What Is Actigraphy and How Is It Used to Track Sleep?

Actigraphy is a method of tracking body movement over days or weeks using a small, watch-like sensor worn on the wrist, and it has become one of the most widely used tools for studying sleep and physical activity outside a laboratory. The technology has been validated against in-lab polysomnography, the gold standard for sleep measurement, across several decades of research.¹ But actigraphy is not a perfect proxy for a sleep study, and its strengths and blind spots shape how clinicians, researchers, and even consumers should interpret the data it produces.

How the Sensor Works

An actigraph contains an accelerometer that detects motion, typically sampling many times per second. The raw acceleration signal gets converted into “activity counts” over set time intervals, usually 30 or 60 seconds. During sleep, your body moves very little, so the count drops toward zero. During wakefulness, counts rise. Software then applies a scoring algorithm that looks at the pattern of counts across a window of time and labels each interval as “sleep” or “wake.”

Most research-grade actigraphs are worn on the non-dominant wrist, though ankle-worn devices exist for specific purposes like measuring leg movements. The devices are small enough to forget you’re wearing one, which is the entire point: they capture your natural behavior at home, at work, and on the weekends, over periods that would be impractical in a sleep lab. A typical study might ask someone to wear an actigraph for one to two weeks straight, pressing a button at bedtime and on waking to help the software identify the sleep window.

How Accurate Is It Compared to a Sleep Lab

Actigraphy is very good at knowing when you’re asleep and less good at knowing when you’re awake. In a large validation study comparing wrist actigraphy to polysomnography, overall sensitivity for detecting sleep was about 0.97 and accuracy was around 0.86, but specificity for detecting wakefulness was only about 0.33.² In plain terms, the device catches almost every minute of genuine sleep, but it mistakes a lot of quiet wakefulness for sleep. If you’re lying still in bed scrolling through anxious thoughts, the actigraph may assume you’re out cold.

This asymmetry creates a predictable bias. Actigraphy tends to overestimate total sleep time and underestimate how long you spent awake after initially falling asleep. Those errors were not trivial in one study comparing four popular scoring algorithms: total sleep time was overestimated by roughly 30 to 45 minutes, and wake time after sleep onset was underestimated by about 22 to 25 minutes across most algorithms.³ Gender and insomnia status made only small differences in these accuracy metrics, but increasing age slightly reduced the device’s already-limited ability to detect wakefulness.⁴

The practical takeaway: actigraphy is reliable for estimating broad sleep patterns, like whether your average bedtime is drifting later or whether you’re getting substantially less sleep this month than last. It’s less reliable for pinpointing exactly how long it took you to fall asleep on a given night.

The Algorithm Matters More Than You’d Think

Not all actigraph software interprets movement data the same way. The four most commonly used scoring algorithms in research are the Cole-Kripke, Rescored Cole-Kripke, Sadeh, and UCSD methods. They all use the same underlying movement counts but apply different rules to classify each epoch as sleep or wake, and they can disagree meaningfully.

When tested against an EEG-based reference, all four had high sensitivity for detecting sleep epochs, in the range of 94 to 98 percent. But their specificity for detecting wake ranged from about 42 to 54 percent, and the sleep parameters they produced differed from one another. The Sadeh algorithm showed the smallest bias for wake after sleep onset, total sleep time, and sleep efficiency, making it generally recommended for healthy adults. The UCSD algorithm, despite its lower overall agreement with EEG, was better suited for tracking changes in sleep over time or in response to treatment.⁵

This means comparing results across studies can be tricky if different researchers used different algorithms, or different firmware versions of the same algorithm. It’s a gap that bothers the field and motivates ongoing work on standardization.

What Doctors Actually Use It For

The American Academy of Sleep Medicine published clinical practice guidelines recommending actigraphy for several specific conditions. These are all conditional recommendations, meaning the evidence supports their use but the strength of that support varies.

Insomnia: Actigraphy helps estimate sleep patterns in adults and children with insomnia, providing objective data that often differs from what patients report in sleep diaries.
Circadian rhythm disorders: For conditions like delayed sleep phase, advanced sleep phase, shift work disorder, and non-24-hour sleep-wake disorder, actigraphy is especially valuable because it can capture weeks of data showing the timing of sleep relative to the clock.
Sleep-disordered breathing: When integrated with a home sleep apnea test device, actigraphy can help estimate total sleep time during the recording, which is important for calculating the severity of the breathing problem.
Hypersomnolence: Before a formal daytime sleepiness test, actigraphy can verify that the patient has been sleeping a normal amount in the preceding days, ruling out simple sleep deprivation as the cause of their excessive sleepiness.
Insufficient sleep syndrome: Actigraphy can document that an adult is actually getting less sleep than they think, or confirm a pattern of chronic short sleep.

The one strong recommendation in the guideline was negative: actigraphy should not replace the electromyography-based measurement used to diagnose periodic limb movement disorder.⁶ The systematic review supporting these guidelines found that actigraphy consistently provides objective information that differs from patient-reported sleep logs for some parameters, reinforcing its value as a complement to self-report rather than a replacement for it.⁷

Earlier practice parameters from 2007 had already established actigraphy’s role in older adults and nursing home residents, populations where traditional overnight sleep studies are logistically difficult, and in pediatric populations where it helps delineate sleep patterns and track treatment responses.⁸

Circadian Rhythm Tracking

Beyond simple sleep-wake scoring, actigraphy data can be analyzed for circadian rhythm characteristics using what researchers call nonparametric methods. These bypass the need to fit a mathematical curve to the data and instead extract straightforward measures of the activity pattern. Three core metrics are commonly used: interdaily stability, which captures how consistent the rhythm is from day to day; intradaily variability, which measures how fragmented the rhythm is within a given day; and relative amplitude, which reflects the contrast between peak activity and the quietest period.⁹ The analysis also identifies the ten most active hours (M10) and the five least active hours (L5) of the day.¹⁰

These metrics turn out to be clinically meaningful. For diagnosing circadian rhythm disorders, actigraphy provides the long-term picture of rest and activity patterns that a single-night sleep study simply cannot. Experts recommend using actigraphy data alongside sleep logs and melatonin measurements when evaluating someone for a circadian disorder.¹¹

Sleep Misperception in Insomnia

One of actigraphy’s more interesting clinical applications is uncovering how people misjudge their own sleep. When researchers compared seven nights of actigraphy and sleep diary data in people with insomnia and good sleepers, they found three distinct subtypes of sleep misperception. Good sleepers tended to overestimate how long they slept, while people with insomnia tended to underestimate it. But the subtypes weren’t simply “correct” versus “wrong.” The groups were best distinguished by how short the person’s worst reported night was, and by the average size and variability of the gap between their diary and the actigraph.¹²

This matters clinically because telling a patient with insomnia “you’re actually sleeping more than you think” is not always the right message. The degree and direction of misperception vary, and the method of measurement influences what you find. When at-home actigraphy was compared with a single in-lab polysomnography night, people with insomnia showed consistent underestimation of total sleep time in both settings. But their misperception of how long it took to fall asleep was actually larger at home than in the lab, and their overestimation of wake time after sleep onset appeared only in the lab, not at home.¹³ The environment shapes the misperception itself, which is a strong argument for using multi-night home-based actigraphy rather than relying solely on a single lab night.

Mood Disorders and Activity Patterns

Actigraphy has become a prominent research tool in psychiatry, particularly for mood disorders where disrupted sleep and altered daily activity are core features. A meta-analysis of 38 studies covering nearly 3,800 participants found that people with depression were substantially less active than healthy controls and spent more time awake after falling asleep. Even during remitted or between-episode phases, people with a history of mood disorders still showed longer total sleep time, longer sleep onset latency, and more nighttime wakefulness compared to controls.¹⁴ That last finding is worth pausing on: the sleep disturbance doesn’t fully normalize between episodes, suggesting it may be a trait marker rather than just a symptom.

In bipolar disorder specifically, actigraphy reveals decreased overall activity and altered sleep patterns that persist across manic, depressed, and stable phases.¹⁵ Daily activity profiles can even distinguish mood states: depressive days in bipolar disorder are marked by lower overall activity, later activity onset in the morning, a midday bump, and low evening activity compared to other mood states.¹⁶ This kind of objective tracking is appealing because mood diaries are subjective and easy to skip, while an actigraph captures data continuously whether the patient remembers to engage with it or not.

Actigraphy in Children

Studying sleep in children poses obvious logistical challenges. Polysomnography requires a child to tolerate electrodes, wires, and an unfamiliar room, which can alter their sleep and upset the parent-child bedtime routine. Actigraphy sidesteps most of that. A review of pediatric validation studies found that sensitivity for detecting sleep was consistently high across age groups, from roughly 83 to 99 percent in infants and around 95 percent in adolescents. Specificity varied much more widely, ranging from as low as 17 percent in some infant studies up to about 77 percent in toddlers and 75 percent in adolescents.¹⁷

The wide specificity range in infants makes sense: babies are often still during quiet wakefulness, and they move around during active sleep, which confuses a motion-based algorithm. Researchers working with infants need to be especially cautious about what the device is really measuring. In older children and adolescents, the performance gap narrows and the device becomes more reliable for tracking real-world sleep schedules, making it useful for studying the effects of school start times, screen use, and other factors that shift sleep timing in young people.

Rest-Activity Rhythms and Neurodegeneration

Actigraphy-derived circadian metrics are showing up in dementia research as potential early warning signs. Higher intradaily variability, meaning a more fragmented rest-activity rhythm within the day, was associated with worse cognitive performance and increased dementia symptoms in adults with Down syndrome, a population at elevated risk for Alzheimer’s disease.¹⁸

In a study that combined actigraphy data with postmortem brain analysis, increasing fragmentation of the rest-activity rhythm over time partially mediated the link between degeneration of the locus coeruleus, a brainstem structure involved in arousal and attention, and cognitive decline in Alzheimer’s disease.¹⁹ Rhythm fragmentation, in other words, may not just be a consequence of neurodegeneration but part of the pathway through which it progresses. This is still an area of active investigation, but it illustrates how a simple wrist sensor can contribute to understanding conditions far beyond insomnia.

Research-Grade Devices Versus Consumer Wearables

If actigraphy sounds a lot like what your fitness tracker already does, that’s because consumer wearables share the same basic principle: an accelerometer on your wrist detecting movement. The crucial difference lies in transparency and validation. Research-grade actigraphs produce raw accelerometer data that can be analyzed with published, peer-reviewed algorithms. Consumer devices process the data through proprietary algorithms that manufacturers don’t disclose.

A study comparing consumer-grade wearables against a research-grade accelerometer in free-living conditions found discrepancies in physical activity metrics that the researchers attributed to fundamental differences in those proprietary algorithms. Because manufacturers don’t reveal how their calculations work, any explanation for why a Fitbit or Apple Watch disagrees with a validated research device remains speculative.²⁰ Your consumer device might tell you that you slept 7 hours and 42 minutes last night, but there’s no way for an outside scientist to verify how that number was derived or how it would compare to a research-grade measurement on the same night.

That said, some consumer wearables have begun publishing validation studies of their own, and the technology in commercial devices is rapidly converging with research devices in terms of sensor hardware. The gap is primarily one of openness: if the algorithm is a black box, the data can’t be independently scrutinized, and that matters in clinical and research contexts where reproducibility is essential.

Data Quality and the Non-Wear Problem

Actigraphy can only measure activity when you’re wearing the device, and one of the trickier data-processing steps involves distinguishing genuine sedentary behavior (sitting on a couch, sleeping) from periods when the device was removed (left on a nightstand, taken off for a shower). Both look like prolonged strings of zero counts.

Validated algorithms for detecting non-wear time typically use a window of consecutive zero counts as the cutoff. In adults, a 90-minute window of consecutive zero or near-zero counts, with an allowance for up to two minutes of artifactual movement, has been recommended as the most accurate approach.²¹ Using data from all three axes of a triaxial accelerometer improved classification accuracy compared to using only the vertical axis.²²

In children and adolescents, a shorter window of 30 minutes of consecutive zeros has been recommended, because young people tend to have more intermittent activity and shorter sedentary bouts. The choice of non-wear rule substantially affects estimates of sedentary time in youth, which matters for studies trying to link sitting behavior to health outcomes.²³ Getting this step wrong could make an active child look sedentary or vice versa, so researchers have to document their non-wear rules carefully and consider their effects on the results.

Machine Learning Is Improving Sleep-Wake Scoring

The traditional scoring algorithms like Sadeh and Cole-Kripke were developed decades ago using relatively simple threshold-based rules. A growing body of work is applying machine learning to raw accelerometer data, and the results are encouraging. In a large-cohort benchmark study, deep-learning architectures including convolutional neural networks and long short-term memory (LSTM) networks achieved accuracy scores significantly higher than both traditional algorithms and the device’s built-in scoring, and were comparable to human annotation.²⁴

The progress extends to children. A pediatric-trained LSTM classifier achieved a balanced accuracy of about 0.85, with sensitivity around 0.88 and specificity of 0.83, outperforming both adult-trained random forest models and standard actigraphy-based algorithms.²⁵ That specificity number is a meaningful jump above the 0.33 typically seen with traditional methods, which suggests machine learning may finally be closing actigraphy’s longstanding weakness in detecting wakefulness.

Physical Activity Measurement

Actigraphy isn’t only about sleep. The same accelerometer data can be used to estimate energy expenditure and classify physical activity intensity, which is why actigraphs appear in large epidemiological studies tracking how much people move. Research has established age-specific cut points for classifying activity as light, moderate, or vigorous based on the vector magnitude of acceleration counts per minute.²⁶

Accelerometer counts and fat-free mass together explain the largest share of variance in activity-related energy expenditure under free-living conditions. In one study, these two variables accounted for about 60 percent of the variance, with additional smaller contributions from questionnaire-based physical activity data and carbohydrate intake, bringing the model to about 71 percent explained variance in total.²⁷ The implication is that actigraphy provides a strong objective backbone for estimating physical activity, but it works best when combined with information about the person’s body composition and self-reported behavior.

Light Sensing and Its Limits

Some research-grade actigraphs include a built-in light sensor, which can capture ambient light exposure across the day and night. This is useful for circadian research because light is the primary signal that synchronizes your internal clock. However, not all light sensors are created equal. A recent evaluation of wearable light dosimeters found that the Actiwatch’s light sensor had the poorest spatial accuracy among the devices tested, with a sharply directional response that dropped to near-zero by 30 degrees off-axis. Dedicated light dosimeters designed for circadian research performed substantially better.²⁸ If your actigraph is on your wrist and your wrist is at your side, the sensor may be pointing at the floor, not at the sky or the overhead lights. This matters less for rough day-versus-night classification but considerably more for any study trying to quantify the biological impact of light exposure on circadian timing.

Leg Actigraphy for Limb Movements

While the clinical guidelines strongly recommend against using wrist actigraphy to diagnose periodic limb movement disorder, ankle-worn actigraphy has been explored as a more targeted alternative. The idea is intuitive: if you want to count leg kicks during sleep, put the sensor on the leg. A systematic review found that leg-worn actigraphy could circumvent many of the practical and economic barriers of polysomnography for quantifying nocturnal leg movements, but the evidence base was limited and heterogeneous. Different studies used different devices, placement positions, and counting methods, and common accelerometers varied considerably in their ability to detect periodic limb movements.²⁹ Standardization is still lacking, and for now, electromyography remains the reference method.