Data sources

We used population-level health administrative data housed at the Institute for Clinical and Evaluative Sciences (ICES) in the Province of Ontario, Canada. ICES is an independent nonprofit research organization that has the capacity to link multiple provincial administrative health databases using unique identifiers for every Ontario resident with a provincial health card number. At ICES, patient-level records are linked anonymously via the Registered Persons Database (RPDB) that contains the sex, age, postal code, and date of birth and death for all Ontario residents. Obstetric deliveries were identified using the MOMBABY dataset, which is derived from the Canadian Institute of Health Information Discharge Abstract Database (CIHI-DAD). MOMBABY captures deliveries and provides linked maternal and newborn health records for all births occurring in an Ontario hospital (>99% of deliveries). It also provides information on gestational age at birth, allowing for accurate estimation of conception date. Psychiatric admissions were captured either in the CIHI-DAD or through Ontario Mental Health Reporting System Database where admissions to designated psychiatric beds have been recorded since 2005. Diagnoses of bipolar disorder and major depressive disorder were captured within 5 years preceding the index pregnancy using the International Classification of Diseases, Ninth Revision (ICD-9) and ICD-10CA diagnoses (CIHI-DAD) and the text revision of the Diagnostic and Statistical Manual of Mental Disorders-IV diagnostic system (Ontario Mental Health Reporting System Database). Perinatal outcomes were captured during the index delivery hospitalization or on subsequent hospitalizations within 42 days after delivery, based on ICD-10CA diagnoses or Canadian Classification of Health Interventions procedure codes recorded in the CIHI-DAD. Mortality outcomes were identified through the RPDB, which identifies death arising in or outside of a hospital setting. Additional datasets used to measure study covariates included the National Ambulatory Care Reporting System Database for emergency department data, and the Ontario Health Insurance Plan Database for outpatient physician service use data. The databases used in this study have been found to be complete, reliable and accurate with respect to demographic information and primary diagnoses for inpatient services.

Participants

We identified 1,032,831 obstetric deliveries among Ontario women ages 15 to 50 years whose estimated date of conception was between April 1, 2002, and March 31, 2010. Deliveries to nonOntario residents, those with an invalid health card number and those where mother-infant matching was unsuccessful were excluded (n = 2216), as were those with a diagnosis of schizophrenia, schizoaffective disorder, or another psychotic disorder in the 5 years preceding conception (n = 1628), the focus of a previous study. We identified the 3 groups within the remaining cohort of 1,028,987 deliveries. The bipolar disorder group comprised 1859 unique women with 2124 deliveries, and who had been hospitalized for bipolar disorder within 5 years before conception in the index pregnancy (kappa = .93; 95% confidence interval [CI], 0.88–0.97). From the remaining women, we identified the major depressive disorder group which comprised 3724 unique women with 4487 deliveries, who had been hospitalized for major depressive disorder within 5 years before conception (kappa = .80; 95% CI, 0.74–0.87). Hospital-based diagnoses were required for this classification because the ability to discriminate between these 2 disorders using only outpatient service claims is likely to be poor. The reference group comprised 432,358 unique women with 553,840 deliveries, and who had no documented mental illness, that is, the absence of a hospitalization or physician service claim from a psychiatrist or family physician associated with any mental health or addictions diagnosis within the 5-year period before the index pregnancy.

Within each of the 3 exposure groups, we randomly selected 1 delivery per woman using a computer-generated randomization list.

Outcomes

The main study outcomes were: (1) PTB <37 weeks’ gestation; (2) severe small for gestational age (SGA), defined as a same-sex, same-gestational age birthweight <3rd percentile; and (3) severe large for gestational age (LGA), defined as a same-sex, same-gestational age birthweight >97th percentile. These outcomes were chosen because they are prevalent, reliably measured, and are major determinants of perinatal morbidity and mortality. As secondary outcomes, we evaluated other key perinatal health indicators: PTB <32 weeks and PTB <28 weeks’ gestation; SGA <10th percentile; LGA >90th percentile; congenital malformations; serious neonatal morbidity, comprising respiratory distress syndrome (RDS), seizure, sepsis, intraventricular hemorrhage, persistent fetal circulation, neonatal abstinence syndrome, admission to a neonatal intensive care unit, neonatal readmission to hospital <28 days, stillbirth after 20 weeks’ gestation and infant mortality <28 days, and <1 year after birth.

Covariates

Study covariates included maternal age and socioeconomic status by neighborhood income quintile and location of residence, according to each woman’s residential postal code at the time of delivery as captured by the RPDB. Prepregnancy hypertension and diabetes mellitus were captured using existing ICES disease registries. We used hospitalization data and physician service claims to capture parity and infant sex as well as problems including venous thromboembolism, and substance and/or alcohol use disorders diagnosed within the year before conception. We also measured maternal gestational diabetes, gestational hypertension, preeclampsia and other obstetric problems during the index hospitalization or within 42 days after delivery as well as mental health and reproductive health service utilization.

Statistical analyses

The distribution of covariates and outcomes for each group was described using means, medians or proportions. For all outcomes except stillbirth, only pregnancies that resulted in live births comprised the denominator of the at-risk population. Logistic regression was used to generate unadjusted and adjusted odds ratios (AORs) and 95% CIs using women with no documented mental illness as the referent. The main outcome ORs were adjusted for maternal age, income quintile, parity, infant sex, obesity, diabetes, hypertension, venous thromboembolic disease, and obstetric complications. These covariates were selected a priori as key prognostic factors that could influence outcome independently of mental health diagnosis (eg, maternal age), or could lie along the explanatory pathway between mental health diagnosis and adverse outcomes (eg, obstetric complications). Secondary outcome ORs were adjusted for maternal age and parity only to maintain appropriate covariate to event ratios, as the event rates for many of these outcomes was projected to be low. SAS version 9.2 (SAS Institute, Cary, NC) was used for all statistical analyses.

Ethics approval

Study approval was obtained from the Research Ethics Board at Women’s College Hospital (Study ID 2011-0055-E) and at Sunnybrook Health Sciences Center (ICES logged study: 2011-0584-419-000). ICES is named as a prescribed entity under Section 45(1) of Ontario’s Personal Health Information Protection Act, 2004 (PHIPA). As a requirement of having this status in PHIPA, ICES policies, practices and procedures are reviewed and approved by the Ontario Information and Privacy Commissioner. Access to raw data is governed by confidentiality agreements between ICES and independent investigators as per PHIPA guidelines. Cell sizes ≤5 are suppressed in publication according to this privacy legislation.