Statistical analysis
All statistical analyses were performed with R software, version
4.0.3.22 Inescapably, if only for their age,
non-pregnant women who responded to our survey did not have the same
characteristics as the pregnant women who responded to it. Therefore, we
used a propensity score approach to control for confounding factors that
might influence our result on their mental well-being levels. We
included all pregnant women but selected non-pregnant women by
stratification by a propensity score.23 We analysed
only questionnaires with sufficient information to calculate this score.
A woman’s propensity score was defined as her probability of being
pregnant based on the individual covariates we measured. This score was
calculated by applying a generalised linear model with current pregnancy
as the dependent variable and considering the following characteristics:
age range, marital status, living alone or with someone else,
psychiatric (including addictions) history, parity, local extent of
pandemic area during weeks 2-3 of lockdown, educational level, and
occupation. We distributed the propensity scores obtained for each woman
into five classes. Finally, we matched non-pregnant women (controls) on
a three-for-one basis, class by class. The early/late pandemic area was
determined retrospectively as early or late by the respondents’
postcodes. Districts with a ratio of more than 2 deaths per 100 000
residents on March 23, 2020, were classified as early pandemic areas by
the French national public health agency, Santé Publique France(https://www.data.gouv.fr/).
Quantitative variables with normal distributions according to the
Shapiro-Wilk test were described by their means and standard deviations
(SD) and then compared with a Welch two-sample t-test. When
distributions were not normal, variables were described according to
their medians with their 25th and
75th percentiles and then compared by a Wilcoxon rank
sum test. Qualitative variables were described as the number of
individuals and percentages and then compared with Fisher’s exact test.
The denominator is reported when it comprises less than 95% of the
total sample size. A multiple regression analysis then assessed the
association of pregnant women’s characteristics with their WEMWBS score.
This score was entered as a dependent variable in the model and all
their other characteristics as independent variables.