Statistical analysis
Data analyses were performed by the SPSS 20.0 (Inc., Chicago, IL, USA). Normally distributed continuous variables were expressed as means ± standard deviation (SD), non-normal distributed variables were presented as median with interquartile (IQR), and categorical variables were presented as number and percentage. Differences between groups were assessed by Mann-Whitney U test, Kruskal Wallis test or Chi-square test as appropriate. Logistic regression analysis was used to explore the risk of pregnancy complications or outcomes in underweight women adjusting the effects of gestational age. As our previous study suggested that TG/HDL-c ratio could be a good marker to predict the risks of GDM and LGA11, we also analyzed the association between TG/HDL-c ratio and perinatal outcomes. A ROC analysis was conducted to evaluate the ability of the special factor in the model from logistic analysis for correctly discriminating the adverse perinatal outcomes. Plots of the sensitivity (true positive) versus 1-specificity (false positive) for the two models were made and the overall diagnostic accuracy between the two models was quantified using AUC. P <0.05 was considered statistically significant.