Study design
The data of this study were derived from the CHNS, which was a prospective household-based study to longitudinally measure how sociological, economic and demographic factors change health and nutritional status across the life span of the Chinese population using a follow-up interval of two or three years. The CHNS selected individuals of various ages in ten rounds of surveys from 1989 to 2015, who lived in twelve diverse provinces including Shandong, Liaoning, Heilongjiang, Jiangsu, Henan, Guizhou, Hunan, Hubei, Zhejiang, Yunnan, Shanxi and Guangxi and three autonomous cities including Beijing, Shanghai and Chongqing (in 2011 and 2015). A multi-stage, stratified, random cluster sampling design was used to ensure a balanced representation of urban, suburban and rural areas. All data were collected by trained and certified health workers.
Study population
We analyzed the 2015 CHNS, the first wave to collect BIA data. We limited the sample to the participants aged between 40 and 64 years old with complete data on total PA and BF%. Participants with implausible BF% values (<5%,>70%) were excluded from our study. We excluded individuals who were disabled or women who were pregnant or breastfeeding during the survey year (n=74) and subjects who had received diagnoses of hypertension, diabetes, myocardial infarction, stroke, cancer, fracture and asthma (n=1551). Individuals with implausible energy intakes values (<800 kcal/day or >6000 kcal/day for men, <600 kcal/day or >4000 kcal/day for women, n=236) (21-23) and BMI (<10 kg/m2, >60 kg/m2, n=5) were excluded. Our final analysis sample included 5763 observations with 2540 men and 3223 women.
Anthropometrics
Trained health workers measured weight, height, body fat percentages (including the total BF%, trunk BF%, and arm and leg BF%). Using a body composition analyzer (BC601, TANITA, method of BIA) weight was measured to the nearest 0.1 kg with the participant standing without shoes and wearing a single layer of clothing. Height was measured without shoes to the nearest 0.1 cm using SECA 206 wall-mounted metal tapes. Percentages of body fat were calculated using software with aproprietary algorithm which requires age, gender, height, and physical activity level inputs by technicians. According to the Chinese definitions, we defined the overweight and obesity based on BMI (the ratio of weight in kilograms and height in meters2) cutoff points of 24 and 28 kg/m2, respectively.
Total physical activity and sedentary behavior assessment
We used the standard physical activity questionnaire to calculate the average metabolic equivalents of task (MET) hours per day to indicate the PA level, which comprised four PA domains: travel, occupational, leisure and domestic. The ratio of a person’s working metabolic rate relative to their resting (basal) metabolic rate defined was the definition of a MET unit. Thus, the average MET-hours per week measurements comprised both the time spent in each activity and the average intensity of each activity (or sub-activity). We categorized the total MET-hours per day into slightly PA, moderate PA, moderate-vigorous PA (MVPA) and vigorous PA according to the interquartile. Details on how these values were calculated were described elsewhere (24-26).
Sedentary behaviors were calculated as the average hours per day (hr/day) spent in various non-occupational recreational activities, including watching TV or movies/videos, reading/writing, board games and computer usage. Time spent engaged in these kinds of activities were summed to obtain total time expenditure on sedentary behaviors, which was not included in the physical activity calculation.
Assessment of covariates
Standard questionnaires were used by trained interviewers to collect sociodemographic characteristics, annual per household income, smoking, alcohol consumption, community information (urbanization index) and dietary intake. We categorized the educational level into low, medium and high for primary school education or less, middle school education and high school education and above, respectively. Marital status was grouped into two statuses (married and single). Participants reported their gross annual per capita household income according to household size, which was inflated to 2015 values and categorized into tertiles. Smoking status was classified as current smoking and no smoking now. Alcohol consumption refers to whether or not participants have drunk alcohol in the past year.
Energy intake per day and percentage of energy from fat were used as continuous variables, which were collected using a weighing method of condiments in combination with three consecutive 24-hour recall calculated using the China Food Composition Table over the same three-day period (21).
Region was grouped as Northern China (Heilongjiang, Liaoning and Beijing), Central China (Shangdong, Jiangsu, Shanghai, Henan and Shanxi) and Southern China (Hubei, Chongqing, Zhejiang, Guizhou, Hunan, Guangxi and Yunnan) due to climate and dietary habit differences. Community urbanization index was calculated based on 12 multidimensional components at the community level reflecting population density, economic activity, traditional markets, modern markets, transportation infrastructure, sanitation, communications, housing, education, diversity, health infrastructure and social services (27, 28), and was categorized into tertiles (high, middle and low). Others details were presented in previous analyses of the CHNS (25).
Statistical analysis
We calculated descriptive statistics for the individual demographic variables, which were stratified by gender. Continuous variables were expressed as medians, the 25th percentile (Q1), and the 75th percentile (Q3). Categorical variables were expressed as percentages. BF% was described as nuclear density figure by gender and BMI groups. Median (Q1, Q3) PA was described by gender and sociodemographic variables and Kruskal-Wallis tests were used for sociodemographic variables. Box plots were used to express the BF% by PA levels.
Due to a significant statistical interaction between gender and overweight/obese across the 10th , 25th , 50th , 75th , and 90th percentiles (P<0.0001), we use separate gender-BMI-stratified QR analyses to assess associations of the total PA, at total body and trunk BF% percentiles. 3 to 6 MET was defined as moderate intensity exercise (29), than we used 4.5 MET·h/d, such as half step jogging (30), to represent the average level of it and tested the association as an additional hour of moderate PA with lower BF%. Compared with the traditional linear regression based on means, QR allowed us to evaluate the distribution of BF% at several cut points, did not require any assumption about the distribution of the regression residuals, and was not influenced by skewness in the distribution of the dependent variable. This method provided greater statistical efficiency when outliers are present and is robust to varying effects of covariates at different percentiles of the response variable (31-33).
We constructed three models. Model 1 was controlled for the total PA only. The individual-level variables were then added to the equation to estimate Model 2, including the sedentary activity time, age, educational level, marital status, household income level, energy intake, energy percentage from fat, BMI, smoking status, alcohol consumption status and region. For Model 3, urbanization index level was add to Model 2. BF% included the total body and the trunk. P<0.05 were considered statistically significant. SAS 9.4 (SAS Institute, Inc., Cary, NC, USA) was conducted in descriptive and QR analyses.