This was a population-based linkage study of kindergarten-aged children (ages 5 to 6 years) with a completed EDI record in 2015 using health data routinely collected from the publicly funded universal healthcare system and other administrative data available in Ontario, Canada. The datasets linked to the EDI used unique encoded identifiers and were analyzed at ICES, an independent, non-profit research institute whose legal status under Ontario’s health information privacy law allows it to collect and analyze health care and demographic data from designated health information custodians, without individual consent, for health system evaluation and improvement.

The EDI is a 103-item validated questionnaire completed by senior kindergarten teachers in publicly funded schools in the second half of the school year that measures children’s ability to meet age-appropriate developmental expectations across five domains: 1) physical health and well-being; 2) social competence; 3) emotional maturity; 4) language and cognitive development; and 5) communication skills and general knowledge [6, 20]. Designed as a population-level research tool, the EDI is a measure of developmental status of children across schools, neighbourhoods, as well as larger geographies, which with repeated implementations, provides indicators of developmental change over time in populations of children. It can be used to help identify cohorts of children who may be struggling, though not yet identified, and may benefit from universal programs. The 2015 implementation of the EDI in Ontario is the 4th provincial level collection in a cycle that began in 2004. The first cycle of EDI data constitutes the Ontario baseline for subsequent years and is used to determine the 10th percentile cut-offs for assessment of vulnerability. Children scoring below this 10th percentile cut-off are at higher odds of not meeting optimal developmental standards [6, 21, 22]. Such children may benefit from cost-effective, universal preventive programs [23]. The EDI also contains parent-reported and teacher-recorded medical and developmental diagnoses for children, including but not limited to, autism spectrum disorder (ASD), asthma, anxiety, cerebral palsy, and attention deficit and hyperactivity disorder (ADHD). These are selected from a list of over 30 diagnoses available to the teacher [23].

The Registered Person’s Database (RPDB), the file of all residents in Ontario who are eligible for universal provincial health care coverage, was used to identify date of birth, sex and alpha-numeric residential postal code via health insurance information. Such coverage includes refugees and immigrants to Canada who have been granted permanent resident status. For health data, we used the Ontario Health Insurance Plan for outpatient physician billings, the National Ambulatory Care Reporting System for emergency department visits, and the Canadian Institute for Health Information’s Discharge Abstract Database for inpatient hospitalizations. The MOMBABY database, an ICES-derived dataset, linked children born in Ontario hospitals to their mothers using birth hospitalization records [24]. We used the mother’s encoded identifier to obtain pre- and post-natal information related to socio-demographic and medical factors of mothers. For children without records in the MOMBABY database, we linked to the Immigration, Refugee and Citizenship Canada’s Permanent Resident Database [25] to determine whether the child entered Canada as an immigrant (e.g., economic class, family class, etc.) or refugee. Children who were not born in a hospital in Ontario or identified as an immigrant or refugee, were categorized as ‘other’, which includes children coming to Ontario through inter-provincial or -territorial migration and children born at home. We derived neighbourhood income quintile, the Ontario Marginalization Index material deprivation quintile [26], and rural residence by linking residential postal code with census data. The Ontario Drug Benefit database identified children and mothers who had ever used social assistance through filling a prescription paid for by the Ontario Drug Benefit Program [27].

In Ontario, children enter Year 1 of kindergarten in September of the calendar year they turn 4 years old (junior kindergarten), and Year 2 the year they turn 5 (senior kindergarten). The majority of children (almost 90%) attend publicly funded schools [28]. All children in publicly funded Year 2 kindergarten (n = 135, 937) [29] in Ontario whose teacher completed the EDI questionnaire in the final term of the year (1^st January to 30th June, 2015, age 5- to 6-years-old) were considered for linkage. Individuals were excluded if their EDI data did not have a minimum number of questions completed (i.e. valid for inclusion filter created by EDI developers that flags useable records) [23]. After linkage of the EDI to health and administrative data, we further excluded records that linked to the MOMBABY database but with a flag indicating a problem with linkage. For example, we excluded children where the infant’s identifier was equivalent to the mother’s identifier.

We retrieved birth date and sex variables for all individuals with a valid Ontario health card number with a year of birth from 2006 to 2010 from the RPDB. These birth years were considered as potentially linkable with children who completed the 2015 EDI in Ontario. For all individuals in the retrieved cohort, we determined their postal code of residence as of 1^st July, 2015 using the RPDB postal code file. We used deterministic linkage and variables of birth date, sex, and postal code to link EDI records to the RPDB file. For records that remained unlinked after this first pass, we expanded the catchment to include the postal code file for 2014. We performed this second pass to account for the possibility of an address change that might not have been reflected in the EDI or 2015 RPDB file. As in the first pass, birth date and sex variables for all individuals with a valid Ontario health card number with year of birth from 2006 to 2010 were retrieved from the RPDB file. For all individuals in the retrieved cohort, we used their postal code of residence as of 1^st July, 2014 using the RPDB postal code file and deterministically linked them to the EDI records. Due to the lack of more specific identifiers in the EDI dataset, such as the child’s name, we were unable to link records with identical values for birth date, sex, and postal code to the RPDB. Thus, multiple birth children of the same sex were excluded from the linked cohort. This approach was used as no unique identifier was available in each database that were common to both datasets.

The outcome of interest was a child’s vulnerability status as determined by their score on each of the five domains measured by the EDI. Children were considered to demonstrate overall vulnerability if they scored less than or equal to the 10th percentile cut-off for Ontario children on at least one domain.

We chose social and medical risk factors for a child’s vulnerability based on previous research that has shown a relationship between these factors [30–35]. Risk factors were categorized as medical or social (Supplementary Appendix A). Medical risk factors related to the birth event (premature birth, low birthweight, neonatal intensive care unit stay at birth) could only be determined for children born in an Ontario hospital. Because extreme prematurity (<28 weeks gestation) and extremely low birth weight (<1500 grams) can contribute to significant neurodevelopmental morbidity, these risk factors were given double the weight of children born preterm at 28 to ≤36 weeks gestation or low birth weight at 1500 to < 2500 grams, and counted as two risk factors each. Other medical risk factors included congenital anomalies, major illnesses, hospitalizations for 6 or more days, and any intensive care unit stay. To determine the presence of a major illness, we used The Johns Hopkins ACG® System Version 10 [36, 37] a case-mix methodology for aggregating conditions into adjusted diagnostic groups (ADGs) that are similar in expected resource use to determine whether children had 2 or more major diagnoses from the first of their birth or their first eligibility for provincial health insurance to their 4th birthday.

Sociodemographic risk factors were derived from both the child’s and, for those born in Ontario, their mother’s encounters with the health system, and involved linkages with census, physician billing, hospitalization, immigration, Ontario Drug Benefit program, and the Narcotic Monitoring System sources [25, 38]. These risk factors include: child or mother was a refugee; child was discharged to social services at birth; child had an outpatient visit or hospitalization for maltreatment; child lives in the lowest neighbourhood income quintile; mother was younger than 19 years of age at first birth; mother had 4 or more children as of the child’s 4th birthday; mother has 10 or more ADGs in the post-natal period; mother has a history of mental illness in the pre- or post-natal period; and mother has a history of an emergency department visit or hospitalization as a result of experiencing violence.

For each child we calculated a score representing the sum of risk factors in each of the medical and social domains where, for example, a score of 0 under the social domain meant the child had no social risk factors. We categorized scores in each of the domains (medical or social) as zero, one, two, and three or more and then created 16 dyads representing a child’s medical and sociodemographic risk factor experience. A dyad of ‘00’ represented a child with no medical or sociodemographic risk factors while the dyad ‘33’ represented a child with both medical and social risk scores of three or more.

We selected five common clinical conditions reported in the EDI by the teacher, often from information provided by the parent, to assess concordance between the EDI-based reports and health conditions recorded in administrative data. Conditions included anxiety, asthma, ADHD, ASD, and cerebral palsy (CP). To determine whether a child had the condition according to administrative data, we observed from when the child had provincial health insurance (birth or migration to Ontario) to 2 years following the completion of the EDI and where the diagnosis was reported in either physician billings, emergency department and hospital discharge records, or in the narcotic monitoring records (ADHD only). For asthma, we used an asthma specific database developed at ICES and compiled based on a validated algorithm for identifying this condition in administrative data [39]. Data sources and definitions of medical diagnoses are provided in Supplementary Appendix A.

To assess potential biases and limitations of the EDI data, we compared the linked EDI data to the unlinked EDI data in a descriptive analysis of variables recorded in the EDI, as well as information obtained from postal code linkages. We used the standardized difference of the mean to test for statistically significant differences where a value equal to or greater than 0.10 was considered significant.

We used descriptive analysis to better understand the relationship between medical and social risk factors and a child’s vulnerability. We plotted risk factor dyads against vulnerability, defined as scoring at or below the 10th percentile on at least one of the five domains or scoring at or below the 10th percentile on two or more domains.

We examined the concordance of health conditions reported on the EDI with the administrative data identification of five health conditions. We calculated the agreement (kappa) between EDI and administrative data in identifying the condition, as well as the positive predictive value of the EDI source to identify a condition relative to the administrative data reference standard.

All data analyses were conducted at ICES using SAS Enterprise Guide 7.1 (SAS Institute Inc., Cary, North Carolina).

This study was approved by the Research Ethics Board at The Hospital for Sick Children, Toronto, Ontario.

Characteristics of children comparing those with linked and unlinked records and only those with linked records included in the study are shown in Tables 1 and 2, respectively. Supplementary Appendix B further details baseline characteristics of linked children who completed the EDI and were included in the study. Children with linked and unlinked records had similar sociodemographic characteristics including age, sex, and rural residence (Table 1). Significant differences were observed between groups by neighbourhood level income quintile with lower linkage rates for children living in the lowest neighbourhood level income quintile (income quintile 1: 22.5% unlinked vs. 18.3% linked; standardized difference [SD] 0.10) and higher linkage rates for those who had been at the school administering the EDI for longer than one month (98.9% linked vs. 94.4% unlinked; SD 0.25). There were no differences between children with linked and unlinked records by special needs designation, mother tongue, by enrollment in French Immersion education, or by positive multiple challenge index assignation (defined as scoring below standardized cut-point on 9 or more of the 16 subdomains). Differences in mean domain scores were observed based on linkage status in physical health and well-being (8.76 ± 1.40 linked, 8.61 ± 1.49 unlinked; standardized difference 0.11), social competence (8.21 ± 1.97 linked, 8.01 ± 2.06 unlinked; SD 0.10), and language and cognitive development (8.82 ± 1.69 linked, 8.61 ± 1.82 unlinked; SD 0.12) domains, with scores being higher in the children with linked records. Differences were not observed between groups in emotional maturity and communication skills and general knowledge domains. When comparing linkage groups based on the proportion below the 10th percentile in any domain, no differences were observed.

Table 3 shows the concordance between select parent reported/teacher recorded medical diagnoses in the EDI and health administrative data records of those same diagnoses. Generally, anxiety, asthma, cerebral palsy, and ADHD were infrequently recorded in the EDI when there was a record in health administrative data. For ASD, records in the EDI occurred in almost half of health administrative data records. ASD showed good concordance with a positive predictive value of 89.4% and a negative predictive value of 98.4% and kappa 0.60. There was generally fair concordance with EDI and administrative data in identifying cerebral palsy with positive and negative predictive values of 91.8% and 99.7% respectively and kappa 0.34.

Among linked individuals, 31,530 (30.3%) demonstrated overall vulnerability (one or more domains), among whom 15,039 (14.4%), 7,240 (7.0%), 4,392 (4.2%), 2,852 (2.7%), and 2,007 (1.9%) had vulnerabilities on 1, 2, 3, 4, and 5 domains, respectively (Table 4). For children born very preterm (< 28 weeks gestation) and preterm (28 to 36 weeks), 54.1% and 36.6%, respectively, were vulnerable on one or more domains; with similar proportions observed in children born with very low or low birth weight. Among children with a prolonged neonatal intensive care unit stay at birth, congenital anomalies, major childhood illnesses, prolonged hospitalizations, and any intensive care unit stay, between 40.1% to 46.0% demonstrated overall vulnerability.

There were 3,366 children who were either themselves or had mothers who were refugees, of whom 1,247 (37.0%) were vulnerable on one or more domains. Out of only 78 children who were discharged to social services at birth, 67.9% were vulnerable on one or more domains (Table 4). Among children with a history of maltreatment, 45.6% were vulnerable on at least one domain. Across all neighbourhood income quintiles, there was a gradient with increasing number of vulnerable domains with decreasing neighbourhood income quintile, with the highest percentage of children (40.0%) with overall vulnerability in the lowest income quintile neighbourhoods. Among children with family members (either themselves or their mother) who received income assistance, 53.9% were vulnerable on at least one domain. About a third of children with mothers with multiple (10+) medical comorbidities (33.7%) and mental illness (33.1%) were overall vulnerable (Table 4). Among only 63 children in the sample with known history of maternal assault, 65.1% of children demonstrated vulnerability.

Supplementary Appendix C shows the number and proportion of the population affected by medical and social risk factors. Most (85.3%) children had no medical risk factors with only 5.3% having two or more medical risk factors. In contrast, 35.1% of children had no social risk factors and 34.1% had two or more social risk factors. Figure 3 shows the proportion of individuals with vulnerability on one or more EDI domains by number of social and medical risk factors. With increasing social or medical risk factors, there was an increase in the proportion with EDI vulnerabilities. Among those with three social risk factors, regardless of the number of medical risk factors, the proportion with overall vulnerability (47.5%–61.7%) was higher than all other combined social and medical risk groups. Among those with three social risk factors and no medical risk factors, 47.5% were vulnerable overall compared with 35.5% of those with three or more medical risk factors with no social risk factors.