A evolução da actigrafia na pesquisa do sono

What actigraphy measures

Actigrafia

an indirect method of estimating sleep and wake from movement. A wrist actigraph records acceleration, converts the signal into time-based intervals called epochs, and applies an algorithm to classify each epoch as more consistent with sleep or wake. Sleep variables such as total sleep time, sleep onset, wake after sleep onset, sleep efficiency, and the sleep period are then derived from those classifications.

The method is valuable because sleep behavior is not limited to the laboratory. A wearable can capture several nights of rest-activity patterns in a participant’s normal environment, supporting longitudinal and free-living research. However, movement is only a proxy for sleep. Quiet wakefulness can be classified as sleep, while movement during sleep can be classified as wake. This fundamental limitation helps explain why Actigrafia generally shows high sensitivity for detecting sleep but lower specificity for detecting wake.

A clear distinction between hardware and software is essential. Two devices may record similar acceleration signals but produce different results because they use different epoch lengths, sensitivity settings, rest-interval rules, or scoring algorithms. Conversely, improved algorithms can extract more useful information from existing hardware when raw or minimally processed data are available.

^{[1], [2], [3], [4]}

A timeline of actigraphy milestones

The early development of actigraphy centered on the practical use of movement as a proxy for sleep and wake. The method expanded sleep assessment beyond the laboratory by allowing recordings in the home environment and across longer periods than a single PSG night (2025). This established the central use case that still defines actigraphy: measuring sleep-wake patterns in real-world conditions rather than directly recording every physiological feature of sleep.

A major milestone was the development and comparison of rule-based algorithms. Scoring-rule validation remains a core theme: in free-living conditions, two widely used actigraphs showed convergent validity for total sleep period, wake after sleep onset, and sleep efficiency (2017). Their performance depends on the device, population, scoring configuration, and reference standard. In a pediatric comparison, both tested actigraph brands showed good sensitivity but poorer specificity, and agreement with PSG varied according to algorithm, sensitivity setting, age group, and sleep-disordered breathing status (2012).

Validation work then moved from asking whether actigraphy could estimate sleep to examining how specific devices and algorithms agree with PSG. In a clinical laboratory sample, a Scripps Clinic algorithm produced epoch-by-epoch agreement of 85–87% with PSG, while wake was underestimated and neither tested approach achieved fully satisfactory agreement (2010). In adults with insomnia, actigraphy was satisfactory for several sleep-pattern measures, but the authors emphasized that the findings applied to a particular instrument and algorithm rather than to actigraphy as a whole (2006).

Another milestone was the recognition that defining the rest interval is itself an analytical decision. Bedtime and rise time determine the window in which many sleep indices are calculated. A study of Actiwatch 2 recordings found that software defaults and manual scoring could differ, particularly for rest-interval duration and activity levels around bedtime and rise time; the investigators proposed manual detection supported by an event diary (2016).

More recent development has focused on open, automated, and machine-learning approaches. An automated algorithm applied to Axivity, GENEActiv, and ActiGraph data produced estimates comparable with PSG for sleep-period timing, sleep duration, sleep onset, and waking time, but performed poorly for wake during the sleep period (2023). Other work has explored hidden Markov models, temporal convolutional networks, and recurrent neural networks to reduce dependence on manually defined rest periods or improve generalization across settings (2020) (2023) (2025).

The latest stage of the evolution is multimodal sensing. Current systems may combine accelerometry with heart rate variability, photoplethysmography, temperature, posture, or other physiological signals. A deep-learning model using activity counts and heart rate variability improved agreement with PSG over comparator algorithms, while a separate multi-sensor wearable study showed improved sleep detection and classification after applying a machine-learning model to raw accelerometer, photoplethysmography, and temperature data (2020) (2022).

^{[1], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14]}

How algorithms changed the actigraph

The actigraph device is only one component of the measurement system. The algorithm determines how acceleration is transformed into sleep-wake estimates, and apparently small choices can affect the final clinical or research outcome. Relevant choices include epoch duration, activity-count processing, sensitivity thresholds, treatment of non-wear, identification of the rest interval, and whether the algorithm uses only movement or additional physiological inputs.

Traditional supervised algorithms were often developed against PSG-labeled data and optimized for particular populations or devices. Their high sleep sensitivity can be useful when the principal question is whether a participant was probably asleep, but low wake specificity can lead to overestimation of sleep and underestimation of quiet wakefulness. This is especially relevant in insomnia, hypersomnolence, fragmented sleep, and other situations in which prolonged quiet wakefulness may be common.

Machine learning has expanded the design space. Hidden Markov models can infer individualized sleep-wake states without requiring a universal supervised classifier, while deep-learning models can use temporal patterns across multiple epochs and sensor streams. In patients with suspected idiopathic hypersomnia, a sequence-to-sequence long short-term memory model outperformed Actiwatch Software on relevant sleep-wake metrics during a 32-hour PSG comparison, although further work is needed before such performance can be generalized to home monitoring.

Algorithm improvements do not eliminate the need for transparent validation. A model can perform well in its development sample yet lose performance in a different population or dataset. For example, a temporal convolutional network that performed favorably in its evaluation data showed reduced performance when applied to the Multi-Ethnic Study of Atherosclerosis dataset, highlighting the importance of external validation and generalization testing.

^{[3], [5], [9], [11], [12], [13]}

From actiwatch systems to multi-sensor devices

Actiwatch systems helped establish wrist actigraphy as a practical sleep-monitoring method in clinical and research settings. In a seven-day comparison, ActiGraph Link and Actiwatch 2 produced highly correlated total sleeping-period estimates regardless of the algorithm used; overall sleep/wake classification accuracy ranged from 93.05% to 96.13% across algorithms. In a separate free-living comparison, ActiGraph GT3X+ showed high agreement with Actiwatch-64 for total sleep time and sleep efficiency, but underestimated wake after sleep onset and sleep-onset latency relative to the Actiwatch. In another validation study, THIM, Fitbit, and Actiwatch devices showed broadly comparable overnight sleep-monitoring performance, but all devices displayed substantial variability between individuals.

Research-grade accelerometers now commonly support tri-axial movement recording and may provide raw or minimally processed data for independent analysis. A study comparing Axivity, GENEActiv, and ActiGraph devices found that an automated algorithm generated comparable estimates for several timing and duration measures, but not for wake after sleep onset. These findings support the value of device comparison and standardized analysis rather than assuming that all accelerometers are interchangeable.

Consumer wearables have increased access to accelerometer-based monitoring, but their suitability depends on data access, documentation, intended use, and independent validation. In a PSG comparison, a consumer-grade Huami Arc produced sleep estimates comparable to a clinical-grade Actiwatch under an optimized threshold, while both devices showed the expected limitation in distinguishing motionless wake from sleep.

For teams planning an actiwatch replacement, the relevant comparison is not only the device body but also raw-data access, export options, wear location, algorithm documentation, and validation for the intended population.

Multi-sensor devices represent the next major development. Combining movement with heart rate variability improved sleep-wake scoring in one deep-learning study, and combining chest acceleration with posture changes improved sleep-wake detection compared with wrist acceleration alone in another study. These systems may also support sleep-stage estimation, although stage-level outputs require separate validation against PSG and should not be treated as equivalent to laboratory scoring without appropriate evidence. ^{[1], [3], [5], [10], [14], [15], [16], [17], [18]}

Choosing actigraphy for clinical and research workflows

The appropriate device is determined by the question being asked. For longitudinal sleep timing, regularity, and rest-activity patterns outside the laboratory, a validated wrist actigraph may provide a practical balance between participant burden and objective measurement. For studies requiring reproducible secondary analysis, raw-data access, documented sampling and epoch settings, and exportable files may be more important than a polished consumer dashboard.

Buyers should examine validation evidence for the intended population, wear location, recording duration, reference standard, algorithm, and outcome measure. Agreement for total sleep time does not guarantee agreement for wake after sleep onset, sleep efficiency, or epoch-by-epoch classification. For example, one multi-brand validation found acceptable agreement for several sleep-period measures but poor reliability for sleep efficiency and wake after sleep onset.

Clinical teams should also define how rest intervals will be identified, how diaries and event markers will be used, and how missing or ambiguous recordings will be handled. These decisions should be specified before data collection where possible, because differences in rest-interval definition and scoring configuration can affect comparisons between participants and studies.

^{[1], [9], [10]}

Limitations and interpretation

Actigraphy is not a replacement for PSG when the research or clinical question requires direct measurement of electroencephalography, respiratory events, arousals, or validated sleep staging. It infers sleep from movement and therefore has difficulty identifying quiet wakefulness, a limitation reflected in low wake specificity and frequent sleep overestimation in validation studies.

Results may vary with age, sleep disruption, insomnia, hypersomnolence, device placement, algorithm, sensitivity setting, and study environment. Inter-device comparisons can also be imperfect: different brands may show poor agreement even when each device appears reasonable against a reference method.

Actigraphy results should therefore be interpreted as estimates whose meaning depends on the measurement protocol. Validation should be specific to the device, algorithm, population, and outcome. Emerging AI and multi-sensor methods require transparent reporting, external validation, and standardized comparison frameworks before their outputs can be treated as broadly interchangeable. ^{[1], [3], [6], [8], [11], [12], [13]}

Frequently Asked Question

Related resources

• actiwatch replacement

Referências

1. Beyond the Sleep Lab: A Narrative Review of Wearable Sleep Monitoring. Maria P Mogavero, Giuseppe Lanza, Oliviero Bruni, Luigi Ferini-Strambi, et al. 2025. PubMed. Read source
2. Sleep assessment by means of a wrist actigraphy-based algorithm: agreement with polysomnography in an ambulatory study on older adults. Giulia Regalia, Giulia Gerboni, Matteo Migliorini, Matteo Lai, et al. 2021. PubMed. Read source
3. PSG Validation of minute-to-minute scoring for sleep and wake periods in a consumer wearable device. Joseph Cheung, Eileen B Leary, Haoyang Lu, Jamie M Zeitzer, et al. 2020. PubMed. Read source
4. Multi-Night at-Home Evaluation of Improved Sleep Detection and Classification with a Memory-Enhanced Consumer Sleep Tracker. Shohreh Ghorbani, Hosein Aghayan Golkashani, Nicholas I Y N Chee, Teck Boon Teo, et al. 2022. PubMed. Read source
5. The convergent validity of Actiwatch 2 and ActiGraph Link accelerometers in measuring total sleeping period, wake after sleep onset, and sleep efficiency in free-living condition. Paul H Lee, Lorna K P Suen. 2017. PubMed. Read source
6. Direct comparison of two new actigraphs and polysomnography in children and adolescents. Lisa J Meltzer, Colleen M Walsh, Joel Traylor, Anna M L Westin. 2012. PubMed. Read source
7. Wrist actigraphic scoring for sleep laboratory patients: algorithm development. Daniel F Kripke, Elizabeth K Hahn, Alexandra P Grizas, Kep H Wadiak, et al. 2010. PubMed. Read source
8. Actigraphy validation with insomnia. Kenneth L Lichstein, Kristen C Stone, James Donaldson, Sidney D Nau, et al. 2006. PubMed. Read source
9. Defining the rest interval associated with the main sleep period in actigraph scoring. Chin Moi Chow, Shi Ngar Wong, Mirim Shin, Rebecca G Maddox, et al. 2016. PubMed. Read source
10. Validation of an automated sleep detection algorithm using data from multiple accelerometer brands. Tatiana Plekhanova, Alex V Rowlands, Melanie J Davies, Andrew P Hall, et al. 2023. PubMed. Read source
11. A novel machine learning unsupervised algorithm for sleep/wake identification using actigraphy. Xinyue Li, Yunting Zhang, Fan Jiang, Hongyu Zhao. 2020. PubMed. Read source
12. Performance of an open machine learning model to classify sleep/wake from actigraphy across ∼24-hour intervals without knowledge of rest timing. Daniel M Roberts, Margeaux M Schade, Lindsay Master, Vasant G Honavar, et al. 2023. PubMed. Read source
13. Actigraphy against 32-hour polysomnography in patients with suspected idiopathic hypersomnia. Tugdual Adam, Jérôme Tanty, Lucie Barateau, Yves Dauvilliers. 2025. PubMed. Read source
14. Deep Neural Network Sleep Scoring Using Combined Motion and Heart Rate Variability Data. Shahab Haghayegh, Sepideh Khoshnevis, Michael H Smolensky, Kenneth R Diller, et al. 2020. PubMed. Read source
15. Free-living cross-comparison of two wearable monitors for sleep and physical activity in healthy young adults. Nicola Cellini, Elizabeth A McDevitt, Sara C Mednick, Matthew P Buman. 2016. PubMed. Read source
16. The Development and Accuracy of the THIM Wearable Device for Estimating Sleep and Wakefulness. Hannah Scott, Nicole Lovato, Leon Lack. 2021. PubMed. Read source
17. Improving Sleep Quality Assessment Using Wearable Sensors by Including Information From Postural/Sleep Position Changes and Body Acceleration: A Comparison of Chest-Worn Sensors, Wrist Actigraphy, and Polysomnography. Javad Razjouyan, Hyoki Lee, Sairam Parthasarathy, Jane Mohler, et al. 2017. PubMed. Read source
18. Predicting Sleep and Sleep Stage in Children Using Actigraphy and Heartrate via a Long Short-Term Memory Deep Learning Algorithm: A Performance Evaluation. R Glenn Weaver, James W White, Olivia Finnegan, Hongpeng Yang, et al. 2026. PubMed. Read source