Alexander H. K. Montoye

Validity of Consumer-Based Physical Activity Monitors for Specific Activity Types.

PURPOSE Consumer-based physical activity (PA) monitors are popular for individual tracking of PA variables. However, current research has not examined how these monitors track energy expenditure (EE) and steps in distinct activities. This study examined the accuracy of the Fitbits One, Zip, and Flex and Jawbone UP24 for estimating EE and steps for specific activities and activity categories. METHODS Thirty subjects completed a structured protocol consisting of three sedentary, four household, and four ambulatory/exercise activities. All subjects began by lying on a bed for 10 min; 10 other activities were performed for 5 min each. Indirect calorimetry (COSMED) and researcher-counted steps were criterion measures for EE and step counts, respectively. The Omron HJ-720IT pedometer was used as a comparison of step count accuracy. EE and steps were compared with criterion measures using the Friedman repeated-measures nonparametric test and mean absolute percent error (MAPE). RESULTS All PA monitors predicted EE within 8% of COSMED for sedentary activity but overestimated EE by 16%-40% during ambulatory activity. All monitors except the Fitbit Flex (within 8% of criterion) underestimated EE by 27%-34% during household activity. EE predictions were accompanied with MAPE >10%. For household activity, the Fitbit Flex estimated steps within 10% of researcher-counted steps; all other monitors underestimated steps by 35%-64%. All monitors estimated steps within 4% of researcher-counted steps and displayed MAPE <10% during ambulatory activity. The Omron underestimated household steps by 74% but was within 1% for ambulatory steps. All monitors severely underestimated EE and steps during cycling. CONCLUSION Consumer-based PA monitors should be used cautiously for estimating EE, although they provide accurate measures of steps for structured ambulatory activity, similar to validated pedometers.

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.

Comparison of four Fitbit and Jawbone activity monitors with a research-grade ActiGraph accelerometer for estimating physical activity and energy expenditure

Background/aim Consumer-based physical activity (PA) monitors have become popular tools to track PA behaviours. Currently, little is known about the validity of the measurements provided by consumer monitors. We aimed to compare measures of steps, energy expenditure (EE) and active minutes of four consumer monitors with one research-grade accelerometer within a semistructured protocol. Methods Thirty men and women (18–80 years old) wore Fitbit One (worn at the waist), Fitbit Zip (waist), Fitbit Flex (wrist), Jawbone UP24 (wrist) and one waist-worn research-grade accelerometer (ActiGraph) while participating in an 80 min protocol. A validated EE prediction equation and active minute cut-points were applied to ActiGraph data. Criterion measures were assessed using direct observation (step count) and portable metabolic analyser (EE, active minutes). A repeated measures analysis of variance (ANOVA) was used to compare differences between consumer monitors, ActiGraph, and criterion measures. Similarly, a repeated measures ANOVA was applied to a subgroup of subjects who didn’t cycle. Results Participants took 3321±571 steps, had 28±6 active min and expended 294±56 kcal based on criterion measures. Comparatively, all monitors underestimated steps and EE by 13%–32% (p<0.01); additionally the Fitbit Flex, UP24, and ActiGraph underestimated active minutes by 35%–65% (p<0.05). Underestimations of PA and EE variables were found to be similar in the subgroup analysis. Conclusion Consumer monitors had similar accuracy for PA assessment as the ActiGraph, which suggests that consumer monitors may serve to track personal PA behaviours and EE. However, due to discrepancies among monitors, individuals should be cautious when comparing relative and absolute differences in PA values obtained using different monitors.

Reporting accelerometer methods in physical activity intervention studies: a systematic review and recommendations for authors.

Objective This systematic review assessed the completeness of accelerometer reporting in physical activity (PA) intervention studies and assessed factors related to accelerometer reporting. Design The PubMed database was used to identify manuscripts for inclusion. Included studies were PA interventions that used accelerometers, were written in English and were conducted between 1 January 1998 and 31 July 2014. 195 manuscripts from PA interventions that used accelerometers to measure PA were included. Manuscript completeness was scored using 12 questions focused on 3 accelerometer reporting areas: accelerometer information, data processing and interpretation and protocol non-compliance. Variables, including publication year, journal focus and impact factor, and population studied were evaluated to assess trends in reporting completeness. Results The number of manuscripts using accelerometers to assess PA in interventions increased from 1 in 2002 to 29 in the first 7 months of 2014. Accelerometer reporting completeness correlated weakly with publication year (r=0.24, p<0.001). Correlations were greater when we assessed improvements over time in reporting data processing in manuscripts published in PA-focused journals (r=0.43, p=0.002) compared to manuscripts published in non-PA-focused journals (r=0.19, p=0.021). Only 7 of 195 (4%) manuscripts reported all components of accelerometer use, and only 132 (68%) reported more than half of the components. Conclusions Accelerometer reporting of PA in intervention studies has been poor and improved only minimally over time. We provide recommendations to improve accelerometer reporting and include a template to standardise reports.

Comparison of Activity Type Classification Accuracy from Accelerometers Worn on the Hip, Wrists, and Thigh in Young, Apparently Healthy Adults

ABSTRACT The purpose of this article is to compare accuracy of activity type prediction models for accelerometers worn on the hip, wrists, and thigh. Forty-four adults performed sedentary, ambulatory, lifestyle, and exercise activities (14 total, 10 categories) for 3–10 minutes each in a 90-minute semi-structured laboratory protocol. Artificial neural networks (ANNs) were developed for four accelerometers (right hip, both wrists, and right thigh,) to predict individual activities and activity categories, with direct observation (DO) as criterion. The wrist-mounted accelerometers achieved the highest accuracy for individual activities (80.9%–81.1%) and activity categories (86.6%–86.7%); accuracy was not different between wrists. The hip-mounted accelerometer had the lowest accuracy (66.2% individual activities, 72.5% activity categories); thigh-mounted accelerometer accuracy (71.4% individual activities, 84.0% activity categories) fell between the wrist- and hip-mounted accelerometers. ANNs developed for accelerometers worn on the wrists and thigh provided high accuracy for activity type prediction and represent a potential approach to physical activity (PA) assessment.

Use of a wireless network of accelerometers for improved measurement of human energy expenditure

Single, hip-mounted accelerometers can provide accurate measurements of energy expenditure (EE) in some settings, but are unable to accurately estimate the energy cost of many non-ambulatory activities. A multi-sensor network may be able to overcome the limitations of a single accelerometer. Thus, the purpose of our study was to compare the abilities of a wireless network of accelerometers and a hip-mounted accelerometer for the prediction of EE. Thirty adult participants engaged in 14 different sedentary, ambulatory, lifestyle and exercise activities for five minutes each while wearing a portable metabolic analyzer, a hip-mounted accelerometer (AG) and a wireless network of three accelerometers (WN) worn on the right wrist, thigh and ankle. Artificial neural networks (ANNs) were created separately for the AG and WN for the EE prediction. Pearson correlations (r) and the root mean square error (RMSE) were calculated to compare criterion-measured EE to predicted EE from the ANNs. Overall, correlations were higher (r = 0.95 vs. r = 0.88, p < 0.0001) and RMSE was lower (1.34 vs. 1.97 metabolic equivalents (METs), p < 0.0001) for the WN than the AG. In conclusion, the WN outperformed the AG for measuring EE, providing evidence that the WN can provide highly accurate estimates of EE in adults participating in a wide range of activities.

Energy-aware activity classification using wearable sensor networks

This paper presents implementation details, system characterization, and the performance of a wearable sensor network that was designed for human activity analysis. Specific machine learning mechanisms are implemented for recognizing a target set of activities with both out-of-body and on-body processing arrangements. Impacts of energy consumption by the on-body sensors are analyzed in terms of activity detection accuracy for out-of-body processing. Impacts of limited processing abilities for the on-body scenario are also characterized in terms of detection accuracy, by varying the background processing load in the sensor units. Impacts of varying number of sensors in terms of activity classification accuracy are also evaluated. Through a rigorous systems study, it is shown that an efficient human activity analytics system can be designed and operated even under energy and processing constraints of tiny on-body wearable sensors.

Validation and Comparison of Accelerometers Worn on the Hip, Thigh, and Wrists for Measuring Physical Activity and Sedentary Behavior

Background Recent evidence suggests that physical activity (PA) and sedentary behavior (SB) exert independent effects on health. Therefore, measurement methods that can accurately assess both constructs are needed. Objective To compare the accuracy of accelerometers placed on the hip, thigh, and wrists, coupled with machine learning models, for measurement of PA intensity category (SB, light-intensity PA [LPA], and moderate- to vigorous-intensity PA [MVPA]) and breaks in SB. Methods Forty young adults (21 female; age 22.0 ± 4.2 years) participated in a 90-minute semi-structured protocol, performing 13 activities (three sedentary, 10 non-sedentary) for 3–10 minutes each. Participants chose activity order, duration, and intensity. Direct observation (DO) was used as a criterion measure of PA intensity category, and transitions from SB to a non-sedentary activity were breaks in SB. Participants wore four accelerometers (right hip, right thigh, and both wrists), and a machine learning model was created for each accelerometer to predict PA intensity category. Sensitivity and specificity for PA intensity category classification were calculated and compared across accelerometers using repeated measures analysis of variance, and the number of breaks in SB was compared using repeated measures analysis of variance. Results Sensitivity and specificity values for the thigh-worn accelerometer were higher than for wrist- or hip-worn accelerometers, > 99% for all PA intensity categories. Sensitivity and specificity for the hip-worn accelerometer were 87–95% and 93–97%. The left wrist-worn accelerometer had sensitivities and specificities of > 97% for SB and LPA and 91–95% for MVPA, whereas the right wrist-worn accelerometer had sensitivities and specificities of 93–99% for SB and LPA but 67–84% for MVPA. The thigh-worn accelerometer had high accuracy for breaks in SB; all other accelerometers overestimated breaks in SB. Conclusion Coupled with machine learning modeling, the thigh-worn accelerometer should be considered when objectively assessing PA and SB.

Variability of Objectively Measured Sedentary Behavior.

INTRODUCTION The primary purpose of this study was to evaluate variability of sedentary behavior (SB) throughout a 7-d measurement period and to determine if <7 d of SB measurement would be comparable with the typical 7-d measurement period. METHODS Retrospective data from Ball State Universitys Clinical Exercise Physiology Laboratory on 293 participants (99 men, 55 ± 14 yr, body mass index = 29 ± 5 kg·m(-2); 194 women, 51 ± 12 yr, body mass index = 27 ± 7 kg·m(-2)) with seven consecutive days of data collected with ActiGraph accelerometers were analyzed (ActiGraph, Fort Walton Beach, FL). Time spent in SB (either <100 counts per minute or <150 counts per minute) and breaks in SB were compared between days and by sex using a two-way repeated-measures ANOVA. Stepwise regression was performed to determine if <7 d of SB measurement were comparable with the 7-d method, using an adjusted R2 of ≥0.9 as a criterion for equivalence. RESULTS There were no differences in daily time spent in SB between the 7 d for all participants. However, there was a significant interaction between sex and days, with women spending less time in SB on both Saturdays and Sundays than men when using the 100 counts per minute cut-point. Stepwise regression showed using any 4 d would be comparable with a 7-d measurement (R2 > 0.90). CONCLUSIONS When assessed over a 7-d measurement period, SB appears to be very stable from day to day, although there may be some small differences in time spent in SB and breaks in SB between men and women, particularly on weekend days. The stepwise regression analysis suggests that a measurement period as short as 4 d could provide comparable data (91% of variance) with a 1-wk assessment. Shorter assessment periods would reduce both researcher and subject burden in data collection.

Evaluating the Responsiveness of Accelerometry to Detect Change in Physical Activity

The responsiveness to change of the Actical and ActiGraph accelerometers was assessed in children and adolescents. Participants (N = 208) aged 6 to 16 years completed two simulated free-living protocols, one with primarily light-to-moderate physical activity (PA) and one with mostly moderate-to-vigorous PA. Time in sedentary, light, moderate, and vigorous PA was estimated using 8 previously developed cut-points (4 for Actical and 4 for ActiGraph) and 5-sec, 15-sec, and 30-sec epochs. Accelerometer responsiveness for detecting differences in PA between protocols was assessed using standardized response means (SRMs). SRM values ≥.8 represented high responsiveness to change. Both accelerometers showed high responsiveness for all PA intensities (SRMs = 1.2–4.7 for Actical and 1.1–3.3 for ActiGraph). All cut-points and epoch lengths yielded high responsiveness, and choice of cut-points and epoch length had little effect on responsiveness. Thus, both the Actical and ActiGraph can detect change in PA in a simulated free-living setting, irrespective of cut-point selection or epoch length.