The CCHS is a nationally representative, cross-sectional survey conducted by Statistics Canada to collect comprehensive data on health status, healthcare utilisation, and the social determinants of health among Canadians [21]. It is the most widely used population health survey in the country and provides essential information to monitor trends in population health, health behaviours, and healthcare access and utilisation, as well as to support policy development and guide health research at the national, provincial, and sub-provincial levels.

Originally administered biennially, the CCHS adopted an annual format in 2007, sampling approximately 65,000 individuals each year across all ten provinces and three territories. The survey uses a multistage, stratified cluster sampling design based on the Labour Force Survey frame, with recent cycles incorporating targeted sampling from the census to enhance representation of equity-deserving populations. The CCHS excludes individuals living on reserves, full-time members of the Canadian Forces, institutionalised populations, and residents of certain remote regions. Although these groups represent less than 3% of the Canadian population, they are important populations whose health experiences are not captured in the survey.

Data collection is conducted in both official languages (i.e., English and French) via telephone (CATI), in-person (CAPI), or electronic questionnaire (EQ), depending on cycle and region [21]. Typical overall response rates range from approximately 60-75% across cycles, comparable to other large national health surveys. Statistics Canada applies extensive quality control and validation procedures, including interviewer training, real-time logic checks, edit and imputation procedures, and post-collection weighting adjustments to mitigate the impact of nonresponse and coverage error.

Survey content includes annually repeated core modules (e.g., general health, chronic conditions, healthcare access) and rotating thematic modules (e.g., mental health, reproductive health, food insecurity), enabling both comparability over time and responsiveness to emerging public health priorities. It captures a wide range of individual-level variables, including age, sex, race, immigration status, education, income, and health behaviours (e.g., smoking, alcohol use, physical activity, diet).

For the creation of this cohort, the CCHS offers several advantages: with linkage, it captures pre-pregnancy exposures not available in clinical data, enables stratified analyses across sociodemographic groups, and provides rich behavioural and contextual variables within a nationally representative sampling framework.

The Discharge Abstract Database (DAD), maintained by the Canadian Institute for Health Information (CIHI), is a national administrative dataset that records all hospital separations (discharges, transfers, and in hospital deaths) from acute care facilities across Canada, with the exception of Quebec [22]. Quebec accounts for approximately 22% of the Canadian population [30], and hospitalisation data for Quebec residents are collected separately by the ministère de la Santé et des Services sociaux and submitted to CIHI for inclusion in the Hospital Morbidity Database (HMDB) [22, 31]; consequently, Quebec births are not captured in this cohort. The DAD contains detailed information on obstetrical deliveries, newborns, and stillbirths from acute inpatient hospitals, and in this cohort was used to identify delivery hospitalisations using validated ICD codes. Given that approximately 98% of births in Canada occur in hospitals, the vast majority of CCHS participants who are pregnant—or go on to become pregnant—can be accurately linked to delivery records in the DAD [32]. The small proportion of births not captured by the DAD primarily includes planned home births or deliveries in birthing centres, which are more common among individuals with low-risk pregnancies and those receiving midwifery-led care [33].

Each DAD record contains demographic, administrative, and clinical information abstracted from hospital charts, including patient age and sex, admission and discharge dates, discharge disposition, length of stay, most responsible diagnosis, up to 25 additional diagnoses, and procedures performed during hospitalisation. Clinical data are coded using the International Statistical Classification of Diseases and Related Health Problems and the Canadian Classification of Health Interventions (CCI). Since 2004–2005, all jurisdictions reporting to the DAD have fully adopted ICD-10-CA and CCI coding standards [31]. Prior to this, ICD-9-CA was used. DAD records are subject to validation rules at both the hospital level and at CIHI, including checks for logical consistency, valid code use, and adherence to abstracting standards.

CAN-Marg is a census-based measure developed by the Centre for Urban Health Solutions at St. Michael’s Hospital and now maintained by Public Health Ontario [23, 34]. It was created to capture neighbourhood-level marginalization and facilitate research on health inequities across Canada. Using principal component analysis of census indicators, CAN-Marg generates standardised scores across four key domains—material deprivation, residential instability, dependency, and ethnic concentration. These scores are available at the dissemination area (DA) level, which represents the smallest standard geographic unit used by Statistics Canada for census data (typically comprising 400–700 residents) [35], and have been widely applied in population health studies to investigate the role of social context in health outcomes.

CANUE produces standardised, pan-Canadian datasets on environmental exposures that can be linked to health and survey data using residential postal codes [24]. These resources cover multiple environmental domains such as ambient air pollution, access to green space, and climate-related conditions. Depending on the measure, exposures are provided either at the postal code level or at a spatial resolution of 1 km². For this study, exposure assignments were aligned with the timing of each respondent’s CCHS interview, using the interview year or the nearest available dataset.

Can-ALE is an area-based measure that evaluates how supportive neighbourhoods are for physical activity [25]. Developed by the Geo-Social Determinants of Health Research Group at McGill University in collaboration with the Public Health Agency of Canada, Can-ALE incorporates open-source geographic data to generate standardised indicators of walkability and access to local amenities. Data are available for all census dissemination areas (DAs) in Canada for the 2006 and 2016 census years. In this study, Can-ALE values were linked to participants’ residential postal codes and matched to the census year closest to the date of their CCHS interview (e.g., a respondent interviewed in 2011 was assigned Can-ALE values from 2006, while a respondent interviewed in 2017 was linked to the 2016 index).

We began by restricting the CCHS to individuals who identified as female at birth, aged 15 to 49 years, who completed a survey cycle between 2000 and 2017 and consented to data linkage. This age range aligns with the reproductive lifespan. Given the evolving structure of the CCHS over time (e.g., changes in sample design, survey content, and variable formats) each cycle was cleaned separately before harmonisation.

The cleaning process involved recoding and standardising key variables to ensure consistency across cycles. In particular, the 2015 CCHS redesign introduced changes in variable naming conventions, question phrasing, and data collection methodology (e.g., expanded use of electronic questionnaires), necessitating careful cross-cycle harmonisation. Variables were organised into the following conceptual domains:

A complete list of variables included in the cohort, along with their definitions, is provided in Table 1. Because the cohort spans a 17-year period (2000–2017), we undertook systematic cross-cycle harmonisation to ensure temporal consistency of key variables. Where questions differed slightly across cycles, we created harmonised variables based on conceptually equivalent items. Major structural changes to the CCHS occurred in 2007 (transition from biennial to annual cycles) and in 2015 (survey redesign, updated questionnaire structure, and expanded electronic data collection) [21]. All variables were harmonised using standardised coding schemes and consistent derived thresholds to preserve comparability over time. Where direct harmonisation was not possible, variables were collapsed into broader categories to ensure stability across cycles. For example, employment status—measured using different question formats in early cycles—was recoded into consistent categories, and categorical variables such as BMI classification and income were recoded using standardised thresholds.

Although only a small proportion of CCHS respondents reported being pregnant at the time of the interview (approximately 400 per year), we flagged these individuals for internal validation (see Results) and descriptive purposes using cycle-specific pregnancy indicators (e.g., “Are you currently pregnant?”). However, current pregnancy was not a requirement for cohort inclusion, which was instead based on 2-year subsequent linkage to delivery hospitalisations.

After cleaning and harmonisation, all eligible cycles were merged into a single dataset.

To identify relevant delivery hospitalisations, we extracted and cleaned records from the DAD between 2000 and 2017. The DAD includes all acute care hospital separations in Canada (excluding Quebec), and delivery-related records were identified using validated diagnostic and procedural codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CA) and the Tenth Revision, Canadian Modification (ICD-10-CA). Clinical coding in the DAD transitioned from ICD-9-CA to ICD-10-CA between 2001 and 2004, with full national adoption by 2004–2005 [22]. All delivery and obstetric outcome definitions used in this cohort were developed using validated crosswalks and harmonised code lists spanning both coding systems to ensure consistency of outcome ascertainment over the full study period.

The full list of codes used to identify delivery hospitalisations is provided in Table 2.

We restricted the dataset to delivery records involving either a live birth or stillbirth. To maintain cohort consistency and reduce clinical heterogeneity, we excluded multifetal gestations (e.g., twins or higher-order multiples), identified through ICD-coded plurality indicators. These pregnancies are associated with elevated risk for complications and adverse outcomes. We also derived supplementary variables where available, such as gestational age at delivery, history of spontaneous abortion, and prior term deliveries.

For the purposes of this manuscript, we have not included any specific pregnancy outcomes, as users of the cohort may define outcomes according to their research objectives. For example, current manuscripts in preparation using this cohort focus on adverse pregnancy outcomes (e.g., gestational diabetes, placental abruption, and preeclampsia) which have been defined using ICD codes and incorporated into the analytic dataset.

Using Statistics Canada’s SDLE, CCHS respondents were linked to DAD records through a probabilistic record linkage process conducted by Statistics Canada. This method uses multiple identifiers (e.g., name, date of birth, sex, postal code) to assign an encrypted unique linkage key for each individual and applies standardised quality assurance procedures, including clerical review and validation thresholds, to minimise false matches and missed links [29].

Linkage was restricted to CCHS respondents who provided informed consent for record linkage at the time of survey participation, consistent with Statistics Canada data governance policies. Across survey cycles, over 80% of CCHS respondents consent to linkage. Among consenting respondents, the linkage success rate between the CCHS and the DAD is approximately 85% (excluding Quebec, which does not participate in the DAD) [36, 37].

While individual-level linkage error rates are not released due to confidentiality restrictions, prior methodological evaluations of CCHS–administrative data linkages have demonstrated high linkage quality and minimal bias at the population-level. For example, a coverage evaluation comparing linked CCHS–hospital records with full hospital discharge data found only a modest under-coverage (~7.7%) of hospitalisations in the linked file, indicating high overall linkage quality at the population-level [38]. Nevertheless, potential sources of linkage bias remain, including differential consent to linkage (not all survey respondents agree to link), and failure to link when personal identifiers are missing or inconsistent. These limitations should be considered when interpreting analyses based on the linked cohort.

For this study, only respondents with a hospital delivery within two years following their CCHS interview date were included in the analytic cohort. This two-year window was chosen to balance temporal alignment between pre-pregnancy exposures and delivery outcomes, while minimising misclassification. In cases where multiple deliveries occurred within the window, only the first was retained to reflect the pregnancy most proximate to the CCHS interview.

Figure 2 illustrates the inclusion and exclusion logic using hypothetical participant timelines. Participant A is included in the cohort, as they delivered within 9 months of their CCHS interview, suggesting they would likely have reported being pregnant at the time of the survey. Participant B is also included, with a delivery that occurred within the two-year window. Participant C is not included in the cohort as they delivered prior to their CCHS interview. Participant D had two deliveries within two years of their CCHS interview; however, only the first delivery is retained, since survey responses—particularly those related to pregnancy—may not correspond to subsequent pregnancies within the two-year follow-up period. Participant E is excluded because their delivery occurred more than two years after the CCHS interview date, falling outside the inclusion window. Finally, Participant F is excluded from the cohort due to the absence of a linked delivery record.

We linked area-level exposures to CCHS participants using their six-digit residential postal code at the time of survey. Postal codes were mapped to standardised geographic units using the Postal Code Conversion File Plus (PCCF+), a probabilistic linkage tool developed by Statistics Canada.

The PCCF+ is an enhanced version of the Postal Code Conversion File (PCCF) [27] and includes a SAS-based program that assigns postal codes to a range of standard geographic identifiers (e.g., dissemination areas, census subdivisions, census metropolitan areas, and health regions). Unlike the PCCF, PCCF+ employs population-weighted random allocation for postal codes that map to more than one geographic area, improving accuracy in urban and multi-jurisdictional settings. It also draws on additional sources, including the Postal Code Population Weight File, Geographic Attribute File, and Health Region boundary files.

Using the CCHS postal code and DA-level generated by the PCCF+, we were able to link CCHS respondents to multiple contextual datasets: CAN-Marg, CANUE, Can-ALE. The PCCF+ itself also generates area-level income variables, which provide geographically standardised proxies for socioeconomic position.

Variables were organised into the following conceptual domains: