Consumer Activity Trackers Research Articles

Methods to Estimate Energy Expenditure, Physical Activity, and Sedentary Time in Pregnant Women: A Validation Study Using Doubly Labeled Water

Background: Activity trackers and the Pregnancy Physical Activity Questionnaire (PPAQ) measures physical activity (PA) and sedentary time (SED). However, none of these tools have been validated against a criterion method in pregnancy. We aimed to compare a consumer activity tracker and the Danish version of PPAQ (PPAQ-DK) and to validate them using the doubly labeled water technique (DLW) as criterion method. Methods: A total of 220 healthy pregnant women participated. Total energy expenditure (TEE), PA energy expenditure (PAEE), and PA level were determined at gestational Weeks 28–29 using DLW and a Garmin Vivosport (Garmin, Olathe, KS) activity tracker. In addition, PAEE, moderate-to-vigorous intensity PA, and SED were determined using the activity tracker and PPAQ-DK during all three trimesters. Results: TEE from the activity tracker and DLW correlated (r = .63; p < .001), but the activity tracker overestimated TEE (503 kcal/day). Also, the activity tracker overestimated PAEE (303 kcal/day) and PA level compared with DLW. Likewise, PPAQ-DK overestimated PAEE (1,513 kcal/day) compared with DLW. Compared to PPAQ-DK, the activity tracker reported lower values of PAEE and moderate-to-vigorous intensity PA and higher values of SED during all three trimesters. Conclusions: When compared to DLW, we found better agreement of PAEE estimates from the activity tracker than from PPAQ-DK. TEE from the tracker and DLW correlated moderately well, but this was not the case for PAEE or PA level. The activity tracker measured lower PA and higher SED than PPAQ-DK throughout pregnancy. The consumer activity tracker performed better than the questionnaire, but both significantly overestimated PA compared to DLW.

Validation of Fitbit Charge 4 for assessing sleep in Chinese patients with chronic insomnia: A comparison against polysomnography and actigraphy.

Our research aims to assess the performance of a new generation of consumer activity trackers (Fitbit Charge 4TM: FBC) to measure sleep variables and sleep stage classifications in patients with chronic insomnia, compared to polysomnography (PSG) and a widely used actigraph (Actiwatch Spectrum Pro: AWS). We recruited 37 participants, all diagnosed with chronic insomnia disorder, for one night of sleep monitoring in a sleep laboratory using PSG, AWS, and FBC. Epoch-by-epoch analysis along with Bland-Altman plots was used to evaluate FBC and AWS against PSG for sleep-wake detection and sleep variables: total sleep time (TST), sleep efficiency (SE), waking after sleep onset (WASO), and sleep onset latency (SOL). FBC sleep stage classification of light sleep (LS), deep sleep (DS), and rapid eye movement (REM) was also compared to that of PSG. When compared with PSG, FBC notably underestimated DS (-41.4, p < 0.0001) and SE (-4.9%, p = 0.0016), while remarkably overestimating LS (37.7, p = 0.0012). However, the TST, WASO, and SOL assessed by FBC presented no significant difference from that assessed by PSG. Compared with PSG, AWS and FBC showed great accuracy (86.9% vs. 86.5%) and sensitivity (detecting sleep; 92.6% vs. 89.9%), but comparatively poor specificity (detecting wake; 35.7% vs. 62.2%). Both devices showed better accuracy in assessing sleep than wakefulness, with the same sensitivity but statistically different specificity. FBC supplied equivalent parameters estimation as AWS in detecting sleep variables except for SE. This research shows that FBC cannot replace PSG thoroughly in the quantification of sleep variables and classification of sleep stages in Chinese patients with chronic insomnia; however, the user-friendly and low-cost wearables do show some comparable functions. Whether FBC can serve as a substitute for actigraphy and PSG in patients with chronic insomnia needs further investigation.

Exploring the Use of Fitbit Consumer Activity Trackers to Support Active Lifestyles in Adults with Type 2 Diabetes: A Mixed-Methods Study.

Background: People with type 2 diabetes are less active than those without the condition. Physical activity promotion within diabetes health care is limited. This project explored the use of Fitbit activity trackers (Fitbit, San Francisco, CA, USA) to support active lifestyles in adults with type 2 diabetes through a mixed-methods study. Methods: Two stages were conducted. In stage 1, adults with type 2 diabetes used a Fitbit Charge 4 (Fitbit, San Francisco, CA, USA) for 4 weeks. Fitbit and self-reported physical activity data was examined through quantitative analysis. Qualitative analysis was conducted to explore the experiences of participants. In stage 2, health professionals were interviewed to examine their views on using Fitbit activity trackers within type 2 diabetes care. Results: Adults with type 2 diabetes were recruited for stage 1 and adult health care and fitness professionals were recruited for stage 2. Stage 1 participants’ self-reported increases in physical activity (mean weekly minutes of walking increased from 358.75 to 507.50 min, p = 0.046) and a decrease in sedentary behaviour (mean daily hours of sedentary behaviour decreased from 10.65 to 10.05 h, p = 0.575). Fitbit activity data ranges identified individuals who led inactive and sedentary lifestyles below levels recommended and in need of physical activity support to reduce the risk to their health. During interviews, participants stated that the Fitbit activity tracker motivated them to be more active. Stage 2 participants intimated that Fitbit activity trackers could improve the promotion of physical activity within type 2 diabetes care. Interventions involving the Fitbit premium service, community prescription and combined use of Fitbits with physical activity behaviour change models were recommended by stage 2 participants. Conclusions: This study found that there is future scope for using Fitbit activity trackers to support active lifestyles in adults diagnosed with type 2 diabetes.

Preliminary Simulation for Obtaining Accurate Heart Rate Data Using Imputation Methods

AbstractIncreasingly data acquired from consumer activity trackers are being used by clinicians to monitor and treat patients. There are many reported cases where medical doctors leverage data retrieved from these devices to successfully treat and diagnose patients with heart conditions. In addition, hospitals are testing these devices in clinical trials to support telemedicine. One challenge is sensor susceptibility to distortions, resulting in missingness and impacting on validity. In this study we are exploring optimal methods of imputation through an experimental simulation using heart rate interval data. For this initial methodology we are utilizing Fitbit heart rate data to investigate performances of four imputation methods using 1,000 Monte Carlo runs. Normalized errors from simulated missingness experiments were determined by comparing observed and predicted values. Preliminary findings show that linear and Kalman imputation methods are most appropriate when imputing heart rate data.

A Multi-Level Classification Approach for Sleep Stage Prediction With Processed Data Derived From Consumer Wearable Activity Trackers.

Consumer wearable activity trackers, such as Fitbit are widely used in ubiquitous and longitudinal sleep monitoring in free-living environments. However, these devices are known to be inaccurate for measuring sleep stages. In this study, we develop and validate a novel approach that leverages the processed data readily available from consumer activity trackers (i.e., steps, heart rate, and sleep metrics) to predict sleep stages. The proposed approach adopts a selective correction strategy and consists of two levels of classifiers. The level-I classifier judges whether a Fitbit labeled sleep epoch is misclassified, and the level-II classifier re-classifies misclassified epochs into one of the four sleep stages (i.e., light sleep, deep sleep, REM sleep, and wakefulness). Best epoch-wise performance was achieved when support vector machine and gradient boosting decision tree (XGBoost) with up sampling were used, respectively at the level-I and level-II classification. The model achieved an overall per-epoch accuracy of 0.731 ± 0.119, Cohen's Kappa of 0.433 ± 0.212, and multi-class Matthew's correlation coefficient (MMCC) of 0.451 ± 0.214. Regarding the total duration of individual sleep stage, the mean normalized absolute bias (MAB) of this model was 0.469, which is a 23.9% reduction against the proprietary Fitbit algorithm. The model that combines support vector machine and XGBoost with down sampling achieved sub-optimal per-epoch accuracy of 0.704 ± 0.097, Cohen's Kappa of 0.427 ± 0.178, and MMCC of 0.439 ± 0.180. The sub-optimal model obtained a MAB of 0.179, a significantly reduction of 71.0% compared to the proprietary Fitbit algorithm. We highlight the challenges in machine learning based sleep stage prediction with consumer wearables, and suggest directions for future research.

Validity and potential clinical utility of a consumer and research-grade activity tracker in insomnia disorder: outside the laboratory

Validity and potential clinical utility of a consumer and research-grade activity tracker in insomnia disorder: outside the laboratory

Validity, potential clinical utility and comparison of a consumer activity tracker and a research-grade activity tracker in insomnia disorder II: Outside the laboratory.

Accurate assessment of sleep can be fundamental for monitoring, managing and evaluating treatment outcomes within diseases. A proliferation of consumer activity trackers gives easy access to objective sleep. We evaluated the performance of a commercial device (Fitbit Alta HR) relative to a research-grade actigraph (Actiwatch Spectrum Pro) in measuring sleep before and after a cognitive behavioural intervention in insomnia disorder. Twenty-five individuals with DSM-5 insomnia disorder (M=50.6±15.9years) wore Fitbit and Actiwatch and completed a sleep diary during an in-laboratory polysomnogram, and for 1week preceding and following seven weekly sessions of cognitive-behavioural intervention for insomnia. Device performance was compared for sleep outcomes (total sleep time, sleep latency, sleep efficiency and wake after sleep onset). The analyses assessed (a) agreement between devices across days and pre- to post-treatment, and (b) whether pre- to post-treatment changes in sleep assessed by devices correlated with clinical measures of change. Devices generally did not significantly differ from each other on sleep variable estimates, either night to night, in response to sleep manipulation (pre- to post-treatment) or in response to changes in environment (in the laboratory versus at home). Change in sleep measures across time from each device showed some correlation with common clinical measures of change in insomnia, but not insomnia diagnosis as a categorical variable. Overall, the Fitbit provides similar estimates of sleep outside the laboratory to a research grade actigraph. Despite the similarity between Fitbit and Actiwatch performance, the use of consumer technology is still in its infancy and caution should be taken in its interpretation.

Validity, potential clinical utility, and comparison of consumer and research-grade activity trackers in Insomnia Disorder I: In-lab validation against polysomnography.

Consumer activity trackers claiming to measure sleep/wake patterns are ubiquitous within clinical and consumer settings. However, validation of these devices in sleep disorder populations are lacking. We examined 1 night of sleep in 42 individuals with insomnia (mean=49.14±17.54years) using polysomnography, a wrist actigraph (Actiwatch Spectrum Pro: AWS) and a consumer activity tracker (Fitbit Alta HR: FBA). Epoch-by-epoch analysis and Bland-Altman methods evaluated each device against polysomnography for sleep/wake detection, total sleep time, sleep efficiency, wake after sleep onset and sleep latency. FBA sleep stage classification of light sleep (N1+N2), deep sleep (N3) and rapid eye movement was also compared with polysomnography. Compared with polysomnography, both activity trackers displayed high accuracy (81.12% versus 82.80%, AWS and FBA respectively; ns) and sensitivity (sleep detection; 96.66% versus 96.04%, respectively; ns) but low specificity (wake detection; 39.09% versus 44.76%, respectively; p=.037). Both trackers overestimated total sleep time and sleep efficiency, and underestimated sleep latency and wake after sleep onset. FBA demonstrated sleep stage sensitivity and specificity, respectively, of 79.39% and 58.77% (light), 49.04% and 95.54% (deep), 65.97% and 91.53% (rapid eye movement). Both devices were more accurate in detecting sleep than wake, with equivalent sensitivity, but statistically different specificity. FBA provided equivalent estimates as AWS for all traditional actigraphy sleep parameters. FBA also showed high specificity when identifying N3, and rapid eye movement, though sensitivity was modest. Thus, it underestimates these sleep stages and overestimates light sleep, demonstrating more shallow sleep than actually obtained. Whether FBA could serve as a low-cost substitute for actigraphy in insomnia requires further investigation.

Step Count and Sedentary Time Validation of Consumer Activity Trackers and a Pedometer in Free-Living Settings

We evaluated the accuracy and precision of wearable activity trackers and a pedometer (ATPs) in estimating steps and sedentary time (ST) in free-living settings. Thirty-two healthy men and women (M...

0313 Validation of a Consumer Activity Tracker Against Polysomnography and Actigraphy in Insomnia.

0313 Validation of a Consumer Activity Tracker Against Polysomnography and Actigraphy in Insomnia.

Sensitivity of a Consumer Activity Tracker to Detect Changes In Lab-based and Free-living Activity Settings

Sensitivity of a Consumer Activity Tracker to Detect Changes In Lab-based and Free-living Activity Settings

Steps per Day Measured by Consumer Activity Trackers Worn at the Non-Dominant and Dominant Wrist Relative to a Waist-Worn Pedometer

Purpose: To compare steps counts recorded by consumer activity trackers when worn on the non-dominant and dominant wrist against a waist-worn pedometer during free-living. Methods: 30 participants ...

A Consumer Activity Tracker Is Sensitive In Detecting Change In Free-living Energy Expenditure And Steps

A Consumer Activity Tracker Is Sensitive In Detecting Change In Free-living Energy Expenditure And Steps

Comparison Of Step Estimates Between A Consumer Activity Tracker And Research Accelerometer On The Dominant And Non-dominant Wrist

Many consumer activity trackers (AT) can be worn at different locations such as the hip, shoe, or wrist. The versatility of wear locations makes these devices attractive to the consumer. However, there is limited data on the differences between wear locations of the dominant (D) and non-dominant (ND) wrist location when estimating steps, a common metric produced by many AT. PURPOSE: To compare estimates of total steps (TS) obtained from a consumer AT and a research accelerometer (RA) worn simultaneously on the dominant (D) and non-dominant (ND) wrist during three simulated free-living conditions. METHODS: Twenty healthy adults wore an AT and RA on the D and ND wrist and completed three 1-hour laboratory conditions: 1) sedentary (SED), 2) sedentary plus walking (SW), and sedentary plus jogging (SJ). During the SED condition, participants completed 60 minutes of sitting. During the SW and SJ conditions, participants completed 30 minutes of sitting plus 30 minutes of continuous walking or jogging at 3.0 or 5.0 mph, respectively. Means and 95% confidence intervals were used to assess differences of TS between the D and ND wrist locations for the AT and RA among all three conditions. RESULTS: Within devices, similar TS estimates were produced from the D and ND wrist across all three conditions. Between devices, the AT and RA placed at the D and ND wrist produced similar estimates of TS during the SED and SW conditions. For the SJ condition, the AT produced significantly higher TS estimates, compared with the RA, which was less sensitive to TS detection with increasing intensity. CONCLUSIONS: Researchers should use caution when using TS detected from a wrist-worn RA. An additional criterion measure, such as manual step counting, would more clearly identify under- and over-reporting of TS output from AT and RA during controlled laboratory settings. Supported by: Seed Grant from the Institute of Applied Life Sciences at University of Massachusetts, Amherst and by Fossil Inc.Figure

A Consumer Activity Tracker is Sensitive to Changes In Steps during Simulated Free-Living

A Consumer Activity Tracker is Sensitive to Changes In Steps during Simulated Free-Living

Reliability and Validity of Ten Consumer Activity Trackers Depend on Walking Speed.

To examine the test-retest reliability and validity of ten activity trackers for step counting at three different walking speeds. Thirty-one healthy participants walked twice on a treadmill for 30 min while wearing 10 activity trackers (Polar Loop, Garmin Vivosmart, Fitbit Charge HR, Apple Watch Sport, Pebble Smartwatch, Samsung Gear S, Misfit Flash, Jawbone Up Move, Flyfit, and Moves). Participants walked three walking speeds for 10 min each; slow (3.2 km·h), average (4.8 km·h), and vigorous (6.4 km·h). To measure test-retest reliability, intraclass correlations (ICC) were determined between the first and second treadmill test. Validity was determined by comparing the trackers with the gold standard (hand counting), using mean differences, mean absolute percentage errors, and ICC. Statistical differences were calculated by paired-sample t tests, Wilcoxon signed-rank tests, and by constructing Bland-Altman plots. Test-retest reliability varied with ICC ranging from -0.02 to 0.97. Validity varied between trackers and different walking speeds with mean differences between the gold standard and activity trackers ranging from 0.0 to 26.4%. Most trackers showed relatively low ICC and broad limits of agreement of the Bland-Altman plots at the different speeds. For the slow walking speed, the Garmin Vivosmart and Fitbit Charge HR showed the most accurate results. The Garmin Vivosmart and Apple Watch Sport demonstrated the best accuracy at an average walking speed. For vigorous walking, the Apple Watch Sport, Pebble Smartwatch, and Samsung Gear S exhibited the most accurate results. Test-retest reliability and validity of activity trackers depends on walking speed. In general, consumer activity trackers perform better at an average and vigorous walking speed than at a slower walking speed.

Reliability and validity of ten consumer activity trackers.

BackgroundActivity trackers can potentially stimulate users to increase their physical activity behavior. The aim of this study was to examine the reliability and validity of ten consumer activity trackers for measuring step count in both laboratory and free-living conditions.MethodHealthy adult volunteers (n = 33) walked twice on a treadmill (4.8 km/h) for 30 min while wearing ten different activity trackers (i.e. Lumoback, Fitbit Flex, Jawbone Up, Nike+ Fuelband SE, Misfit Shine, Withings Pulse, Fitbit Zip, Omron HJ-203, Yamax Digiwalker SW-200 and Moves mobile application). In free-living conditions, 56 volunteers wore the same activity trackers for one working day. Test-retest reliability was analyzed with the Intraclass Correlation Coefficient (ICC). Validity was evaluated by comparing each tracker with the gold standard (Optogait system for laboratory and ActivPAL for free-living conditions), using paired samples t-tests, mean absolute percentage errors, correlations and Bland-Altman plots.ResultsTest-retest analysis revealed high reliability for most trackers except for the Omron (ICC .14), Moves app (ICC .37) and Nike+ Fuelband (ICC .53). The mean absolute percentage errors of the trackers in laboratory and free-living conditions respectively, were: Lumoback (−0.2, −0.4), Fibit Flex (−5.7, 3.7), Jawbone Up (−1.0, 1.4), Nike+ Fuelband (−18, −24), Misfit Shine (0.2, 1.1), Withings Pulse (−0.5, −7.9), Fitbit Zip (−0.3, 1.2), Omron (2.5, −0.4), Digiwalker (−1.2, −5.9), and Moves app (9.6, −37.6). Bland-Altman plots demonstrated that the limits of agreement varied from 46 steps (Fitbit Zip) to 2422 steps (Nike+ Fuelband) in the laboratory condition, and 866 steps (Fitbit Zip) to 5150 steps (Moves app) in the free-living condition.ConclusionThe reliability and validity of most trackers for measuring step count is good. The Fitbit Zip is the most valid whereas the reliability and validity of the Nike+ Fuelband is low.

W2E--Wellness Warehouse Engine for Semantic Interoperability of Consumer Health Data.

Novel health monitoring devices and applications allow consumers easy and ubiquitous ways to monitor their health status. However, technologies from different providers lack both technical and semantic interoperability and hence the resulting health data are often deeply tied to a specific service, which is limiting its reusability and utilization in different services. We have designed a Wellness Warehouse Engine (W2E) that bridges this gap and enables seamless exchange of data between different services. W2E provides interfaces to various data sources and makes data available via unified representational state transfer application programming interface to other services. Importantly, it includes Unifier--an engine that allows transforming input data into generic units reusable by other services, and Analyzer--an engine that allows advanced analysis of input data, such as combining different data sources into new output parameters. In this paper, we describe the architecture of W2E and demonstrate its applicability by using it for unifying data from four consumer activity trackers, using a test base of 20 subjects each carrying out three different tracking sessions. Finally, we discuss challenges of building a scalable Unifier engine for the ever-enlarging number of new devices.