Open the sleep summary in any wearable app tomorrow morning and count where your eye goes first. For almost everyone: deep sleep. Whether the number is high or low is the first thing you check, the first thing you post about, the first thing you optimize for. Coaches build products around it. Whole ring companies market themselves on it.
The undervalued stage is REM.
This is contrarian to the wearable narrative and it is straightforwardly correct on the biology. REM is where emotional memories get processed and defanged. REM is where creative recombination happens. REM is where complex procedural skills consolidate. REM is what alcohol destroys, what SSRIs blunt, and what disappears first when you cut sleep short. The people fixating on deep sleep numbers are, in almost every case, losing more REM than deep sleep to the same underlying behaviors — and do not notice because their tracker undercounts REM and overcounts deep.
Here is the case, honestly.
Why deep sleep gets all the attention
To be fair to deep sleep: it earned some of its reputation. N3 slow-wave sleep is where growth hormone spikes, where declarative memory consolidates, where the glymphatic system flushes metabolic waste most efficiently. Cutting N3 has real consequences. Xie's 2013 Science paper on metabolite clearance is what a lot of people cite when they talk about sleep, aging, and long-term neurological risk. That mechanism is real.
Deep sleep is also what wearables report best in the marketing view — a single dramatic-sounding number, easily gamified. It sounds serious. "90 minutes of deep sleep" reads better on a coach's Instagram than "good sleep spindle density." If you sleep more hours in absolute terms, your deep sleep number goes up. If you sleep less, it drops. The relationship is linear enough to feel controllable.
But two things.
First, deep sleep is largely front-loaded in the night. It is remarkably resilient to modest sleep restriction — cut an hour off the end of your night and you probably lost mostly REM, not N3. Your deep sleep budget was mostly spent by 3 a.m.
Second, most consumer devices overestimate deep sleep against polysomnography. Validation studies (de Zambotti's group and others) consistently show wrist and ring trackers reporting more N3 than PSG confirms. The number that feels controllable is also the number most likely to be inflated on your dashboard.
What REM actually does
REM handles a specific class of cognitive work that does not happen elsewhere.
Emotional memory processing — the "overnight therapy"
Walker's group has published for two decades on REM's role in decoupling emotional memories from their affective charge. The mechanism, sketched: during REM, noradrenergic activity in the locus coeruleus drops to near zero — REM is the only state where this happens — while limbic areas remain highly active. In practical terms, the memory of what happened gets consolidated, but the emotional intensity attached to it gets processed down. You remember the event without re-feeling it at the same volume.
Chronically low REM appears to interrupt that decoupling. Emotional memories stay hot.
The clinical corner of this is PTSD. Nightmares in PTSD are REM-dominant, and one hypothesis for why the disorder maintains itself is that normal REM processing of the traumatic memory is disrupted — so each night the memory reconsolidates without ever losing its charge. The prazosin literature, which works by reducing noradrenergic tone during REM, is built on this framework. The full picture is not mapped, but the REM-emotion link is one of the strongest signals in the whole sleep literature.
Creative recombination and insight
Cai's 2009 PNAS study had participants nap after training on a remote-associates task. The REM nappers improved measurably on the task afterward. Non-REM nappers and quiet-rest controls did not. The proposed mechanism: REM increases activity across weakly-linked neural networks, so distant associations become easier — the state that lets you wake up with the solution you could not see the night before.
Wagner's 2004 Nature paper on sleep-dependent insight showed the same class of effect on a hidden-rule discovery task. Sleep — particularly sleep that included substantial REM — more than doubled the rate of insight relative to matched wake time.
This is the biology behind the "sleep on it" folk wisdom. It is REM doing the work.
Procedural memory for complex sequences
N2's sleep spindles handle simple motor consolidation — the piano scale, the swing tempo, the isolated movement. REM contributes to the more complex, integrated procedural knowledge — chains of motor sequences, coordinated skill patterns, the version of skill practice where you are integrating several sub-skills at once. Rasch and Born's 2013 Physiological Reviews summary walks through the division of labor.
Why REM disappears when you cut sleep short
Here is the math most people never see written out plainly.
REM makes up about 20 to 25 percent of a full eight-hour night. But it is not distributed evenly. The first REM period after sleep onset is short — often five to ten minutes. Each subsequent REM period is longer than the last. The final REM period of the night can be 30 to 45 minutes.
Now apply that to what happens when you cut sleep short.
An eight-hour sleeper who moves to six hours does not lose 25 percent of their sleep evenly across stages. They lose the last two hours, and those last two hours were mostly REM. The math on typical distributions: an eight-hour night contains roughly 90 to 120 minutes of REM. A six-hour night contains roughly 30 to 45 minutes. Cutting two hours off the back of the night cuts REM by roughly 60 to 70 percent while cutting deep sleep by maybe 10 to 20 percent.
This is not a small difference. This is a night that looks 75 percent normal by total sleep time and is actually delivering roughly 40 percent of the REM the biology expects. Do that four nights in a row — as most people do during a normal work week with Sunday-through-Wednesday sleep debt — and REM debt accumulates fast. Try to "catch up" on Saturday morning and you get a REM-rebound sleep-in with vivid dreams, which is your brain telling you what it was starving for.
The wearable does not usually spot this cleanly. It reports total sleep as slightly reduced. It reports deep sleep as slightly reduced. It reports REM as reduced too — often by less than reality, because REM is the stage consumer trackers underestimate most.
Alcohol, SSRIs, and other REM suppressors
Two categories of everyday exposure suppress REM more than deep sleep. Most people do not know this, and their trackers do not clearly report it.
Alcohol
Even one or two drinks in the evening reliably suppress REM in the first half of the night, sometimes eliminate it entirely for the first sleep cycle. The second half of the night rebounds — REM percentage jumps and can be higher than baseline — but total REM across the night is typically reduced, architecture is fragmented, and rebound REM comes with more frequent awakenings. Ebrahim's 2013 meta-analysis in Alcoholism: Clinical and Experimental Research is the reference — alcohol shortens sleep latency, disrupts architecture, and cuts REM in a clean dose-dependent way.
You can drink a glass of wine at 6 p.m., feel like it did nothing to your sleep, and still have measurably less REM at 3 a.m. Your wearable may show reduced sleep quality but often blame it on the wrong stage.
SSRIs and SNRIs
Selective serotonin reuptake inhibitors and their SNRI cousins suppress REM directly and reliably. REM latency lengthens. Total REM shortens. This is not controversial — it is in every psychopharmacology textbook. The clinical question is whether the antidepressant benefit outweighs the REM suppression on a case-by-case basis, and for many people with moderate to severe depression the answer is clearly yes. But if you are on an SSRI, understand that your REM is chronically reduced, and adjust the other REM inputs accordingly. Do not compound it with regular alcohol. Do not chronically sleep six hours on top of it.
Bupropion (Wellbutrin) has a different profile — it does not suppress REM the way SSRIs do, and REM may even increase slightly. Worth a conversation with your prescriber if REM specifically is a concern and clinical fit allows the option.
Others worth knowing
Benzodiazepines and z-drugs (Ambien, Lunesta, Sonata) reduce REM. THC suppresses REM acutely, and stopping regular THC use produces intense REM rebound — the vivid dreams for a week or two that regular users notice on cessation. Beta-blockers can affect REM in some people. High-dose melatonin (5 mg and up) has been reported to alter architecture in some individuals; low doses (0.3 to 0.5 mg) are cleaner in the timing literature.
What to actually do
The intervention list for protecting REM is not exotic.
- Sleep the last two hours. This is the single biggest lever. Chronically cutting sleep by two hours at the back end of the night is chronically cutting REM by 60 to 70 percent. Weekend catch-up sleep does not fully make it up. Give yourself the back half of the night, most nights.
- Stop drinking within three hours of bed. The dose-response is real. Even moderate drinking near bedtime measurably suppresses REM. Time it earlier if you are going to drink at all.
- Talk to your prescriber if you are on chronic REM-suppressing medication. Not to stop taking it — that is not the point — but to know it is happening and to protect other inputs. A person on an SSRI who also sleeps six hours and drinks four times a week is compounding three REM hits.
- Consistent wake time. The circadian drive for REM is tied to core body temperature rhythm, which is anchored to your wake time. Irregular wake times fragment REM architecture even when total sleep is constant.
- Measure the trend, not the number. If you track, watch your REM trend across weeks. Absolute REM minutes on any given night are noisy and estimated. But a week-over-week drop of 20 percent that lines up with a behavior change is real signal.
Tracking REM at home — honest options
The Vyvata catalog has two options that give you a defensible picture of REM at home without going into a sleep lab.
Two honest caveats. First, both devices estimate REM — they do not measure it directly, and no consumer device does. Neither is a substitute for a diagnostic sleep study if you have real concerns about apnea, insomnia, or a movement disorder. Second, watch trends, not single-night numbers. Any given night's REM figure carries meaningful measurement error. A trend across two to four weeks tied to a behavior change is where the useful signal lives.
The bottom line
Deep sleep is important. It is not the stage most people are actually losing.
REM is what alcohol suppresses, what SSRIs blunt, what most people chronically underget by sleeping six hours instead of eight, and what most consumer trackers underreport. The people optimizing for their deep sleep number are, in almost every case, tolerating REM losses that are two to three times larger — and those losses do not show up cleanly in the app.
Protect the back half of your night. Stop drinking earlier. Understand what your medications are doing. And if you are going to use a tracker to reason about this, use it for trend rather than absolute number, and pick a device that has been validated against polysomnography rather than one built for step counts.
The stage worth optimizing for is the one you are most likely to be quietly starving.
