Statistical analysis
Variables were checked for normal distribution by Kolmogorov-Smirnov test. Comparisons between cases with and without adverse neonatal outcomes were performed by Student-T (assuming unequal variances), Mann-Whitney U, Pearson Chi-squared and Fisher-F, as appropriate.
The likelihood of neonatal complications was modeled by logistic regression (with robust estimation of the standard errors). The explained uncertainty for the occurrence of adverse neonatal outcomes was calculated as the R2-Naegelkerke.
The predictive performance was determined by receiver–operating characteristic (ROC) curve analysis. Paired ROC curves were compared by the DeLong method (33).
Statistical analyses and graph constructions were performed using STATA 13.0 (StataCorp LT, Texas, USA) and R 3.1.2 (The R Foundation for Statistical Computing) [package “pROC”].