Statistical analysis
All data were analyzed with Stata 16.1 (Stata Corp, College Station, TX, USA) and SPSS 25.0 software (SPSS, Chicago, IL, USA). Continuous variables were presented as mean ± SD and categorical variables were presented as numbers (proportion). The comparisons of serum TG levels among different periods and among groups were performed by one-way ANOVA for continuous variables. Bonferroni test was used for pairwise comparisons if homogeneity of variance is satisfied, and Tamhane T2 test was used otherwise. Pearson correlation and stepwise multiple liner regression analysis were carried out to analyze the associations between serum TG levels during pregnancy and 42 days postpartum. Serum TG level at 42 days postpartum was the dependent variable. Serum TG level at each gestational week, age, GWG, pre-BMI, GDM and gestational age were the independent variables. If α ≤ 0.05, the variable was entered into the model, and the variable was ruled out if α ≥ 0.10. A two-stage approach 12 was carried out to analyze the associations between TG trend of change during pregnancy and TG level and hypertriglyceridemia at 42 days postpartum. In the first stage, a linear mixed-effect model (LME) was constructed for TG as a function of gestational week at sampling and predicted the best linear unbiased predictor (BLUP) of random intercept and slope. The predicted intercept represents the mean TG level at the 6-8th gestational week. The predicted slope represents the trend of TG level changes throughout gestation. Secondly, the BLUP estimates of intercept and slope were used as predictors in the linear and logistic regression models including TG level at 42 days postpartum and hypertriglyceridemia as the outcomes, respectively. The odds ratios (ORs) and 95% confidence intervals (95% CIs) of postpartum hypertriglyceridemia in association with gestational TG levels were calculated by using logistic regression analysis. The model was adjusted for age, GWG, pre-BMI and GDM. Cutoff points of serum TG level at each gestational week were chosen by the point nearest to the top-left most corner of the receiver operating characteristic (ROC) curves, which was selected by the biggest Youden Index 13. Sensitivity, specificity, and 95% CIs were calculated. The area under the ROC curve (AUC) was used to evaluated discrimination of the predictive model of gestational TG level on postpartum hypertriglyceridemia. The Hosmer-Lemeshow goodness-of-fit test 14 was carried out for assessment of calibration. Hierarchical logistic regression model was used for subgroup analysis among different pre-BMI subgroups and between GDM subgroups.P value of < 0.05 was considered statistically significant.