Actigraphy in Sleep Medicine and Clinical Research: Triaxial Accelerometers, Cole-Kripke vs. Sadeh Scoring, WASO/TST Endpoints, and CPT 95803

Editorial and Clinical Notice: This reference guide summarizes biomedical instrumentation, sleep-wake scoring algorithms, and clinical practice guidelines from the American Academy of Sleep Medicine (AASM) and peer-reviewed sleep research literature. It is intended for clinical researchers and healthcare professionals and does not provide individual medical advice.

Actigraphy is a non-invasive, ambulatory method of monitoring human rest-activity cycles and circadian rhythms over multiple consecutive days or weeks using a wrist-worn or ankle-worn accelerometer (actigraph). Governed clinically by the American Academy of Sleep Medicine (AASM) Clinical Practice Guideline and deployed widely as a digital health technology (DHT) in decentralized clinical trials (DCTs), actigraphy translates continuous movement counts or raw triaxial gravitational acceleration (g-units) into objective sleep continuity endpoints—including Total Sleep Time (TST), Wake After Sleep Onset (WASO), and Sleep Efficiency (SE).

What Is Actigraphy and How Is It Used in Clinical Research and Healthcare?

Unlike in-laboratory overnight polysomnography (PSG)—which integrates electroencephalography (EEG), electrooculography (EOG), and submental electromyography (EMG) for a single night—an actigraph is worn continuously on the non-dominant wrist for 72 hours to 14 consecutive days (or longer in longitudinal trials) in the participant’s natural home environment, often paired with a daily sleep diary or the Pittsburgh Sleep Quality Index (PSQI).

Modern research-grade actigraphs use micro-electro-mechanical systems (MEMS) triaxial accelerometers sampling omnidirectional acceleration across three orthogonal axes (X, Y, Z) at 30 to 100 Hz, frequently accompanied by an ambient photopic/blue light sensor (lux) to detect bedtime light exposure and an off-wrist capacitive or skin-temperature sensor to distinguish true sleep from device removal.

Classical Epoch-Based vs. Raw-Data Accelerometry Pipelines

• Legacy Epoch Count Mode (Zero-Crossing / Time-Above-Threshold / Digital Integration): Band-pass filters the accelerometer signal (typically 0.25–2.5 Hz to isolate voluntary human limb movement from gravity and high-frequency tremor) and aggregates movement into 30-second or 60-second epoch activity counts.
• Raw Triaxial Acceleration Mode (e.g., ENMO and Angle-z in GGIR): Stores uncompressed raw acceleration in gravitational units (g) along all three axes. Open-source research pipelines such as GGIR compute the Euclidean Norm Minus One (ENMO = max(0, √(x² + y² + z²) − 1g)) and nocturnal arm-angle changes (angle-z) to derive vendor-agnostic sleep and 24-hour physical activity metrics across multi-site clinical trials.

Comparison of Sleep-Wake Scoring Algorithms: Cole-Kripke vs. Sadeh vs. Raw Arm-Angle Heuristic

Because actigraphy measures physical quiescence rather than cortical EEG sleep stages, mathematical scoring algorithms classify each 30-second or 1-minute epoch as either Sleep (S) or Wake (W):

Algorithm	Primary Reference & Target Population	Mathematical Window & Variables	Strengths & Known Limitations
Cole-Kripke Algorithm	Cole et al. (Sleep, 1992); validated primarily in adults (ages 35–65)	Weighted moving sum over 7 one-minute epochs (4 preceding epochs A-4..A-1, current epoch A0, and 2 subsequent epochs A+1, A+2) scaled by a constant factor P: D = P × (w-4A-4 + ... + w0A0 + w+1A+1 + w+2A+2) Scored as Sleep if D < 1.0; Wake if D ≥ 1.0, followed by Webster rescoring rules.	Standard default for adult insomnia, psychiatry, and oncology clinical trials; high sensitivity for sleep (~88–95%) but lower specificity for quiet wakefulness (lying still awake in bed).
Sadeh Algorithm	Sadeh et al. (Sleep, 1994); validated in children, adolescents, and young adults (ages 10–25)	Linear discriminant function over an 11-minute window: PS = 7.601 − 0.065·MEAN-W-5 − 1.08·NAT − 0.056·SD-LAST6 − 0.703·ln(ACT + 1) (where NAT is the number of epochs with activity ≥ 50 and < 100). Scored as Sleep if PS ≥ 0.	Accounts for higher nocturnal motor activity in pediatric and adolescent cohorts; widely used in pediatric sleep medicine.
HDCZA / van Hees Arm-Angle Algorithm (GGIR)	van Hees et al. (2015 / 2018); device-agnostic raw triaxial accelerometry	Identifies sustained inactivity bouts (SIBs) where the z-axis arm angle changes by < 5 degrees over a 5-minute rolling window, bounded by the Sleep Period Time (SPT) window.	Eliminates proprietary vendor count discrepancies across multi-device decentralized clinical trials.

Core Sleep Continuity and Circadian Digital Endpoints Derived from Actigraphy

In interventional clinical trials evaluating hypnotics, chronobiotics, or digital cognitive behavioral therapy for insomnia (dCBT-I), actigraphy generates two families of endpoints:

1. Nocturnal Sleep Continuity Endpoints (Within the Rest Interval):
   • Time in Bed (TIB) / Rest Interval: Duration (minutes) from “lights out” (marked by the patient pressing the actigraph event-marker button, sleep log, or ambient light drop) to final “out of bed” time.
   • Sleep Onset Latency (SOL): Minutes from lights out to the first epoch of sustained scored sleep.
   • Wake After Sleep Onset (WASO): Total minutes scored as wake between initial sleep onset and final awakening—a primary endpoint for sleep maintenance insomnia.
   • Total Sleep Time (TST): Total minutes scored as sleep within the rest interval (TST = TIB − SOL − WASO − terminal wakefulness).
   • Sleep Efficiency (SE%): Percentage of time in bed spent asleep (SE = [TST / TIB] × 100%).
2. Nonparametric Circadian Rhythm Analysis (NPCRA) Endpoints (Across Multiple 24-Hour Days):
   • Interdaily Stability (IS): Quantifies the consistency of the 24-hour rest-activity rhythm across days (ranging from 0 for Gaussian noise to 1 for perfect daily synchronization to environmental zeitgebers).
   • Intradaily Variability (IV): Quantifies rhythm fragmentation (frequency and extent of transitions between rest and activity within 24 hours; elevated in neurodegenerative diseases and sundowning).
   • Relative Amplitude (RA): Normalized ratio between the most active 10-hour period (M10) and the least active 5-hour period (L5): RA = (M10 − L5) / (M10 + L5).

What Standards, AASM Guidelines, and CPT 95803 Rules Govern Actigraphy?

Clinical use, reimbursement, and trial regulatory submissions for actigraphy are governed by three frameworks:

• 2018 AASM Clinical Practice Guideline (Smith et al., J Clin Sleep Med): The AASM recommends actigraphy for estimating sleep parameters in adult and pediatric patients with insomnia disorder, circadian rhythm sleep-wake disorders (CRSWDs) (such as delayed or advanced sleep-wake phase disorder, shift work disorder, and non-24-hour sleep-wake rhythm disorder), and insufficient sleep syndrome, as well as for monitoring total sleep time prior to Multiple Sleep Latency Testing (MSLT) in suspected narcolepsy/hypersomnia and estimating total sleep time during home sleep apnea testing (HSAT) (enhancing respiratory event index accuracy alongside STOP-Bang screening). Conversely, the AASM issues a strong recommendation against using actigraphy in place of EMG for diagnosing periodic limb movement disorder (PLMD).
• CPT Code 95803 Billing Requirements: Under AMA Current Procedural Terminology, CPT 95803 covers “Actigraphy testing, recording, analysis, interpretation, and report (minimum of 72 hours to 14 consecutive days of recording).” Payers require use of an FDA-cleared medical actigraph (typically regulated under 21 CFR § 882.1400 or § 880.6310) rather than a consumer fitness tracker, a minimum of 72 hours of continuous recording, and a signed physician interpretation report.
• FDA Guidance on Digital Health Technologies (DHTs) for Remote Data Acquisition in Clinical Investigations: When used as a trial endpoint in wearable DHT clinical trials, sponsors must document analytical validation of the scoring algorithm against simultaneous overnight polysomnography (epoch-by-epoch sensitivity, specificity, and Bland-Altman bias for TST and WASO) and define non-wear imputation rules in the Statistical Analysis Plan.