Activity classification using the GENEA: optimum sampling frequency and number of axes

Variations in Concurrent Validity of Two Independent Inertial Measurement Units Compared to Gold Standard for Upper Body Posture during Computerised Device Use

Inertial measurement units (IMUs) may provide an objective method for measuring posture during computer use, but research is needed to validate IMUs' accuracy. We examine the concurrent validity of two different IMU systems in measuring three-dimensional (3D) upper body posture relative to a motion capture system (Mocap) as a potential device to assess postures outside a laboratory environment. We used 3D Mocap and two IMU systems (Wi-Fi and Bluetooth) to capture the upper body posture of twenty-six individuals during three physical computer working conditions (monitor correct, monitor raised, and laptop). Coefficient of determination (R2) and root-mean-square error (RMSE) compared IMUs to Mocap. Head/neck segment [HN], upper trunk segment [UTS], and joint angle [HN-UTS] were the primary variables. Wi-Fi IMUs demonstrated high validity for HN and UTS (sagittal plane) and HN-UTS (frontal plane) for all conditions, and for HN rotation movements (both for the monitor correct and monitor raised conditions), others moderate to poor. Bluetooth IMUs for HN, and UTS (sagittal plane) for the monitor correct, laptop, and monitor raised conditions were moderate. Frontal plane movements except UTS (monitor correct and laptop) and all rotation had poor validity. Both IMU systems were affected by gyroscopic drift with sporadic data loss in Bluetooth IMUs. Wi-Fi IMUs had more acceptable accuracy when measuring upper body posture during computer use compared to Mocap, except for trunk rotations. Variation in IMU systems' performance suggests validation in the task-specific movement(s) is essential.

Evaluating Behavior Recognition Pipeline of Laying Hens Using Wearable Inertial Sensors

Recently, animal welfare has gained worldwide attention. The concept of animal welfare encompasses the physical and mental well-being of animals. Rearing layers in battery cages (conventional cages) may violate their instinctive behaviors and health, resulting in increased animal welfare concerns. Therefore, welfare-oriented rearing systems have been explored to improve their welfare while maintaining productivity. In this study, we explore a behavior recognition system using a wearable inertial sensor to improve the rearing system based on continuous monitoring and quantifying behaviors. Supervised machine learning recognizes a variety of 12 hen behaviors where various parameters in the processing pipeline are considered, including the classifier, sampling frequency, window length, data imbalance handling, and sensor modality. A reference configuration utilizes a multi-layer perceptron as a classifier; feature vectors are calculated from the accelerometer and angular velocity sensor in a 1.28 s window sampled at 100 Hz; the training data are unbalanced. In addition, the accompanying results would allow for a more intensive design of similar systems, estimation of the impact of specific constraints on parameters, and recognition of specific behaviors.

Pedometers and Accelerometers in Multiple Sclerosis: Current and New Applications

Pedometers and accelerometers have become commonplace for the assessment of physical behaviors (e.g., physical activity and sedentary behavior) in multiple sclerosis (MS) research. Current common applications include the measurement of steps taken and the classification of physical activity intensity, as well as sedentary behavior, using cut-points methods. The existing knowledge and applications, coupled with technological advances, have spawned new opportunities for using those motion sensors in persons with MS, and these include the utilization of the data as biomarkers of disease severity and progression, perhaps in clinical practice. Herein, we discuss the current state of knowledge on the validity and applications of pedometers and accelerometers in MS, as well as new opportunities and strategies for the improved assessment of physical behaviors and disease progression, and consequently, personalized care.

Physical Activity in Centenarians beyond Cut-Point-Based Accelerometer Metrics

This study described and compared physical activity (PA) characteristics at the end of the human lifespan using conventional cut-point-based versus cut-point-free accelerometer metrics. Eighteen institutionalized centenarians (101.5 ± 2.1 years, 72.2% female, 89% frail) wore the wrist GENEActiv accelerometer for 7 days. Conventional metrics, such as time spent in light-intensity PA (LiPA) and moderate-to-vigorous intensity PA (MVPA) were calculated according to published cut-points for adults and older adults. The following cut-point-free metrics were evaluated: average acceleration, intensity gradient and Mx metrics. Depending on the cut-point, centenarians accumulated a median of 15-132 min/day of LiPA and 3-15 min/day of MVPA. The average acceleration was 9.2 mg [Q1: 6.7 mg-Q3: 12.6 mg] and the intensity gradient was -3.19 [-3.34--3.12]. The distribution of Z-values revealed positive skew for MVPA, indicating a potential floor effect, whereas the skew magnitude was attenuated for cut-point-free metrics such as intensity gradient or M5. However, both cut-point-based and cut-point-free metrics were similarly positively associated with functional independence, cognitive and physical capacities. This is the first time that PA has been described in centenarians using cut-point-free metrics. Our results suggest that new analytical approaches could overcome cut-point limitations when studying the oldest-old. Future studies using these new cut-point-free PA metrics are warranted to provide more complete and comparable information across groups and populations.

Prevalence of Metabolic Syndrome and Association with Physical Activity and Frailty Status in Spanish Older Adults with Decreased Functional Capacity: A Cross-Sectional Study

Metabolic syndrome (MetS) is a cluster of medical conditions associated with several health disorders. MetS and frailty can be related to prolonged physical deconditioning. There is a need to know whether there is concordance between the different ways of diagnosing it and to know their prevalence in Spanish older adults. Thus, the aims of this study were to describe the prevalence of MetS; to analyse the concordance between different definitions to diagnose MetS; and to study the associations between MetS, frailty status, and physical activity (PA) in older adults with decreased functional capacity. This report is a cross-sectional study involving 110 Spanish older adults of ages ≥65 years with decreased functional capacity. Clinical criteria to diagnose MetS was defined by different expert groups. Anthropometric measurements, blood biochemical analysis, frailty status, functional capacity, and PA were assessed. The Kappa statistic was used to determine the agreement between the five MetS definitions used. Student’s t-test and the Pearson chi-square test were used to examine differences between sex, frailty, and PA groups. The sex-adjusted prevalence of MetS assessed by the National Cholesterol Education Program—Third Adult Treatment Panel was 39.4% in men and 32.5% in women. The International Diabetes Federation and the Harmonized definitions had the best agreement (k = 1.000). The highest odds ratios (ORs) of cardiometabolic risk factors to develop MetS were elevated triglycerides (37.5) and reduced high-density lipoprotein cholesterol (27.3). Central obesity and hypertension prevalence were significantly higher in the non-active group (70.7% and 26.8%, respectively), compared to the active group (50.0% and 7.7%, respectively). Moreover, the active group (OR = 0.85, 95% CI = 0.35, 2.04) and active women group (OR = 0.77, 95% CI = 0.27, 2.20) appeared to show a lower risk of developing this syndrome. MetS is highly prevalent in this sample and changes according to the definition used. It seems that sex and frailty do not influence the development of MetS. However, PA appears to decrease central obesity, hypertension, and the risk of developing MetS.

Accuracy and Acceptability of Wrist-Wearable Activity-Tracking Devices: Systematic Review of the Literature

Background

Numerous wrist-wearable devices to measure physical activity are currently available, but there is a need to unify the evidence on how they compare in terms of acceptability and accuracy.

Objective

The aim of this study is to perform a systematic review of the literature to assess the accuracy and acceptability (willingness to use the device for the task it is designed to support) of wrist-wearable activity trackers.

Methods

We searched MEDLINE, Embase, the Cochrane Central Register of Controlled Trials, and SPORTDiscus for studies measuring physical activity in the general population using wrist-wearable activity trackers. We screened articles for inclusion and, for the included studies, reported data on the studies’ setting and population, outcome measured, and risk of bias.

Results

A total of 65 articles were included in our review. Accuracy was assessed for 14 different outcomes, which can be classified in the following categories: count of specific activities (including step counts), time spent being active, intensity of physical activity (including energy expenditure), heart rate, distance, and speed. Substantial clinical heterogeneity did not allow us to perform a meta-analysis of the results. The outcomes assessed most frequently were step counts, heart rate, and energy expenditure. For step counts, the Fitbit Charge (or the Fitbit Charge HR) had a mean absolute percentage error (MAPE) <25% across 20 studies. For heart rate, the Apple Watch had a MAPE <10% in 2 studies. For energy expenditure, the MAPE was >30% for all the brands, showing poor accuracy across devices. Acceptability was most frequently measured through data availability and wearing time. Data availability was ≥75% for the Fitbit Charge HR, Fitbit Flex 2, and Garmin Vivofit. The wearing time was 89% for both the GENEActiv and Nike FuelBand.

Conclusions

The Fitbit Charge and Fitbit Charge HR were consistently shown to have a good accuracy for step counts and the Apple Watch for measuring heart rate. None of the tested devices proved to be accurate in measuring energy expenditure. Efforts should be made to reduce the heterogeneity among studies.

Detailed analysis and comparison of different activity metrics

Actigraphic measurements are an important part of research in different disciplines, yet the procedure of determining activity values is unexpectedly not standardized in the literature. Although the measured raw acceleration signal can be diversely processed, and then the activity values can be calculated by different activity calculation methods, the documentations of them are generally incomplete or vary by manufacturer. These numerous activity metrics may require different types of preprocessing of the acceleration signal. For example, digital filtering of the acceleration signals can have various parameters; moreover, both the filter and the activity metrics can also be applied per axis or on the magnitudes of the acceleration vector. Level crossing-based activity metrics also depend on threshold level values, yet the determination of their exact values is unclear as well. Due to the serious inconsistency of determining activity values, we created a detailed and comprehensive comparison of the different available activity calculation procedures because, up to the present, it was lacking in the literature. We assessed the different methods by analysing the triaxial acceleration signals measured during a 10-day movement of 42 subjects. We calculated 148 different activity signals for each subject’s movement using the combinations of various types of preprocessing and 7 different activity metrics applied on both axial and magnitude data. We determined the strength of the linear relationship between the metrics by correlation analysis, while we also examined the effects of the preprocessing steps. Moreover, we established that the standard deviation of the data series can be used as an appropriate, adaptive and generalized threshold level for the level intersection-based metrics. On the basis of these results, our work also serves as a general guide on how to proceed if one wants to determine activity from the raw acceleration data. All of the analysed raw acceleration signals are also publicly available.

Responsiveness of an activity tracker as a measurement tool in a knee osteoarthritis clinical trial (ACTIVe-OA study)

BACKGROUND

In osteoarthritis (OA) clinical trials, reliable and responsive outcome measures to document physical and functional improvements are limited.

OBJECTIVE

This study aimed to assess whether the use of an activity tracker in an OA clinical trial is a responsive measurement tool. Secondary objectives assessed feasibility and validity.

METHODS

We recruited 65 participants in a prospective cohort study nested in a placebo-controlled clinical trial of platelet-rich plasma injection in knee OA. Participants wore an activity tracker (Fitbit Flex 2), and a smartphone was preloaded with a mobile application (OApp) designed to monitor load rates as a surrogate of knee loading. Participants used the systems for 7 days at baseline and for 7 days before the 2-month follow-up assessment. Effect size (ES) and standardised response mean (SRM) were calculated for change in step count and knee loading rate and regularly used knee OA outcome measures. Correlation coefficients (r) were calculated to examine the strength of the association between outcome measures.

RESULTS

Step count showed a trivial ES and SRM and mean knee loading rate measurements a moderate ES and SRM. We found a weak but significant correlation between change in mean steps per day and global improvement overall (r = 0.28) and Western Western Ontario and McMaster Universities Osteoarthritis Index function (r = -0.28). Compliance was high with the activity trackers.

CONCLUSIONS

Despite good feasibility, this study did not show significant responsiveness or validity of the activity trackers as compared with currently recommended outcome measures in OA clinical trials. The main challenge is the lack of a gold standard outcome measure to validate against, and because of the complex interplay between pain and measured function, a lack of correlation does not necessarily represent a failed validation in this context. Australian New Zealand Clinical Trials Registry: ACTRN12617000853347. This trial is a substudy of the "Platelet-rich plasma as a symptom-and disease-modifying treatment for knee osteoarthritis - the RESTORE trial".

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.

Associations between Daily Movement Distribution, Bone Structure, Falls, and Fractures in Older Adults: A Compositional Data Analysis Study

With aging, bone density is reduced, increasing the risk of suffering osteoporosis and fractures. Increasing physical activity (PA) may have preventive effects. However, until now, no studies have considered movement behaviors with compositional data or its association to bone mass and structure measured by peripheral computed tomography (pQCT). Thus, the aim of our study was to investigate these associations and to describe movement behavior distribution in older adults with previous falls and fractures and other related risk parameters, taking into account many nutritional and metabolic confounders. In the current study, 70 participants above 65 years old (51 females) from the city of Zaragoza were evaluated for the EXERNET-Elder 3.0 project. Bone mass and structure were assessed with pQCT, and PA patterns were objectively measured by accelerometry. Prevalence of fear of falling, risk of falling, and history of falls and fractures were asked through the questionnaire. Analyses were performed using a compositional data approach. Whole-movement distribution patterns were associated with cortical thickness. In regard to other movement behaviors, moderate-to-vigorous PA (MVPA) showed positive association with cortical thickness and total true bone mineral density (BMD) at 38% (all p < 0.05). In addition, less light PA (LPA) and MVPA were observed in those participants with previous fractures and fear of falling, whereas those at risk of falling and those with previous falls showed higher levels of PA. Our results showed positive associations between higher levels of MVPA and volumetric bone. The different movement patterns observed in the groups with a history of having suffered falls or fractures and other risk outcomes suggest that different exercise interventions should be designed in these populations in order to improve bone and prevent the risk of osteoporosis and subsequent fractures.

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.

Current clinical utilisation of wearable motion sensors for the assessment of outcome following knee arthroplasty: a scoping review

OBJECTIVES

Wearable motion sensors are used with increasing frequency in the evaluation of gait, function and physical activity within orthopaedics and sports medicine. The integration of wearable technology into the clinical pathway offers the ability to improve post-operative patient assessment beyond the scope of current, questionnaire-based patient-reported outcome measures. This scoping review assesses the current methodology and clinical application of accelerometers and inertial measurement units for the evaluation of patient activity and functional recovery following knee arthroplasty.

DESIGN

This is a systematically conducted scoping review following Joanna Briggs Institute methodology for scoping reviews and reported consulting the Preferred Reporting Items for Systematic Review and Meta-Analyses extension for scoping reviews. A protocol for this review is registered with the Open Science Framework (https://osf.io/rzg9q).

DATA SOURCES

CINAHL, EMBASE, MEDLINE and Web of Science databases were searched for manuscripts published between 2008 and 2019.

ELIGIBILITY CRITERIA

We included clinical studies reporting the use of any combination of accelerometers, pedometers or inertial measurement units for patient assessment at any time point following knee arthroplasty.

DATA EXTRACTION AND SYNTHESIS

Data extracted from manuscripts included patient demographics, sensor technology, testing protocol and sensor-based outcome variables.

RESULTS

45 studies were identified, including 2076 knee arthroplasty patients, 620 patients with end-stage osteoarthritis and 449 healthy controls. Primary aims of the identified studies included functional assessment, physical activity monitoring and evaluation of knee instability. Methodology varied widely between studies, with inconsistency in reported sensor configuration, testing protocol and output variables.

CONCLUSIONS

The use of wearable sensors in evaluation of knee arthroplasty procedures is becoming increasingly common and offers the potential to improve clinical understanding of recovery and rehabilitation. While current studies lack consistency, significant opportunity exists for the development of standardised measures and protocols for function and physical activity evaluation.

Cut-point-free accelerometer metrics to assess children's physical activity: An example using the school day

The aims were to (a) investigate associations between a novel accelerometer metric: the minimum acceleration value above which the most active 30-minutes were accumulated during the school day (M30_ACC ), and health indicators, and (b) demonstrate that applying an equivalent cut-point to the M30_ACC metric gives comparable prevalence results as a moderate-to-vigorous physical activity (MVPA) cut-point approach. Two hundred and ninety-six children (age 9-10-years) wore wrist-mounted accelerometers for 7-days. School day MVPA and M30_ACC were calculated. Body mass index (BMI), waist-to-height ratio (WHtR), and cardiorespiratory fitness (CRF) were also measured. Mixed linear models investigated associations between M30_ACC and health indicators. Agreement between ranked MVPA and M30_ACC values was assessed using percent agreement, kappa, sensitivity, and specificity statistics. M30_ACC thresholds associated with health indicators were 213 mg (BMI), 206 mg (WHtR), and 269 mg (CRF) for girls. The equivalent values for boys were 234 mg (BMI), 230 mg (WHtR), and 327 mg (CRF). Less than half of girls and 75% of boys accumulated 30 minutes of school day MVPA. Just <50% of girls and >80% of boys had M30_ACC values ≥200 mg, which is equivalent to brisk walking. Agreement between MVPA and M30_ACC tertiles was high, reflected by the sensitivity and specificity values of >90%. Results demonstrate the utility of M30_ACC as a PA metric that is not heavily influenced by researcher decisions. M30_ACC has potential as an accelerometer-specific metric for generating PA guidelines related to health indicators and easily understood forms of activity such as brisk walking.

Evaluating and Enhancing the Generalization Performance of Machine Learning Models for Physical Activity Intensity Prediction From Raw Acceleration Data

PURPOSE

To evaluate and enhance the generalization performance of machine learning physical activity intensity prediction models developed with raw acceleration data on populations monitored by different activity monitors.

METHOD

Five datasets from four studies, each containing only hip- or wrist-based raw acceleration data (two hip- and three wrist-based) were extracted. The five datasets were then used to develop and validate artificial neural networks (ANN) in three setups to classify activity intensity categories (sedentary behavior, light, and moderate-to-vigorous). To examine generalizability, the ANN models were developed using within dataset (leave-one-subject-out) cross validation, and then cross tested to other datasets with different accelerometers. To enhance the models' generalizability, a combination of four of the five datasets was used for training and the fifth dataset for validation. Finally, all the five datasets were merged to develop a single model that is generalizable across the datasets (50% of the subjects from each dataset for training, the remaining for validation).

RESULTS

The datasets showed high performance in within dataset cross validation (accuracy 71.9-95.4%, Kappa K = 0.63-0.94). The performance of the within dataset validated models decreased when applied to datasets with different accelerometers (41.2-59.9%, K = 0.21-0.48). The trained models on merged datasets consisting hip and wrist data predicted the left-out dataset with acceptable performance (65.9-83.7%, K = 0.61-0.79). The model trained with all five datasets performed with acceptable performance across the datasets (80.4-90.7%, K = 0.68-0.89).

CONCLUSIONS

Integrating heterogeneous datasets in training sets seems a viable approach for enhancing the generalization performance of the models. Instead, within dataset validation is not sufficient to understand the models' performance on other populations with different accelerometers.

Calibration and validation of accelerometer-based activity monitors: A systematic review of machine-learning approaches

BACKGROUND

Objective measures using accelerometer-based activity monitors have been extensively used in physical activity (PA) and sedentary behavior (SB) research. To measure PA and SB precisely, the field is shifting towards machine learning-based (ML) approaches for calibration and validation of accelerometer-based activity monitors. Nevertheless, various parameters regarding the use and development of ML-based models, including data type (raw acceleration data versus activity counts), sampling frequency, window size, input features, ML technique, accelerometer placement, and free-living settings, affect the predictive ability of ML-based models. The effects of these parameters on ML-based models have remained elusive, and will be systematically reviewed here. The open challenges were identified and recommendations are made for future studies and directions.

METHOD

We conducted a systematic search of PubMed and Scopus databases to identify studies published before July 2017 that used ML-based techniques for calibration and validation of accelerometer-based activity monitors. Additional articles were manually identified from references in the identified articles.

RESULTS

A total of 62 studies were eligible to be included in the review, comprising 48 studies that calibrated and validated ML-based models for predicting the type and intensity of activities, and 22 studies for predicting activity energy expenditure.

CONCLUSIONS

It appears that various ML-based techniques together with raw acceleration data sampled at 20-30 Hz provide the opportunity of predicting the type and intensity of activities, as well as activity energy expenditure with comparable overall predictive accuracies regardless of accelerometer placement. However, the high predictive accuracy of laboratory-calibrated models is not reproducible in free-living settings, due to transitive and unseen activities together with differences in acceleration signals.

Comparison of self-reported and accelerometer-assessed measurements of physical activity according to socio-demographic characteristics in Korean adults

OBJECTIVES

Previous studies have shown relatively low correlations between self-reported and accelerometer-assessed physical activity (PA). However, this association differs by socio-demographic factors, and this relationship has not been fully investigated in the general population. Thus, we investigated the correlation between self-reported and accelerometer-assessed PA and whether it differed by demographic and socioeconomic factors among the Korean general population.

METHODS

This cross-sectional study included 623 participants (203 men and 420 women) aged 30 to 64 years, who completed a PA questionnaire and wore a wrist-worn accelerometer on the non-dominant wrist for 7 days. We examined the agreement for metabolic equivalent task minutes per week (MET-min/wk) between the 2 measures and calculated Spearman correlation coefficients according to demographic and socioeconomic factors.

RESULTS

The kappa coefficient between tertiles of self-reported and accelerometer-assessed total MET-min/wk was 0.16 in the total population, suggesting overall poor agreement. The correlation coefficient between the 2 measurements was 0.26 (p<0.001) in the total population, and the correlation tended to decrease with increasing age (p for trend <0.001) and depression scores (p for trend <0.001).

CONCLUSIONS

We found a low correlation between self-reported and accelerometer-assessed PA among healthy Korean adults, and the correlation decreased with age and depression score. When studying PA using accelerometers and/or questionnaires, age and depression need to be considered, as should differences between self-reported and accelerometer-assessed PA.

Physical activity characterization: does one site fit all?

BACKGROUND

It is evident that a growing number of studies advocate a wrist-worn accelerometer for the assessment of patterns of physical activity a priori, yet the veracity of this site rather than any other body-mounted location for its accuracy in classifying activity is hitherto unexplored.

OBJECTIVE

The objective of this review was to identify the relative accuracy with which physical activities can be classified according to accelerometer site and analytical technique.

METHODS

A search of electronic databases was conducted using Web of Science, PubMed and Google Scholar. This review included studies written in the English language, published between database inception and December 2017, which characterized physical activities using a single accelerometer and reported the accuracy of the technique.

RESULTS

A total of 118 articles were initially retrieved. After duplicates were removed and the remaining articles screened, 32 full-text articles were reviewed, resulting in the inclusion of 19 articles that met the eligibility criteria.

CONCLUSION

There is no 'one site fits all' approach to the selection of accelerometer site location or analytical technique. Research design and focus should always inform the most suitable location of attachment, and should be driven by the type of activity being characterized.

Associations Between Objectively Measured Physical Activity, Body Composition and Sarcopenia: Findings from the Hertfordshire Sarcopenia Study (HSS)

Regular physical activity (PA) is associated with reduced risk of the development and progression of musculoskeletal, metabolic and vascular disease. However, PA declines with age and this can contribute to multiple adverse outcomes. The aims of this study were to describe the relationship between accelerometer-determined PA, body composition and sarcopenia (the loss of muscle mass and function with age). Seven-day PA was measured using the GENEactiv accelerometer among 32 men and 99 women aged 74-84 years who participated in the Hertfordshire Sarcopenia Study. We measured mean daily acceleration and minutes/day spent in non-sedentary and moderate-to-vigorous physical activity (MVPA) levels. Body composition was measured by dual-energy X-ray absorptiometry, muscle strength by grip dynamometry and function by gait speed. Sarcopenia was defined according to the EWGSOP diagnostic algorithm. Men and women spent a median (inter-quartile range) of 138.8 (82, 217) and 186 (122, 240) minutes/day engaging in non-sedentary activity but only 14.3 (1.8, 30.2) and 9.5 (2.1, 18.6) min in MVPA, respectively. Higher levels of PA were associated with reduced adiposity, faster walking speed and decreased risk of sarcopenia. For example, a standard deviation (SD) increase in mean daily acceleration was associated with an increase in walking speed of 0.25 (95% CI 0.05, 0.45) SDs and a reduction in the risk of sarcopenia of 35% (95% CI 1, 57%) in fully adjusted analyses. PA was not associated with hand grip strength. Community-dwelling older adults in this study were largely sedentary but there was evidence that higher levels of activity were associated with reduced adiposity and improved function. PA at all intensity levels in later life may help maintain physical function and protect against sarcopenia.

Establishing cut-points for physical activity classification using triaxial accelerometer in middle-aged recreational marathoners

The purpose of this study was to establish GENEA (Gravity Estimator of Normal Everyday Activity) cut-points for discriminating between six relative-intensity activity levels in middle-aged recreational marathoners. Nighty-eight (83 males and 15 females) recreational marathoners, aged 30–45 years, completed a cardiopulmonary exercise test running on a treadmill while wearing a GENEA accelerometer on their non-dominant wrist. The breath-by-breath V̇O₂ data was also collected for criterion measure of physical activity categories (sedentary, light, moderate, vigorous, very vigorous and extremely vigorous). GENEA cut-points for physical activity classification was performed via Receiver Operating Characteristic (ROC) analysis. Spearman’s correlation test was applied to determine the relationship between estimated and measured intensity classifications. Statistical analysis were done for all individuals, and separating samples by sex. The GENEA cut-points established were able to distinguish between all six-relative intensity levels with an excellent classification accuracy (area under the ROC curve (AUC) values between 0.886 and 0.973) for all samples. When samples were separated by sex, AUC values were 0.881–0.973 and 0.924–0.968 for males and females, respectively. The total variance in energy expenditure explained by GENEA accelerometer data was 78.50% for all samples, 78.14% for males, and 83.17% for females. In conclusion, the wrist-worn GENEA accelerometer presents a high capacity of classifying the intensity of physical activity in middle-aged recreational marathoners when examining all samples together, as well as when sample set was separated by sex. This study suggests that the triaxial GENEA accelerometers (worn on the non-dominant wrist) can be used to predict energy expenditure for running activities.

Calibration of raw accelerometer data to measure physical activity: A systematic review

Most of calibration studies based on accelerometry were developed using count-based analyses. In contrast, calibration studies based on raw acceleration signals are relatively recent and their evidences are incipient. The aim of the current study was to systematically review the literature in order to summarize methodological characteristics and results from raw data calibration studies. The review was conducted up to May 2017 using four databases: PubMed, Scopus, SPORTDiscus and Web of Science. Methodological quality of the included studies was evaluated using the Landis and Koch's guidelines. Initially, 1669 titles were identified and, after assessing titles, abstracts and full-articles, 20 studies were included. All studies were conducted in high-income countries, most of them with relatively small samples and specific population groups. Physical activity protocols were different among studies and the indirect calorimetry was the criterion measure mostly used. High mean values of sensitivity, specificity and accuracy from the intensity thresholds of cut-point-based studies were observed (93.7%, 91.9% and 95.8%, respectively). The most frequent statistical approach applied was machine learning-based modelling, in which the mean coefficient of determination was 0.70 to predict physical activity energy expenditure. Regarding the recognition of physical activity types, the mean values of accuracy for sedentary, household and locomotive activities were 82.9%, 55.4% and 89.7%, respectively. In conclusion, considering the construct of physical activity that each approach assesses, linear regression, machine-learning and cut-point-based approaches presented promising validity parameters.

Assessment of Physical Activity of Hospitalised Older Adults: A Systematic Review

BACKGROUND

The assessment of physical activity levels of hospitalised older people requires accurate and reliable measures. Physical activities that older people in hospital commonly engage in include exercises and walking. Measurement of physical activity levels of older inpatients is essential to evaluate the impact of interventions to improve physical activity levels and to determine associations between physical activity in hospital and other health-related outcome measures.

OBJECTIVE

To determine which measures are used to measure physical activity of older people in hospital, and to describe their properties and applications.

METHOD

A systematic review of four databases: Medline, Embase, CINAHL and AMED was conducted for papers published from 1996 to 2016. Inclusion criteria were participants aged ≥ 65 years and studies which included measures of physical activity in the acute medical inpatient setting. Studies which specifically assessed the activity levels of surgical patients or patients with neurological conditions such as stroke or brain injury were excluded. All study designs were included in the review.

RESULTS

18 studies were included from 127 articles selected for full review. 15 studies used objective measures to measure the physical activity of older inpatients: 11 studies used accelerometers and four used direct systematic observations. Seven accelerometers were identified including the StepWatch Activity Monitor, activPAL, GENEActiv, Kenz Lifecorder EX, Actiwatch-L, Tractivity and AugmenTech Inc. Pittsburgh accelerometer. Three studies used a subjective measure (interviews with nurses and patients) to classify patients into low, intermediate and high mobility groups. The StepWatch Activity Monitor was reported to be most accurate at step-counting in patients with slow gait speed or altered gait. The activPAL was reported to be highly accurate at classifying postures.

CONCLUSION

Physical activity levels of older inpatients can be measured using accelerometers. The accuracy of the accelerometers varies between devices and population-specific validation studies are needed to determine their suitability in measuring physical activity levels of hospitalised older people. Subjective measures are less accurate but can be a practical way of measuring physical activity in a larger group of patients.

Wrist-worn triaxial accelerometry predicts the energy expenditure of non-vigorous daily physical activities

OBJECTIVES

Triaxial accelerometry is commonly used to estimate oxygen uptake (VO₂) and energy expenditure in health and fitness studies. We tested the correlation of a triaxial accelerometer in terms of a summation of vector magnitudes with gravity subtracted (SVMgs) and measured VO₂ for different daily physical activities.

DESIGN

Original research, cross-sectional.

METHODS

Twenty volunteers wore a triaxial accelerometer on both wrists while performing 12 assigned daily physical activities for 6min for each activity. The VO₂ was determined by indirect calorimetry using a portable metabolic measurement system. The last 3min of each activity was assumed to represent steady-state. The VO₂ measured during these periods was averaged and converted into metabolic equivalents (METs).

RESULTS

The range of VO₂ for all activities was 0.18-3.2L/min (0.8-12.2 METs). Significant differences in SVMgs existed between accelerometer placements on the dominant (120.9±8.7gmin) versus non-dominant hand (99.9±6.8gmin; P=0.016) for the lowest levels of physical activity defined as <1.5 METs. Piecewise linear regression model using 6 METs as the transition point showed similar significant correlations for the non-dominant wrist (r²=0.85; P<0.001) and the dominant wrist (r²=0.86; P<0.001). Using the non-dominant wrist below 6 METs, the slope of the relationship between SVMgs and METs was 105.3±4.3 (95% CI 96.9 to 113.7) indicating an increase in SVMgs of approximately 100 units for every MET increase in oxygen uptake.

CONCLUSIONS

Wrist-worn triaxial accelerometry reliably predicted energy expenditure during common physical activities <6 METs. More consistent correlations were found when the accelerometer was worn on the non-dominant wrist rather than the dominant wrist.

Performance of Activity Classification Algorithms in Free-Living Older Adults

PURPOSE

The objective of this study is to compare activity type classification rates of machine learning algorithms trained on laboratory versus free-living accelerometer data in older adults.

METHODS

Thirty-five older adults (21 females and 14 males, 70.8 ± 4.9 yr) performed selected activities in the laboratory while wearing three ActiGraph GT3X+ activity monitors (in the dominant hip, wrist, and ankle; ActiGraph, LLC, Pensacola, FL). Monitors were initialized to collect raw acceleration data at a sampling rate of 80 Hz. Fifteen of the participants also wore GT3X+ in free-living settings and were directly observed for 2-3 h. Time- and frequency-domain features from acceleration signals of each monitor were used to train random forest (RF) and support vector machine (SVM) models to classify five activity types: sedentary, standing, household, locomotion, and recreational activities. All algorithms were trained on laboratory data (RFLab and SVMLab) and free-living data (RFFL and SVMFL) using 20-s signal sampling windows. Classification accuracy rates of both types of algorithms were tested on free-living data using a leave-one-out technique.

RESULTS

Overall classification accuracy rates for the algorithms developed from laboratory data were between 49% (wrist) and 55% (ankle) for the SVMLab algorithms and 49% (wrist) to 54% (ankle) for the RFLab algorithms. The classification accuracy rates for SVMFL and RFFL algorithms ranged from 58% (wrist) to 69% (ankle) and from 61% (wrist) to 67% (ankle), respectively.

CONCLUSIONS

Our algorithms developed on free-living accelerometer data were more accurate in classifying the activity type in free-living older adults than those on our algorithms developed on laboratory accelerometer data. Future studies should consider using free-living accelerometer data to train machine learning algorithms in older adults.

Accelerometer wear-site detection: When one site does not suit all, all of the time

OBJECTIVES

Choice of accelerometer wear-site may facilitate greater compliance in research studies. We aimed to test whether a simple method could automatically discriminate whether an accelerometer was worn on the hip or wrist from free-living data.

DESIGN

Cross-sectional.

METHODS

Twenty-two 10-12y old children wore a GENEActiv at the wrist and at the hip for 7-days. The angle between the forearm and the total acceleration vector for the wrist-worn monitor and between the pelvis and the total acceleration vector for the hip-worn monitor (i.e. the angle between the Y-axis component of the acceleration and the total acceleration vector) was calculated for each 5s epoch. The standard deviation of this angle (SDangle) was calculated over time for the wrist-worn and hip-worn monitor for windows of varying lengths. We hypothesised that the wrist angle would be more variable than the hip angle.

RESULTS

Wear site could be discriminated based on SDangle; the shorter the time window the lower the optimal threshold and Area under the Receiver-Operating-Characteristic curve (AUROC) for discrimination of wear-site (AUROC=0.833 (1min) - 0.952 (12h)). Classification accuracy was good for windows of 8min (sensitivity=90%, specificity=87%, AUROC=0.92) and plateaued for windows of ≥60min (sensitivity and specificity >90%, AUROC=0.95-0.96).

CONCLUSIONS

We have presented a robust, computationally simple method that detects whether an accelerometer is being worn on the hip or wrist from 8 to 60min of data. This facilitates the use of wear-site specific algorithms to analyse accelerometer data.

Energy Expenditure Prediction Using Raw Accelerometer Data in Simulated Free Living

PURPOSE

The purpose of this study was to develop, validate, and compare energy expenditure (EE) prediction models for accelerometers placed on the hip, thigh, and wrists using simple accelerometer features as input variables in EE prediction models.

METHODS

Forty-four healthy adults participated in a 90-min, semistructured, simulated free-living activity protocol. During the protocol, participants engaged in 14 different sedentary, ambulatory, lifestyle, and exercise activities for 3-10 min each. Participants chose the order, duration, and intensity of activities. Four accelerometers were worn (right hip, right thigh, as well as right and left wrists) to predict EE compared with that measured by the criterion measure (portable metabolic analyzer). Artificial neural networks (ANNs) were created to predict EE from each accelerometer using a leave-one-out cross-validation approach. Accuracy of the ANN was evaluated using Pearson correlations, root mean square error, and bias. Several ANNs were developed using different input features to determine those most relevant for use in the models.

RESULTS

The ANNs for all four accelerometers achieved high measurement accuracy, with correlations of r > 0.80 for predicting EE. The thigh accelerometer provided the highest overall accuracy (r = 0.90) and lowest root mean square error (1.04 METs), and the differences between the thigh and the other monitors were more pronounced when fewer input variables were used in the predictive models. None of the predictive models had an overall bias for prediction of EE.

CONCLUSIONS

A single accelerometer placed on the thigh provided the highest accuracy for EE prediction, although monitors worn on the wrists or hip can also be used with high measurement accuracy.

Comparability of measured acceleration from accelerometry-based activity monitors

BACKGROUND

Accelerometers that provide triaxial measured acceleration data are now available. However, equivalence of output between brands cannot be assumed and testing is necessary to determine whether features of the acceleration signal are interchangeable.

PURPOSE

This study aimed to establish the equivalence of output between two brands of monitor in a laboratory and in a free-living environment.

METHODS

For part 1, 38 adults performed nine laboratory-based activities while wearing an ActiGraph GT3X+ and GENEActiv (Gravity Estimator of Normal Everyday Activity) at the hip. For part 2, 58 children age 10-12 yr wore a GT3X+ and GENEActiv at the hip for 7 d in a free-living setting.

RESULTS

For part 1, the magnitude of time domain features from the GENEActiv was greater than that from the GT3X+. However, frequency domain features compared well, with perfect agreement of the dominant frequency for 97%-100% of participants for most activities. For part 2, mean daily acceleration measured by the two brands was correlated (r = 0.93, P < 0.001, respectively) but the magnitude was approximately 15% lower for the GT3X+ than that for the GENEActiv at the hip.

CONCLUSIONS

Frequency domain-based classification algorithms should be transferable between monitors, and it should be possible to apply time domain-based classification algorithms developed for one device to the other by applying an affine conversion on the measured acceleration values. The strong relation between accelerations measured by the two brands suggests that habitual activity level and activity patterns assessed by the GENE and GT3X+ may compare well if analyzed appropriately.

Modelling a response as a function of high-frequency count data: The association between physical activity and fat mass

Accelerometers are widely used in health sciences, ecology and other application areas. They quantify the intensity of physical activity as counts per epoch over a given period of time. Currently, health scientists use very lossy summaries of the accelerometer time series, some of which are based on coarse discretisation of activity levels, and make certain implicit assumptions, including linear or constant effects of physical activity. We propose the histogram as a functional summary for achieving a near lossless dimension reduction, comparability between individual time series and easy interpretability. Using the histogram as a functional summary avoids registration of accelerometer counts in time. In our novel method, a scalar response is regressed on additive multi-dimensional functional predictors, including the histogram of the high-frequency counts, and additive non-linear predictors for other continuous covariates. The method improves on the current state-of-the art, as it can deal with high-frequency time series of different lengths and missing values and yields a flexible way to model the physical activity effect with fewer assumptions. It also allows the commonly made modelling assumptions to be tested. We investigate the relationship between the response fat mass and physical activity measured by accelerometer, in data from the Avon Longitudinal Study of Parents and Children. Our method allows testing of whether the effect of physical activity varies over its intensity by gender, by time of day or by day of the week. We show that meaningful interpretation requires careful treatment of identifiability constraints in the light of the sum-to-one property of a histogram. We find that the (not necessarily causal) effect of physical activity on kg fat mass is not linear and not constant over the activity intensity.

Automatic Stress Detection in Working Environments From Smartphones' Accelerometer Data: A First Step

Increase in workload across many organizations and consequent increase in occupational stress are negatively affecting the health of the workforce. Measuring stress and other human psychological dynamics is difficult due to subjective nature of selfreporting and variability between and within individuals. With the advent of smartphones, it is now possible to monitor diverse aspects of human behavior, including objectively measured behavior related to psychological state and consequently stress. We have used data from the smartphone's built-in accelerometer to detect behavior that correlates with subjects stress levels. Accelerometer sensor was chosen because it raises fewer privacy concerns (e.g., in comparison to location, video, or audio recording), and because its low-power consumption makes it suitable to be embedded in smaller wearable devices, such as fitness trackers. About 30 subjects from two different organizations were provided with smartphones. The study lasted for eight weeks and was conducted in real working environments, with no constraints whatsoever placed upon smartphone usage. The subjects reported their perceived stress levels three times during their working hours. Using combination of statistical models to classify selfreported stress levels, we achieved a maximum overall accuracy of 71% for user-specific models and an accuracy of 60% for the use of similar-users models, relying solely on data from a single accelerometer.

Assessing sedentary behavior with the GENEActiv: introducing the sedentary sphere

BACKGROUND

The Sedentary Sphere is a method for the analysis, identification, and visual presentation of sedentary behaviors from a wrist-worn triaxial accelerometer.

PURPOSE

This study aimed to introduce the concept of the Sedentary Sphere and to determine the accuracy of posture classification from wrist accelerometer data.

METHODS

Three samples were used: 1) free living (n = 13, ages 20-60 yr); 2) laboratory based (n = 25, ages 30-65 yr); and 3) hospital inpatients (n = 10, ages 60-90 yr). All participants wore a GENEActiv on their wrist and activPAL on their thigh. The free-living sample wore an additional GENEActiv on the thigh and completed the Multimedia Activity Recall for Children and Adults. The laboratory-based sample wore the monitors while seated at a desk for 7 h, punctuated by 2 min of walking every 20 min. The free-living and inpatient samples wore the monitors for 24 h. Posture was classified from wrist-worn accelerometry using the Sedentary Sphere concept.

RESULTS

Sitting time did not differ between the wrist GENEActiv and the activPAL in the free-living sample and was correlated in the three samples combined (rho = 0.9, P < 0.001), free-living and inpatient samples (r ≃ 0.8, P < 0.01). Mean intraindividual agreement was 85% ± 7%. In the laboratory-based and inpatient samples, sitting time was underestimated by the wrist GENEActiv by 30 min and 2 h relative to the activPAL, respectively (P < 0.05). Posture classification disagreed during reading while standing, cooking while standing, and brief periods during driving. Posture allocation validity was excellent when the GENEActiv was worn on the thigh, evidenced by the near-perfect agreement with the activPAL (96% ± 3%).

CONCLUSIONS

The Sedentary Sphere enables determination of the most likely posture from the wrist-worn GENEActiv. Visualizing behaviors on the sphere displays the pattern of wrist movement and positions within that behavior.

Long-term activity recognition from wristwatch accelerometer data

With the development of wearable devices that have several embedded sensors, it is possible to collect data that can be analyzed in order to understand the user's needs and provide personalized services. Examples of these types of devices are smartphones, fitness-bracelets, smartwatches, just to mention a few. In the last years, several works have used these devices to recognize simple activities like running, walking, sleeping, and other physical activities. There has also been research on recognizing complex activities like cooking, sporting, and taking medication, but these generally require the installation of external sensors that may become obtrusive to the user. In this work we used acceleration data from a wristwatch in order to identify long-term activities. We compare the use of Hidden Markov Models and Conditional Random Fields for the segmentation task. We also added prior knowledge into the models regarding the duration of the activities by coding them as constraints and sequence patterns were added in the form of feature functions. We also performed subclassing in order to deal with the problem of intra-class fragmentation, which arises when the same label is applied to activities that are conceptually the same but very different from the acceleration point of view.

Classification accuracy of the wrist-worn gravity estimator of normal everyday activity accelerometer

PURPOSE

The purpose of this study was to determine whether the published left-wrist cut points for the triaxial Gravity Estimator of Normal Everyday Activity (GENEA) accelerometer are accurate for predicting intensity categories during structured activity bouts.

METHODS

A convenience sample of 130 adults wore a GENEA accelerometer on their left wrist while performing 14 different lifestyle activities. During each activity, oxygen consumption was continuously measured using the Oxycon mobile. Statistical analysis used Spearman's rank correlations to determine the relationship between measured and estimated intensity classifications. Cross tabulations were constructed to show the under- or overestimation of misclassified intensities. One-way χ2 tests were used to determine whether the intensity classification accuracy for each activity differed from 80%.

RESULTS

For all activities, the GENEA accelerometer-based physical activity monitor explained 41.1% of the variance in energy expenditure. The intensity classification accuracy was 69.8% for sedentary activities, 44.9% for light activities, 46.2% for moderate activities, and 77.7% for vigorous activities. The GENEA correctly classified intensity for 52.9% of observations when all activities were examined; this increased to 61.5% with stationary cycling removed.

CONCLUSIONS

A wrist-worn triaxial accelerometer has modest-intensity classification accuracy across a broad range of activities when using the cut points of Esliger et al. Although the sensitivity and the specificity are less than those reported by Esliger et al., they are generally in the same range as those reported for waist-worn, uniaxial accelerometer cut points.

Association between questionnaire- and accelerometer-assessed physical activity: the role of sociodemographic factors

The correlation between objective and self-reported measures of physical activity varies between studies. We examined this association and whether it differed by demographic factors or socioeconomic status (SES). Data were from 3,975 Whitehall II (United Kingdom, 2012–2013) participants aged 60–83 years, who completed a physical activity questionnaire and wore an accelerometer on their wrist for 9 days. There was a moderate correlation between questionnaire- and accelerometer-assessed physical activity (Spearman's r = 0.33, 95% confidence interval: 0.30, 0.36). The correlations were higher in high-SES groups than in low-SES groups (P 's = 0.02), as defined by education (r = 0.38 vs. r = 0.30) or occupational position (r = 0.37 vs. r = 0.29), but did not differ by age, sex, or marital status. Of the self-reported physical activity, 68.3% came from mild activities, 25% from moderate activities, and only 6.7% from vigorous activities, but their correlations with accelerometer-assessed total physical activity were comparable (range of r 's, 0.21–0.25). Self-reported physical activity from more energetic activities was more strongly associated with accelerometer data (for sports, r = 0.22; for gardening, r = 0.16; for housework, r = 0.09). High-SES persons reported more energetic activities, producing stronger accelerometer associations in these groups. Future studies should identify the aspects of physical activity that are most critical for health; this involves better understanding of the instruments being used.