Real-World Use of Consumer Sleep Devices: A Rapid Review

Abstract

Topic Importance:

As awareness of sleep’s role in health grows, consumer sleep technologies (CSTs) have gained wide use among both researchers and the public. These devices offer continuous, non-invasive sleep tracking in real-world environments. However, understanding their accuracy compared to polysomnography (PSG)—the gold standard in sleep measurement—remains limited, particularly in free-living settings and clinical populations. Assessing CST performance and usability in these contexts is essential for guiding future adoption.

Review Findings:

This review identified 29 studies evaluating CSTs in ambulatory settings. CSTs showed moderate accuracy in estimating total sleep time (TST) and time in bed (TIB), but lower precision for sleep efficiency (SE), wake after sleep onset (WASO), and stage classification. REM and deep sleep estimates were particularly unreliable. Multi-sensor devices modestly improved basic sleep-wake detection, while contactless devices offered lower user burden but limited portability. Few studies included clinical populations such as individuals with insomnia, depression, or older adults. Only a small number explored CSTs in intervention delivery, and integration into clinical care remains minimal.

Summary:

CSTs may support general sleep monitoring under real-world conditions, especially for duration and circadian timing. However, accuracy remains limited for sleep staging and fragmentation, and most devices have not been validated in diverse or clinical populations. Ongoing improvements in device design, signal integration, and algorithm transparency are needed to enhance CST reliability and clinical relevance.

As awareness of the profound impact of sleep on overall health continues to grow^1–10, the popularity of sleep tracking technologies has surged. Alongside advancements in nearable and contactless devices, these technologies have transformed sleep health engagement, with increasing use in clinical and non-clinical settings^11–13. By enabling continuous, non-invasive monitoring outside clinical settings, consumer sleep tracking (CST) technologies hold significant potential for personalized, long-term health monitoring and advancing sleep medicine.

CSTs are commercially available sleep-tracking devices designed for general consumers, often in the form of wearable (e.g., smartwatches, rings, headbands) or contactless (e.g., bedside or under-mattress sensors) technologies. In 2023, over one-third of Americans reported using wearable devices to monitor their sleep¹⁴. Although CST wearables remain the most convenient, user-friendly, and affordable sleep tracking option for the public, research comparing wearables to polysomnography (PSG) has produced mixed results^15–18. As such, implementation of CST for clinical and research purposes has lagged behind the rate at which the general population has adopted their use. Importantly, while CST’s are not intended to replace PSG in clinical diagnosis, CSTs have gained popularity among both health-conscious individuals and patients with sleep concerns, prompting a growing interest in their real-world performance across diverse populations.

Despite the accessibility and convenience of wearables, the complexity of sleep health measurement demands more precise tools. Sleep health encompasses multiple dimensions including regularity, satisfaction, alertness, timing, efficiency, and duration¹⁹. Currently, polysomnography (PSG) is the gold standard for quantifying sleep. PSG is typically conducted in laboratories by trained technicians. PSG quantifies neurological and physiological changes associated with sleep, not sleep itself. Although PSG is the accepted gold standard which other devices measuring sleep should be quantified against, limitations including cost, the requirement to sleep in the lab or clinic, limited longitudinal data, and the technical expertise required, drive market demand for alternative and more accessible sleep monitoring strategies, like CSTs.

Given PSG’s limitations within ambulatory settings, alternative methods such as wearables have been implemented to quantify sleep since the 1970s^20–22 and have greatly accelerated over the past decades²³. These devices have been broadly categorized into traditional research/clinical-grade actigraphy, consumer-grade devices, and contemporary research/clinical-grade devices²³. They offer a wide range of sleep metrics, including scores, duration, sleep cycles, depth, and alarms, but many lack parallel clinical metrics, leading to questions about their accuracy and relevance to health. Historically, studies assessing these metrics were called “validation” studies, which could mislead by suggesting inherent device reliability. Now, the field uses the term “performance evaluation” to reflect that different populations and outcomes require different performance levels^23–25.

Best practice is to rigorously benchmark all new devices against PSG, as comparing devices alone does not meet the gold standard²⁶. In 2021, a standardized framework for analyzing performance evaluation studies was introduced, including R-based code and tools like Bland-Altman plots and epoch-by-epoch analysis²⁵. In 2024, a state-of-the-science paper provided recommendations for using wearable technology in sleep and circadian research²³. While this review doesn’t cover the specifics, we refer to these comprehensive resources for detailed instructions.

This demand for more accessible, long-term sleep monitoring solutions has spurred significant advancements in wearable and contactless devices. Recent reviews have highlighted the potential of these technologies as effective alternatives to PSG and traditional actigraphy²³. Zhai et al. (2023) found CST devices performed comparably to research-grade actigraphy in distinguishing between wakefulness and sleep, even surpassing it in accuracy^15,27. Similarly, a review by Birrer et al. (2024) evaluated the performance of wearable devices for sleep stage classification and found that the combination of accelerometry and photoplethysmography (PPG) enhances sleep state classification accuracy²⁸, a domain where wearable sleep tracking devices have had generally poor performance against PSG^15,29,30. However, concerns persist about the limitations of single-signal monitoring¹³, with researchers advocating for the integration of multi-sensor³¹, or dynamic monitoring³². Importantly, multi-sensor device monitoring achieves higher granularity and detail in sleep-related metrics²³. In instances where multi-modal or dynamic monitoring is not feasible, EEG-based systems, despite being single-signal, remain the runner up to PSG for sleep monitoring due to their higher performance in identifying all sleep stages²⁹. Encouragingly, some CSTs have shown substantial agreement with PSG for specific outcomes, reflecting notable advances in hardware and algorithms³³.

Increasingly, real-world studies are needed to understand how CSTs perform in ambulatory environments and among diverse populations. This review addresses that gap by evaluating CST performance in free-living conditions. Specifically, we examine: (1) how CSTs quantify sleep behaviors in naturalistic contexts, (2) the accuracy and limitations of CST-derived metrics, (3) CST usability and user experience, and (4) the potential for CSTs to improve sleep tracking and outcomes in specific populations. We conclude with key recommendations for future research and device design in this rapidly evolving field.

Literature Search

Methods

This review followed established rapid-review methodology as described by Tricco et al. (2017)^34,35. Inclusion criteria were studies that (1) assessed CST/devices in ambulatory settings and (2) quantified device performance and/or utility. Exclusions included duplicates or non-research abstracts at screening, studies published more than 10 years before the search window, evaluations focused on sleep-apnea detection or therapy devices, and studies that did not assess CSTs in ambulatory settings or report quantitative performance or utility. Separate searches were conducted across Medline, Scopus, and Web of Science using Boolean operators and MeSH terms where applicable. Search terms included combinations of “consumer sleep technology,” “consumer sleep tracker,” “wearable sleep device,” “actigraphy,” “sleep monitoring,” “sleep disorder,” “sleep health,” “performance,” “accuracy,” and “reliability.” Two reviewers independently verified the selection process to ensure accuracy and consistency. These terms were selected to identify studies that evaluated CST performance and/or utility in real-world or ambulatory settings. The initial search returned 5,760 publications. After removing 1,893 duplicates 29 publications met our final eligibility criteria. The original search was performed in September 2024 and then updated in April 2025. The flowchart in figure 1 below illustrates the full study selection process.

Figure 1. Flow diagram of search design and screening.

This figure summarizes the systematic search and screening process used to identify studies for inclusion. An initial total of 5,760 publications was retrieved from the Central Discovery Index. After removing duplicates and irrelevant abstracts, 1,458 articles remained. Studies older than 10 years were excluded, reducing the pool to 1,146 articles. A final screening identified 29 studies that assessed consumer sleep technologies in real-world, ambulatory settings.

Evidence Review

Performance Evaluation of Consumer Sleep Trackers

Figure 2 displays the feature availability and battery life of the CSTs included in this review. Figure 3 presents a heatmap based on studies that directly compared CSTs to PSG. Most studies reported accuracy metrics, bias estimates, or Bland–Altman analyses to assess agreement with PSG. To facilitate comparison across heterogeneous reporting formats, we applied study-specific thresholds defining high accuracy (≥85% agreement, no significant bias, or mean deviations ≤15 minutes), moderate accuracy (70–84% agreement or deviations of 16–30 minutes), and low accuracy (<70% agreement, significant directional bias, or statistically significant differences from PSG). These categories summarize CST performance across key sleep metrics, including TST, SE, SOL, WASO, and staging (REM and deep sleep). Full study listings and raw findings are provided in the e-table 1 and e-table 2.

Figure 2.

This figure displays the reported battery life and feature availability for a range of consumer sleep devices. Devices are listed on the y-axis, and features are shown along the x-axis, including accelerometry, photoplethysmography, heart rate variability, skin temperature, oxygen saturation, and other physiological signals. Green cells indicate that a feature is available, while gray cells indicate that a feature is not supported by the device. Battery life is shown using horizontal blue bars and is expressed in days. ACC = accelerometry; PPG = photoplethysmography; HRV = heart rate variability.

Figure 3.

This heatmap illustrates the level of agreement between consumer sleep technologies and polysomnography across several sleep measures. Each row represents a specific device, and each column represents a sleep metric, including total sleep time, time in bed, wake after sleep onset, sleep efficiency, and sleep staging accuracy. Green cells indicate high agreement, yellow cells indicate moderate agreement, and red cells indicate low agreement. Devices include both commercial products such as Fitbit, Withings, and Oura, and research-grade tools such as Bodymedia and Micro Motionlogger. TST = total sleep time; TIB = time in bed; WASO = wake after sleep onset; SE = sleep efficiency; PSG = polysomnography.

Total Sleep Time and Time in Bed

The included 29 studies, summarized e-table 3, collectively represent approximately 4,855 participants across North America, Europe, and Asia, with individual sample sizes ranging from 5 to 3,551. In a broader context comparing CSTs to benchmarks other than PSG, consumer sleep technologies demonstrate moderate accuracy when estimating total sleep time (TST) and time in bed (TIB) compared to reference standards such as sleep diaries or sleep logs. In a four-night validation study under free-living conditions, Klier and Wagner (2022) reported that the Fitbit Versa 2 underestimated TST by 5.7% and TIB by 5.3%, with substantial agreement based on Lin’s concordance correlation coefficients (TST CCC_Lin = 0.83; TIB CCC_Lin = 0.75)³⁶. Comparable performance was observed for the Polar Ignite, which underestimated TST by 3.6% and showed nearly perfect agreement (CCC_Lin = 0.85), and for the Garmin Fēnix 5X Plus, which overestimated TST by 5.9% (CCC_Lin = 0.76). In a 14-day study of physicians-in-training, Hassinger et al. (2024) found that the Fitbit Inspire showed minimal bias compared to sleep logs for TST (−2.7 minutes) and TIB (−4.3 minutes); however, agreement remained low (TST CCC = 0.22; TIB CCC = 0.20)³⁷. In contrast, Rajput et al. (2023) found that the Garmin Vivoactive 4 overestimated TST by approximately 70 minutes, with Bland-Altman limits of agreement ranging from −115 to +256 minutes, indicating substantial variability and low precision³⁸. In contrast, the Fitbit Alta HR showed no significant difference from the Actiwatch Spectrum Pro in estimating TST in a sample with DSM-5 insomnia disorder. Multiple studies using ActiGraph or Dreem 2 headbands as references (e.g., Kastelic et al., Gruwez et al., Kuosmanen et al.) reported TST overestimations ranging from ~22 to ~70 minutes, although some fell within a 10% mean absolute error range^39–41. TIB estimates similarly varied across devices. While Garmin and Polar devices underestimated TIB by ~2% and ~5.9%, respectively, Fitbit models ranged from underestimating TIB slightly (Versa 2) to not reaching equivalence thresholds in other studies⁴² (e.g., Intel Basis Peak, Garmin Vivosport).

Sleep Efficiency

Sleep efficiency (SE) estimates demonstrated generally less agreement across most consumer sleep technologies evaluated when compared to sleep logs. In Klier and Wagner (2022), the Fitbit Versa 2 underestimated SE by 3.0% (CCC_Lin = 0.12), the Garmin Fēnix 5X Plus showed a CCC_Lin of 0.05, and the Polar Ignite overestimated SE by 2.3% (CCC_Lin = 0.34), indicating weak to negligible concordance with sleep diary values³⁶. Similarly, Hassinger et al. (2024) found that the Fitbit Inspire overestimated SE by 2.4% compared to sleep logs, with a CCC of just 0.12³⁷. Gruwez et al. (2017) also reported statistically significant SE overestimation by the Withings Pulse (p = 0.02) and Jawbone UP, relative to PSG-based assessments³⁹. Ravindran et al. (2023) observed a 9.19% overestimation of SE (SD = 14.26%) by the Withings Sleep Analyzer, describing overall agreement as modest and unsatisfactory⁴³. Although Liang et al. (2018) reported relatively good agreement for SE using the Fitbit Charge 2, they found that the Neuroon EEG headband significantly over- or underestimated SE depending on signal quality, highlighting the importance of hardware reliability⁴⁴. Collectively, these findings underscore the limited reliability of SE outputs from CSTs (particularly with sleep diaries), even when duration metrics such as TST or TIB show moderate concordance with reference standards.

Wake After Sleep Onset

Wake after sleep onset (WASO) was among the least accurate metrics. Klier and Wagner (2022) found no agreement between wearable-derived WASO and diary values (CCCLin = −0.01 to 0.10)³⁶. Liang et al. (2018) reported WASO overestimation by Fitbit, and a signal-quality–dependent discrepancy in Neuroon estimates⁴⁴. In most other studies listed in the comprehensive table, WASO showed either no statistical significance or poor agreement relative to PSG, Actiwatch, or diary reference values. Sleep onset latency (SOL) and timing measures such as sleep onset and wake time were inconsistently reported but showed modest agreement in some studies. Hassinger et al. (2024) reported higher concordance for the Fitbit Inspire’s sleep onset (CCC = 0.78) and wake time (CCC = 0.72), despite low agreement for TST and SE³⁷.

Circadian and Rest-Activity Rhythm Metrics

Regarding rest-activity rhythms and circadian phase estimation, McMahon et al. (2023)⁴⁵ demonstrated that Fitbit Charge 2 and Inspire HR correlated strongly with Actiwatch 2.0 on metrics such as amplitude (r = 0.578), acrophase (r = 0.715), and relative amplitude (r = 0.72), suggesting that wearable-derived circadian features are suitable for rhythm analysis even when nightly sleep metrics are imprecise⁴⁵. Wu et al. (2025) similarly found that smartwatch-derived acrophases from activity, heart rate, and core body temperature correlated with Actigraph-derived acrophase (r = 0.96, 0.87, and 0.79, respectively), and significantly predicted self-reported chronotype (R² = 0.64, p < 0.001)⁴⁶.

Sleep Staging

Sleep staging remains the most problematic outcome to validate. Most devices did not provide access to raw data for independent stage classification and studies lacked PSG comparisons. As explicitly noted by Klier and Wagner (2022), staging could not be assessed due to lack of diary-based benchmarks and inaccessible manufacturer algorithms. von Gall et al. (2024) used Fitbit-generated staging data to explore circadian timing of REM and deep sleep but did not validate those stages against PSG, limiting conclusions about staging accuracy⁴⁷.

User Interaction and Experience

Sleep-tracking devices are typically categorized as wearable (e.g., wristbands, rings, headbands) or contactless (e.g., mattress-based or bedside sensors). Wearables use accelerometry, photoplethysmography (PPG), and body temperature to estimate sleep metrics, while contactless devices rely on ambient sensors such as pressure, sound, or radar. Although wearable devices offer greater portability and often higher data granularity, their short battery life, discomfort, or displacement during sleep can contribute to missing data and reduced adherence⁴¹.

Contactless devices are less intrusive and typically more comfortable, especially for long-term use or among older adults and individuals who find wearables bothersome. Their ease of use, lack of battery limitations, and passive operation make them ideal for population-scale deployment, although they often collect fewer physiological signals than wearable counterparts and may have limited portability^27,43.

Beyond hardware, user interface design and data presentation significantly affect perceptions of credibility and engagement. Devices that are intuitive, aesthetically pleasing, and easy to interpret tend to be used more consistently⁴⁸. However, technical barriers, such as issues downloading software or syncing data, can undermine engagement, especially in older or less tech-savvy users⁴¹.

Emerging evidence also points to potential psychological consequences of sleep tracking. While rare, conditions such as orthosomnia, an unhealthy preoccupation with improving sleep through technology, have been observed in some users⁴⁹. Additionally, device notifications and continuous tracking may contribute to phantom phone signals (PPS) and heightened device awareness, which could interfere with sleep or increase stress^50,51. More research is needed to evaluate the long-term psychological impact of CSTs, especially in populations with sleep disturbances.

To improve both user experience and data quality, some researchers recommend streamlining data directly to researchers to reduce participant burden, while still allowing limited self-monitoring to enhance engagement. Overall, balancing user autonomy, ease of use, and data reliability remains critical for CST adherence and real-world performance evaluation.

Population-Specific Findings

Insomnia Populations

Five studies focused on individuals with insomnia. In one study, Hamill et al. (2020) evaluated the Fitbit^® Alta HR against the Actiwatch^™ Spectrum Pro among 25 adults diagnosed with DSM-5 insomnia disorder⁵². Data were collected in a home setting over two weeks, including one week before and after completing cognitive behavioral therapy for insomnia (CBT-I). No significant differences were observed between devices, suggesting clinical utility of the CST for insomnia populations. Two studies assessed the use of CSTs during CBT-I intervention delivery. Kang et al. (2017) examined CBT-I delivered through a mobile application with or without a wearable device (Fitbit^® Charge HR)⁵³. Wearable use was not randomized. Participants using wearables had lower baseline insomnia severity and PSQI scores, but wearable use did not significantly impact treatment efficacy, adherence, or satisfaction. Similarly, Luik et al. (2018) analyzed 3,551 adults with insomnia enrolled in digital CBT-I (Sleepio^™)⁵⁴. Among those, 378 connected a personal wearable device (Fitbit^® or Jawbone^™ UP^®). Participants who chose not to use a wearable had better sleep outcomes at baseline and post-therapy, suggesting that wearable use was not necessary to derive benefit from the intervention. Ojalvo et al. (2023) conducted a randomized controlled trial comparing Fitbit^® Inspire^™ 2 use plus a self-help sleep guide to a sleep diary–only condition in patients meeting ICD-10 criteria for insomnia⁵⁵. Both groups improved on sleep quality (measured via the PSQI), anxiety, quality of life, and sleep reactivity. However, the diary-only group had greater improvements in sleep duration, and no differences were observed in any psychological or sleep outcomes between groups after four weeks, indicating minimal added value from the wearable. In a smaller feasibility study, Pulantara et al. (2018) evaluated a mobile health application (iREST) that integrated clinician messaging and insomnia-tailored content, such as a nightmare module⁵⁶. Among 19 active-duty service members and veterans with insomnia, only 7 connected a wearable (Fitbit^® Charge).

Older Adults and Contactless Devices

Ravindran et al. (2023) assessed sleep estimation among 35 older adults (mean age = 70.8) using the Withings^® Sleep Analyzer and Emfit QS^®, both contactless under-mattress sensors⁴³. Comparison devices included sleep diaries and Actiwatch^™ Spectrum. CSTs overestimated TST (92.7–101.5 minutes) and SE (9.2–9.4%), while underestimating WASO and SOL. While agreement was modest, the contactless sensors showed better concordance with diary-reported in-bed naps than wrist-worn actigraphy, and participants reported the contactless format as more comfortable.

Depression Populations

Zhang et al. (2021) studied the relationship between sleep parameters and depression in 368 participants across the Netherlands, Spain, and the UK, all with a recent history of depression. Participants wore Fitbit^® Charge 2 or 3 devices for at least 12 nights⁵⁷. Higher TST, longer time in bed, increased awake time, and lower sleep efficiency were all significantly associated with worse depression scores on the PHQ-8 (p<.001). Deep sleep, REM sleep, and REM latency showed no significant associations with depression, while lower sleep efficiency and higher nighttime awakenings were significantly associated with poorer sleep subscale scores. These findings indicate that extended or fragmented sleep may be predictive of mood deterioration.

Short Sleep Duration Interventions

Our team has explored the use of CSTs for sleep extension among adults with short sleep duration. In a factorial design trial, participants were randomized to one of four groups: Fitbit-only, coaching-only, Fitbit plus coaching, or self-management. Results indicated no significant differences in sleep extension across groups. However, feasibility assessments revealed high satisfaction with the Fitbit and coaching interventions⁵⁸, suggesting that CSTs are acceptable tools in short sleep populations even if their additive effects on outcomes may be limited.

Future Directions

Future advancements in consumer sleep technologies (CSTs) may rely heavily on machine learning, deep learning, and personalized algorithms. These methods have the potential to improve sleep scoring, staging, predictive modeling, and adaptation to individual sleep profiles^15,30,59,60. Integrating multi-modal data, such as accelerometry and photoplethysmography (PPG), has shown improved staging accuracy^27,28, especially when used in personalized rather than generalized models⁶⁰.

Open access to raw data and open-source algorithms could enhance scientific rigor, lower barriers to innovation, and facilitate clinical applications⁶¹. However, future CST evaluations must account for real-world settings, population diversity, and the influence of contextual factors. Providing users the option to indicate intended sleep onset and wake times would help distinguish actual sleep from time in bed, improving algorithmic accuracy and clinician interpretation⁶². Additional features such as environmental sensors (e.g., light, noise, temperature), behavioral inputs (e.g., caffeine, screen use, physical activity), and real-time feedback on signal quality may reduce data loss, particularly among technology-naïve users^63,64.

CSTs often employ non-clinical terms like “deep sleep” or black box scores, which may confuse users and clinicians. Establishing standardized terminology and adopting multi-signal validation frameworks are essential for alignment across consumer and clinical domains²⁸. CSTs now commonly appear in clinical contexts, for example, the Apple Watch’s FDA-cleared sleep apnea detection and Samsung’s pending cuffless blood pressure feature, despite clinicians frequently lacking training in wearable data interpretation⁶⁴. Integration into care pathways, clinician education, and professional guidance from groups like CHEST and AASM will be essential for safe adoption⁶³.

Longitudinal field studies underscore the value of CSTs in detecting behavior-driven disruptions that would be missed in the lab⁶⁵. For example, a 65-day field study using a wearable ring revealed user-identified disruptions like irregular schedules and evening screen use that are often undetected in controlled settings⁶⁵. Massive app datasets have also shown seasonal variation in sleep duration and quality⁶⁶. Validation studies suggest CSTs can track sleep reliably in irregular-schedule populations like shift workers, though contactless devices often underperform^61,67. In clinical samples, devices like the Jawbone UP3 have shown moderate agreement with PSG, while others have demonstrated variable accuracy⁶⁷.

Real-world evaluations must increasingly involve diverse populations, including older adults, adolescents, and individuals with disrupted sleep patterns, where device performance is often weaker^67–69. Studies such as Hamill et al. (2020) show varied results across insomnia and depression cohorts, reinforcing the need for population-tailored assessments⁵². The rapid pace of hardware and software updates also complicates efforts to maintain up-to-date evaluation, highlighting the need for continuous validation.

Emerging innovations such as graphene electronic tattoos and nanofiber-based tactile sensors may eventually offer ultra-thin, skin-integrated alternatives to traditional wearables. These flexible biosensors can noninvasively measure EEG, ECG, respiration, hydration, and other metrics with high fidelity and comfort^70–73. While some CSTs are expanding into cardiopulmonary monitoring, this review focused solely on sleep detection in ambulatory settings. Together, advances in personalized modeling, transparent data access, wearable design, and ethical implementation can support the evolution of CSTs from wellness tools to clinically viable adjuncts that bridge patient behavior and medical decision-making.

Summary

In conclusion, CSTs offer transformative potential for sleep health monitoring but remain estimations of PSG, with no consensus on performance standards. Clinical interpretation should be cautious, given the need for broader validation across populations. Still, CSTs may perform more reliably in healthy individuals and provide unique insights into behavioral and environmental factors often missed by PSG, supporting health behavior awareness, contextualized interpretation, and engagement with real-world, longitudinal sleep data.