In the current investigation, accelerometer-derived physical activity estimates and relationships with health outcomes were compared by length of time sampling interval (i.e., 60s vs. 10s epoch). Findings suggest that although there was a significant difference in physical activity estimates by epoch length, the relationship between physical activity and most health outcomes did not vary by length of data collection. There were, however, a few health outcomes that were more strongly correlated with MVPA presented as 60s epochs when compared to 10s epoch-derived estimates. This finding may relate to the underlying role of physical activity in the causation of obesity such that 60s estimates may better reflect total energy expenditure. However, when examining the equality of the regression slope in these variables by epoch length, no significant differences in relationships were noted. This suggests that measuring the rate of change in health outcomes as a function of physical activity is not dependent on the length of data collection based on current processing methods. Further, additional adjustment for BMI did not elicit strikingly different findings when compared with the unadjusted results.
The current study is not the first to observe differences of intensity classification based on epoch length, but builds upon previous research that was limited to children and adolescents[27–29]. Given the nature and primary purpose of the WOMAN Study, a criterion measure of physical activity was not included for additional comparison of study findings. Regardless, several key sources of error in physical activity outcomes based on 60s data are worthy of noting. First, intermittent activities are easily misclassified, with brief moderate intensity activities often categorized as light intensity once data are summed over a full minute. Similarly, brief periods of movement can result in misclassification of predominately sedentary minutes as light intensity. Second, sustained bouts of activity lasting at least 60s rarely start and stop in synchronization with the accelerometer's internal clock. As a result, a 60s bout of walking, or the beginning and end of a longer walking bout, can easily be split over more than one 60s interval, which could result in misclassification of moderate activities as light intensity. These sources of misclassification can result in substantial shifts between 60 and 10s epoch for minutes detected as light vs. inactive and smaller, but important, shifts between minutes detected as light vs. MVPA. Perhaps most striking is the transfer of roughly 100 min/d detected as light intensity based on 60s epochs to inactive minutes, raising the proportion of inactive or sedentary time from 60% to 72% of monitored time, or an additional 1 hour and 39 minutes of sedentary time per day among these women. Understanding the role that epoch length has in defining physical activity estimates is of critical importance in order to reduce sources of systematic error in research studies.
It is important to note that physical activity estimates were derived from activity count threshold-based processing methods that have been used almost exclusively by researchers over the past decade. However, advancements in accelerometer technology have enhanced the level of sophistication of available data processing methods and interpretation of derived summary estimates. Early efforts by Crouter et al., utilized a two regression approach that predicted intensity from separate equations based on the variability of the detected activity sampled in 10s epochs[30, 31]. However, as previously stated, the ActiGraph and other devices are now capable of sampling data using 1s epochs or in raw acceleration mode to provide a dense data stream for use in emerging signal processing and pattern recognition models (e.g., artificial neural networks)[2, 32]. Recent progress in accelerometer processing methods offer promise for increased accuracy of physical activity estimates from body worn accelerometers; precision that could potentially improve the ability to observe significant relationships with health outcomes beyond those observed in the current study. In the current report, accelerometer data was sampled using 10s epochs in order to utilize data processing methods that were available at the time the 48 month follow-up visits were initiated. It is important to note that these results were gleaned from 2 epoch lengths and examined in a specific population sub-group. However, since high frequency accelerometer data can be reintegrated to a lower resolution, as was done in the current report (i.e., 10s to 60s epochs), with no additional burden, we would encourage researchers to collect data at the maximum allowable resolution for the desired monitoring period in order to preserve the ability to utilize the latest advancements in accelerometer data processing methods for use in epidemiological studies as they become available.
When interpreting the findings, several limitations need to be considered. Given the cross-sectional nature of the data, no inferences should be made about the causality of relationships between physical activity and health outcomes. Due to limited resources, accelerometer data was only collected in one-quarter of WOMAN Study participants who attended the 48 month follow-up visit. Although collecting accelerometer data in all participants attending the 48 month follow-up visit would have increased the power to see more significant physical activity-health outcome relationships, few differences were noted between those who were included in the accelerometer sub-study and those who were not. Finally, waist-worn, uni-axial accelerometers provide an accurate measure of predominantly ambulatory activities and, thus, do not capture all physical activities that may contribute to an improvement in health outcomes. Furthermore, comparisons with self-reported leisure physical activity may also be limited as the past-year MAQ includes non-ambulatory activities, but not lower intensity activities (i.e., household chores), which may be captured with accelerometers. Therefore, the weak or null relationships between health outcome and physical activity that were observed in the current study may be a reflection of the limited quantification of total physical activity or simply the result of the relatively homogeneous nature of the study sample.