Physical activity, entertainment screen time, and dietary fat profiles and concurrent weight status among school-attending Chinese children and adolescents: CHNS 2004–2011

Abstract

Physical activity, screen-based leisure, and diet co-occur, but profiles estimated from a small set of indicators may be sensitive to historical period and model assumptions. We identified profiles defined by total physical activity, reported entertainment screen time, and dietary fat contribution to energy intake in the 2004–2011 China Health and Nutrition Survey (CHNS), and examined their concurrent associations with weight status. This pooled cross-sectional analysis included 4,112 person-wave observations from school-attending participants aged 6–17 years. A prespecified Gaussian mixture analysis compared one through six profiles. Associations with measured overweight/obesity were estimated in 4,068 observations from 2,757 participants using posterior-probability pseudo-class draws and participant-clustered robust standard errors. Wave-, age-, model-specification-, missing-data-, and alternative-outcome analyses assessed robustness. The prespecified model retained five profiles: very low activity (6.9%), low-to-moderate activity/lower fat (37.8%), no reported entertainment screen time (5.4%), higher fat (27.8%), and highly active (22.2%). The exact BIC-preferred number of profiles varied across survey waves, age groups, and model specifications, and 48.4% of observations had a maximum posterior probability below 0.70. The estimated prevalence of overweight/obesity, accounting for profile-membership uncertainty, was highest in the very-low-activity profile (24.4%) and was 16.4% in the reference profile. This descriptive difference attenuated after adjustment (odds ratio, 1.18; 95% confidence interval, 0.81–1.69), and the fully adjusted global profile test was non-significant ( p  = 0.806). Multinomial models separating overweight and obesity, continuous BMI and approximate WHO BMI-for-age z-score models, multiple imputation, and inverse-probability weighting produced the same overall conclusion. In the 2004–2011 CHNS analytic sample, the measured behaviors formed heterogeneous combinations, but the selected profiles did not independently distinguish concurrent weight status after adjustment. The findings are historical and sample-specific; they should not be interpreted as current national profile distributions or as validation of profile-based interventions.