Study population and data extraction
A retrospective cohort study was carried out using two data sources. The first source pertained to a birth-cohort, with information on the parents, that was selected from the civil registry database of Statistics Netherlands (Centraal Bureau voor de Statistiek, CBS), consisting of all live births in the urban area of Utrecht and surrounding semi-rural municipalities between January 1, 1998 and December 31, 2007 (n = 106 953; hereafter: CBS-cohort). The CBS is responsible for collecting and processing all individual and population health care data in the Netherlands for the purpose of research by health authorities and academic institutions . Under Dutch privacy law, use of personal (health care) data for the purpose of scientific research is allowed, provided that data cannot be traced to the individual. For 514 individuals (0.5%), the mother could not be identified, for 3735 (3.5%) the father and for 378 individuals (0.4%) neither the mother nor the father.
The second source of the data originates from The Psychiatric Case Registry of the central part of the Netherlands (PCR-MN), which is in operation since 1999. It contains anonymized information on all patients who attended in- or out-patient facilities for mental health care until December 31st 2009, including date of birth, gender, postal code and at least one DSM-IV diagnosis . The PCR-MN does not receive information from GPs, paediatricians or from the small number of psychiatrists working in private practice who rarely see patients with autism. Thus, it is likely that the vast majority of children referred to psychiatric services are recorded by PCR-MN.
Linkage between PCR-MN and CBS-cohort was based on date of birth of the child, gender and a part of the postal code. The postal codes included in the CBS-cohort are identical to those covered by the PCR-MN region. A unique match with a record in the CBS-cohort could be established in 89% of the patients registered in the PCR-MN with autistic disorder (DSM-IV code 299.00), Asperger syndrome (299.80) or PDD-NOS (299.80). Country of birth and parental country of birth are known for all children in the CBS-cohort, because it is part of the information provided at compulsory child registration after birth. In the PCR-MN, parental country of birth was incomplete for 27% of the children. However, we used the available information from the PCR-MN to verify whether ethnic minorities were disproportionally present among cases that could not be linked to the CBS-cohort. The proportions of children for whom successful record linkage was possible were 84% for the native Dutch children, 86% for the Moroccan, 83% for the Turkish, 100% for the Surinamese and Dutch Antillean children and 91% for those from other countries of origin. Thus, in the cohort resulting from the merge between PCR-MN and CBS-cohort, the occurrence of diagnosed ASD may be slightly underestimated, but not differentially between immigrants or native-born children.
For children with a diagnosis of ASD, the period at risk ended at the time of diagnosis, i.e. at the age at diagnosis. For children without a diagnosis of ASD, the period at risk ended at emigration outside the catchment area, death or 31 December 2009, whichever came first. In addition, to evaluate whether any difference between migrants and non-migrants in rates of ASD or its subtypes may be explained by differences in time until diagnosis, age at diagnosis was estimated for each migrant group.
In line with the CBS definition of immigrant status, a Dutch-born subject with two Dutch-born parents was considered native Dutch. A Dutch-born subject with at least one foreign-born parent was classified according to the country of birth of that parent. When the parents were born in different foreign countries, the maternal country of birth was decisive for assignment to a particular group. In an effort to distinguish the effect of maternal from paternal migration, we consecutively reclassified children based on the country of birth of the mother only. Most migrants in the study area originated from Turkey, Morocco, Surinam and the Netherlands Antilles. The latter two groups of migrants from former Dutch colonies with relative geographical proximity were collapsed into a single group in the analyses. Migrants were categorized according to their source country's economic situation to capture the putative influence of discrepancy in the level of development between source and host society. Using the United Nations Conference on Trade and Development (UNCTAD) classification for the year 2011 , countries were divided into three categories: 1) developing economies, 2) transition economies (e.g., Croatia, Ukraine), and 3) developed economies. UNCTAD group 2 including predominantly Eastern European countries was adjoined with group 3, reasoning that these countries mostly resemble countries with developed economies in terms of cultural and ethnic characteristics. This combined group is hereafter referred to as ‘developed countries’. UNCTAD group 1 is referred to as ‘developing countries’.
Group comparisons on demographic variables and age at diagnosis were performed using t tests for independent groups. Incidence rates of any ASD, and separately for narrowly defined autism vs. Asperger syndrome and PDD-NOS, were estimated for each migrant group. Multivariable Poisson regression analysis was used to estimate rate ratios (RRs), adjusted for gender and paternal age. These variables were selected a priori based on the reported association with the incidence of ASD . We adjusted for advanced maternal age at birth, which has also been related to an increased risk for ASD , in separate analyses to minimize the possible effect of collinearity with paternal age. RRs with a confidence interval (CI) not including unity were considered statistically significant. Time of observation was used as offset and the variance was scaled to take possible over-dispersion into account. The Poisson regression analysis was performed with the Generalized Linear Models module of STATA, version 10.0 . The overall between-group differences in risk were tested with the Wald chi-square statistic.
We used the Kaplan–Meier method to estimate the cumulative incidence for a diagnosis of ASD at age 10 in the entire birth-cohort. The age cut-off was set at a maximum of 10 years, because for an older age, the majority of the cohort was not followed up through the period of risk. Because migrants tend to settle in urban areas and urbanicity has been associated with an increased risk for autism , we performed subanalyses on a restricted urban sample of the inhabitants of the city of Utrecht only.