Gabriel M Jouan

Background: Previous research has attempted to determine the minimum number of days of accelerometry required to reliably reflect an individual’s physical activity. However, human behaviors on a day-to-day basis can be highly variable. As a consequence, the number of days required to reliably predict habitual physical activity is dependent on the variability that exists within an individual. There is a concern that adopting generic recommendations from previous research could provide unreliable estimates by failing to represent individuals with specific physical activity patterns. Objective: The main aim of this study was to identify clusters of individuals with distinct physical activity patterns and to determine if the number of days of accelerometry data required to reliably estimate short- (7 days) and medium-term (28 days) physical activity differed between each unique cluster. Methods: Accelerometry data were retrieved from 2 independent research studies. Participants during each study had their physical activity recorded using a Withings Scanwatch (Withings Health Solutions). Following a data eligibility process, agglomerative hierarchical clustering was used to identify clusters of individuals based on their physical activity. The clusters were determined using 4 dimensions; mean, SD, skewness, and kurtosis of the step count data. Intraclass correlation coefficients (ICCs) of step count were then calculated within each physical activity cluster. A series of ICCs were computed by separately comparing the average step count across the full periods (7 and 28, for the short- and medium-term analysis, respectively) to a series of averaged subsamples (ranging from 1-6 days and 1-27 days, for the short- and medium-term analysis, respectively). For each subsample, 500 random combinations were generated and compared, providing a distribution of ICCs for each subsample. An ICC of ≥0.80 identified when the subsample of days was sufficient to achieve appropriate reliability. Results: Of 258 participant datasets, 149 were eligible for the short-term analysis and 64 were eligible for the medium-term analysis. Following agglomerative hierarchical clustering, 4 and 3 clusters of sufficient size (n≥12) were identified in the short-term and medium-term analyses, respectively. When considering the short-term analysis, to achieve a mean ICC score greater than or equal to 0.80, using all randomized combinations, the number of days ranged from 2 to 6 days depending on the physical activity cluster. For the medium-term analysis, the number of days required to achieve a mean ICC score greater than or equal to 0.80 ranged from 6 to 11 days. The short-term analysis clusters displayed more diversity in physical activity patterns than the medium-term analysis. Conclusions: Physical activity patterns influence the number of days required to estimate habitual physical activity. Thus, to avoid unreliable estimates of physical activity, which could significantly impact the interpretation of results, researchers should be mindful of the physical activity patterns of their sample before adopting generic recommendations.

BACKGROUND Previous research has attempted to determine the minimum number of days of accelerometry required to reliably reflect an individual’s physical activity. However, human behaviors on a day-to-day basis can be highly variable. As a consequence, the number of days required to reliably predict habitual physical activity is dependent on the variability that exists within an individual. There is a concern that adopting generic recommendations from previous research could provide unreliable estimates by failing to represent individuals with specific physical activity patterns. OBJECTIVE The main aim of this study was to identify clusters of individuals with distinct physical activity patterns and to determine if the number of days of accelerometry data required to reliably estimate short- (7 days) and medium-term (28 days) physical activity differed between each unique cluster. METHODS Accelerometry data were retrieved from 2 independent research studies. Participants during each study had their physical activity recorded using a Withings Scanwatch (Withings Health Solutions). Following a data eligibility process, agglomerative hierarchical clustering was used to identify clusters of individuals based on their physical activity. The clusters were determined using 4 dimensions; mean, SD, skewness, and kurtosis of the step count data. Intraclass correlation coefficients (ICCs) of step count were then calculated within each physical activity cluster. A series of ICCs were computed by separately comparing the average step count across the full periods (7 and 28, for the short- and medium-term analysis, respectively) to a series of averaged subsamples (ranging from 1-6 days and 1-27 days, for the short- and medium-term analysis, respectively). For each subsample, 500 random combinations were generated and compared, providing a distribution of ICCs for each subsample. An ICC of ≥0.80 identified when the subsample of days was sufficient to achieve appropriate reliability. RESULTS Of 258 participant datasets, 149 were eligible for the short-term analysis and 64 were eligible for the medium-term analysis. Following agglomerative hierarchical clustering, 4 and 3 clusters of sufficient size (n≥12) were identified in the short-term and medium-term analyses, respectively. When considering the short-term analysis, to achieve a mean ICC score greater than or equal to 0.80, using all randomized combinations, the number of days ranged from 2 to 6 days depending on the physical activity cluster. For the medium-term analysis, the number of days required to achieve a mean ICC score greater than or equal to 0.80 ranged from 6 to 11 days. The short-term analysis clusters displayed more diversity in physical activity patterns than the medium-term analysis. CONCLUSIONS Physical activity patterns influence the number of days required to estimate habitual physical activity. Thus, to avoid unreliable estimates of physical activity, which could significantly impact the interpretation of results, researchers should be mindful of the physical activity patterns of their sample before adopting generic recommendations.