Statistical analysis
A Chi-square test was used to assess the differences in socio-demographic and pregnancy characteristics between the exposed and unexposed groups. A generalized linear model (GLM) was applied to estimate the associations of Level I lockdown exposure with gestational length (linear regression) and PTB risk (binary logistic regression), after adjusting for potential confounders. A multinomial logistic regression model was used when PTB was further divided into MPTB and VPTB, with term birth as the reference. An interaction test was conducted to examine the potential modification effects of infant sex by comparing the association coefficients between male and female infants (27).
Similarly, GLM and multinomial logistic regression models were employed to examine the association of cumulative exposure dose with gestational length or PTB. The cumulative exposure dose in the exposed group was divided into four groups by quartiles (Q1, Q2, Q3, and Q4). The association of each quartile of cumulative exposure (vs unexposed) with gestational length or PTB were estimated. A trend test was conducted by assuming the values of quartiles as a continuous variable.
All analyses were performed using R3.6.1 (R Development Core Team 2019, https://www.r-project.org). All the tests were two-sided and a P<0.05 was considered to be statistically significant.