We used the ‘Education and Child Health Insights from Linked Data’ (ECHILD) database which comprises linked de-identified data from Hospital Episode Statistics (HES), Office for National Statistics (ONS) mortality records and the National Pupil Database (NPD) [15–17]. HES and NPD contain information recorded on state-funded hospital attendances and school activity in England. HES Admitted Patient Care (APC) contains clinical information on inpatient stays (such as diagnoses and procedures) and captures 97% of all birth admissions in England, meaning that these data can be used to define population-based birth cohorts [15, 18]. Linkage to subsequent hospital admissions in HES APC allows for longitudinal follow-up. ONS mortality data, which is routinely linked to HES, provides information on the cause and date of deaths registered in England and Wales (including those that occur outside of hospital) [19]. The NPD contains several school censuses (Supplementary Table 1, Supplementary Appendix 1), which provide information on school enrolments, as well as pupil-level data, including recorded SEN provision, attainment, absences, and exclusions [16]. We used the Department for Education’s opensource ‘Get Information about Schools’ (GIAS) register to add school type to ECHILD using each school’s unique reference number [20]. We also downloaded congenital anomaly prevalence data submitted by England-based registries to the ‘European Network of Population-based Registries for Congenital Malformations’ (EUROCAT) for external cohort validation [3].

Our study population (the “birth cohort”) included all live singleton births recorded in NHS-funded hospitals in England between 1 September 2003 and 31 August 2013. Birth admissions were identified using a combination of diagnostic and procedure codes, healthcare resource group codes and administrative variables (as outlined in Zylbersztejn et al. [18]). We excluded infants from multiple births, to minimise the risk of linkage error, and infants who were not resident in England, to minimise the risk of loss to follow up [18]. To create the sub-cohort of children attending state-school (the “school cohort”), we then excluded children who: did not link to the NPD; were not present in any NPD census in school years 1 to 6 (ages 5/6 to 10/11 years); or were only present in NPD censuses ≥2 years outside the expected age (see Supplementary Figure 1, Supplementary Appendix 1). Pupils not following the national curriculum do not have a school year listed in the NPD, so were instead assigned a year based on their age in each academic year. The maximum date of follow up was 31 August 2019 (the last full academic year before the COVID-19 pandemic).

We used International Classification of Diseases 10th Revision (ICD-10) codes based on EUROCAT guide 1.5 to define MCAs in this study (Supplementary Table 2, Supplementary Appendix 1) [21]. A child was defined as having “any MCA” if they had one or more relevant ICD-10 code(s) in hospital admission or mortality records during the first year of life (i.e. infancy). The same criteria were applied across the whole cohort, i.e. for all birth years. MCAs were classified by 12 system-specific subgroups. Infants could belong to multiple subgroups (e.g. a child with a nervous system and eye anomaly is counted in both groups), but for each of the 12 subgroups we also identified children who did not have MCAs in any of the other systems (‘isolated’ anomalies). Within the 12 subgroups we also defined 25 conditions (including isolated and non-isolated cases), which were selected based on a group size of ≥200 at the end of follow up to avoid problems with statistical disclosure. Conditions were defined using EUROCAT guidelines, apart from congenital diaphragmatic hernias and anorectal malformations, where we used definitions developed previously [22, 23]. We were unable to apply EUROCAT minor anomaly exclusion rules where 5-digit ICD-10 codes were required (see Supplementary Table 2, Supplementary Appendix 1).

We reported death from any cause before the age of 5 years to coincide with the expected age at entry into year 1, the first full year of compulsory education and the start of key stage 1 in state-funded schools in England. We also present survival rates to age 7 years, corresponding to the expected age at entry into year 3 (and the start of key stage 2). Occurrence and date of death were derived from HES-ONS linked mortality records. Follow-up began at birth and ended at the earliest of: death, the end of the study period (31 August 2019), or each child’s 5^th birthday (or 7^th birthday for survival rates to age 7 years).

In state-funded schools in England, children are entitled to receive SEN provision if they have “a significantly greater difficultly in learning than the majority of others of the same age, or have a disability which prevents them from making use of facilities generally provided by mainstream schools” [24]. There are two broad categories of SEN provision [24]: SEN support (previously, ‘School Action’/‘School Action Plus’), which is arranged by the school and may include different educational materials or small group support; and Education, Health and Care Plans (EHCP; previously, ‘statement of SEN’), which is arranged by local authorities for children whose needs cannot be met by the lower level of provision and may include one-to-one support in the classroom and therapies outside school. To this, we added a third category, ‘specialist provision’, to differentiate between children attending non-mainstream schools where support for all children differs, including smaller classroom sizes and an adapted curriculum (Supplementary Table 1, Supplementary Appendix 1).

We used the NPD alternative provision, pupil referral unit and termly school censuses linked to the GIAS register to define four categories of recorded SEN provision in each school year: (1) none, where there was no recorded SEN nor evidence of attendance at a special school or alternative provision in any NPD census; (2) SEN support in mainstream school, where ‘School Action’, ‘School Action Plus’ or ‘SEN support’ (following the 2014 legislative changes) was recorded in at least one census and there was no record of an EHCP nor a record of attending a special school or alternative provision in any NPD census; (3) EHCP in mainstream school, where a ‘statement of SEN’ or ‘EHCP’ (following the 2014 legislative changes) was recorded in at least one census and there was no record of attending a special school or alternative provision in any NPD census; (4) specialist provision, where there is a record of attendance at a special school or alternative provision in any census.

We derived the following variables from each child’s HES APC record at birth to compare socio-demographic characteristics of the birth and school cohorts: year of birth; phenotypic sex (female or male); region of residence; five-group index of multiple deprivation, based on the lower super output area of residential address recorded at birth [25]; and racial-ethnic group (six major groups aggregated from 16 categories). We use the term racial-ethnic group purposefully to emphasise that this covariate includes both race- and ethnicity-based identifiers (e.g. “White” and “British“, respectively). This paper is descriptive and therefore we do not include adjustment for any covariates in the main analyses.

After excluding multiple births and non-England residents, our birth cohort comprised 6,180,400 singleton live births in NHS-funded hospitals in England from 1st September 2003 to 31st August 2013 (Figure 1). Of these, 3.5% (219,249) had evidence of at least one MCA identified in hospital or death records in the first year of life. The most common anomalies were those of the cardiac (53,741; 0.9% of the birth cohort), limb (34,366; 0.6%), and urinary systems (29,158; 0.5%; Table 1). The estimated prevalence of MCAs in this study was about twice the average prevalence reported by regional registries for available MCAs over the same period, with some exceptions (Supplementary Table 3, Supplementary Appendix 1). Similar prevalence rates were reported for the orofacial and abdominal wall MCA subgroups and for severe cardiac anomalies, congenital diaphragmatic hernia, limb reduction defect and Down syndrome. The prevalence of children with an identified MCA increased from 3.3% amongst births in 2003/04 to 3.8% in 2012/13, with the largest system specific sub-group increase among children with any cardiac anomaly (from 0.7% to 1.1%; Supplementary Figure 2, Supplementary Appendix 1).

For children without an MCA, survival to age 5 and 7 years was stable at 99.7% (95% CI, 99.7-99.7 for each age). Children with at least one MCA had survival rates of 95.1% (95% CI, 95.0-95.2) at 5 years and 95.0% (95% CI, 94.9-95.1) at 7 years. Among the 12 system-specific subgroups, average 5-year survival was under 90% for those with any nervous system, cardiac, respiratory, digestive or chromosomal anomaly. Children with isolated anomalies had higher survival rates than on average for children in their anomaly subgroup. Conversely, survival rates were lower for the isolated chromosomal anomaly group, compared with specific conditions (Down syndrome and Turner syndrome). The condition with the lowest survival rate was congenital diaphragmatic hernia (66.9% to age 5 years, 95% CI 64.4-69.2).

Of the 6,180,400 children in the birth cohort, 16.0% (990,478) were excluded from the school cohort, including: 29,253 children who died before the expected age of entry into year 1; 734,563 children without a linked NPD record; 225,991 children (with a linked NPD record) who were not present in any NPD census during years 1 to 6 within the study period; and 372 present in NPD after date of death (evidence of possible linkage error; Figure 1). The final school cohort consisted of 5,189,922 children, including 3.5% (181,324) with evidence of at least one MCA. Children were more likely to be included in the school cohort if they were born in later study years, lived outside London and were from a White racial-ethnic group (Supplementary Table 4, Supplementary Appendix 1). The proportion of children in the birth cohort who were included in the school cohort was 84.0% among children without an MCA compared with 82.7% of children with any MCA (rising to 84.3% and 86.9%, respectively, when accounting for deaths before age 5 years; Supplementary Table 5, Supplementary Appendix 1). Of all studied MCA sub-groups and conditions only one (indeterminate sex) had inclusion rates under 84% after accounting for deaths before age 5 years.

1 Of the 5,008,598 children in the school cohort without an MCA, 25.7% (1,285,572) had any recorded SEN provision during years 1 to 6 (22.5% with SEN support in mainstream school, 1.8% with an EHCP in mainstream school and 1.4% in specialist provision; Table 2; Figure 2). In comparison, of the 181,324 children with at least one MCA, 41.6% (75,381) had any recorded SEN (27.6% with SEN support in mainstream school, 6.6% with an EHCP in mainstream school and 7.3% in specialist provision). Within system-specific subgroups, any recorded SEN provision was highest among children with any chromosomal (93.1%; 6,712/7,208), nervous system (73.3%; 6,358/8,676) or eye anomalies (66.0%; 2373/3,599). 8.7% (624/7,208) of children with any chromosomal anomaly had SEN support in mainstream school recorded, compared with 25.8% to 39.1% among other system-specific anomaly subgroups.

EHCPs in mainstream school were most frequently recorded among children with any chromosomal (35.3%; 3,548/7,208), nervous system (16.6%; 1,440/8,676) and eye anomalies (15.4%; 555/3,599). Those with abdominal wall, urinary system, genital or limb anomalies were the least likely to have such provision (<5%). Broadly, the proportion of each MCA group with recorded EHCPs in mainstream school was similar to the proportion attending specialist provision, except for children with nervous system and chromosomal anomalies. Among children with microcephaly, 57.4% (952/1,659) attended specialist provision compared with 13.0% (215/1,659) with a recorded EHCP in a mainstream setting. Within the system-specific subgroups, children with isolated anomalies had lower proportions of recorded SEN than on average for the subgroup. The overall pattern of results is comparable when restricting the results to key stage 1 only (years 1 and/or 2), with a slight decrease in the prevalence of recorded SEN across all MCA subgroups and conditions (Supplementary Table 6 and Supplementary Figure 3, Supplementary Appendix 1).

Overall, the prevalence of children with SEN provision recorded in year 1 decreased from 21.1% in 2009/10 to 15.0% in 2018/19 (Figure 3), with the same decreasing pattern observed for children with and without an MCA. As shown in Supplementary Figure 4 (Supplementary Appendix 1), when split by category of recorded SEN provision, this decrease is only observable for SEN support (the largest group; 88.3% of all recorded SEN provision in year 1).

The proportion of children with any recorded SEN provision in year 1 decreased by 4.2% (99% CI -4.3,- 4.2) after the 2014 SEN reforms for children without an MCA, compared with 4.8% (99% CI -5.4,- 4.3) for those with any MCA (Figure 4; Supplementary Table 7, Supplementary Appendix 1). Of the MCA subgroups, the proportion of children with any recorded SEN provision decreased most for abdominal wall (-8.2%, 99% CI -12.2,- 4.1), respiratory system (-6.3%, 99% CI -11.0,- 1.6) and cardiac (-6.1%, 99% CI -7.3,- 4.8) anomalies. Of the categories of SEN provision, these decreases are only observed in SEN support in mainstream school for children without an MCA (-4.6%, 99% CI -4.6,- 4.5) (Figure 5; Supplementary Table 7, Supplementary Appendix 1). Children with any MCA had a similar decrease in SEN support (-4.4%, 99% CI -4.9,- 3.9), together with slight decrease (-0.4%, 99% CI -0.7,- 0.1) in the prevalence of EHCPs in mainstream school.

Patterns of survival for congenital anomalies in our study are similar to pooled estimates presented in a 2020 meta-analysis, highlighting increasing survival for children with many of these conditions over time [6]. Estimates of SEN provision for children with MCAs are similar to those provided from registry data linked to national administrative educational records in four areas of England (41.6% in our study compared with 44.0% from the EUROlinkCAT report) [10]. The decline in the proportion of children with SEN support (formerly ‘School Action’/‘School Action Plus’) following 2014 government reforms fits with patterns reported elsewhere [26]. We add to this evidence by quantifying this decrease (approximately 1 in 20 fewer children receiving SEN provision) and showing that changes affected children with and without MCAs who had additional learning needs at the lower level of provision.

As illustrated in our time series plot, the decrease in SEN support had already begun several years before formal introduction of reforms. Sweeping public sector austerity policies in the early 2010’s, including cuts to school budgets, alongside criticism of the current system of SEN identification likely contributed to this decrease [27]. Prior to 2010, SEN provision had been rising [20], culminating in a report by the Office for Standards in Education, Children’s Services and Skills, which stated that there was “over-identification” of children at the lower level of SEN provision in lieu of “better teaching” [21]. Decreases in recorded SEN provision over this period may therefore reflect changing criteria of the children who should receive SEN provision and at what age. This may explain the trend differences across MCA groupings, although one should not over interpret these differences given the relatively small numbers involved. The intended impact of the 2014 SEN reform on better teaching contrasts with evidence on the experiences of children, their families and teachers, which indicate that changes in SEN provision reflects rising unmet need [28–30]. The possible detrimental impact of these changes is further supported by reports published in 2019 by the House of Commons Education Committee and the National Audit Office that present a picture of a fragmented and increasingly unsustainable SEN system [28, 29].

A strength of our study is the cohort size and comprehensive national coverage. We used a cohort of over six million children (5,189,922 in the SEN provision analysis) to describe less commonly investigated MCA subgroups and conditions. Longitudinal records over a 10-year study period meant that we could examine SEN provision before and after a major policy change. A major limitation relates to the recording of SEN provision not being based on clear criteria for service delivery. Firstly, a record of SEN provision in the educational data does not necessarily indicate that SEN provision was received, and there is no information about the precise elements of support received (hence our label of ‘recorded’ SEN provision). Secondly, in practice, SEN provision varies by factors beyond a child’s learning needs, as highlighted by the time-varying patterns presenting in our study. Other research has shown SEN provision to vary by school, geographical location, child’s ethnicity and previous contact with social services, amongst other factors [31]. A further limitation is that diagnostic codes of MCAs are not necessarily indicate additional learning needs, as they do not capture severity or functional impairment. For example, the diagnostic codes used to define conditions in the eye anomaly group (anophthalmos/microphthalmos, congenital cataract and congenital glaucoma) do not indicate whether a child has unilateral or bilateral eye disease, with the former unlikely to require SEN provision in isolation [32]. We are undertaking further work, with comprehensive clinical input, to define groups of children who are most likely to require SEN provision.

Using hospital admission and mortality records during the first year of life, we identify MCAs in 354.7 per 10,000 live singleton births, compared with 185.5 per 10,000 live and still births in registry data over the same period [3]. Differences in completeness of reporting, coverage (England-wide compared with one third of the population) and ages at data collection (infancy compared with predominately birth or newborn examinations) over the time period studied may partially explain our higher rates [4]. However, our study also misclassifies some children with minor anomalies due to the granularity of ICD-10 codes required to exclude some conditions. We also find an increase in the prevalence of MCAs in our study over time, which may indicate both increased detection and depth of coding in hospital records over time, particularly for milder congenital anomalies [7, 22]. Arguably, changes to the population of children with MCAs captured over time (potentially with less need for education support, on average), may have led to an overestimation in the difference of SEN provision after SEN reforms. However, the same pattern of results are observed for children with orofacial and abdominal anomalies—MCA subgroups with comparable rates to registry data and stable prevalence rates over the study period—, which give us confidence in the interpretation of our results. That is, that children with lower level SEN needs (with and without MCAs) were less likely to have SEN provision recorded in the latter half of the 2010’s. Further studies, including validation via linkage to the national congenital anomaly and rare disease registration service (national coverage available since 2018 [33]) and primary care records, are required to improve case ascertainment and confidence in, or adjustment for, the precision of results for a wider group of congenital anomalies [34].

Overall, 84% of children in the birth cohort and alive at age 5 years were included in the school cohort; meaning the results may not be representative of all children born in England [17]. Approximately half of the children excluded would be expected to: attend non-state funded, or private, school (7% of the population) [17]; be home schooled (estimated to be 0.6% of the population) [17]; or have emigrated out of England before starting school (no reliable estimates are available but, based on emigration statistics for all ages in England, rates are unlikely to exceed 0.5%) [35]. The remaining non-links (~8%) represent missed links or potential duplicates between the health and educational datasets (leading to an inflated denominator). Nonetheless, rates of linkage across MCA groups and children without an MCA are similar, giving us confidence in our comparative results. The rate of inclusion in the school cohort was below 70% for children with indeterminate sex, suggesting that findings for this condition should be interpreted with some caution. We theorise that lower rates of follow-up in this group are the result of linkage error, since the algorithm for linkage between datasets in ECHILD uses child sex and children with indeterminate sex are more likely to have changes to sex/gender assignment over time. Continued improvements to linkage in the ECHILD database over time, will further strengthen the national application of results to children attending state-funded schools in England [17].

Clinicians and families want precise information about the prognosis of children with MCAs beyond infancy. The work presented here shows how large administrative datasets can be harnessed to provide such information, albeit through a rationed service that is subject to policy change. As the ECHILD database is accessible to government and UK researchers [36], the approach in this study could be developed to use de-identified data routinely for planning and monitoring of SEN provision. For example, together with our ongoing work to phenotype children in need of SEN provision, the need for early specialised support could be identified in HES by the end of infancy to inform service planning by local authorities. Specialist paediatric services could use such information to prepare parents for the type of support likely to be received. Administrative data offers an important resource for understanding variation and inequities in the provision of SEN services and, if combined with qualitative evidence of areas of good practice, and randomised or quasi randomised methods, to assess SEN practices that have a beneficial effect and for whom. The application of these methods could extend to examining the impact of changes in clinical practices or support (for example, a new drug or therapy) on educational outcomes, such as attainment and school absence.

Future research comparing MCA cohorts across UK countries and internationally will provide evidence on how education and health outcomes in children with MCAs, compared with peers, vary in different jurisdictions with contrasting timing, intensity and pedagogical approaches to SEN provision. Continued work on validating coding for specific MCA groups is important to support this comparative work. SEN provision has been called ‘a postcode lottery’ in England [31], but the extent to which this represents underlying inequities has been difficult to capture to date. Our approach of defining similar groups of children using health records could be used to explore whether children with the same underlying needs are systematically managed differently in educational settings, depending on the school or local authority or both. Extension of this work to children without MCAs who are in need of SEN provision is also important, given that we find that children with MCAs only account for a minority (5.5%) of all children with any recorded SEN provision and less than two in ten children with EHCPs.