Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2025 Feb 28;34(6):e14480. doi: 10.1111/jsr.14480

Performance evaluation of an under‐mattress sleep sensor versus polysomnography in > 400 nights with healthy and unhealthy sleep

Jack Manners 1,✉, Eva Kemps 2, Bastien Lechat 1, Peter Catcheside 1, Danny J Eckert 1, Hannah Scott 1
PMCID: PMC12592840  PMID: 40017337

Summary

Consumer sleep trackers provide useful insight into sleep. However, large‐scale performance evaluation studies are needed to properly understand sleep tracker accuracy. This study evaluated performance of an under‐mattress sensor to estimate sleep and wake versus polysomnography in a large sample, including individuals with and without sleep disorders and during day versus night sleep opportunities, across multiple in‐laboratory studies. One‐hundred and eighty‐three participants (51%/49% male/female, mean [SD] age = 45 [18] years) attended the sleep laboratory for a research study including simultaneous polysomnography and under‐mattress sensor (Withings Sleep Analyser) recordings. Epoch‐by‐epoch analyses determined accuracy, sensitivity and specificity of the Withings Sleep Analyser versus polysomnography. Bland–Altman plots examined bias in sleep duration, efficiency, onset‐latency, and wake after sleep onset. Overall Withings Sleep Analyser sleep–wake classification accuracy was 83%, sensitivity 95% and specificity 37%. The Withings Sleep Analyser significantly overestimated total sleep time (48 [81] min), sleep efficiency (9 [15]%) and sleep‐onset latency (6 [26] min), and underestimated wake after sleep onset (54 [78] min). Accuracy and specificity were higher for night versus daytime sleep opportunities in healthy individuals (89% and 47% versus 82% and 26%, respectively, p < 0.05). Accuracy and sensitivity were also higher for healthy individuals (89% and 97%) versus those with sleep disorders (81% and 91%, p < 0.05). Withings Sleep Analyser performance is comparable to other consumer sleep trackers, with high sensitivity but poor specificity compared with polysomnography. Withings Sleep Analyser performance was reasonably stable, but more variable in daytime sleep opportunities and in people with a sleep disorder. Contactless, under‐mattress sleep sensors show promise for accurate sleep monitoring, noting the tendency to over‐estimate sleep particularly where wake time is high.

Keywords: performance evaluation, polysomnography, sleep, sleep measures, sleep trackers, validation study, wearables

1. INTRODUCTION

Sleep is critical for optimal daytime function, performance, safety and health. Thus, accurate and reliable estimates of sleep are important to help manage these key outcomes. However, sleep is difficult to objectively evaluate, particularly over extended multi‐day periods. This is partly because gold‐standard quantification of sleep, polysomnography, is expensive and complex to administer over multiple nights. Wearable and “nearable” (devices that are not in direct contact with an individual) sleep tracking devices are used to simplify sleep estimation in the home, such as wrist‐based actigraphy, bedside radar, or mattress sensor devices (Tobin et al., 2021; Yoon & Choi, 2023). Deciding which devices are best suited for sleep estimation requires critical appraisal of device performance and practicality. Accordingly, the present study evaluated the performance of an under‐mattress device, the Withings Sleep Analyser (WSA), to evaluate sleep in a diverse cohort and at various sleep opportunities (day and night).

To help circumvent cost and measurement complexity for multi‐day assessments, wearable sleep trackers, such as actigraphy‐based devices, are often used to infer wake and sleep from body movements. This approach is sensitive to detect sleep, but cannot reliably discriminate wake from sleep when people lie still awake (Paquet et al., 2007). Additional signals such as heart rate, heart rate variability and breathing can improve device performance, but poor specificity remains a key limitation of this technology. Another problem is that wearable devices require the user to ensure the device is charged and properly worn. Consequently, data loss is common (Fleur et al., 2021), variable (Huang et al., 2022), and performance can drastically drop when devices are not worn correctly (Sun et al., 2017). Nearable sleep trackers that infer sleep from detected motion, including respiratory and cardiac motion, avoid the need to charge or wear any device. These devices are typically designed to be placed on or under the mattress, or at the bedside, with little subsequent intervention required once properly set up (Schutte et al., 2021; Yoon & Choi, 2023). However, some evidence suggests that these devices may be less accurate than wearable counterparts (Lee et al., 2023). Thus, rigorous performance evaluation is essential to determine device reliability, practical benefits and sleep tracking performance.

The WSA has been previously validated in smaller trials (N = 18–118 [49–118 nights]) against polysomnography to estimate sleep and identify breathing disturbances (Edouard et al., 2021; Lee et al., 2023; Ravindran et al., 2021; Scott et al., 2023). The device uses a pneumatic sensor placed under the mattress to detect air‐pressure changes, from which movement, sleep, respiration and heart rate are inferred. The device is attractive for both clinical and research use, as once set up and connected to Wi‐Fi and mains power, minimal ongoing user input is required. Indeed, recent large‐scale, long‐term monitoring studies that have used this technology have been able to address key sleep health questions that were not previously possible with polysomnography or existing wearables (Lechat et al., 2022; Lechat et al., 2023; Lechat et al., 2024; Scott et al., 2023; Scott et al., 2024). Furthermore, this technology could markedly reduce data loss compared with wearable sleep trackers that require regular device charging and correct daily wear. The unobtrusive nature of this approach is also potentially amenable to patient/participant monitoring in environments less conducive to wearable devices, such as hospitals, nursing homes, and in high‐risk on‐call workplace settings. Given the poor specificity of sleep tracking devices (Chinoy et al., 2020; De Zambotti et al., 2024), sleep classification may be less accurate in these environments as sleep is more likely to be impaired, compared with in the home or laboratory. Despite this, few devices have been rigorously evaluated across multi‐night sleeps, and fewer devices have been assessed to any degree with non‐standard sleep schedules. Device accuracy under these conditions is particularly important if they are to be used for readiness evaluation in shift workers, or to examine clinical outcomes in people with sleep disorders. Therefore, comprehensive performance evaluation of the WSA during sleep opportunities of varied timing and in people with and without sleep disorders remains important to establish the potential utility of nearable sleep trackers for extended multi‐day monitoring in more challenging sleep assessment settings.

This study used comprehensive objective measurements of sleep in several laboratory sleep research studies to evaluate the classification accuracy of the WSA compared with polysomnography. Secondary aims explored how classification accuracy: (1) was affected by the timing of the sleep opportunity; (2) differed in people with versus without sleep disorders; (3) was explained by polysomnography sleep efficiency (SE); and (4) compared with a validated consumer wearable, the Fitbit Charge 4 (FC4) device.

2. METHODS

Data were utilized from 13 studies conducted at Flinders Health and Medical Research Institute: Sleep Health, from 2021 to 2023, where data from the WSA were collected during the sleep study. These comprised of 10 studies with between 1 and 3 nights in the sleep laboratory for participants with suspected or diagnosed obstructive sleep apnea (OSA), insomnia, co‐morbid insomnia and OSA (COMISA), cardiovascular disease, or general sleep complaint; two studies with single‐night laboratory visits with healthy volunteers; and one study with two 8‐day laboratory visits comprised of one nighttime sleep followed by five daytime sleep opportunities. Across all research studies, 416 sleep recordings (224 nighttime and 192 daytime recordings) with simultaneous polysomnography and WSA recordings were available for analysis. Data were time‐matched based on the clock‐times in the polysomnography and WSA recordings. Subsets of data were used to address secondary aims. Specifically, only data from healthy participants were available for the time of recording comparisons (given that only one study collected data from daytime sleep opportunities), only nighttime recordings were available for the healthy versus sleep disorder comparisons, and only one study of healthy participants had data available from both the WSA and the FC4 to enable device comparisons. Detailed study information can be found in Table S1.

2.1. Equipment

2.1.1. Polysomnography

Polysomnography was collected during all sleep opportunities. Polysomnography set‐ups were conducted in accordance with the standard 10–20 electroencephalography electrode placement system, using Compumedics Grael 4K PSG:EEG devices (Compumedics, Victoria, Australia). Sleep studies were independently scored using Profusion Compumedics software (v4.0) according to standardized American Academy of Sleep Medicine polysomnography scoring criteria (Iber, 2007). Polysomnography sleep stages were extracted as wake, rapid eye movement (REM) sleep, stage 1 sleep (N1), stage 2 sleep (N2) and Stage 3 sleep (N3).

2.1.2. Withings Sleep Analyser

The WSA is an under‐mattress device that uses a pneumatic sensor to detect changes in pressure in an air‐bladder relative to atmospheric pressure (i.e. ballistography). The device uses inferred movement based on this information and sound from an in‐built microphone to estimate respiration, heart rate and sleep stages. These estimations are conducted using proprietary algorithms that are not publicly available. The WSA was placed under the mattress, level with the chest of the sleeping individual. WSA‐derived total sleep time (TST), SE and wake after sleep onset (WASO) have been validated, compared with polysomnography (Edouard et al., 2021; Kainec et al., 2024; Lee et al., 2023; Scott et al., 2023), but extensive performance evaluation of sleep characteristics and sleep staging accuracy is lacking.

2.1.3. Fitbit Charge 4

The FC4 is a wrist‐worn device that contains a tri‐axial accelerometer to track movement (i.e. actigraphy) and a photoplethysmography sensor to estimate heart rate. This information is used to infer sleep stages and wake. FC4 devices were placed on each participant's non‐dominant wrist, with appropriate band‐sizes chosen to ensure proper fit. The FC4 model has been validated against polysomnography (Dong et al., 2022), and earlier Fitbit models similarly show reasonable accuracy to detect sleep compared with polysomnography. Individual validation studies show such devices typically overestimate TST by 30–60 min and SE by 5%–10%, and underestimate WASO by 20–60 min (Moreno‐Pino et al., 2019; de Zambotti et al., 2017; Stucky et al., 2021). Meta‐analyses suggest that newer Fitbit devices may not significantly differ from polysomnography in TST, SE and WASO (Haghayegh et al., 2019), yet the latest validated devices still show poor specificity (e.g. 62% of wake correctly identified; Dong et al., 2022). Accordingly, Fitbit devices are generally at least on‐par with highly validated research‐grade actiwatch devices (Conley et al., 2019; Haghayegh et al., 2019).

2.2. Statistical analysis

The WSA performance was evaluated based on recommended guidelines (de Zambotti et al., 2019; Menghini et al., 2021). Device data (WSA and FC4) were extracted via custom application programming interface software developed in Python (v3.11). Polysomnography data were extracted as European Data Format files using Python, and time‐matched to WSA and FC4 data to provide concurrent epoch‐by‐epoch data. As the WSA device only provides sleep stage classification data at 60‐s intervals, FC4 and polysomnography data were converted from 30‐s to 60‐s epochs. Where combined epochs differed, wake was scored if present in either, otherwise the first value was used (Sadeh et al., 1994). Lights on and off were derived from polysomnography sleep reports to determine sleep opportunities. Where device data started or ended within these limits, preceding and trailing epochs were designated as wake.

The TST was calculated as the sum of sleep epochs within the sleep opportunity. SE was calculated as the sum of sleep epochs divided by the total number of epochs within the sleep opportunity. Sleep‐onset latency (SOL) was calculated as the sum of wake epochs before the first sleep epoch, within the sleep opportunity. WASO was calculated as the sum of wake epochs between the first and last sleep epoch within the sleep opportunity. Wake and sleep classification performance was determined for each sleep recording as accuracy (proportion of correctly scored epochs), sensitivity (proportion of polysomnography‐derived sleep epochs that the WSA correctly scored as sleep) and specificity (proportion of polysomnography‐derived wake epochs that the WSA correctly scored as wake), compared with polysomnography. Additionally, four‐stage (“wake”, “light”, “deep”, “REM”) sleep classification from the WSA was compared with polysomnography sleep stages, where polysomnography N2 and N3 sleep epochs were combined as “deep” sleep.

Linear mixed model (LMM) analyses examined: (1) performance (WSA versus polysomnography) by time of recording (daytime versus nighttime) in healthy participants; (2) performance (WSA versus polysomnography) by sleep disorder status (healthy sleep versus sleep disorder); and (3) performance by consumer sleep tracker ([WSA versus polysomnography] versus [FC4 versus polysomnography]). LMM analyses were performed using the lme4 package (Bates et al., 2015; R v4.2.2) to examine how sleep estimates (TST, SOL, WASO, SE) and WSA performance (accuracy, sensitivity and specificity) were affected by these three fixed effects. Participant ID was entered as random effects in all models. Marginal R 2 was calculated to estimate variance explained by fixed effects. Secondary analyses were conducted using mixed models adjusted for age, body mass index (BMI) and polysomnography‐derived SE to examine potential confounders in the relationship between sleep opportunity timing, sleep disorder status and WSA performance.

Bland–Altman plots were used to examine bias and limits of agreement, calculated as ±1.96 the standard deviation of mean differences and their 95% confidence limits, between polysomnography and device‐derived estimates of TST, SE, WASO and SOL for each recording. Proportional bias was also calculated and tested, where significant bias indicated that the mean difference between device and polysomnography increased or decreased as a function of the size of measurement (Euser et al., 2008).

For participants with more than one recording, the coefficients of variation (CV; standard deviation divided by mean) for accuracy, sensitivity and specificity were calculated for each individual. To examine differences in performance variability across sleep disorder statuses and sleep opportunity timings, group differences in mean CV and the distribution of CVs were compared using linear regression and Levene tests, respectively.

Finally, in the subset where both FC4 and WSA devices were used, data‐loss was quantified and compared using a paired‐samples t‐test. Post‐hoc comparisons with Bonferroni corrections were conducted where main effects were significant. All data are reported as mean (SD) unless otherwise specified. A value of p < 0.05 was considered statistically significant.

3. RESULTS

Final data included 416 recordings from 183 participants, collected across 13 sleep research studies at Flinders Health and Medical Research Institute: Sleep Health. Participant demographics can be found in Table 1.

TABLE 1.

Participant demographics

Healthy sleep Sleep disorder Total
Sample size 82 101 183
Nighttime recordings 96 128 224
Daytime recordings 192 0 192
Age (years), mean (SD) 30.9 (11.9) 56.5 (14.2) 45 (18.4)
Sex, n (%) male 35 (43) 58 (57) 94 (51)
n (%) female 47 (57) 43 (43) 89 (49)
BMI (kg m−2), mean (SD) 24 (4) 31.7 (10.4) 28.3 (9)
AHI (events per hr), mean (SD) 2.9 (7.1) 25.9 (26.6) 16.2 (23.6)
Open in a new tab

For a full breakdown of sleep disorders, refer to Table S1.

AHI, apnea–hypopnea index; BMI, body mass index; SD, standard deviation.

3.1. Accuracy, sensitivity and specificity

The WSA had high overall accuracy and sensitivity but moderately poor specificity, compared with polysomnography (Table 2). LMMs showed significantly lower mean (standard error [SEM]) accuracy of 7% (± 1.4%), F 272 = 25.2, p < 0.001, marginal R 2 = 0.08, and specificity of 19.7% (± 2.5%), F 277.2 = 64.2, p < 0.001, marginal R 2 = 0.18, during daytime sleep opportunities compared with nighttime sleep opportunities. There was no significant difference in sensitivity (p > 0.05). LMMs also showed significantly higher accuracy of 7.2% (± 1.7%), F 181 = 18.5, p < 0.001, marginal R 2 = 0.08, and sensitivity of 4.4% (± 1.5%), F 190.9 = 8.6, p = 0.004, marginal R 2 = 0.04, for healthy sleepers compared with those with a diagnosed or suspected sleep disorder.

TABLE 2.

Epoch‐by‐epoch performance of WSA versus polysomnography

Accuracy % Sensitivity % Specificity %
Healthy sleeps
Nighttime 88.7 (7.8) 97.2 (3.3) 46.7 (19.9)
Daytime 81.8 (12) 96.5 (6.6) 26 (20.2)
Sleep disorders
Nighttime 80.6 (13.3) 91.4 (12.6) 46 (26.3)
Overall 83 (12) 95.1 (8.8) 36.9 (24.3)
Open in a new tab

There were no data available for daytime sleep opportunities in patients with sleep disorders.

Fully adjusted models (Table 3) show that the group effect of sleep disorder status on accuracy was fully explained by age, BMI and polysomnography‐derived SE. The effect on specificity was partially explained by age. The group effect of nighttime versus daytime sleep opportunity on accuracy was partially explained by age and SE. The effect on specificity was partially explained by SE.

TABLE 3.

Predictors of epoch‐by‐epoch model performance

Accuracy % Sensitivity % Specificity %
Predictor Estimate p Estimate p Estimate p
Sleep disorder status
Sleep disorder 1.95 0.271 4.38 0.031 −9.73 0.046
Age −0.10 0.042 −0.01 0.844 −0.41 0.002
BMI −0.18 0.018 −0.12 0.142 −0.12 0.550
SE 0.49 < 0.001 −0.08 0.121 0.02 0.896
Sleep opportunity timing
Daytime −5.50 < 0.001 −0.86 0.252 −20.95 < 0.001
Age −0.16 0.002 −0.00 0.941 −0.31 0.071
BMI 0.12 0.328 0.09 0.385 0.06 0.896
SE 0.56 <0.001 0.02 0.458 −0.20 0.020
Open in a new tab

Note: Bolded p values represent statistically significant differences. BMI, body mass index; SE, sleep efficiency.

3.2. Sleep characteristics

Sleep characteristics as measured by polysomnography and estimated by the WSA are found in Table 4, including subsets of data for secondary analyses. Overall, the WSA significantly overestimated TST, SE and SOL, and significantly underestimated WASO, compared with polysomnography. This was also reflected in sleep stages, where light, deep and REM sleep are typically underestimated, and wake is overestimated.

TABLE 4.

Sleep characteristics of WSA versus polysomnography during nighttime and daytime recordings

PSG WSA Difference (WSA – PSG)
Overall
TST 413.1 (84.4) 460.6 (75.3) 47.5 (81.2)*
SE (%) 79.3 (14) 88.5 (11.5) 9.2 (14.8)*
SOL 17.9 (30.6) 24.2 (29.7) 6.4 (26.1)*
WASO 90.1 (71.8) 36.2 (59.6) −53.9 (77.9)*
Wake 108 (76.2) 60.4 (67.3) −47.6 (81.2)*
Light sleep 214.2 (63.3) 238 (67.2) 23.8 (86.4)*
Deep sleep 113.7 (44.7) 119.4 (52.4) 5.7 (60)*
REM sleep 84.7 (35) 102.7 (49.2) 18 (45.4)*
Healthy sleep – nighttime
TST 436 (81) 472 (70.1) 36 (45.9)*
SE (%) 82.3 (12) 89.2 (9.4) 7 (8.6)*
SOL 30 (38.9) 34.1 (39) 4.1 (20.6)
WASO 64.3 (56) 24.1 (37.8) −40.2 (42.5)*
Wake 94.3 (68.9) 58.3 (56.8) −36 (45.9)*
Light sleep 253.5 (57) 236.1 (67.2) −17.4 (68.8)
Deep sleep 98.1 (35.8) 129.8 (48) 31.8 (51.2)*
REM sleep 84.2 (34) 105.9 (38.4) 21.7 (30.9)*
Healthy sleep – daytime
TST 420.6 (84.5) 486.8 (58.4) 66.2 (79)*
SE (%) 78.9 (14.5) 91.5 (9.4) 12.7 (13.6)*
SOL 5.3 (5.1) 13.5 (11.2) 8.3 (10)*
WASO 107.4 (78.2) 32.9 (61.4) −74.5 (79.2)*
Wake 112.7 (78.4) 46.3 (62.5) −66.4 (78.9)*
Light sleep 192.3 (50.1) 239.9 (60.2) 47.6 (77.1)*
Deep sleep 132.1 (34.2) 125.2 (38.3) −6.9 (46.3)
REM sleep 95 (30.6) 120.7 (45.3) 25.7 (47.1)*
Disordered sleep – nighttime
TST 384.6 (79.5) 412.7 (79.2) 28.1 (97.9)*
SE (%) 77.9 (14.6) 83.5 (14.1) 5.7 (18.8)*
SOL 27.6 (38) 32.9 (35.2) 5.2 (41.8)
WASO 83.5 (65.5) 50.2 (67.4) −33.4 (88.4)*
Wake 111.2 (77.3) 83.1 (75.5) −28.1 (97.9)*
Light sleep 217.4 (70.7) 236.5 (77.1) 19.2 (98.6)
Deep sleep 97.7 (53.6) 102.9 (68.1) 5.2 (76.4)
REM sleep 69.5 (36.5) 73.2 (48.4) 3.8 (48.7)
Open in a new tab

Data reflect mean (SD), and units are minutes unless otherwise indicated.

PSG, polysomnography; REM, rapid eye movement; TST, total sleep time; SE, sleep efficiency; SOL, sleep‐onset latency; WASO, wake after sleep onset; WSA, Withings Sleep Analyser.

*

p < 0.05 in mixed model analyses between WSA and polysomnography.

Participants had lower TST, SE and SOL, and higher WASO in daytime sleep opportunities compared with nighttime sleep opportunities, and healthy participants had higher TST and SE, and lower WASO compared with those with a diagnosed or suspected sleep disorder. There were significant device by sleep opportunity timing interactions for all but SOL and REM sleep duration, where metrics were over‐ or underestimated to a greater degree during daytime sleep opportunities. There were also significant device by sleep disorder status interactions for light, deep and REM sleep duration, where light sleep was overestimated, and deep and REM sleep were underestimated to a greater degree in individuals with a sleep disorder, compared with healthy individuals. A detailed breakdown of interactions can be found in Table S2.

Bland–Altman plots in Figure 1 highlight the significant increase in mean bias for TST and SE estimates, and decrease for WASO estimates, during daytime versus nighttime sleep opportunities. Furthermore, these plots show the significant proportional bias that was evident in daytime and nighttime estimated TST, SOL and WASO, as well as daytime SE. Bland–Altman plots of healthy sleepers compared with those with a diagnosed or suspected sleep disorder did not show notable differences, as shown in Figure S1.

FIGURE 1.

FIGURE 1

Open in a new tab

Bland–Altman plots of TST, SE, SOL and WASO during nighttime (left) and daytime (right) sleep opportunities. Red solid lines indicate mean bias, with dashed red 95% CIs. Grey solid lines indicate limits of agreement (mean bias ±1.96 standard deviation), with dashed grey 95% CIs. CI, confidence interval; SE, sleep efficiency; SOL, sleep‐onset latency; TST, total sleep time; WASO, wake after sleep onset.

3.3. Sleep stages

Confusion matrices in Figure 2 demonstrate that during nighttime sleep opportunities with healthy sleepers, the WSA accurately classified 50% of wake and 91% of sleep epochs. For daytime sleep opportunities, the WSA accurately classified 27% of wake and 90% of sleep. Sensitivity for sleep stage classification ranged from 62% to 75% during nighttime sleep opportunities, and 60%–69% during daytime sleep opportunities. For individuals with a diagnosed or suspected sleep disorder, the WSA accurately classified 43% of wake and 78% of sleep, with sleep stage accuracy ranging from 45% to 59%.

FIGURE 2.

FIGURE 2

Open in a new tab

Confusion matrices showing Withings Sleep Analyser (WSA) versus polysomnography four‐stage classification for healthy sleepers during nighttime and daytime recordings, and people with a sleep disorder during nighttime recordings.

3.4. WSA versus FC4 accuracy

Data from a subset of 22 healthy participants (nighttime recordings = 35, daytime = 160, mean [SD] age = 31.3 [12.4] years, 10 male, 12 female) were used to compare sleep estimation performance of the WSA and FC4 versus polysomnography. Figure 3 shows the sleep‐stage classification confusion matrices between devices for both nighttime and daytime sleep opportunities. There was a significant interaction effect of device by timing of sleep opportunity for accuracy, F 363.5 = 6.5, p = 0.01, marginal R 2 = 0.08, and specificity, F 365.3 = 4.3, p = 0.04, marginal R 2 = 0.17, whereby the FC4 showed significantly higher accuracy (5% ± 1%) and specificity (19.1% ± 2.2%) than the WSA during daytime sleep opportunities, but did not significantly differ during nighttime sleep opportunities. In addition, the WSA had significantly poorer accuracy (8.7% ± 1.7%) and specificity (15.7% ± 3.6%) during daytime sleep opportunities, compared with night, while the FC4 did not. This effect is evident in the confusion matrices, where daytime sleep opportunities saw a larger reduction in wake classification accuracy for the WSA (43% to 28%) compared with the FC4 (53% to 47%). The interaction effect of device by timing of sleep opportunity was also significant, p = 0.006, but post‐hoc analyses revealed no significant comparisons.

FIGURE 3.

FIGURE 3

Open in a new tab

Confusion matrices showing four‐stage classification, compared with polysomnography, for WSA versus FC4. Sleep stage classification compared four‐stage estimation (device “wake” = PSG wake; device “light” = PSG N1; device “deep” = PSG N2 + N3; device “REM” = PSG REM). FC4, Fitbit Charge 4; N1, stage 1 sleep; N2, stage 2 sleep; N3, Stage 3 sleep; PSG, polysomnography; REM, rapid eye movement; WSA, Withings Sleep Analyser.

Comparisons were conducted between WSA and FC4 estimations of TST, SE and WASO, with mean (SD) values seen in Table 5. Main effects of sleep opportunity timing and device were significant for all sleep outcomes (p < 0.05), but there were no significant interactions between timing of sleep opportunity and device. Post‐hoc analyses revealed that both the WSA and FC4 overestimated TST and SE, and underestimated WASO, while only the WSA significantly overestimated SOL (p < 0.05). Post‐hoc analyses also showed that the FC4 was significantly closer than the WSA to polysomnography in estimates of TST by 20 (± 8.1) min, p = 0.04, SE by 3.5% (± 1.3%), p = 0.03, and WASO by 21.5 (± 7.5) min, p = 0.01.

TABLE 5.

Sleep characteristics of WSA and FC4 versus polysomnography during nighttime and daytime recordings

PSG WSA WSA–PSG FC4 FC4–PSG WSA–FC4
Overall
TST 426.1 (82.9) 486.3 (57) +60.2 (77.4) 467.1 (68.6) +41 (57.5) +9.6 (33.6)
SE 79.9 (14.2) 91.4 (9.2) +11.5 (13.3) 88 (10.6) +8.2 (11.1) +1.7 (5.3)
SOL 8.8 (11.8) 17 (15.8) +8.2 (12.6) 11.6 (23.8) +2.8 (20.6) +2.7 (10.6)
WASO 98.7 (77.8) 30.2 (60.1) −68.5 (78.3) 52.3 (52.5) −46.5 (57.1) −11 (31)
Night
TST 454.9 (57.5) 494.5 (30.9) +39.6 (42.1) 473.4 (51.1) +18.5 (52.7) +10.6 (22.4)
SE 84.7 (10.8) 92 (5.8) +7.4 (7.9) 88.3 (8.6) +3.7 (9.3) +1.9 (3.7)
SOL 23.1 (20.2) 29.1 (24.1) +6 (19.7) 27.4 (41.9) +4.3 (37.2) +0.9 (19.2)
WASO 59.4 (54) 13.8 (18.2) −45.6 (45.1) 34.5 (25.8) −24.9 (49.7) −10.4 (11.7)
Day
TST 419.8 (86.3) 484.5 (61.2) +64.8 (82.6) 465.7 (71.9) +45.9 (57.5) +9.4 (35.6)
SE 78.8 (14.6) 91.2 (9.8) +12.4 (14) 87.9 (11) +9.1 (11.3) +1.6 (5.6)
SOL 5.7 (5.2) 14.4 (11.9) +8.7 (10.5) 8.2 (15.9) +2.5 (14.9) +3.1 (7.6)
WASO 107.3 (79.6) 33.8 (65.3) −73.5 (83.1) 56.1 (56) −51.2 (57.7) −11.1 (33.9)
Open in a new tab

FC4, Fitbit Charge 4; PSG, polysomnography; SE, sleep efficiency; SOL, sleep‐onset latency; TST, total sleep time; WASO, wake after sleep onset; WSA, Withings Sleep Analyser.

3.5. Data loss

Within the subset where both devices were used (n = 25, 248 possible recordings), a total of 17 recordings (mean [SD] = 0.68 [0.9] per participant) were lost with the WSA compared with 53 (2.12 [3.38] per participant) with the FC4. The paired‐samples t‐test confirmed that this was a significant difference, t 24 = 2.42, p = 0.02, d = 0.58. WSA data loss occurred entirely due to mats being unintentionally left unpowered from prior sleep studies. FC4 data loss occurred due to improper charging (both user and mechanical error), syncing errors and improper wear.

3.6. Multi‐night performance

For participants with more than one recording (n = 49, 282 nights), variability in accuracy, sensitivity and specificity is demonstrated in Figure 4. Individuals with healthy sleep during a nighttime sleep opportunity had mean (SD) variance in accuracy, sensitivity and specificity of 5.01% (9.89%), 0.98% (0.85%) and 22.3% (19.6%), respectively. Mixed models showed significantly greater variability (represented as the mean CV) in accuracy, 2.33% ± 2.59%, p = 0.03, and specificity, 6.03% ± 6.85%, p < 0.001, for daytime compared with nighttime sleep opportunities. Sensitivity was more variable in individuals with a sleep disorder, 2.42% ± 3.02%, p = 0.02. The Levene's test also indicated that the distribution of sensitivity variability differed (6.12, p = 0.02), seen in Figure 4 as the wider distribution of variability for individuals with disordered sleep. No other significant differences in distribution were found.

FIGURE 4.

FIGURE 4

Open in a new tab

Withings Sleep Analyser (WSA) versus polysomnography variability in accuracy for (a) daytime versus nighttime sleep opportunities; and (b) individuals with healthy versus disordered sleep. Larger mean coefficients of variation reflect greater multi‐night variability in WSA performance. Wider plots reflect a greater range of variability in performance across individuals in each category.

4. DISCUSSION

This study comprehensively evaluated the performance of the WSA to estimate sleep and wake compared with polysomnography. Appropriate use of such consumer sleep technology requires adequate performance evaluation (de Zambotti et al., 2019). Here, we completed one of the largest independent performance evaluations of a consumer sleep tracker to date. Results indicate that the WSA was accurate compared with polysomnography to classify sleep and wake states, with moderate‐to‐poor accuracy for sleep staging, which was comparable to existing wearable devices (Chinoy et al., 2020; Lee et al., 2023). The WSA systematically overestimated TST, SE and SOL, and underestimated WASO, compared with polysomnography. Accuracy and specificity were comparable to existing devices during nighttime sleep opportunities with healthy sleepers, but poorer during daytime sleep opportunities and poorer to estimate sleep in people with sleep disorders. These effects were partially explained by lower polysomnography‐derived SE during these instances. Overall, the WSA demonstrates relatively minimal bias in TST and SE estimates that is comparable to other validated consumer sleep tracking devices (Haghayegh et al., 2019; Lee et al., 2023), but performance was worse for sleep periods with more wakefulness.

These data are consistent with previous studies in which movement‐based sleep trackers often misclassify motionless wake as sleep (Chinoy et al., 2020; Miller et al., 2022). However, few studies have investigated circumstances where poorer than normal sleep is likely to impact reliable sleep estimation. Compared with one study with actigraphy devices (Gao et al., 2022), the WSA performs relatively poorly for reliably detecting wake during daytime recordings, with 72% of wake misclassified as sleep. However, in the subset of participants with a diagnosed or suspected sleep disorder, the lower accuracy was driven by a reduction in sensitivity (i.e. relatively more sleep being misclassified as wake). This may be due to the especially high misclassification of light sleep by the WSA as wake or deeper sleep, and the relatively high proportion of such sleep in people with a sleep disorder (OSA in particular; Eckert & Younes, 2014; Ratnavadivel et al., 2009; Shahveisi et al., 2018). This suggests that the WSA, and likely other sleep tracking devices (Cook et al., 2019; Dong et al., 2022; Kang et al., 2017; Te Lindert et al., 2020), are less accurate to classify sleep and wake in individuals with a sleep disorder. Whether these devices are accurate enough for a given purpose is difficult to determine from the accuracy data alone. Performance evaluation, together with investigation into the potential utility of such sleep estimations, is required to make this judgement. We have shown that the WSA has utility to estimate cognitive performance under simulated shift work conditions (Manners et al., 2024), and other studies have found associations between wearable device estimations of sleep and cognitive fatigue or associated outcomes (Adão Martins et al., 2021; Zhu et al., 2017). Thus, further work is needed to test how the proportion of sleep misclassification observed in the current study may impact the device utility in clinical, occupational and related contexts.

Multi‐night performance results demonstrated that WSA accuracy and sensitivity are reasonably stable across nights, while specificity is more variable, particularly in contexts of poor sleep. This has significant implications for longitudinal monitoring with this, and similar devices, where performance compared with polysomnography is generally considered to be stable. Given that performance was more variable in daytime sleep opportunities and people with sleep disorders, observed (or a lack of) differences in objective sleep pre–post an event (e.g. shift work initiation, sleep disorder treatment) may be biased. For example, inconsistent findings in objective sleep changes after cognitive behavioural therapy for insomnia (Mitchell et al., 2019; Squires et al., 2022) may be due to poorer performance on poorer nights of sleep. Similarly, evidence suggests small or inconsistent differences in objective sleep quality between shift workers and daytime workers (Chang & Peng, 2021; Sathvik et al., 2022; Shin et al., 2021), but this could potentially be due to more variable device performance in shift workers, resulting in poor sleep being more often misclassified as adequate. Future work is likely necessary to test the within‐subject accuracy of sleep tracking devices pre–post interventions, particularly in populations prone to producing poorer performance from sleep tracking devices such as shift workers or those with chronic insomnia.

This study also showed that the FC4 had more reliable sleep classification accuracy than the WSA. Specifically, the FC4 showed higher accuracy and specificity (more accurate wake classification) during daytime sleep opportunities. This was further evidenced in sleep characteristics, where the FC4 overestimated TST by about 20 minutes less, overestimated SE by about 4% less, and underestimated WASO by about 20 min less than the WSA. This may be due to limitations in the modality of the two devices, where relatively motionless wake could be interpreted more accurately by a wrist‐worn device than an under‐mattress sensor. It may be possible that, given perceived restriction to movement imposed by polysomnography equipment, participants were less inclined to change position or move during wake. This effect would likely be more prominent in chest movement, as the WSA primarily estimates, than in wrist movements measured by the FC4. Other differences between device algorithms may also account for the differences between WSA and FC4 accuracy, particularly in the incorporation of heart rate and breathing signals. Overall, the FC4 device was more accurate at classifying sleep than the WSA, particularly during daytime sleep opportunities, but further comparisons are warranted to examine whether this is maintained in a naturalistic environment.

There are additional considerations that should be noted when critically evaluating sleep tracking devices. The WSA, unlike wrist‐worn devices, does not need charging or manual synchronization of data. As such, in the study used for the sub‐sample device comparison, we found a data‐loss rate of 7% (compared with 21% data loss with the FC4). As opposed to the laboratory environment with technicians ensuring suitable device set‐up and suitable use to protect data fidelity, data loss would be expected to be higher in a naturalistic environment. This greater data loss may be higher for the FC4 device than the WSA, given that, once set up, the WSA does not require user input or effort to continue recording, whereas wrist‐worn devices must be charged, worn correctly and routinely synchronized for lossless data capture. Additionally, the WSA has been found to be highly accurate at bed occupancy timing (i.e. in/out of bed times) and duration compared with polysomnography with video (Ravindran et al., 2021), which wearable devices are typically less accurate at detecting in home environments (Feehan et al., 2018; Haghayegh et al., 2019; Hamill et al., 2020). Thus, the WSA has practical advantages over other forms of sleep tracking devices that should be taken into consideration when selecting a device for use, alongside its accuracy for sleep/wake detection.

Although there are strengths to the large sample size, there are still limitations to this study. Firstly, as with most consumer sleep technologies, the specifics of the sleep and respiratory algorithms that the WSA and FC4 use are not publicly available. As such, we are limited in our interpretation of what is driving inaccuracies, particularly in terms of how real versus inferred movement, heart rate and breathing interact with sleep staging. Secondly, some of the included research studies with individuals with a diagnosed or suspected sleep disorder had pre‐sleep events that may have impacted sleep, such as drug versus placebo interventions and respiratory testing. Given that SE partially explained WSA performance compared with polysomnography (by about 30%–35%), the direct effects of such covariates on sleep quality likely account for some variability in WSA classification performance. Similarly, most of the participants with a sleep disorder were older than the healthy cohort, which limits the generalizability of the present findings to healthy older adults. The extensive and heterogeneous sample provides opportunity to evaluate performance with the expected variability in real‐world use (e.g. variable continuous positive airway pressure use, acute sedative use, noise disruptions, etc.), but also highlights the need for further testing in populations not covered here, including children and adolescents, or individuals with other sleep disorders. It should also be considered that this study was conducted using data from entirely in‐laboratory sleep research protocols, and in many cases single‐night laboratory studies. Given known effects of comfort and sleeping position on sleep and breathing (Mueller et al., 2022), as well as known first‐night effects (Agnew Jr. et al., 1966), sleep in a naturalistic setting will differ, and the extent to which this impacts WSA performance is unknown. Finally, there were insufficient numbers to compare device accuracy between individual sleep disorders, nor were there data from other consumer sleep trackers, and this should be elucidated in further work.

4.1. Conclusions

This study extensively compared the WSA with polysomnography in a large sample of healthy individuals and people with a diagnosed or suspected sleep disorder during nighttime and daytime sleep opportunities to provide novel insights into performance characteristics. Overall, the WSA was accurate at sleep and wake detection during nighttime recordings compared with other consumer sleep trackers, but was less accurate at wake classification during daytime recordings. The WSA was also less accurate at sleep classification in people with a suspected or diagnosed sleep disorder. WSA accuracy and sensitivity was relatively stable, but specificity was variable. Performance variability was stronger in daytime sleep opportunities and in people with disordered sleep. The WSA overestimated TST, SE and SOL, and underestimated WASO to comparable levels seen with other consumer sleep trackers (Chinoy et al., 2020; Lee et al., 2023). Lower polysomnography‐derived SE was associated with worse WSA wake classifications, but this only partially explained the effect of wake on classification accuracy. In a subset of participants, the FC4 was highly comparable to the WSA during nighttime recordings but, consistent with better wake classification performance, was more accurate during daytime recordings. Ultimately, the WSA may be suitable to estimate sleep and wake in a variety of naturalistic environments, and the final choice should depend on the performance, logistics and pragmatic needs of the clinical practice or research study. The current study findings provide important novel data that may inform these decisions.

AUTHOR CONTRIBUTIONS

Jack Manners: Conceptualization; writing – original draft; methodology; visualization; writing – review and editing; data curation; formal analysis; software; investigation. Eva Kemps: Writing – review and editing; conceptualization; supervision; resources; data curation; funding acquisition. Bastien Lechat: Formal analysis; methodology; writing – review and editing; data curation; resources; conceptualization; validation; investigation. Peter Catcheside: Supervision; data curation; funding acquisition; writing – review and editing; methodology; conceptualization; project administration; resources. Danny J. Eckert: Supervision; resources; data curation; writing – review and editing; funding acquisition; investigation; project administration. Hannah Scott: Investigation; conceptualization; project administration; supervision; data curation; methodology; visualization; validation; writing – review and editing; resources.

CONFLICT OF INTEREST STATEMENT

Financial Disclosure: DJE and BL have had an investigator‐initiated research grant supported by Withings. Non‐Financial Disclosures: None.

Supporting information

DATA S1. Supplementary Appendix.

JSR-34-e14480-s001.docx (493.8KB, docx)

ACKNOWLEDGEMENTS

The authors thank and acknowledge the support from our colleagues at Flinders University and FHMRI: Sleep Health. This study was investigator‐driven work supported by the Defence Science and Technology Group (DSTG). The authors also thank all participants involved in this study. Open access publishing facilitated by Flinders University, as part of the Wiley ‐ Flinders University agreement via the Council of Australian University Librarians.

Manners, J. , Kemps, E. , Lechat, B. , Catcheside, P. , Eckert, D. J. , & Scott, H. (2025). Performance evaluation of an under‐mattress sleep sensor versus polysomnography in > 400 nights with healthy and unhealthy sleep. Journal of Sleep Research, 34(6), e14480. 10.1111/jsr.14480

Danny J. Eckert and Hannah Scott are co‐senior authors.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

REFERENCES

  1. Adão Martins, N. R. , Annaheim, S. , Spengler, C. M. , & Rossi, R. M. (2021). Fatigue monitoring through wearables: A state‐of‐the‐art review. Frontiers in Physiology, 12, 790292. 10.3389/fphys.2021.790292 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Agnew, H. W., Jr. , Webb, W. B. , & Williams, R. L. (1966). The first night effect: An eeg study of SLEEP. Psychophysiology, 2, 263–266. 10.1111/j.1469-8986.1966.tb02650.x [DOI] [PubMed] [Google Scholar]
  3. Bates, D. , et al. (2015). Package ‘Lme4’. convergence 12, 2.
  4. Chang, W. P. , & Peng, Y. X. (2021). Meta‐analysis of differences in sleep quality based on actigraphs between day and night shift workers and the moderating effect of age. Journal of Occupational Health, 63(1), e12262. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Chinoy, E. D. , Cuellar, J. A. , Huwa, K. E. , Jameson, J. T. , Watson, C. H. , Bessman, S. C. , Hirsch, D. A. , Cooper, A. D. , Drummond, S. P. A. , & Markwald, R. R. (2020). Performance of seven consumer sleep‐tracking devices compared with polysomnography. Sleep, 44(5), zsaa291. 10.1093/sleep/zsaa291 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Conley, S. , et al. (2019). Agreement between Actigraphic and polysomnographic measures of sleep in adults with and without chronic conditions: A systematic review and meta‐analysis. Sleep Medicine Reviews, 46, 151–160. 10.1016/j.smrv.2019.05.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Cook, J. D. , Eftekari, S. C. , Leavitt, L. A. , Prairie, M. L. , & Plante, D. T. (2019). Optimizing Actigraphic estimation of sleep duration in suspected idiopathic hypersomnia. Journal of Clinical Sleep Medicine, 15, 597–602. 10.5664/jcsm.7722 04. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. de Zambotti, M. , Cellini, N. , Goldstone, A. , Colrain, I. M. , & Baker, F. C. (2019). Wearable sleep Technology in Clinical and Research Settings. Medicine and Science in Sports and Exercise, 51, 1538–1557. 10.1249/MSS.0000000000001947 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. De Zambotti, M. , Goldstein, C. , Cook, J. , Menghini, L. , Altini, M. , Cheng, P. , & Robillard, R. (2024). State of the science and recommendations for using wearable technology in sleep and circadian research. Sleep, 47(4), zsad325. 10.1093/sleep/zsad325. [DOI] [PubMed] [Google Scholar]
  10. de Zambotti, M. , Goldstone, A. , Claudatos, S. , Colrain, I. M. , & Baker, F. C. (2017). A validation study of Fitbit charge 2™ compared with polysomnography in adults. Chronobiology International, 35(4), 465–476. 10.1080/07420528.2017.1413578 [DOI] [PubMed] [Google Scholar]
  11. Dong, X. , et al. (2022). Validation of Fitbit charge 4 for assessing sleep in Chinese patients with chronic insomnia: A comparison against polysomnography and Actigraphy. PLoS One, 17, e0275287. 10.1371/journal.pone.0275287 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Eckert, D. J. , & Younes, M. K. (2014). Arousal from sleep: Implications for obstructive sleep apnea pathogenesis and treatment. Journal of Applied Physiology, 116(3), 302–313. 10.1152/japplphysiol.00649.2013 [DOI] [PubMed] [Google Scholar]
  13. Edouard, P. , Campo, D. , Bartet, P. , Yang, R.‐Y. , Bruyneel, M. , Roisman, G. , & Escourrou, P. (2021). Validation of the Withings sleep analyzer, an under‐the‐mattress device for the detection of moderate‐severe sleep apnea syndrome. Journal of Clinical Sleep Medicine, 17(6), 1217–1227. 10.5664/jcsm.9168 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Euser, A. M. , Dekker, F. W. , & Le Cessie, S. (2008). A practical approach to bland‐Altman plots and variation coefficients for log transformed variables. Journal of Clinical Epidemiology, 61(10), 978–982. 10.1016/j.jclinepi.2007.11.003 [DOI] [PubMed] [Google Scholar]
  15. Feehan, L. M. , et al. (2018). Accuracy of Fitbit devices: Systematic review and narrative syntheses of quantitative data. JMIR mHealth and uHealth, 6, e10527. 10.2196/10527 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Fleur, R. G. S. , George, S. M. S. , Leite, R. , Kobayashi, M. , Agosto, Y. , & Jake‐Schoffman, D. E. (2021). Use of Fitbit devices in physical activity intervention studies across the life course: Narrative review. JMIR mHealth and uHealth, 9(5), e23411. 10.2196/23411 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Gao, C. , Li, P. , Morris, C. J. , Zheng, X. , Ulsa, M. C. , Gao, L. , Scheer, F. A. J. L. , & Hu, K. (2022). Actigraphy‐based sleep detection: Validation with polysomnography and comparison of performance for nighttime and daytime sleep during simulated shift work. Nature and Science of Sleep, 14, 1801–1816. 10.2147/NSS.S373107 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Haghayegh, S. , Khoshnevis, S. , Smolensky, M. H. , Diller, K. R. , & Castriotta, R. J. (2019). Accuracy of wristband Fitbit models in assessing sleep: Systematic review and meta‐analysis. Journal of Medical Internet Research, 21(11), e16273. 10.2196/16273 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Hamill, K. , Jumabhoy, R. , Kahawage, P. , de Zambotti, M. , Walters, E. M. , & Drummond, S. P. A. (2020). Validity, potential clinical utility and comparison of a consumer activity tracker and a research‐grade activity tracker in insomnia disorder II: Outside the laboratory. Journal of Sleep Research, 29, e12944. 10.1111/jsr.12944 [DOI] [PubMed] [Google Scholar]
  20. Huang, Y. , Upadhyay, U. , Dhar, E. , Kuo, L.‐J. , & Syed‐Abdul, S. (2022). A scoping review to assess adherence to and clinical outcomes of wearable devices in the cancer population. Cancers, 14, 4437. 10.3390/cancers14184437 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Iber, C. (2007). The AASM manual for the scoring of sleep and associated events: Rules. In Terminology and technical specification. American Academy of Sleep Medicine. [Google Scholar]
  22. Kainec, K. A. , et al. (2024). Evaluating accuracy in five commercial sleep‐tracking devices compared to research‐grade Actigraphy and polysomnography. Sensors, 24(2), 635. 10.3390/s24020635 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Kang, S.‐G. , et al. (2017). Validity of a commercial wearable sleep tracker in adult insomnia disorder patients and good sleepers. Journal of Psychosomatic Research, 97, 38–44. 10.1016/j.jpsychores.2017.03.009 [DOI] [PubMed] [Google Scholar]
  24. Lechat, B. , Naik, G. , Appleton, S. , Manners, J. , Scott, H. , Nguyen, D. P. , Escourrou, P. , Adams, R. , Catcheside, P. , & Eckert, D. J. (2024). Regular snoring is associated with uncontrolled hypertension. Npj Digital Medicine, 7, 1–8. 10.1038/s41746-024-01026-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Lechat, B. , et al. (2022). Multinight prevalence, variability, and diagnostic misclassification of obstructive sleep apnea. American Journal of Respiratory and Critical Care Medicine, 205(5), 563–569. 10.1164/rccm.202107-1761OC [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Lechat, B. , et al. (2023). Multi‐night measurement for diagnosis and simplified monitoring of obstructive sleep Apnoea. Sleep Medicine Reviews, 72, 101843. 10.1016/j.smrv.2023.101843 [DOI] [PubMed] [Google Scholar]
  27. Lee, T. , Cho, Y. , Cha, K. S. , Jung, J. , Cho, J. , Kim, H. , Kim, D. , Hong, J. , Lee, D. , Keum, M. , Kushida, C. A. , Yoon, I.‐Y. , & Kim, J.‐W. (2023). Accuracy of 11 wearable, nearable, and Airable consumer sleep trackers: Prospective multicenter validation study. JMIR mHealth and uHealth, 11, e50983. 10.2196/50983 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Manners, J. , et al. (2024). Estimating vigilance from the pre‐work shift sleep using an under‐mattress sleep sensor. Journal of Sleep Research, 33(5), e14138. 10.1111/jsr.14138 [DOI] [PubMed] [Google Scholar]
  29. Menghini, L. , Cellini, N. , Goldstone, A. , Baker, F. C. , & de Zambotti, M. (2021). A standardized framework for testing the performance of sleep‐tracking technology: Step‐by‐step guidelines and open‐source code. Sleep, 44(2), zsaa170. 10.1093/sleep/zsaa170 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Miller, D. J. , Sargent, C. , & Roach, G. D. (2022). A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults. Sensors, 22(16), 6317. 10.3390/s22166317 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Mitchell, L. J. , Bisdounis, L. , Ballesio, A. , Omlin, X. , & Kyle, S. D. (2019). The impact of cognitive behavioural therapy for insomnia on objective sleep parameters: A meta‐analysis and systematic review. Sleep Medicine Reviews, 1(47), 90–102. [DOI] [PubMed] [Google Scholar]
  32. Moreno‐Pino, F. , Porras‐Segovia, A. , López‐Esteban, P. , Artés, A. , & Baca‐García, E. (2019). Validation of Fitbit charge 2 and Fitbit Alta HR against polysomnography for assessing sleep in adults with obstructive sleep apnea. Journal of Clinical Sleep Medicine, 15(11), 1645–1653. 10.5664/jcsm.8032 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Mueller, C. E. , Li, H. , Begasse, S. M. , et al. (2022). Sleep position, patient comfort, and technical performance with two established procedures for home sleep testing. Sleep & Breathing, 26, 1673–1681. 10.1007/s11325-021-02530-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Paquet, J. , Kawinska, A. , & Carrier, J. (2007). Wake detection capacity of Actigraphy during sleep. Sleep, 30, 1362–1369. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Ratnavadivel, R. , et al. (2009). Marked reduction in obstructive sleep apnea severity in slow wave sleep. Journal of Clinical Sleep Medicine, 5(6), 519–524. 10.5664/jcsm.27651 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Ravindran, K. K. G. , Monica, C. , Atzori, G. , Enshaeifar, S. , Mahvash‐Mohammadi, S. , Dijk, D.‐J. , & Revell, V. (2021). Validation of technology to monitor sleep and bed occupancy in older men and women. Alzheimer's & Dementia, 17, e056018. 10.1002/alz.056018 [DOI] [Google Scholar]
  37. Sadeh, A. , Sharkey, M. , & Carskadon, M. A. (1994). Activity‐based sleep‐wake identification: An empirical test of methodological issues. Sleep, 17(3), 201–207. 10.1093/sleep/17.3.201 [DOI] [PubMed] [Google Scholar]
  38. Sathvik, S. , Krishnaraj, L. , & Irfan, M. (2022). Evaluation of sleep quality and duration using wearable sensors in shift laborers of construction industry: A public health perspective. Frontiers in Public Health, 10, 952901. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
  39. Schutte, S. , Deak, M. C. , Khosla, S. , Goldstein, C. A. , Yurcheshen, M. , Chiang, A. , Gault, D. , Kern, J. , O'Hearn, D. , Ryals, S. , Verma, N. , Kirsch, D. B. , Baron, K. , Holfinger, S. , Miller, J. , Patel, R. , Bhargava, S. , & Ramar, K. (2021). Evaluating consumer and clinical sleep technologies: An American Academy of sleep medicine update. Journal of Clinical Sleep Medicine, 17(11), 2275–2282. 10.5664/jcsm.9580 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Scott, H. , Lechat, B. , Guyett, A. , Reynolds, A. C. , Lovato, N. , Naik, G. , Appleton, S. , Adams, R. , Escourrou, P. , Catcheside, P. , & Eckert, D. J. (2023). Sleep irregularity is associated with hypertension: Findings from over 2 million nights with a large global population sample. Hypertension, 80, 1117–1126. 10.1161/HYPERTENSIONAHA.122.20513 [DOI] [PubMed] [Google Scholar]
  41. Scott, H. , Naik, G. , Lechat, B. , Manners, J. , Fitton, J. , Nguyen, D. P. , Hudson, A. L. , Reynolds, A. C. , Sweetman, A. , Escourrou, P. , Catcheside, P. , & Eckert, D. J. (2024). Are we getting enough sleep? Frequent irregular sleep found in an analysis of over 11 million nights of objective in‐home sleep data. Sleep Health, 10(1), 91–97. 10.1016/j.sleh.2023.10.016 [DOI] [PubMed] [Google Scholar]
  42. Shahveisi, K. , Jalali, A. , Moloudi, M. R. , Moradi, S. , Maroufi, A. , & Khazaie, H. (2018). Sleep architecture in patients with primary snoring and obstructive sleep apnea. Basic and Clinical Neuroscience, 9, 147–156. 10.29252/NIRP.BCN.9.2.147 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Shin, S. , Kim, S. H. , & Jeon, B. (2021). Objective assessment of sleep patterns among night‐shift workers: A scoping review. International Journal of Environmental Research and Public Health, 18(24), 13236. 10.3390/ijerph182413236 [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Squires, L. R. , Rash, J. A. , Fawcett, J. , & Garland, S. N. (2022). Systematic review and meta‐analysis of cognitive‐behavioural therapy for insomnia on subjective and actigraphy‐measured sleep and comorbid symptoms in cancer survivors. Sleep Medicine Reviews, 63(1) Jun 2022, 101615. 10.1016/j.smrv.2022.101615 [DOI] [PubMed] [Google Scholar]
  45. Stucky, B. , et al. (2021). Validation of Fitbit charge 2 sleep and heart rate estimates against polysomnographic measures in shift workers: Naturalistic study. Journal of Medical Internet Research, 23(10), e26476. 10.2196/26476 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Sun, X. , Qiu, L. , Wu, Y. , Tang, Y. , & Cao, G. (2017). SleepMonitor. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 1(3), 1–22. 10.1145/3130969 [DOI] [Google Scholar]
  47. Te Lindert, B. H. W. , et al. (2020). Optimizing Actigraphic estimates of polysomnographic sleep features in insomnia disorder. Sleep, 43(11), zsaa090. 10.1093/sleep/zsaa090 [DOI] [PubMed] [Google Scholar]
  48. Tobin, S. Y. , Williams, P. G. , Baron, K. G. , Halliday, T. M. , & Depner, C. M. (2021). Challenges and opportunities for applying wearable technology to sleep. Sleep Medicine Clinics, 16, 607–618. 10.1016/j.jsmc.2021.07.002 [DOI] [PubMed] [Google Scholar]
  49. Yoon, H. , & Choi, S. H. (2023). Technologies for Sleep Monitoring at home: Wearables and nearables. Biomedical Engineering Letters, 13(3), 313–327. 10.1007/s13534-023-00305-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Zhu, Y. , Jankay, R. R. , Pieratt, L. C. , & Mehta, R. K. (2017). Wearable sensors and their metrics for measuring comprehensive occupational fatigue: A scoping review. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 61(1), 1041–1045. 10.1177/1541931213601744 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

DATA S1. Supplementary Appendix.

JSR-34-e14480-s001.docx (493.8KB, docx)

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Articles from Journal of Sleep Research are provided here courtesy of Wiley

RESOURCES