All identified citations were downloaded to EndNote© software and titles and abstracts
screened for relevance by one reviewer (EM). The full text versions of potentially relevant
articles were obtained and assessed for eligibility. A second reviewer (CT) independently
assessed the eligibility of 100 randomly selected citations and eligibility decisions were
compared. Authors were contacted to determine whether identified conference papers or
abstracts had been published in peer reviewed journals and raw data was requested if
published studies did not presented information in the required format. Where duplicate
publications were identified, the data based on the largest sample size was included. One
reviewer (EM) extracted data on study characteristics and results using a piloted form. Two
reviewers (EM and SA-W) independently assessed the quality of included studies (low,
moderate or high) using a tool adapted from previous measures (see Appendix 2,available
online at http://links.lww.com/xxx).(10,11) Discrepancies were resolved through discussion
with the senior author (LH). Studies were assessed for risk of selection and measurement
bias and those with high risk of bias in either area were defined as low quality and excluded
from the meta-analysis. This included studies that used non-validated screening measures,
studies which excluded women with a current or previous diagnosis of the mental disorder
in question and case-control studies which selected participants on a characteristic likely to
bias findings (e.g. preeclampsia, fetal mortality). Studies were defined as high quality if they
had population-based samples and used either diagnostic measures of mental disorders or
well-validated and widely used screening measures (such as the Edinburgh Postnatal
Depression Scale).