Data sources
The Manitoba Population Research Data Repository housed at the Manitoba Centre for Health Policy (MCHP; University of Manitoba, Canada) contains de-identified personal information about Manitoba residents enrolled in the Provincial health plan which provides coverage of insured health services for virtually the entire population of Manitoba [28]. Each individual is assigned a unique, nine-digit Personal Health Identification Number (PHIN) and an encrypted version of this PHIN is attached to person-level records in the Repository data files. This permits linkage of an individual across several data sources and years while still maintaining complete confidentiality. The Repository itself contains data collected by a variety of local agencies in Manitoba which have been previously validated for population-based research [29]. Nine data sources housed in the Repository were accessed for this study: Manitoba Health Insurance Registry (maternal age and insurance coverage), Canada Census File (area-level income quintiles), Drug Program Information Network (DPIN; prescriptions filled), Hospital Abstracts (delivery and birth records, mood/anxiety disorders, diabetes, postpartum psychological distress), Medical Services (mood/anxiety disorders, diabetes, postpartum psychological distress), Diagnostic Services of Manitoba (DSM; diabetes), Manitoba Diabetes Education Resource for Children and Adolescents (DERCA; diabetes), and the Manitoba Maternal Serum Screening Program (MMSSP; maternal pregnancy weights).
Cohort Formation
Hospital Abstracts were used to identify all records where a woman (N=139,362) had an in-hospital single live birth between April 1, 2002 and March 31, 2019 (N = 250,746) (Figure 1). The Hospital Newborn to Mother Link Registry (nblink) was then used to link the mother’s delivery record and the child’s birth record. This allows for pregnancy start date calculation and gestational period identification, based on the difference between the gestational age in weeks and the child’s birth date or, if missing, the mother’s hospital separation (discharge) date. Women with a missing pregnancy start date were excluded (n=5,532 excluded) (Figure 1). In order to assure proper data capture, births were then limited to those where the mother had continuous health coverage under Manitoba Health from five years before their pregnancy start date up to one year following their delivery date (up to March 31, 2019; n=51,930 excluded) (Figure 1). Births were also limited to those where the mother’s age was 15 years or greater at the pregnancy start date (n=779 excluded) (Figure 1); an age of 15 years or greater was determined in order to apply a previously validated mood/anxiety disorder definition (see Primary outcome below). This gave a total of N=192,505 births to N=105,898 mothers (Figure 1).
Manitoba Maternal Serum Screening Program Data were utilized to determine the mother’s weight during the gestational period where a first trimester and/or a second trimester weight is recorded. First trimester weight was selected however if missing, the second trimester weight was used. Weight ranges to classify women with and without obesity were chosen initially by using average female heights and weights reported by the Centers for Disease Control and Prevention (CDC) [30], and choosing weights that would reflect traditional BMI values used in studying weight classes in the “average population”. In 2013-2014, American women aged 20 years or greater weighed on average 169.8 pounds (77.0 kg) and were 1.616 meters tall (63.6 inches; 5 feet 3 inches), which translates into a BMI of 29.5 kg/m2 (overweight). Based on that value, weight cut-offs were determined based on a height of 64 inches (±4 inches). For a BMI of 22 (lean/no obesity), the estimated weight range was 51.0 – 65.6 kg. For a BMI of 35 (with obesity; class II), the estimated weight range was 81.2 – 104.3 kg (see Additional File 1 for details). Therefore, in an effort to appropriately capture all women and avoid misclassification, we chose a recorded maternal weight range of >38 – 65.6 kg to represent women in the lean group (without obesity), and a weight range of ≥85.0 – 186.0 kg for the obesity group (with obesity). These weight cut-offs were initially validated with pre-pregnancy height, weight and BMI information taken from a cohort of women (n=62) classified as lean (n=23) and obese (n=39) in two previous studies of ours that gave us a capture rate of 90% for lean and 100% for obese groups (respectively; see Additional File 1 for details) [20, 21]. These weight cut-offs were then validated against a large 2017 publicly available dataset from the United States of female weights which also contained height, thus permitting BMI’s to be calculated [31]. Use of the same weight ranges in this dataset resulted in 9% of women misclassified as having obesity (who were not), and 0% of women misclassified as lean (who were not) according to BMI. Women with unlikely weights (≤37 kg or ≥187 kg), or who do not fall into one of the identified weight groups were excluded (n=96,411 excluded). Of the remaining births, one birth per mother was randomly selected (n=53,865 excluded). Births to mothers with pre-existing type I or type II diabetes (n=1,218) as well as a pre-existing mood/anxiety disorder (n=13,128) were excluded, since both pre-existing conditions are predictors for postpartum depression [32–34]. Therefore, the final weight group cohorts are N=17,975 for the lean group, and N=9,908 for the with obesity group (Figure 1).
Primary outcome
A woman was considered to have PPD if in the 12 months after giving birth she had (i) one or more hospitalizations with a diagnosis of depressive disorder, affective psychoses, neurotic depression, adjustment reaction, anxiety disorder, anxiety states, phobic disorders, or obsessive-compulsive disorders according to International Classification of Disease, 9th revision, Clinical Modification (ICD-9-CM) codes: 296.2-296.8, 300, 300.0, 300.2, 300.3, 300.4, 309, 311, or ICD 10th revision Canada (ICD-10-CA) codes: F31, F32, F32.0, F33, F34.1, F38.0, F38.1, F40, F41, F41.0, F41.1, F41.2, F41.3, F41.8, F41.9, F42, F43.1, F43.2, F43.8, F44, F45.0, F451, F452, F48, F53.0, F68.0, F93.0, F99; (ii) one or more physician visits with a diagnosis of depressive disorder, affective psychoses, or adjustment reaction according to ICD-9-CM codes 296, 309 or 311; (iii) a prescription for an antidepressant or mood stabilizer based on Anatomical Therapeutic Chemical (ATC) codes N03AB02, N03AB52, N03AF01, N05AN01, N06A; (iv) one or more physician visits with a diagnosis for anxiety disorders (ICD-9-CM code 300) as well as one or more prescriptions for an antidepressant or mood stabilizer (ATC codes N03AB02, N03AB52, N03AF01, N05AN01, N06A); and (v) two or more physician visits with a diagnosis for anxiety disorders (ICD-9-CM code 300). This definition has been used in previous studies at MCHP (http://mchp-appserv.cpe.umanitoba.ca/viewDefinition.php?definitionID=104493).
Covariates
The following information was extracted for all women: maternal age at pregnancy start date (continuous and 5-year age groups), area-level income (urban and rural quintiles), insulin treatment in pregnancy (yes/no; based on a dispensed prescription with ATC code A10A filled in the 168 days following pregnancy start date and before the child’s birth date), pre-existing diabetes (yes/no), and pre-existing mood/anxiety disorders (yes/no).
Pre-existing diabetes and pre-existing mood/anxiety disorders were determined before limiting data to one random birth per mother so that if either of these conditions were present on any pregnancy they could be “carried forward” to all subsequent pregnancies for that mother. Thus, pre-existing diabetes was defined as (i) having any type of diabetes in the two years prior to the pregnancy start date, or in the 168 days (24-weeks) following the pregnancy start date or (ii) having pre-existing diabetes diagnosed in a previous pregnancy. Pre-existing diabetes was identified using an existing algorithm at MCHP [35] (see Additional File 2 for full definition). Pre-existing mood/anxiety disorder was defined (i) as having a mood and/or anxiety disorder diagnosed in the 5-years prior to pregnancy start or (ii) having a pre-existing mood and/or anxiety disorder diagnosed in a previous pregnancy. A mood/anxiety disorder was defined using an existing algorithm at MCHP [35] (see Additional File 2 for full definition).
Statistical Analysis
All data management, programming and analyses were performed on MCHP’s secure servers, using SAS® version 9.4 software. The percentage (%) and adjusted rate of PPD, as well as 95% confidence intervals (CIs), were estimated using Poisson distributions containing a log link function and the logarithm of the number of mothers as the offset in the models. Models were adjusted for maternal age group (at pregnancy start date) and area-level income (at delivery). While it would have been ideal to adjust for other factors associated with both insulin treatment and obesity, the sample size did not allow for additional adjustment. The analyses were stratified by insulin treatment (yes/no) in pregnancy. Significance is represented by a p-value being <0.05 (two-tailed).