Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Method for Activity Sleep Harmonization (MASH): a novel method for harmonizing data from two wearable devices to estimate 24-h sleep–wake cycles

Method for Activity Sleep Harmonization (MASH): a novel method for harmonizing data from two... Background Daily 24‑h sleep–wake cycles have important implications for health, however researcher preferences in choice and location of wearable devices for behavior measurement can make 24‑h cycles difficult to estimate. Further, missing data due to device malfunction, improper initialization, and/or the participant forgetting to wear one or both devices can complicate construction of daily behavioral compositions. The Method for Activity Sleep Harmonization (MASH) is a process that harmonizes data from two different devices using data from women who concurrently wore hip (waking) and wrist (sleep) devices for ≥ 4 days. Methods MASH was developed using data from 1285 older community‑ dwelling women (ages: 60–72 years) who concurrently wore a hip‑ worn ActiGraph GT3X + accelerometer (waking activity) and a wrist‑ worn Actiwatch 2 device (sleep) for ≥ 4 days (N = 10,123 days) at the same time. MASH is a two‑tiered process using (1) scored sleep data (from Actiwatch) or (2) one‑ dimensional convolutional neural networks (1D CNN) to create predicted wake intervals, recon‑ cile sleep and activity data disagreement, and create day‑level night ‑ day‑night pairings. MASH chooses between two different 1D CNN models based on data availability (ActiGraph + Actiwatch or ActiGraph‑ only). MASH was evaluated using Receiver Operating Characteristic (ROC) and Precision‑Recall curves and sleep–wake intervals are compared before (pre‑harmonization) and after MASH application. Results MASH 1D CNNs had excellent performance (ActiGraph + Actiwatch ROC‑AUC = 0.991 and ActiGraph‑ only ROC‑AUC = 0.983). After exclusions (partial wear [n = 1285], missing sleep data proceeding activity data [n = 269], and < 60 min sleep [n = 9]), 8560 days were used to show the utility of MASH. Of the 8560 days, 46.0% had ≥ 1‑min disagreement between the devices or used the 1D CNN for sleep estimates. The MASH waking intervals were cor‑ rected (median minutes [IQR]: − 27.0 [− 115.0, 8.0]) relative to their pre‑harmonization estimates. Most correction (− 18.0 [− 93.0, 2.0] minutes) was due to reducing sedentary behavior. The other waking behaviors were reduced a median (IQR) of − 1.0 (− 4.0, 1.0) minutes. Erin E. Dooley and J. F. Winkles share first authorship. *Correspondence: Erin E. Dooley edooley@uab.edu Full list of author information is available at the end of the article © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 2 of 12 Conclusions Implementing MASH to harmonize concurrently worn hip and wrist devices can minimizes data loss and correct for disagreement between devices, ultimately improving accuracy of 24‑h compositions necessary for time‑use epidemiology. Keyword Actigraphy, Accelerometer, Sleep, Physical activity, Harmonization, Machine learning, 24‑h activity, Time ‑ use epidemiology Background such as independently funded ancillary studies to large Time-use movement behaviors (e.g., sleep, sedentary prospective cohorts with initial protocols proposing dif- behavior, and physical activity) [1–3] are modifiable fac - ferent devices. To reduce participant burden of wearing tors associated with numerous health outcomes and all- devices across many weeks of data collection, researchers cause mortality [4–8]. Wearable accelerometers are used may need to collaborate on a multiple-device wear proto- to measure time-use behaviors in free-living settings for col occurring over the course of one week, for example. a variety of populations [9, 10]. While there are numer- To characterize 24-h sleep–wake compositions for ous consumer wearables (e.g., smart watches, fitness multiple-device protocols, approaches to harmonizing monitors) that have the capacity to estimate waking and simultaneously collected data from multiple devices are sleep behaviors, potential issues such as user feedback needed. Unfortunately, data collection in naturalistic set- biases and data extraction limit their research utility [11, tings increases the potential for protocol deviations that 12]. Additionally, consumer device (e.g., Apple, Fitbit, undermine accuracy. Specifically, missing data due to Garmin) accuracy depends on manufacturer and device device malfunction, improper initialization, and/or the type and are liable to algorithm changes that may occur participant forgetting to wear one or both devices on numerous times throughout a study and without notifica - one or more days complicates construction of day-night tion to investigators, creating unmeasurable noise in the pairings. Another common issue researchers face is that data [13–17]. Research-grade devices (e.g., ActiGraph, despite being instructed to remove the device assessing activPAL, Actiwatch, GENEActiv) provide substantial waking activity during sleep periods, it is not uncommon flexibility over consumer devices for processing and re- for participants to wear these devices longer than neces- processing of data to achieve current best-practices. sary (i.e., to bed), which can inflate estimates of seden - However, devices for quantifying 24-h time-use behav- tary behavior time. Through this lens, a multiple-device iors can be placed on different anatomical locations (e.g., data harmonization process needs to facilitate two types hip, wrist, thigh) and can be worn for different amounts of adjustment when sleep and activity data are joined of time (e.g., waking only, sleep only, 24  h/day) depend- for 24-h cycle development. First, the frame of refer- ing on the primary outcome(s) of interest. For example, ence for a day should be defined as the concatenation of researchers only interested in physical activity behaviors a dynamic sleep–wake interval rather than a constrained may utilize waking hip placements, whereas those inter- period (e.g., midnight to midnight) that may be utilized ested in sedentary behaviors may want to consider pos- when behaviors are viewed separately [18, 22]. Second, tural positions and therefore need a device with a thigh harmonization should correct any overlap between mon- placement, and those interested in sleep and/or circadian itors if there is behavior categorization disagreement, phase may utilize wrist placements [18, 19]. which may address situations where sleep is incorrectly While a single wrist-worn device to capture movement classified as sedentary behavior or non-wear for one behaviors have become increasingly popular, validity for device. The latter could result in inaccuracies due to the measuring physical activity across the intensity spec- device (still) recording after it was removed. trum against criterion measures (e.g., indirect calorim- Herein, we present the Method for Activity Sleep Har- etry, doubly labeled water) has a wide range of accuracy monization (MASH) process, a novel method that har- (r = 0.17–0.93) [20]. This may lead some researchers to monizes data from multiple devices to create coherent implement protocols that have participants switch the sleep-activity pairings. MASH is a multiple device (hip device between the hip (during the day) and wrist (at and wrist) harmonization method that addresses many of night), potentially increasing participant non-compliance the issues described above including, missing data (e.g., [21], or consider protocols in which participants wear not wearing one device) or discordant behavior charac- two or more devices concurrently, as in the current study. terization (e.g., one device characterizes sleep whereas Other instances when a multiple-device protocol may be the other device characterizes sedentary behavior), while necessary include when multiple funded studies are being also accommodating both regular and irregular sleep pat- conducted simultaneously on a single study population, terns and minimizing data loss. We detail this method D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 3 of 12 using data from 1285 older women who concurrently Data collection wore an ActiGraph accelerometer on the hip and an Acti- Waking behaviors were quantified using the hip-worn watch 2 device on the wrist for up to seven days as part of ActiGraph wGT3X+ (ActiGraph, Pensacola, FL) device the Study of Women’s Health Across the Nation (SWAN). during all waking hours, except for water-based activities, for up to seven days. Raw acceleration data were sampled Methods at 40 Hz and were downloaded and reintegrated to a 60-s Parent study epoch using ActiLife6 software [18]. Wear and non-wear Data are from SWAN, an ongoing longitudinal multi- (time periods in which participants did not wear the site cohort study of women, which has been previously hip device, such as sleeping and water-based activities) described [23]. Briefly, 3302 women ages 42–52  years were defined using the Choi algorithm with the ‘Physi - (mean age ± standard deviation [SD]: 46.4 ± 2.7  years) calActivity’ R package [24]. Evenson vector magnitude were recruited from seven geographic sites across the (VM) cut point values [25] were used to classify minutes −1 U.S.: Boston, MA; Chicago, IL; Southeast area, Michi- as sedentary behavior (< 76 VMct·min ), low light (76 −1 gan; Los Angeles, CA; Newark, NJ; Pittsburgh, PA; and to < 903 VMct·min ) intensity (LLPA), high light (903 −1 Oakland, CA. Each site recruited White women and to < 2075 VMct·min ) intensity (HLPA),  and  moderate women of one other race/ethnicity. Cohort members to vigorous intensity physical activity (MVPA) (≥ 2075 −1 have been followed through 16 follow-up visits approxi- VMct·min ). The original 15-s thresholds were multi - mately every year. Data for these analyses were collected plied by four to account for the longer epoch (60-s), with at the SWAN follow-up visit 15 (2015–2017) (N = 2091 slight adjustments to obtain mutually exclusive threshold women), in which a subsample of women were invited ranges [26]. For the waking interval, days were classified to concurrently wear two devices to quantify waking and as adherent if they had ≥ 600  min of wear time. Partici- sleep behaviors. A total of 1285 women had valid data pants were included if they had ≥ 4 adherent days [18]. for MASH development and evaluation (Fig.  1). Ethics These days did not need to be consecutive. approval was obtained from Institutional Review Boards The sleep interval was quantified using the wrist-worn at each of the seven SWAN sites and all participants pro- Actiwatch 2 (Philips Respironics, Murrysville, PA) device vided written informed consent at each visit. worn for 24  h/day on the non-dominant wrist. Partici- pants completed a diary and were asked to press an event Fig. 1 Participant flow diagram for data harmonization Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 4 of 12 marker on the watch to indicate when they went to bed occurred for nights that had a Actiwatch malfunction, with the intention to sleep and when they rose from bed the Actiwatch was removed, or there was no/poor sleep. for the final time each day. The sleep diary included ques - In these cases, one-dimensional Convolutional Neural tions regarding when they got into bed, the time they Network (1D CNN) models [28] are used. 1D CNNs were tried to go to sleep, the time they woke up for the day, chosen because they have been previously employed on and the time when they rose from bed. The Actiwatch a variety of actigraphy data for algorithm detection [29– was set at 0.05  g for 3–11  Hz and data were sampled in 31]. The 1D CNN models read the epoch-level ActiGraph 60-s epochs. To determine total scored sleep time, the and Actiwatch data and assign each epoch with a prob- Actiwatch 2 data were processed, evaluated for quality, ability of being ‘within a wake interval’. and scored with the event marker, default sleep detection Two 1D CNN models were created for MASH. One algorithm (wake threshold: 40 ct/min) [27], and the sleep 1D CNN model accommodates both ActiGraph + Acti- diary (if available) in Actiware 5.0.9 using procedures watch data, using data from both devices, and a separate consistent with the Society of Behavioral Sleep Medicine 1D CNN model uses ActiGraph-only data, for situations guidelines [19]. Clock times for sleep onset (beginning when the Actiwatch data were invalid or missing. Once of sleep interval) and sleep offset (end of sleep inter - the 1D CNN models generate epoch-level predictions, val) were determined using the start of the first minute a simple optimization procedure is used to determine (onset) or the last minute (offset) of 10 consecutive min - which clusters of epochs were most likely to represent utes of immobility. In addition, all sleep records were vis- the true waking interval. See Additional file  1: Appendix ually inspected for quality. Sleep records were removed SA for a conceptual framework of the MASH process. if there was a Actiwatch malfunction, the Actiwatch was The 1D CNN models were trained using all days that removed prior to sleep (e.g., non-wear), there was no/ had valid sleep data surrounding them. This sample was poor sleep, or there was < 60  min of total scored sleep. randomly divided into training, test, and validation data- These records are henceforth referred to as ‘valid scored sets of mutually exclusive individuals. Each 1D CNN sleep data’. model created epoch-level predictions by evaluating cen- tered 101-epoch windows of time surrounding the epoch MASH development in question [30]. We considered the costs of misclassify- MASH utilizes data from three sources: (1) ActiGraph– ing each epoch as ‘within wake interval’ or ‘outside wake count data from Axes 1–3, (2) Actiwatch–lux (white interval’ as equal; therefore, the optimal cutoff probabil - light) and count data, and (3) sleep onset and sleep offset ity differentiating these statuses was determined using clock times–from the valid scored sleep data correspond- Youden’s J-statistic [32]. See Additional file  1: Appendix ing to the beginning (sleep onset) and end (sleep offset) SB and SC for a detailed description of the approaches of the primary sleep interval. used to join the sleep and waking datasets and model Harmonizing the sleep and activity data through building. MASH is a two-tiered approach that addresses two states of data availability. At its core, MASH reconciles the two Removing the hip device at night prior to sleep onset datasets (i.e., ActiGraph- and Actiwatch-derived) by While developing MASH, we noticed the predicted sleep determining the bounds of the ‘waking interval’ for each intervals resulting from the 1D CNNs were more likely to day. The creation of these intervals represents the coher - have shorter waking intervals (both pre-harmonization ent fusion of the two datasets: nig ht -day -night . and after MASH application) and longer sleeping inter- (t-1) (t) (t) Any ‘correction’ to the waking behaviors (sedentary vals compared to valid scored sleep-derived intervals behavior, LLPA, HLPA, MVPA) associated with the (e.g., Actiwatch dataset). While it could be that records imposition of these intervals resulted from a disagree- missing sleep data might have shorter waking intervals ment between the sleep and activity data (e.g., the Acti- because women who were less likely wear the Actiwatch watch says a person is asleep and the ActiGraph says they 2 wrist device might not be as diligent at wearing the are engaging in sedentary behavior). ActiGraph hip device (thus having shorter wake inter- The first tier of MASH applies to instances where the vals), having longer sleep intervals is problematic because 24-h period has valid scored sleep data preceding and the act of removing the hip device was being confused proceeding it. The wake interval is built using the pre - with sleep onset. vious day’s sleep onset time and the current day’s sleep For records that had valid scored sleep data, the aver- offset time (night -day -night ). The second tier is age difference between removing the hip device and sleep (t-1) (t) (t) used for all other instances where a wake interval does onset was 44.4  min. For all records that did not have not have valid scored sleep data (e.g., missing sleep onset valid scored sleep data, the average duration of the sleep or sleep offset) immediately surrounding it. This typically interval was 45.2 min longer (P < 0.001) than the intervals D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 5 of 12 where valid scored sleep data were present. Given the chi-square tests for categorical variables. Selected par- similar sizes of each effect, we therefore used records ticipant characteristics included self-reported age (years), with valid scored sleep data to construct a bivariate prob- race/ethnicity (Black, Chinese, Hispanic, Japanese, ability distribution for wake interval length by the size of White), education (< high school, high school, some col- the difference between removing the hip device and sleep lege, college, post-college), self-rated health (poor, fair, onset. The probability distribution was constructed using good, very good, excellent), difficulty walking one mile bounded 2-dimensional kernel density estimation with a (yes/no), and obesity (body mass index [BMI] ≥ 30  kg/ minimum value equivalent to what was in the data and m ) calculated using height and weight at visit 15. Model an imposed maximum value of 200  min for each varia- performance was examined using the C-statistic, i.e., ble (~ 3 SD from the mean of 44.4 min). Given the size of the area under the Receiver Operating Characteristic each wake interval, the probability distribution was used (AUC-ROC) curve. To account for slight data imbalance to generate an estimate of the amount of time that exists (roughly a 66/33 split between ‘within wake interval’ and between removing the hip device and sleep onset. This ‘outside of wake interval’), Precision-Recall curves were estimate was added to the predicted timestamp for sleep also used [33]. Paired t-tests were used to examine esti- onset (thus shortening the duration of the sleep interval). mates of time-use movement behaviors between days This process was replicated ten times for all records that were MASH-corrected and when the estimates were that did not have scored sleep data indicating sleep onset. not corrected. The MASH intervals were then constructed using the average of these ten samples. For more information on Results this process, please consult Additional file  1: Appendix 1D CNN construction and accuracy SD  and SE. MASH user documentation is available at The sample used to build the prediction models included https:// github. com/ jsw70/ MASH. 1112 older women who had both valid scored sleep data preceding and following each wake interval in ques- MASH evaluation tion. Participants were similarly distributed (P > 0.05) for The full sample included 10,123 days with useable accel - demographics and selected health characteristics across erometry data across 1285 participants. In order to eval- the training (n = 625), test (n = 278), and validation uate MASH, for this analysis we focused on days that had (n = 209) sets (Table 1). full sleep–wake compositions consisting of a sleeping The AUC-ROC for both 1D CNN models developed for interval followed by a waking interval. This was done to MASH (ActiGraph + Actiwatch or ActiGraph-only) were maximize the number of full compositions we could eval- considered excellent (Fig.  2) with values of 0.991 and uate. For example, while we could have chosen to view a 0.983 for the ActiGraph + Actiwatch and ActiGraph-only composition as being a day proceeded by night (wake- models, respectively. In addition, the accompanying Pre- sleep) this would have led to fewer compositions for cision-Recall AUC were 0.993 and 0.989. Using Youden’s evaluation as the last day of data collection would likely J-statistic to determine a cutoff probability threshold be excluded (no sleep data). However, of note, the MASH (0.698 and 0.729), the sensitivity of the models at each process creates a bi-directional dataset that allows for optimal point was 95.7% for the ActiGraph + Actiwatch flexibility in examining both sleep–wake or wake-sleep model and 92.8% for the ActiGraph-only model. The compositions. specificity of the 1D CNN models was 95.5% and 95.6%, To evaluate the harmonization process, three exclu- respectively. sions were applied to the sleep–wake data: (1) the first day of data collection for the waking interval was removed because it was a partial day with the first Data harmonization instance of detected wear corresponding to when the Of the 8560 sleep–wake compositions, 84.9% (n = 7270 devices were distributed and placed on the participant records) had valid scored sleep data (i.e., had both sleep during the in-person exam visit (n = 1285  days), (2) any onset and sleep offset). Of the remaining 15.1% of days observation that did not have sleep data preceding the (n = 1290 records), either of the 1D CNN models was wake interval (n = 269), and (3) any instances where the applied to estimate (1) sleep offset (n = 32 days), (2) sleep sleep data was less than 60  min (n = 9). The analytical onset (n = 503  days), or (3) both sleep offset and onset dataset for the evaluation of MASH included 8560 sleep– (n = 755 days) (Table 2). wake compositions (Fig. 1). With the sleep and wake intervals defined, 46.0% (3934 Potential differences in participant characteris - of 8560) of days needed correction to the waking inter- tics between the training, test, and validation datasets val. This was to address improper classification of at least were assessed using t-tests for continuous variables or one minute-level epoch as both sleep and wake (82.9%; Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 6 of 12 Table 1 One‑ dimensional Convolutional Neural Network (1D CNN) SWAN follow‑up visit 15 (2015–2017) participant characteristics, overall and by dataset Characteristic Sample Training set Test set Validation set (N = 1112) (n = 625) (n = 278) (n = 209) % (n) % (n) % (n) % (n) Age (M ± SD) 65.5 ± 2 65.4 ± 2 65.2 ± 2 65.8 ± 2 Race/ethnicity Black 25.8 (287) 27.8 (174) 23.7 (66) 22.5 (47) Chinese 12.9 (143) 13.0 (81) 15.1 (42) 9.6 (20) Hispanic 3.0 (33) 3.0 (19) 2.5 (7) 3.3 (7) Japanese 12.1 (134) 11.2 (70) 13.3 (37) 12.9 (27) White 46.3 (515) 45.0 (281) 45.3 (126) 51.7 (108) Education < High school 4.0 (45) 5.0 (31) 2.5 (7) 3.3 (7) High school 14.9 (166) 13.9 (87) 16.2 (45) 16.3 (34) Some college 31.3 (348) 28.8 (180) 33.1 (92) 36.4 (76) College 22.8 (253) 23.5 (147) 21.6 (60) 22.0 (46) Post‑ college 26.3 (292) 28.5 (178) 25.2 (70) 21.1 (44) Missing 0.7 (8) 0.3 (2) 1.4 (4) 1.0 (2) Obesity (BMI ≥ 30 kg/m ) 36.1 (401) 38.2 (239) 32.7 (91) 34.0 (71) Missing 1.0 (11) 1.0 (6) 1.1 (3) 1.0 (2) Self‑rated health Poor, fair, or good 47.3 (526) 47.8 (299) 49.3 (137) 43.1 (90) Missing 0.8 (9) 1.1 (7) 0.7 (2) Difficulty walking one mile 36.5 (406) 39.4 (246) 34.2 (95) 31.1 (65) BMI body mass index There were no statistically significant differences between the training, test, and validation datasets at the P = 0.05 level using t-tests for continuous variables or chi- square tests for categorical variables Fig. 2 Receiver Operating Characteristic (ROC) curves with cutoff thresholds and sensitivity and specificity values D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 7 of 12 Table 2 Number of days in the sample, MASH model used, and sleep–wake interval size Reason Number of days MASH model applied Sleep–wake (N = 8560) interval size, hours Mean (SD) No sleep data are missing 7270 Valid scored sleep 24.04 (1.41) Sleep data are missing 1290 Only sleep onset data are missing 503 1D CNN 23.20 (2.35) Sleep offset data are missing on any day besides first day 32 1D CNN 23.92 (2.52) Both sleep onset and sleep offset data are missing 755 1D CNN 23.88 (1.68) 1D CNN one-dimensional convolutional neural network; MASH method for activity sleep harmonization 3262 of 3934  days) or due to missing sleep data (672 of (median [IQR] = −  76.0 [−  150.0, −  14.0] min of total 3934 days). wake time) for prediction, even though the distribution For days requiring correction, the average correc- of the wake intervals requiring 1D CNN correction were tion applied included a median (interquartile range relatively more skewed. [IQR]) of −  27.0 (−  115.0, 8.0) minutes of total wake Table  3 presents the time-use behavior estimates pre- time. The distribution of the MASH-corrected wake harmonization (e.g., prior to MASH implementation) intervals smoothed out a cluster of days that the Choi and the estimates once harmonized using MASH. When algorithm (applied to ActiGraph data) classified as hav - compared to other waking behavior estimates (i.e., LLPA, ing > 1200 min (20 h) of waking wear (Fig. 3). This finding HLPA, and MVPA) the distribution of the sedentary is consistent regardless of whether the wake interval was behavior estimate was most influenced once the wear corrected using scored sleep data (median [IQR] = − 21.0 intervals were corrected (Fig.  4). Specifically, sedentary [−  98.0, 12.0] min of total wake time) or 1D CNN behavior was corrected a median (IQR) of − 18.0 (− 93.0, Fig. 3 Distribution of wake interval sizes between the uncorrected days and the corrected days, overall and by correction method Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 8 of 12 Table 3 Pre‑harmonization and post MASH harmonization time ‑use estimates (N = 8560) days Time-use behavior Pre-harmonization, MASH harmonization, minutes minutes Median (IQR) Median (IQR) Sedentary behavior 438 (344, 547) 419 (329, 512) Low light intensity physical activity (LLPA) 277 (219, 337) 274 (216, 333) High light intensity physical activity (HLPA) 137 (100, 181) 136 (99, 181) Moderate to vigorous intensity physical activity (MVPA) 46 (24, 78) 45 (24, 78) Sleep 440 (378, 497) 444 (383, 502) MASH method for activity sleep harmonization a −1 −1 Evenson vector magnitude (VM) cut point values were used to classify minutes as sedentary behavior (< 76 VMct·min ), low light (76 to < 903 VMct·min ) intensity −1 −1 (LLPA), high light (903 to < 2075 VMct·min ) intensity (HLPA), and MVPA (≥ 2075 VMct·min ) Pre-harmonization included 8024 days due to missing data Fig. 4 Distribution of waking activity lengths between the uncorrected and corrected activity variables 2.0) min, whereas the other waking behavior types missing), it was not possible to perform t-test analysis on (LLPA + HLPA + MVPA) were corrected a median (IQR) sleep because a substantial portion of the data did not of −  1.0 (−  4.0, 1.0) min. Paired t-test analysis results have an ‘uncorrected’ sleep measurement. demonstrate MASH resulted in statistically significant Participants had a mean (SD) of 6.7 (1.5) sleep–wake reductions in all forms of activity; however, only seden- compositions. The final MASH harmonized dataset tary behavior [t(8559) = −  34.2, P < 0.001] had a mean resulted in a mean (SD) sleep–wake composition interval difference greater than 3 min. Because the wake interval size of 23.97 (1.52) hours. The interval sizes of the sleep– correction process within MASH also simultaneously wake compositions were similar across the MASH model creates sleep intervals (in cases where the sleep data are applied (Table 2). D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 9 of 12 Discussion disorders (e.g., insomnia, sleep apnea), and excess health- Measuring time-use movement behaviors accurately care costs [8, 49, 50]. across the continuous 24-h period is critical as these Choosing between a single device protocol or a pro- behaviors are interrelated and evidence suggests that tocol in which participants wear two or more devices the combined effects of these behaviors on health may concurrently, as in the current study, is largely up to be greater than their individual effects [34–36]. This researcher preferences. However, multiple-device proto- had led to 24-h public health guidelines released by the cols may be necessary for independently funded ancillary World Health Organization [37] and by some countries studies to large prospective cohorts to reduce participant (e.g., Australia [38], Canada [39, 40], New Zealand [41], burden of wearing devices across many weeks of data South Africa [42]). Thus, accurate estimation of 24-h collection. Placing multiple devices at different anatomi - sleep–wake cycles, including the contributing behaviors, cal locations to increase precision of time-use behaviors is paramount. We developed the Method for Activity has been previously implemented on the hip + thigh Sleep Harmonization (MASH) to harmonize time-series [51] and chest + thigh [52]. The MASH method provides data from two accelerometers that use two different an integral step in 24-h time-use assessment by joining placements (wrist and hip) to estimate behaviors com- accelerometry data from the hip + wrist for increased prising the 24-h period. This method creates night-day- measurement precision. Prior to this method develop- night pairings rather than constraining data to a fixed ment, researchers would need to weigh the decision of time period (e.g., midnight to midnight). We analyzed an which behavior outcome could have poor performance interval of sleep followed by a subsequent waking inter- accuracy within their study or have participants switch val as a 24-h sleep–wake composition. This accounts between wear locations, with the possibility of increas- for the compositional nature of behaviors used in time- ing non-compliance [21]. Further, providing open-source use epidemiology [1–3]. Developed on a large sample of code may help with data harmonization and protocol older women, the findings suggest the MASH approach development across studies. (1) minimized data loss due to missing sleep data and In addition to increasing 24-h measurement precision, (2) improve precision of 24-h sleep–wake compositions. MASH also minimizes data loss. MASH evaluates the Together, these findings support the utility of MASH to epoch-level data and applies a classification algorithm harmonize sleep–wake data obtained from two devices that scores the epoch as either within a waking interval and correct these data, as needed, to more precisely esti- or within a sleeping interval. Therefore, although diary/ mate time-use movement behaviors for further analysis. sleep data may be missing, night-day-night pairings could Physical activity and sleep have largely been separate still be constructed. In our sample, a total of 15.1% of disciplines, with each preferring certain devices and days (n = 1290 records) were missing either sleep onset, anatomical placement for field-based data collection. sleep offset, or both times from the sleep dataset. With - For measuring waking activities, specifically time spent out this classification algorithm, those days would be lost within intensity categories, triaxial accelerometer place- during analysis or would need to be imputed. Further, ment is most accurate at the hip [43, 44]. However, for MASH can be used to create daily 24-h compositions sleep detection, reliable accelerometry measurement rather than a single averaged daily estimate, which can occurs on the wrist [19, 45, 46]. Findings from Full and have important implications for examining time-use pat- colleagues suggest estimates of sleep duration using an terns of behaviors across the week and development of ActiGraph worn on the hip were significantly higher from future intervention studies targeting these behaviors. polysomnography (PSG), overestimating total sleep time The limitations of MASH should be noted. This method by 37.8 (SD = 61.3) min [47]. This could be the reason the was developed in a sample of community-dwelling older majority of corrections attributed to MASH were to fix adult women (age range: 60–72  years). However, we do instances where sleep was coded as sedentary behavior. not believe this would change model development, and Further, total volume of physical activity measured by given the flexibility and utility of 1D CNN models, imple - wrist-worn devices (e.g., Actiwatch 2) have a weak cor- mentation of MASH in other studies and populations is relation (r = 0.26) with hip-worn devices and thus are achievable. In addition, MASH only removes waking data not favorable for measuring physical activity [48]. Over- when the wake interval sizes are longer than the MASH estimation of sedentary behavior and underestimation of prediction, which occurs in instances where the partici- sleep can have detrimental effects to outcomes research. pant was likely wearing the ActiGraph monitor on the Sleep, in the absence of disturbances or disorders, is hip while sleeping. MASH is unable to determine activ- thought to be a restorative and health-promoting pro- ity behavior when data are missing due to not wearing cess for the body [7], whereas excessive sedentary behav- the device during the waking interval, for example, if a ior is associated with several diseases, including sleep participant woke up and did not immediately put on the Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 10 of 12 SWAN Study of Women’s Health Across the Nation waking device. We did not calculate these data from the SD Standard deviation wrist-worn device as the Actiwatch 2 is not accurate for VM Vector magnitude measuring physical activity [48] and we did not impute LLPA Low light intensity physical activity HLPA High light intensity physical activity these minutes as we were unable to determine the true 1D CNN One ‑ dimensional Convolutional Neural Network waking activity behavior. However, statistical techniques BMI Body mass index such as compositional data analysis (CoDA), which treat AUC‑ROC Area under the Receiver Operating Characteristic IQR Interquartile range daily time-use movement behaviors as a composition that are translated to real space through the applica- Supplementary Information tion of coordinate systems and constrained to 1440  min The online version contains supplementary material available at https:// doi. [53], overcome non-wear issues. With MASH, the period org/ 10. 1186/ s44167‑ 023‑ 00017‑5. within the resting and sleep intervals (e.g., sleep onset latency) is classified as sedentary behavior, which is sup - Additional file 1: SA. Method of Activity Sleep Harmonization (MASH) ported by the Sedentary Behavior Research Network conceptual framework. SB. Determining valid scored sleep data validity. SC. Building the 1D CNN models. Figure S1. The network structure for (SBRN)’s definition as ‘any waking behavior character - both 1D CNN models with and without the Actiwatch data. Table S1. ized by an energy expenditure ≤ 1.5 metabolic equiva- Hyperparameters for both One‑ dimensional Convolutional Neural Net‑ lents, while in a sitting, reclining, or lying posture’ [54]. work (1D CNN) models. Figure S2. Precision‑Recall curves. SD. Account‑ ing for the 1D CNN tendency to confuse hip device removal as sleep However, long sleep latency may have differential effects onset. Figure S3. Bivariate probability distribution for the amount of time on health than sedentary behavior [55] and the health- found between hip device removal and sleep onset for all valid scored related importance of distinguishing this period requires sleep data. Figure S4. Comparing the distribution of sleep intervals created by the 1D CNN’s and the scored sleep data. The two graphs are further study [56]. Lastly, MASH only classifies the main separated by whether or not the bivariate sampling was used to adjust (overnight) sleep interval and did not attempt to classify for confusing hip device removal with sleep onset. SE. Determining wake daytime sleeping (napping) from either device. Despite intervals from epoch‑level 1D CNN predictions. the limitations, we built MASH and the 1D CNN predic- tion models using a large sample of women (N = 1285) Acknowledgements who wore two devices (hip + wrist) concurrently. The We thank the study staff at each site and all the women who participated in SWAN. models had excellent classification and there were no significant differences between the scored sleep and 1D Author contributions CNN prediction models. Using the 1D CNN can help ED and JW contributed equally to this work. ED conceptualized the study, supported the formal analysis, investigation, and methodology and was the minimize data loss for days when participants forget to lead in writing the manuscript. JW was the lead in the analysis, investigation, wear the wrist device or there was a device malfunction. methodology, and visualization of the figures, and supported in preparing the Methods and Results sections of the text. AC and KPG conceptualized the study, supported the formal analysis, investigation, and methodology, reviewed the manuscript, and provided supervision for the study. CK, SB, KD, Conclusions CKG, BS, ST, and MH provided a critical role in reviewing and editing the text. MASH is a dataset harmonization method for merging HK reviewed and edited the text and was a primary investigator during fund‑ sleep and waking activity behaviors measured concur- ing acquisition. All authors have read and approved the final manuscript. rently from multiple devices (hip + wrist). The devices Funding were chosen because of their accuracy in measuring wak- The Study of Women’s Health Across the Nation (SWAN) has grant support ing activity behaviors (ActiGraph GT3X+) and sleep from the National Institutes of Health (NIH), DHHS, through the National Institute on Aging (NIA), the National Institute of Nursing Research (NINR) and (Actiwatch 2). We built MASH to merge separate, inde- the NIH Office of Research on Women’s Health (ORWH) (Grants U01NR004061; pendent datasets, minimize data loss for missing sleep U01AG012505, U01AG012535, U01AG012531, U01AG012539, U01AG012546, data, and with the flexibility that this process can be rep - U01AG012553, U01AG012554, U01AG012495, and U19AG063720). The content of this article is solely the responsibility of the authors and does not licated in other studies that simultaneously collect sleep necessarily represent the official views of the NIA, NINR, ORWH or the NIH. and waking behaviors using two devices. Researchers can Clinical Centers: University of Michigan, Ann Arbor–Carrie Karvonen‑ use the MASH approach to correct sleep–wake harmo- Gutierrez, PI 2021–present, Siobán Harlow, PI 2011–2021, MaryFran Sowers, PI 1994–2011; Massachusetts General Hospital, Boston, MA–Sherri‐Ann nization, construct daily-level compositions, and aggre- Burnett‐Bowie, PI 2020–Present; Joel Finkelstein, PI 1999–2020; Robert Neer, gate to averaged daily values as needed. Ultimately, this PI 1994–1999; Rush University, Rush University Medical Center, Chicago, IL – approach increases precision of the physical activity and Imke Janssen, PI 2020–Present; Howard Kravitz, PI 2009–2020; Lynda Powell, PI 1994–2009; University of California, Davis/Kaiser–Elaine Waetjen and Monique sleep estimates which may improve the accuracy of the Hedderson, PIs 2020–Present; Ellen Gold, PI 1994–2020; University of California, observed measures of association with health outcomes. Los Angeles‑Arun Karlamangla, PI 2020–Present; Gail Greendale, PI 1994–2020; Albert Einstein College of Medicine, Bronx, NY–Carol Derby, PI 2011–present, Rachel Wildman, PI 2010–2011; Nanette Santoro, PI 2004–2010; University of Abbreviations Medicine and Dentistry‑New Jersey Medical School, Newark–Gerson Weiss, PI LIPA Light intensity physical activity 1994–2004; and the University of Pittsburgh, Pittsburgh, PA—Rebecca Thurs‑ MVPA Moderate to vigorous intensity physical activity ton, PI 2020–Present; Karen Matthews, PI 1994–2020. MASH Method for Activity Sleep Harmonization D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 11 of 12 NIH Program Office: National Institute on Aging, Bethesda, MD—Rosaly 5. World Health Organization. Global action plan on physical activity Correa‑ de‑Araujo 2020–present; Chhanda Dutta 2016–present; Winifred Rossi 2018–2030: more active people for a healthier world. Geneva: World 2012–2016; Sherry Sherman 1994–2012; Marcia Ory 1994–2001; National Health Organization; 2018. Institute of Nursing Research, Bethesda, MD—Program Officers. 6. World Health Organization. WHO guidelines on physical activity and Central Laboratory: University of Michigan, Ann Arbor–Daniel McConnell sedentary behaviour. Geneva: World Health Organization; 2020. (Licence: (Central Ligand Assay Satellite Services). CC BY-NC-SA 3.0 IGO). Coordinating Center: University of Pittsburgh, Pittsburgh, PA–Maria Mori 7. Consensus Conference Panel. Recommended amount of sleep for a Brooks, PI 2012–present; Kim Sutton‑ Tyrrell, PI 2001–2012; New England healthy adult: a joint consensus statement of the American academy Research Institutes, Watertown, MA–Sonja McKinlay, PI 1995–2001. of sleep medicine and sleep research society. Sleep. 2015;38(6):843–4. Steering Committee: Susan Johnson, Current Chair; Chris Gallagher, Former https:// doi. org/ 10. 5665/ sleep. 4716. Chair. 8. Patterson R, McNamara E, Tainio M, et al. Sedentary behaviour and risk of all‑ cause, cardiovascular and cancer mortality, and incident type 2 Availability of data and materials diabetes: a systematic review and dose response meta‑analysis. Eur J Epi‑ SWAN provides access to public use datasets that include data from SWAN demiol. 2018;33(9):811–29. https:// doi. org/ 10. 1007/ s10654‑ 018‑ 0380‑1. screening, the baseline visit and follow‑up visits (https:// aging resea rchbi 9. Troiano RP, Stamatakis E, Bull FC. How can global physical activity surveil‑ obank. nia. nih. gov/). To preserve participant confidentiality, some, but not all, lance adapt to evolving physical activity guidelines? Needs, challenges of the data used for this manuscript are contained in the public use datasets. and future directions. Br J Sports Med. 2020;54(24):1468–73. https:// doi. A link to the public use datasets is also located on the SWAN web site: http:// org/ 10. 1136/ bjspo rts‑ 2020‑ 102621. www. swans tudy. org/ swan‑ resea rch/ data‑ access/. Investigators who require 10. Omura JD, Whitfield GP, Chen TJ, et al. Surveillance of physical activity and assistance accessing the public use dataset may contact the SWAN Coordinat‑ sedentary behavior among youth and adults in the United States: history ing Center at the following email address: swanaccess@edc.pitt.edu. and opportunities. J Phys Act Health. 2021;18(S1):S6–24. https:// doi. org/ 10. 1123/ jpah. 2021‑ 0179. 11. Strain T, Wijndaele K, Pearce M, Brage S. Considerations for the use of Declarations consumer‑ grade wearables and smartphones in population surveillance of physical activity. J Meas Phys Behav. 2022. https:// doi. org/ 10. 1123/ Ethics approval and consent to participate jmpb. 2021‑ 0046. Ethics approval was obtained from Institutional Review Boards at each of the 12. Hyde ET, Omura JD, Fulton JE, et al. Physical activity surveillance using seven SWAN sites and all participants provided written informed consent at wearable activity monitors: are US adults willing to share their data? Am each visit. J Health Promot. 2020;34(6):672–6. https:// doi. org/ 10. 1177/ 08901 17119 Consent for publication 13. Fuller D, Colwell E, Low J, et al. Reliability and validity of commercially Not applicable. available wearable devices for measuring steps, energy expenditure, and heart rate: Systematic review. JMIR Mhealth Uhealth. 2020;8(9): e18694. Competing interests https:// doi. org/ 10. 2196/ 18694. The authors declare that they have no competing interests. 14. Straiton N, Alharbi M, Bauman A, et al. The validity and reliability of consumer‑ grade activity trackers in older, community‑ dwelling adults: a Author details systematic review. Maturitas. 2018;112:85–93. https:// doi. org/ 10. 1016/j. Department of Epidemiology, The University of Alabama at Birmingham, matur itas. 2018. 03. 016. Birmingham, AL, USA. Epidemiology Data Center, The University of Pittsburgh, 15. Scott H, Lack L, Lovato N. A systematic review of the accuracy of sleep Pittsburgh, PA, USA. Department of Epidemiology, The University of Pitts‑ wearable devices for estimating sleep onset. Sleep Med Rev. 2020;49: burgh, Pittsburgh, PA, USA. Department of Health and Human Development, 101227. https:// doi. org/ 10. 1016/j. smrv. 2019. 101227. The University of Pittsburgh, Pittsburgh, PA, USA. Division of Research, Kaiser 16. O’Driscoll R, Turicchi J, Beaulieu K, et al. How well do activity monitors Permanente Northern California, Oakland, CA, USA. Center for Behavioral estimate energy expenditure? A systematic review and meta‑analysis of Cardiovascular Health, Columbia University Medical Center, New York, NY, USA. the validity of current technologies. Br J Sports Med. 2020;54(6):332–40. Department of Epidemiology, University of Michigan, Ann Arbor, MI, USA. 17. Evenson KR, Goto MM, Furberg RD. Systematic review of the validity and Department of Psychiatry and Behavioral Sciences and Department of Pre‑ reliability of consumer‑ wearable activity trackers. Int J Behav Nutr Phys ventive Medicine, Rush University Medical Center, Chicago, IL, USA. Depar t‑ Act. 2015;12(1):159. https:// doi. org/ 10. 1186/ s12966‑ 015‑ 0314‑1. ment of Psychiatry and Behavioral Neurobiology, The University of Alabama 18. Migueles JH, Cadenas‑Sanchez C, Ekelund U, et al. Accelerometer data at Birmingham, Birmingham, AL, USA. Department of Psychiatry, School collection and processing criteria to assess physical activity and other of Medicine, The University of Pittsburgh, Pittsburgh, PA, USA. outcomes: a systematic review and practical considerations. Sports Med. 2017;47(9):1821–45. https:// doi. org/ 10. 1007/ s40279‑ 017‑ 0716‑0. Received: 21 November 2022 Accepted: 2 February 2023 19. Ancoli‑Israel S, Martin JL, Blackwell T, et al. The SBSM guide to actigra‑ phy monitoring: clinical and research applications. Behav Sleep Med. 2015;13(Suppl 1):S4–38. https:// doi. org/ 10. 1080/ 15402 002. 2015. 10463 56. 20. Liu F, Wanigatunga AA, Schrack JA. Assessment of physical activity in adults using wrist accelerometers. Epidemiol Rev. 2021;43(1):65–93. References https:// doi. org/ 10. 1093/ epirev/ mxab0 04. 1. Rosenberger ME, Fulton JE, Buman MP, et al. The 24‑hour activity 21. De Craemer M, Verbestel V. Comparison of outcomes derived from the cycle: a new paradigm for physical activity. Med Sci Sports Exerc. actigraph gt3x+ and the axivity ax3 accelerometer to objectively meas‑ 2019;51(3):454–64. https:// doi. org/ 10. 1249/ MSS. 00000 00000 001811. ure 24‑hour movement behaviors in adults: a cross‑sectional study. Int J 2. Pedišić Ž, Dumuid D, Olds TS. Integrating sleep, sedentary behav‑ Environ Res Public Health. 2022;19(1):271. https:// doi. org/ 10. 3390/ ijerp iour, and physical activity research in the emerging field of time ‑use h1901 0271. epidemiology: definitions, concepts, statistical methods, theoretical 22. National Center for Health Statistics (NCHS). National health and nutrition framework, and future directions. Kinesiology. 2017;49(2):252–69. examination survey: 2011–2012 data documentation, codebook, and 3. Falck RS, Davis JC, Khan KM, et al. A Wrinkle in measuring time use for frequencies physical activity monitor—day (PAXDAY_G). Hyattsville: Cent‑ cognitive health: how should we measure physical activity, sedentary ers for Disease Control and Prevention; 2020. behaviour and sleep? Am J Lifestyle Med. 2021. https:// doi. org/ 10. 23. Sowers MF, Crawford SL, Sternfeld B, et al. SWAN: a multicenter, multi‑ 1177/ 15598 27621 10314 95. ethnic, community‑based cohort study of women and the menopausal 4. U.S. Department of Health and Human Services. Physical activity transition. In: Lobo R, Marcus R, Kelsey J, editors., et al., Menopause: biol‑ guidelines for Americans. 2nd ed. Washington: U.S. Department of ogy and pathobiology. San Diego: Academic Press; 2000. p. 175–88. Health and Human Services; 2018. Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 12 of 12 24. Choi L, Beck C, Liu Z, et al. Physical activity: process accelerometer data 44. Rosenberger ME, Haskell WL, Albinali F, et al. Estimating activity and for physical activity measurement. R package version 0.2–4. 2021. https:// sedentary behavior from an accelerometer on the hip or wrist. Med Sci CRAN.R‑ proje ct. org/ packa ge= Physi calAc tivity. Sports Exerc. 2013;45(5):964–75. https:// doi. org/ 10. 1249/ MSS. 0b013 25. Evenson KR, Wen F, Herring AH, et al. Calibrating physical activity intensity e3182 7f0d9c. for hip‑ worn accelerometry in women age 60 to 91 years: the women’s 45. Littner M, Kushida CA, Anderson WM, et al. Practice parameters for the health initiative OPACH calibration study. Prev Med Rep. 2015;2:750–6. role of actigraphy in the study of sleep and circadian rhythms: an update https:// doi. org/ 10. 1016/j. pmedr. 2015. 08. 021. for 2002. Sleep. 2003;26(3):337–41. https:// doi. org/ 10. 1093/ sleep/ 26.3. 26. Stewart A, Sternfeld B, Lange‑Maia BS, et al. Reported and device ‑based 337. physical activity by race/ethnic groups in young‑ old women. J Meas Phys 46. Lehrer HM, Yao Z, Krafty RT, et al. Comparing polysomnography, actig‑ Behav. 2020;3(2):118–27. https:// doi. org/ 10. 1123/ jmpb. 2019‑ 0062. raphy, and sleep diary in the home environment: the study of women’s 27. Kushida CA, Chang A, Gadkary C, et al. Comparison of actigraphic, health across the nation (SWAN) sleep study. Sleep Adv. 2022. https:// doi. polysomnographic, and subjective assessment of sleep parameters in org/ 10. 1093/ sleep advan ces/ zpac0 01. sleep‑ disordered patients. Sleep Med. 2001;2(5):389–96. https:// doi. org/ 47. Full KM, Kerr J, Grandner MA, et al. Validation of a physical activity accel‑ 10. 1016/ s1389‑ 9457(00) 00098‑8. erometer device worn on the hip and wrist against polysomnography. 28. LeCun Y, Bengio Y. Convolutional networks for images, speech, and time Sleep Health. 2018;4(2):209–16. https:// doi. org/ 10. 1016/j. sleh. 2017. 12. series. In: Arbib MA, editor. The handbook of brain theory and neural 007. networks. Cambridge: The MIT Press; 1995. p. 255–8. 48. Lambiase MJ, Gabriel KP, Chang YF, et al. Utility of actiwatch sleep monitor 29. Granovsky L, Shalev G, Yacovzada N, et al. Actigraphy‑based sleep/wake to assess waking movement behavior in older women. Med Sci Sports pattern detection using convolutionalneural networks. arXiv preprint Exerc. 2014;46(12):2301–7. https:// doi. org/ 10. 1249/ mss. 00000 00000 arXiv: 1802. 07945. 2018 Feb 22.Access date: July 19, 2021. 000361. 30. Palotti J, Mall R, Aupetit M, et al. Benchmark on a large cohort for sleep‑ 49. Nguyen P, Le LK‑D, Ananthapavan J, et al. Economics of sedentary behav‑ wake classification with machine learning techniques. NPJ Digit Med. iour: A systematic review of cost of illness, cost‑ effectiveness, and return 2019;2(1):50. https:// doi. org/ 10. 1038/ s41746‑ 019‑ 0126‑9. on investment studies. Prev Med. 2022;156: 106964. https:// doi. org/ 10. 31. Greenwood‑Hickman MA, Nakandala S, Jankowska MM, et al. The CNN 1016/j. ypmed. 2022. 106964. hip accelerometer posture (CHAP) method for classifying sitting pat‑ 50. Yang Y, Shin JC, Li D, et al. Sedentary behavior and sleep problems: a terns from hip accelerometers: a validation study. Med Sci Sports Exerc. systematic review and meta‑analysis. Int J Behav Med. 2017;24(4):481–92. 2021;53(11):2445–54. https:// doi. org/ 10. 1249/ mss. 00000 00000 002705.https:// doi. org/ 10. 1007/ s12529‑ 016‑ 9609‑0. 32. Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3(1):32–5. 51. Micklesfield LK, Westgate K, Smith A, et al. Physical activity behaviors of 33. Davis J, Goadrich M. The relationship between Precision‑Recall and ROC a middle‑age south african cohort as determined by integrated hip and curves. In: Proceedings of the 23rd international conference on machine thigh accelerometry. Med Sci Sports Exerc. 2022;54(9):1493–505. https:// learning; 2006.doi. org/ 10. 1249/ MSS. 00000 00000 002940. 34. Janssen I, Clarke AE, Carson V, et al. A systematic review of compositional 52. Bassett DR, John D, Conger SA, et al. Detection of lying down, sitting, data analysis studies examining associations between sleep, sedentary standing, and stepping using two activPAL monitors. Med Sci Sports behaviour, and physical activity with health outcomes in adults. Appl Exerc. 2014;46(10):2025–9. https:// doi. org/ 10. 1249/ MSS. 00000 00000 Physiol Nutr Metab. 2020;45(10):S248–57. https:// doi. org/ 10. 1139/ 000326. apnm‑ 2020‑ 0160. 53. Dumuid D, Pedišić Ž, Palarea‑Albaladejo J, et al. Compositional data analy‑ 35. Dumuid D, Stanford TE, Martin‑Fernández J‑A, et al. Compositional data sis in time‑use epidemiology: what, why, how. Int J Environ Res Public analysis for physical activity, sedentary time and sleep research. Stat Health. 2020;17(7):2220. https:// doi. org/ 10. 3390/ ijerp h1707 2220. Methods Med Res. 2018;27(12):3726–38. https:// doi. org/ 10. 1177/ 09622 54. Tremblay MS, Aubert S, Barnes JD, et al. Sedentary behavior research 80217 710835. network (SBRN) ‑ terminology consensus project process and out ‑ 36. Chastin SF, Palarea‑Albaladejo J, Dontje ML, et al. Combined effects of come. Int J Behav Nutr Phys Act. 2017;14(1):75. https:// doi. org/ 10. 1186/ time spent in physical activity, sedentary behaviors and sleep on obesity s12966‑ 017‑ 0525‑8. and cardio‑metabolic health markers: a novel compositional data analysis 55. Bubu OM, Brannick M, Mortimer J, et al. Sleep, cognitive impairment, and approach. PLoS ONE. 2015;10(10): e0139984. https:// doi. org/ 10. 1371/ alzheimer’s disease: a systematic review and meta‑analysis. Sleep. 2016. journ al. pone. 01399 84.https:// doi. org/ 10. 1093/ sleep/ zsw032. 37. World Health Organization. Guidelines on physical activity, sedentary 56. Barone Gibbs B, Kline CE. When does sedentary behavior become sleep? behaviour and sleep for children under 5 years of age. Geneva: World A proposed framework for classifying activity during sleep‑ wake transi‑ Health Organization; 2019. tions. Int J Behav Nutr Phys Act. 2018;15(1):81. https:// doi. org/ 10. 1186/ 38. Okely AD, Ghersi D, Hesketh KD, et al. A collaborative approach to adopt‑s12966‑ 018‑ 0712‑2. ing/adapting guidelines—the Australian 24‑hour movement guidelines for the early years (Birth to 5 years): an integration of physical activity, Publisher’s Note sedentary behavior, and sleep. BMC Public Health. 2017;17(Suppl 5):869. Springer Nature remains neutral with regard to jurisdictional claims in pub‑ https:// doi. org/ 10. 1186/ s12889‑ 017‑ 4867‑6. lished maps and institutional affiliations. 39. Ross R, Chaput JP, Giangregorio LM, et al. Canadian 24‑hour movement guidelines for adults aged 18–64 years and adults aged 65 years or older: an integration of physical activity, sedentary behaviour, and sleep. Appl Physiol Nutr Metab. 2020;45(10):S57‑s102. https:// doi. org/ 10. 1139/ Re Read ady y to to submit y submit your our re researc search h ? Choose BMC and benefit fr ? Choose BMC and benefit from om: : apnm‑ 2020‑ 0467. 40. Tremblay MS, Carson V, Chaput JP, et al. Canadian 24‑hour movement fast, convenient online submission guidelines for children and youth: an integration of physical activity, sed‑ thorough peer review by experienced researchers in your field entary behaviour, and sleep. Appl Physiol Nutr Metab. 2016;41(6 Suppl 3):S311–27. https:// doi. org/ 10. 1139/ apnm‑ 2016‑ 0151. rapid publication on acceptance 41. Ministry of Health. Sit less, move more, sleep well: active play guidelines support for research data, including large and complex data types for under‑fives. Wellington, New Zealand. 2017. • gold Open Access which fosters wider collaboration and increased citations 42. Draper CE, Tomaz SA, Biersteker L, et al. The South African 24‑hour move ‑ ment guidelines for birth to 5 years: an integration of physical activity, sit‑ maximum visibility for your research: over 100M website views per year ting behavior, screen time, and sleep. J Phys Act Health. 2020;17(1):109– 19. https:// doi. org/ 10. 1123/ jpah. 2019‑ 0187. At BMC, research is always in progress. 43. Cleland I, Kikhia B, Nugent C, et al. Optimal placement of accelerometers Learn more biomedcentral.com/submissions for the detection of everyday activities. Sensors. 2013;13(7):9183–200. https:// doi. org/ 10. 3390/ s1307 09183. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Activity Sedentary and Sleep Behaviors Springer Journals

Method for Activity Sleep Harmonization (MASH): a novel method for harmonizing data from two wearable devices to estimate 24-h sleep–wake cycles

Download PDF
12 pages

Loading next page...
 
/lp/springer-journals/method-for-activity-sleep-harmonization-mash-a-novel-method-for-mrRPI8Cl4l

References (63)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2023
eISSN
2731-4391
DOI
10.1186/s44167-023-00017-5
Publisher site
See Article on Publisher Site

Abstract

Background Daily 24‑h sleep–wake cycles have important implications for health, however researcher preferences in choice and location of wearable devices for behavior measurement can make 24‑h cycles difficult to estimate. Further, missing data due to device malfunction, improper initialization, and/or the participant forgetting to wear one or both devices can complicate construction of daily behavioral compositions. The Method for Activity Sleep Harmonization (MASH) is a process that harmonizes data from two different devices using data from women who concurrently wore hip (waking) and wrist (sleep) devices for ≥ 4 days. Methods MASH was developed using data from 1285 older community‑ dwelling women (ages: 60–72 years) who concurrently wore a hip‑ worn ActiGraph GT3X + accelerometer (waking activity) and a wrist‑ worn Actiwatch 2 device (sleep) for ≥ 4 days (N = 10,123 days) at the same time. MASH is a two‑tiered process using (1) scored sleep data (from Actiwatch) or (2) one‑ dimensional convolutional neural networks (1D CNN) to create predicted wake intervals, recon‑ cile sleep and activity data disagreement, and create day‑level night ‑ day‑night pairings. MASH chooses between two different 1D CNN models based on data availability (ActiGraph + Actiwatch or ActiGraph‑ only). MASH was evaluated using Receiver Operating Characteristic (ROC) and Precision‑Recall curves and sleep–wake intervals are compared before (pre‑harmonization) and after MASH application. Results MASH 1D CNNs had excellent performance (ActiGraph + Actiwatch ROC‑AUC = 0.991 and ActiGraph‑ only ROC‑AUC = 0.983). After exclusions (partial wear [n = 1285], missing sleep data proceeding activity data [n = 269], and < 60 min sleep [n = 9]), 8560 days were used to show the utility of MASH. Of the 8560 days, 46.0% had ≥ 1‑min disagreement between the devices or used the 1D CNN for sleep estimates. The MASH waking intervals were cor‑ rected (median minutes [IQR]: − 27.0 [− 115.0, 8.0]) relative to their pre‑harmonization estimates. Most correction (− 18.0 [− 93.0, 2.0] minutes) was due to reducing sedentary behavior. The other waking behaviors were reduced a median (IQR) of − 1.0 (− 4.0, 1.0) minutes. Erin E. Dooley and J. F. Winkles share first authorship. *Correspondence: Erin E. Dooley edooley@uab.edu Full list of author information is available at the end of the article © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 2 of 12 Conclusions Implementing MASH to harmonize concurrently worn hip and wrist devices can minimizes data loss and correct for disagreement between devices, ultimately improving accuracy of 24‑h compositions necessary for time‑use epidemiology. Keyword Actigraphy, Accelerometer, Sleep, Physical activity, Harmonization, Machine learning, 24‑h activity, Time ‑ use epidemiology Background such as independently funded ancillary studies to large Time-use movement behaviors (e.g., sleep, sedentary prospective cohorts with initial protocols proposing dif- behavior, and physical activity) [1–3] are modifiable fac - ferent devices. To reduce participant burden of wearing tors associated with numerous health outcomes and all- devices across many weeks of data collection, researchers cause mortality [4–8]. Wearable accelerometers are used may need to collaborate on a multiple-device wear proto- to measure time-use behaviors in free-living settings for col occurring over the course of one week, for example. a variety of populations [9, 10]. While there are numer- To characterize 24-h sleep–wake compositions for ous consumer wearables (e.g., smart watches, fitness multiple-device protocols, approaches to harmonizing monitors) that have the capacity to estimate waking and simultaneously collected data from multiple devices are sleep behaviors, potential issues such as user feedback needed. Unfortunately, data collection in naturalistic set- biases and data extraction limit their research utility [11, tings increases the potential for protocol deviations that 12]. Additionally, consumer device (e.g., Apple, Fitbit, undermine accuracy. Specifically, missing data due to Garmin) accuracy depends on manufacturer and device device malfunction, improper initialization, and/or the type and are liable to algorithm changes that may occur participant forgetting to wear one or both devices on numerous times throughout a study and without notifica - one or more days complicates construction of day-night tion to investigators, creating unmeasurable noise in the pairings. Another common issue researchers face is that data [13–17]. Research-grade devices (e.g., ActiGraph, despite being instructed to remove the device assessing activPAL, Actiwatch, GENEActiv) provide substantial waking activity during sleep periods, it is not uncommon flexibility over consumer devices for processing and re- for participants to wear these devices longer than neces- processing of data to achieve current best-practices. sary (i.e., to bed), which can inflate estimates of seden - However, devices for quantifying 24-h time-use behav- tary behavior time. Through this lens, a multiple-device iors can be placed on different anatomical locations (e.g., data harmonization process needs to facilitate two types hip, wrist, thigh) and can be worn for different amounts of adjustment when sleep and activity data are joined of time (e.g., waking only, sleep only, 24  h/day) depend- for 24-h cycle development. First, the frame of refer- ing on the primary outcome(s) of interest. For example, ence for a day should be defined as the concatenation of researchers only interested in physical activity behaviors a dynamic sleep–wake interval rather than a constrained may utilize waking hip placements, whereas those inter- period (e.g., midnight to midnight) that may be utilized ested in sedentary behaviors may want to consider pos- when behaviors are viewed separately [18, 22]. Second, tural positions and therefore need a device with a thigh harmonization should correct any overlap between mon- placement, and those interested in sleep and/or circadian itors if there is behavior categorization disagreement, phase may utilize wrist placements [18, 19]. which may address situations where sleep is incorrectly While a single wrist-worn device to capture movement classified as sedentary behavior or non-wear for one behaviors have become increasingly popular, validity for device. The latter could result in inaccuracies due to the measuring physical activity across the intensity spec- device (still) recording after it was removed. trum against criterion measures (e.g., indirect calorim- Herein, we present the Method for Activity Sleep Har- etry, doubly labeled water) has a wide range of accuracy monization (MASH) process, a novel method that har- (r = 0.17–0.93) [20]. This may lead some researchers to monizes data from multiple devices to create coherent implement protocols that have participants switch the sleep-activity pairings. MASH is a multiple device (hip device between the hip (during the day) and wrist (at and wrist) harmonization method that addresses many of night), potentially increasing participant non-compliance the issues described above including, missing data (e.g., [21], or consider protocols in which participants wear not wearing one device) or discordant behavior charac- two or more devices concurrently, as in the current study. terization (e.g., one device characterizes sleep whereas Other instances when a multiple-device protocol may be the other device characterizes sedentary behavior), while necessary include when multiple funded studies are being also accommodating both regular and irregular sleep pat- conducted simultaneously on a single study population, terns and minimizing data loss. We detail this method D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 3 of 12 using data from 1285 older women who concurrently Data collection wore an ActiGraph accelerometer on the hip and an Acti- Waking behaviors were quantified using the hip-worn watch 2 device on the wrist for up to seven days as part of ActiGraph wGT3X+ (ActiGraph, Pensacola, FL) device the Study of Women’s Health Across the Nation (SWAN). during all waking hours, except for water-based activities, for up to seven days. Raw acceleration data were sampled Methods at 40 Hz and were downloaded and reintegrated to a 60-s Parent study epoch using ActiLife6 software [18]. Wear and non-wear Data are from SWAN, an ongoing longitudinal multi- (time periods in which participants did not wear the site cohort study of women, which has been previously hip device, such as sleeping and water-based activities) described [23]. Briefly, 3302 women ages 42–52  years were defined using the Choi algorithm with the ‘Physi - (mean age ± standard deviation [SD]: 46.4 ± 2.7  years) calActivity’ R package [24]. Evenson vector magnitude were recruited from seven geographic sites across the (VM) cut point values [25] were used to classify minutes −1 U.S.: Boston, MA; Chicago, IL; Southeast area, Michi- as sedentary behavior (< 76 VMct·min ), low light (76 −1 gan; Los Angeles, CA; Newark, NJ; Pittsburgh, PA; and to < 903 VMct·min ) intensity (LLPA), high light (903 −1 Oakland, CA. Each site recruited White women and to < 2075 VMct·min ) intensity (HLPA),  and  moderate women of one other race/ethnicity. Cohort members to vigorous intensity physical activity (MVPA) (≥ 2075 −1 have been followed through 16 follow-up visits approxi- VMct·min ). The original 15-s thresholds were multi - mately every year. Data for these analyses were collected plied by four to account for the longer epoch (60-s), with at the SWAN follow-up visit 15 (2015–2017) (N = 2091 slight adjustments to obtain mutually exclusive threshold women), in which a subsample of women were invited ranges [26]. For the waking interval, days were classified to concurrently wear two devices to quantify waking and as adherent if they had ≥ 600  min of wear time. Partici- sleep behaviors. A total of 1285 women had valid data pants were included if they had ≥ 4 adherent days [18]. for MASH development and evaluation (Fig.  1). Ethics These days did not need to be consecutive. approval was obtained from Institutional Review Boards The sleep interval was quantified using the wrist-worn at each of the seven SWAN sites and all participants pro- Actiwatch 2 (Philips Respironics, Murrysville, PA) device vided written informed consent at each visit. worn for 24  h/day on the non-dominant wrist. Partici- pants completed a diary and were asked to press an event Fig. 1 Participant flow diagram for data harmonization Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 4 of 12 marker on the watch to indicate when they went to bed occurred for nights that had a Actiwatch malfunction, with the intention to sleep and when they rose from bed the Actiwatch was removed, or there was no/poor sleep. for the final time each day. The sleep diary included ques - In these cases, one-dimensional Convolutional Neural tions regarding when they got into bed, the time they Network (1D CNN) models [28] are used. 1D CNNs were tried to go to sleep, the time they woke up for the day, chosen because they have been previously employed on and the time when they rose from bed. The Actiwatch a variety of actigraphy data for algorithm detection [29– was set at 0.05  g for 3–11  Hz and data were sampled in 31]. The 1D CNN models read the epoch-level ActiGraph 60-s epochs. To determine total scored sleep time, the and Actiwatch data and assign each epoch with a prob- Actiwatch 2 data were processed, evaluated for quality, ability of being ‘within a wake interval’. and scored with the event marker, default sleep detection Two 1D CNN models were created for MASH. One algorithm (wake threshold: 40 ct/min) [27], and the sleep 1D CNN model accommodates both ActiGraph + Acti- diary (if available) in Actiware 5.0.9 using procedures watch data, using data from both devices, and a separate consistent with the Society of Behavioral Sleep Medicine 1D CNN model uses ActiGraph-only data, for situations guidelines [19]. Clock times for sleep onset (beginning when the Actiwatch data were invalid or missing. Once of sleep interval) and sleep offset (end of sleep inter - the 1D CNN models generate epoch-level predictions, val) were determined using the start of the first minute a simple optimization procedure is used to determine (onset) or the last minute (offset) of 10 consecutive min - which clusters of epochs were most likely to represent utes of immobility. In addition, all sleep records were vis- the true waking interval. See Additional file  1: Appendix ually inspected for quality. Sleep records were removed SA for a conceptual framework of the MASH process. if there was a Actiwatch malfunction, the Actiwatch was The 1D CNN models were trained using all days that removed prior to sleep (e.g., non-wear), there was no/ had valid sleep data surrounding them. This sample was poor sleep, or there was < 60  min of total scored sleep. randomly divided into training, test, and validation data- These records are henceforth referred to as ‘valid scored sets of mutually exclusive individuals. Each 1D CNN sleep data’. model created epoch-level predictions by evaluating cen- tered 101-epoch windows of time surrounding the epoch MASH development in question [30]. We considered the costs of misclassify- MASH utilizes data from three sources: (1) ActiGraph– ing each epoch as ‘within wake interval’ or ‘outside wake count data from Axes 1–3, (2) Actiwatch–lux (white interval’ as equal; therefore, the optimal cutoff probabil - light) and count data, and (3) sleep onset and sleep offset ity differentiating these statuses was determined using clock times–from the valid scored sleep data correspond- Youden’s J-statistic [32]. See Additional file  1: Appendix ing to the beginning (sleep onset) and end (sleep offset) SB and SC for a detailed description of the approaches of the primary sleep interval. used to join the sleep and waking datasets and model Harmonizing the sleep and activity data through building. MASH is a two-tiered approach that addresses two states of data availability. At its core, MASH reconciles the two Removing the hip device at night prior to sleep onset datasets (i.e., ActiGraph- and Actiwatch-derived) by While developing MASH, we noticed the predicted sleep determining the bounds of the ‘waking interval’ for each intervals resulting from the 1D CNNs were more likely to day. The creation of these intervals represents the coher - have shorter waking intervals (both pre-harmonization ent fusion of the two datasets: nig ht -day -night . and after MASH application) and longer sleeping inter- (t-1) (t) (t) Any ‘correction’ to the waking behaviors (sedentary vals compared to valid scored sleep-derived intervals behavior, LLPA, HLPA, MVPA) associated with the (e.g., Actiwatch dataset). While it could be that records imposition of these intervals resulted from a disagree- missing sleep data might have shorter waking intervals ment between the sleep and activity data (e.g., the Acti- because women who were less likely wear the Actiwatch watch says a person is asleep and the ActiGraph says they 2 wrist device might not be as diligent at wearing the are engaging in sedentary behavior). ActiGraph hip device (thus having shorter wake inter- The first tier of MASH applies to instances where the vals), having longer sleep intervals is problematic because 24-h period has valid scored sleep data preceding and the act of removing the hip device was being confused proceeding it. The wake interval is built using the pre - with sleep onset. vious day’s sleep onset time and the current day’s sleep For records that had valid scored sleep data, the aver- offset time (night -day -night ). The second tier is age difference between removing the hip device and sleep (t-1) (t) (t) used for all other instances where a wake interval does onset was 44.4  min. For all records that did not have not have valid scored sleep data (e.g., missing sleep onset valid scored sleep data, the average duration of the sleep or sleep offset) immediately surrounding it. This typically interval was 45.2 min longer (P < 0.001) than the intervals D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 5 of 12 where valid scored sleep data were present. Given the chi-square tests for categorical variables. Selected par- similar sizes of each effect, we therefore used records ticipant characteristics included self-reported age (years), with valid scored sleep data to construct a bivariate prob- race/ethnicity (Black, Chinese, Hispanic, Japanese, ability distribution for wake interval length by the size of White), education (< high school, high school, some col- the difference between removing the hip device and sleep lege, college, post-college), self-rated health (poor, fair, onset. The probability distribution was constructed using good, very good, excellent), difficulty walking one mile bounded 2-dimensional kernel density estimation with a (yes/no), and obesity (body mass index [BMI] ≥ 30  kg/ minimum value equivalent to what was in the data and m ) calculated using height and weight at visit 15. Model an imposed maximum value of 200  min for each varia- performance was examined using the C-statistic, i.e., ble (~ 3 SD from the mean of 44.4 min). Given the size of the area under the Receiver Operating Characteristic each wake interval, the probability distribution was used (AUC-ROC) curve. To account for slight data imbalance to generate an estimate of the amount of time that exists (roughly a 66/33 split between ‘within wake interval’ and between removing the hip device and sleep onset. This ‘outside of wake interval’), Precision-Recall curves were estimate was added to the predicted timestamp for sleep also used [33]. Paired t-tests were used to examine esti- onset (thus shortening the duration of the sleep interval). mates of time-use movement behaviors between days This process was replicated ten times for all records that were MASH-corrected and when the estimates were that did not have scored sleep data indicating sleep onset. not corrected. The MASH intervals were then constructed using the average of these ten samples. For more information on Results this process, please consult Additional file  1: Appendix 1D CNN construction and accuracy SD  and SE. MASH user documentation is available at The sample used to build the prediction models included https:// github. com/ jsw70/ MASH. 1112 older women who had both valid scored sleep data preceding and following each wake interval in ques- MASH evaluation tion. Participants were similarly distributed (P > 0.05) for The full sample included 10,123 days with useable accel - demographics and selected health characteristics across erometry data across 1285 participants. In order to eval- the training (n = 625), test (n = 278), and validation uate MASH, for this analysis we focused on days that had (n = 209) sets (Table 1). full sleep–wake compositions consisting of a sleeping The AUC-ROC for both 1D CNN models developed for interval followed by a waking interval. This was done to MASH (ActiGraph + Actiwatch or ActiGraph-only) were maximize the number of full compositions we could eval- considered excellent (Fig.  2) with values of 0.991 and uate. For example, while we could have chosen to view a 0.983 for the ActiGraph + Actiwatch and ActiGraph-only composition as being a day proceeded by night (wake- models, respectively. In addition, the accompanying Pre- sleep) this would have led to fewer compositions for cision-Recall AUC were 0.993 and 0.989. Using Youden’s evaluation as the last day of data collection would likely J-statistic to determine a cutoff probability threshold be excluded (no sleep data). However, of note, the MASH (0.698 and 0.729), the sensitivity of the models at each process creates a bi-directional dataset that allows for optimal point was 95.7% for the ActiGraph + Actiwatch flexibility in examining both sleep–wake or wake-sleep model and 92.8% for the ActiGraph-only model. The compositions. specificity of the 1D CNN models was 95.5% and 95.6%, To evaluate the harmonization process, three exclu- respectively. sions were applied to the sleep–wake data: (1) the first day of data collection for the waking interval was removed because it was a partial day with the first Data harmonization instance of detected wear corresponding to when the Of the 8560 sleep–wake compositions, 84.9% (n = 7270 devices were distributed and placed on the participant records) had valid scored sleep data (i.e., had both sleep during the in-person exam visit (n = 1285  days), (2) any onset and sleep offset). Of the remaining 15.1% of days observation that did not have sleep data preceding the (n = 1290 records), either of the 1D CNN models was wake interval (n = 269), and (3) any instances where the applied to estimate (1) sleep offset (n = 32 days), (2) sleep sleep data was less than 60  min (n = 9). The analytical onset (n = 503  days), or (3) both sleep offset and onset dataset for the evaluation of MASH included 8560 sleep– (n = 755 days) (Table 2). wake compositions (Fig. 1). With the sleep and wake intervals defined, 46.0% (3934 Potential differences in participant characteris - of 8560) of days needed correction to the waking inter- tics between the training, test, and validation datasets val. This was to address improper classification of at least were assessed using t-tests for continuous variables or one minute-level epoch as both sleep and wake (82.9%; Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 6 of 12 Table 1 One‑ dimensional Convolutional Neural Network (1D CNN) SWAN follow‑up visit 15 (2015–2017) participant characteristics, overall and by dataset Characteristic Sample Training set Test set Validation set (N = 1112) (n = 625) (n = 278) (n = 209) % (n) % (n) % (n) % (n) Age (M ± SD) 65.5 ± 2 65.4 ± 2 65.2 ± 2 65.8 ± 2 Race/ethnicity Black 25.8 (287) 27.8 (174) 23.7 (66) 22.5 (47) Chinese 12.9 (143) 13.0 (81) 15.1 (42) 9.6 (20) Hispanic 3.0 (33) 3.0 (19) 2.5 (7) 3.3 (7) Japanese 12.1 (134) 11.2 (70) 13.3 (37) 12.9 (27) White 46.3 (515) 45.0 (281) 45.3 (126) 51.7 (108) Education < High school 4.0 (45) 5.0 (31) 2.5 (7) 3.3 (7) High school 14.9 (166) 13.9 (87) 16.2 (45) 16.3 (34) Some college 31.3 (348) 28.8 (180) 33.1 (92) 36.4 (76) College 22.8 (253) 23.5 (147) 21.6 (60) 22.0 (46) Post‑ college 26.3 (292) 28.5 (178) 25.2 (70) 21.1 (44) Missing 0.7 (8) 0.3 (2) 1.4 (4) 1.0 (2) Obesity (BMI ≥ 30 kg/m ) 36.1 (401) 38.2 (239) 32.7 (91) 34.0 (71) Missing 1.0 (11) 1.0 (6) 1.1 (3) 1.0 (2) Self‑rated health Poor, fair, or good 47.3 (526) 47.8 (299) 49.3 (137) 43.1 (90) Missing 0.8 (9) 1.1 (7) 0.7 (2) Difficulty walking one mile 36.5 (406) 39.4 (246) 34.2 (95) 31.1 (65) BMI body mass index There were no statistically significant differences between the training, test, and validation datasets at the P = 0.05 level using t-tests for continuous variables or chi- square tests for categorical variables Fig. 2 Receiver Operating Characteristic (ROC) curves with cutoff thresholds and sensitivity and specificity values D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 7 of 12 Table 2 Number of days in the sample, MASH model used, and sleep–wake interval size Reason Number of days MASH model applied Sleep–wake (N = 8560) interval size, hours Mean (SD) No sleep data are missing 7270 Valid scored sleep 24.04 (1.41) Sleep data are missing 1290 Only sleep onset data are missing 503 1D CNN 23.20 (2.35) Sleep offset data are missing on any day besides first day 32 1D CNN 23.92 (2.52) Both sleep onset and sleep offset data are missing 755 1D CNN 23.88 (1.68) 1D CNN one-dimensional convolutional neural network; MASH method for activity sleep harmonization 3262 of 3934  days) or due to missing sleep data (672 of (median [IQR] = −  76.0 [−  150.0, −  14.0] min of total 3934 days). wake time) for prediction, even though the distribution For days requiring correction, the average correc- of the wake intervals requiring 1D CNN correction were tion applied included a median (interquartile range relatively more skewed. [IQR]) of −  27.0 (−  115.0, 8.0) minutes of total wake Table  3 presents the time-use behavior estimates pre- time. The distribution of the MASH-corrected wake harmonization (e.g., prior to MASH implementation) intervals smoothed out a cluster of days that the Choi and the estimates once harmonized using MASH. When algorithm (applied to ActiGraph data) classified as hav - compared to other waking behavior estimates (i.e., LLPA, ing > 1200 min (20 h) of waking wear (Fig. 3). This finding HLPA, and MVPA) the distribution of the sedentary is consistent regardless of whether the wake interval was behavior estimate was most influenced once the wear corrected using scored sleep data (median [IQR] = − 21.0 intervals were corrected (Fig.  4). Specifically, sedentary [−  98.0, 12.0] min of total wake time) or 1D CNN behavior was corrected a median (IQR) of − 18.0 (− 93.0, Fig. 3 Distribution of wake interval sizes between the uncorrected days and the corrected days, overall and by correction method Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 8 of 12 Table 3 Pre‑harmonization and post MASH harmonization time ‑use estimates (N = 8560) days Time-use behavior Pre-harmonization, MASH harmonization, minutes minutes Median (IQR) Median (IQR) Sedentary behavior 438 (344, 547) 419 (329, 512) Low light intensity physical activity (LLPA) 277 (219, 337) 274 (216, 333) High light intensity physical activity (HLPA) 137 (100, 181) 136 (99, 181) Moderate to vigorous intensity physical activity (MVPA) 46 (24, 78) 45 (24, 78) Sleep 440 (378, 497) 444 (383, 502) MASH method for activity sleep harmonization a −1 −1 Evenson vector magnitude (VM) cut point values were used to classify minutes as sedentary behavior (< 76 VMct·min ), low light (76 to < 903 VMct·min ) intensity −1 −1 (LLPA), high light (903 to < 2075 VMct·min ) intensity (HLPA), and MVPA (≥ 2075 VMct·min ) Pre-harmonization included 8024 days due to missing data Fig. 4 Distribution of waking activity lengths between the uncorrected and corrected activity variables 2.0) min, whereas the other waking behavior types missing), it was not possible to perform t-test analysis on (LLPA + HLPA + MVPA) were corrected a median (IQR) sleep because a substantial portion of the data did not of −  1.0 (−  4.0, 1.0) min. Paired t-test analysis results have an ‘uncorrected’ sleep measurement. demonstrate MASH resulted in statistically significant Participants had a mean (SD) of 6.7 (1.5) sleep–wake reductions in all forms of activity; however, only seden- compositions. The final MASH harmonized dataset tary behavior [t(8559) = −  34.2, P < 0.001] had a mean resulted in a mean (SD) sleep–wake composition interval difference greater than 3 min. Because the wake interval size of 23.97 (1.52) hours. The interval sizes of the sleep– correction process within MASH also simultaneously wake compositions were similar across the MASH model creates sleep intervals (in cases where the sleep data are applied (Table 2). D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 9 of 12 Discussion disorders (e.g., insomnia, sleep apnea), and excess health- Measuring time-use movement behaviors accurately care costs [8, 49, 50]. across the continuous 24-h period is critical as these Choosing between a single device protocol or a pro- behaviors are interrelated and evidence suggests that tocol in which participants wear two or more devices the combined effects of these behaviors on health may concurrently, as in the current study, is largely up to be greater than their individual effects [34–36]. This researcher preferences. However, multiple-device proto- had led to 24-h public health guidelines released by the cols may be necessary for independently funded ancillary World Health Organization [37] and by some countries studies to large prospective cohorts to reduce participant (e.g., Australia [38], Canada [39, 40], New Zealand [41], burden of wearing devices across many weeks of data South Africa [42]). Thus, accurate estimation of 24-h collection. Placing multiple devices at different anatomi - sleep–wake cycles, including the contributing behaviors, cal locations to increase precision of time-use behaviors is paramount. We developed the Method for Activity has been previously implemented on the hip + thigh Sleep Harmonization (MASH) to harmonize time-series [51] and chest + thigh [52]. The MASH method provides data from two accelerometers that use two different an integral step in 24-h time-use assessment by joining placements (wrist and hip) to estimate behaviors com- accelerometry data from the hip + wrist for increased prising the 24-h period. This method creates night-day- measurement precision. Prior to this method develop- night pairings rather than constraining data to a fixed ment, researchers would need to weigh the decision of time period (e.g., midnight to midnight). We analyzed an which behavior outcome could have poor performance interval of sleep followed by a subsequent waking inter- accuracy within their study or have participants switch val as a 24-h sleep–wake composition. This accounts between wear locations, with the possibility of increas- for the compositional nature of behaviors used in time- ing non-compliance [21]. Further, providing open-source use epidemiology [1–3]. Developed on a large sample of code may help with data harmonization and protocol older women, the findings suggest the MASH approach development across studies. (1) minimized data loss due to missing sleep data and In addition to increasing 24-h measurement precision, (2) improve precision of 24-h sleep–wake compositions. MASH also minimizes data loss. MASH evaluates the Together, these findings support the utility of MASH to epoch-level data and applies a classification algorithm harmonize sleep–wake data obtained from two devices that scores the epoch as either within a waking interval and correct these data, as needed, to more precisely esti- or within a sleeping interval. Therefore, although diary/ mate time-use movement behaviors for further analysis. sleep data may be missing, night-day-night pairings could Physical activity and sleep have largely been separate still be constructed. In our sample, a total of 15.1% of disciplines, with each preferring certain devices and days (n = 1290 records) were missing either sleep onset, anatomical placement for field-based data collection. sleep offset, or both times from the sleep dataset. With - For measuring waking activities, specifically time spent out this classification algorithm, those days would be lost within intensity categories, triaxial accelerometer place- during analysis or would need to be imputed. Further, ment is most accurate at the hip [43, 44]. However, for MASH can be used to create daily 24-h compositions sleep detection, reliable accelerometry measurement rather than a single averaged daily estimate, which can occurs on the wrist [19, 45, 46]. Findings from Full and have important implications for examining time-use pat- colleagues suggest estimates of sleep duration using an terns of behaviors across the week and development of ActiGraph worn on the hip were significantly higher from future intervention studies targeting these behaviors. polysomnography (PSG), overestimating total sleep time The limitations of MASH should be noted. This method by 37.8 (SD = 61.3) min [47]. This could be the reason the was developed in a sample of community-dwelling older majority of corrections attributed to MASH were to fix adult women (age range: 60–72  years). However, we do instances where sleep was coded as sedentary behavior. not believe this would change model development, and Further, total volume of physical activity measured by given the flexibility and utility of 1D CNN models, imple - wrist-worn devices (e.g., Actiwatch 2) have a weak cor- mentation of MASH in other studies and populations is relation (r = 0.26) with hip-worn devices and thus are achievable. In addition, MASH only removes waking data not favorable for measuring physical activity [48]. Over- when the wake interval sizes are longer than the MASH estimation of sedentary behavior and underestimation of prediction, which occurs in instances where the partici- sleep can have detrimental effects to outcomes research. pant was likely wearing the ActiGraph monitor on the Sleep, in the absence of disturbances or disorders, is hip while sleeping. MASH is unable to determine activ- thought to be a restorative and health-promoting pro- ity behavior when data are missing due to not wearing cess for the body [7], whereas excessive sedentary behav- the device during the waking interval, for example, if a ior is associated with several diseases, including sleep participant woke up and did not immediately put on the Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 10 of 12 SWAN Study of Women’s Health Across the Nation waking device. We did not calculate these data from the SD Standard deviation wrist-worn device as the Actiwatch 2 is not accurate for VM Vector magnitude measuring physical activity [48] and we did not impute LLPA Low light intensity physical activity HLPA High light intensity physical activity these minutes as we were unable to determine the true 1D CNN One ‑ dimensional Convolutional Neural Network waking activity behavior. However, statistical techniques BMI Body mass index such as compositional data analysis (CoDA), which treat AUC‑ROC Area under the Receiver Operating Characteristic IQR Interquartile range daily time-use movement behaviors as a composition that are translated to real space through the applica- Supplementary Information tion of coordinate systems and constrained to 1440  min The online version contains supplementary material available at https:// doi. [53], overcome non-wear issues. With MASH, the period org/ 10. 1186/ s44167‑ 023‑ 00017‑5. within the resting and sleep intervals (e.g., sleep onset latency) is classified as sedentary behavior, which is sup - Additional file 1: SA. Method of Activity Sleep Harmonization (MASH) ported by the Sedentary Behavior Research Network conceptual framework. SB. Determining valid scored sleep data validity. SC. Building the 1D CNN models. Figure S1. The network structure for (SBRN)’s definition as ‘any waking behavior character - both 1D CNN models with and without the Actiwatch data. Table S1. ized by an energy expenditure ≤ 1.5 metabolic equiva- Hyperparameters for both One‑ dimensional Convolutional Neural Net‑ lents, while in a sitting, reclining, or lying posture’ [54]. work (1D CNN) models. Figure S2. Precision‑Recall curves. SD. Account‑ ing for the 1D CNN tendency to confuse hip device removal as sleep However, long sleep latency may have differential effects onset. Figure S3. Bivariate probability distribution for the amount of time on health than sedentary behavior [55] and the health- found between hip device removal and sleep onset for all valid scored related importance of distinguishing this period requires sleep data. Figure S4. Comparing the distribution of sleep intervals created by the 1D CNN’s and the scored sleep data. The two graphs are further study [56]. Lastly, MASH only classifies the main separated by whether or not the bivariate sampling was used to adjust (overnight) sleep interval and did not attempt to classify for confusing hip device removal with sleep onset. SE. Determining wake daytime sleeping (napping) from either device. Despite intervals from epoch‑level 1D CNN predictions. the limitations, we built MASH and the 1D CNN predic- tion models using a large sample of women (N = 1285) Acknowledgements who wore two devices (hip + wrist) concurrently. The We thank the study staff at each site and all the women who participated in SWAN. models had excellent classification and there were no significant differences between the scored sleep and 1D Author contributions CNN prediction models. Using the 1D CNN can help ED and JW contributed equally to this work. ED conceptualized the study, supported the formal analysis, investigation, and methodology and was the minimize data loss for days when participants forget to lead in writing the manuscript. JW was the lead in the analysis, investigation, wear the wrist device or there was a device malfunction. methodology, and visualization of the figures, and supported in preparing the Methods and Results sections of the text. AC and KPG conceptualized the study, supported the formal analysis, investigation, and methodology, reviewed the manuscript, and provided supervision for the study. CK, SB, KD, Conclusions CKG, BS, ST, and MH provided a critical role in reviewing and editing the text. MASH is a dataset harmonization method for merging HK reviewed and edited the text and was a primary investigator during fund‑ sleep and waking activity behaviors measured concur- ing acquisition. All authors have read and approved the final manuscript. rently from multiple devices (hip + wrist). The devices Funding were chosen because of their accuracy in measuring wak- The Study of Women’s Health Across the Nation (SWAN) has grant support ing activity behaviors (ActiGraph GT3X+) and sleep from the National Institutes of Health (NIH), DHHS, through the National Institute on Aging (NIA), the National Institute of Nursing Research (NINR) and (Actiwatch 2). We built MASH to merge separate, inde- the NIH Office of Research on Women’s Health (ORWH) (Grants U01NR004061; pendent datasets, minimize data loss for missing sleep U01AG012505, U01AG012535, U01AG012531, U01AG012539, U01AG012546, data, and with the flexibility that this process can be rep - U01AG012553, U01AG012554, U01AG012495, and U19AG063720). The content of this article is solely the responsibility of the authors and does not licated in other studies that simultaneously collect sleep necessarily represent the official views of the NIA, NINR, ORWH or the NIH. and waking behaviors using two devices. Researchers can Clinical Centers: University of Michigan, Ann Arbor–Carrie Karvonen‑ use the MASH approach to correct sleep–wake harmo- Gutierrez, PI 2021–present, Siobán Harlow, PI 2011–2021, MaryFran Sowers, PI 1994–2011; Massachusetts General Hospital, Boston, MA–Sherri‐Ann nization, construct daily-level compositions, and aggre- Burnett‐Bowie, PI 2020–Present; Joel Finkelstein, PI 1999–2020; Robert Neer, gate to averaged daily values as needed. Ultimately, this PI 1994–1999; Rush University, Rush University Medical Center, Chicago, IL – approach increases precision of the physical activity and Imke Janssen, PI 2020–Present; Howard Kravitz, PI 2009–2020; Lynda Powell, PI 1994–2009; University of California, Davis/Kaiser–Elaine Waetjen and Monique sleep estimates which may improve the accuracy of the Hedderson, PIs 2020–Present; Ellen Gold, PI 1994–2020; University of California, observed measures of association with health outcomes. Los Angeles‑Arun Karlamangla, PI 2020–Present; Gail Greendale, PI 1994–2020; Albert Einstein College of Medicine, Bronx, NY–Carol Derby, PI 2011–present, Rachel Wildman, PI 2010–2011; Nanette Santoro, PI 2004–2010; University of Abbreviations Medicine and Dentistry‑New Jersey Medical School, Newark–Gerson Weiss, PI LIPA Light intensity physical activity 1994–2004; and the University of Pittsburgh, Pittsburgh, PA—Rebecca Thurs‑ MVPA Moderate to vigorous intensity physical activity ton, PI 2020–Present; Karen Matthews, PI 1994–2020. MASH Method for Activity Sleep Harmonization D ooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 11 of 12 NIH Program Office: National Institute on Aging, Bethesda, MD—Rosaly 5. World Health Organization. Global action plan on physical activity Correa‑ de‑Araujo 2020–present; Chhanda Dutta 2016–present; Winifred Rossi 2018–2030: more active people for a healthier world. Geneva: World 2012–2016; Sherry Sherman 1994–2012; Marcia Ory 1994–2001; National Health Organization; 2018. Institute of Nursing Research, Bethesda, MD—Program Officers. 6. World Health Organization. WHO guidelines on physical activity and Central Laboratory: University of Michigan, Ann Arbor–Daniel McConnell sedentary behaviour. Geneva: World Health Organization; 2020. (Licence: (Central Ligand Assay Satellite Services). CC BY-NC-SA 3.0 IGO). Coordinating Center: University of Pittsburgh, Pittsburgh, PA–Maria Mori 7. Consensus Conference Panel. Recommended amount of sleep for a Brooks, PI 2012–present; Kim Sutton‑ Tyrrell, PI 2001–2012; New England healthy adult: a joint consensus statement of the American academy Research Institutes, Watertown, MA–Sonja McKinlay, PI 1995–2001. of sleep medicine and sleep research society. Sleep. 2015;38(6):843–4. Steering Committee: Susan Johnson, Current Chair; Chris Gallagher, Former https:// doi. org/ 10. 5665/ sleep. 4716. Chair. 8. Patterson R, McNamara E, Tainio M, et al. Sedentary behaviour and risk of all‑ cause, cardiovascular and cancer mortality, and incident type 2 Availability of data and materials diabetes: a systematic review and dose response meta‑analysis. Eur J Epi‑ SWAN provides access to public use datasets that include data from SWAN demiol. 2018;33(9):811–29. https:// doi. org/ 10. 1007/ s10654‑ 018‑ 0380‑1. screening, the baseline visit and follow‑up visits (https:// aging resea rchbi 9. Troiano RP, Stamatakis E, Bull FC. How can global physical activity surveil‑ obank. nia. nih. gov/). To preserve participant confidentiality, some, but not all, lance adapt to evolving physical activity guidelines? Needs, challenges of the data used for this manuscript are contained in the public use datasets. and future directions. Br J Sports Med. 2020;54(24):1468–73. https:// doi. A link to the public use datasets is also located on the SWAN web site: http:// org/ 10. 1136/ bjspo rts‑ 2020‑ 102621. www. swans tudy. org/ swan‑ resea rch/ data‑ access/. Investigators who require 10. Omura JD, Whitfield GP, Chen TJ, et al. Surveillance of physical activity and assistance accessing the public use dataset may contact the SWAN Coordinat‑ sedentary behavior among youth and adults in the United States: history ing Center at the following email address: swanaccess@edc.pitt.edu. and opportunities. J Phys Act Health. 2021;18(S1):S6–24. https:// doi. org/ 10. 1123/ jpah. 2021‑ 0179. 11. Strain T, Wijndaele K, Pearce M, Brage S. Considerations for the use of Declarations consumer‑ grade wearables and smartphones in population surveillance of physical activity. J Meas Phys Behav. 2022. https:// doi. org/ 10. 1123/ Ethics approval and consent to participate jmpb. 2021‑ 0046. Ethics approval was obtained from Institutional Review Boards at each of the 12. Hyde ET, Omura JD, Fulton JE, et al. Physical activity surveillance using seven SWAN sites and all participants provided written informed consent at wearable activity monitors: are US adults willing to share their data? Am each visit. J Health Promot. 2020;34(6):672–6. https:// doi. org/ 10. 1177/ 08901 17119 Consent for publication 13. Fuller D, Colwell E, Low J, et al. Reliability and validity of commercially Not applicable. available wearable devices for measuring steps, energy expenditure, and heart rate: Systematic review. JMIR Mhealth Uhealth. 2020;8(9): e18694. Competing interests https:// doi. org/ 10. 2196/ 18694. The authors declare that they have no competing interests. 14. Straiton N, Alharbi M, Bauman A, et al. The validity and reliability of consumer‑ grade activity trackers in older, community‑ dwelling adults: a Author details systematic review. Maturitas. 2018;112:85–93. https:// doi. org/ 10. 1016/j. Department of Epidemiology, The University of Alabama at Birmingham, matur itas. 2018. 03. 016. Birmingham, AL, USA. Epidemiology Data Center, The University of Pittsburgh, 15. Scott H, Lack L, Lovato N. A systematic review of the accuracy of sleep Pittsburgh, PA, USA. Department of Epidemiology, The University of Pitts‑ wearable devices for estimating sleep onset. Sleep Med Rev. 2020;49: burgh, Pittsburgh, PA, USA. Department of Health and Human Development, 101227. https:// doi. org/ 10. 1016/j. smrv. 2019. 101227. The University of Pittsburgh, Pittsburgh, PA, USA. Division of Research, Kaiser 16. O’Driscoll R, Turicchi J, Beaulieu K, et al. How well do activity monitors Permanente Northern California, Oakland, CA, USA. Center for Behavioral estimate energy expenditure? A systematic review and meta‑analysis of Cardiovascular Health, Columbia University Medical Center, New York, NY, USA. the validity of current technologies. Br J Sports Med. 2020;54(6):332–40. Department of Epidemiology, University of Michigan, Ann Arbor, MI, USA. 17. Evenson KR, Goto MM, Furberg RD. Systematic review of the validity and Department of Psychiatry and Behavioral Sciences and Department of Pre‑ reliability of consumer‑ wearable activity trackers. Int J Behav Nutr Phys ventive Medicine, Rush University Medical Center, Chicago, IL, USA. Depar t‑ Act. 2015;12(1):159. https:// doi. org/ 10. 1186/ s12966‑ 015‑ 0314‑1. ment of Psychiatry and Behavioral Neurobiology, The University of Alabama 18. Migueles JH, Cadenas‑Sanchez C, Ekelund U, et al. Accelerometer data at Birmingham, Birmingham, AL, USA. Department of Psychiatry, School collection and processing criteria to assess physical activity and other of Medicine, The University of Pittsburgh, Pittsburgh, PA, USA. outcomes: a systematic review and practical considerations. Sports Med. 2017;47(9):1821–45. https:// doi. org/ 10. 1007/ s40279‑ 017‑ 0716‑0. Received: 21 November 2022 Accepted: 2 February 2023 19. Ancoli‑Israel S, Martin JL, Blackwell T, et al. The SBSM guide to actigra‑ phy monitoring: clinical and research applications. Behav Sleep Med. 2015;13(Suppl 1):S4–38. https:// doi. org/ 10. 1080/ 15402 002. 2015. 10463 56. 20. Liu F, Wanigatunga AA, Schrack JA. Assessment of physical activity in adults using wrist accelerometers. Epidemiol Rev. 2021;43(1):65–93. References https:// doi. org/ 10. 1093/ epirev/ mxab0 04. 1. Rosenberger ME, Fulton JE, Buman MP, et al. The 24‑hour activity 21. De Craemer M, Verbestel V. Comparison of outcomes derived from the cycle: a new paradigm for physical activity. Med Sci Sports Exerc. actigraph gt3x+ and the axivity ax3 accelerometer to objectively meas‑ 2019;51(3):454–64. https:// doi. org/ 10. 1249/ MSS. 00000 00000 001811. ure 24‑hour movement behaviors in adults: a cross‑sectional study. Int J 2. Pedišić Ž, Dumuid D, Olds TS. Integrating sleep, sedentary behav‑ Environ Res Public Health. 2022;19(1):271. https:// doi. org/ 10. 3390/ ijerp iour, and physical activity research in the emerging field of time ‑use h1901 0271. epidemiology: definitions, concepts, statistical methods, theoretical 22. National Center for Health Statistics (NCHS). National health and nutrition framework, and future directions. Kinesiology. 2017;49(2):252–69. examination survey: 2011–2012 data documentation, codebook, and 3. Falck RS, Davis JC, Khan KM, et al. A Wrinkle in measuring time use for frequencies physical activity monitor—day (PAXDAY_G). Hyattsville: Cent‑ cognitive health: how should we measure physical activity, sedentary ers for Disease Control and Prevention; 2020. behaviour and sleep? Am J Lifestyle Med. 2021. https:// doi. org/ 10. 23. Sowers MF, Crawford SL, Sternfeld B, et al. SWAN: a multicenter, multi‑ 1177/ 15598 27621 10314 95. ethnic, community‑based cohort study of women and the menopausal 4. U.S. Department of Health and Human Services. Physical activity transition. In: Lobo R, Marcus R, Kelsey J, editors., et al., Menopause: biol‑ guidelines for Americans. 2nd ed. Washington: U.S. Department of ogy and pathobiology. San Diego: Academic Press; 2000. p. 175–88. Health and Human Services; 2018. Dooley et al. Journal of Activity, Sedentary and Sleep Behaviors (2023) 2:8 Page 12 of 12 24. Choi L, Beck C, Liu Z, et al. Physical activity: process accelerometer data 44. Rosenberger ME, Haskell WL, Albinali F, et al. Estimating activity and for physical activity measurement. R package version 0.2–4. 2021. https:// sedentary behavior from an accelerometer on the hip or wrist. Med Sci CRAN.R‑ proje ct. org/ packa ge= Physi calAc tivity. Sports Exerc. 2013;45(5):964–75. https:// doi. org/ 10. 1249/ MSS. 0b013 25. Evenson KR, Wen F, Herring AH, et al. Calibrating physical activity intensity e3182 7f0d9c. for hip‑ worn accelerometry in women age 60 to 91 years: the women’s 45. Littner M, Kushida CA, Anderson WM, et al. Practice parameters for the health initiative OPACH calibration study. Prev Med Rep. 2015;2:750–6. role of actigraphy in the study of sleep and circadian rhythms: an update https:// doi. org/ 10. 1016/j. pmedr. 2015. 08. 021. for 2002. Sleep. 2003;26(3):337–41. https:// doi. org/ 10. 1093/ sleep/ 26.3. 26. Stewart A, Sternfeld B, Lange‑Maia BS, et al. Reported and device ‑based 337. physical activity by race/ethnic groups in young‑ old women. J Meas Phys 46. Lehrer HM, Yao Z, Krafty RT, et al. Comparing polysomnography, actig‑ Behav. 2020;3(2):118–27. https:// doi. org/ 10. 1123/ jmpb. 2019‑ 0062. raphy, and sleep diary in the home environment: the study of women’s 27. Kushida CA, Chang A, Gadkary C, et al. Comparison of actigraphic, health across the nation (SWAN) sleep study. Sleep Adv. 2022. https:// doi. polysomnographic, and subjective assessment of sleep parameters in org/ 10. 1093/ sleep advan ces/ zpac0 01. sleep‑ disordered patients. Sleep Med. 2001;2(5):389–96. https:// doi. org/ 47. Full KM, Kerr J, Grandner MA, et al. Validation of a physical activity accel‑ 10. 1016/ s1389‑ 9457(00) 00098‑8. erometer device worn on the hip and wrist against polysomnography. 28. LeCun Y, Bengio Y. Convolutional networks for images, speech, and time Sleep Health. 2018;4(2):209–16. https:// doi. org/ 10. 1016/j. sleh. 2017. 12. series. In: Arbib MA, editor. The handbook of brain theory and neural 007. networks. Cambridge: The MIT Press; 1995. p. 255–8. 48. Lambiase MJ, Gabriel KP, Chang YF, et al. Utility of actiwatch sleep monitor 29. Granovsky L, Shalev G, Yacovzada N, et al. Actigraphy‑based sleep/wake to assess waking movement behavior in older women. Med Sci Sports pattern detection using convolutionalneural networks. arXiv preprint Exerc. 2014;46(12):2301–7. https:// doi. org/ 10. 1249/ mss. 00000 00000 arXiv: 1802. 07945. 2018 Feb 22.Access date: July 19, 2021. 000361. 30. Palotti J, Mall R, Aupetit M, et al. Benchmark on a large cohort for sleep‑ 49. Nguyen P, Le LK‑D, Ananthapavan J, et al. Economics of sedentary behav‑ wake classification with machine learning techniques. NPJ Digit Med. iour: A systematic review of cost of illness, cost‑ effectiveness, and return 2019;2(1):50. https:// doi. org/ 10. 1038/ s41746‑ 019‑ 0126‑9. on investment studies. Prev Med. 2022;156: 106964. https:// doi. org/ 10. 31. Greenwood‑Hickman MA, Nakandala S, Jankowska MM, et al. The CNN 1016/j. ypmed. 2022. 106964. hip accelerometer posture (CHAP) method for classifying sitting pat‑ 50. Yang Y, Shin JC, Li D, et al. Sedentary behavior and sleep problems: a terns from hip accelerometers: a validation study. Med Sci Sports Exerc. systematic review and meta‑analysis. Int J Behav Med. 2017;24(4):481–92. 2021;53(11):2445–54. https:// doi. org/ 10. 1249/ mss. 00000 00000 002705.https:// doi. org/ 10. 1007/ s12529‑ 016‑ 9609‑0. 32. Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3(1):32–5. 51. Micklesfield LK, Westgate K, Smith A, et al. Physical activity behaviors of 33. Davis J, Goadrich M. The relationship between Precision‑Recall and ROC a middle‑age south african cohort as determined by integrated hip and curves. In: Proceedings of the 23rd international conference on machine thigh accelerometry. Med Sci Sports Exerc. 2022;54(9):1493–505. https:// learning; 2006.doi. org/ 10. 1249/ MSS. 00000 00000 002940. 34. Janssen I, Clarke AE, Carson V, et al. A systematic review of compositional 52. Bassett DR, John D, Conger SA, et al. Detection of lying down, sitting, data analysis studies examining associations between sleep, sedentary standing, and stepping using two activPAL monitors. Med Sci Sports behaviour, and physical activity with health outcomes in adults. Appl Exerc. 2014;46(10):2025–9. https:// doi. org/ 10. 1249/ MSS. 00000 00000 Physiol Nutr Metab. 2020;45(10):S248–57. https:// doi. org/ 10. 1139/ 000326. apnm‑ 2020‑ 0160. 53. Dumuid D, Pedišić Ž, Palarea‑Albaladejo J, et al. Compositional data analy‑ 35. Dumuid D, Stanford TE, Martin‑Fernández J‑A, et al. Compositional data sis in time‑use epidemiology: what, why, how. Int J Environ Res Public analysis for physical activity, sedentary time and sleep research. Stat Health. 2020;17(7):2220. https:// doi. org/ 10. 3390/ ijerp h1707 2220. Methods Med Res. 2018;27(12):3726–38. https:// doi. org/ 10. 1177/ 09622 54. Tremblay MS, Aubert S, Barnes JD, et al. Sedentary behavior research 80217 710835. network (SBRN) ‑ terminology consensus project process and out ‑ 36. Chastin SF, Palarea‑Albaladejo J, Dontje ML, et al. Combined effects of come. Int J Behav Nutr Phys Act. 2017;14(1):75. https:// doi. org/ 10. 1186/ time spent in physical activity, sedentary behaviors and sleep on obesity s12966‑ 017‑ 0525‑8. and cardio‑metabolic health markers: a novel compositional data analysis 55. Bubu OM, Brannick M, Mortimer J, et al. Sleep, cognitive impairment, and approach. PLoS ONE. 2015;10(10): e0139984. https:// doi. org/ 10. 1371/ alzheimer’s disease: a systematic review and meta‑analysis. Sleep. 2016. journ al. pone. 01399 84.https:// doi. org/ 10. 1093/ sleep/ zsw032. 37. World Health Organization. Guidelines on physical activity, sedentary 56. Barone Gibbs B, Kline CE. When does sedentary behavior become sleep? behaviour and sleep for children under 5 years of age. Geneva: World A proposed framework for classifying activity during sleep‑ wake transi‑ Health Organization; 2019. tions. Int J Behav Nutr Phys Act. 2018;15(1):81. https:// doi. org/ 10. 1186/ 38. Okely AD, Ghersi D, Hesketh KD, et al. A collaborative approach to adopt‑s12966‑ 018‑ 0712‑2. ing/adapting guidelines—the Australian 24‑hour movement guidelines for the early years (Birth to 5 years): an integration of physical activity, Publisher’s Note sedentary behavior, and sleep. BMC Public Health. 2017;17(Suppl 5):869. Springer Nature remains neutral with regard to jurisdictional claims in pub‑ https:// doi. org/ 10. 1186/ s12889‑ 017‑ 4867‑6. lished maps and institutional affiliations. 39. Ross R, Chaput JP, Giangregorio LM, et al. Canadian 24‑hour movement guidelines for adults aged 18–64 years and adults aged 65 years or older: an integration of physical activity, sedentary behaviour, and sleep. Appl Physiol Nutr Metab. 2020;45(10):S57‑s102. https:// doi. org/ 10. 1139/ Re Read ady y to to submit y submit your our re researc search h ? Choose BMC and benefit fr ? Choose BMC and benefit from om: : apnm‑ 2020‑ 0467. 40. Tremblay MS, Carson V, Chaput JP, et al. Canadian 24‑hour movement fast, convenient online submission guidelines for children and youth: an integration of physical activity, sed‑ thorough peer review by experienced researchers in your field entary behaviour, and sleep. Appl Physiol Nutr Metab. 2016;41(6 Suppl 3):S311–27. https:// doi. org/ 10. 1139/ apnm‑ 2016‑ 0151. rapid publication on acceptance 41. Ministry of Health. Sit less, move more, sleep well: active play guidelines support for research data, including large and complex data types for under‑fives. Wellington, New Zealand. 2017. • gold Open Access which fosters wider collaboration and increased citations 42. Draper CE, Tomaz SA, Biersteker L, et al. The South African 24‑hour move ‑ ment guidelines for birth to 5 years: an integration of physical activity, sit‑ maximum visibility for your research: over 100M website views per year ting behavior, screen time, and sleep. J Phys Act Health. 2020;17(1):109– 19. https:// doi. org/ 10. 1123/ jpah. 2019‑ 0187. At BMC, research is always in progress. 43. Cleland I, Kikhia B, Nugent C, et al. Optimal placement of accelerometers Learn more biomedcentral.com/submissions for the detection of everyday activities. Sensors. 2013;13(7):9183–200. https:// doi. org/ 10. 3390/ s1307 09183.

Journal

Journal of Activity Sedentary and Sleep Behaviors – Springer Journals

Published: Apr 5, 2023

Keywords: Actigraphy; Accelerometer; Sleep; Physical activity; Harmonization; Machine learning; 24-h activity; Time-use epidemiology

There are no references for this article.