You are looking at 31 - 40 of 174 items for :

  • Journal for the Measurement of Physical Behaviour x
  • Sport and Exercise Science/Kinesiology x
  • Psychology and Behavior in Sport/Exercise x
  • Refine by Access: All Content x
Clear All
Free access

Evaluation of Within- and Between-Site Agreement for Direct Observation of Physical Behavior Across Four Research Groups

Sarah Kozey Keadle, Julian Martinez, Scott J. Strath, John Sirard, Dinesh John, Stephen Intille, Diego Arguello, Marcos Amalbert-Birriel, Rachel Barnett, Binod Thapa-Chhetry, Melanna Cox, John Chase, Erin Dooley, Rob Marcotte, Alexander Tolas, and John W. Staudemayer

Direct observation (DO) is a widely accepted ground-truth measure, but the field lacks standard operational definitions. Research groups develop project-specific annotation platforms, limiting the utility of DO if labels are not consistent. Purpose: The purpose was to evaluate within- and between-site agreement for DO taxonomies (e.g., activity intensity category) across four independent research groups who have used video-recorded DO. Methods : Each site contributed video files (508 min) and had two trained research assistants annotate the shared video files according to their existing annotation protocols. The authors calculated (a) within-site agreement for the two coders at the same site expressed as intraclass correlation and (b) between-site agreement, the proportion of seconds that agree between any two coders regardless of site. Results: Within-site agreement at all sites was good–excellent for both activity intensity categories (intraclass correlation range: .82–.9) and posture/whole-body movement (intraclass correlation range: .77–.98). Between-site agreement for intensity categories was 94.6% for sedentary, 80.9% for light, and 82.8% for moderate–vigorous. Three of the four sites had common labels for eight posture/whole-body movements and had within-site agreements of 94.5% and between-site agreements of 86.1%. Conclusions: Distinct research groups can annotate key features of physical behavior with good-to-excellent interrater reliability. Operational definitions are provided for core metrics for researchers to consider in future studies to facilitate between-study comparisons and data pooling, enabling the deployment of deep learning approaches to wearable device algorithm calibration.

Restricted access

Comparison of Six Accelerometer Metrics for Assessing the Temporal Patterns of Children’s Free-Play Physical Activity

Katherine L. McKee, Karin A. Pfeiffer, Amber L. Pearson, and Kimberly A. Clevenger

Accelerometers are frequently used to measure physical activity in children, but lack of uniformity in data processing methods, such as the metric used to summarize accelerometer data, limits comparability between studies. The objective was to compare six accelerometer metrics (raw: mean amplitude deviation, Euclidean norm minus one, activity index, monitor-independent movement summary units; count: vertical axis, vector magnitude) for characterizing the intensity and temporal patterns of first and second graders’ (n = 88; age = 7.8 ± 0.7 years) recess physical activity. At a 5-s epoch level, Pearson’s correlations (r) between metrics ranged from .66 to .98. When each epoch was classified into one of four intensity levels based on quartiles, agreement between metrics as indicated by weighted kappa ranged from .81 to .96. When collapsed to time spent in each intensity level, metrics were strongly correlated (r = .76–.99) and most often statistically equivalent for estimating time spent in Quartile 3 or 4. Children were ranked from least to most active, and agreement between metrics was strong (Spearman’s correlation ≥ .87). Temporal patterns were characterized using five fragmentation indices calculated using each of the six metrics, which were fair-to-strongly correlated (r = .53–.99), with the strongest associations for number of high-intensity activity bouts (r ≥ .89). Most fragmentation indices were not statistically equivalent between metrics. While metrics captured similar trends in activity intensity and temporal patterns, caution is warranted when making comparisons of point estimates derived from different metrics. However, all metrics were able to similarly capture higher intensity activity (i.e., Quartile 3 or 4), the most common outcome of interest in intervention studies.

Restricted access

Impact of Continuous Focal Sampling Time Thresholds on Physical Activity Metrics When Using Video-Recorded Direct Observation

Julian Martinez, John Staudenmayer, and Scott J. Strath

Purpose: To determine differences in physical activity metrics between 1-, 5-, and 10-s direct observation (DO) time thresholds and compare annotation completion time between different time thresholds. Methods: Participants (n = 10, mean age 40.7 ± 22.3 years, five males) were video recorded for 2 hr within a free-living setting. DO videos were annotated by one experienced annotator with a priori developed Posture and Behavior schemas. The annotation order of video, time threshold, and schema used was randomized. For analysis, annotations were collapsed into posture and behavior domains. Total video time is reported. Time to code videos, overall percent agreement, and statistical bias of each posture and behavior domain for the 5-s time threshold and 10-s time threshold were compared to 1-s time threshold output. Results: 19.7 hr of DO were recorded. On average, the 1-s time threshold took 183.9 ± 34.2 min to annotate with the Posture schema and 118.8 ± 23.6 min with the Behavior schema. Under the Posture schema, the 5-s time threshold was 31.7% faster, had 91.5% agreement, and all biases were <±5 min, while the 10-s time threshold was 43.6% faster, had 89.2% agreement, and had biases ranging from −7.59 to 5.21 min. Under the Behavior schema, the 5-s time threshold was 16.0% faster, had 92.0% agreement, and had all biases <±2.1 min, while the 10-s time threshold was 27.6% faster, had 88.3% agreement, and had all biases <±3.9 min. Conclusion: Longer DO annotation time thresholds are accurate and faster but less precise for certain posture and behavior domains when compared to criterion 1-s time threshold in healthy adults.

Restricted access

Comparative Analysis and Conversion Between Actiwatch and ActiGraph Open-Source Counts

Paul H. Lee, Ali Neishabouri, Andy C.Y. Tse, and Christine C. Guo

Body-worn sensors have contributed to a rich and growing body of literature in public health and clinical research in the last decades. A major challenge in sensor research is the lack of consistency and standardization of the collection and reporting of the sensor data. The algorithms used to derive these activity counts can be vastly different between manufactures and not always transparent to the researchers. With Philips, one of the major research-grade wearable device manufacturers, discontinuing this product line, many researchers are left in need of alternative solutions and at the risk of not being able to relate their historical data using the Philips Actiwatch 2 devices to future findings with other devices. We herein provide a comparison analysis and conversion method that can be used to convert activity counts from Philips to those from ActiGraph, another major manufacturer who provide both raw acceleration data and count data based on their open-source algorithm to the research community. This work provides an approach to maximize the scientific value of historical actigraphy data collected by the Actiwatch devices to support research continuity in this community. The conversion, however, is not perfect and only offers an approximation, due to the intrinsic difference in the count algorithms between the two accelerometers, and the permanent information loss during data reduction. We encourage future research using body-worn sensors to retain the raw sensor data to ensure data consistency, comparability, and the ability to leverage future algorithm improvement.

Restricted access

Applying Average Real Variability to Quantifying Day–Day Physical Activity and Sedentary Postures Variability: A Comparison With Standard Deviation

Madeline E. Shivgulam and Myles W. O’Brien

Intraindividual activity variability is often overlooked, with some existing work using SD as a variability metric. However, average real variability (ARV) may be a more suitable metric as it accounts for temporal variability. The purpose of this exploratory study was to (a) apply ARV analyses to habitual activity outcomes; (b) assess the agreement between ARV and SD for habitual step counts, standing time, and sedentary time; and (c) determine the relationship between activity variability (SD and ARV) with average activity values. One hundred and eighty-nine participants (37 ± 22 years, 109 females) wore the activPAL inclinometer on their thigh 24 hr/day for 6.4 ± 0.9 days. SD and ARV were calculated for each participant across their wear time. A Wilcoxon signed-rank test revealed that ARV was significantly higher than SD for step count, standing time, and sedentary time (all, p < .001). Equivalence testing demonstrated mixed equivalence for step counts (10%), standing time (12%), and sedentary time (14%). SD and ARV were highly correlated to each other for all activity metrics (all, ρ > .857, p < .001). SD was moderately (ρ = .601, p < .001) and weakly (ρ = .296, p < .001) correlated with average step count and standing time, respectively. ARV was weakly correlated with average step count and standing time (both: ρ < .499, p < .001). However, average sedentary time was not associated with SD or ARV (both, p > .177). While the two measurements of variability were strongly correlated, they cannot be used interchangeably. More monitoring research should consider intraindividual activity variability and use methods, such as ARV, that consider the temporal nature of day–day activity.

Restricted access

Interchangeability of Research and Commercial Wearable Device Data for Assessing Associations With Cardiometabolic Risk Markers

Andrew P. Kingsnorth, Elena Moltchanova, Jonah J.C. Thomas, Maxine E. Whelan, Mark W. Orme, Dale W. Esliger, and Matthew Hobbs

Introduction: While there is evidence on agreement, it is unknown whether commercial wearables can be used as surrogates for research-grade devices when investigating links with markers of cardiometabolic risk. Therefore, the aim of this study was to investigate whether data from a commercial wearable device could be used to assess associations between behavior and cardiometabolic risk markers, compared with physical activity from a research-grade monitor. Methods: Forty-five adults concurrently wore a wrist-worn Fitbit Charge 2 and a waist-worn ActiGraph wGT3X-BT during waking hours over 7 consecutive days. Log-linear regression models were fitted, and predictive fit via a one-out cross-validation was performed for each device between behavioral (steps, and light and moderate-to-vigorous physical activity) and cardiometabolic variables (body mass index, weight, body fat percentage, systolic and diastolic blood pressure, glycated haemoglobin, grip strength, estimated maximal oxygen uptake, and waist circumference). Results: Overall, step count was the most consistent predictor of cardiometabolic risk factors, with negative associations across both Fitbit and ActiGraph devices for body mass index (−0.017 vs. −0.020, p < .01), weight (−0.014 vs. −0.017, p < .05), body fat percentage (−0.021 vs. −0.022, p < .01), and waist circumference (−0.013 vs. −0.015, p < .01). Neither device was found to provide a consistently better prediction across all included cardiometabolic risk markers. Conclusions: Step count data from a commercial-grade wearable device showed similar associations and predictive relationships with cardiometabolic risk markers compared with a research-grade wearable device, providing preliminary support for their use in health research.

Open access

Prediction Strength for Clustering Activity Patterns Using Accelerometer Data

Jingzhi Yu, Kristopher Kapphahn, Hyatt Moore, Farish Haydel, Thomas Robinson, and Manisha Desai

Background: Clustering, a class of unsupervised machine learning methods, has been applied to physical activity data recorded by accelerometers to discover unique patterns of physical activity and health outcomes. The prediction strength metric provides a criterion to determine the optimal number of clusters for clustering methods. The aim of this study is to provide specific guidance for applying prediction strength to time series accelerometer data. Methods: For this purpose, we designed an extensive simulation study. We created a synthetic data set of accelerometer data using data from a childhood obesity management trial. We evaluated the role of a prespecified threshold of the prediction strength metric as a key input parameter. We compared the recommended threshold (between 0.8 and 0.9) with an approach we developed (Local Maxima). Results: The choice of threshold had a large impact on performance. When the noise level increased (greater overlap between true clusters), lower thresholds outperformed the recommended threshold, which tended to underestimate the true number of clusters. In addition, we found that sorting the data by magnitude of intensity in windows within the time series of interest prior to clustering alleviated sensitivity to threshold choice. Furthermore, for accelerometer data, we recommend that the Local Maxima approach be utilized together with a graphical evaluation of the prediction strength metric function over values of k. Finally, we strongly suggest sorting of the data prior to clustering if sorting retains meaning for the research question at hand. Conclusion: Our recommendations can help future researchers discover more robust patterns from accelerometer data.

Restricted access

Volume 6 (2023): Issue 2 (Jun 2023)

Restricted access

Calibrating Physical Activity and Sedentary Behavior for Hip-Worn Accelerometry in Older Women With Two Epoch Lengths: The Women’s Health Initiative Objective Physical Activity and Cardiovascular Health Calibration Study

Kelly R. Evenson, Fang Wen, Christopher C. Moore, Michael J. LaMonte, I-Min Lee, Andrea Z. LaCroix, and Chongzhi Di

Purpose: The purpose of this study was to develop 60-s epoch accelerometer intensity cut points for vertical axis count and vector magnitude (VM) output from hip-worn triaxial accelerometers among women 60–91 years old. We also compared these cut points against cut points derived by multiplying 15-s epoch cut points by four. Methods: Two hundred apparently healthy women wore an ActiGraph GT3X+ accelerometer on their hip while performing a variety of laboratory-based activities that were sedentary (watching television and assembling a puzzle), low light (washing/drying dishes), high light (laundry and dust mopping), or moderate-to-vigorous physical activity (400-m walk) intensity. Oxygen uptake was measured using an Oxycon portable calorimeter. Sedentary behavior and physical activity intensity cut points for vertical axis and VM counts were derived for 60-s epochs from receiver operating characteristic and by multiplying the 15-s cut points by four; both were compared with oxygen uptake. Results: The median age was 74.5 years (interquartile range 70–83). The 60-s epoch cut points for vertical counts were 0 sedentary, 1–73 low light, 74–578 high light, and ≥579 moderate-to-vigorous physical activity. The 60-s epoch cut points for VM were 0–88 sedentary, 89–663 low light, 664–1,730 high light, and ≥1,731 moderate-to-vigorous physical activity. For both sets of cut points, the receiver operating characteristic approach yielded more accurate estimates than the multiplication approach. Conclusion: The derived 60-s epoch cut points for vertical counts and VM can be applied to epidemiologic studies to define sedentary behavior and physical activity intensities in older adult populations.

Restricted access

Conceptualizing, Defining, and Measuring Before-School Physical Activity: A Review With Exploratory Analysis of Adolescent Data

James Woodforde, Sjaan Gomersall, Anna Timperio, Venurs Loh, Hannah Browning, Francisco Perales, Jo Salmon, and Michalis Stylianou

Physical activity (PA) among children and adolescents is often reported by time segments centered around the school day, including before school. However, there is no consistent approach to defining the before-school segment, to accurately capture PA levels and facilitate synthesis of results across studies. Therefore, this study aimed to (a) examine how studies with children and adolescents have defined the before-school segment, and (b) compare adolescents’ before-school PA using various segment definitions. We conducted a systematic search and review of literature from six databases, and subsequently analyzed accelerometer data from Australia (n = 472, mean age 14.9 years, 40% male), to compare PA across five before-school definitions. Our review found 69 studies reporting before-school PA, 59 of which used device-based measures. Definitions ranged widely, but justifications were rarely reported. Our empirical comparison of definitions resulted in a range of participants meeting wear time criteria (≥3 days at >50% of segment length) from the latest-starting definition (30 min prior to school; n = 443) to the earliest-starting definition (6:00 a.m.–school start; n = 155), implying that for many participants, accelerometer wear was low in the early hours due to sleep or noncompliance. Statistically significant differences in light and moderate-to-vigorous PA (mean minutes/school day, proportion of segment length, and proportion of wear time) were found between definitions, indicating that before-school PA could potentially be underestimated depending on definition choice. We recommend that future studies clearly report and justify segment definition, apply segment-specific wear time criteria, and collect wake time data to enable individualized segment start times and minimize risk of data misclassification.