To be enrolled in the study, women had to be at least 18 years of age and not yet completed their 17th week of pregnancy. Women were ineligible if they had a known multi-fetal pregnancy, had insulin-dependent diabetes, did not speak English or Spanish, did not have access to a telephone, had plans to relocate, or intended to terminate their pregnancy.
Recruitment and Assessment Procedures
We recruited women from 137 obstetrical practices and hospital-based clinics in Connecticut and western Massachusetts. Obstetrical providers or a project screener gave patients a form on which patients could indicate interest in participation. The form also included the estimated date of delivery. Providers faxed forms to the central data collection site where research staff obtained consent by phone and completed a screening interview. The structured screening questionnaire included questions about pregnancy dates, mood, psychotherapy, antidepressant use, medical conditions, and plans to relocate or terminate the pregnancy. The interviewer confirmed eligibility criteria and administered the gateway questions for a depressive disorder from the World Mental Health Composite International Diagnostic Interview v2.1 (WMH-CIDI)5
): (1) felt sad, empty or depressed; (2) felt discouraged or (3) lost interest in most things they enjoy for most of the day during any two-week interval in the current pregnancy. The screener also asked respondents about past episodes or treatment for a depressive disorder. We used these results to enrich the cohort with women who used antidepressants or were at risk of a major depressive episode. To do this, we offered participation to women who were undergoing antidepressant treatment or had a current or prior history of a depressive disorder. We randomly selected and invited participation from one of every three women who neither provided a positive response to screening questions for a depressive disorder nor reported treatment for a depressive disorder in the last five years.
We obtained written consent during a face-to-face interview before 17 completed weeks gestation. We re-interviewed participants by phone at 28 (± 2) weeks’ gestation (“monitoring” phone interview) and again at 8 (± 4) weeks postpartum (“postpartum” phone interview). Pregnancies in the cohort occurred between March of 2005 and May of 2009; follow-up continued until September 2009. Institutional Review Boards at Yale University School of Medicine and affiliated hospitals provided approval for the study.
At each assessment point we administered the depressive disorder, post-traumatic stress disorder (PTSD), generalized anxiety disorder, and panic disorder modules of the WMH-CIDI, 5
to participants. The Composite International Diagnostic Interview is a valid and reliable, fully structured lay interview instrument,6
and has been administered to over 150,000 persons from 28 countries. Although the interview has not been specifically validated for use in pregnant women, such women have been well represented among those interviewed. In a validity study of a mixture of participants, the interview had high concordance with a semi-structured clinical psychiatric interview for 12-month period prevalence. 7
The area under the receiver operating curve between the semi- structured clinical interview and the Composite International Diagnostic Interview was between 0.8 to 0.9 for any depressive or anxiety disorder. The specificity for any depressive disorder in the prior 12 months was 97% (se=0.9) and the sensitivity was 69% (se=11.8). The Diagnostic Interview is similarly reliable when administered over the telephone.8
At the initial face-to-face interview we obtained a lifetime history of prior depressive disorder and the self-reported number of lifetime episodes by repeating the gateway questions outlined above. We asked greater detail for depressive episodes in the six months before and during pregnancy. The latter information included questions about all criteria for a major depressive episode according to the Diagnostic and Statistical Manual for DSM-IV.9
At this interview, we asked women about symptoms in pregnancy months 1–3; at the monitoring phone interview we asked about symptoms in months 4–7; and at the postpartum phone interview we asked about symptoms during pregnancy months 8–10 and the first two months after delivery. We gave women specific dates to aid recall. We applied the algorithm for a major depressive episode from the Composite International Diagnostic Interview to determine, on a monthly basis, whether a participant was in an episode of illness. A diagram of the interview schedule and time period queried is shown in .
For information on antidepressant use or psychotherapy, we relied primarily on participant interviews. Interviewers collected information about antidepressant use for time periods corresponding to those for the depressive episodes. We asked participants to show us pill bottles, if available, at the home interview. We also showed participants pictures of various antidepressant pills in order to obtain more accurate information. While we attempted to collect records from outpatient behavioral health clinicians, many clinicians would not provide this information. Because of missing data, we deemed data from this source inadequate for analysis. Information about covariates, such as drug and alcohol use, race, ethnicity and socioeconomic status was also collected by interview.
Interviewers and Quality Control
Interviewers received a minimum of four days of didactic training followed by at least six practice interviews and a minimum of four supervised interviews before becoming eligible to conduct independent interviews. We audiotaped interviews with permission of participants. We randomly selected participants for quality-control assessment. For five percent of interviews, supervising staff subsequently called the participant and confirmed demographic and other information. For an additional five percent, the entire interview tape was reviewed for quality of data collection. Finally, all interviews were reviewed by second- and third-level coders. Any inconsistencies, unresolved questions, or missing information triggered review of the audiotape or a call-back to the participant. We used the same reliability procedures for phone interviews.
Study Recruitment and Analytic Group
Study recruitment is illustrated in . We enrolled 9525 women, of whom 3087 (32%) were ineligible, declined to enroll, or were not screened. Another 1905 (20%) screened positive for a depressive episode in the last five years or were currently undergoing antidepressant treatment. The remaining 4533 women (48%) screened negative for a depressive episode or antidepressant treatment. We invited all women who screened positive to participate in the study, and we randomly selected 1612 (36%) of the women who screened negative to participate. Of these 3517 women, we interviewed 2793 (79%). We excluded 1928 of these women because they were not relevant for the analysis. These included participants who were already experiencing a major depressive episode in Month 1, and women with no history of a depressive disorder (who would have less reason to consider antidepressant use in pregnancy). In addition, we excluded participants with one of the following conditions that may appear similar to a major depressive episode and could have led to misclassification errors: (n=87): HIV (n=4), hypothyroidism (n=31), sickle cell anemia (n=4), pancreatitis (n=1), substance abuse (n=12), or alcohol abuse (n=7). Substance abuse was defined as use of an illegal drug 4 or more days per week in month 3 or later; alcohol abuse was use of alcohol 4 days or more per week or 5 or more drinks per occasion. Finally, we excluded participants who used antipsychotic (n=24) or anticonvulsant (n=19) agents that would indicate a history of bipolar or other psychotic disorder. Some participants in our sample met more than one exclusion criteria.
Study Recruitment and Selection of Analytic Sample
The remaining group of 778 women constituted our analytic sample (). Of this group, 21 women (3%) miscarried, 617 (79%) successfully completed both subsequent interviews, and 689 (89%) completed at least one of the two remaining interviews. When a monitoring phone interview was missed, we did not collect data for pregnancy months 4–7 and when a postpartum phone interview was missed we did not collect information for pregnancy months 9 and 10 and postpartum months 1 and 2.
We used Cox regression to model risk for a major depressive episode. An event required full DSM IV criteria for at least two weeks. The risk period began in the second month of pregnancy and ended two months after delivery. Women with missing data for a major depressive episode were treated as censored. In a sensitivity analysis, we considered censorship to be an event. For each model, we used the exact method of computation to account for a large number of ties in times to event.
The primary exposure of interest was antidepressant treatment, a time-dependent variable. We examined models comparing treatment with onset of a major depressive episode during the corresponding months. We also created models for time-lagged antidepressant indicators of 1, 2 and 3 months to resolve issues of causal ordering. In the end, we chose a lag of 2 months, so as to relate risk of episodes with antidepressant use two months earlier. The longer lag was chosen to avoid artificially inflating the hazard for antidepressant users by including antidepressants taken following onset of a major depressive episode. Lags of 2 and 3 months produced similar hazard ratios for antidepressant use but a lag of 2 months was more sensitive to changes in use.
We pre-selected covariates that might confound the relationship between antidepressant use and onset of a major depressive episode. A history of prior depressive episodes (categorized as 1, 2, 3 or 4 or more lifetime episodes), illness onset before the age of 14, and psychiatric hospitalization are correlates of severe depressive illness.10
We included age, socioeconomic status and race/ethnicity in our adjusted models because perinatal depressive episodes may be higher in poor, ethnic-minority women11
and in adolescents.12
We included factors related to physical health: pre-pregnancy body mass index (BMI) and binary indicators for doctor visits related to diabetes and hypertension in the 12 months before pregnancy. Finally, we included an indicator for a major depressive episode in the 6 months before pregnancy because women with more recent illness appear to have greater risk after medication discontinuation. 13
All two-way interactions between antidepressant use and other factors, including the antidepressant-time interaction, were not statistically significant. We asked about psychotherapy during pregnancy but only by trimester, making it difficult to model this factor to account for causal ordering.
For each covariate, we computed an unadjusted hazard ratio (HR) for a major depressive episode with the associated 95% confidence interval (CI). We also computed an adjusted HR that assessed the jointly modeled effects of each factor. Missing data were uncommon due to extensive data quality checks; age of initial depressive illness onset and BMI were the only predictors with missing values, and this was due to participant recall. Missing values for these factors were included as a separate category. Missing values for antidepressant use always corresponded to missed interviews and censored data. Analysis was performed using SAS version 9.2 (SAS Institute Inc., Cary, NC, USA).