Statistical analysis
Where data permitted, Stata 14 was used to conduct meta-analyses of
proportions using a random-effects model. We analysed the available data
from three main population groups. The ‘treated’ group were women with
CID who required treatment with biologic medications throughout their
pregnancy. The ‘disease matched’ group were the group of women with CID
not prescribed biologics in pregnancy. The ‘disease free’ group was a
group representative of the general population, (i.e. women who were
pregnant and did not have a diagnosis of CID). We modelled data using
the program metaprop which augments the metan program. This
allowed for computation of 95% confidence intervals16and pooling of proportions, presenting a weighted sub-group and overall
pooled estimates with inverse-variance weights from a random effects
model. The primary analysis was performed on all eligible studies, with
subgroup analyses by biologic type and by chronic inflammatory disease
subtype where appropriate. We used linear regression analysis to compare
the mean of proportions of outcomes across exposure groups using the
‘treated’ group as the reference category.