Study population and sampling frame

Our study utilized data from the Maternity Experiences Survey (MES), the first Canadian pan-maternity study. The MES was an initiative of the Canadian Perinatal Surveillance System, which is managed by the Public Health Agency of Canada. The sampling frame was developed from the 2006 Census file. Details of the survey methodology are available elsewhere. In brief, births occurring between Feb. 15, 2006, and May 15, 2006, were chosen for inclusion in the study, after excluding multiple births and stillbirths, mothers younger than 15 years of age, and mothers living on First Nations reserves or in institutions or no longer living with their baby.

A total of 58,972 mother-baby pairs were identified. The sample was stratified on province or territory in which the mother resided at the time of the census and on maternal age (<20 years, ≥20 years old). The target population was women who had given birth approximately 6 months before the data were collected. This time frame was chosen to enable exploration of postpartum depression and minimize recall error.

Proportional sampling was undertaken with a parameter of 0.7, using the Kish allocation method. The sample was first allocated to the primary strata and then allocated to the substrata proportionally to the population in the substrata. Young mothers were oversampled. Samples sizes for each province and territory were calculated separately based on an expected response rate of 75% and a minimum reportable proportion of 10% in the provinces and 15% in the territories. The required level of precision of all minimum proportions was set at a coefficient of variation of 16.5% or lower. This meant that characteristics or events that were reported by 10% or more of respondents in a province, or by 15% or more or respondents in a territory, could be reliably estimated. Higher minimum reportable percentages were allowed for First Nations off-reserve, Inuit, and Métis mothers. Based on these requirements, an initial sample of 8542 potential respondents was randomly selected from the sampling frame.

Prior to data collection, potential respondents were mailed an introductory letter and a survey pamphlet requesting their participation in English, French, or Inuktitut. Interviews were conducted by female Statistics Canada employees by telephone in the provinces and either by telephone or in person in the territories. Interviews were conducted by telephone in the provinces and either by telephone or in person in the territories. Interviews were approximately 45 minutes in length. A maximum of 25 attempts was made to reach each potential respondent, and attempts to trace respondents were made if they were not reached at the phone number listed on the sample frame. In addition to English and French, multilingual interviewers used translated glossaries to conduct 150 interviews in 13 nonofficial languages using computer assisted telephone interview application software.

Overall, complete responses were obtained from 6421 women, 77.9% of those identified for participation. Only 1% of women with whom contact was made refused participation. Babies of participant mothers were on average 7.3 months of age, with the majority between 6 and 9 months (84%) and a full range between 5 and 14 months.

Violence exposure

Recent violence exposure was evaluated in the MES using a series of questions that addressed both threatened and actual violence. Women were asked to consider the previous 2 years. In the current study, we defined exposure as actual violence, including having been slapped, bitten, kicked, hit, beaten, choked, threatened with a gun or knife, or had one used against them or forced into sexual activity.

These questions were adapted from the Violence Against Women Survey, conducted in Canada in 1993 and are similar in content to the Abuse Assessment Screen. Women were also asked about their relationship with the perpetrator, and we restricted our analysis to responses that identified a husband, partner, boyfriend, or family member. We included family members because in some cultures family members may engage in abuse of women on behalf of women’s partners. Women were asked about their experience of violence during the 2 years preceding the survey and about their exposure specifically before or during pregnancy and since the birth.

Postpartum depression

Depression was assessed using the Edinburgh Post-Natal Depression Scale (EPDS), a 10 item screening tool designed to identify postpartum depression at the time of its administration. A score of 13 or more of a maximum possible of 30 was used to indicate the presence of postpartum depression. The EPDS has been validated in numerous studies of postpartum women and has been shown to have a sensitivity of 86% and specificity of 78%.

Approval for the study was granted by Health Canada’s Science Advisory Board and Research Ethics Board and by the Federal Privacy Commissioner as well as from Statistics Canada’s Policy Committee. The current secondary analysis was approved by the Research Data Centre Access Granting Committee of Statistics Canada, St Michael’s Hospital Research Ethics Board, and the University of British Columbia Behavioural Research Ethics Board.

Statistical analysis

Data were analyzed at the Toronto Regional Statistics Canada Research Data Centre using SAS 9.2. Records in the sample were assigned a design weight based on their Census weight divided by its probability of selection from the stratified sampling frame. Weighting adjustments were performed to adjust for rates of nonresponse. After calibration of sample weights, the estimated target population was 76,508 in-scope mother-baby pairs.

We compared risk factors for postpartum depression among women exposed to vs not exposed to family or intimate partner or family perpetrated violence. Within the cohort of women exposed to intimate partner violence, we also assessed timing of abuse in relation to postpartum depression. We examined sociodemographic characteristics of survey participants including age, marital status, education, employment, income, wantedness of pregnancy, and depression prior to pregnancy within study cohorts.

Point estimates of prevalence were calculated using final sampling weights. Proportions and odds ratios and their 95% confidence intervals were weighted and calculated using the jackknife method of variance estimation. Missing values were very rare and were therefore not considered in the analysis with the exception of low income for which we created a “missing” category. Variables with a statistically significant association with postpartum depression in a bivariate association were retained for testing in a multivariable model among both abused and nonabused women. We report adjusted odds ratios for those that remained statistically significant in the multivariable analysis.