Research Methods Guide

Actigraphy: how it works, what it measures, and when to use it

A complete guide to actigraphy for researchers, clinicians and study coordinators — from accelerometer physics to sleep parameters, scoring algorithms, circadian metrics and the honest limitations. With peer-reviewed sources throughout.

By the Condor Research Team · Updated August 2026 · All cited claims link to published sources

How does actigraphy work?

An actigraph contains a micro-electro-mechanical (MEMS) tri-axial accelerometer that samples motion continuously — research devices typically sample between 25 and 100 Hz. The raw acceleration signal is reduced to activity counts per epoch using one of a few well-described methods:

• PIM (Proportional Integrating Measure) — integrates the area under the rectified acceleration curve within each epoch; sensitive to both the amount and intensity of movement.
• ZCM (Zero-Crossing Mode) — counts how many times the signal crosses a threshold near zero; sensitive to movement frequency.
• TAT (Time Above Threshold) — accumulates the time the signal spends above a set threshold; sensitive to movement duration.

A scoring algorithm then classifies each epoch as sleep or wake based on the pattern of activity counts in and around it. Because the classification is movement-based, the choice of epoch length, count method and algorithm all influence the final sleep estimates — which is why methods sections should report all three. (For a deeper treatment, see our guides to epoch length selection and feature extraction from raw signals.)

A brief history

Actigraphy grew out of 1970s ambulatory monitoring research, and its modern era began when Webster and colleagues published the first automated sleep-scoring algorithm for wrist activity in 1982.^[8] Four decades of validation studies, algorithm refinements and shrinking hardware later, actigraphy has moved from research curiosity to guideline-supported clinical tool^[1] — a trajectory reviewed in “40 years of actigraphy” by Patterson et al.^[2]

The scoring algorithms, briefly

Three classical algorithms dominate the literature, all published and openly described:

• Cole-Kripke (1992) — a weighted sum of activity in the scored epoch and surrounding epochs, validated against polysomnography in adults.^[3]
• Sadeh (1994) — a discriminant model using mean activity, variability and counts in a moving window; historically favored in adolescent and young-adult studies.^[4]
• Oakley / Actiware-style (1997) — a weighted window compared against a sensitivity threshold; the approach used by the widely cited Philips Actiware software. Documented in a technical report rather than a journal article.^[7]

Machine-learning and deep-learning classifiers are an active research area — but in a head-to-head comparison of eight algorithms on the same dataset, simple regression-based and heuristic algorithms performed slightly better than the machine-learning and deep-learning models, whose performance seemed to suffer from poor generalization.^[2] That result is one reason the classical algorithms remain the comparability standard across decades of published studies. Our overview of sleep-wake classification models covers this evolution, and algorithm validation against gold-standard measures explains how agreement is quantified.

What does actigraphy measure?

From an epoch-by-epoch sleep/wake classification plus rest-interval markers, actigraphy software derives the standard sleep parameters:

Because recordings span weeks, actigraphy also yields circadian rhythm variables that no single-night study can produce:

• Non-parametric variables (NPCRA): interdaily stability (IS), intradaily variability (IV), relative amplitude (RA), and the timing of the least active 5 hours (L5) and most active 10 hours (M10).
• Cosinor analysis: MESOR, amplitude and acrophase of the fitted 24-hour rhythm.
• Light-derived metrics: time above illuminance thresholds, timing of light exposure relative to the rest interval and, on spectrally capable devices, melanopic EDI.^[5,6]

These variables underpin research on circadian biomarker discovery and sleep fragmentation.

Where is the device worn?

The non-dominant wrist is the standard placement for sleep and circadian applications — most validation literature, including the classical algorithms above, assumes it. Other placements serve different goals: the hip is common in physical-activity research (step counts, MVPA), the ankle in infant studies and restless-legs protocols, and the chest when actigraphy rides along with cardiac monitors. Placement matters for comparability: counts recorded at different body sites are not interchangeable, and a light sensor reads a different exposure at the wrist than at the chest — a second, separate source of non-comparability beyond the device-specific counts discussed below. Report placement (including dominant vs non-dominant wrist) in the methods section, and see our guide to device placement and wear compliance.

Actigraphy vs polysomnography: which one, when?

Polysomnography (PSG) remains the gold standard for characterizing sleep architecture — it records brain activity, eye movements, muscle tone, breathing and oxygenation in a lab or clinic. Actigraphy answers a different question: what does sleep look like across weeks of ordinary life?

The 2018 AASM clinical practice guideline supports actigraphy for evaluating circadian rhythm sleep-wake disorders, for supplementing sleep logs in insomnia and hypersomnia work-ups, and for estimating TST when PSG is not feasible — while PSG remains indicated for suspected sleep-disordered breathing and parasomnias.^[1] In practice the two are complements, not substitutes: many protocols use actigraphy to characterize habitual sleep before and after a PSG night. Our side-by-side analysis of actigraphy vs polysomnography goes deeper, and multimodal data fusion covers protocols that combine both.

How accurate is actigraphy?

Honestly stated: good at detecting sleep, weaker at detecting quiet wakefulness. Across validation studies, classical algorithms show high sensitivity for sleep (commonly above 90%) but substantially lower specificity for wake, because a motionless awake person looks like a sleeping one to an accelerometer.^[2] The practical consequences:

• TST and SE tend to be overestimated in populations with long quiet wake — insomnia, older adults, shift workers lying in bed awake.
• Agreement with PSG is strongest in healthy sleepers and weakest where wake-in-bed is common; report population-specific validation where it exists.
• Accuracy claims should always name the algorithm, epoch length, device and reference standard — the same hardware can produce different numbers under different scoring settings.

Two design choices improve defensibility: pairing actigraphy with a sleep diary (the AASM-recommended practice^[1]), and using devices with off-wrist detection so non-wear is not scored as sleep. When gaps do occur, principled imputation strategies keep the analysis honest.

Light measurement: the capability that separates modern actigraphs

Because circadian physiology responds to light — particularly short-wavelength light acting on melanopsin-expressing retinal ganglion cells — modern circadian research increasingly requires quantifying personal light exposure, not just movement. Since CIE S 026:2018, the reference quantity is melanopic equivalent daylight illuminance (melanopic EDI), which weights light by the melanopsin action spectrum.^[5]

This is where actigraphs differ most. Single-channel photopic sensors measure brightness as the eye’s cone system sees it; deriving melanopic EDI requires spectral information that only multi-channel sensors provide. An independent NIH benchmark of four wearable light sensors found large differences between devices in detection range, accuracy against a criterion spectrometer, and inter-device variability — differences large enough to change study conclusions about light exposure.^[6] When light is an outcome variable, sensor selection deserves the same scrutiny as algorithm selection. Our guides to quantifying circadian light exposure and wearable light data quality cover the practical decisions.

What is actigraphy used for?

• Circadian rhythm sleep-wake disorders — delayed/advanced sleep phase, irregular sleep-wake rhythm, shift work disorder; the indication with the strongest guideline support.^[1]
• Insomnia and hypersomnia evaluation — objective complement to sleep diaries across multiple weeks.^[1]
• Clinical trials — objective, low-burden sleep endpoints at scale; actigraphy features in pharmaceutical and behavioral trials as a digital outcome measure. See regulatory considerations for wearables in trials.
• Chronobiology and light research — long-term rhythm stability, light-exposure ecology, seasonal and occupational studies.
• Population and remote research — decentralized protocols where participants never visit a lab; see remote sleep research at scale.
• Special populations — pediatrics, older adults, psychiatric and neurodegenerative conditions, where lab studies are burdensome and habitual patterns matter.

Clinical use and reimbursement: CPT 95803

In the United States, clinical actigraphy has a dedicated Category I billing code — CPT 95803 (“Actigraphy testing, recording, analysis, interpretation, and report”). Per the AMA CPT descriptor and its parenthetical instructions, the code requires a minimum of 72 hours and a maximum of 14 consecutive days of recording, cannot be reported together with home sleep apnea testing or polysomnography codes (95806–95811), and cannot be reported more than once in any 14-day period; the AASM guideline recommends recording durations consistent with these requirements.^[1]

The code existing does not mean the service is routinely paid. Several payers classify actigraphy as investigational — Blue Shield of California’s medical policy 2.01.73 (effective August 2025), for example, deems actigraphy investigational when used as the sole technique to record and analyze body movement — and practitioners in billing forums report frequent denials and limited success obtaining payment. Reimbursement also differs between Medicare contractors and between private payers. Before building a clinical workflow around CPT 95803, locate the specific medical policy of each payer involved and verify its current position. Research use, by contrast, involves no billing codes at all. This section describes the landscape as of August 2026 and is not billing advice.

Actigraphy in children and infants

Pediatric sleep research leans on actigraphy heavily, because lab studies are burdensome for children and night-to-night variability is high. The method carries specific considerations by age: the Sadeh algorithm originated in adolescent and young-adult samples,^[4] infant-specific algorithms exist (for example Sazonov et al., validated with ankle placement in infants^[9]), and scoring rules that perform well in adults can misclassify the fragmented, motile sleep of infants. Meltzer et al.’s review of pediatric actigraphy remains the standard methodological reference — including the finding that device, algorithm and scoring-rule choices materially change pediatric sleep estimates.^[10] Practical constraints matter too: smaller wrists, school schedules, and parent-reported diaries as the compliance backstop.

Limitations worth designing around

• No sleep staging. Movement cannot resolve N1–N3 or REM; questions about sleep architecture need PSG or EEG-based wearables.
• Quiet wakefulness inflates sleep estimates — mitigate with diaries, event markers and population-appropriate algorithms.^[2]
• Counts are device-specific. Activity counts are not directly comparable across manufacturers; switching hardware mid-study introduces a device confound unless bridged deliberately (parallel wear, shared algorithm implementations).
• Wear compliance is the real failure mode. Non-wear, water exposure and forgotten devices cost more data than any algorithm choice; see wear compliance and data integrity.

Actigraphy’s value is not that it replaces the sleep lab — it is that it measures the sleep people actually get, night after night, in the life they actually live.

Choosing a research actigraph: what to compare

Specification sheets rarely use the same definitions, so compare on evidence:

1. Validation trail — peer-reviewed validations of the device (or its algorithms) in populations like yours, not just manufacturer specifications.
2. Light sensing — single photopic channel vs multi-channel spectral sensing with melanopic EDI, and any independent accuracy data.^[6]
3. Data access — raw data export for reproducible analysis (R, Python, MATLAB), documented formats, no lock-in.
4. Wear logistics — battery life at your epoch setting, waterproofing, off-wrist detection, event markers.
5. Continuity — if migrating from discontinued hardware, whether legacy files and scoring methods carry over; see our Actiwatch migration guide.

Condor Instruments builds two research actigraphs — ActLumus, with a 10-channel spectral light sensor reporting melanopic EDI, and ActTrust 2 — analyzed in ActStudio, which implements Cole-Kripke, PIM/ZCM/TAT, an Actiware-style scoring algorithm, cosinor and NPCRA, with raw export for independent analysis.

Frequently asked questions

What is actigraphy in simple terms?

It is sleep and activity tracking based on movement. A watch-like device records how much you move, continuously, for days or weeks; software then estimates when you were asleep, when you were awake, and how stable your daily rhythm is.

What is actigraphy used for?

Evaluating circadian rhythm sleep-wake disorders, complementing sleep diaries in insomnia and hypersomnia work-ups, estimating sleep in clinical trials, and studying sleep, activity and light exposure in free-living conditions over long periods.^[1]

How accurate is actigraphy compared with polysomnography?

Validation studies consistently show high sensitivity for detecting sleep (typically above 90%) and lower specificity for detecting wake, so actigraphy tends to overestimate sleep in people who lie still while awake. Accuracy depends on the device, algorithm, epoch length and population, so studies should report all four.^[2]

How long should an actigraphy recording be?

Most protocols record 7–14 consecutive days, 24 hours per day, to capture night-to-night variability and both work and free days. Circadian variables such as interdaily stability benefit from the longer end of that range.

Is a consumer smartwatch the same as a research actigraph?

No. Consumer wearables estimate sleep with proprietary, changeable algorithms and rarely expose raw data, which makes methods reproducibility difficult. Research actigraphs provide documented algorithms, raw data export, epoch-level control and published validations — requirements for peer-reviewed and regulated research.

Does actigraphy measure light exposure?

Many research actigraphs include a light sensor. Capabilities vary widely: single-channel sensors record photopic illuminance only, while multi-channel spectral sensors can derive melanopic EDI per CIE S 026 — the quantity modern circadian research expects. Independent testing has found large between-device differences in light accuracy.^[5,6]

Do I need ethics approval to use actigraphy?

Actigraphy is non-invasive and low-risk, but it continuously records behavioral data, so research use requires the usual ethics review and informed consent covering data collection, storage and sharing — typically as expedited/minimal-risk review.

Last reviewed August 2026 by the Condor Research Team. Actiware and Actiwatch are trademarks of Philips Respironics, referenced for algorithm identification. Condor Instruments manufactures the ActLumus and ActTrust actigraphs; product mentions are identified as such.