Advancing accelerometer-based measurement and analysis of 24-hour physical behaviors in early childhood

Early childhood is a crucial period for establishing healthy 24-hour physical behaviors (i.e., physical activity, sedentary behavior, and sleep), yet evidence on the optimal 24-hour physical behavior patterns for health in young children is currently lacking. A key challenge in advancing this research is the adequate measurement and analysis of these behaviors. Accelerometers provide a promising solution as they continuously collect high-dimensional data on bodily movements. However, there is no consensus on the measurement protocol for young children, and traditional analytical approaches—such as total time-use estimates—fail to capture the temporal accumulation of physical behaviors (i.e., physical behavior patterns). To understand how specific physical behavior sequences are associated with healthy growth and development, more advanced data processing and analytical methods are needed, applicable to diverse and underrepresented populations. This thesis aims to advance the measurement and analysis of 24-hour physical behaviors using accelerometers in early childhood. The first part (Chapters 2—4) focuses on developing an evidence-based protocol. A systematic review showed that no existing accelerometer-based method was both reliable and valid across all 24-hour physical behaviors and age groups, and that differences in accelerometer brands, wear locations, signal axes, and analytical approaches hinder comparability. The comparability of raw acceleration signals across different devices and settings was examined in a laboratory study, demonstrating that consistent data collection and processing protocols are essential for accurate measurement. Building on these insights, the My Little Moves measurement protocol was developed, combining raw accelerometer data from hip- and wrist-worn devices with parent-reported time-use diaries. This protocol captures both physical behavior and their context, including what childen do, how long, how often, where, and with whom. Reliability analyses indicated that at least two full days of parent-reported data are required to obtain valid estimates, which aligned reasonably with accelerometer measurements, although distinguishing self-initiated movement from movement caused by others remained challenging. The second part (Chapters 5—6) focuses on advancing the analysis of temporal physical behavior patterns. A two-dimensional fused targeted ridge estimator quantified the frequency of bouts defined by duration and intensity, treating similar bouts as related. Motif probability, derived using the forward algorithm and parameters from a hidden semi-Markov model, quantified the likelihood of specific sequences of physical behaviors, revealing clusters of children with similar physical behavioral volumes and complexity but distinct motif probabilities. In conclusion, this thesis provides an evidence-based accelerometer protocol and two analytical approaches that integrate intensity and duration to capture capture temporal physical behavior patterns. Future research should focus validation and refinement of these analytical approaches for age-specific segmentation, while promoting transparency throughout the accelerometer data pipeline, thereby supporting more consistent and comparable research across studies and enabling a deeper understanding of how physical behavior patterns relate to healthy growth and development.