This is a population-based study using administrative data linkages to relate delivery/birth data with maternal immigration characteristics.
Ontario is Canada’s most populous province (12.2 million in 2006) and receives annually about half of all immigrants to Canada (approximately 120,000 each year), with more than 90% of them concentrated in urban areas. The study population was composed of all 474,614 singleton live births born in Ontario hospitals to mothers living in any of the 11 Ontario Census Metropolitan Areas (Great Sudbury, Hamilton, Kingston, Kitchener, London, Oshawa, St. Catharines–Niagara, Ottawa-Gatineau, Thunder Bay, Toronto, and Windsor)18 at the time of delivery, between April 1, 2002 and March 31, 2007.
Preterm birth and some maternal and obstetric characteristics were obtained from the Discharge Abstract Database of the Canadian Institute of Health Information that captures around 98% of all deliveries that occurred in the province. Birth data were limited to the most recent 5-year period available (April 1, 2002 to March 31, 2007) in order to ensure that the measurement of the outcomes was the same for all the study subjects and not affected by secular trends and changes in coding schemes and reporting practices over time. Data on immigrants’ landings to Ontario were available for the calendar years 1985–2000 through the Landed Immigrant Data System, which is the official immigration registry compiled by Citizenship and Immigration Canada. Immigration data were probabilistically linked with the healthcare databases based on surname, given names, sex, and date of birth. Special algorithms to account for unique characteristics of foreign names (e.g., inversion of first and family name, standardization of names, conversion to upper case, diverse text functions) were incorporated into the linkage process. Different matching strategies were tried and records were classified as excellent matches, matches needed to be reviewed, and obvious non-matches. The last two groups were reviewed manually and added to a subsequent matching cycle. This strategy resulted in 84% of immigrants successfully matched to entries in the provincial healthcare registry. Matched and non-matched individuals did not differ substantially across most characteristics in the immigrant dataset.19 Some non-matched immigrants may have moved back to their countries or to other provinces shortly after arrival and others may have been classified as non-immigrants. While this miscategorization may affect comparisons between immigrants and non-immigrants, comparisons made between immigrant groups are not likely to be biased because the reference group is internal to immigrants. To avoid misclassification of immigrant status regarding immigrants obtaining their permanent residence after December 31, 2000, we excluded 74,961 infants whose mothers were first registered into the provincial Health Insurance Plan after March 31, 2001 (to account for the 3-month registration waiting period for immigrants), who may have been newcomers either from abroad or from other provinces and who would have been otherwise classified as non-immigrants. After excluding another 2,183 records with missing or invalid information on the outcomes and covariates, the final population size for analyses was 397,470 births. These data were merged with small-area data (census tracts as neighborhoods) from the 2001 Canadian census to obtain a neighborhood deprivation measure.
Use of the data was approved by the Sunnybrook Health Sciences Centre and by the Research Ethics Board of the University of Toronto, Toronto, Ontario.
Preterm birth is a major predictor of perinatal morbidity and mortality, and is associated with childhood disabilities, neurodevelopmental disorders, and adult onset of diseases.20
21 Preterm birth (PTB) was defined as a delivery before 37 completed weeks of gestation. Gestational age is largely estimated by ultrasound dating in Canada and finally determined by the attending physician’s best interpretation of all of the clinical data, backed up by documentation from the nursing staff as a secondary source.22 In Ontario in 2006, 95% of women had their first ultrasound by week 20 of gestation.23
We used a material-deprivation score 24 with mean 0 and standard deviation 1 based on information of the 2001 Census aggregated at the census tract level, which combined the following characteristics: percent of population below the Statistics Canada low income cutoff, percent of population 20 years and over without high school diploma, percent of single-parent families, percent of income comprised of government transfer payments, percent of population unemployed (15 years and over), and percent of homes needing major repairs. Census tracts are relatively stable urban neighborhoods with a typical population of 2,500–8,000 and are relatively homogeneous with respect to population characteristics and living conditions. Census tract boundaries have been found to correspond well to those of “natural” neighborhoods.25 Census tracts were ordered according to their material deprivation score and collapsed into three approximately equal-sized groups (tertiles).
Immigrants’ Duration of Residence
Mothers were categorized as foreign-born if present in the immigration database. “Non-immigrants” or “Canadian-born” included mostly the Canadian-born and a few immigrant women who obtained their permanent residence before 1985, whose immigration status could not be determined due to lack of data. We measured duration of residence in Canada as the difference between the date of delivery and the date of arrival, thus reflecting the time of exposure to the local environment. We used duration of residence as a continuous measure and also collapsed it in approximately 5-year duration groups (15 months–4 years, 5–9 years, 10–14 years, and 15 years and more).
Information from hospital records available for the entire study population was measured at the time of delivery/birth and included infant sex (male vs. female), maternal age at delivery (15–19, 20–24, 25–29, 35–39, ≥40 years vs. 30–34), and parity (primiparae versus multiparae; Table 1).
Information available for immigrants only was obtained from their landing records and last updated at the port of entry on the date of landing to Canada. With the exception of language knowledge, which was self-reported, the remaining information was ascertained based on legal documentation provided by the immigrants during their application processes.
Relevant covariates included maternal country of birth, age at arrival groups (<12, 12–18, 19–25, 26–30, and >30 years), high school graduation (no versus yes), marital status (single, widowed, or separated versus married or common law), immigrant class (economic class, refugee status versus family class), and knowledge of either official Canadian language (none vs. English or French). Countries of birth were grouped into world regions using a modified version of the UNICEF classification (Appendix 1).26
To determine neighborhood effects for each comparison group, we obtained separate variance estimates for non-immigrants and immigrants according to their duration of residence by conducting stratified analyses. We used a two-level model for non-immigrants with births nested within neighborhoods (model 1), and cross-classified random effects models (CCREM) for immigrants (model 2), to account for the simultaneous membership of births to maternal countries of birth and neighborhoods, and stratified by duration of residence groups to obtain separate variance estimates at each length of residence. CCREM are appropriate to model non-nested data structures,27
28 such as ours. Immigrant mothers living in a particular neighborhood may have come from several different countries, and mothers coming from a particular country may settle in different neighborhoods. Thus, this data structure presents a cross-classification of countries and neighborhoods. Details of this approach are given elsewhere.17 For each model, the total variance of preterm birth was partitioned into the variance between maternal countries of birth and the variance between neighborhoods. P values for the variance components were calculated using the Wald test (one-sided).27
To assess how the neighborhood deprivation gradient in preterm birth varied according to immigrant status and immigrants’ duration of residence, we constructed an additional model for all immigrants (model 3). This was a CCREM with births nested within neighborhoods and maternal countries of birth, including a product term between duration subgroups and deprivation tertiles, and adjusted for infant sex, maternal age, parity, and socio-demographic and immigration characteristics that were only available for immigrants. In exploratory analyses, we examined the distribution of births across maternal countries of births and neighborhoods, by immigrant status and immigrants’ duration of residence to ensure that the data met the minimum sample size requirements for multilevel logistic regression.29
From the logistic model, we calculated predicted probabilities of preterm birth and adjusted absolute risk differences in preterm birth (both expressed as percents) between neighborhood deprivation tertiles, with 95% confidence intervals using normal approximation.31 We chose to report absolute rather than relative measures of effect because absolute risk differences better reflect the excess population burden of the outcome attributable to differences in neighborhood deprivation.
In exploratory analyses restricted to immigrants, we tested whether the associations between duration of residence and neighborhood deprivation with preterm birth differed across maternal regions of birth and cohorts of arrival (pre 1988, 1989–1992, 1993–1996, and 1997–2000). Since none of these interactions were statistically significant, we therefore considered maternal regions of birth and cohorts of arrival as potential confounding factors. Cohort of arrival was dropped out because it was not associated with the outcome. We also performed sensitivity analyses considering the effect of age at arrival but this variable was not associated with preterm birth, particularly among young immigrants. Duration of residence was associated with preterm birth across strata of age at arrival and cohorts of arrival. An additional reason to exclude age at arrival is that this variable is a linear combination of maternal age at delivery and duration of residence, which were included in the adjusted models. We also explored whether year of birth was a confounder of the association between duration of residence and preterm birth, since secular increases in preterm birth rates have been reported in Canada.32 The association between year of birth and preterm birth disappeared after immigrants’ duration of residence was added to the model but the unadjusted association between duration of residence and preterm birth did not change after year of birth was added to the model and therefore we dropped year of birth from the models.