Improving Hip-Worn Accelerometer Estimates of Sitting Using Machine Learning Methods

An active sitting chair can increase energy expenditure while performing standardized data entry work

BACKGROUND

Active sitting chairs have been proposed as an effective approach for reducing sedentary behaviour in the workplace.

OBJECTIVE

This cross-sectional study evaluated how an active sitting chair altered energy expenditure compared to a traditional office chair during seated computer work.

METHODS

Sixteen participants (8M/8F) completed two 20-min sessions of seated standardized computer work in an active sitting chair, with a multiaxial rotating seat pan, and traditional office chair. Metabolic and ventilatory variables were collected with a customized metabolic cart and cardiac variables were collected by a Hexoskin© shirt. Average ventilatory, metabolic, and cardiac variables from the last 15-min of each block were compared between chairs and sexes.

RESULTS

Statistically significant increases in oxygen uptake (V˙O2) emerged in active sitting (0.02 L/min; 7.6%), and ultimately led to a 1.5 kcal increase in energy expenditure compared to traditional sitting. Proportional and significant changes in minute ventilation (V˙E; + 0.9 L/min), heart rate (HR; + 5.8 bpm), and heart rate variability (HRV; -0.05 s) occurred, which further support the greater metabolic demand in active sitting.

CONCLUSIONS

A 1.5 kcal per 15-min increase in energy expenditure translates to 6 kcal/hour and 48 kcal/day. Compared to other literature, this change is similar to caloric expenditure when climbing three to six flights of stairs and when using alternative workstation designs (e.g., standing or sitting on a stability ball). An active sitting chair with a multiaxial rotating seat pan and no back support, appears to be a good alternative for increasing energy expenditure at a workstation.

A week-long field study of seated pelvis and lumbar spine kinematics during office work

The study objective was to quantify "natural" seated pelvis and lumbar spine kinematics over multiple days of work at individuals' workstations. Twenty participants completed five days of their usual office work while seated time was characterized from a thigh-worn activity monitor. Seated pelvic tilt and lumbar spine flexion-extension were measured from tri-axial accelerometers. Seated time accounted for approximately 90% of participants' workdays. Sitting was characterized by posterior pelvic tilt and lumbar flexion (43-79% of maximum flexion) with an average of 9 shifts and 13 fidgets every 15 min. No significant differences emerged by sex or between days indicating that a single representative day can capture baseline sitting responses in the field. Average field kinematics tended to agree with the laboratory-collected kinematics, but postural variability was larger in the field. These kinematic values could be useful for designing interventions aimed at reducing spine flexion and increasing spine movement in occupational sitting.

Comparative Analysis of Force-Sensitive Resistors and Triaxial Accelerometers for Sitting Posture Classification

Sedentary behaviors, including poor postures, are significantly detrimental to health, particularly for individuals losing motion ability. This study presents a posture detection system utilizing four force-sensitive resistors (FSRs) and two triaxial accelerometers selected after rigorous assessment for consistency and linearity. We compared various machine learning algorithms based on classification accuracy and computational efficiency. The k-nearest neighbor (KNN) algorithm demonstrated superior performance over Decision Tree, Discriminant Analysis, Naive Bayes, and Support Vector Machine (SVM). Further analysis of KNN hyperparameters revealed that the city block metric with K = 3 yielded optimal classification results. Triaxial accelerometers exhibited higher accuracy in both training (99.4%) and testing (99.0%) phases compared to FSRs (96.6% and 95.4%, respectively), with slightly reduced processing times (0.83 s vs. 0.85 s for training; 0.51 s vs. 0.54 s for testing). These findings suggest that, apart from being cost-effective and compact, triaxial accelerometers are more effective than FSRs for posture detection.

Application of wearables for remote monitoring of oncology patients: A scoping review

Objective

This review aims to systematically map and categorize the current state of wearable applications among oncology patients and to identify determinants impeding clinical implementation.

Methods

A Medline, Embase and clinicaltrials.gov search identified journal articles, conference abstracts, letters, reports, dissertations and registered studies on the use of wearables in patients with malignancies published up to 10 November 2021.

Results

Of 2509 records identified, 112 met the eligibility criteria. Of these, 9.8% (11/112) were RCTs and 47.3% (53/112) of publications were observational. Wearables were investigated pre-treatment (2.7%; 3/112), during treatment (34.8%; 39/112), post-treatment (17.9%; 20/112), in survivors (27.7%; 31/112) and in non-specified or multiple treatment phases (17.0%; 19/112). Medical-grade wearables were applied in 22.3% (25/112) of publications. Primary objectives ranged from technical feasibility (8.0%; 9/112), user feasibility (42.9%; 48/112) and correlational analysis (40.2%; 45/112) to outcome change analysis (8.9%; 10/112). Outcome change was mostly investigated regarding physical activity improvement (80.0%; 8/10). Most publications (42.9%; 48/112) and registered studies (39.3%; 24/61) featured multiple cancer types, with breast cancer as the most prevalent specific type (22.3% in publications, 16.4% in registered studies).

Conclusions

Most studies among oncology patients using wearables are focused on assessing the user feasibility of consumer-grade wearables, whereas rates of RCTs assessing clinical efficacy are low. Substantial improvements in clinically relevant endpoints by the use of wearables, such as morbidity and mortality are yet to be demonstrated.

Measuring nurses' on-shift physical activity and sedentary time by accelerometry or heart rate monitoring: a descriptive case study illustrating the importance of context

BACKGROUND

There is debate whether nurses are active enough stemming from differences in measurement tools, clinical contexts, and nursing tasks. A descriptive case study concerning the use of device-based measures in combination with direct observation is presented to examine the effect of the nursing context and the discrepancies between different measurement tools for identifying nurses' on-shift activity levels.

METHODS

Data were collected across seven shifts in medical and surgical wards. Nurses' activity was assessed using accelerometry and heart rate monitoring, in addition to direct observation. Data graphs were plotted for each shift and measurement device, with direct observations used to contextualise the data and identify discrepancies.

RESULTS

Higher activity levels were recorded on-shift through heart rate monitoring (87%) compared to accelerometry (27%). This pattern was also observed specifically on early, late, and medical ward shifts. Data discrepancies between the two devices stemmed from the shift and (or) ward type, highlighting the importance of understanding the context of nursing duties when assessing nurses' activity levels.

CONCLUSIONS

It is also vital that researchers, policymakers, and practitioners consider how they will measure nurses' occupational physical activity, which consequently will influence outcomes, and therefore, decisions around the need (or not) for intervention.

A Standardised Core Outcome Set for Measurement and Reporting Sedentary Behaviour Interventional Research: The CROSBI Consensus Study

Heterogeneity of descriptors and outcomes measured and reported in sedentary behaviour (SB) research hinder the meta-analysis of data and accumulation of evidence. The objective of the Core Research Outcomes for Sedentary Behaviour Interventions (CROSBI) consensus study was to identify and validate, a core outcome set (COS) to report (what, how, when to measure) in interventional sedentary behaviour studies. Outcomes, extracted from a systematic literature review, were categorized into domains and data items (COS v0.0). International experts (n = 5) provided feedback and identified additional items, which were incorporated into COS v0.1. A two round online Delphi survey was conducted to seek consensus from a wider stakeholder group and outcomes that achieved consensus in the second round COS (v0.2), were ratified by the expert panel. The final COS (v1.0) contains 53 data items across 12 domains, relating to demographics, device details, wear-time criteria, wear-time measures, posture-related measures, sedentary breaks, sedentary bouts and physical activity. Notably, results indicate that sedentary behaviour outcomes should be measured by devices that include an inclinometry or postural function. The proposed standardised COS is available openly to enhance the accumulation of pooled evidence in future sedentary behaviour intervention research and practice.

Emerging Artificial Intelligence-Empowered mHealth: Scoping Review

BACKGROUND

Artificial intelligence (AI) has revolutionized health care delivery in recent years. There is an increase in research for advanced AI techniques, such as deep learning, to build predictive models for the early detection of diseases. Such predictive models leverage mobile health (mHealth) data from wearable sensors and smartphones to discover novel ways for detecting and managing chronic diseases and mental health conditions.

OBJECTIVE

Currently, little is known about the use of AI-powered mHealth (AIM) settings. Therefore, this scoping review aims to map current research on the emerging use of AIM for managing diseases and promoting health. Our objective is to synthesize research in AIM models that have increasingly been used for health care delivery in the last 2 years.

METHODS

Using Arksey and O'Malley's 5-point framework for conducting scoping reviews, we reviewed AIM literature from the past 2 years in the fields of biomedical technology, AI, and information systems. We searched 3 databases, PubsOnline at INFORMS, e-journal archive at MIS Quarterly, and Association for Computing Machinery (ACM) Digital Library using keywords such as "mobile healthcare," "wearable medical sensors," "smartphones", and "AI." We included AIM articles and excluded technical articles focused only on AI models. We also used the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) technique for identifying articles that represent a comprehensive view of current research in the AIM domain.

RESULTS

We screened 108 articles focusing on developing AIM models for ensuring better health care delivery, detecting diseases early, and diagnosing chronic health conditions, and 37 articles were eligible for inclusion, with 31 of the 37 articles being published last year (76%). Of the included articles, 9 studied AI models to detect serious mental health issues, such as depression and suicidal tendencies, and chronic health conditions, such as sleep apnea and diabetes. Several articles discussed the application of AIM models for remote patient monitoring and disease management. The considered primary health concerns belonged to 3 categories: mental health, physical health, and health promotion and wellness. Moreover, 14 of the 37 articles used AIM applications to research physical health, representing 38% of the total studies. Finally, 28 out of the 37 (76%) studies used proprietary data sets rather than public data sets. We found a lack of research in addressing chronic mental health issues and a lack of publicly available data sets for AIM research.

CONCLUSIONS

The application of AIM models for disease detection and management is a growing research domain. These models provide accurate predictions for enabling preventive care on a broader scale in the health care domain. Given the ever-increasing need for remote disease management during the pandemic, recent AI techniques, such as federated learning and explainable AI, can act as a catalyst for increasing the adoption of AIM and enabling secure data sharing across the health care industry.

Quality Evaluation of Free-living Validation Studies for the Assessment of 24-Hour Physical Behavior in Adults via Wearables: Systematic Review

Background

Wearable technology is a leading fitness trend in the growing commercial industry and an established method for collecting 24-hour physical behavior data in research studies. High-quality free-living validation studies are required to enable both researchers and consumers to make guided decisions on which study to rely on and which device to use. However, reviews focusing on the quality of free-living validation studies in adults are lacking.

Objective

This study aimed to raise researchers’ and consumers’ attention to the quality of published validation protocols while aiming to identify and compare specific consistencies or inconsistencies between protocols. We aimed to provide a comprehensive and historical overview of which wearable devices have been validated for which purpose and whether they show promise for use in further studies.

Methods

Peer-reviewed validation studies from electronic databases, as well as backward and forward citation searches (1970 to July 2021), with the following, required indicators were included: protocol must include real-life conditions, outcome must belong to one dimension of the 24-hour physical behavior construct (intensity, posture or activity type, and biological state), the protocol must include a criterion measure, and study results must be published in English-language journals. The risk of bias was evaluated using the Quality Assessment of Diagnostic Accuracy Studies-2 tool with 9 questions separated into 4 domains (patient selection or study design, index measure, criterion measure, and flow and time).

Results

Of the 13,285 unique search results, 222 (1.67%) articles were included. Most studies (153/237, 64.6%) validated an intensity measure outcome such as energy expenditure. However, only 19.8% (47/237) validated biological state and 15.6% (37/237) validated posture or activity-type outcomes. Across all studies, 163 different wearables were identified. Of these, 58.9% (96/163) were validated only once. ActiGraph GT3X/GT3X+ (36/163, 22.1%), Fitbit Flex (20/163, 12.3%), and ActivPAL (12/163, 7.4%) were used most often in the included studies. The percentage of participants meeting the quality criteria ranged from 38.8% (92/237) to 92.4% (219/237). On the basis of our classification tree to evaluate the overall study quality, 4.6% (11/237) of studies were classified as low risk. Furthermore, 16% (38/237) of studies were classified as having some concerns, and 72.9% (173/237) of studies were classified as high risk.

Conclusions

Overall, free-living validation studies of wearables are characterized by low methodological quality, large variability in design, and focus on intensity. Future research should strongly aim at biological state and posture or activity outcomes and strive for standardized protocols embedded in a validation framework. Standardized protocols for free-living validation embedded in a framework are urgently needed to inform and guide stakeholders (eg, manufacturers, scientists, and consumers) in selecting wearables for self-tracking purposes, applying wearables in health studies, and fostering innovation to achieve improved validity.

Accelerometer-Derived Daily Life Movement Classified by Machine Learning and Incidence of Cardiovascular Disease in Older Women: The OPACH Study

Background Current physical activity guidelines focus on volume and intensity for CVD prevention rather than common behaviors responsible for movement, including those for daily living activities. We examined the associations of a machine-learned, accelerometer-measured behavior termed daily life movement (DLM) with incident CVD. Methods and Results Older women (n=5416; mean age, 79±7 years; 33% Black, 17% Hispanic) in the Women's Health Initiative OPACH (Objective Physical Activity and Cardiovascular Health) study without prior CVD wore ActiGraph GT3X+ accelerometers for up to 7 days from May 2012 to April 2014 and were followed for physician-adjudicated incident CVD through February 28th, 2020 (n=616 events). DLM was defined as standing and moving in a confined space such as performing housework or gardening. Cox models estimated hazard ratios (HR) and 95% CI, adjusting for age, race and ethnicity, education, alcohol use, smoking, multimorbidity, self-rated health, and physical function. Restricted cubic splines examined the linearity of the DLM-CVD dose-response association. We examined effect modification by age, body mass index, Reynolds Risk Score, and race and ethnicity. Adjusted HR (95% CIs) across DLM quartiles were: 1.00 (reference), 0.68 (0.55-0.84), 0.70 (0.56-0.87), and 0.57 (0.45-0.74); p-trend<0.001. The HR (95% CI) for each 1-hour increment in DLM was 0.86 (0.80-0.92) with evidence of a linear dose-response association (p non-linear>0.09). There was no evidence of effect modification by age, body mass index, Reynolds Risk Score, or race and ethnicity. Conclusions Higher DLM was independently associated with a lower risk of CVD in older women. Describing the beneficial associations of physical activity in terms of common behaviors could help older adults accumulate physical activity.

The CNN Hip Accelerometer Posture (CHAP) Method for Classifying Sitting Patterns from Hip Accelerometers: A Validation Study

INTRODUCTION

Sitting patterns predict several healthy aging outcomes. These patterns can potentially be measured using hip-worn accelerometers, but current methods are limited by an inability to detect postural transitions. To overcome these limitations, we developed the Convolutional Neural Network Hip Accelerometer Posture (CHAP) classification method.

METHODS

CHAP was developed on 709 older adults who wore an ActiGraph GT3X+ accelerometer on the hip, with ground-truth sit/stand labels derived from concurrently worn thigh-worn activPAL inclinometers for up to 7 d. The CHAP method was compared with traditional cut-point methods of sitting pattern classification as well as a previous machine-learned algorithm (two-level behavior classification).

RESULTS

For minute-level sitting versus nonsitting classification, CHAP performed better (93% agreement with activPAL) than did other methods (74%-83% agreement). CHAP also outperformed other methods in its sensitivity to detecting sit-to-stand transitions: cut-point (73%), TLBC (26%), and CHAP (83%). CHAP's positive predictive value of capturing sit-to-stand transitions was also superior to other methods: cut-point (30%), TLBC (71%), and CHAP (83%). Day-level sitting pattern metrics, such as mean sitting bout duration, derived from CHAP did not differ significantly from activPAL, whereas other methods did: activPAL (15.4 min of mean sitting bout duration), CHAP (15.7 min), cut-point (9.4 min), and TLBC (49.4 min).

CONCLUSION

CHAP was the most accurate method for classifying sit-to-stand transitions and sitting patterns from free-living hip-worn accelerometer data in older adults. This promotes enhanced analysis of older adult movement data, resulting in more accurate measures of sitting patterns and opening the door for large-scale cohort studies into the effects of sitting patterns on healthy aging outcomes.

The Prediction of Running Velocity during the 30-15 Intermittent Fitness Test Using Accelerometry-Derived Metrics and Physiological Parameters: A Machine Learning Approach

Measuring exercise variables is one of the most important points to consider to maximize physiological adaptations. High-intensity interval training (HIIT) is a useful method to improve both cardiovascular and neuromuscular performance. The 30–15_IFT is a field test reflecting the effort elicited by HIIT, and the final velocity reached in the test is used to set the intensity of HIIT during the training session. In order to have a valid measure of the velocity during training, devices such as GPS can be used. However, in several situations (e.g., indoor setting), such devices do not provide reliable measures. The aim of the study was to predict exact running velocity during the 30–15_IFT using accelerometry-derived metrics (i.e., Player Load and Average Net Force) and heart rate (HR) through a machine learning (ML) approach (i.e., Support Vector Machine) with a leave-one-subject-out cross-validation. The SVM approach showed the highest performance to predict running velocity (r = 0.91) when compared to univariate approaches using PL (r = 0.62), AvNetForce (r = 0.73) and HR only (r = 0.87). In conclusion, the presented multivariate ML approach is able to predict running velocity better than univariate ones, and the model is generalizable across subjects.

Machine-learning models for activity class prediction: A comparative study of feature selection and classification algorithms

PURPOSE

Machine-learning (ML) approaches have been repeatedly coupled with raw accelerometry to classify physical activity classes, but the features required to optimize their predictive performance are still unknown. Our aim was to identify appropriate combination of feature subsets and prediction algorithms for activity class prediction from hip-based raw acceleration data.

METHODS

The hip-based raw acceleration data collected from 27 participants was split into training (70 %) and validation (30 %) subsets. A total of 206 time- (TD) and frequencydomain (FD) features were extracted from 6-second non-overlapping windows of the signal. Feature selection was done using seven filter-based, two wrapper-based, and one embedded algorithm, and classification was performed with artificial neural network (ANN), support vector machine (SVM), and random forest (RF). For every combination between the feature selection method and the classifiers, the most appropriate feature subsets were found and used for model training within the training set. These models were then validated with the left-out validation set.

RESULTS

The appropriate number of features for the ANN, SVM, and RF ranged from 20 to 45. Overall, the accuracy of all the three classifiers was higher when trained with feature subsets generated using filter-based methods compared with when they were trained with wrapper-based methods (range: 78.1 %-88 % vs. 66 %-83.5 %). TD features that reflect how signals vary around the mean, how they differ with one another, and how much and how often they change were more frequently selected via the feature selection methods.

CONCLUSIONS

A subset of TD features from raw accelerometry could be sufficient for ML-based activity classification if properly selected from different axes.

Application of Convolutional Neural Network Algorithms for Advancing Sedentary and Activity Bout Classification

Background

Machine learning has been used for classification of physical behavior bouts from hip-worn accelerometers; however, this research has been limited due to the challenges of directly observing and coding human behavior "in the wild." Deep learning algorithms, such as convolutional neural networks (CNNs), may offer better representation of data than other machine learning algorithms without the need for engineered features and may be better suited to dealing with free-living data. The purpose of this study was to develop a modeling pipeline for evaluation of a CNN model on a free-living data set and compare CNN inputs and results with the commonly used machine learning random forest and logistic regression algorithms.

Method

Twenty-eight free-living women wore an ActiGraph GT3X+accelerometer on their right hip for 7 days. A concurrently worn thigh-mounted activPAL device captured ground truth activity labels. The authors evaluated logistic regression, random forest, and CNN models for classifying sitting, standing, and stepping bouts. The authors also assessed the benefit of performing feature engineering for this task.

Results

The CNN classifier performed best (average balanced accuracy for bout classification of sitting, standing, and stepping was 84%) compared with the other methods (56% for logistic regression and 76% for random forest), even without performing any feature engineering.

Conclusion

Using the recent advancements in deep neural networks, the authors showed that a CNN model can outperform other methods even without feature engineering. This has important implications for both the model's ability to deal with the complexity of free-living data and its potential transferability to new populations.

The Relationship of Accelerometer-Assessed Standing Time With and Without Ambulation and Mortality: The WHI OPACH Study

BACKGROUND

Self-reported time spent standing has been associated with lower risk of mortality. No previous studies have examined this association using device-measured standing.

METHOD

This was a prospective cohort study of 5878 older (median age = 80 years), racial/ethnically diverse, community-dwelling women in the WHI Objective Physical Activity and Cardiovascular Health Study (OPACH). Women wore accelerometers for 1 week and were followed for mortality. The study applied previously validated machine learning algorithms to ActiGraph GT3X+ accelerometer data to separately measure time spent standing with and without ambulation. Cox proportional hazards models were used to estimate mortality risk adjusting for potential confounders. Effect modification by age, body mass index, moderate-to-vigorous physical activity, sedentary time, physical functioning, and race/ethnicity was evaluated.

RESULTS

There were 691 deaths during 26 649 person-years of follow-up through March 31, 2018 (mean follow-up = 4.8 years). In fully adjusted models, all-cause mortality risk was lower among those with more standing without ambulation (quartile [Q] 4 vs Q1 HR = 0.63; 95% CI = 0.49-0.81, p-trend = .003) and more standing with ambulation (Q4 vs Q1 HR = 0.50; 95% CI = 0.35-0.71, p-trend < .001). Associations of standing with ambulation and mortality were stronger among women with above-median sedentary time (HR = 0.51; 95% CI = 0.38-0.68) compared to women with below-median sedentary time (HR = 0.80; 95% CI = 0.59-1.07; p-interaction = .02).

CONCLUSIONS

In this prospective study among older women, higher levels of accelerometer-measured standing were associated with lower risks of all-cause mortality. Standing is an achievable approach to interrupting prolonged sedentary time, and if not contraindicated, is a safe and feasible behavior that appears to benefit health in older ages.

Endothelial-derived cardiovascular disease-related microRNAs elevated with prolonged sitting pattern among postmenopausal women

Time spent sitting is positively correlated with endothelial dysfunction and cardiovascular disease risk. The underlying molecular mechanisms are unknown. MicroRNAs contained in extracellular vesicles (EVs) reflect cell/tissue status and mediate intercellular communication. We explored the association between sitting patterns and microRNAs isolated from endothelial cell (EC)-derived EVs. Using extant actigraphy based sitting behavior data on a cohort of 518 postmenopausal overweight/obese women, we grouped the woman as Interrupted Sitters (IS; N = 18) or Super Sitters (SS; N = 53) if they were in the shortest or longest sitting pattern quartile, respectively. The cargo microRNA in EC-EVs from the IS and SS women were compared. MicroRNA data were weighted by age, physical functioning, MVPA, device wear days, device wear time, waist circumference, and body mass index. Screening of CVD-related microRNAs demonstrated that miR-199a-5p, let-7d-5p, miR-140-5p, miR-142-3p, miR-133b level were significantly elevated in SS compared to IS groups. Group differences in let-7d-5p, miR-133b, and miR-142-3p were validated in expanded groups. Pathway enrichment analyses show that mucin-type O-glycan biosynthesis and cardiomyocyte adrenergic signaling (P < 0.001) are downstream of the three validated microRNAs. This proof-of-concept study supports the possibility that CVD-related microRNAs in EC-EVs may be molecular transducers of sitting pattern-associated CVD risk in overweight postmenopausal women.

How Accurate and Precise Can We Measure the Posture and the Energy Expenditure Component of Sedentary Behaviour with One Sensor?

Sedentary behaviour is an emergent public health topic, but there is still no method to simultaneously measure both components of sedentary behaviour-posture and energy expenditure-with one sensor. This study investigated the accuracy and precision of measuring sedentary time when combining the proprietary processing of a posture sensor (activPAL) with a new energy expenditure algorithm and the proprietary processing of a movement sensor (ActiGraph) with a published posture algorithm. One hundred office workers wore both sensors for an average of 7 days. The activPAL algorithm development used 38 and the subsequent independent method comparison 62 participants. The single sensor sedentary estimates were compared with Bland-Atman statistics to the Posture and Physical Activity Index, a combined measurement with both sensors. All single-sensor methods overestimated sedentary time. However, adding the algorithms reduced the overestimation from 129 to 21 (activPAL) and from 84 to 7 min a day (ActiGraph), with far narrower 95% limits of agreements. Thus, combining the proprietary data with the algorithms is an easy way to increase the accuracy and precision of the single sensor sedentary estimates and leads to sedentary estimates that are more precise at the individual level than those of the proprietary processing are at the group level.

Agreement of sedentary behaviour metrics derived from hip-worn and thigh-worn accelerometers among older adults: with implications for studying physical and cognitive health

Little is known about how sedentary behaviour (SB) metrics derived from hip-worn and thigh-worn accelerometers agree for older adults. Thigh-worn activPAL micro monitors were concurrently worn with hip-worn ActiGraph GT3X+ accelerometers (with SB measured using the 100 count-per-minute (cpm) cut-point; ActiGraph_100cpm) by 953 older adults (age 77±6.6, 54% women) for 4-to-7 days. Device agreement for sedentary time and 5 SB pattern metrics was assessed using mean error and correlations. Logistic regression tested associations with 4 health outcomes using standardized (i.e., z-scores) and unstandardized SB metrics. Mean errors (activPAL-ActiGraph_100cpm) and 95% limits of agreement were: sedentary time -54.7(-223.4,113.9) min/d; time in 30+ minute bouts 77.6(-74.8,230.1) min/d; mean bout duration 5.9(0.5,11.4) min; usual bout duration 15.2(0.4,30) min; breaks in sedentary time -35.4(-63.1,-7.6) breaks/d; and alpha -0.5(-0.6,-0.4). Respective Pearson correlations were: 0.66, 0.78, 0.73, 0.79, 0.51, 0.40. Concordance correlations were: 0.57, 0.67, 0.40, 0.50, 0.14, 0.02. The statistical significance and direction of associations was identical for ActiGraph_100cpm and activPAL metrics in 46 of 48 tests, though significant differences in the magnitude of odds ratios were observed among 9 of 24 tests for unstandardized and 2 of 24 for standardized SB metrics. Caution is needed when interpreting SB metrics and associations with health from ActiGraph_100cpm due to the tendency for it to overestimate breaks in sedentary time relative to activPAL. However, high correlations between activPAL and ActiGraph_100cpm measures and similar standardized associations with health outcomes suggest that studies using ActiGraph_100cpm are useful, though not ideal, for studying SB in older adults.

Concurrent and discriminant validity of ActiGraph waist and wrist cut-points to measure sedentary behaviour, activity level, and posture in office work

Background

Sedentary Behaviour (SB) gets an increasing attention from ergonomics and public health due to its associated detrimental health effects. A large number of studies record SB with ActiGraph counts-per-minute cut-points, but we still lack valid information about what the cut-points tell us about office work. This study therefore analysed the concurrent and discriminant validity of commonly used cut-points to measure SB, activity level, and posture.

Methods

Thirty office workers completed four office tasks at three workplaces (conventional chair, activity-promoting chair, and standing desk) while wearing two ActiGraphs (waist and wrist). Indirect calorimetry and prescribed posture served as reference criteria. Generalized Estimation Equations analysed workplace and task effects on the activity level and counts-per-minute, and kappa statistics and ROC curves analysed the cut-point validity.

Results

The activity-promoting chair (p < 0.001, ES ≥ 0.66) but not the standing desk (p = 1.0) increased the activity level, and both these workplaces increased the waist (p ≤ 0.003, ES ≥ 0.63) but not the wrist counts-per-minute (p = 0.74) compared to the conventional chair. The concurrent and discriminant validity was higher for activity level (kappa: 0.52–0.56 and 0.38–0.45, respectively) than for SB and posture (kappa ≤0.35 and ≤ 0.19, respectively). Furthermore, the discriminant validity for activity level was higher for task effects (kappa: 0.42–0.48) than for workplace effects (0.13–0.24).

Conclusions

ActiGraph counts-per-minute for waist and wrist placement were – independently of the chosen cut-point – a measure for activity level and not for SB or posture, and the cut-points performed better to detect task effects than workplace effects. Waist cut-points were most valid to measure the activity level in conventional seated office work, but they showed severe limitations for sit-stand desks. None of the placements was valid to detect the increased activity on the activity-promoting chair. Caution should therefore be paid when analysing the effect of workplace interventions on activity level with ActiGraph waist and wrist cut-points.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-021-10387-7.

Is Sitting Always Inactive and Standing Always Active? A Simultaneous Free-Living activPal and ActiGraph Analysis

Sedentary Behavior (SB), defined as sitting with minimal physical activity, is an emergent public health topic. However, the measurement of SB considers either posture (e.g., activPal) or physical activity (e.g., ActiGraph), and thus neglects either active sitting or inactive standing. The aim of this study was to determine the true amount of active sitting and inactive standing in daily life, and to analyze by how much these behaviors falsify the single sensors' sedentary estimates. Sedentary time of 100 office workers estimated with activPal and ActiGraph was therefore compared with Bland-Altman statistics to a combined sensor analysis, the posture and physical activity index (POPAI). POPAI classified each activPal sitting and standing event into inactive or active using the ActiGraph counts. Participants spent 45.0% [32.2%-59.1%] of the waking hours inactive sitting (equal to SB), 13.7% [7.8%-21.6%] active sitting, and 12.0% [5.7%-24.1%] inactive standing (mean [5th-95th percentile]). The activPal overestimated sedentary time by 30.3% [12.3%-48.4%] and the ActiGraph by 22.5% [3.2%-41.8%] (bias [95% limit-of-agreement]). The results showed that sitting is not always inactive, and standing is not always active. Caution should therefore be paid when interpreting the activPal (ignoring active sitting) and ActiGraph (ignoring inactive standing) measured time as SB.

Random forest algorithms for recognizing daily life activities using plantar pressure information: a smart-shoe study

Background

Wearable activity trackers are regarded as a new opportunity to deliver health promotion interventions. Indeed, while the prediction of active behaviors is currently primarily relying on the processing of accelerometer sensor data, the emergence of smart clothes with multi-sensing capacities is offering new possibilities. Algorithms able to process data from a variety of smart devices and classify daily life activities could therefore be of particular importance to achieve a more accurate evaluation of physical behaviors. This study aims to (1) develop an activity recognition algorithm based on the processing of plantar pressure information provided by a smart-shoe prototype and (2) to determine the optimal hardware and software configurations.

Method

Seventeen subjects wore a pair of smart-shoe prototypes composed of plantar pressure measurement insoles, and they performed the following nine activities: sitting, standing, walking on a flat surface, walking upstairs, walking downstairs, walking up a slope, running, cycling, and completing office work. The insole featured seven pressure sensors. For each activity, at least four minutes of plantar pressure data were collected. The plantar pressure data were cut in overlapping windows of different lengths and 167 features were extracted for each window. Data were split into training and test samples using a subject-wise assignment method. A random forest model was trained to recognize activity. The resulting activity recognition algorithms were evaluated on the test sample. A multi hold-out procedure allowed repeating the operation with 5 different assignments. The analytic conditions were modulated to test (1) different window lengths (1–60 seconds), (2) some selected sensor configurations and (3) different numbers of data features.

Results

A window length of 20 s was found to be optimum and therefore used for the rest of the analysis. Using all the sensors and all 167 features, the smart shoes predicted the activities with an average success of 89%. “Running” demonstrated the highest sensitivity (100%). “Walking up a slope” was linked with the lowest performance (63%), with the majority of the false negatives being “walking on a flat surface” and “walking upstairs.” Some 2- and 3-sensor configurations were linked with an average success rate of 87%. Reducing the number of features down to 20 does not alter significantly the performance of the algorithm.

Conclusion

High-performance human behavior recognition using plantar pressure data only is possible. In the future, smart-shoe devices could contribute to the evaluation of daily physical activities. Minimalist configurations integrating only a small number of sensors and computing a reduced number of selected features could maintain a satisfying performance. Future experiments must include a more heterogeneous population.

The Use of Inertial Measurement Units for the Study of Free Living Environment Activity Assessment: A Literature Review

This article presents an overview of fifty-eight articles dedicated to the evaluation of physical activity in free-living conditions using wearable motion sensors. This review provides a comprehensive summary of the technical aspects linked to sensors (types, number, body positions, and technical characteristics) as well as a deep discussion on the protocols implemented in free-living conditions (environment, duration, instructions, activities, and annotation). Finally, it presents a description and a comparison of the main algorithms and processing tools used for assessing physical activity from raw signals.

Sedentary Behavior and Public Health: Integrating the Evidence and Identifying Potential Solutions

In developed and developing countries, social, economic, and environmental transitions have led to physical inactivity and large amounts of time spent sitting. Research is now unraveling the adverse public health consequences of too much sitting. We describe improvements in device-based measurement that are providing new insights into sedentary behavior and health. We consider the implications of research linking evidence from epidemiology and behavioral science with mechanistic insights into the underlying biology of sitting time. Such evidence has led to new sedentary behavior guidelines and initiatives. We highlight ways that this emerging knowledge base can inform public health strategy: First, we consider epidemiologic and experimental evidence on the health consequences of sedentary behavior; second, we describe solutions-focused research from initiatives in workplaces and schools. To inform a broad public health strategy, researchers need to pursue evidence-informed collaborations with occupational health, education, and other sectors.

Application of Raw Accelerometer Data and Machine-Learning Techniques to Characterize Human Movement Behavior: A Systematic Scoping Review

BACKGROUND

Application of machine learning for classifying human behavior is increasingly common as access to raw accelerometer data improves. The aims of this scoping review are (1) to examine if machine-learning techniques can accurately identify human activity behaviors from raw accelerometer data and (2) to summarize the practical implications of these machine-learning techniques for future work.

METHODS

Keyword searches were performed in Scopus, Web of Science, and EBSCO databases in 2018. Studies that applied supervised machine-learning techniques to raw accelerometer data and estimated components of physical activity were included. Information on study characteristics, machine-learning techniques, and key study findings were extracted from included studies.

RESULTS

Of the 53 studies included in the review, 75% were published in the last 5 years. Most studies predicted postures and activity type, rather than intensity, and were conducted in controlled environments using 1 or 2 devices. The most common models were support vector machine, random forest, and artificial neural network. Overall, classification accuracy ranged from 62% to 99.8%, although nearly 80% of studies achieved an overall accuracy above 85%.

CONCLUSIONS

Machine-learning algorithms demonstrate good accuracy when predicting physical activity components; however, their application to free-living settings is currently uncertain.

Quantifying Physical Activity in Young Children Using a Three-Dimensional Camera

The purpose of this study was to determine the feasibility and validity of using three-dimensional (3D) video data and computer vision to estimate physical activity intensities in young children. Families with children (2–5-years-old) were invited to participate in semi-structured 20-minute play sessions that included a range of indoor play activities. During the play session, children’s physical activity (PA) was recorded using a 3D camera. PA video data were analyzed via direct observation, and 3D PA video data were processed and converted into triaxial PA accelerations using computer vision. PA video data from children (n = 10) were analyzed using direct observation as the ground truth, and the Receiver Operating Characteristic Area Under the Curve (AUC) was calculated in order to determine the classification accuracy of a Classification and Regression Tree (CART) algorithm for estimating PA intensity from video data. A CART algorithm accurately estimated the proportion of time that children spent sedentary (AUC = 0.89) in light PA (AUC = 0.87) and moderate-vigorous PA (AUC = 0.92) during the play session, and there were no significant differences (p > 0.05) between the directly observed and CART-determined proportions of time spent in each activity intensity. A computer vision algorithm and 3D camera can be used to estimate the proportion of time that children spend in all activity intensities indoors.

Total Sitting Time and Sitting Pattern in Postmenopausal Women Differ by Hispanic Ethnicity and are Associated With Cardiometabolic Risk Biomarkers

Background Sedentary behavior is pervasive, especially in older adults, and is associated with cardiometabolic disease and mortality. Relationships between cardiometabolic biomarkers and sitting time are unexplored in older women, as are possible ethnic differences. Methods and Results Ethnic differences in sitting behavior and associations with cardiometabolic risk were explored in overweight/obese postmenopausal women (n=518; mean±SD age 63±6 years; mean body mass index 31.4±4.8 kg/m2). Accelerometer data were processed using validated machine-learned algorithms to measure total daily sitting time and mean sitting bout duration (an indicator of sitting behavior pattern). Multivariable linear regression was used to compare sitting among Hispanic women (n=102) and non-Hispanic women (n=416) and tested associations with cardiometabolic risk biomarkers. Hispanic women sat, on average, 50.3 minutes less/day than non-Hispanic women (P<0.001) and had shorter (3.6 minutes less, P=0.02) mean sitting bout duration. Among all women, longer total sitting time was deleteriously associated with fasting insulin and triglyceride concentrations, insulin resistance, body mass index and waist circumference; longer mean sitting bout duration was deleteriously associated with fasting glucose and insulin concentrations, insulin resistance, body mass index and waist circumference. Exploratory interaction analysis showed that the association between mean sitting bout duration and fasting glucose concentration was significantly stronger among Hispanic women than non-Hispanic women (P-interaction=0.03). Conclusions Ethnic differences in 2 objectively measured parameters of sitting behavior, as well as detrimental associations between parameters and cardiometabolic biomarkers were observed in overweight/obese older women. The detrimental association between mean sitting bout duration and fasting glucose may be greater in Hispanic women than in non-Hispanic women. Corroboration in larger studies is warranted.

Sedentary Behavior and Chronic Disease: Mechanisms and Future Directions

Background: Recent updates to physical activity guidelines highlight the importance of reducing sedentary time. However, at present, only general recommendations are possible (ie, “Sit less, move more”). There remains a need to investigate the strength, temporality, specificity, and dose–response nature of sedentary behavior associations with chronic disease, along with potential underlying mechanisms. Methods: Stemming from a recent research workshop organized by the Sedentary Behavior Council themed “Sedentary behaviour mechanisms—biological and behavioural pathways linking sitting to adverse health outcomes,” this paper (1) discusses existing challenges and scientific discussions within this advancing area of science, (2) highlights and discusses emerging areas of interest, and (3) points to potential future directions. Results: A brief knowledge update is provided, reflecting upon current and evolving thinking/discussions, and the rapid accumulation of new evidence linking sedentary behavior to chronic disease. Research “action points” are made at the end of each section—spanning from measurement systems and analytic methods, genetic epidemiology, causal mediation, and experimental studies to biological and behavioral determinants and mechanisms. Conclusion: A better understanding of whether and how sedentary behavior is causally related to chronic disease will allow for more meaningful conclusions in the future and assist in refining clinical and public health policies/recommendations.

Sedentary Behavior and Prevalent Diabetes in 6,166 Older Women: The Objective Physical Activity and Cardiovascular Health Study

BACKGROUND

We examined associations of sedentary time and sedentary accumulation patterns (ie, how sedentary time is accumulated) with prevalent diabetes in an ethnically diverse cohort of older women.

METHODS

Community-dwelling women aged 63-99 (n = 6,116; median age = 79) wore ActiGraph GT3X+ accelerometers 24 h/day for up to 7 days from which we derived average daily sedentary time and three measures of sedentary accumulation patterns: breaks in sedentary time, usual sedentary bout duration, and alpha. Odds ratios (ORs) and 95% confidence intervals (CIs) for prevalent diabetes were estimated using multivariable logistic regression.

RESULTS

Twenty-one percent (n = 1,282) of participants had diabetes. Women in the highest quartile of sedentary time (≥10.3 h/day) had higher odds of diabetes (OR = 2.18; 95% CI = 1.77-2.70) than women in the lowest quartile (≤8.3 h/day). Prolonged accumulation patterns (ie, accumulating sedentary time in longer sedentary bouts) was associated with higher odds of diabetes than regularly interrupted patterns (comparing quartiles with the most vs least prolonged patterns: usual bout duration OR = 1.57, 95% CI = 1.28-1.92; alpha OR = 1.61, 95% CI = 1.32-1.97); however, there was no significant association for breaks in sedentary time (OR = 1.00, 95% CI = 0.82-1.20).

CONCLUSIONS

High levels of sedentary time and accumulating it in prolonged patterns were associated with increased odds of diabetes among older women.

Detecting prolonged sitting bouts with the ActiGraph GT3X

The ActiGraph has a high ability to measure physical activity; however, it lacks an accurate posture classification to measure sedentary behavior. The aim of the present study was to develop an ActiGraph (waist-worn, 30 Hz) posture classification to detect prolonged sitting bouts, and to compare the classification to proprietary ActiGraph data. The activPAL, a highly valid posture classification device, served as reference criterion. Both sensors were worn by 38 office workers over a median duration of 9 days. An automated feature selection extracted the relevant signal information for a minute-based posture classification. The machine learning algorithm with optimal feature number to predict the time in prolonged sitting bouts (≥5 and ≥10 minutes) was searched and compared to the activPAL using Bland-Altman statistics. The comparison included optimized and frequently used cut-points (100 and 150 counts per minute (cpm), with and without low-frequency-extension (LFE) filtering). The new algorithm predicted the time in prolonged sitting bouts most accurate (bias ≤ 7 minutes/d). Of all proprietary ActiGraph methods, only 150 cpm without LFE predicted the time in prolonged sitting bouts non-significantly different from the activPAL (bias ≤ 18 minutes/d). However, the frequently used 100 cpm with LFE accurately predicted total sitting time (bias ≤ 7 minutes/d). To study the health effects of ActiGraph measured prolonged sitting, we recommend using the new algorithm. In case a cut-point is used, we recommend 150 cpm without LFE to measure prolonged sitting and 100 cpm with LFE to measure total sitting time. However, both cpm cut-points are not recommended for a detailed bout analysis.

Day-level sedentary pattern estimates derived from hip-worn accelerometer cut-points in 8-12-year-olds: Do they reflect postural transitions?

Improving sedentary measurement is critical to understanding sedentary-health associations in youth. This study assessed agreement between the thigh-worn activPAL and commonly used hip-worn ActiGraph accelerometer methods for assessing sedentary patterns in children. Both devices were worn by 8-12-year-olds (N = 195) for 4.6 ± 1.9 days. Two ActiGraph cut-points were applied to two epoch durations: ≤25 counts (c)/15 s, ≤75c/15s, ≤100c/60s, and ≤300c/60s. Bias, mean absolute deviation (MAD), and intraclass correlation coefficients (ICCs) tested agreement between devices for total sedentary time and 11 sedentary pattern variables (usual bout duration, sedentary time accumulated in various bout durations, breaks/day, break rate, and alpha). For most sedentary pattern variables, ActiGraph 25c/15s, 75c/15s, and 100c/60s had poor ICCs, with bias and MAD >20%. ActiGraph 300c/60s had a better agreement than the other cut-points, but all ICCs were <0.587. ActiGraph underestimated sedentary time in longer bouts and usual bout duration, and overestimated sedentary time in shorter bouts, breaks/day, and alpha. For total sedentary time, ActiGraph 25c/15s, 300c/60s, and 75c/15s had good/fair ICCs, with bias and MAD <20%. Sedentary patterns derived from two commonly used ActiGraph cut-points did not appear to reflect postural changes. These differences between measurement devices should be considered when interpreting findings from sedentary pattern studies.

Breast cancer survivors reduce accelerometer-measured sedentary time in an exercise intervention

PURPOSE

Cancer survivors are highly sedentary and have low physical activity. How physical activity interventions impact sedentary behavior remains unclear. This secondary analysis examined changes in sedentary behavior among breast cancer survivors participating in a physical activity intervention that significantly increased moderate-to-vigorous physical activity (MVPA).

METHODS

Insufficiently active breast cancer survivors were randomized to a 12-week physical activity intervention (exercise arm) or control arm. The intervention focused solely on increasing MVPA with no content targeting sedentary behavior. Total sedentary behavior, light physical activity (LPA), and MVPA were measured at baseline and 12 weeks (ActiGraph GT3X+ accelerometer). Separate linear mixed-effects models tested intervention effects on sedentary behavior, intervention effects on LPA, the relationship between change in MVPA and change in sedentary behavior, and potential moderators of intervention effects on sedentary behavior.

RESULTS

The exercise arm had significantly greater reductions in sedentary behavior than the control arm (mean - 24.9 min/day (SD = 5.9) vs. - 4.8 min/day (SD = 5.9), b = - 20.1 (SE = 8.4), p = 0.02). Larger increases in MVPA were associated with larger decreases in sedentary behavior (b = - 1.9 (SE = 0.21), p < 0.001). Women farther out from surgery had significantly greater reductions in sedentary behavior than women closer to surgery (b = - 0.91 (SE = 0.5), p = 0.07). There was no significant group difference in change in LPA from baseline to 12 weeks (b = 5.64 (SE = 7.69), p = 0.48).

CONCLUSIONS

Breast cancer survivors in a physical activity intervention reduced total sedentary time in addition to increasing MVPA.

IMPLICATIONS FOR CANCER SURVIVORS

Both increasing physical activity and reducing sedentary behavior are needed to promote optimal health in cancer survivors. These results show that MVPA and sedentary behavior could be successfully targeted together, particularly among longer-term cancer survivors.

CLINICAL TRIAL REGISTRATION

This study is registered at www.ClinicalTrials.gov (NCT02332876).