This study used data from CHNS, an ongoing population-based cohort study carried out by the national and local government of China. From 1993 to 2009, six waves of cross-sectional surveys were completed in nine provinces (Liaoning, Jiangsu, Henan, Hunan, Guizhou, Heilongjiang, Shandong, Hubei and Guangxi) using a multistage, random-clustering process, with more than 12,000 participants enrolled. The study takes into account Chinese geographic distribution, economic development level and public health resources, and the sample can be considered to provide a representative data set reflecting ordinary Chinese adolescents. The survey collects comprehensive demographic data, including gender, age, education level, dietary and nutritional status, health behavior and clinical data. Each participant signed an informed consent form, and the relevant information related to this study has been published elsewhere .
The study staff conducted face-to-face interviews with participants in 1997, 2000, 2004, 2006 and 2009 to measure height and weight to calculate BMI. In 2009, 9549 fasting blood samples of participants were collected by study staff. We excluded individuals who were either aged < 10 years or > 20 years (n = 8,484) in 1993, pregnant women (n = 24), people without fasting blood glucose (FBG) or glycosylated hemoglobin (HbA1c) information and people whose waist circumference and hip circumference measurements were conducted less than 3 times of follow-up visits (n = 447) (Figure 1).
Measurement and definition
BMI was calculated as weight (kg) divided by height squared (m2). Blood collection and examination were done by professional study staff. All subjects provided 12 mL of blood (in three 4-mL test tubes) after an overnight fast. The levels of high-density lipoprotein cholesterol (HDL-c), total cholesterol (TC), and FBG were assessed using a Hitachi 7,600 machine (Randox, Crumlin, UK; Kyowa, Tokyo, Japan). HbA1c was measured using an HLC-723 G7/D10/PDQ A1c Automated Glycohemoglobin Analyzer (Tosoh Bioscience LLC, Osaka, Japan; Bio-Rad Laboratories, Hercules, CA, USA; Primus Electronics, Morris, IL, USA). Intakes of total energy, carbohydrates, fat and protein were all calculated from participants’ average 3-day dietary intake data obtained by questionnaire. Trained health workers or nurses measured subjects' right arm blood pressure following a standardized procedure using a regularly calibrated mercury sphygmomanometer with a suitable cuff size. Systolic blood pressure (SBP) at the first appearance of pulse sound (Korotkoff stage 1) and diastolic blood pressure (DBP) at disappearance of the pulse sound (Korotkoff stage 5) were recorded. SBP and DBP measurements were repeated 3 times, and the mean value was taken to reduce the influence of measurement error. Height, weight were measured while the subjects were wearing light clothing and no shoes. Waist circumference was measured with tape located at the level of the umbilicus when the participants stood normally with their feet 25-30 cm apart and at the end of the expiratory phase. Educational level and history of hypertension were obtained through self-report. Smoking was defined as previous smoking. Alcohol consumption was defined as drinking alcohol > 3 times/week.
MetS was diagnosed in 2009 according to the International Diabetes Federation criteria , which defined as central obesity (waist ≥ 90 cm in men or ≥ 80 cm in women) plus any two of the following: 1) elevated TGs (> 1.7 mmol/L) or specific treatment for TG abnormality; 2) reduced HDL cholesterol (in men < 1.0 mmol/L and in women < 1.3 mmol/L) or specific treatment for TG abnormality; 3) elevated blood pressure (SBP ≥ 130 mm Hg or DBP ≥ 85 mmHg) or treatment of previously diagnosed hypertension; and 4) elevated fasting plasma glucose (FBG ≥ 5.6 mmol/L) or previously diagnosed type 2 diabetes).
For continuous variables, Student’s t test (for normally distributed variables) /Mann-Whitney U test (for variables with a skewed distribution) and analysis of variance (for normally distributed variables)/Kruskal-Wallis H (for variables with a skewed distribution) were used to detect differences between groups. For categorical variables, the chi-square test/Fisher’s exact test was used to detect differences between groups. We explored the relationship between different BMI development trajectories and incident MetS through logistic regression to calculate the odds ratios (ORs) with 95% confidence intervals (CIs), and the effects of age, sex, BMI, waist circumference, residence, education background, smoking status, alcohol consumption, and nutritional intake were adjusted according to different models.
A latent class growth mixed model (LCGMM) was used to identify different trajectory patterns of BMI since it can take into account random individual variation and within-group variance. Based on the BMI measured three times or more by the participants, a growth model was established according to the measurements over time, while sex was taken as a covariate. Multiple LCGMMs were analyzed with different trajectories to obtain linear and nonlinear model parameters. The model selection criterion was based on the Bayesian information criterion, Vuong-Lo-Mendell-Rubin likelihood ratio test, efficiency of classification and posterior probabilities. After that, the estimated slope and variance in different trajectory classifications were obtained. P < 0.05 (two-sided) indicated statistical significance. All of the analyses were performed with Mplus 8, Stata 15.0 and R (version 3.6.1).