Implications of Wearable Sleep Health Technology for Cardiovascular Health Assessment and Interventions: A Narrative Review

Abstract

Sleep health is a multidimensional component of cardiovascular health, and modulating sleep as a cardiometabolic risk factor requires accurate measurement. Numerous technologies have emerged in the last 50 years to longitudinally characterize sleep over time in ambulatory settings, and their use by the public has grown. The impact of this democratization of health information through wearables has not been adequately explored in the literature. This narrative review equips clinicians and researchers with a mechanistic understanding of wearable sleep health technology to augment precision cardiovascular health assessment and intervention.

Peer-reviewed literature was synthesized from PubMed, spanning foundational actigraphy studies through recent multi-sensor and electroencephalography (EEG)–based device evaluations, with emphasis on device performance against polysomnography (PSG). Consistent with a narrative review, sources were selected through domain expertise rather than systematic screening. Wearable devices measure peripheral proxies to determine if the wearer is awake or asleep. The simplest method, actigraphy, uses motor activity to differentiate sleep from wake. Newer multi-sensor devices utilize actigraphy in addition to other physiological proxies of sleep (e.g. heart rate, heart rate variability, skin temperature). Current device heterogeneity can diminish the validity of conclusions can be drawn from wearable data; thus performance evaluation against polysomnography is vital to determine the accuracy, sensitivity, and specificity of a device. Across evaluations, devices consistently show high sensitivity for detecting sleep but lower specificity for detecting wake; several consumer devices nonetheless perform comparably to research-grade actigraphy, some have received regulatory authorization for sleep-apnea screening, and headband EEG systems approach PSG-level staging. Wearable devices show promise as objective anchors to drive provider-patient interactions in cardiometabolic lifestyle interventions; offering scalable, valid measurement tools to quantify sleep health and inform cardiovascular risk assessment. Interpretation requires awareness of structural device limitations, optimally through the understanding of the physiological concepts which underpin the function of these devices.

Graphical Abstract

INTRODUCTION

The American Heart Association (AHA) now includes sleep duration within Life’s Essential 8 framework ¹. A recent scientific statement asserted that this framework should move beyond sleep duration alone, instead conceptualizing sleep health as a multidimensional construct ². Expanding and clarifying Life’s Essential 8 framework to consider many sleep dimensions – duration, timing, regularity, satisfaction, functioning, continuity – provides a more complete risk profile to reflect how sleep supports cardiometabolic health (CMH) across the lifespan. How best to assess multidimensional sleep health remains an open question, especially given the recent advancement, abundance, and availability of different research- and consumer-grade devices.

Sleep wearables leverage the stability of physiologic signals that occur during sleep to make sleep/wake determinations. This paradigm is represented visually in Figure 1. In 2025, the World Sleep Society has published recommendations on wearable consumer health trackers that monitor sleep. The document includes actionable recommendations for consumers, clinicians, researchers and device manufacture to promote consistency between device outputs and user experience. To align sleep health technology with clinical applications the authors organize their robust recommendations in a case specific manner. This review aims to compliment that work by synthesizing a mechanistic interpretation that is otherwise scattered amongst the literature³. Advancements in sleep technology offer greater opportunities for intervention, an increasingly valuable tool for the modulation of sleep variables as a modifiable risk factor of disease. This narrative review thus outlines how current wearable technology measures sleep health and discusses the use of wearable technology in cardiometabolic risk assessment, management, and research.

Figure 1.

Sleep Measurement Paradigms

Note. Schematic representation of sleep measurement models comparing central nervous system (CNS) changes, captured by EEG-based wearables, to peripheral changes measured by actigraphy and multi-sensor devices (e.g., movement transducers, PPG, SpO2, EDA). The figure illustrates parallel assessment with polysomnography (PSG), including detection modalities (EEG, EMG, EOG, ECG, respiratory effort, airflow, SpO2) and indicates performance evaluation pathways. PPG = photoplethysmography; SpO2 = pulse oximetry; EDA = electrodermal activity; EEG = electroencephalography; EMG = electromyography; EOG = electrooculography; ECG = electrocardiography. Figure by Michael Amadei, 2025.

It is critical that evaluations of newer objective sleep measurement devices be assessed relative to the gold standard of PSG⁴, but it is prudent to note that the gold standard itself is imperfect and there may be benefits of other devices that improve upon the gold standard in some ways, while falling short in others^5,6. Compared to discrete contributors to cardiometabolic health, sleep is measured indirectly through the presence or absence of physiologic features across numerous organ systems. The effects of insufficient sleep occur both acutely and chronically, heavily influenced by behavior on a day-to-day basis. The result is in a less stable signal – presenting unique challenges compared to other CMH factors. The inability of PSG to characterize the aggregate effect of sleep over time is exactly why sleep wearables are of great interest. Another longitudinal sleep measurement tool, a sleep diary, filled out in the morning upon awakening can be subject to self-report bias and requires active engagement, potentially contributing to nonadherence and makes them arguably less suitable for ultra-long term sleep characterization⁷.

This work is a narrative rather than systematic review. Relevant literature was identified primarily through PubMed/MEDLINE, supplemented by review of key reference lists and primary sources, spanning foundational actigraphy studies of the 1970s–1980s through contemporary device evaluations published to 2025. Consistent with the aims of a narrative synthesis, sources were selected through author expertise to represent the conceptual and historical development of wearable sleep measurement and its application to cardiometabolic health, rather than through exhaustive systematic screening. This approach is well suited to a heterogeneous and rapidly evolving evidence base and to providing the mechanistic, clinically oriented context that a systematic review’s narrower scope may not capture⁸

BACKGROUND

As described in Grandner and Fernandez (2021)⁹, “Sleep is generally defined as a naturally recurring, reversible state of perceptual disengagement, reduced consciousness, and relative immobility, the propensity of which is patterned by homeostatic and circadian factors.” As such, there is no single, direct measurement of sleep in humans. Instead, sleep is measured using relatively reliable and valid proxies that reflect sleep physiology, see Figure 1. The gold standard measurement of sleep is polysomnography (PSG), an approach primarily based on electroencephalography (EEG) but that also leverages other signals, including electrooculography (EOG), electromyography (EMG), electrocardiogram (ECG), body position, and respiration. Although PSG captures many physiologic signals related to sleep in real-time, PSG does not directly measure sleep-wake regulation or completely capture all the physiology that differentiates sleep from wake ¹⁰. Sleep-wake regulation is driven by subcortical processes that result in diffuse and sometimes disparate physiologic sequelae.

Modern wearable sleep assessment devices are built upon accelerometer technology known as actigraphy, from which sleep-wake status is inferred based on the absence of motor activity. Actigraphy relies on the stability between the physiological effect of sleep on motor tone ¹¹ and the neural phenomena of sleep as brain activity is modulated by structures in the midbrain and brainstem ¹². First investigated in the context of psychomotor changes and sleep changes in depression, tethered movement transducers showed correlation between measurement movement and clinical state ¹³. Later, a set of actigraphy records were compared to epoch-matched PSG, wherein the two methods agreed on sleep-wake determinations 94.5% of the time ¹⁴, suggesting that wrist-based movements produced reliable estimations of sleep vs. wake. In 1982, an algorithm was applied to the actigraphic signal in which the changes in activity level were summed and then scored, agreeing with PSG 93.9% of the time ¹⁵, offering the potential of an accurate, economical and scalable solution to sleep measurement. Consequently, actigraphy is relatively inexpensive and enables micro-longitudinal observation in free-living conditions, making it a valuable resource in sleep and circadian research ¹⁶.

Simple actigraphy continues to be used and – through iterative improvements in technology – now uses more sensitive accelerometers and includes photo plethysmographs, skin temperature sensors, ambient light sensors, pulse oximetry, among other additional technologies. Alongside the implementation of more complex algorithms that consider an increasing number of variables, a distinct class of multi-sensor sleep wearables has emerged, henceforth referred to as multi-sensor devices. ¹⁷. In simple actigraphy, when the device records more movement, the wearer has a higher probability of being awake; when the device records a static state of non-movement, the wearer has a higher probability of being asleep. This assumption is augmented in multi-sensor devices by updating probabilities of sleep vs. wake determinations using additional physiological signals ¹⁸. Refer to Figure 2 for an overview of wearable classes.

Figure 2.

Common Sleep Wearables

Note. Overview of wearable sleep health technology modalities, including actigraphy, wrist-worn and finger-worn multi-sensor devices, and headband-based EEG solutions. Key features, measurement principles, and comparative values and limitations are summarized for each device type. HR = heart rate; HRV = heart rate variability; Temp = skin temperature; SpO2 = pulse oximetry; RR = respiratory rate; EDA = electrodermal activity; EEG = electroencephalography. On-board sensors may vary across devices Figure by Michael Amadei, 2025.

Photoplethysmography (PPG) is a non-invasive optical technique which relies on the reflection of a specific wavelength of light emitted from a photodiode in a device to measures blood volume changes in a microvascular bed within tissue ¹⁹. This allows for the calculation of pulse rate (PR) pulse rate variability (PRV), and pulse oximetry (SpO2), functioning both as independently relevant measures and informative to sleep architecture determinations ²⁰. Most often at a distal extremity, PR and PRV are often reported as a proxy for heart rate though the measure accumulates signal artifact as the pulse wave travels, PRV has been reported in the literature to be a somewhat acceptable surrogate for HRV, in the case of sleep assessment pulse measures appear sufficient to inform relative autonomic activation as it pertains to the staging of sleep²¹. PRV and PR are informative as parasympathetic activity increases and sympathetic activity withdraws as sleep deepens from light to slow-wave sleep. During non-rapid eye movement (NREM) sleep, PR typically declines and PRV increases, reflecting greater vagal tone and cardiac autonomic stability, whereas rapid eye movement (REM) sleep is characterized by autonomic instability marked by transient PR elevations and reduced, more erratic PRV that mimics wake-like variability²². Thus, PR and PRV trends allow proprietary algorithms to probabilistically classify epochs into wake, light, deep, and REM stages by incorporating prior epoch context and stage transition probabilities, rather than derived from a single physiological signal alone ^16,23.

Several devices measure skin temperature as part of their multisensory array. There is a lack of evidence for the utility of skin temperature on its own as a sleep-wake or circadian signal. Still, some devices use skin temperature to derive core body temperature²⁴. In addition, skin temperature may aid in the detection of sleep-wake and circadian states even if not useful on its own^25,26.

Multi-sensor devices have rapidly proliferated into the consumer space, offering individuals access to their sleep and other health-related data. This class of wearables are typically combined with smartphone-based apps, offering user interfaces with features extending beyond sleep tracking ²⁷. Sleep tracking features are also present in many products that are not explicitly purchased for that purpose. Yet, devices vary widely and evolve quickly, and the clinical implications remain unclear. Increased user engagement with their health data presents both opportunities and challenges for advancing cardiovascular health. A consensus must be reached on how to interpret this relative data abundance within clinical context, especially for large scale disease screening and individualized risk projections.

Sleep characterization by digital devices is not limited to devices directly attached to skin, because a broader category of “nearables” also exists. An example of nearables is in the pocket of many people – the modern, ubiquitous smartphone can leverage onboard sensors to infer the sleep state of the user, including accelerometers, microphones, and moment-to-moment data about pickups and use²⁸. Several applications, both those native to devices and third party apps, use these internal functions to discern sleep-related variables. In addition, sleep determinations based solely on smartphone use data (separate from sensor data) also shows promise and is a testament to the integration of technology into the lives of modern humans²⁹. Other nearables include under-mattress sensors and instrumented mattress toppers, which infer sleep from motor activity and ballistocardiographic signals, as well as bedside devices that use sonar or radar to detect respiratory effort and gross body movement.²⁷ Because these devices do not contact the skin, signal attribution becomes a practical limitation: a sensor spanning the sleep surface cannot readily distinguish the intended user from a bed partner, an ambiguity that does not arise with worn devices. The validation paradigm for this class also remains immature, with fewer performance evaluations against polysomnography and less methodological standardization than exists for wearables. Notably, some temperature-modulating mattress toppers incorporate on-board sleep measurement to inform thermal intervention, blurring the distinction between measurement and intervention that otherwise characterizes consumer sleep technology.

EVALUATION OF SLEEP TECHNOLOGY

The concept of “validation” as it pertains to wearable sleep devices has changed somewhat over the past several decades. Initially, “validation” studies of actigraphy compared epoch-by-epoch sleep-wake determinations to PSG ³⁰. Researchers later recognized that simple agreement rates overestimate accuracy despite suboptimal performance of the device. A typical night of sleep in healthy individuals (the subjects of most “validation” studies) consists almost entirely of sleep. An algorithm that is optimized to detect sleep in the sleep period will thus generate high rates of agreement despite flaws in the detection algorithm. For example, if a typical night consists of 90% sleep epochs, and an algorithm scored every epoch as “sleep” without variability, the algorithm would have achieved 90% agreement despite being a terrible algorithm.

Epoch-by-epoch analyses evolved to incorporate separate evaluations of “sensitivity,” “specificity,” and “accuracy.” Sensitivity refers to the % of epochs identified as sleep by the PSG that are correctly identified by the comparison device; specificity refers to the number of wake epochs correctly identified; and accuracy refers to the total number of epochs correctly identified ³⁰. As such, specificity has emerged as the key metric differentiating devices. The test of a device is not the ability to detect sleep when sleep is likely – the test of a device is instead the ability to detect wake where wake is unlikely. See Figure 3, a confusion matrix, which is a common reporting format for the results of performance evaluations.

Figure 3.

Performance Evaluation Confusion Matrix

Note. Epoch-by-epoch comparison framework for sleep/wake classification between wearable devices and polysomnography (PSG). True positive (TP) and true negative (TN) values represent correctly identified sleep and wake epochs by the device, while false positive (FP) and false negative (FN) values denote misclassifications relative to gold standard PSG. Sensitivity reflects the proportion of detected sleep epochs (TP/[TP+FN]), and specificity reflects the proportion of detected wake epochs (TN/[TN+FP]). Figure by Michael Amadei, 2025.

Sleep assessment technology achieves high sensitivity in detecting sleep yet achieves lower specificity ³¹. Given the basic mechanics of actigraphy, quiet wakefulness is a key limitation, when movement is low but the wearer is not asleep ³². Quiet wakefulness is most salient when determining sleep onset latency, that is, how long it takes the wearer to fall asleep from the start of their sleep attempt ³². A person lies still, relaxes, and prepares for sleep until they fall asleep. An algorithm based on movement alone may have difficulty distinguishing quiet wakefulness from sleep; for this reason sleep-wake detection in insomnia is especially difficult, making sleep diaries more appropriate most of the time ¹⁰.

Comparing devices to PSG is an additional layer of abstraction between the device of interest and the sleep it intends to measure. As with any comparison, it is important to scrutinize the quality of the reference or control measure. PSG is considered the gold standard tool to measure sleep as it captures both the central and peripheral effects of the state of sleep on the human body. Consequentially, PSG has been vital to the field’s understanding of sleep, with many current guidelines are based on it ³³. However, PSG still relies on human scoring, and interrater variability may be present among PSG records, such that two scorers may score the same record differently even though they used the same guidelines. One meta-analysis found the highest rates of consensus for wake and REM sleep, the lowest rate of consensus for N1, and overall assessed agreement to be substantial on the Cohen’s kappa at 0.76 ³⁴. Manual scoring is reliable, although this variability should be appreciated.

The field has largely departed from considering any single study a “validation” of the specific device, instead considering the study a “performance evaluation” ³⁵, formally describing how a specific device performs within a specific context. This is a critical distinction. On one hand, it cannot be assumed that a device that detects sleep-wake differences in a highly controlled environment among healthy individuals will perform equally well in sleep-disturbed individuals at home. When critically appraising a wearable use case it is prudent to ask, When and where is the device accurate? And how can results be interpreted differently in different contexts?³⁶ On the other hand, formal evaluations in every possible sub-population do not need to exist in order to be able to clinically or empirically evaluate the data obtained ³⁵. These concepts define guidelines for assessing and reporting device performance ³⁷.

ESTIMATION OF SLEEP STAGE FROM WEARABLE DEVICES

To predict sleep stage (e.g., N1, N2, N3, REM), algorithms incorporate variables like PR, PRV, circadian proxies, past epoch staging, etc., in addition to movement ¹⁸. A meta-analysis of 62 evaluated wearable configurations found that sleep stage determination is superior in devices utilizing PPG in addition to actigraphy rather than those using purely actigraphy ¹⁸. An evaluation of multi-sensor wearables found that the devices had an accuracy of 60–80% to stage sleep ³¹. Notably, most devices cannot differentiate all sleep stages and typically combine multiple stages (e.g., “light” sleep comprising of stages N1 and N2 ¹⁸, which may be difficult to distinguish phenomenologically yet nonetheless represent very different physiological processes ³⁸. Figures 4,5, and 6 present the performance of six multi-sensor devices against PSG in staging Light, Deep and REM sleep respectively.

Figure 4.

Performance of 6 Devices Against PSG in Detecting Light Sleep

Note. Comparison of epoch-by-epoch (EBE) agreement in light sleep detection across six consumer sleep monitoring devices (Fitbit Alta HR, Garmin Fenix 5S, Garmin Vivosmart 3, EarlySense Live, ResMed S+, SleepScore Max). Accuracy, sensitivity, and specificity metrics are presented for each device relative to polysomnography. Sensitivity indicates the proportion of true light sleep epochs detected; specificity reflects correct identification of non-light sleep epochs; accuracy is the overall proportion of correct classifications. Adapted from Table 9 ³¹

Figure 5.

Performance of 6 Devices Against PSG in Detecting Deep Sleep

Note. Comparison of epoch-by-epoch (EBE) agreement in deep sleep detection across six consumer sleep monitoring devices (Fitbit Alta HR, Garmin Fenix 5S, Garmin Vivosmart 3, EarlySense Live, ResMed S+, SleepScore Max). Accuracy, sensitivity, and specificity metrics are presented for each device relative to polysomnography. Sensitivity indicates the proportion of true deep sleep epochs detected; specificity reflects correct identification of non-deep sleep epochs; accuracy is the overall proportion of correct classifications. Adapted from Table 10 ³¹.

Figure 6.

Performance of 6 Devices Against PSG in Detecting Rapid Eye Movement (REM) Sleep

Note. Comparison of epoch-by-epoch (EBE) agreement in rapid eye movement (REM) sleep detection across six consumer sleep monitoring devices (Fitbit Alta HR, Garmin Fenix 5S, Garmin Vivosmart 3, EarlySense Live, ResMed S+, SleepScore Max). Accuracy, sensitivity, and specificity metrics are presented for each device relative to polysomnography. Sensitivity indicates the proportion of true REM sleep epochs detected; specificity reflects correct identification of non-REM epochs; accuracy is the overall proportion of correct classifications. Adapted from Table 11 ³¹.

A recent assessment of six commercially available multi-sensor wearables compared with PSG found the devices to be variable in their agreement with PSG ³⁹. Of the devices tested, the Apple Watch Series 8, the Fitbit Charge 5, and the Fitbit Sense demonstrated clinically acceptable estimation of total sleep time and sleep efficiency but had discrepancies across other sleep metrics. The Cohen’s kappa coefficients of these devices were 0.53, 0.41, 0.42, respectively, indicating moderate agreement ³⁹. The assessment suggested that these devices’ performance is likely sufficient to identify gross trends, especially across the long-term, but fail to meet many thresholds of clinical utility ³⁹.

A recent performance evaluation, which have examined multiple devices worn at the same time, found that the consumer devices, Fatigue Science Readiband, Fitbit Alta HR, EarlySense Live, ResMed S+, SleepScore Max, with the exception of the Garmin devices, performed similarly or better than actigraphy when compared with PSG ³¹. This supports that these consumer grade devices are equivalent to actigraphy, which has been the research standard for ambulatory sleep measurement for years. Notably, these devices, like actigraphy still struggle with wake identification, resulting in lower specificities, resulting in poorer performance on nights with poorer sleep ³¹. A performance evaluation of the Oura Ring Gen 3 against ambulatory PSG found that the device has a sleep-wake sensitivity of 94.5% and a sleep-wake specificity of 73.0% ⁴⁰. The Huawei Watch GT2 has been evaluated relative to polysomnography, showing high sensitivity (95.9%) and accuracy (87.3%) but relatively low specificity (47.9%) for sleep/wake detection, with agreement being less consistent for detailed sleep-stage classification; the device also tended to overestimate total sleep time and sleep efficiency ⁴¹. In addition, the watch was compared to another actigraphy device and showed good agreement ⁴²

The Happy Ring is a finger worn multi-sensor that uses two algorithms, one personalized and one generalized to process data from on-board sensors. The Happy Ring, similar to the Oura, Whoop, Fitbit, and Actiwatch, which all demonstrated high levels sensitivity against PSG and the Dreem 2 Headband,. The personalized algorithm leverages machine learning to dynamically adapt to the wearer, yielding high specificity observed in lab (0.83) ⁴³.The specificities of Oura, Whoop, Fitbit, and Actiwatch were 0.49, 0.54, 0.49, 0.19 respectively against in-lab PSG. Refer to Figures 7&8 for device performance across both settings.

Figure 7.

Performance of 6 Devices Against In-lab PSG: Sleep vs. Wake

Note. Comparison of accuracy, sensitivity, and specificity for six consumer and research sleep monitoring devices (Happy Ring—Personalized, Happy Ring—Generalized, Oura, Whoop, Fitbit, Actiwatch) evaluated against in-lab polysomnography (PSG). Metrics reflect each device’s ability to detect sleep and wake epochs relative to gold-standard PSG. Sensitivity represents correct detection of sleep epochs; specificity measures correct detection of wake epochs; accuracy is the overall rate of correct classifications. Adapted from Table 2 ⁴³

Figure 8.

Performance of 6 Devices Against In-home Dreem 2: Sleep vs. Wake

Note. Comparison of accuracy, sensitivity, and specificity for six sleep monitoring devices (Happy Ring—Personalized, Happy Ring—Generalized, Oura, Whoop, Fitbit, Actiwatch) evaluated against in-home Dreem 2 EEG-headband. Metrics reflect each device’s capability to correctly classify sleep and wake epochs relative to Dreem EEG. Sensitivity indicates the proportion of true sleep epochs detected; specificity represents correct identification of wake epochs; accuracy is the overall proportion of correct classifications. Adapted from Table 2 ⁴³.

Ultimately, several consumer-grade devices have shown to be sensitive, specific, and accurate enough to facilitate application and assessment of sleep recommendations as discussed in Life’s Essential 8. Characterization of sleep habits over time is less subject to recall bias than self-report by patients. Sleep diaries may ameliorate some of this recall bias but accordingly present adherence concerns.

WRIST-BASED VS. FINGER-BASED WEARABLES

Two of most common form-factors that wearable devices take are that of a wrist-worn bracelet or watch and that of a finger-worn ring. An assessment of signal quality between the two sampling locations based on PPG to determine HR, SpO2 and RR found the finger to have significantly less median error across the three metrics than the wrist ⁴⁴. The combined median error for walking and at rest was 41% for wrist-worn PPG and 13% for finger-worn PPG ⁴⁴. Still, the use of a ring-worn sensor can be limited by several factors. Rings are typically not adjustable, meaning that an improperly sized ring may impact signal quality. Ring devices cannot adapt to a change in body habitus over time. Finger circumference is often not fixed throughout the day, and a ring that is appropriately sized for good signal quality might become too tight as a finger swells due to limb dependency. Additionally, clinicians and researchers can struggle with resource allocation for which ring sizes to purchase. Rings are less amenable to certain forms of exercise due to risk of injuries to the finger or damage to the device ⁴⁵. Signal quality and the wearer’s preference of formfactor should be considered to promote adherence.

HEADBAND-BASED WEARABLES

Headband-based wearables are ambulatory EEG monitors, a cornerstone measure of PSG, worn on the head that measure changes in transcranial voltage over fewer channels than traditional, lab-based EEGs ⁴⁶. The primary rationale for the use of these devices over other wearables is that the direct measure of brain waves is superior to the measure of peripheral physiological changes in the measure of sleep, especially as it pertains to sleep scoring. Indeed, headband-based wearables are often used in performance evaluations of multi-sensor and actigraphy devices in home-based settings ⁴³.

A systematic review of forty-two publications evaluating 24 different EEG-based wearables demonstrated that ~70% of headband-based wearables studied (17 devices) could reliably differentiate between the three distinct stages of non-REM sleep ⁴⁷.The performance evaluation of the Dreem Headband, a limited-montage dry EEG device with an on-board accelerometer, demonstrated comparable performance to PSG for sleep scoring, with an 83.5% agreement with PSG ⁴⁸. In this evaluation, each sleep expert’s manual scoring was compared to a consensus hypnogram built from the other four scorers, ensuring fair evaluation against a collective standard rather than a single individual; this consensus scoring yielded an average of 86.7% for any one scorer relative to the others. This finding highlights inter-rater variability in PSG scoring. Thus, the Dreem Headband’s automated sleep staging was similarly compared to a consensus derived from the scorers. Another ambulatory EEG device, the MUSE-S, was evaluated for PSG agreement in both heath sleepers and people with sleep apnea. The performance evaluation revealed substantial agreement with PSG in both groups, with the lowest agreement in NREM1 sleep. This transitional phase is a common point of ambiguity in wearable devices, with the MUSE-S scoring a Cohen’s Kappa, a measure of agreement, of 0.41 in the healthy sleeper group and 0.33 in the sleep apnea group ⁴⁹, with the device scoring 0.74 across all epochs and 0.87 for wake in a another trial⁵⁰. A performance evaluation that considered the night-to-night variability and accuracy of the Sleep Profiler, a frontopolar ambulatory EEG, found that substantial agreement between a consensus of reviewers and the automated scoring system ⁵¹. Notably, the clinical population of this evaluation consisted of various sleep comorbidities, including insomnia, obstructive sleep apnea, and depression, indicating that the device is appropriate for use in populations with disordered sleep.

Headband-based wearables that use EEG are closely aligned with PSG both mechanistically and in their performance evaluations. Their cumbersome form factor, high cost, and absent accelerometers all generally make headband-based wearables inferior to smaller devices that can be worn continuously, especially as they pertain to application in cardiovascular medicine. To a certain extent they may even be considered over-powered for sleep assessment that is currently relevant to heart health. The singular use of these devices for more accurate characterization of sleep may leave clinicians without other metrics useful in the assessment of CMH, like daily activity, HR, HRV, SpO2, EDA, etc. However, these devices ultimately enable high fidelity sleep characterization that is vital for various use cases in research and beyond.

USE CASES

Screening for Sleep Disorders with Wearables

In general, commercially available devices are not approved as diagnostic devices for sleep disorders. The information generated by devices may be useful, but most devices have not yet risen to the level of demonstrated diagnostic utility, with a few exceptions. The WatchPAT was the first wrist-based wearable device to gain approval for the diagnosis of sleep apnea ⁵², although the WatchPAT is not a commercially available device and must be prescribed by a physician. Similarly, the Sleep Image ring obtained approval for diagnosis of obstructive sleep apnea ⁵³. This device calculates apnea/hypopnea using a cardiopulmonary coupling technique; the presence of low-frequency coupling is characteristically linked to OSA ⁵³. Although this device has the form factor of a commercial device, it is also classed as a medical device.

A growing number of consumer wearables have obtained regulatory authorization for sleep apnea screening, This emerging capability extends their relevance to CMH assessment. The Apple Watch has received FDA clearance for its Sleep Apnea Notification Feature⁵⁴.Samsung also received FDA authorization for sleep apnea detection via its Galaxy Watch using the Samsung Health Monitor app ⁵⁵. Other consumer brands, such as Fitbit, Oura, WHOOP, Garmin among others have not yet obtained FDA clearance for use in sleep monitoring or screening at this time, though the rapid pace of device development may soon change this. The ability of devices to screen for sleep disorders reinforces utility in the practice of multi-dimensional cardiometabolic risk mitigation, especially considering the morbidity and prevalence of OSA.

Sleep Coaching and Sleep Health Promotion Programs

Many digital platforms integrate sleep reminders, feedback on sleep timing, and other strategies to modulate the bidirectional relationship between sleep and behavior ⁵⁶. While individual outcomes vary, evidence shows that digital-based interventions can improve multiple sleep metrics ⁵⁷. Application-based interventions for insomnia have demonstrated potential benefit but did not explicitly support the efficacy of minor behavioral interventions that are common among consumer application-paired wearable devices. Though minor application-based behavioral interventions have shown to be largely ineffective at improving sleep parameters, patient education on healthy sleep behaviors remain indicated to be inclusive to individuals with lower levels of health literacy.

Improving Insomnia Treatment

Wearables can facilitate insomnia management by capturing objective, longitudinal data on sleep patterns in the home environment as well as by assessing treatment adherence ⁵⁸. These devices offer a more convenient alternative than traditional sleep diaries for monitoring sleep patterns, although self-reported assessments such as sleep diaries remain the gold standard for insomnia assessment ¹⁰.

Of note, the integration of wearables with digital cognitive behavioral therapy for insomnia platforms is emerging. Pilot studies suggest that mobile applications synchronized with wearables are efficacious in improving both subjective and objective sleep measures ⁵⁹; however, large-scale validation is still warranted. While promising, the integration of wearables into clinical practice is hindered by their potential to heighten sleep-related anxiety and the need for further empirical evaluation. ¹⁰

Improving Sleep Apnea Treatment

Many devices measure blood oxygen saturation levels, with studies exploring the role of wrist-worn reflective PPG in estimating the apnea-hypopnea index ⁶⁰. While these devices cannot yet – and may never – replace standard PSG or home sleep apnea testing, they serve as useful data collection tools that capture night-to-night variability while additionally monitoring for arrhythmias that are more prevalent in this population ⁶¹. As not all sleep apnea diagnoses require CPAP, wearables offer an opportunity to track disease management in the absence of on-board measurement present in most CPAP machines ⁶². Moreover, rather than relying on single-night assessments, wearable devices that track oxygen saturation over time may provide useful information to guide sleep apnea treatment ⁶³.

Improving Behavioral Cardiometabolic Risk Factors/Integrating into Cardiovascular Medicine

It is necessary to consider the reasons for suboptimal sleep to understand sleep as a behavioral cardiometabolic risk factor ¹. If an overt sleep pathology is not identified, then sleep opportunity, need, and ability should be considered ⁶⁴. Lifestyle counseling is likely to be insufficient to overcome social determinants of health contributing to sleep disturbance^65,66,such as shift-workers whose natural circadian rhythm is dictated by their employment, light pollution, air pollution, neighborhood violence, among others. Patients often produce their own wearable records for a clinician. Due to device variability and the nuanced interpretation required, many providers are often reluctant to discuss wearable data ⁶⁷, especially in the context of a rushed clinic visit. A simple approach in this case would be to identify whether the patient has an actual sleep complaint or if they are simply concerned about device scores yet are otherwise well. If a sleep complaint is present, then provider should use appropriate practice guidelines to treat, refer, or educate the patient. If the patient is concerned solely about a device-generated score and is otherwise well, then it is recommended to communicate that consumer sleep technology is not diagnostic ⁶⁸. Accordingly, providers should reassure patients that their sense of well-being and lack of symptoms are a more reliable indicator of sleep health than the device’s reported metrics ⁶⁸. The behavior of seeking a wearable sleep tool may represent willingness to improve sleep health but having the wearable alone does not mean they are equipped with adequate tools ⁶⁹. Recognizing this distinction is essential for providers when crafting collaborative plans to modify risk factors for cardiovascular disease. Wearables offer both an initial intervention opportunity and the quantitative assessment of longitudinal adherence to lifestyle interventions. This increased ability to effectively apply secondary prevention aligns well with the goals of Life’s Essential 8.

Benefits of co-witnessed exercise monitoring

Wearable platforms have the additional benefit of often being integrated with exercise promotion tools ⁷⁰, including step count trackers and daily reminders to move. Accordingly, these features enable providers to have informed conversations with their patients about lifestyle management of disease ⁷⁰. Sleep and exercise have a bidirectional relationship, both playing dynamic roles in the development and progression of a variety of chronic conditions ⁷¹. A benefit of patients engaging with a wearable ecosystem is that they can interact with modifiable risk factors for health ⁷². Evidence suggests that the use of a physical activity tracker can modestly improve total physical activity ⁷³; however, long-term adherence typically requires ongoing engagement and follow-up. Prior use of a wearable alone does not appear to produce lasting effects after use has been discontinued, even in a setting where usage was specifically gamified ⁷⁴. The ability of multidimensional platforms that consider sleep, exercise, and other health metrics presents providers with an opportunity to set goals with patients and quantitatively track adherence over time.

Addressing Orthosomnia

Use of wearables may develop into orthosomnia, a problematic hyperfocus on one’s sleep parameters that may paradoxically decrease subjective sleep quality ⁷⁵. Orthosomnia as a condition is a reminder to proceed with caution, not an argument against patient use of wearables. However, overzealous attitudes held by providers and patients alike should be tempered with appropriate messaging ⁷⁶. One key aspect is that sleep measurement often records wake that cannot be recalled by the wearer ⁷⁷, see Figure 9. Patients should first use their own agency and interoception rather than build an approval-based relationship with a digital tool. An excess of information about one’s sleep does not proportional increase understanding or health ⁶⁸.

Figure 9.

Conceptualization of Measured Sleep vs Patient Recalled Sleep

Note. Conceptual model comparing subjective experience and device-recorded measures of sleep and wake. Subjective experience is divided into recalled sleep and recalled wake, while device records separate measured sleep and measured wake. The overlap highlights that measured wake includes both recalled wake and periods of wakefulness without conscious awareness. Figure by Michael Amadei, 2025.

CONCLUSIONS AND FUTURE DIRECTIONS

This review is intended to provide readers with information about wearables as measurement tools as opposed to the interventions as which they are often portrayed. Across sleep research, wearable devices are used to reliably characterize sleep parameters over longer periods of time than PSG. In clinical medicine, their implementation may be less linear due to data quality, data variability, interoperability, health data privacy, and lack of actionable guides based on wearable data.

Sleep technology enhances understanding of sleep and circadian science at both a population and individual level. Digital platforms can organize sleep records at scale, increase health engagement, and potentially serve as an objective anchor in clinician-patient conversations about navigating lifestyle factors. Sleep wearables have also shown promise in disease screening as well as an adjuvant to potentiate interventions, enabling users to catalog and reinforce healthy habits over time. Biometric data shows promise for use in disease prediction tools in the future ⁷⁸, while also prompting concerns about privacy and data governance.

Digital sleep monitoring devices are measurement tools. They are not sleep health interventions, as they are often advertised by many consumer brands. An important limitation of sleep wearables is that their outputs do not inherently fit into clinically actionable protocols. For example, if a patient comes to a provider worried about suboptimal levels of some sleep metric identified by the device, in the absence of a clinical sleep pathology, there is not a reliable strategy they can offer beyond coaching sleep hygiene ⁶⁸. This example is especially salient in actigraphy-based wearables that often underestimate sleep compared to individual recall of how long they slept ⁷⁹. Marginal benefit over standard of care is limited in cases where outputs from wearables guide practitioners to advice they would be giving patients anyway. A vital caveat being that CMH risk factors are not adequately being discussed in routine office visits⁸⁰

This narrative review is limited by the significant lag between the release of new devices and the published performance evaluations. Oftentimes, by the time the evaluation is published, the software and/or hardware versions on which the device was originally evaluated may be potentially unavailable, replaced by newer, untested versions⁸¹. Moreover, consumer wearable devices are often aggressively marketed in ways that can give consumers misguided or incorrect impressions about device capabilities⁸².

For the past several decades, wearable sleep tracking technology has been used to successfully characterize sleep-wake states across multiple days in free-living conditions. Since that time, advances in technology have diversified and expanded the features of these useful tools to include multiple peripheral signals beyond movement, as well as headband devices that ambulatorily capture EEG signals. Existing wearable devices are useful in the assessment of a wide range of sleep parameters in free-living conditions. Therefore, incorporating these devices into cardiovascular research and clinical practice may aid in the characterization and/or management of patients given the link between sleep assessment cardiometabolic health risk assessment.

Researchers are currently exploring how changes in tracked variables over long periods could be used to infer conditions about internal states of health ⁷⁸, and advancements in machine learning algorithms could enable relationships to be delineated that are not currently apparent ⁸³.

Skeptics argue that complex and expensive tools would offer only marginal benefit over more readily obtained metrics like age, family history, lifestyle choices, and social determinants of health ⁸⁴. In many cases, that information would generate the same lifestyle advice that is given to every patient. For example, the STOP-BANG screening tool for sleep apnea is simple and can be accomplished in any clinic visit and offers strong predictive utility, not requiring any technology ⁸⁵. Therefore, if a technological solution added time, cost, and complication to sleep disorder screening and still could not outperform a simple clinical screener, its utility would be questionable.

Increasing the amount of data does not inherently increase understanding of internal state ⁸⁶. Care must be taken in implementation to ensure that the use of these wearable devices offers clinical utility rather than unnecessarily adding a surplus of health information without commensurate improvements in understanding. Real-time, granular measurement may offer benefits in clinical management, but current modeling should not be mistaken for a complete picture of health ⁸⁷. Instead, the promise of digital tools is to monitor ongoing disease processes or direct resources to patients who have an increased likelihood of disease ⁸⁸, with the aim to positively impact patient outcomes.

As wearable devices become increasingly ubiquitous, patients will increasingly bring in their data to clinic visits. Likewise, the increasing demonstrability of these devices as data collection tools will increase their utilization in research. It is imperative that the incorporation of these devices into both clinical and research environments accounts for these devices’ strengths and limitations. In the practice of cardiovascular medicine, wearable sleep health technology can sufficiently characterize sleep habits in a manner that exceeds characterization from a clinical history, in many cases offering the provider more information about a patient’s sleep than they could otherwise obtain. This information alongside numerous other metrics gleaned from wearables positions these devices as compelling tools to add to a cardiologist’s armamentarium.

Highlights.

Wearable sleep health technology infers sleep from peripheral proxies including motor activity, photoplethysmography-derived cardiac signals, and skin temperature; since sleep is being indirectly measured, outputs from devices should be interpreted based on the physiological signals being measured.

Evaluated against polysomnography, wearables consistently show high sensitivity for detecting sleep but lower specificity for detecting wake; specificity, rather than overall agreement, is the metric that meaningfully differentiates devices.

Wearables are measurement tools, not interventions. Their outputs do not map directly onto actionable clinical protocols, and their utility depends on critical appraisal of device-specific performance evaluations.

Applied with appropriate caveats, wearable-derived sleep data offer scalable, longitudinal, objective characterization of sleep health that can support cardiometabolic risk assessment consistent with AHA’s Life’s Essential 8.

Abbreviations:

CMH

cardiometabolic health

CPAP

continuous positive airway pressure

EBE

epoch-by-epoch

EDA

electrodermal activity

EEG

electroencephalography

EMG

electromyography

EOG

electrooculography

HRV

heart rate variability

NREM

non-rapid eye movement

OSA

obstructive sleep apnea

PPG

photoplethysmography

PR

pulse rate

PRV

pulse rate variability

PSG

polysomnography

REM

rapid eye movement

RR

respiratory rate

SpO2

peripheral oxygen saturation