Dinesh John

Validation and comparison of ActiGraph activity monitors

OBJECTIVE To compare activity counts from the ActiGraph GT3X to those from the ActiGraph GT1M during treadmill walking/running. A secondary aim was to develop tri-axial vector magnitude (VM3) cut-points to classify physical activity (PA) intensity. METHODS Fifty participants wore the GT3X and the GT1M on the non-dominant hip and exercised at 4 treadmill speeds (4.8, 6.4, 9.7, and 12 km h(-1)). Vertical (VT) and antero-posterior (AP) activity counts (counts min(-1)) as well as the vector magnitudes of the two axes (VM2) from both monitors were tested for significant differences using two-way ANOVAs. Bland-Altman plots were used to assess agreement between activity counts from the GT3X and GT1M. Linear regression analysis between VM3 countsmin(-1) and oxygen consumption data was conducted to develop VM3 cut-points for moderate, hard and very hard PA. RESULTS There were no significant inter-monitor differences in VT activity counts at any speed. AP and VM2 activity counts from the GT1M were significantly higher (p<0.01) than those from the GT3X at 4.8, 9.7 and 12 km h(-1). High inter-monitor agreement was found for VT activity counts but not for AP and VM2 activity counts. VM3 cut-points for moderate, hard, and very hard PA intensities were 2690-6166, 6167-9642, >9642 counts min(-1). CONCLUSION Due to the lack of congruence between the AP and VM2 activity counts from the GT1M and the GT3X, comparisons of data obtained with these two monitors should be avoided when using more than just the VT axis. VM3 cut-points may be used to classify PA in future studies.

Actigraph and Actical Physical Activity Monitors: A Peek under the Hood

Since the 1980s, accelerometer-based activity monitors have been used by researchers to quantify physical activity. The technology of these monitors has continuously evolved. For example, changes have been made to monitor hardware (type of sensor (e.g., piezoelectric, piezoresistive, capacitive)) and output format (counts vs raw signal). Commonly used activity monitors belong to the ActiGraph and the Actical families. This article presents information on several electromechanical aspects of these commonly used activity monitors. The majority of the article focuses on the evolution of the ActiGraph activity monitor by describing the differences among the 7164, the GT1M, and the GT3X models. This is followed by brief descriptions of the influences of device firmware and monitor calibration status. We also describe the Actical, but the discussion is short because this device has not undergone any major changes since it was first introduced. This article may help researchers gain a better understanding of the functioning of activity monitors. For example, a common misconception among physical activity researchers is that the ActiGraph GT1M and GT3X are piezoelectric sensor-based monitors. Thus, this information may also help researchers to describe these monitors more accurately in scientific publications.

Comparison of Four ActiGraph Accelerometers during Walking and Running

UNLABELLED Currently, researchers can use the ActiGraph 7164 or one of three different versions of the ActiGraph GT1M to objectively measure physical activity. PURPOSE To determine whether differences exist between activity counts from the ActiGraph 7164 and the three versions of the GT1M at given walking and running speeds. METHODS Ten male participants (23.6 +/- 2.7 yr) completed treadmill walking and running at 10 different speeds (3-min stages) while wearing the ActiGraph 7164 and the latest GT1M (GT1M-V3) or the GT1M version one (GT1M-V1) and the GT1M version two (GT1M-V2). Participants walked at 3, 5, and 7 km x h(-1) followed by running at 8, 10, 12, 14, 16, 18, and 20 km x h(-1). The accelerometers were worn on an elastic belt around the waist over the left and right sides of the hip. Testing was performed on different days using a counterbalanced within-subjects design to account for potential differences attributable to accelerometer placement. At each speed, a one-way repeated-measures ANOVA was used to examine differences between activity counts in counts per minute (cpm). Post hoc pairwise comparisons with Bonferroni adjustments were used where appropriate. RESULTS There were no significant differences between activity counts at any given walking or running speed (P < 0.05). At all running speeds, activity counts from the ActiGraph 7164 and GT1M-V2 displayed the lowest and highest values, respectively. Output from all accelerometers peaked at 14 km x h(-1) (mean range = 8974 +/- 677 to 9412 +/- 982 cpm) and then gradually declined at higher speeds. The mean difference score at peak output between the ActiGraph 7164 and GT1M-V2 was 439 +/- 565 cpm. CONCLUSIONS There were no statistically significant differences between outputs from all the accelerometers, indicating that researchers can select any of the four ActiGraph accelerometers in doing research.

Use of pedometers and accelerometers in clinical populations: validity and reliability issues

Abstract Background: Pedometers and accelerometers are often used in clinical research studies, and have the potential to provide valid and objective information on physical activity. Objectives: The purposes of this review are to introduce clinicians to various activity monitors that are commercially available, and to discuss their strengths and limitations. Scientific articles over the past 20 years were reviewed, and the most common types of physical activity monitors were identified. Emphasis was placed on devices with established validity and reliability, which were acceptable to participants and did not interfere with their activities. Ease-of-use from the investigators standpoint was also considered. Major findings: A number of activity monitors were identified, including four pedometers (Yamax SW digiwalker, New Lifestyles NL-2000, Omron HJ-720 ITC, and StepWatch), three accelerometer-based activity monitors (ActiGraph, Actical, and RT3), and a multisensor device (Sensewear Armband). Validity and reliability of these devices is discussed. Conclusions: Pedometers and accelerometers are useful for tracking ambulatory physical activity in clinical populations, and those that display steps and/or calories are useful in motivating patients to increase their activity levels.

Multisensor Data Fusion for Physical Activity Assessment

This paper presents a sensor fusion method for assessing physical activity (PA) of human subjects, based on support vector machines (SVMs). Specifically, acceleration and ventilation measured by a wearable multisensor device on 50 test subjects performing 13 types of activities of varying intensities are analyzed, from which activity type and energy expenditure are derived. The results show that the method correctly recognized the 13 activity types 88.1% of the time, which is 12.3% higher than using a hip accelerometer alone. Also, the method predicted energy expenditure with a root mean square error of 0.42 METs, 22.2% lower than using a hip accelerometer alone. Furthermore, the fusion method was effective in reducing the subject-to-subject variability (standard deviation of recognition accuracies across subjects) in activity recognition, especially when data from the ventilation sensor were added to the fusion model. These results demonstrate that the multisensor fusion technique presented is more effective in identifying activity type and energy expenditure than the traditional accelerometer-alone-based methods.

Validation of the Fitbit wireless activity tracker for prediction of energy expenditure.

OBJECTIVE The purpose of this study was to examine the accuracy of the Fitbit wireless activity tracker in assessing energy expenditure (EE) for different activities. METHODS Twenty participants (10 males, 10 females) wore the Fitbit Classic wireless activity tracker on the hip and the Oxycon Mobile portable metabolic system (criterion). Participants performed walking and running trials on a treadmill and a simulated free-living activity routine. Paired t tests were used to test for differences between estimated (Fitbit) and criterion (Oxycon) kcals for each of the activities. RESULTS Mean bias for estimated energy expenditure for all activities was -4.5 ± 1.0 kcals/6 min (95% limits of agreement: -25.2 to 15.8 kcals/6 min). The Fitbit significantly underestimated EE for cycling, laundry, raking, treadmill (TM) 3 mph at 5% grade, ascent/descent stairs, and TM 4 mph at 5% grade, and significantly overestimated EE for carrying groceries. Energy expenditure estimated by the Fitbit was not significantly different than EE calculated from the Oxycon Mobile for 9 activities. CONCLUSION The Fitbit worn on the hip significantly underestimates EE of activities. The variability in underestimation of EE for the different activities may be problematic for weight loss management applications since accurate EE estimates are important for tracking/monitoring energy deficit.

Comment on "estimating activity and sedentary behavior from an accelerometer on the hip and wrist".

I n the article, ‘‘Estimating Activity and Sedentary Behavior from an Accelerometer on the Hip or Wrist’’ (3), the authors provide evidence that a hip-worn accelerometer estimates activity energy expenditure and classifies activity type and sedentary behavior more accurately than a wrist-worn accelerometer. This study is an important and timely contribution improving our understanding of how accelerometer placement affects the accuracy in estimating physical activity and sedentary behavior metrics. As acknowledged by the authors, this conclusion is based on using simple analytic procedures and summary data from a derived acceleration measure that eliminates much of the raw acceleration signal features that could potentially improve the estimates. Other studies using the raw acceleration signal have shown that the wrist-worn monitor provides reasonably accurate and precise estimates of activity type (4), energy expenditure, activity intensity, and sedentary behavior (1). Use of a derived summary measure from the accelerometer has other limitations. Average accelerometer count values may lead to activity intensity misclassification (2). Specifically, similar average activity counts for activities with different energy expenditure levels and similar energy expenditure levels with different activity counts have been reported. Nevertheless, the authors used these derived accelerometer counts since accelerometer data are typically reported in this manner. Additional evidence about the performance of newer machine learning processing of the raw acceleration signal is necessary to ultimately determine the utility of wrist-worn accelerometers. The authors point out that a wrist-worn accelerometer may be particularly inaccurate for detecting sedentary behaviors since arm movements may occur during sedentary time (i.e., seated and talking with arm movements). It is possible, however, that advanced computational methods may be able to detect these kinds of arm movements and differentiate them from arm activity where energy expenditure is increased. There are some advantages to using a wrist-worn accelerometer. NHANES is currently collecting data using an accelerometer worn on the wrist so that the monitor is worn 24 hId, making it possible to collect sleep measures. In addition, wear compliance is improved with a monitor worn on the wrist. In the first NHANES hip-worn accelerometer study (2003–2006), 40%–70% of participants (compliance varied by age group) had 6+ d of data with 10 hId of wear time. In the current NHANES wrist-worn accelerometer study (2011–2012), compliance has been 70%–80% (6+ d of data) with a median wear time of 21–22 hId (R. Troiano and J. McClain, personal communication). Traditionally, cut points developed and validated in laboratory calibration studies were applied to the data collected in the field. This approach has limitations and likely leads to field-based activity intensity misclassification. Using advanced computation methods applied to the raw acceleration signal may improve estimates of free-living physical activity and sedentary behavior measures. To address this issue, algorithms should first be developed using a broad range of activities that are most prevalent in day-to-day activity. Second, cross validation of the algorithms in the free-living setting is needed where activities do not occur in fixed time intervals and are often performed in sporadic and unpredictable patterns. This strategy will ensure a fair comparison of findings across studies by eliminating the strong influence of activity type on prediction accuracy. While this new method may challenge the traditional research paradigm of calibrating activity monitors exclusively in controlled conditions, it can maximally harness the potential of advanced data processing techniques that adapt and improve prediction. This will be particularly important when there is variability in acceleration patterns due to random movements Address for correspondence: Patty S. Freedson, Ph.D., Department of Kinesiology, University of Massachusetts, Amherst, MA 01003-9258; E-mail: psf@kin.umass.edu.

Effects of body mass index and step rate on pedometer error in a free-living environment.

UNLABELLED Pedometers could provide great insights into walking habits if they are found to be accurate for people of all weight categories. PURPOSE the purposes of this study were to determine whether the New Lifestyles NL-2000 (NL) and the Digi-Walker SW-200 (DW) yield similar daily step counts as compared with the StepWatch 3 (SW) in a free-living environment and to determine whether pedometer error is influenced by body mass index (BMI) and speed of walking. The SW served as the criterion because of its accuracy across a range of speeds and BMI categories. Slow walking was defined as ≤80 steps per minute. METHODS fifty-six adults (mean ± SD: age = 32.7 ± 14.5 yr) wore the devices for 7 d. There were 20 normal weight, 18 overweight, and 18 obese participants. A two-way repeated-measures ANOVA was performed to determine whether BMI and device were related to number of steps counted per day. Stepwise linear regressions were performed to determine what variables contributed to NL and DW error. RESULTS both the NL and the DW recorded fewer steps than the SW (P < 0.001). In the normal weight and overweight groups, error was similar for the DW and NL. In the obese group, the DW underestimated steps more than the NL (P < 0.01). DW error was positively related to BMI and percentage of slow steps, whereas NL error was linearly related to percentage of slow steps. A surprising finding was that many healthy, community-dwelling adults accumulated a large percentage of steps through slow walking. CONCLUSIONS the NL is more accurate than the DW for obese individuals, and neither pedometer is accurate for people who walk slowly. Researchers and practitioners must weigh the strengths and limitations of step counters before making an informed decision about which device to use.

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 Spearmans 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.

Comparison of Raw Acceleration from the GENEA and ActiGraph™ GT3X+ Activity Monitors

Purpose: To compare raw acceleration output of the ActiGraph™ GT3X+ and GENEA activity monitors. Methods: A GT3X+ and GENEA were oscillated in an orbital shaker at frequencies ranging from 0.7 to 4.0 Hz (ten 2-min trials/frequency) on a fixed radius of 5.08 cm. Additionally, 10 participants (age = 23.8 ± 5.4 years) wore the GT3X+ and GENEA on the dominant wrist and performed treadmill walking (2.0 and 3.5 mph) and running (5.5 and 7.5 mph) and simulated free-living activities (computer work, cleaning a room, vacuuming and throwing a ball) for 2-min each. A linear mixed model was used to compare the mean triaxial vector magnitude (VM) from the GT3X+ and GENEA at each oscillation frequency. For the human testing protocol, random forest machine-learning technique was used to develop two models using frequency domain (FD) and time domain (TD) features for each monitor. We compared activity type recognition accuracy between the GT3X+ and GENEA when the prediction model was fit using one monitor and then applied to the other. Z-statistics were used to compare the proportion of accurate predictions from the GT3X+ and GENEA for each model. Results: GENEA produced significantly higher (p < 0.05, 3.5 to 6.2%) mean VM than GT3X+ at all frequencies during shaker testing. Training the model using TD input features on the GENEA and applied to GT3X+ data yielded significantly lower (p < 0.05) prediction accuracy. Prediction accuracy was not compromised when interchangeably using FD models between monitors. Conclusions: It may be inappropriate to apply a model developed on the GENEA to predict activity type using GT3X+ data when input features are TD attributes of raw acceleration.