Profiling the littlest movers: Quantity and predictors of toddlers' physical activity levels measured using a novel machine learning model

ObjectiveThe objectives of this study were to: (1) quantify toddlers total physical activity (TPA) and guideline adherence using a machine learning method; and (2) explore socio-ecological predictors (e.g., sex, childcare) of TPA. MethodsToddlers (n=103, 21.4 {+/-} 6.9 months, 52% female) from the Hamilton, Canada region completed a gross motor assessment (Peabody Developmental Motor Scales 2nd ed; PDMS-2) and wore an ActiGraph wGT3X-BT accelerometer on the right hip for 4-8 days. Parents completed demographics and physical activity surveys. TPA was estimated using a validated machine learning model and reported using descriptive statistics. Multiple linear regression explored potential predictors of TPA: age, sex, household income, older sibling, BMI-for-age z-score, gross motor z-score, childcare arrangement, parent physical activity, and temperature, controlling for accelerometer wear time. ResultsToddlers had an average of 200.3 {+/-} 44.0 minutes of daily TPA. Most (72%) met the PA guideline of 180 min/day when averaged across days, while only 27% met the guideline on all days. The regression model was significant and explained 57% of the variation in TPA (F13,79 = 8.09, p < 0.0001). Controlling for wear time, the only significant positive predictors were age and PDMS-2 z-score. ConclusionAlmost three quarters of toddlers met the TPA guidelines. Older toddlers and toddlers with more advanced gross motor skills for their age participated in more daily TPA. Future research should continue to apply machine learning methods in more diverse samples and could build on modifiable predictors (e.g., motor skill) to design interventions to improve toddlers PA levels.