2.5 Statistical analyses
Categorical variables were compared by using the Chi-square test. For continuous variables, normality was tested by the Kolmogorov-Smirnov test. The independent sample t -test was used for normal data, the Wilcoxon-Mann-Whitney test was used for nonnormal data. To assess the relationship between exercise during pregnancy and PTB risk, Multiple logistic regression was used to calculate odds ratios (ORs ) and 95% of the corresponding confidence intervals (95%CIs ). Variable selection for the multiple models was guided by the directed acyclic graph (DAG). The DAG was plotted to recognize potential causality, mediation, or confounding factors[17]. Furthermore, women’s exercise activity during pregnancy was grouped by time spent and energy expenditure, and the none and low group was used as a reference. The effect of each trimester was observed separately. For subgroup analysis, the subjects were stratified by whether they had pregnancy complications or not. The median effect was analyzed by sequential test.
Statistical analyses were performed using the statistical computing environment R 4.0.5. P-value of <0.05 was considered statistically significant.