Performance of consumer wrist-worn sleep tracking devices compared to polysomnography: a meta-analysis

Abstract

Study Objectives:

The use of sleep tracking devices is increasing as people become more aware of the importance of sleep and interested in monitoring their patterns. With many devices on the market, we conducted a meta-analysis comparing sleep scoring data from consumer wrist-worn sleep tracking devices with polysomnography to validate the accuracy of these devices.

Methods:

We retrieved studies from the databases of SCOPUS, EMBASE, Cochrane Library, PubMed, Web of Science, and KoreaMed and OVID Medline up to March 2024. We compared personal data about participants and information on objective sleep parameters.

Results:

From 24 studies, data of 798 patient using Fitbit, Jawbone, myCadian watch, WHOOP strap, Garmin, Basis B1, Zulu Watch, Huami Arc, E4 wristband, Fatigue Science Readiband, Apple Watch, or Xiaomi Mi Band 5 were analyzed. There were significant differences in total sleep time (mean difference, −16.854; 95% confidence interval, [−26.332; −7.375]), sleep efficiency (mean difference, −4.691; 95% confidence interval, [−7.079; −2.302]), sleep latency (mean difference, 2.574; 95% confidence interval, [0.606; 4.542]), and wake after sleep onset (mean difference, 13.255; 95% confidence interval, [4.522; 21.988]) between all consumer sleep tracking devices and polysomnography. In subgroup analysis, there was no significant difference in wake after sleep onset between Fitbit and polysomnography. There was also no significant difference in sleep latency between other devices and polysomnography. Fitbit measured sleep latency longer than other devices, and other devices measured wake after sleep onset longer. Based on Begg and Egger’s test, there was no publication bias in total sleep time and sleep efficiency.

Conclusions:

Wrist-worn sleep tracking devices, although popular, are not as reliable as polysomnography in measuring key sleep parameters such as total sleep time, sleep efficiency, and sleep latency. Physicians and consumers should be aware of their limitations and interpret results carefully, though they can still be useful for tracking general sleep patterns. Further improvements and clinical studies are needed to enhance their accuracy.

Citation:

Lee YJ, Lee JY, Cho JH, Kang YJ, Choi JH. Performance of consumer wrist-worn sleep tracking devices compared to polysomnography: a meta-analysis. J Clin Sleep Med. 2025;21(3):573–582.

BRIEF SUMMARY

Current Knowledge/Study Rationale: Previous studies on consumer wrist-worn sleep tracking devices have shown inconsistent results, and only a few meta-analysis studies have compared their effectiveness to polysomnography. This study aimed to compare sleep-related indicators from these devices with polysomnographic results and a recent literature review, focusing on the validity of these devices and their effectiveness compared to polysomnography.

Study Impact: Wrist-worn sleep tracking devices, similar to polysomnography, have limitations and varying accuracy in assessing sleep parameters such as total sleep time, efficiency, latency, and wake after sleep onset. Further algorithm improvement and clinical studies are needed in the consumer wrist-worn sleep tracking device market.

INTRODUCTION

Sleep is a complex physiological process that reduces responsiveness due to perceptual separation from the environment and is vital for maintaining life.^1,2 The lack of sleep can negatively affect immune function, appetite control, and overall health, increasing the risk of disease.³ Polysomnography, a standard test for diagnosing sleep disorders such as obstructive sleep apnea, narcolepsy, and insomnia, records various biological signals such as electroencephalography (EEG), electrooculography, electromyography, electrocardiography, airflow, respiratory effort, oxygen saturation, body position, and snoring. It is widely used as a standard test for diagnosing sleep diseases.^2,4,5 Although highly accurate, polysomnography is expensive and inconvenient and requires a skilled expert.^6,7 Consumer sleep tracking devices are widely used for their convenience, but there is limited information on how their results differ from polysomnography.

Recently, as awareness of the importance of sleep increases, the market for consumer sleep tracking devices that can be used conveniently has been growing significantly.⁸ Consumer sleep tracking devices have the advantage of being able to record the user’s sleep and waking behaviors while the user is active in a familiar sleep environment or living daily life.⁹ There are many types of consumer sleep tracking devices, including wrist-worn types such as watches and bracelets; contact types such as rings, eye patches, headbands, headphones, or belts; and noncontact type such as smartphone applications or materials that can be placed around the bed and mattress.^9,10 Previously, various consumer wrist-worn-type sleep tracking devices such as Fitbit, Garmin, Jawbone, Fatigue Science Readiband, myCadian watch, WHOOP strap, Basis B1, E4 wristband, and Zulu watch have been developed and used and studies on their effectiveness have been conducted.^11–34 However, studies examining the validity of consumer wrist-worn-type sleep tracking devices have shown inconsistent results. Although a few previous meta-analyses have examined the effectiveness of consumer wrist-worn sleep tracking devices compared to polysomnography,³⁵ our study analyzed the latest studies comparing a broader range of consumer wrist-worn-type sleep tracking devices and sleep parameters. We compared various sleep indicators measured by these devices with polysomnographic results, incorporating the most recent literature review.

METHODS

Study design

This study was based on a comparative evaluation of sleep-related indicators (sleep-scoring data) including total sleep time, sleep efficiency, sleep latency, and wake after sleep onset in healthy adults or adult patients with various diseases such as insomnia and sleep-related breathing disorders. A meta-analysis was conducted along with a literature review on clinical studies that investigated the validity of consumer wrist-worn-type sleep tracking devices compared to polysomnography. The Preferred Reporting Items Guidelines for Systematic Review and Meta-Analysis were used to conduct this systematic review and meta-analysis.

Search for literature sources

The meta-analysis was conducted on clinical studies that evaluated validity of various consumer wrist-worn-type sleep tracking devices through comparison with polysomnography. A data search was conducted in SCOPUS, EMBASE, Cochrane Library, PubMed (Medline), Web of Science, KoreaMed, and OVID Medline to collect studies on various consumer sleep tracking devices. References of collected studies were also investigated to conduct a more comprehensive data search. The following terms were retrieved using a combination of MeSH (Medical Subject Headings) terms and keywords: (Polysomnogra* OR Sleep monitoring OR Somnograph* OR Sleep measur*) AND (Wearable* OR Electronic skin OR Smartphone OR Actiwatch OR ActiGraph* OR Wrist-Worn OR Wrist-Worn OR Wristband OR Garmin OR Fitbit OR Smartphone* OR Sleep Track* OR Sleep scoring OR Xiaomi smart band OR Xiaomi Mi Band OR Galaxy watch OR Apple watch OR Zulu watch OR Jawbone UP* OR Whoop Strap OR Fatigue science readiband). The exact search strategies used for each literature database are provided in Table S1 and Table S2 in the supplemental material. The search language was limited to English. We conducted the search strategies on October 21, 2022, and then searched for updated results on March 15, 2024. Related studies published during the designated period were collected.

Inclusion and exclusion criteria of literature sources

The inclusion criteria of relevant studies for meta-analysis were as follows: (1) studies targeting healthy adults or adult patients with various diseases; (2) studies evaluating the validity of consumer sleep tracking devices; (3) studies presented with objective sleep-related indicators (eg, total sleep time, sleep efficiency, sleep latency, wake after sleep onset, etc); and (4) studies compared with polysomnography (including sleep tests with EEG analysis). Exclusion criteria were as follows: (1) if the age of the target group in the study was unclear; (2) studies evaluating other types of consumer sleep tracking devices instead of a wrist-worn type; (3) studies that did not present objective indicators (eg, total sleep time, sleep efficiency, sleep latency, wake after sleep onset etc); and (4) studies published in languages other than English.

Literature search and data extraction were independently conducted by 2 researchers. After primary screening based on the title and abstract of the study, the original text of the selected study was reviewed and data necessary for meta-analysis were extracted from the final selected study. When extracted data were different, a third researcher reviewed them and an agreement among researchers was reached.

Data extraction

Data were extracted from the final selected studies based on a standardized format. Extracted data included information on the model of consumer wrist-worn-type sleep tracking devices, personal data about participants, total number of participants, age (years), sex (male:female), study design, information on the level of evidence, and information on objective sleep indicators (such as total sleep time, sleep efficiency, sleep latency, and wake after sleep onset).

Quality evaluation

We conducted a quality assessment of nonrandomized controlled studies using the Newcastle–Ottawa Scale and subsequently converted the assessments into the Agency for Healthcare Research and Quality standards, categorizing them as either good, fair, or poor. The Newcastle–Ottawa Scale assesses research quality using 3 categories: selection of study groups, comparability of groups, and outcome measurement. Criteria in each category receive star ratings, with a maximum of 9 stars indicating research quality.

Statistical analysis

To evaluate the validity of consumer wrist-worn-type sleep tracking devices, average and standard deviation values of total sleep time, sleep efficiency, sleep latency, wake after sleep onset were collected from targeted studies. The degree of heterogeneity of selected studies was judged by Cochrane’s Q test and the I² statistic. If the I² statistic was less than 25%, the heterogeneity was low. If the I² statistic was 50%, the heterogeneity was moderate. If the I² statistic was 75% or more, the heterogeneity was high.¹² After examining the heterogeneity of individual studies, a random effect model was applied when heterogeneity was high. All results of this study are presented with a 95% confidence interval (CI). All significance probabilities (P values) were determined with a 2-sided test. The statistical significance of publication bias was confirmed through 3 methods: funnel plot, Egger’s linear regression asymmetry test, and Begg’s test. In Egger’s test and Begg’s test, a 1-sided test was used. It was judged that there was asymmetry at P < .05. In addition, we used the Duval and Tweedie trim-and-fill methods to account for publication bias in the summed effect size calculation. All statistical analyses were performed using Comprehensive Meta-Analysis V2 software (Biostat, Englewood, New Jersey).

RESULTS

Search results and characteristics

A total of 3,727 related studies were retrieved from databases after removing duplicates. After primary evaluation of title and abstract, 3,640 studies were excluded. After examining original texts of selected studies and applying inclusion and exclusion criteria presented in the method, 63 studies were additionally excluded. Finally, 24 studies were selected for the meta-analysis (Table 1 and Figure S1 in the supplemental material). Among consumer wrist-worn-type sleep tracking devices used in studies, Fitbit (Classic, Flex, Versa, Versa 2, Charge 2, Charge 4, Alta HR, and Inspire HR) was the most common one, followed by Garmin (Vivosmart 3, Vivosmart 4, Fenix 5S), Jawbone (UP, UP3), WHOOP strap, E4 wristband, myCadian watch, Basis B1, Fatigue Science Readiband, Zulu Watch, Xiaomi Mi Band 5, Huami Arc, and Apple Watch (Series 8). Most of the study participants were healthy adults. Studies conducted on adults with or suspected of various diseases such as insomnia, sleep-related breathing disorders, periodic limb movement disorder, hypersomnia, and depression were included. We used a random effects model due to high statistical heterogeneity, and subgroup analyses between healthy and unhealthy groups, Fitbit and other devices, depending on published year, were performed to explain the volatility. Fitbit was analyzed in most of the included studies, and it was possible that recent studies published after 2021 included devices with improved algorithms. Therefore, we performed subgroup analysis to determine whether the heterogeneity was due to the algorithm or to Fitbit, which accounts for the majority.

Table 1.

Characteristics of included clinical research studies.

References	Year	Model	Participants	No. of Samples	Age (years)	Sex (M:F)	Study Design	Level of Evidence
Montgomery-Downs et al11	2012	Fitbit (Classic)	Healthy group	24	19 to 41	14:10	Prospective study	Level 2
de Zambotti et al12	2015	Jawbone (UP)	Midlife healthy females	28	50.1 ± 3.9	0:28	Prospective study	Level 2
Cook et al13	2017	Fitbit (Flex)	Mild to moderate major depressive disorder	21	26.5 ± 4.6	4:17	Prospective study	Level 2
Kang et al14	2017	Fitbit (Flex)	Insomnia disorder group & healthy group	33 17	38.4 ± 11.2 32.1 ± 7.4	14:19 6:11	Prospective study	Level 2
de Zambotti et al15	2018	Fitbit (Charge 2)	Healthy group & PLMS group	35 9	35 ± 12 42 ± 15	12:23 6:3	Prospective study	Level 2
Cook et al16	2018	Jawbone (UP3)	Suspected central disorders of hypersomnolence	43	33.3 ± 11.0	14:29	Prospective study	Level 2
Pigeon et al17	2018	myCadian watch	Healthy group	20	30.1 ± 13.1	14:6	Prospective study	Level 2
Svensson et al18	2019	Fitbit (Versa)	Healthy Japanese group	20	25 to 67	10:10	Prospective study	Level 2
Moreno-Pino et al19	2019	Fitbit (Charge 2 + Alta HR)	Group presenting with symptoms of OSA	65	58.8 ± 13.8	42:23	Prospective study	Level 2
Miller et al20	2020	WHOOP strap	Healthy group	12	22.9 ± 3.4	6:6	Prospective study	Level 2
Mouritzen et al21	2020	Garmin (Vivosmart 4)	Healthy group	18	56.1 ± 12.0	5:13	Prospective study	Level 2
Kanady et al22	2020	Basis B1	Healthy group	18	26.8 ± 3.4	5:13	Prospective study	Level 2
Berryhill et al24	2020	WHOOP strap	Healthy group	32	23.8 ± 5.0	11:21	Prospective study	Level 2
Devine et al27	2020	Zulu Watch	Healthy group	8	30.4 ± 3.2	4:4	Prospective study	Level 2
Cheung et al31	2020	Huami Arc	Unhealthy group (training group:test group)	41 (20:21)	42.2 ± 14.7	17:24	Prospective study	Level 2
Regalia et al23	2021	E4 wristband (ACTS1) E4 wirstband (Sedah)	Clinically diverse population of older adults	46	66.3 ± 10.3	25:21	Prospective study	Level 2
Ellender et al25	2021	Jawbone (UP3)	Suspected sleep disorder	54	48.1 ± 18.1	23:31	Prospective study	Level 2
Chinoy et al26	2021	Fatigue Science Readiband Fitbit (Alta HR) Garmin (Fenix 5S) Garmin (Vivosmart 3)	Healthy group	34	28.1 ± 3.9	12:22	Prospective study	Level 2
Stucky et al28	2021	Fitbit (Charge 2)	Police officers and paramedics undergoing shift work	59	33.5 ± 8.1	26:33	Prospective study	Level 2
Scott et al29	2021	Fitbit (Flex) Fitbit (Alta)	Healthy group	25	25.4 ± 6.4	10:15	Prospective study	Level 2
Dong et al30	2022	Fitbit (Charge 4)	Diagnosed with chronic insomnia (DSM-5 criteria)	37	48.8 ± 2.1	17:20	Prospective study	Level 2
Jaworski et al33	2023	Apple Watch (Series 8)	Healthy group	1	31	1:0	Prospective study	Level 2
Concheiro-Moscoso et al32	2023	Xiaomi Mi Band 5	Mixed (sleep disorder group:healthy group)	45 (25:20)	53.2 ± 15.4	23:22	Prospective study	Level 2
Kainec et al34	2024	Fitbit (Inspire HR) Fitbit (Versa 2) Garmin (Vivosmart 4)	Healthy group	53	22.5 ± 3.5	31:22	Prospective study	Level 2

DSM-5 = Diagnostic and Statistical Manual of Mental Disorders, fifth edition, F = female, M = male, OSA = obstructive sleep apnea, PLMS = periodic limb movement of sleep.

We assessed the quality of these studies using the Newcastle–Ottawa Scale and then aligned them with Agency for Healthcare Research and Quality standards. All studies were evaluated to be of good quality concerning the risk of bias.

Comparison of total sleep time

Statistical heterogeneity was found in 22 studies comparing total sleep time between consumer wrist-worn-type sleep tracking devices and polysomnography (I² = 67.5%, P < .001 by Q test). As a result of meta-analysis, which applied a random effects model in consideration of the high heterogeneity among studies included in the analysis, it was found that there was a significant difference in total sleep time between consumer wrist-worn-type sleep tracking devices and polysomnography (mean difference [MD], −16.854; 95% CI, [−26.332; −7.375]; P < .001) (Figure 1). To evaluate whether there was a publication bias in studies, a funnel plot was created for each estimated effect size. Results of Begg’s test (P = .214) and Egger’s test (P = .280) confirmed that there was no publication bias in total sleep time (Figure S2 in the supplemental material).

Figure 1. Forest plots comparing sleep parameters between consumer wrist-worn-type sleep tracking devices and polysomnography.

(A) Total sleep time, (B) sleep efficiency, (C) sleep latency, and (D) wake after sleep onset. CI = confidence interval.

Comparison of sleep efficiency

Statistical heterogeneity was found in 18 studies comparing sleep efficiency between consumer wrist-worn-type sleep tracking devices and polysomnography (I² = 84.1%, P < .001 by Q test). As a result of the meta-analysis, which applied the random effects model in consideration of the high heterogeneity among studies included in the analysis, it was found that there was a significant difference in sleep efficiency between consumer wrist-worn-type sleep tracking devices and polysomnography (MD, −4.691; 95% CI, [−7.079; −2.302]; P < .001) (Figure 1). To evaluate whether there was a publication bias in studies, a funnel plot was created for each estimated effect size. Results of Begg’s test (P = .1312) and Egger’s test (P = .169) confirmed that there was no publication bias in sleep efficiency (Figure S2).

Comparison of sleep latency

Statistical heterogeneity was found in 18 studies comparing sleep latency between consumer wrist-worn-type sleep tracking devices and polysomnography (I² = 57.9%, P < .001 by Q test). As a result of the meta-analysis, which applied the random effects model in consideration of the high heterogeneity among studies included in the analysis, it was found that there was a significant difference in sleep latency between consumer wrist-worn-type sleep tracking devices and polysomnography (MD, 2.574; 95% CI, [0.606; 4.542]; P = .010) (Figure 1). To evaluate whether there was a publication bias in the studies, a funnel plot was created for each estimated effect size. Results of Begg’s test (P = .337) and Egger’s test (P = .017) confirmed that there was no publication bias in sleep latency (Figure S2).

Because publication bias was suspected as a result of Egger test and Begg funnel plot analysis, Duval and Tweedie’s trim-and-fill test was performed. The results showed that sleep latency was significantly different between polysomnography and devices (MD, 0.611; 95% CI, [−1.577; 2.788]). Therefore, included studies reporting sleep latency could be influenced by bias and may not demonstrate significant differences between polysomnography and consumer wrist-worn-type sleep tracking devices.

Comparison of wake after sleep onset

Statistical heterogeneity was found in 20 studies comparing wake after sleep onset between consumer wrist-worn-type sleep tracking devices and polysomnography (I² = 67.4%, P < .001 by Q test). As a result of the meta-analysis, which applied the random effects model in consideration of the high heterogeneity among studies included in the analysis, it was found that there was a significant difference in wake after sleep onset between consumer wrist-worn-type sleep tracking devices and polysomnography (MD, 13.255; 95% CI, [4.522; 21.988]; P = .003) (Figure 1). To evaluate whether there was a publication bias in studies, a funnel plot was created for each estimated effect size. Results of Begg’s test (P = .082) and Egger’s test (P = .061) confirmed that there was no publication bias in wake after sleep onset (Figure S2).

Because publication bias was suspected as a result of Egger’s test and Begg funnel plot analysis, Duval and Tweedie’s trim-and-fill test was performed. The results showed that wake after sleep onset was significantly different between polysomnography and devices (MD, 0.171; 95% CI, [−0.122; 0.464]). Therefore, included studies reporting wake after sleep onset could be influenced by bias and may not demonstrate significant differences between polysomnography and consumer wrist-worn-type sleep tracking devices.

Comparison of healthy group with unhealthy group

As a result of the subgroup analysis between healthy and unhealthy groups, there was no significant difference in total sleep time between healthy (MD, −12.504; 95% CI, [−21.420; −3.587]; P = .006) and unhealthy groups (MD, −26.188; 95% CI, [−50.158; −2.217]; P = .032). There was also no significant difference in sleep efficiency between healthy (MD, −6.493; 95% CI, [−8.989; −3.996]; P < .001) and unhealthy groups (MD, −2.159; 95% CI, [−6.731; 2.413]; P = .355). For sleep latency and wake after sleep onset, there were no significant differences between healthy (sleep latency MD, 2.005; 95% CI, [−9.452; 13.452]; P = .054) and wake after sleep onset (MD, 13.910; 95% CI, [3.036; 24.782]; P = .012) and unhealthy groups (sleep latency MD, 3.120; 95% CI, [−2.231; 8.470]; P = .253) and wake after sleep onset (MD, 11.955; 95% CI, [−4.192; 26.102]; P = .147).

In subgroup analysis, there was no significant difference between the polysomnography and consumer wrist-worn-type sleep tracking devices in terms of sleep latency and efficiency for the unhealthy group (Figure 2).

Figure 2. Forest plots comparing sleep parameters between consumer wrist-worn-type sleep tracking devices and polysomnography in healthy and unhealthy group.

(A) Total sleep time, (B) sleep efficiency, (C) sleep latency, and (D) wake after sleep onset. CI = confidence interval.

Comparison of Fitbit with other devices

As a result of the subgroup analysis between Fitbit and other devices, there was no significant difference in total sleep time between Fitbit (MD, −6.564; 95% CI, [−18.890; 5.763]; P = .297) and other devices (MD, −26.674; 95% CI, [−39.051; −14.297]; P < .001). There was also no significant difference in sleep efficiency between Fitbit (MD, −3.626; 95% CI, [−7.639; 0.387]; P = .077) and other devices (MD, −5.498; 95% CI, [−8.050; −2.946]; P < .001). For sleep latency and wake after sleep onset, there were significant differences between Fitbit (sleep latency MD, 5.602; 95% CI, [2.105; 9.098]; P = .002) and wake after sleep onset (MD, 0.533; 95% CI, [−12.517; 13.583]; P = .936) and other devices group (sleep latency MD, −0.122; 95% CI, [−1.493; 1.249]; P = .861) and wake after sleep onset (MD, 24.140; 95% CI, [15.735; 32.545]; P < .001).

In subgroup analysis, other devices reported sleep latency more accurately than Fitbit, and Fitbit reported longer sleep latency than polysomnography. Fitbit reported wake after sleep onset with high accuracy, and other devices reported longer wake after sleep onset than polysomnography (Figure 3).

Figure 3. Forest plots comparing sleep parameters between consumer wrist-worn-type sleep tracking devices and polysomnography in Fitbit and other devices.

(A) Total sleep time, (B) sleep efficiency, (C) sleep latency, and (D) wake after sleep onset. CI = confidence interval.

Comparison of studies published after 2021 with studies published before 2020

As a result of the subgroup analysis depending on published year, there was no significant difference in total sleep time between recent (MD, −20.544; 95% CI, [−34.832; 6.255]; P = .005) and previous studies (MD, −13.174; 95% CI, [−25.443; −0.905]; P = .035). There was also no significant difference in sleep efficiency between recent (MD, −5.830; 95% CI, [−9.502; −2.158]; P = .002) and previous studies (MD, −3.744; 95% CI, [−7.057; −0.431]; P = .027). For sleep latency and wake after sleep onset, there were no significant differences between recent studies (sleep latency MD, 1.782; 95% CI, [−0.452; 4.016]; P = .118) and wake after sleep onset (MD, 14.929; 95% CI, [0.404; 29.453]; P = .044) and previous studies (sleep latency MD, 3.406; 95% CI, [0.102; 6.710]; P = .043) and wake after sleep onset (MD, 12.634; 95% CI, [2.798; 22.469]; P = .012).

In subgroup analysis, studies that were recently published after 2021 showed no significant difference in total sleep time or sleep latency between consumer wrist-worn sleep tracking devices and polysomnography (Figure 4).

Figure 4. Forest plots comparing sleep parameters between consumer wrist-worn-type sleep tracking devices and polysomnography in recent and previous studies.

(A) Total sleep time, (B) sleep efficiency, (C) sleep latency, and (D) wake after sleep onset. CI = confidence interval.

DISCUSSION

Polysomnography is a valuable diagnostic tool for sleep disorders and treatment monitoring, particularly in conditions such as sleep apnea, where precise measurement of respiratory and sleep parameters is crucial. However, its high cost and inconvenience make it less practical for routine use, especially in cases such as insomnia where self-reported sleep perception plays a significant role, and the benefits of polysomnography may be limited to ruling out comorbid disorders.

In this study, we compared various consumer wrist-worn devices with polysomnography to demonstrate their effectiveness by analyzing objective sleep parameters such as total sleep time, sleep efficiency, sleep latency, and wake after sleep onset, using data from the latest studies compared to previous meta-analyses. There were significant differences in total sleep time, sleep efficiency, sleep latency, and wake after sleep onset between consumer wrist-worn-type sleep tracking devices and polysomnography. After the subgroup analysis, there was no significant difference in wake after sleep onset between Fitbit and polysomnography or in sleep latency between other devices and polysomnography. Fitbit reported sleep latency longer than other devices, and other devices reported wake after sleep onset longer. According to our results, we determined whether the consumer wrist-worn-type sleep tracking devices we use are reliable for diagnosing sleep disorders and interpret the results carefully.

Sleep-related conditions are assessed using indicators such as total sleep time, sleep efficiency, sleep latency, and wake after sleep onset. Total sleep time represents actual sleep time, sleep efficiency is the ratio of total sleep time to bedtime, and wake after sleep onset is the time awakened in the middle of sleep.² Sleep quality is generally considered good in all age groups with 85% efficiency, 15-minute latency, and 20-minute wake after sleep onset, whereas poor sleep quality is indicated by 74% efficiency.³⁶

Studies comparing consumer wrist-worn sleep tracking devices to polysomnography revealed inconsistent results and varied validity across various studies. Pigeon et al¹⁷ compared sleep-related indicators between myCadian watch, a wrist-worn sleep tracking device, and polysomnography in 20 healthy adults and found no significant differences in total sleep time, efficiency, or latency. Dong et al³⁰ conducted a study comparing Fitbit (Charge 4) and polysomnography in 37 patients with chronic insomnia and found no significant difference in total sleep time, sleep latency, or wake after sleep onset. Two studies suggest that wrist-worn sleep tracking devices can be cost-effective for measuring sleep-related indicators.^17,30 However, a subgroup analysis comparing Fitbit and other devices revealed a significant difference in sleep latency and wake after sleep onset.

Other previous studies also showed conflicting results. Cook et al¹³ have compared Fitbit (Flex), a wrist-worn sleep tracking device, with polysomnography in 21 adult patients with mild to moderate major depressive disorders. The results showed significant differences in total sleep time, sleep efficiency, and wake after sleep onset. Svensson et al¹⁸ conducted a study comparing Fitbit, a wrist-worn sleep tracking device, with EEG analysis in 20 healthy adults. They found significant differences in sleep efficiency, latency, and wake after sleep onset, suggesting further research is needed to determine Fitbit’s cost-effectiveness.

The following reasons can be used to explain inconsistent results and high heterogeneity of included studies. First, in polysomnography, a standard test, the technician/physician manually scores sleep-related indicators such as total sleep time, sleep efficiency, sleep latency, and wake after sleep onset based on EEG, electrooculography, and chin electromyography findings. Second, in consumer wrist-worn-type sleep tracking devices, sleep-related indicators are generally analyzed by an automatic algorithm of the sleep tracking device based on findings such as movement, heart rate, and heart rate variability based on accelerometer and photoplethysmography. However, because sleep latency and wake after sleep onset showed significant differences between Fitbit and other devices, careful interpretation of sleep latency and wake after sleep onset is needed. Third, measurement indicators or automatic reading algorithms of consumer wrist-worn-type sleep tracking devices might be different for each manufacturer and each model. Fourth, all studies used for this meta-analysis targeted adults as study participants. However, some studies targeted healthy adults, whereas others targeted adults with various diseases or suspected diseases such as insomnia, sleep-related breathing disorders, periodic limb movement disorders, hypersomnolence, and depression. Thus, polysomnography measures sleep-related indicators directly, whereas consumer wrist-worn sleep tracking devices measure and analyze them indirectly. Inconsistent research results may occur due to different measurement parameters or automatic analyzing algorithms for each manufacturer and study participant. However, there was no significant difference in subgroup analysis comparing healthy and unhealthy groups.

Wearable sleep tracking devices offer several advantages over traditional in-laboratory polysomnography. They are portable, cost-effective, and user-friendly, making them more accessible to a larger population. They are designed for home use, providing greater convenience and less expense compared to in-laboratory polysomnography. However, their accuracy and clinical efficacy have not yet been verified, which is a significant limitation. Despite this, wearable sleep tracking devices offer a more comfortable and cost-effective alternative to traditional polysomnography. The recent recommendation for using Wearable technology in sleep research also pointed out this limitation.³⁷ Wearable sleep tracking technology outcomes, especially from consumer-grade devices, require careful interpretation due to limited access to raw data and opaque algorithms, potentially affecting accuracy and reproducibility.³⁷ Evaluation studies often lack diversity and consider real-world challenges, requiring cautious interpretation and integration of consumer-grade and research/clinical-grade devices. Consumer-grade devices pose complexities such as data access and privacy concerns, emphasizing the need for proper data preprocessing for meaningful results interpretation.

Wearable sleep tracking devices could potentially replace standard polysomnography in regions where it is not easily accessible. Future research should focus on improving the accuracy of the automatic algorithm of these devices. Accurate measurement and analysis of sleep-related indicators such as total sleep time, sleep efficiency, sleep latency, and wake after sleep onset are crucial for understanding sleep-related conditions or quality.

This study also had several limitations. First, a random effects model was used due to statistical heterogeneity in the meta-analysis of sleep-related indicators such as total sleep time, sleep efficiency, sleep latency, and wake after sleep onset. To explain the volatility, we performed a subgroup analysis such as healthy vs unhealthy group, Fitbit vs other devices, and studies published after 2021 vs before 2020. However, further review including more prospective studies in a large cohort are needed to reduce heterogeneity. Second, although we confirmed that no publication bias existed, this could be the most significant potential confounder. We attempted to assess the extent to which transparent reporting of same funding could influence trial design, implementation, and results. Finally, sleep staging via consumer wrist-worn-type sleep tracking device performance was not considered. Because the ability to identify sleep staging is a significant advantage of both polysomnography and consumer wrist-worn-type sleep tracking device over standard actigraphy, further study related to sleep staging performance is needed.

Consumer wrist-worn sleep tracking devices did not show a significant advantage over gold-standard polysomnography statistically. However, it is unclear whether these devices are necessarily inaccurate. Subgroup analysis indicated they could potentially replace polysomnography. Although there were small differences in parameters, it is crucial to consider the clinical significance and cost-effectiveness of the device before determining its clinical effectiveness.

In conclusion, total sleep time, sleep efficiency, sleep latency, and wake after sleep onset were significantly different between consumer wrist-worn sleep tracking devices and polysomnography. Other devices measured sleep latency more accurately than Fitbit, which tended to overestimate it, whereas Fitbit reported wake after sleep onset more accurately than others. Overall, the metrics of these devices are not yet considered valid compared to polysomnography, which analyzes EEG, electrooculography, and electromyography. Physicians and consumers should use wrist-worn sleep trackers cautiously, acknowledging their accuracy limitations and interpreting results carefully. However, given the increasing popularity of these devices, their role in tracking general sleep trends is still valuable, though further research and algorithm improvements are needed to approach the accuracy of polysomnography.

DISCLOSURE STATEMENT

All authors have seen and approved the manuscript. Work for this study was performed at the Soonchunhyang University College of Medicine. This study was supported by the Soonchunhyang University Research Fund. The authors report no conflicts of interest.

ABBREVIATIONS

CI

confidence interval

EEG

electroencephalography

MD

mean difference