Statistical analysis
Mean and standard deviation were reported for continuous variables, and frequencies and percentage were reported for categorical variables. Chi-square tests were employed to compare the distribution of BMI categories according to different baseline characteristics. Log-binomial models based on Generalized Estimating Equations (GEE) were employed to estimate the adjusted risk ratios (RRs) and 95% CIs of the six outcomes for underweight, overweight and obesity. In each outcome, three models were fitted.
In Model 1, we adjusted for participants’ sociodemographic characteristics, including age at baseline (19-24 years, 25-29 years, 30-34 years, 35-39 years, or 40-50 years), ethnicity (Han or others), educational level (primary school or below, junior high school, senior high school or college or above), occupation (farmer, worker, servicer or others), region (pearl river delta, non-pearl river delta), migrant population (yes or no). In model 2, we additionally adjusted for history of pregnancy and adverse pregnancy outcomes except for primary caesarean delivery,13 including first pregnancy (yes or no), primipara (yes or no); history of PTB (yes or no), miscarriage (yes or no), induced abortion (yes or no), birth defects (yes or no), or stillbirth (yes or no). In model 3, we additionally adjusted for the lifestyles of the women and the husband, including smoking status of husband before pregnancy and during the early stage of pregnancy (yes or no), smoking and alcohol consumption of women before pregnancy and during the early stage of pregnancy (yes or no), and passive smoking of women before pregnancy (yes or no). Because infant’s sex is associated with the six outcomes, we adjusted for this variable in all analysis in addition to others listed.