Detecting Sleep and Nonwear in 24-h Wrist Accelerometer Data from the National Health and Nutrition Examination Survey

Abstract

Estimating physical activity, sedentary behavior, and sleep from wrist-worn accelerometer data requires reliable detection of sensor nonwear and sensor wear during both sleep and wake. Purpose: This study aimed to develop an algorithm that simultaneously identifies sensor wake-wear, sleep-wear, and nonwear in 24-h wrist accelerometer data collected with or without filtering. Methods: Using sensor data labeled with polysomnography (n = 21) and directly observed wake-wear data (n = 31) from healthy adults, and nonwear data from sensors left at various locations in a home (n = 20), we developed an algorithm to detect nonwear, sleep-wear, and wake-wear for "idle sleep mode" (ISM) filtered data collected in the 2011-2014 National Health and Nutrition Examination Survey. The algorithm was then extended to process original raw data collected from devices without ISM filtering. Both algorithms were further validated using a polysomnography-based sleep and wake-wear data set (n = 22) and diary-based wake-wear and nonwear labels from healthy adults (n = 23). Classification performance (F1 scores) was compared with four alternative approaches. Results: The F1 score of the ISM-based algorithm on the training data set using leave-one-subject-out cross-validation was 0.95 ± 0.13. Validation on the two independent data sets yielded F1 scores of 0.84 ± 0.60 for the data set with sleep-wear and wake-wear and 0.94 ± 0.04 for the data set with wake-wear and nonwear. The F1 score when using original, raw data was 0.96 ± 0.08 for the training data sets and 0.86 ± 0.18 and 0.97 ± 0.04 for the two independent validation data sets. The algorithm performed comparably or better than the alternative approaches on the data sets. Conclusions: A novel machine-learning algorithm was designed to recognize wake-wear, sleep-wear, and nonwear in 24-h wrist-worn accelerometer data that are applicable for ISM-filtered data or original raw data.

Does Wrist-Worn Accelerometer Wear Compliance Wane over a Free-Living Assessment Period? An NHANES Analysis

“…Although the SWaN algorithm developers have demonstrated that their algorithm performs similarly to or better than several popular nonwear and sleep detection methods, both training and validation datasets were comprised of several disparate datasets, each with small sample sizes (<31 participants) of healthy, young participants (mean ages, 18.8–23.8 yr) (24). Additionally, data from a variety of activity monitors (Axivity, ActiGraph GT9X and GT3X+) were used, and the target classes within the datasets were unbalanced and blended from multiple sources to achieve a training dataset with all three labeled behavior classes (sleep wear, wake wear, and nonwear) (24). Thus, additional independent validation of the SWaN algorithm against ground-truth observation of all classified behaviors (wake wear, sleep wear, and nonwear) in a single dataset collected in a broader population would likely enable a better definition of confidence limits surrounding our observations of nonwear and wear fatigue.…”

“…A three-step, open-source algorithm (24) was used to classify each minute of PAM data as nonwear, sleep wear, wake wear, or unknown. The algorithmically scored wear classifications were applied before the NCHS data release, with greater detail on how the algorithm handled ISM and made predictions of the target classes provided elsewhere (24).…”

Companion: A Pilot Randomized Clinical Trial to Test an Integrated Two-Way Communication and Near-Real-Time Sensing System for Detecting and Modifying Daily Inactivity among Adults >60 Years—Design and Protocol

“…These devices will be worn on the non-dominant wrist. Wrist sensor data will be processed using Monitor-Independent Movement Summary unit (MIMS) to yield overall movement volume estimations [ 40 ] and machine learning algorithms to yield sensor wear, sleep, and wake behavior characterization (i.e., sedentary, ambulation, and upright behaviors) [ 41 , 42 , 43 , 44 ] during the 16-week intervention period. (ii) Health outcomes : (a) anthropometrics (height, weight, and waist/hip circumference); (b) body composition (via bioelectrical impedance) [ 45 ]; (c) seated resting heart rate and blood pressure; (d) fasting blood biomarkers (blood glucose and lipid profile); (e) cognition (via The NIH Toolbox Cognition Battery) [ 46 ]; (f) physical function (via The Short Physical Performance Battery) [ 47 , 48 ]; (g) upper and lower body muscle endurance (via maximal push-up and squat tests, respectively) [ 49 ]; (h) aerobic fitness (via the six-minute walk test) [ 50 , 51 ]; and (i) other questionnaire-based health outcomes (see supplementary study protocol ).…”

“…(iii) Motion data will be extracted from Centrepoint and processed on our backend located on Northeastern University’s high-performance computing cluster. This backend includes signal visualization software (i.e., Signaligner Pro v2.3.10-beta; Signaligner.org) [ 42 ], a signal quality control (QC) algorithm to identify anomalous sensor signals [ 75 ] that are unrepresentative of human movement, motion summarization algorithms for total activity volume (i.e., MIMS) [ 40 ], and activity classification using machine learning algorithms: (a) SWaN (sleep, wear, and non-wear) [ 44 ] and (b) MUSS (i.e., multi-site sensing for activity recognition: lying, sitting movement, non-wear, ambulation, and upright movement) [ 41 , 43 ]. The three algorithmic outputs are up sampled to an event resolution of one second and paired with a composite SWaN plus MUSS prediction ( Figure 3 ).…”

“…Similar and consecutive one second events with the same neighboring predictions and timestamps are combined into single events, resulting in continuously labeled activity classification events of variable duration. Combining the outputs of SWaN and MUSS into a composite prediction will allow us to retain MUSS’ granularity (ambulation vs. sitting movement vs. upright movement) during wear hours, and SWaN’s more accurate and sophisticated classification of sleep vs. non-wear [ 41 , 43 , 44 ]. (iv) Processed motion summary and raw sensor data will be visualized by a health coach using the backend interface to temporally display the data with overlaid automatically processed behavior labels.…”