In Australia, and other high income countries, core primary prevention services for young children include the provision of a universal child health service, led by child health nurses [26], and high-quality Early Childhood Education and Care (ECEC), led by early childhood teachers [27]. In Australia, all children are offered child health services as part of state and territory government funded primary health services. These services start in infancy and are free-of-charge to children’s caregivers at point of use. There is no legislative requirement for families to use these services and participation is voluntary [28].

In Australia, all children are offered 600 hours of ECEC in the year before they start full-time school. This is equivalent to 15 hours of play-based learning a week for 40 weeks of the year in school settings. There is no legislative requirement for families to enrol children in ECEC and participation is voluntary [27].

Parents and the home learning environment are the most proximal and formative influence on young children’s development [12]. In Australia, parent-child early education programmes are provided by government, non-government and not-for-profit organisations. There is no national universal parent-child early education programme and the provision of parent-child early education programmes is the responsibility of state and territory governments. The delivery of these programs by government departments is guided by the Commonwealth Department of Education and Training Family-School Partnerships Framework [29] and Early Years Learning Framework [30]. Tasmania is the only state/territory in Australia that delivers a universal parent-child early education programme. Launching into Learning (LiL) is a school-based early learning program for children from birth to age 4 years developed by the Tasmanian Department of Education [31].

The health and education sectors provide complementary universal services for the same population of children and their caregivers [15, 30, 32], yet research has largely focused on service use within the health or education sector and not across both sectors, effectively dividing early childhood into silos. Studies that have used population-wide linkage of administrative data records to investigate patterns of participation and non-participation in universal nurse-led child health checks [17, 18] and early childhood education and care [20, 21], have reported socio-economic inequalities in access and use of these services. This siloed view of universal service coverage limits opportunities for cross-sectoral collaboration and coordinated efforts to improve children’s access and use of core universal services.

Given the consistent finding that the burden of developmental vulnerability is concentrated in the most disadvantaged population groups combined with the convergent evidence that the uptake of universal child health and early education services favours more advantaged children; the logical next step is to investigate service use and developmental outcomes at the interface of the health and education universal service systems. The overarching aim of this study was to produce evidence about children’s service use across the health and education sectors with the aim of supporting cross-sectoral collaboration and coordination of core universal early childhood services.

The aims of this study were to investigate (1) patterns of health and education service use from birth through Kindergarten (age four years); (2) associations between cumulative risks and service use patterns; and (3) associations between service use patterns and developmental vulnerability in the first year of full-time school (age five years).

The study was conducted in Tasmania, Australia. The study used a linked dataset comprising five de-identified unit-record administrative datasets collected by the Tasmanian Department of Health (Tasmanian Perinatal Data Collection, Child Health and Parenting Service), Tasmanian Department of Education (Launching into Learning, Kindergarten) and the Commonwealth Department of Education and Training (Australian Early Development Census).

The AEDC is a national teacher-report measure of child development collected in children’s first year of full-time school enrolment across government, independent and Catholic schools. In Tasmania, children start full-time school (Preparatory Year) in the year when they are five years of age on or by 1 January, in any year. The AEDC also collects school and demographic variables from state/territory governments, independent and Catholic schools. The AEDC comprises 96 licenced items across five child development domains: Physical health and wellbeing, social competence, emotional maturity, language and cognitive skills (school-based) and communication skills and general knowledge, and has satisfactory validity and reliability [37]. Children receive a score between zero and ten, where zero is the most developmentally vulnerable. Scores are classified into percentiles, determined using cut-off points established in 2009 [38].

Children who score below the 10th percentile on one or more AEDC domain(s) are classified as developmentally vulnerable (DV1). This was the outcome measure used in this study. Cut-off scores are based on all children who participated in the first national AEDC data collection in 2009, excluding children with special needs, and apply to all AEDC data collections. Children with special needs are those who require special assistance due to a medical diagnosis, a diagnosis of a chronic medical, physical or intellectually disabling condition. Children with special needs are not included in the calculation of the results and do not receive a domain category (e.g., developmentally on track, at risk or vulnerable). The number of children with special needs is reported in each AEDC collection. Children who are younger than 4 years receive a domain score, but are not assigned to a category as their age cannot be validated and cut-offs are age-dependent. Children whose teachers have answered less than 75% of the items in any domain will not receive a score at all. If the teacher indicated that a child had been in their class for less than one month and they did not know the child well enough to complete the AEDC instrument, then that child will have no instrument data [35].

The study used a cumulative risk index [3] comprising 11 binary birth, health behaviour and socio-demographic risk variables. Variable selection was based on evidence of associations between the risk factors and developmental vulnerability in previous studies that used population-level linkage of administrative data records [8, 39, 40], including a prior study of this study population [41]. The risk factors identified in previous studies include biological risk factors (e.g., low birth weight), sociodemographic risk factors (e.g., young maternal age, low maternal education, socioeconomic disadvantage), and health behaviour risks (e.g., smoking in pregnancy) [6, 39, 40]. The risk factors selected for modelling in this study feature in international child development monitoring frameworks [14, 42] and Australian monitoring frameworks. Parental educational attainment is monitored in Australia’s National Assessment Program [43], socioeconomic disadvantage is monitored in Australia’s Health reporting framework [44], and low birth weight, teenage mothers, Indigenous mothers, parity, smoking in pregnancy, alcohol use in pregnancy and language background other than English, are monitored in Australia’s Mothers and Babies reporting framework [45]. For all the risk variables, children were coded with ‘1’ if the risk factor was present and ‘0’ if the risk factor was not present. The cumulative risk index was computed by summing the number of risk exposures (0-11). Cumulative risks were reported as zero, one, two, three and four or more risks.

The service use measure comprised the child’s first recorded use (yes/no) of each of the eight Child Health Assessments (two weeks, four weeks, eight weeks, four months, six months, twelve months, two years and four years), use of Launching into Learning (yes/no), and use of Kindergarten (yes/no) for a total of ten possible service events. For each of the ten service events, children were coded with ‘1’ for participation and ‘0’ for non-participation. Thus each child could record zero to ten service events in total across the first 4 years of childhood. This approach to measuring service use is consistent with Australian government population monitoring and reporting systems [28, 35, 45] and studies with a principal interest in quantifying and understanding differential use of universal early childhood services [17–19, 49, 50].

The analysis proceeded in four steps: (1) Latent Class Analysis (LCA) was used to identify and describe service use groups; (2) information from the latent class model was used to assign children to service use groups; (3) multinomial logistic regression was used to estimate the odds of service use group membership associated with cumulative risks, relative to the reference group; (4) multivariable logistic regression was used to estimate the adjusted odds of developmental vulnerability on one or more AEDC domains associated with service use group membership, relative to the reference group. All LCA analyses were conducted in SAS PROC LCA V1.3.2 [51]. The multinomial logistic regression and multivariable logistic regression analyses were conducted in SAS V.9.4 [52].

The proportions of missing data for risk factors ranged from 0.02% (birth weight) to 24.89% (primary caregiver education). Missing values for risk factors were imputed using PROC MI in SAS V.9.4 [52]. We generated 25 imputed data sets, with results averaged according to Rubin’s rule [53], using PROC MIANALYZE. To account for arbitrary patterns of missing-ness, we used the fully conditional specification method [54]. No auxiliary variables were added to the imputation process. AEDC outcomes were not imputed. The distribution of developmental vulnerability on one or more AEDC domains (DV1) and individual risk factors for children with observed and imputed data is reported in Supplementary Appendix A.

Service use in this population was heterogenous. Across ten service events, 2¹⁰ (1,024) patterns of service use and non-use were possible, and 493 combinations were observed in the data.

LCA describes a series of distinct classes (i.e., groups) and assigns an item-response probability within each class [55–57]. Entropy and the A-BIC (Adjusted Bayesian information criterion) were used as the statistical measures of model fit in this paper. Given the A-BIC indicated a 5-class model and Entropy suggested a 2-class model, all options between two and five classes were considered for interpretability. A 4-class model was selected on the basis of statistical criteria, reported in Supplementary Appendix B, and the researchers’ judgements that: The 4-class model produced a clear distinction between classes, all classes were substantial in size, and each class could be assigned a meaningful descriptive label [58]. Information from the LCA model was used to assign children to each of the four classes. Individual children were assigned to latent classes based on their maximum posterior probability [59].

The four latent classes were given descriptive labels based on the pattern of service use that distinguished each class: Regular service use, Low service use, Declining service use and Selective service use. Figure 1 shows the probability of participation in each service event by service use group, and Table 1 shows the distribution of service events by service use group.

In this study population, the number of risks varied from zero to eight risks. 27% of children were exposed to zero risks, 29% to one risk, 22% to two risks, 13% to three risks and 9% to four or more risks.

The distribution of cumulative risks varied by service use group. Children in the Regular service use group were exposed to an average of 1.2 risks (95% CI 1.2 to 1.3), children in the Declining service use group were exposed to an average of 1.7 risks (95% CI 1.6 to 1.7), children in the Low service use group were exposed to an average of 2.0 risks (95% CI 1.9 to 2.1), and children in the Selective service use group were exposed to an average of 1.4 risks (95% CI 1.3 to 1.5). The distribution of individual and cumulative risks by service use group is reported in Supplementary Appendix C.

Multinomial logistic regression was used to examine the associations between cumulative risks (zero to four or more risks) and service use group membership. Regular service use was the reference category. Table 2 shows the odds of service use group membership associated with increasing levels of cumulative risks. The odds of Declining service use group membership increased with two or more risks. The odds of Low service use group membership increased with one, two, three, and four or more risks. The odds of Selective service use group membership increased with three or more risks.

In this study population, 20.9% of children were classified as developmentally vulnerable on one or more AEDC domains, compared to 22.0% nationally (39). In the Regular service use group 411 (17.2%) children were classified as developmentally vulnerable: 295 (24.2%) children in the Declining service use group; 270 (28.1%) children in the Low service use group; and 111 (17.3%) children in the Selective service use group.

Further insights into universal early childhood use would be gained from linkage of data records that include measures of the intensity, timing and duration of service use.

A limitation of this study is that there was no gold-standard external dataset defining true match status that could be used to compare to the research data set [65, 66]. However, the close concordance between the Australian Early Development Census National Report 2015 [35], and the study cohort, suggests that bias due to linkage errors did not disproportionately lead to an underestimate of the service needs of more disadvantaged population groups. The study was limited to universal services provided by government departments and did not include services provided by the non-government, not-for-profit, and for-profit sectors. The study only used single records of attendance at CHAs and did not include use of the CHaPS for purposes other than CHAs. The service use measure comprised a single record of a child’s service use (yes/no) and did not capture important dimensions such as the intensity, timing and duration of service use. Nevertheless, harmonizing records of 10 child health and early education service events across the first four years of childhood into patterns of service use did reveal important differences in the uptake of early childhood services by groups of children with different levels of cumulative risk exposures. Latent class analysis is a data-driven approach to identifying patterns of service use that carries the risk of the results being sample-dependent. The use of a whole population sampling frame and use of administrative service records reduced the risk of selection bias and recall bias from the records of service events.