Physical Activity Measurement in Toddlers

Physical inactivity tends to track over time and many young children aged 3 to 4 years are physically inactive. Understanding when and how physical activity habits develop requires investigation starting at an even younger age. However, a knowledge gap exists in physical activity evaluation among toddlers (aged 1 and 2 years), which primarily resulted from a major methodological weakness in physical activity measurement for toddlers, particularly related to accelerometer data processing. To process accelerometer data for physical activity estimation among toddlers, an intensity-based accelerometer count cut-point approach has been widely used. However, our pilot study showed that this cut-point approach could misclassify toddler’s unique activities. For example, based on the suggested cut-points, being “carried” by adults would be misclassified as moderateto vigorous-intensity activity. Recently, machine learning techniques have been adopted to recognize activity types. A few studies among preschoolers as well as older children have demonstrated a high accuracy of machine learning activity recognition algorithms. Our pilot study also showed that the machine learning approach offers great potential for recognition of activity types among toddlers. Thus, it is promising that machine learning techniques can complement the cut-point method for physical activity measurement among toddlers. The objective of this project is to develop and validate an accelerometer-based activity recognition algorithm for habitual physical activity estimation among toddlers, using the machine learning approach. The first aim is to develop and validate accelerometer-based activity recognition algorithms for toddlers. The second aim is to describe physical activity, sedentary behavior, and weight trajectories from age 12 to 36 months. To achieve these aims, 124 toddlers at age 12 months will be recruited. They will participate in five waves of assessment at age 12, 18, 24, 30, and 36 months. At each assessment, participants will wear an ActiGraph accelerometer on the hip during an activity trial, a 20-minute free-play session, and a 7-day activity monitoring period. Using the activity trial data, an activity recognition algorithm will be developed to classify activity types into ambulation, other physical activity, sedentary behavior, and being “carried.” Using the freeplay data, the developed algorithm will be validated. The validated algorithm will then be applied to the 7-day accelerometer data to estimate daily time spent in physical activity and sedentary behavior among toddlers. We will explore whether physical activity development is related to changes in body mass index percentile from age 12 to 36 months. This study will help to fill the methodological gap in physical activity measurement among toddlers and produce necessary epidemiologic data on physical activity in early childhood.