One third of American adults regularly sleep fewer than seven hours a night. The clinical literature on what this does to them is not ambiguous. It predicts cardiovascular disease, type 2 diabetes, obesity, several cancers, and Alzheimer’s disease through mechanisms that are now well characterized at the cellular level. The problem is not ignorance of the fact that sleep matters. The problem is that most people think of it as a lifestyle variable, a choice to be optimized against other demands, rather than a physiological process whose disruption has measurable consequences in blood, tissue, and cognitive performance.

This article covers two things. The first is the science: what sleep actually is, what it does, and what disrupted sleep predicts. The second is the practice: how to track it and what interventions the evidence actually supports.

Part 1: The Science

Sleep architecture

Sleep is not a uniform state. It is a structured sequence of stages that cycle through the night, each with distinct electrophysiological signatures and distinct biological functions. A full night of sleep contains four to six cycles, each roughly 90 minutes long, though the composition of those cycles shifts as the night progresses: early cycles are dominated by deep slow-wave sleep, while later cycles contain progressively more REM.

The stages are classified as NREM (non-rapid eye movement) sleep, which is itself divided into three substages, and REM (rapid eye movement) sleep.

N1 is the lightest stage, the transition from wakefulness. It typically occupies five percent or less of total sleep time. Muscle tone reduces, the eyes move slowly, and the brain produces theta waves. N1 is easily disrupted; arousal from N1 often leaves the person feeling they were never asleep.

N2 accounts for roughly 45 to 55 percent of total sleep time in healthy adults. The EEG shows two characteristic features: sleep spindles, bursts of oscillatory activity at 12 to 14 Hz generated by thalamocortical circuits, and K-complexes, sharp biphasic waveforms. Sleep spindles are not inert. They play an active role in memory consolidation: their density correlates with procedural and declarative learning gains across the sleep period. N2 is also when the body begins the cardiovascular downshift of sleep: heart rate and blood pressure fall, and the parasympathetic nervous system increases its dominance.

N3, also called slow-wave sleep (SWS) or deep sleep, is characterized by delta waves, high-amplitude low-frequency oscillations below 4 Hz. It occupies roughly 15 to 20 percent of total sleep time in young adults, declining with age. N3 is when the most consequential physiological work of sleep occurs.

The most important function of slow-wave sleep, identified in a 2013 paper in Science by Lulu Xie and colleagues, is glymphatic clearance. The glymphatic system is a network of perivascular channels through which cerebrospinal fluid circulates through brain tissue, flushing metabolic waste into the venous system. Xie et al. demonstrated in mice that glymphatic activity increased by approximately 60 percent during sleep relative to wakefulness. The same study showed that the interstitial space of the sleeping brain expanded by approximately 60 percent compared to the waking state, facilitating the convective flow of cerebrospinal fluid through tissue. Among the metabolic byproducts cleared by this system is amyloid-beta, the peptide that aggregates into the plaques characteristic of Alzheimer’s disease.^[1]

The clinical implication is direct. Slow-wave sleep is when the brain clears amyloid-beta. Chronic sleep deprivation, or conditions that reduce the proportion of slow-wave sleep, impair glymphatic function and allow amyloid-beta to accumulate. A 2017 study in humans by Ju and colleagues found that even a single night of sleep deprivation produced a measurable increase in amyloid-beta burden in the brain, detected by PET imaging, relative to a night of normal sleep.^[2] This is not a long-term trajectory inference. It is an acute finding.

N3 is also when the pituitary releases the majority of the day’s growth hormone in a single large pulse. Growth hormone during deep sleep drives tissue repair, protein synthesis in skeletal muscle, and contributes to the overnight reduction in cortisol. Disrupted slow-wave sleep, whether from age-related decline in delta activity, alcohol consumption, or poor sleep quality more broadly, suppresses this growth hormone pulse. The hormonal balance article on this site covers how IGF-1, which integrates growth hormone output, is measured and what it predicts: sleep quality is one of the three primary behavioral levers.

REM sleep is the stage where most vivid dreaming occurs. Its EEG signature resembles wakefulness: low-amplitude, high-frequency activity. Motor output to skeletal muscles is actively inhibited by the brainstem, producing the muscle atonia that prevents acting out dreams. REM serves two well-documented functions. The first is consolidation of episodic and emotional memories. During REM, the hippocampus replays recent experience and transfers it to distributed cortical storage, a process that requires the low-norepinephrine, high-acetylcholine neurochemical environment characteristic of REM. The second function is emotional processing: REM sleep strips the emotional charge from difficult memories while preserving the factual content, a process that Matthew Walker describes as “therapy that occurs in your sleep.” Walker’s lab found that participants who slept between two emotional memory encoding sessions showed substantially reduced amygdala reactivity to those memories relative to those who remained awake, and that the reduction correlated with REM sleep duration.^[3]

The architectural feature that most people underappreciate is the skewing of the cycles. Because early cycles contain more slow-wave sleep and late cycles contain more REM, truncating the night from the front (going to bed late) disproportionately removes slow-wave sleep, while truncating from the back (waking too early, or using an alarm) disproportionately removes REM. Both types of truncation are common; both produce different functional deficits.

Sleep efficiency

Sleep efficiency is the percentage of time in bed that is actually spent asleep. It is calculated as total sleep time divided by time in bed, multiplied by 100.

The commonly cited clinical target is 85 percent or above. The reasoning is straightforward: time in bed is a rough proxy for the opportunity for sleep, and efficiency captures how well that opportunity is used. An efficiency below 85 percent means the person is spending a substantial proportion of bed time awake, whether at sleep onset, during nighttime awakenings, or at the end of the night. Each of these patterns has different implications.

Low sleep efficiency due to extended time to fall asleep (sleep onset latency above 20 minutes) suggests difficulty transitioning from wakefulness, often involving sympathetic nervous system activation, elevated cortisol in the first half of the night, or misalignment between the circadian system and the sleep timing. Low sleep efficiency due to frequent nighttime awakenings is a different signal: it points toward sleep fragmentation, which is associated with poor slow-wave and REM sleep even when total sleep time is adequate, and which is a characteristic feature of obstructive sleep apnea. These distinctions matter because they point toward different causes and interventions.

Consumer wearables report sleep efficiency, though with the accuracy caveats discussed in Part 2. The number is useful primarily as a trend indicator: a consistent efficiency above 90 percent suggests good sleep architecture; one that is frequently below 80 percent is a signal worth investigating.

Nocturnal HRV

Heart rate variability (HRV) is the variation in time between consecutive heartbeats. A high HRV indicates that the autonomic nervous system is flexible, able to modulate heart rate rapidly in response to changing demands. A low HRV indicates a more rigid system, dominated by sympathetic tone.

Daytime HRV readings are contaminated by the behavioral and physiological demands of the waking state: posture, stress, food, exercise, and cognitive load all influence the reading within minutes. The measurement is real but noisy.

Nocturnal HRV is cleaner. During deep sleep, vagal tone, the parasympathetic input to the heart via the vagus nerve, reaches its daily maximum. The heart rate slows, and the high-frequency component of HRV, reflecting beat-to-beat variation driven by respiratory sinus arrhythmia, rises substantially. This nocturnal vagal dominance is not a passive consequence of reduced activity; it is an active physiological process. The brainstem nuclei that regulate autonomic output shift toward parasympathetic predominance during NREM sleep, and this shift is necessary for the cardiovascular recovery that sleep provides.

What suppressed nocturnal HRV predicts is well documented. Low overnight HRV is associated with increased all-cause mortality, incident cardiovascular events, and poor glycemic control. A 2018 analysis from the ARIC study found that lower nighttime HRV was significantly associated with incident type 2 diabetes, independently of established risk factors including BMI, waist circumference, and daytime physical activity.^[4] The nocturnal reading is more predictive than daytime HRV for cardiovascular outcomes in most prospective studies, likely because it reflects the quality of the autonomic recovery process rather than the acute response to challenge.

The practical value of tracking nocturnal HRV from a wearable is that it provides a sensitive early signal of physiological stress, illness, or inadequate recovery before subjective symptoms appear. HRV typically drops 24 to 48 hours before the onset of a symptomatic upper respiratory illness, before body temperature rises and before subjective impairment is evident. It also responds to alcohol, late-night exercise, poor sleep timing, and chronic training load with more consistency and earlier signal than most subjective measures.

Chronotype: the CLOCK gene and social jetlag

Chronotype is an individual’s intrinsic preference for the timing of sleep and activity. Early chronotypes (colloquially, morning types) have a naturally earlier circadian phase, sleeping and waking earlier. Late chronotypes (evening types) have a delayed phase, with later preferred sleep times. The preference is not a behavioral habit. It reflects genetic variation in the core clock genes that drive the circadian oscillator.

The most thoroughly studied polymorphism is in the PER3 gene. The PER3 5-repeat allele, found in approximately 10 percent of the European population, is associated with morningness. The 4-repeat allele is associated with eveningness. Homozygous carriers of the 5-repeat allele show substantially more slow-wave sleep in the first half of the night, more pronounced sleep pressure accumulation, and poorer performance on cognitive tasks when sleep-deprived relative to 4-repeat homozygotes. The genetic determinism is not absolute, but the heritability of chronotype is estimated at 50 percent, meaning that roughly half of the variance in chronotype across individuals is accounted for by genetic factors.^[5]

The CLOCK gene encodes a transcription factor that is part of the core molecular oscillator. Variants in CLOCK are associated with altered circadian period length, with some alleles associated with the ultra-long periods (greater than 24.5 hours) that favor evening chronotypes. Because the earth rotates on a 24-hour cycle and human social schedules are largely fixed to it, a person whose intrinsic clock runs slow is in permanent partial circadian misalignment.

This misalignment is called social jetlag. It is the discrepancy between the timing of sleep that the circadian system wants and the timing that social and occupational obligations impose. A late chronotype who works a standard schedule may have a biological sleep preference of midnight to 8 AM but an obligation to be awake by 6:30 AM, producing 90 minutes of daily social jetlag. A study by Roenneberg and colleagues analyzing self-reported chronotype data from over 65,000 Europeans found that social jetlag was associated with a higher body mass index, independent of total sleep duration: for every hour of social jetlag, the odds of being overweight or obese increased by approximately 33 percent.^[6]

The health risks of evening chronotype extend beyond BMI. Walker’s work, synthesizing data from multiple prospective cohorts, found that evening chronotypes had higher rates of depression, anxiety, cardiovascular disease, type 2 diabetes, and all-cause mortality than morning chronotypes. The mechanism is not primarily about sleep duration, which can be held constant between chronotypes for analysis. It is about the mismatch between the internal timing of physiological processes, including cortisol secretion, glucose metabolism, immune function, and body temperature regulation, and the external schedule that determines when those processes actually occur.

Evening chronotypes cannot simply choose to become morning chronotypes. Bright light therapy in the morning and careful sleep timing can shift the circadian phase by one to two hours over several weeks, and this is the primary evidence-based intervention for social jetlag. But a late chronotype shifted two hours earlier is still likely to be a late chronotype on an earlier schedule, not a morning type. The structural health disadvantage of being a late chronotype in a world organized around early schedules is real and largely unavoidable without schedule flexibility.

Sleep debt: it does not fully repay

Sleep debt is the cumulative shortfall between the sleep an individual requires and the sleep they obtain. The widespread assumption is that this debt can be cleared by sleeping longer on weekends or on recovery days. The evidence says otherwise.

The most rigorous data comes from a study by Gregory Belenky and colleagues, published in 2003 in Journal of Sleep Research. Sixty-six subjects were randomized to seven sleep conditions ranging from three to nine hours per night for seven days, followed by three days of recovery sleep at eight hours per night. Reaction time, measured by psychomotor vigilance task, deteriorated progressively across the restriction period and then improved during recovery. The critical finding was the shape of that recovery: subjects who had been restricted to seven hours per night recovered substantially over three days of recovery sleep. Subjects restricted to five or six hours per night showed improvement but did not return to baseline performance by the third recovery day. Performance remained significantly impaired relative to the group that had been sleeping nine hours throughout.^[7]

The more troubling finding from the sleep debt literature is the subjective/objective decoupling. When subjects are chronically sleep-restricted to six hours per night for two weeks and assessed for sleepiness and cognitive performance, their subjective ratings of sleepiness stabilize after a few days. They report feeling adapted. Their objective performance, measured by reaction time and sustained attention tasks, continues to deteriorate throughout the restriction period. At the end of two weeks of six-hour nights, their performance is equivalent to subjects who have been kept awake for 24 hours straight, but they do not feel as impaired as acutely sleep-deprived subjects feel, because chronic deprivation blunts the subjective perception of impairment.^[8]

This decoupling has practical consequences. People who are chronically sleep-restricted are not well positioned to accurately assess their own impairment. They feel fine. They are not fine. The objective indices, including nocturnal HRV, wearable-estimated sleep metrics, and cognitive performance tests, provide information their subjective sense does not.

What sleep predicts

The epidemiological literature connecting sleep to downstream health outcomes is extensive, consistent, and dose-dependent. The findings cluster around five areas.

All-cause mortality. A 2010 meta-analysis of 16 prospective studies covering over 1.3 million participants found that both short sleep (below six hours) and long sleep (above nine hours) were associated with increased all-cause mortality. Short sleep was associated with a 12 percent increase; long sleep with a 30 percent increase. The long-sleep association is generally interpreted as reflecting reverse causation, with undiagnosed illness driving extended sleep, rather than long sleep driving mortality.^[9]

Cancer. Walker’s synthesis of the cancer literature found that sleeping fewer than six hours a night was associated with substantially increased cancer risk across multiple cancer types, with relative risk increases in some studies exceeding 40 percent. The mechanism most studied involves natural killer (NK) cell activity. NK cells are the immune system’s primary surveillance mechanism for cells that have undergone malignant transformation. A 2012 study found that a single night of restricted sleep (four hours) reduced NK cell activity by approximately 70 percent relative to a normal night. The activity returned over subsequent nights of adequate sleep, but the window of suppression represents a real reduction in immune surveillance.^[10]

Cardiovascular disease. Sleep duration below six hours is associated with a roughly 20 percent increased risk of myocardial infarction and stroke in multiple large prospective cohorts. The mechanisms include elevated sympathetic tone, higher overnight blood pressure (the normal nocturnal dip in blood pressure is attenuated in poor sleepers), and elevated inflammatory markers including hsCRP and IL-6. Short sleepers show elevated hsCRP in prospective studies even after adjustment for BMI, smoking, and physical activity.

Insulin resistance. Sleep restriction raises fasting insulin and reduces insulin sensitivity through mechanisms that are partially independent of changes in diet or physical activity. A controlled study by Spiegel and colleagues restricted sleep to four hours per night for six nights. After six nights of restriction, glucose tolerance was substantially impaired: the glucose response to an intravenous glucose tolerance test was 40 percent slower, and acute insulin response was 30 percent lower, relative to a well-rested baseline in the same subjects. The authors noted that the metabolic profile of the sleep-restricted young adults resembled that of older adults with impaired glucose tolerance.^[11]

Alzheimer’s disease. Beyond the acute amyloid-beta finding from the Ju study, longitudinal data support the relationship between poor sleep and Alzheimer’s risk. A 2021 study following over 7,000 participants in the UK Biobank found that those who consistently slept fewer than six hours at age 50 and 60 had a 30 percent increased risk of dementia compared to those sleeping seven hours. The association held after excluding dementia cases identified within the first 10 years of follow-up to reduce the likelihood of reverse causation.^[12] The glymphatic mechanism provides a plausible causal pathway, not just an association.

---

Part 2: Tracking and Improving

What wearables actually measure

Consumer wearables, including the Oura ring, Whoop, and Apple Watch, have transformed sleep monitoring from a clinical procedure to a continuous daily data stream. The transformation comes with a significant caveat: these devices do not measure sleep directly.

Polysomnography (PSG) is the clinical gold standard for sleep staging. It records EEG activity from multiple scalp electrodes, eye movements (EOG), and muscle tone (EMG) simultaneously, providing the electrophysiological data from which sleep stages are classified. Consumer wearables have no EEG capability. They infer sleep and wakefulness from accelerometry (movement detection), heart rate, and heart rate variability. From these signals, proprietary algorithms estimate sleep stages.

The accuracy of this estimation varies by device and by study. The most generous independent validation studies find that consumer wearables achieve roughly 75 to 80 percent agreement with PSG for wake-versus-sleep classification, which is adequate for population-level trend tracking. Sleep stage agreement is worse: for specific stage identification, particularly distinguishing N2 from N3 and detecting brief awakenings, accuracy drops substantially. A 2020 meta-analysis comparing consumer devices to PSG found that most devices overestimated total sleep time and underestimated wakefulness after sleep onset, meaning they systematically make sleep look better than it is.^[13]

This matters for interpreting the data they produce. A single night’s Oura ring report showing 85 minutes of “deep sleep” and 90 minutes of “REM” should not be read as a precise staging of that night’s architecture. It is an estimate with meaningful error at the level of individual nights. What is more reliable is the trend across many nights, and the relative changes in the estimates over time in the same person using the same device.

The most useful metrics to track from a wearable are three things that do not require accurate sleep stage estimation:

Sleep consistency. The standard deviation of bedtime and wake time across nights. The circadian system is more sensitive to timing regularity than to total sleep duration. Variable sleep and wake times undermine circadian entrainment and reduce the predictability of the internal timing cues that the clock genes depend on. Consistent anchor times, particularly a consistent wake time, are the primary intervention for improving circadian alignment. A wake time that varies by more than 30 minutes night-to-night is worth targeting before optimizing anything else.

Total sleep time. The device’s estimate of total sleep time is imprecise for any given night but provides a reasonable signal averaged over a week or more. The population-level evidence strongly supports seven to nine hours as the range associated with lowest all-cause mortality. Total sleep time below six hours averaged across a week warrants attention.

Nocturnal HRV trend. Absolute HRV values are highly individual; comparing your HRV to population norms is less useful than tracking your own trajectory over weeks and months. A declining trend in baseline nocturnal HRV is a signal of cumulative physiological stress, whether from illness, overtraining, poor sleep quality, or lifestyle factors. Devices that compare your current reading to your personal 30- or 60-night baseline provide the most actionable signal.

Improving sleep: what the evidence supports

What follows are interventions with a mechanistic basis and replicated evidence. This is not a list of sleep hygiene generalities.

Morning bright light

The circadian system is entrained primarily by light, specifically by photic input to intrinsically photosensitive retinal ganglion cells (ipRGCs) in the retina that express melanopsin, a photopigment with peak sensitivity around 480 nm (blue wavelengths). These cells project directly to the suprachiasmatic nucleus (SCN) of the hypothalamus, the master circadian clock, via the retinohypothalamic tract.

Morning bright light, in the range of 10,000 lux, advances the circadian phase when received in the first two hours after waking. The mechanism is a phase-response curve: light early in the subjective morning shifts the clock earlier (phase advance); light in the evening shifts it later (phase delay). Twenty to thirty minutes of 10,000 lux exposure through a dedicated light therapy lamp, or outdoor light under clear sky conditions (which provides this intensity naturally), produces measurable phase advances and has been used to treat social jetlag, delayed sleep phase syndrome, and seasonal affective disorder.

The mistaken simplification is to frame this as a blue light problem. The issue is not specifically blue light; it is retinal melanopsin stimulation, and melanopsin is sensitive to the overall amount of light incident on the retina, particularly at short wavelengths. Blocking blue light with amber-tinted glasses is one intervention. A more complete approach is reducing all light intensity in the evening, not just blocking one wavelength from screens. A well-lit living room at 200 to 300 lux, even if the light is warm-toned, provides enough short-wavelength photons to meaningfully suppress melatonin production relative to dim light conditions.

The practical implication: dim all artificial light after sunset or at least two hours before bed, not just screens, and get bright natural light as early as possible in the morning.

Temperature and the core body temperature drop

Sleep onset requires a drop in core body temperature of approximately 1°C. This cooling is not a consequence of sleep; it is a prerequisite for it. The circadian clock drives a programmed afternoon peak in core body temperature followed by a decline in the evening, reaching its nadir in the early morning hours. Environments that interfere with this drop, including rooms that are too warm, delay sleep onset and reduce slow-wave sleep depth.

The optimal sleeping room temperature is approximately 65 to 68°F (18 to 20°C) for most adults. Individual variation exists, and the relevant target is skin temperature during sleep, which should be somewhat warmer than the air temperature due to peripheral vasodilation.

The warm bath finding is counterintuitive enough to warrant explanation. Taking a warm bath or shower one to two hours before bed, at a water temperature of approximately 104 to 108°F (40 to 42°C), reduces sleep onset latency and improves slow-wave sleep. The mechanism is not the bath warming the core; it is the bath accelerating the peripheral vasodilation that dumps core heat to the environment. The hot water dilates the blood vessels in the hands and feet, dramatically increasing blood flow to the extremities, which act as radiators, dissipating core body heat to the environment more rapidly. Core body temperature drops faster than it would without the bath, which is the condition the brainstem needs to initiate sleep. A 2019 meta-analysis of 17 studies confirmed the effect: body bathing or showering in warm water (40 to 42.5°C) one to two hours before bed was associated with significantly improved subjective and objective sleep quality, and specifically with reduced sleep onset latency and increased slow-wave sleep.^[14]

Caffeine: the quarter-life

Caffeine blocks adenosine receptors. Adenosine is a byproduct of neuronal metabolism that accumulates in the brain across the waking day, driving increasing sleep pressure. Caffeine does not reduce sleep pressure; it masks it by competitively blocking the receptors through which adenosine signals. When caffeine is metabolized and the blockade lifts, the accumulated adenosine binds its receptors in a rush, which explains the crash that follows caffeine wearing off.

The half-life of caffeine is five to seven hours in most adults, meaning half of an ingested dose remains active after that interval. Less well known is the quarter-life, which is ten to twelve hours. A 200 mg dose of caffeine consumed at 2 PM still has 50 mg active at 9 PM and 25 mg active at midnight. Most people dramatically underestimate the residual caffeine burden from afternoon consumption.

The practical cutoff for most people is before noon or at the latest early afternoon. For evening chronotypes or people with slower caffeine metabolism (CYP1A2 slow metabolizers, identifiable by genetic testing), the appropriate cutoff may be earlier still. The common experience of being able to fall asleep after an afternoon coffee proves nothing about sleep quality: caffeine’s impact on sleep architecture, particularly slow-wave sleep depth, persists even when sleep onset is not obviously delayed.

Alcohol: net negative on every measure

Alcohol is the most common sleep aid in use. Its reputation for helping sleep is based on a real pharmacological effect: it reduces sleep onset latency, meaning it genuinely makes people fall asleep faster. The mechanism is GABA-ergic, the same as benzodiazepines. This part works.

What happens next does not. Alcohol is metabolized relatively quickly. As blood alcohol concentration falls in the second half of the night, the sedative effect wears off and a rebound in alertness and sympathetic activity occurs. Sleep in the second half of the night, the half that contains most of the night’s REM, becomes fragmented, with more frequent awakenings and substantially reduced REM duration. The REM suppression is dose-dependent: a study by Ebrahim and colleagues quantifying the dose-response relationship found that high doses of alcohol (blood alcohol concentration above 0.10) reduced REM by approximately 24 percent in the first sleep cycle.^[15]

The net effect: alcohol may reduce the time it takes to fall asleep by 10 to 15 minutes while substantially degrading the quality of the sleep that follows. HRV is typically suppressed for the entire night after even moderate alcohol consumption. Wearable devices report this reliably: nights with alcohol almost universally show reduced overnight HRV relative to the individual’s baseline. This is one of the more consistent signals consumer wearables produce.

Consistency over duration

Of all the variables in sleep, consistency of timing has the clearest mechanistic basis for its central importance. The circadian system is a molecular oscillator that anticipates, rather than reacts to, environmental cycles. Its predictive capacity depends on the timing cues, primarily light and meal timing, arriving at predictable intervals. When sleep and wake times vary substantially night-to-night, the circadian system receives conflicting phase information and cannot stably entrain. The result is internal desynchrony between the clock and behavior, analogous to mild chronic jetlag.

The evidence that circadian disruption causes metabolic harm is robust. Shift workers, who experience repeated phase reversals, have substantially elevated rates of metabolic syndrome, type 2 diabetes, cardiovascular disease, and cancer relative to day workers matched for other risk factors. The harm is not simply from reduced total sleep: studies controlling for total sleep duration find that the timing irregularity itself contributes to metabolic dysfunction.

For most people, the most effective intervention is to anchor wake time first and hold it constant, including on weekends. Waking at the same time every day, regardless of what time sleep began, maintains a fixed phase reference point around which the circadian system can organize. Keeping bedtime variable but wake time fixed means some nights will be shorter, but the circadian consistency it produces is more protective than attempting to sleep in on weekends, which advances the circadian phase and produces Monday morning social jetlag.

Exercise timing

Exercise advances and consolidates sleep through multiple mechanisms: it increases adenosine accumulation, slightly elevates core body temperature (which must then drop, facilitating sleep onset later), and augments slow-wave sleep depth in the subsequent night. Morning and afternoon exercise are consistently associated with improved sleep quality.

Late-evening vigorous exercise is more variable. The sympathetic activation produced by vigorous exercise, elevated heart rate, circulating catecholamines, and elevated core body temperature, can delay sleep onset by one to two hours in individuals who are sensitive to exercise-induced arousal. The temperature mechanism is the most robust: vigorous exercise raises core body temperature, and that elevation takes one to three hours to dissipate. If exercise finishes close to bedtime, the residual core temperature elevation may prevent or delay the temperature drop that sleep onset requires.

The evidence is not uniformly negative for evening exercise: some individuals tolerate it without sleep disruption, and some studies find no effect. The safest approach for someone troubleshooting sleep onset difficulties is to avoid vigorous exercise within two to three hours of bed. Light walking or yoga in the evening does not produce the same sympathetic or thermal response and appears to be neutral or mildly beneficial for sleep.

---

Sleep generates a daily data stream that most people discard. The overnight HRV trend tells you whether your autonomic system is recovering. The sleep consistency metric tells you whether your circadian system is entrained. The glymphatic system runs a clearance cycle that depends on slow-wave sleep. These are not abstractions. They are processes with downstream consequences for brain function, metabolic health, and immune surveillance that can be tracked, and to a meaningful degree, influenced.

The wearables article on this site covers the specific devices and their accuracy profiles in more detail: What Wearables Actually Track. The blood tests series starting point, including why normal results can coexist with significant risk, is at Blood Tests: Normal Is Not the Same as Healthy.

References

1. Xie L, Kang H, Xu Q, et al. (2013). Sleep drives metabolite clearance from the adult brain. Science, 342(6156), 373–377. https://pubmed.ncbi.nlm.nih.gov/24136970/
2. Ju YS, Ooms SJ, Sutphen C, et al. (2017). Slow wave sleep disruption increases cerebrospinal fluid amyloid-beta levels. Brain, 140(8), 2104–2111. https://pubmed.ncbi.nlm.nih.gov/28899014/
3. van der Helm E, Yao J, Dutt S, Rao V, Saletin JM, Walker MP. (2011). REM sleep depotentiates amygdala activity to previous emotional experiences. Current Biology, 21(23), 2029–2032. https://pubmed.ncbi.nlm.nih.gov/22078101/
4. Carnethon MR, Golden SH, Folsom AR, Haskell W, Liao D. (2003). Prospective investigation of autonomic nervous system function and the development of type 2 diabetes: the Atherosclerosis Risk In Communities study, 1987–1998. Circulation, 107(17), 2190–2195. https://pubmed.ncbi.nlm.nih.gov/12695302/
5. Viola AU, Archer SN, James LM, et al. (2007). PER3 polymorphism predicts sleep structure and waking performance. Current Biology, 17(7), 613–618. https://pubmed.ncbi.nlm.nih.gov/17346965/
6. Roenneberg T, Allebrandt KV, Merrow M, Vetter C. (2012). Social jetlag and obesity. Current Biology, 22(10), 939–943. https://pubmed.ncbi.nlm.nih.gov/22578422/
7. Belenky G, Wesensten NJ, Thorne DR, et al. (2003). Patterns of performance degradation and restoration during sleep restriction and subsequent recovery: a sleep dose-response study. Journal of Sleep Research, 12(1), 1–12. https://pubmed.ncbi.nlm.nih.gov/12603781/
8. Van Dongen HP, Maislin G, Mullington JM, Dinges DF. (2003). The cumulative cost of additional wakefulness: dose-response effects on neurobehavioral functions and sleep physiology from chronic sleep restriction and total sleep deprivation. Sleep, 26(2), 117–126. https://pubmed.ncbi.nlm.nih.gov/12683469/
9. Cappuccio FP, D’Elia L, Strazzullo P, Miller MA. (2010). Sleep duration and all-cause mortality: a systematic review and meta-analysis of prospective studies. Sleep, 33(5), 585–592. https://pubmed.ncbi.nlm.nih.gov/20469800/
10. Irwin M, Mascovich A, Gillin JC, Willoughby R, Pike J, Smith TL. (1994). Partial sleep deprivation reduces natural killer cell activity in humans. Psychosomatic Medicine, 56(6), 493–498. https://pubmed.ncbi.nlm.nih.gov/7871104/
11. Spiegel K, Leproult R, Van Cauter E. (1999). Impact of sleep debt on metabolic and endocrine function. Lancet, 354(9188), 1435–1439. https://pubmed.ncbi.nlm.nih.gov/10543671/
12. Sabia S, Fayosse A, Dumurgier J, et al. (2021). Association of sleep duration in middle and old age with incidence of dementia. Nature Communications, 12, 2289. https://pubmed.ncbi.nlm.nih.gov/33879784/
13. de Zambotti M, Cellini N, Goldstone A, Colrain IM, Baker FC. (2019). Wearable sleep technology in clinical and research settings. Medicine and Science in Sports and Exercise, 51(7), 1538–1557. https://pubmed.ncbi.nlm.nih.gov/30789435/
14. Haghayegh S, Khoshnevis S, Smolensky MH, Diller KR, Castriotta RJ. (2019). Before-bedtime passive body heating by warm shower or bath to improve sleep: a systematic review and meta-analysis. Sleep Medicine Reviews, 46, 124–135. https://pubmed.ncbi.nlm.nih.gov/31102877/
15. Ebrahim IO, Shapiro CM, Williams AJ, Fenwick PB. (2013). Alcohol and sleep I: effects on normal sleep. Alcoholism: Clinical and Experimental Research, 37(4), 539–549. https://pubmed.ncbi.nlm.nih.gov/23347102/