LDL cholesterol measures the cargo. ApoB counts the actual particles — and it is the particle that lodges in an artery wall. Two people can have identical LDL-C and meaningfully different risk. ApoB tells you which one you are. Lp(a), once. Largely genetic, barely moved by diet or statins, and a meaningful independent driver of early cardiovascular events. Measure it one time in your life. Most people never have. If it's high, it changes how aggressively you treat everything else — and it means your children should be tested. Fasting insulin. Glucose is a lagging indicator. Insulin rises for years while the body compensates, and glucose stays obediently "normal" the entire time. By the time glucose drifts, you've been insulin resistant for the better part of a decade. Insulin resistance is not just a diabetes problem — it is a brain problem. The pattern of glucose hypometabolism in the Alzheimer's brain is not incidental. Blood-based neuro biomarkers — NfL, GFAP. Neurofilament light for axonal injury, GFAP for glial activation. Ten years ago these lived in research labs. They're clinically available now. Almost nobody orders them. Now the part most physicians won't say out loud: none of this is a knowledge failure. Your doctor knows about ApoB. It's a structural failure. A 7-to-15-minute visit reimbursed by insurance cannot order eleven tests, interpret them against your trajectory rather than a reference range, and then sit with you for an hour explaining the plan. The system isn't ignoring your future. It isn't built to see it. That's the entire reason I stopped practicing inside it. If you take one thing from this: stop asking your physician "are my labs normal?" Ask "what did you not order, and why?" The answer to the second question is where your actual risk lives.

Most consumer genetic testing is not medicine. It's a report. The distinction matters: a result is only useful if it changes a decision. Most of what those panels return — carrier status you'll never act on, trait predictions, ancestry-adjacent risk scores with no intervention attached — changes nothing. And the handful of variants that genuinely do change management are exactly the ones that should never arrive as a PDF in your inbox with no one to interpret them.

“The medical standard of care is the floor, not the ceiling." I say some version of that sentence to almost every one of my patients, and it is the single most misunderstood idea in preventive medicine. Standard of care is not a synonym for best available. It's the minimum defensible practice. The way I actually think about testing is a ladder with four rungs. Tier 1 — Standard of Care. Blood pressure, lipids, A1c, cancer screening at guideline intervals. Non-negotiable. Anyone skipping this rung has no business paying for what’s above it. Tier 2 — Guideline-Supported. Endorsed in the literature and by professional societies, inconsistently ordered in practice. ApoB, Lp(a), hs-CRP, fasting insulin, Cystatin C, homocysteine, omega-3 index. None of these are exotic. Most people reading this could have them drawn next week. Tier 3 — Strong Emerging. Real evidence, not yet consensus. Advanced imaging, continuous glucose data in non-diabetics, blood-based neuro markers. Useful in the right patient with someone who knows how to read them. Tier 4 — Frontier. Genetics, epigenetic aging measures, the newest biomarkers. Genuinely informative and genuinely easy to misuse. The failure I see most often isn't people who tested and got bad news. It's people who bought a Tier 4 test while skipping Tiers 1 and 2 entirely. The ladder only works from the bottom up.

The polarized-training literature is one of the few areas in exercise science where the evidence base is unusually consistent across decades and populations. A core finding: across studies of endurance athletes at multiple levels — from competitive amateurs to Olympic-caliber performers — the training distribution that most consistently predicts performance gains is roughly 80% low-intensity, 20% high-intensity, with comparatively little time spent in the moderate zone in between. This has held up across running, cycling, rowing, and cross-country skiing cohorts, and across multiple independent research groups, which is part of why it's more durable than most single-study fitness claims that circulate online. What it doesn't tell you: almost all of this data comes from highly trained endurance athletes, not from sedentary adults starting a program, and not from a longevity or all-cause-mortality lens. Low-intensity work builds the aerobic base efficiently with low recovery cost. High-intensity work drives the ceiling-raising adaptations. The middle zone taxes recovery without maximizing either. The specific 80/20 ratio is athlete-derived, and I treat it as a strong starting heuristic for my patients. Does it change my practice? Yes. It's the reason I actively steer patients away from "moderate and constant" as a default, even when moderate feels more virtuous or measurable in a fitness app. For someone starting from a sedentary baseline, the honest translation is ”make your easy days genuinely easy and your hard days genuinely hard," which most people have never actually done.

A composite from my practice. Details changed, pattern exactly as it presents. 52-year-old executive, on a GLP-1 for fourteen months. Down 38 pounds. Labs improved across the board — glucose, triglycerides, blood pressure, all better. His prior physician called it a major win and had no further recommendations. What we found when we looked at what actually changed: — A DEXA scan showed roughly a quarter of the total weight lost was lean mass, not fat. That's materially worse than what's achievable with the same weight loss paired with resistance training. — Grip strength had declined measurably from a baseline taken two years earlier — a marker I track partly because it correlates with all-cause functional decline, and partly because it's the cheapest, fastest proxy for whole-body muscle quality. — He described new fatigue on stairs and a decline in his usual weekend hikes that he'd attributed to "getting older," not the medication. — He had done ZERO resistance training in the fourteen months. Nobody had told him it mattered while he was losing weight this efficiently. None of this is an argument against GLP-1s. The metabolic improvement was real and worth having. It's an argument against treating rapid weight loss as automatically synonymous with improved health, when a meaningful fraction of what's being lost is the tissue that keeps you functionally capable as you age. The fix wasn't complicated: resistance training 2-3x/week and roughly 0.7 grams of protein per pound of lean body weight. Muscle preservation during rapid weight loss isn't automatic. It has to be deliberately trained for at the same time.

Everyone will tell you to move more. Almost nobody will tell you that *how hard* matters more than *how often* — and that getting it wrong is worse than doing nothing. Training constantly at moderate intensity doesn't split the difference between easy and hard. It captures the cost of both and the adaptive benefit of neither. Your body doesn't grade on effort. It adapts to the specific stimulus you give it, or it doesn't adapt at all.

Most of my patients who "work out consistently and still aren't getting results" are training at exactly the wrong intensity, every single day. Not too little. Not too lazily. Wrong. Here's the pattern I see constantly: someone runs 5-6 miles at a pace that leaves them "pretty tired," most days of the week. Heart rate sits around 75-85% of max the entire session. They feel like they worked hard. They did work hard. And a year later their resting heart rate hasn't substantially improved, their VO2max is flat, and they've quietly lost muscle they'll spend the next few years trying to get back. This is Zone 3-4 — the moderate intensity trap. It is hard enough to generate real fatigue and a real load, but not hard enough to drive the specific adaptation that zone is supposedly training. It's the exercise equivalent of revving an engine in neutral. The physiology, plainly: aerobic adaptation (mitochondrial density, capillary growth, fat oxidation) is trained most efficiently at genuinely easy effort — Zone 2, where you could hold a conversation, not gasp through one. Anaerobic and VO2max adaptation is trained at genuinely hard effort — Zone 5, short intervals near or at max heart rate. The zone in between produces some of each and optimizes neither. Elite endurance athletes train this way deliberately: roughly 80% of volume easy, 20% hard, almost nothing in the middle. It's called polarized training, and it isn't a trend — it's been the dominant model in exercise physiology for two decades. Most high performers do the reverse. They don't have time to "waste" on an easy Zone 2 session, so they push every session into the moderate range to make it "count." The result is accumulated fatigue without proportional adaptation — training volume that produces tiredness, not fitness. What I actually prescribe to my patients: — Strength, 2-3x/week. Non-negotiable, not optional cardio-adjacent extra. — Zone 2, 3-4x/week, 150-200 minutes total. Genuinely easy. The talk-test is the free version: if you can't hold a full sentence, you're not in Zone 2. — Zone 5, 1-2x/week, 16-25 minutes of actual work (not counting warm-up/cooldown). Short, hard, done. — Mobility, daily. Five minutes counts. — Avoid Zone 3-4 as a default. It's fine occasionally. It should not be where most of your training volume lives. The strange part for anyone tracking this in a wearable: the app doesn't tell you your training is misallocated. It tells you calories burned and minutes active, and by both of those metrics, moderate-intensity training looks like the most productive thing you did all week. It's the metric that's misinforming you, not your effort. If your training has plateaued despite consistency, the fix usually isn't more. It's redistributing the same hours toward the two ends and getting out of the middle.

Perhaps the most reversible driver of cognitive decline we have is also the one almost nobody is measuring. Everything I've shared this week has been about mechanism — why insulin resistance quietly degrades brain fuel delivery years before any standard lab flags it. Most of the risk factors that show up in a longevity conversation are frustratingly slow to move, or not really modifiable at all. Genetics doesn't change. Chronological age doesn't reverse. Lp(a), for most people today, barely budges with anything short of an emerging drug class still in trials. Insulin resistance is different. It is one of the few meaningfully reversible contributors to cognitive risk on the entire list — and it moves on a timescale you can actually observe. What "reversible" looks like in practice, from repeat testing at 90 days: fasting insulin dropping by a third to a half in patients who take resistance training seriously three times a week. HOMA-IR normalizing in people who were clearly insulin resistant at baseline. Post-meal glucose excursions (when we track them) shortening back toward a normal return-to-baseline window. Why it moves that fast, mechanistically: skeletal muscle is your body's largest site of glucose disposal. Building and actually using more of it gives insulin more places to do its job efficiently, which is a large part of why resistance training outperforms cardio alone for this specific number. Protein at the start of a meal blunts the glucose spike that follows. A short walk after eating activates muscle glucose uptake independent of insulin entirely — one of the few genuinely "free" interventions in medicine. None of this requires a prescription to start. Reversing insulin resistance is not a guarantee against cognitive decline. It's one lever in a disease process with more than one cause. But it is a lever you can cheaply pull yourself with a number that will tell you within three months whether it worked — which is a genuinely rare combination in preventive medicine.

I used to order a fasting glucose, see it in range, and move on. For years. Not out of laziness — out of training. That's the standard-of-care screen, and it's genuinely what medical education teaches you to trust. What changed my mind wasn't a lecture. It was ordering fasting insulin on myself out of curiosity more than concern, and watching a number sit in a range I wouldn't have predicted, next to a glucose that looked completely unremarkable. If it surprised me — someone who reads this literature for a living — I stopped assuming patients with normal glucose had nothing worth investigating further. That's an uncomfortable thing to admit in public, because the honest version is: the gap between what I was trained to check and what actually predicts risk isn't a knowledge problem unique to other physicians. I had the same blind spot until I went looking for it in my own bloodwork. Building my practice around a 90-minute intake instead of a rushed 10-minute visit isn't really about time as a luxury. It's about time as the only thing that makes it possible to ask the right questions instead of stopping at the first normal number — on a patient, or, it turns out, on myself.

Well-established: insulin receptors are densely expressed in the hippocampus and other memory-relevant regions. Insulin resistance is associated with reduced hippocampal glucose uptake on PET imaging. Type 2 diabetes is an independent, replicated risk factor for both vascular dementia and Alzheimer's disease across large longitudinal cohorts. None of that is controversial. Genuinely still open: whether insulin resistance is a meaningful *cause* of Alzheimer's pathology specifically, versus a strong risk marker that travels alongside other causal factors — vascular disease, inflammation, genetics — that are doing real independent work. The mechanistic story is compelling. Causal proof is harder to establish for any single contributor to a disease this multifactorial. Where I land: I don't need the causal question fully settled to act on this. Insulin resistance is common, it's measurable cheaply, it's meaningfully reversible in 90 days with intervention, and it plausibly contributes to cognitive risk through more than one pathway even if the exact weight of "brain-specific" versus "vascular" contribution isn't nailed down yet. That's the actual argument for treating it aggressively.

Pretty excited about this. Thanks @elonmusk !

A composite from my practice. Details changed, pattern exactly as it presents. A man in his late 30s, no diabetes history, unremarkable annual labs, asked me to put him on a continuous glucose monitor because "everyone in my world is wearing one." Fair request. I said yes, on one condition — we'd use it as a diagnostic instrument, not a wellness accessory, and we'd interpret it properly instead of just staring at the app. Two weeks of data showed something his annual physical never would have: a fasting glucose that looked fine on paper, but a pattern of glucose spikes after meals that took nearly three hours to return to baseline — normally that should happen within about ninety minutes. That's not a diabetes-range number. It's a pattern that shows up years before diabetes-range numbers do, and it corresponds to exactly the kind of chronic post-meal insulin surge that, sustained for years, is implicated in the hippocampal fuel-delivery problem I wrote about Monday. We paired the CGM data with fasting insulin, which his physical had never ordered. It was elevated. HOMA-IR confirmed real insulin resistance sitting quietly behind a completely unremarkable annual bloodwork panel. Here's the honest part, because I don't think CGMs deserve blanket praise: for someone like this, worn short-term and interpreted by a clinician who can pair it with the right labs, it was genuinely useful — it caught a pattern no single fasting number would have shown. For most people wearing one indefinitely with no clinical interpretation, watching a number bounce around without context, it's expensive theater. The device isn't the intervention. What you do with the data is. His plan wasn't a supplement. It was resistance training three days a week, protein at the front of every meal instead of the end, and a 10-minute walk after dinner — the single cheapest post-meal glucose intervention that exists. Repeat CGM and fasting insulin at 90 days showed real, measurable improvement. The lesson isn't "everyone needs a CGM." It's that a tool is only as good as the question it's being used to answer — and most people wearing .

"Alzheimer's is type 3 diabetes" gets thrown around online as either a mic-drop fact or a debunked myth, and both versions are wrong in the same way: they're treating a nuanced mechanistic finding as a slogan. What's true: insulin resistance independently degrades glucose delivery to the hippocampus, and glucose hypometabolism is one of the earliest measurable changes in the Alzheimer's brain — years before symptoms. What's not true: that fixing your insulin sensitivity prevents Alzheimer's, full stop. Dementia has too many contributing pathways for any single mechanism to own that claim. The honest version isn't nearly as quotable.

Alzheimer's disease has been called "type 3 diabetes" for over a decade. Most people who've heard the phrase have no idea how literal it is. Your brain is roughly 2% of your body weight. It burns about 20% of your resting energy. It cannot store meaningful fuel reserves the way muscle or fat can — it needs a continuous, well-regulated supply of glucose getting inside neurons, minute to minute, for as long as you're alive. Insulin is part of how that delivery happens. And the region of the brain most dependent on that signaling — the hippocampus — is exactly where short-term memory is formed and exactly where Alzheimer's pathology begins. So here's the sequence, and it matters that you see it as a sequence, not a coincidence: Insulin resistance develops. The pancreas compensates by producing more insulin, which keeps fasting glucose looking normal for years — sometimes a decade or more — while the compensation itself quietly does damage. Eventually the compensation fails and glucose starts to drift, which is the point most people find out they have a "metabolic problem." But the earliest, most consistent imaging finding in Alzheimer's disease isn't plaque. It's regional glucose hypometabolism — parts of the brain failing to take up fuel efficiently — and it's measurable on PET imaging years to decades before the first symptom. That is the honest, careful version of "type 3 diabetes." Not that Alzheimer's *is* diabetes, as there are many other risk factors. But that the same failure of insulin signaling driving type 2 diabetes in the body is independently degrading fuel delivery in the brain — often before a single lab value outside a specialty test would ever flag it. The caveat I want to be precise about, because this claim gets oversold constantly: insulin resistance is not sufficient or necessary to cause Alzheimer's. Dementia is multifactorial — genetics, vascular health, sleep, inflammation, and more, all contribute. What the evidence supports is narrower and still enormously actionable: insulin resistance is one of the most common, most measurable, and most reversible contributors to cognitive decline risk that we have. That's a different claim than "sugar causes Alzheimer's," and it's the one I can actually defend. What changes in how I practice because of this: — I order fasting insulin alongside fasting glucose on essentially every patient over 35, not just the ones with a diabetes risk profile. Glucose alone misses the entire compensation phase. — I calculate HOMA-IR rather than eyeballing either number alone. — When it's elevated, the first-line intervention isn't a drug. It's resistance training (skeletal muscle is your body's largest glucose disposal site), protein-forward eating, and fixing sleep — a single short night measurably worsens insulin sensitivity the next day. — I retest in 90 days, because unlike most things on a longevity panel, this number actually moves — and moves fast enough to prove the intervention worked before a patient loses faith in it. If you remember one thing from this: your fasting glucose being normal tells you almost nothing about whether this process has already started. It's the test that looks fine right up until the moment it doesn't — and by then, the brain has already been running on a failing fuel-delivery system for years.

Spend some time today reflecting on your week. What did you accomplish? What could you have done better? Did it move you closer to your goals. Set a target for next week, and repeat. When you look back in another 6 or 12 months your progress will shock you.

An especially important thought for the weekend: sleep regularity may be more important than duration. Meaning—try to go to bed and wake up around the same time every day, even on the weekends. People who do this have a lower risk of dying earlier than those who interrupt that pattern. pubmed.ncbi.nlm.nih.gov/3773…

Uncomfortable admission for a guy who now writes about sleep for a living: I spent most of residency treating sleep as the thing you cut first. Post-call, running on caffeine and adrenaline, telling myself the deficit would even out eventually. It doesn't even out. It compounds. Nobody teaches you that in training, because training is explicitly designed around the opposite assumption — that willpower can substitute for rest indefinitely. It can't. It just gets better at hiding the bill. What actually changed my mind wasn't a study. It was watching how consistently the sickest patients I admitted *also* had the worst sleep for years beforehand — not as a coincidence riding alongside their diagnosis, but as a plausible contributor. When you see the pattern enough, you can't un-see it. Building this practice forced me to actually live what I now tell patients, because the alternative — advising sleep hygiene from a state of chronic short sleep myself — felt like the kind of hypocrisy patients can smell in about four seconds. So I fixed my own sleep architecture before I ever put "sleep" on a first-visit intake form. Alcohol first (it was doing more damage to my sleep than I wanted to admit), then a consistent wake time, then finally addressing the light and screen exposure that was pushing my own sleep onset later than I realized. None of that required a device or a supplement stack. It required admitting the thing I already knew and had been ignoring on purpose, because ignoring it was more convenient than the alternative. If you're the kind of person who treats rest as the reward for finishing everything else — I get it. I ran that program for over a decade. I just don't think it's true anymore, and I'm not willing to pretend otherwise for the sake of looking unbreakable.

Been getting asked constantly whether sleep trackers (Oura, Whoop, Apple Watch) are actually measuring what they claim. Here's the honest verdict, not the marketing one. What they're good at: total sleep time and consistency. Accelerometer- and heart-rate-based trackers correlate reasonably well with polysomnography (the gold-standard sleep-lab study) for overall duration and for flagging a genuinely disrupted night. If your ring says you slept five broken hours, believe it. What they're not good at: sleep staging. The "deep sleep" and "REM" percentages that show up in your app every morning are inferred from heart rate variability and movement, not from EEG — which is what actually defines sleep stages clinically. Validation studies comparing consumer wearables to polysomnography consistently find the stage-by-stage accuracy is modest at best, and worse specifically at detecting deep sleep, which is the stage that matters most for the glymphatic clearance mechanism I shared on Monday. So the number people fixate on most — "I only got 40 minutes of deep sleep" — is the number these devices are least equipped to give you reliably. What they cannot do at all: diagnose sleep apnea. Some newer devices flag possible disrupted breathing and that's a legitimate, useful nudge to get evaluated. None of them replace a sleep study. I've now seen this pattern enough times that it's not an edge case: a patient with a genuinely dangerous, undiagnosed apnea whose wearable reported "good" sleep scores for years, because the metric it's actually built to measure — movement and heart rate pattern — doesn't capture repeated micro-arousals from airway collapse. Does this mean don't wear one? No. I certainly do. But trends over months, and duration/consistency specifically, are the genuinely useful self-data. Just hold the nightly "sleep score" the way you'd hold a home blood pressure cuff reading — directionally useful, not diagnostic — and don't let a green checkmark talk you out of a sleep study your symptoms are asking for.