Hi {{ first_name| strip | default: there}},

Most people optimizing their sleep are optimizing the wrong thing.

They're chasing eight hours. Proud of a 90% sleep score. Comparing nightly averages with their partner like it's some kind of wellness Olympics. (My Oura score beat yours. Sit down.)

A wave of studies over the past year points somewhere else entirely: not at how much you sleep, but at the shape of how you sleep.

This single metric — called REM latency (how long it takes your brain to reach its first dream cycle) — independently correlates with amyloid burden, p-tau levels, and lower BDNF (brain-derived neurotrophic factor which repairs and protects neurons — fertilizer for your brain).

Three Alzheimer's biomarkers. And almost no wearable shows it to you.

There is one wearable that earns its place for brain health tracking more than the others. I'll tell you which one — and what to actually look for — in the protocol section below.

The most commonly tracked sleep metric is total sleep time. The second is sleep efficiency — the ratio of time asleep to time in bed. Most wearables mash these into a composite "sleep score" and call it a night.

Both are useful. Neither is what the latest research is flagging.

Two studies in the past year have identified REM sleep architecture as independently predictive of Alzheimer's biomarkers — and the mechanism is very specific.

The first study, published in Alzheimer's & Dementia (January 2025), looked at REM latency: the time from sleep onset to your first REM episode. People with delayed REM sleep had 16% more amyloid and 29% more tau (the sticky plaques and tangles that accumulate in Alzheimer's brains) than those who reached REM early — and 39% less BDNF. Not just one biomarker. All three, simultaneously.

A separate preprint from Montagne et al. extended the story in a different direction: in cognitively normal older adults, those with amyloid-β positivity on PET or CSF had significantly greater night-to-night variability in their REM duration compared to amyloid-negative individuals. Not less REM — less consistent REM. The irregularity was the flag, not the amount. (Note: preprint, not yet peer-reviewed.)

A 2026 meta-analysis of 15,000+ people reinforced the cognitive side of this: higher sleep variability across all metrics correlated with worse cognitive performance, with older adults showing the strongest effect.

Taken together, these reframe what you should actually be optimizing for. Chasing more REM matters less than chasing stable REM, reached early in the night.

REM gets the cultural attention. Deep sleep is doing the work.

Another 2025 paper in Alzheimer's & Dementia looked at slow-wave synchrony — how coordinated your deep sleep oscillations are across the brain — and found it tracks Alzheimer's progression from mild cognitive impairment all the way to severe dementia.

Wave amplitude correlated with MoCA scores (a standard cognitive screening test — memory, attention, word recall) independent of age and sex. The brain doesn't care how many stars your wearable gave your sleep. It cares about the depth and coordination of those waves.

The mechanism is the glymphatic system (your brain's overnight cleaning crew). During deep sleep, cerebrospinal fluid pulses through the brain, physically flushing amyloid and tau. Inactive while you're awake. Runs poorly on fragmented sleep. Think: dishwasher on rinse-only, wondering why the plates aren't clean.

This is why "I got 8 hours" and "I got 2 hours of deep sleep" are two very different biological situations.

Duration still matters. It's just not sufficient on its own.

Multiple large cohort analyses from 2024-2025 have landed on the same shape: an inverted U, with cognitive risk rising at both ends. The sweet spot is 7-8 hours, with the inflection toward elevated risk appearing around 6 hours.

The oversleeping finding is worth sitting with. Sleeping 9+ hours is consistently associated with worse cognitive outcomes in longitudinal cohorts. The interpretation isn't positive: in many cases, people who shift to longer sleep have already entered early cognitive decline, and the brain is demanding more recovery time because the recovery isn't working well. Long sleep can be a symptom, not a perk.

If you're a 7-hour sleeper who feels rested, you're probably fine on duration. The optimization target shifts to architecture.

Let's be honest about what consumer sleep trackers do. They use accelerometry (movement) and photoplethysmography (heart rate and HRV). They do not measure brain waves. Sleep stage estimates are inferences, not measurements.

That said, some inferences are better than others.

A 2025 systematic review and meta-analysis found the Oura Ring shows comparable accuracy to PSG (polysomnography — scalp electrodes measuring actual brain waves in a sleep lab) for total sleep time, deep sleep, and sleep onset latency. A 2024 validation study across 96 participants and 421,000 epochs confirmed the Gen3 doesn't significantly differ from that standard on most measures. For a device you wear on your finger to bed, that's a meaningful result.

The current pecking order:

Oura Gen3: Best overall accuracy across all stages. Most rigorously validated. If you're using wearable data to make decisions about your brain health, this is the device the research supports most.

Apple Watch: Strong overall PSG agreement, solid for trends. Second-best of the four in head-to-head comparisons.

Fitbit: Third in the De Zambotti 2024 head-to-head (κ=0.41), which is "moderate" agreement — better than its reputation.

Whoop: Lowest overall accuracy in that comparison (κ=0.37), but highest sensitivity specifically for detecting deep sleep. If that's your single focus, it outperforms what you'd expect.

One universal caveat: wearables are better at trends than single nights. One bad sleep score tells you less than two weeks of data. If you're consistently seeing lower deep sleep percentage, higher REM variability, or delayed first REM — that pattern is signal. A single anomalous night is noise.

The one metric almost no consumer wearable surfaces clearly: REM latency. Oura includes your first REM episode timing in its staging chart — you can approximate latency by noting how many minutes post-sleep-onset it appears. It's buried, but it's there.

"Sleep affects health" is a correlation people have been arguing about for decades. With Mendelian randomization, we can do better.

A 2025 MR study in Scientific Reports confirmed a one-way causal path: insomnia causally accelerates GrimAge (a DNA-methylation clock that estimates how fast your cells are actually aging — think speedometer, not odometer). Mendelian randomization uses genetic variants that predate any disease as instruments, which gets you causality, not just association. The directional finding held.

A separate 2024 analysis in Sleep across 336,000+ UK Biobank participants confirmed the same.

Sleep isn't just a recovery tool. It is an active biological aging intervention. Chronic poor sleep isn't just unpleasant — it is measurably moving your aging clock in the wrong direction, independent of everything else you're doing.

What I'm actually doing. The evidence behind each piece. And the cautionary tale about cooling mattresses like 8Sleep and ChiliPad.

I changed what I check first in my wearable after going through this research. It's not the summary score. It's the sleep staging chart — specifically, how soon I hit my first REM episode.

My target: first REM cycle within 90 minutes of sleep onset (FYI this is simply a widely cited normal expectation for REM latency, but we’re still awaiting further evidence of what “best” range of time it should be). Luckily, I’ve had a fairly consistent REM latency of 60-75 minutes for as long as I have been tracking.

On wearables: I use an Apple Watch. I've tried ring-based trackers — spent real time with an Ultrahuman and couldn't get used to the sensation on my finger. The Watch is already on my wrist, and for trend tracking over weeks and months (which is where wearable data actually gets useful), it does what I need. I check the staging chart manually rather than relying on the summary score.

Temperature. This is the single highest-leverage lever most people aren't using correctly. Your core body temperature needs to drop to trigger slow-wave onset. I keep my bedroom at 66-70°F via Nest scheduling — it drops automatically before bed.

I also tried a ChiliPad. Then I woke up at 3am in what I can only describe as a full cryotherapy panic. I was so cold I couldn't figure out what was wrong with me. Pulled it off, threw it on the floor, went back to sleep on a normal mattress like a normal person. The device still works perfectly, but like most things, it’s not right for everyone.

Sauna 2-3x per week, timed right. I usually go at the end of the work day, and always make sure to finish at least 90 minutes before bed. The post-sauna temperature drop accelerates sleep onset and deepens early SWS. The timing matters: too close to sleep and you're fighting elevated core temp, which is the exact opposite of what you want.

Magnesium malate or glycinate, 400mg before bed. I rotate between the two — malate has a slight edge for muscle recovery, glycinate is calmer. Both are well-absorbed and don't cause GI issues. The evidence for magnesium's role in GABA modulation and slow-wave sleep is solid. GABA is your brain's primary calming neurotransmitter — the brake pedal. Magnesium helps activate it, which is why deficiency (extremely common) tends to show up as restlessness and light sleep. In a double-blind RCT, magnesium supplementation significantly improved sleep efficiency and early-morning awakening vs. placebo. Top-three sleep intervention for most people.

Phosphatidylserine (PS), 100-200mg before bed. I added this specifically for early morning awakenings — the 3-4am wake-up that won't let you back in. If you've ever been wide awake at 3:47am mentally narrating your to-do list, this is the mechanism. It's not anxiety. It's cortisol arriving four hours too early.

PS blunts the HPA axis cortisol response (HPA axis — the hormonal chain of command that orders your adrenal glands to release cortisol). It's not a sedative. It's turning down the alarm that's going off four hours too early. This pattern is extremely common in high-performing, high-stress people. As a note, however, the cortisol-blunting mechanism is documented; direct evidence for sleep specifically is limited — this one's more N=1 for me (given I doubt there will ever be an RCT on it!).

This is where I have to be honest with you. I don't have REM variability issues — my latency is consistent and my architecture is generally solid. My problem is simpler and harder to fix: I work too late, and I eat too late 😆.

I also don't use an alarm clock. My philosophy has become less about forcing a rigid schedule and more about giving myself the best possible environment for sleep and enough time to actually get through my stages. Regular exercise is non-negotiable. Bedroom temperature is set. The protocol above is in place.

But I'm not going to pretend I go to bed at the same time every night. Some nights it's 10:30pm. Others it's closer to midnight or…a lot later. I know the late eating isn't ideal — digestion competes with deep sleep, and a full stomach raises core temperature, which is the opposite of what you need for slow-wave onset. I'm working on it. And I’m telling you because don’t you just hate it when doctors pretend they’re perfect?

What I've actually changed and stuck with:

Light management front and back. Bright light within 30 minutes of waking anchors your circadian clock. I don't go outside in the morning (I live in NYC, let's be real), so I use a light therapy lamp instead. I also do 50-300 jumping jacks right after waking up, depending on my ambition level that day — the movement accelerates cortisol clearance and core temp regulation. Blue-light blocking and dimmed lighting in the 90 minutes before bed. This costs almost nothing and it works.

Exercise daily, intensity distributed. Hard training always ends long before bedtime. Intense exercise too close to bed elevates cortisol and core temp — both slow-wave disruptors.

And alcohol? Gone. Even one drink suppresses REM sleep in the first half of the night. The wearable data on this is genuinely unambiguous, and honestly I haven’t missed alcohol enough (skipping it is also great for body recomposition, brain health, and numerous other longevity-related reasons).

I'm still working on the consistency piece. Ask me in six months.

Most sleep guides skip this entirely. They shouldn't.

Estrogen and progesterone aren't just reproductive hormones — they're deeply involved in sleep architecture. Progesterone in particular has direct sleep-promoting properties: it metabolizes to allopregnanolone, which is a positive allosteric modulator of GABA-A receptors (the same pathway targeted by benzodiazepines, without the dependency risk). Estrogen supports REM stability and thermal regulation — which is why hot flashes are so disruptive: they're essentially thermal shocks that pull you out of deep sleep at the worst possible time.

When both decline during perimenopause and menopause, sleep architecture takes a measurable hit: more fragmentation, more early awakenings, less deep sleep, more REM instability.

Oral micronized progesterone is the form with the most direct evidence for sleep. A systematic review and meta-analysis of RCT data found that micronized progesterone improves sleep onset latency and overall sleep quality in peri- and post-menopausal women. The mechanism is the allopregnanolone conversion — it's genuinely sedating in a physiological, not pharmacological, way.

I've tried oral micronized progesterone myself. The sleep effect is real and noticeable — earlier sleep onset, fewer middle-of-the-night arousals. This isn't something you buy OTC; it requires a prescription and a conversation with a clinician who understands the evidence. If you've done everything in this protocol and your sleep still looks like a battlefield, and you're in perimenopause or beyond, hormones are worth a serious conversation.

If you want clinical support interpreting your wearable data alongside actual biomarkers, that's exactly what I do in my practice. hillarylinmd.com/longevity-practice

💊 The Alzheimer's drugs targeting amyloid may not actually help patients. A Cochrane systematic review of lecanemab (Leqembi) and donanemab (Kisunla) — the two FDA-approved anti-amyloid drugs — found clinical benefit is "absent or trivial" on validated scales, despite plaque clearance. The amyloid-removal strategy works biochemically. Whether it works clinically is now an open war in the field. Worth following closely if you care about brain health.

🏋️ A 5-year exercise habit buys 4 years of brain protection even after you stop. A RCT in Sports Medicine found that structural brain volume gains from aerobic exercise — hippocampal preservation, prefrontal protection — persisted 4 years after participants stopped training. Exercise isn’t a maintenance drug. It’s a structural investment with a long residual return. Start early.

🥩 Tirzepatide loses more lean muscle than semaglutide. A head-to-head preprint found tirzepatide users lost proportionally more lean mass than semaglutide users — a significant tradeoff given that sarcopenia (muscle loss with age) is one of the strongest predictors of longevity outcomes. If you’re on any GLP-1 agonist: resistance training and protein intake aren’t optional. They’re the co-intervention.

🧬 Your epigenetic clock trend matters more than your current reading. An 8-year longitudinal study in Clinical Epigenetics found that epigenetic age acceleration trajectory over time predicts all-cause mortality better than a single snapshot — independent of baseline health and chronological age. One test tells you where you are. Serial testing tells you where you're going. The 'is my protocol working?' question now has a mortality-validated answer.

🎤 AI can distinguish Alzheimer's from a different dementia just by listening to you speak. A study in JPAD found that ML analysis of speech patterns — phonemic fluency, semantic content, prosody — can stratify Alzheimer's disease from frontotemporal dementia (FTD, a separate condition that gets misdiagnosed as Alzheimer's regularly) before clinical diagnosis, validated against CSF and PET biomarkers. Non-invasive. Scalable. If this holds, it changes early cognitive screening entirely.

I just came from the LiveLong Summit — five times, where I spoke on-stage brilliant doctors and scientists from Drs. Stacy Sims, Mary Clare Haver, Jessica Shepherd, Vonda Wright, Wei-Wu He, Eric Verdin, Nicki Byrnes, and so many more. If you were there, thank you — the conversations in the room were some of the best I've had this year. If you weren't, it's one of the few spaces doing serious longevity work specifically for women, and it's worth following: livelongwoman.com.

We went deep on this exact topic — and I think it's one of the most important episodes we've done.

Deep Sleep: How to Cure Insomnia & Live Longer | Stanford MD

I made this episode because I kept seeing patients doing everything right on paper — eating clean, exercising, managing stress — and still sleeping terribly. We get into the mechanisms, the mistakes people make thinking they're helping, and what actually moves the needle.

Watch on YouTube → | Subscribe to The Longevity Show

The cultural conversation about sleep still orbits the wrong question.

"How many hours did you get?" is a necessary condition, not a sufficient one. Seven hours of fragmented, REM-late, low-synchrony sleep is a different biological reality than seven hours of well-architected sleep that starts cleaning your brain within the first 90 minutes.

I find this weirdly encouraging. The optimization target is more specific than we thought. The research is closing in on exactly which failure modes matter. We're getting actionable.

Forward this to one person currently proud of their 88 sleep score. They deserve to know.

Keep learning,

Hillary Lin, MD

Co-Founder & CEO, CareCore

Was this useful? Forward The Longevity Letter to someone who would benefit.

Follow me for more longevity insights: YouTube | LinkedIn | Instagram | X | TikTok

Want to turn your wellness brand into a full-service health destination? Learn about CareCore's platform or Get Started Here

The Aging Code Summit, May 26-27, Mosesian Center for the Arts, Greater Boston. I'll be on the clinician panel discussing how aging research translates into real interventions.

NYC Tech Week: "Sick Care Is Dead" — June 2, New York, NY
A panel on what actually comes after the old model. If you're in NYC and building in health, this is the room to be in.

Dry Eye Society of the Americas, July 10-11, 2026, NYC. Dry eyes are a growing problem and symptom of deeper issues tied to longevity. For clinicians - join us.

Science of Skin Summit — September 17–20, Austin, TX
I'm speaking twice — one session on AI in dermatology, one on skin and hair as windows into biological age. For clinicians.

🐟 Seatopia — sustainable, sashimi-grade seafood checked for microplastics and mercury (Code HILLARYLINMD for $20 off)

⏱️ Timeline — patented Urolithin A for scientifically proven mitochondrial support via mitophagy (Code CARECORE for 10% off)

🍺 ZBiotics — pro-longevity probiotics for when you do have a little fun (Code CARECORE for 10% off)

🧲 CoreViva — whole-body MRI screening for early disease detection (Code CARECORE for $200 off)

🧠 NeuroAgeTx — the most comprehensive brain aging package at up to 61% off for Longevity Letter readers

🧬 GlycanAge — measures your immune system's biological age and recommends how to fix it

Want to know your biological age? Check out our testing page for epigenetic age testing with physician consultation, including TruDiagnostic TruAge and SystemAge by Generation Lab.

Have a question or topic you want covered? Reply to this email.

Was this forwarded to you? Subscribe here to get The Longevity Letter every week.