Participants
This study utilized the data from (China Health and Nutrition Survey, CHNS)[15], which is an ongoing cohort study jointly conducted by the China Center for Disease Control and Prevention (China CDC) and Carolina Population Center at the University of North Carolina. The project started in 1989 and a multistage, random cluster process was employed to draw samples from 9 provinces including Heilongjiang, Liaoning, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, and Guizhou. Thereafter, nine rounds of follow-up surveys were conducted in 1991, 1993, 1997, 2000, 2004, 2006, 2009, 2011 and 2015. Since 2011, samples from three megacities (Beijing, Shanghai and Chongqing) have been added. The recruitment efforts and sampling strategies have been described previously [13]. To facilitate comprehensive assessment of PA, questions related to sedentary behaviors were added since 2004. For this reason, we only included children aged 6-17 years who participated in at least one round of follow-up surveys from 2004 to 2015. Records with illogical or missing data were excluded. A total of 298 primary sampling units, commonly known as “communities”, were included. This study was approved by the Medical Ethics Committee of the National Institute for Nutrition and Health, China CDC and University of North Carolina.
Data Collection
Data on PA, sedentary behaviors and socio-demographic characteristics were collected using structured questionnaires.
PA outcome measures comprises PA level, the average daily time spent in PA (min/day) and average weekly volume of PA (MET-hrs/week). To estimate the average weekly volume of PA, the average weekly time spent in various PA was multiplied by the corresponding metabolic equivalent of task (MET) values based on the “Youth Compendium of Physical Activities” developed by the US Center for Chronic Disease Prevention and Health Promotion in 2017[16]. In the compendium, MET values of 196 activities are presented for four age-groups: 6–9, 10–12, 13–15, and 16–18 years. With reference to the Physical Activity Guidelines for Chinese Children[17], moderate- and vigorous-intensity PA (MVPA) were identified, and the average time spent in MVPA per day (min/d) were estimated. Using the 60 minutes of accumulated MVPA per day proposed by WHO as cut off point, we defined less than 60 minutes of accumulated MVPA per day as physical inactivity. PA was categorized into three levels based on the average time spent in MVPA per day: <60 min, 60-120 min and > 120 min for low, medium and high levels of PA, respectively. PA was classified into four domains: active leisure, in-school PA, active travel and domestic PA. Active leisure and in-school PA included strolling, gymnastics, track and field, martial arts, ball games and others; active travel included walking and cycling; domestic PA included cleaning the house, doing laundry, cooking and buying food.
Sedentary behaviors were measured as the average weekly time (hrs/week) spent in sedentary behaviors before school, after school and during weekends. Sedentary activities were grouped into four categories: education, screen-based entertainment, transportation, and Arts&Play. Activities in the education category included extracurricular reading, writing, and painting; Activities in the screen-based entertainment category included TV (watching TV, videos, video discs), games (playing computer or smartphone games, playing game consoles), and internet (chatting online, watching programs online or on smartphones); Activities in the transportation category included traveling as passenger by bicycles, buses, subways, cars, taxis, motorcycles; Activities in the Arts&Play category included chess, toy cars, puppets etc.
Socio-demographic characteristics included age, gender, ethnicity, annual family income, parental education levels and geographical factors. Since the average of 12 years old is the beginning of the development of puberty and the secondary study, age was stratified into 6-11 year-old and 12-17 year-old groups. The annual family income was stratified into low (<50,000 yuan) and high (>50,000 yuan) income groups. Parental education was stratified into low (primary or below), middle (secondary completed), and high (college or higher) education levels. Geographical factors included residential areas (urban / rural), region (north / south) and urbanization level of community. Urbanization level of community was assessed using an urbanization index developed by Jones Smith and Barry Popkin. The index was produced from a comprehensive evaluation of 12 dimensions such as population density, education, and transportation infrastructure[18]. In this study, the urbanization index for each survey year was compiled and stratified into three groups: low, medium and high levels of urbanization.
Data Analysis
All data analyses were performed using SAS 9.4 software and p-value < 0.05 was considered as statistically significant. Categorical data were presented as frequency (percentage). The Cochran-Armitage trend test was used to analyse the trends in proportion of gender, residential areas and region over time. The Fisher’s exact test was used for ethnicity as the expected count of cells in this group was less than 5. As paternal education levels and maternal education levels are ordinal multilevel variables, the Mantel-Haenszel chi-square test was used. As age, annual family income and urbanization index of community are continuous variables, the mean ± standard deviation was presented and the multiple linear model was used to test the trends after adjusting other socio-demographic characteristics.
The random-effects ordinal regression model was conducted to examine the trends in PA levels across survey years. The Newton-Raphson Ridge optimization method was used to optimize parameter estimation, controlling for the random effect of communities and adjusting for socio-demographic factors including gender, age, ethnicity, paternal education level, maternal education level, annual family income, urbanization index of community, residential area and region. The trends were further analysed after applying stratification on gender.
Repeated measures mixed models were conducted using volume of PA and time spent in sedentary behaviors as dependent variables, survey year as an independent variable, controlling for all socio-demographic characteristics. Community was included as random effect in the models. The adjusted means of volume of PA and time spent in sedentary behaviors was reported. The trends in both outcomes were examined and the differences across survey years were compared using Bonferroni method. The trends were further analysed after applying stratification on gender, age, urbanization level of community, residential area and region. The repeated measures mixed effects model considers the internal connection of the observations in different survey years, and the aggregation of the observations in the community level. The hypothesis test showed that the change of the -2log-likelihood value of the random intercept model and the random slope model is statistically significant (p < 0.001), indicating that the fitting of the random slope model was better than the random intercept model, so the random slope model was adopted.
Quantile regression models were used to examine trends and the differential effects of correlates at different quantiles of volume of PA and time spent in sedentary behaviors. Quantile regression uses conditional quantile modelling of the dependent variable to estimate the regression parameters by minimizing the weighted sum of the absolute values of the residuals. Since there is no special requirement for the distribution of the dependent variable or the homogeneity of variance, it is not affected by outliers. This is a robust method that can reflect the influence of independent variables on dependent variables at different levels[19].