Gestational age and child development at school entry

Studies have reported a dose-dependent relationship between gestational age and poorer school readiness. The study objective was to quantify the risk of developmental vulnerability for children at school entry, associated with gestational age at birth and to understand the impact of sociodemographic and other modifiable risk factors on these relationships. Linkage of population-level birth registration, hospital, and perinatal datasets to the Australian Early Development Census (AEDC), enabled follow-up of a cohort of 64,810 singleton children, from birth to school entry in either 2009, 2012, or 2015. The study outcome was teacher-reported child development on the AEDC with developmental vulnerability defined as domain scores < 10th percentile of the 2009 AEDC cohort. We used modified Poisson Regression to estimate relative risks (RR) and risk differences (RD) of developmental vulnerability between; (i) preterm birth and term-born children, and (ii) across gestational age categories. Compared to term-born children, adjustment for sociodemographic characteristics attenuated RR for all preterm birth categories. Further adjustment for modifiable risk factors such as preschool attendance and reading status at home had some additional impact across all gestational age groups, except for children born extremely preterm. The RR and RD for developmental vulnerability followed a reverse J-shaped relationship with gestational age. The RR of being classified as developmentally vulnerable was highest for children born extremely preterm and lowest for children born late-term. Adjustment for sociodemographic characteristics attenuated RR and RD for all gestational age categories, except for early-term born children. Children born prior to full-term are at a greater risk for developmental vulnerabilities at school entry. Elevated developmental vulnerability was largely explained by sociodemographic disadvantage. Elevated vulnerability in children born post-term is not explained by sociodemographic disadvantage to the same extent as in children born prior to full-term.

. It is a teacher-completed instrument collected for all children in their first year of full-time school (known as pre-primary in Western Australia, the year level prior to grade 1). The AEDC is conducted every three years, and results are reported at the national, state and territory, community, and local community levels. AEDC cut-off scores are based on the 2009 data collection, and apply to all subsequent AEDC data collections 36 . Children who score < 10 th percentile in a given domain are classified as 'developmentally vulnerable' 37 . Domain scores are not calculated for those students classified as 'special needs' , as the AEDC has not been validated for use in this population. A child is classified as 'special needs' if they require special assistance because of chronic medical, physical, or intellectually disabling conditions. We used two aggregated outcome measures; developmentally vulnerable on one or more AEDC domains (DV1), and developmentally vulnerable on two or more AEDC domains (DV2), and assessed developmental vulnerability on each of the five AEDC domains.
Exposure variables. Gestational age was assessed as the clinical best estimate, based on the last menstrual period when ultrasonography-based estimates at 16-19 weeks' gestation were not available 38 . To assess the consistency of findings between gestational age and developmental vulnerability in the analysis, we first assessed the relationship between preterm birth, categorised as; (i) extremely preterm (< 28 weeks), (ii) very preterm (28-31 weeks), (iii) moderate/late preterm birth (32-36 weeks), and (iv) term birth (≥ 37 weeks). We subsequently assessed associations between gestational age and child development outcomes by categorising gestational age into seven categories; (i) extremely preterm (< 28 weeks), (ii) very preterm (28-31 weeks), (iii) moderate/late preterm birth (32-36 weeks), (iv) early term (37-38 weeks), (v) full-term (39-40 weeks; the reference category), (vi) late-term (40-41 weeks), and (vii) post-term (≥ 42 weeks), in line with clinical conventions 39,40 . Other analysis variables. We classified a range of maternal-, child-, and family-level characteristics, as potential confounders, selected on the basis of previous study findings and data availability (Table 1) [41][42][43] . We considered maternal smoking during pregnancy and pregnancy complications (such as gestational diabetes) as factors that consistently affect gestational age therefore these variables were not included in the analysis. Likewise, covariates such as birthweight and 5-min Apgar scores, considered to be associated with both gestational age and child development, were not included in the analysis, to minimise collider bias 44,45 .  The total number of siblings were derived as the number of live births each mother had prior to the year that the cohort child's AEDC was conducted. The Index of Relative Socioeconomic Disadvantage (IRSD) 51 using residential address at the time of the child's birth was obtained from Birth Registrations. IRSD is derived from Australian Census data and reflects area-level disadvantage and given a score from 1 (most disadvantaged) to 5 (least disadvantaged).
Multiple imputation. Complete covariate information was available for 81.5% (n = 52,819) of the study population. A total of six covariates had missing data; (i) maternal immigration status, (ii) maternal marital status at birth, (iii) maternal occupation status scale, (iv) preschool attendance, (v) reading status at home, and (vi) IRSD. All variables used in the analysis had < 2.5% missing data, apart from maternal occupation status (5.7%), preschool attendance (4.1%) and reading status at home (8.0%). Multiple imputation with chained equations, using 20 imputed datasets, was applied to minimise bias attributable to missing data 52 , and the adjusted analyses presented were performed by pooling estimates from imputed datasets. Statistical modelling. The association between gestational age and risk of developmental vulnerability was modelled using modified Poisson regression with robust error variance to estimate the relative risk 53,54 . We specified a series of models to adjust the results, (i) Model 0 was unadjusted; (ii) Model 1 adjusted for child's sex and age at time of AEDC, (iii) Model 2 additionally adjusted for potentially confounding sociodemographic and maternal variables (maternal age, marital status, occupational status, immigration status, parity, maternal ethnicity, child speaks a language other than English, total number of siblings and IRSD category) and (iv) Model 3 additionally adjusted for potentially modifiable variables (preschool attendance and child reading status at home). Relative Risk (RR) and Risk Difference (RD) and associated 95% confidence intervals (CIs) were estimated for developmental vulnerability within each gestational age category compared to children born fullterm. All statistical analyses were conducted in SAS 9.4 55 .

Sensitivity analysis.
As a sensitivity analysis to assess the effect of multiple imputation, we compared the main outcomes based on (i) the imputed data (n = 64,810) to (ii) the complete cases only (n = 52,819; Supplementary Table 1). Children classified as 'special needs' on the AEDC or with a WARDA record lacked outcome data, however, it is possible that some of the conditions which result in children being classified as 'special needs' are related to gestational age at birth. Therefore, we also conducted a sensitivity analysis that assumed a 'worst-case' scenario, whereby all children with an otherwise complete/valid AEDC record, classified as either 'special needs' on the AEDC or with a WARDA record were classed as developmentally vulnerable (Supplementary Table 2; n = 71,196). To investigate outlier influence, we removed all records of children with a proportion of optimal birthweight (POBW) 56,57 less than two standard deviations from the mean for the cohort (Mean POBW: 99.3%; standard deviation [SD], 12.6; i.e., children with a POBW < 74.1% were excluded; Supplementary Table 3). Finally, we excluded records of children with reported maternal smoking during pregnancy (Supplementary  Table 4), as maternal smoking may confound the gestational age-child developmental vulnerability relationship disproportionally and as prevalence rates of maternal smoking during pregnancy are higher for mothers of low socioeconomic status, and in Indigenous Australian women 46 .

Results
The mean gestational age at birth was 39 weeks (SD: 2). Overall, a total of 25,242 (38.9%) of children were born prior to full-term, whilst 7520 (11.6%) were born post full-term (Table 1). Children born prior to full-term were more likely to have indicators of socioeconomic disadvantage including, the mother not being married at the birth of the child, lower maternal occupational status, younger maternal age at child's birth and being an Indigenous Australian (Table 1) (Table 1). Overall, 21.4%, and 10.3% of children born full-term were classified as DV1 and DV2, respectively (Fig. 2). The RD and the RR of children being classified as DV1 and DV2 exhibited reverse J-shaped relationships with gestational age (Fig. 2). In the unadjusted models children born prior to full-term had an increased RR of being classified as DV1 and DV2 (Fig. 2) compared to children born full-term. Adjustment for sex and age of child (Model 1) attenuated the associations between gestational age and developmental vulnerability for all gestational age categories except extremely preterm birth (Supplementary Fig. 1; p < 0.001 for all statistically significant results). Adjustment for potential socioeconomic and maternal confounders (Model 2) also attenuated associations between gestational age for all gestational age categories (Supplementary Fig. 1; p < 0.05 for all statistically significant results). Similarly, adjustment for modifiable confounders (Model 3) further attenuated the associations between gestational age for all gestational age categories except extremely preterm birth (Supplementary Fig. 1; p < 0.05 for all statistically significant results). Compared to children born full-term, children Unadjusted and Adjusted, Relative Risk and Risk Difference from interaction models for the association between developmental vulnerability on the Australian Early Developmental Census (AEDC), and gestational age. The proportion of the study population classified as developmentally vulnerable (a) on one or more AEDC domains, and (b) on two or more AEDC domains, overlayed with relative risk of developmental vulnerability for each outcome, compared to children born at full-term, and the risk difference of developmental vulnerability on (c) one or more the AEDC domains, and (d) two or more AEDC domains. Developmental vulnerability was defined as scores in the bottom decile, based on the 2009 AEDC cut-offs. Adjusted model based on pooled analysis from 20 imputed datasets controlling for; maternal age at time of child's birth, maternal marital status at time of child's birth, maternal ethnicity, maternal immigration status, maternal occupational status at time of child's birth, parity, age of child at time of AEDC completion, child's sex, preschool attendance, child speaks a language other than English at home, child's reading status, total number of siblings, and Index of Relative Socioeconomic Disadvantage category. All relative risk and risk difference data is presented with 95% confidence intervals; modified Poisson Regression.  Fig. 1).

Sensitivity analysis.
The sensitivity analysis revealed that the overall associations between gestational age and developmental vulnerability at age five were not substantially different between the complete cases and the imputed cases (Supplementary Table 1). Furthermore, the pattern of association between gestational age and developmental vulnerability was similar to the main analysis when children classified as either 'special needs' on the AEDC or with a WARDA record were included in the developmentally vulnerable group (Supplementary Table 2).

Discussion
In this study we found the risk of developmental vulnerability for all five AEDC domains, and the aggregated measures (DV1 and DV2) of child development, followed a reverse J-shaped relationship with respect to gestational age at birth. The risk of developmental vulnerability was highest in children born extremely premature and decreased with increasing gestational age through to children born late-term. Although not statistically significant for children born post-term, the risk of developmental vulnerability generally increased. Adjustment for sociodemographic and maternal characteristics accounted for a substantial proportion of the risk associated with birth prior to full-term. Furthermore, adjustment for modifiable risk factors (preschool attendance and reading status at home) attenuated the relative risk and risk differences for all gestational ages, except children born extremely preterm and late-term. Several studies have reported an association between preterm birth and poor school readiness [5][6][7][8][9][10][11][12] . For example, a US study of children from the Early Childhood Longitudinal Study Birth Cohort (ECLS-B) reported children born very preterm (< 32 weeks) had an increased odds of poor school readiness for both reading and maths (aOR: 2.58 and 3.38, respectively) when compared to term-born children (39-41 weeks) 8 . However, there was no statistically significant difference in school readiness for children born moderate/late preterm (32-36 weeks) or early term (37-38 weeks) compared to term-born children 8 . We also found that children born prior to full-term had an increased risk of being classified as developmentally vulnerable when compared to children born fullterm. However, we also reported that all children born prior to full-term (including children born early term) had increased risk of developmental vulnerability on both aggregated AEDC measures. Variations between our study and the ECLS-B study may be attributable to differences in the definition of the reference category and differences in methodology, such as covariates included in the adjusted models. Furthermore, the results of the ECLS-B study are in contrast to our findings and other non-US 9-11 and US 12,17 studies which have reported a dose-dependent relationship between gestational age and poor school readiness, however, these studies did not include analysis of children born post 41 weeks of gestation [9][10][11][12]17 .
A study of Australian children enrolled in government schools in New South Wales (approximately 70% of the state) examined the association between gestational age and risk of developmental vulnerability on one or more (DV1) AEDC domains 14 . This study reported, as we did, that developmental vulnerability followed a reverse J-shaped relationship between gestational age; no statistically significant risk of child developmental vulnerability was observed for children born late-term or post-term 14 . The categorisation of gestational age in the New South Wales study does not adhere to clinical guidelines, and thus results between this study and our study are not directly comparable due to variations in definitions of gestational age categories used. Furthermore, the results of the New South Wales study were restricted to children attending government schools, therefore are not generalisable to the whole population. Similarly, a South Australian study examining effects of gestational age at birth on child development outcomes for children born ≥ 37 weeks' gestation reported, as we have, that children born after ≥ 42 weeks' gestation had an elevated risk of being classified as DV1, but this risk was not statistically significant 19 . In fact, the study reported no statistically significant risk of child developmental vulnerability among children born after ≥ 37 weeks' gestation 19 . Differences in results between our study and the South Australian study may be attributable to the relatively smaller sample size of the South Australian study (n = 12,601) which resulted in wider confidence intervals. Albeit, statistically insignificant the results of the South Australian and New South Wales studies, combined with the results of our study add to the evidence   www.nature.com/scientificreports/ base that the risk of developmental vulnerability of children born after full-term is not equivalent to being born at 40-41 weeks' gestation.
As the exact aetiology is yet to be fully elucidated, there are likely to be numerous, multifactorial mechanisms by which gestational age can impact child development outcomes in children born early term, late-term, and post-term. We found that early term born children had an elevated risk of being classified as DV1, DV2 and developmentally vulnerable on all AEDC domains, expect Emotional Maturity. Embryology studies have reported that total brain volume increases considerably during the final weeks of gestation 60 . Absolute cortical grey matter volume increases four-fold between 30 and 40 weeks of gestation whilst, the absolute volume of white matter increases five-fold between 25 and 41 weeks of gestation 61 . It can be postulated that early term births may result in disruptions to the in utero neural development and thus, increase the risk of child developmental vulnerabilities. As this disruption in rapid neural development does not occur in late-term born children, this may also explain the reduced risk of child developmental vulnerabilities observed in this population. Furthermore, late preterm and early term born children may have shortened breastfeeding durations compared to term/post-term children 62 . Shorter breastfeeding durations 63 may also explain the increased risk of developmental vulnerability in children born prior to full-term. In the case of children born post-term it is possible that the observed increased developmental vulnerability may be related to a decreased perfusion 64 . Decreased umbilical blood flow, increasing foetal weight and reduced oxygen transport capacity 64 , may result in a suboptimal intrauterine environment. Increased exposure to this suboptimal environment may further increase the likelihood of adverse short-and long-term child development outcomes for post-term born children.
A study using data from the Longitudinal Study of Australian Children found that children born preterm had significantly lower cognitive school readiness after controlling for social and perinatal risk factors 9 . Preschool enrolment was observed to be positively associated with cognitive skills at kindergarten entry although the positive effect was not found to be more pronounced in premature infants from socially disadvantaged backgrounds, suggesting that current developmental interventions may not be equipped to mitigate the effects of preterm birth, but have the potential to improve school readiness across gestational age 9 . A limitation of this study was that the Peabody Picture Vocabulary Test used in the study can underestimate the cognitive skills of children for whom English is not their first language or spoken at home 9 . Thus, it is unclear if the beneficial effects of preschool enrolment extend into other domains of early childhood development and across population sub-groups. The results of our study reported that preschool attendance and child's reading status at home were associated not only with a decreased risk of children being classified as DV1 and DV2, but also reduced risk of children being classified as developmentally vulnerable on each of the five AEDC domains, for all gestational age categories, except extremely preterm and late-term born child. Thus, mechanistically our findings suggest that although children born preterm are at the greatest risk for development vulnerabilities, emphasising the increased risk of poorer neurocognitive development and their associated downstream effects, modifiable risk factors (such as preschool attendance and reading status at home) can improve child development outcomes across multiple developmental domains. Furthermore, the finding that modifiable factors do not attenuate the risk of developmental disadvantage in children born extremely preterm maybe indicative of the fact that poorer neurocognitive development in these children cannot be mitigated via modifiable risk factors.

Strengths and limitations.
The main strengths of this study were the large representative sample, the range of sociodemographic variables considered, the categorisation of gestational age in line with clinical conventions, and the assessment of a broad range of child developmental outcomes. A limitation was that although we adjusted for several indicators of socioeconomic disadvantage, any residual confounding related to unmeasured family-level characteristics cannot be ruled out. Additionally, we were unable to explore the effect of other important social risk factors such as parenting experience and/or practices, stability and quality of housing, and the total number of people residing within a household.
Interpretation. The results of this study add to the current knowledge base of the relationship between gestational age and the behavioural, emotional, physical, and cognitive capacities that develop in the first five years of life. The reverse J-shaped relationship observed reinforces the adverse effects of prematurity on later life outcomes and supports recent concerns that the risk of developmental vulnerability for children that are born prior to full-term or post-term (≥ 42 weeks gestational age) is greater compared to those children who are born full-term. Thus, there may also be unmet developmental needs for children born moderate/late preterm and post-term. Our findings also suggest that much of the relative gaps in developmental outcomes for children born prior to full-term can be accounted for by socioeconomic disadvantage and that modifiable factors such as preschool attendance and reading status at home are likely to improve child development in this vulnerable population. In Australia, approximately 7.0% of singletons are classified as preterm, whilst 0.6% of all births are classified as post-term 46 . Given that clinical practice guidelines rarely consider the longer-term implications of prolonged pregnancies 65 , and there are limited studies examining the effects of prolonged pregnancy on later outcomes, further well-designed research is required to fully elucidate the implications of prolonged pregnancies for children. Given the cumulative nature of school-based learning, children who begin school with poor school readiness often fail to catch up with their peers and fall further behind as they progress through schooling 28 . This research highlights the importance of gestational age as an important risk factor for child developmental vulnerability, school readiness and thus, later educational outcomes.

Conclusions
Children born prior to full-term are at an elevated risk of developmental vulnerability at school starting age. In addition, we reported a small increased risk of developmental vulnerability for children born post-term. Elevated vulnerability in children born prior to full-term is largely explained by sociodemographic disadvantage. Children born post-term have an elevated risk of developmental vulnerability compared to children born prior to fullterm. However, adjustment for sociodemographic factors did not reduce the risk of developmental vulnerability in children born post-term to the same extent as children born prior to full-term.

Data availability
The linked administrative data are owned by the government departments who approved the linkage and use of the data for this study. Use of the study data is restricted to named researchers. The current Human Research Ethics Committee approvals were obtained for public sharing and presentation of data on group level only, meaning the data used in this study cannot be shared by the authors. Collaborative research may be conducted according to the ethical requirements and relevant privacy legislations. Potential collaborators should contact author G.P. with their expression of interest. The steps involved in seeking permission for linkage and use of the data used in this study are the same for all researchers.