Education systems worldwide need to balance delivery of effective mainstream education with additional activities or support for children who experience difficulties with learning compared to their peers [1]. In this study of the English context, we refer to such additional support as special educational needs or disability (SEND) provision (see Appendix 1, Table A1 Section 5) [2]. Typically, SEND provision involves a wide range of types and intensity of activities, from continuous one-to-one support to occasional group work, across various types of needs. Activities change as the child responds and develops [3]. SEND provision also includes collaboration between multi-sector professionals and families, all within a complex, publicly funded system [4, 5].

Children likely to need SEND provision include those with disability, preterm birth or chronic physical or mental health conditions that affect learning [5–7]. In many countries, children with behaviour problems, speech or language difficulties, low school attainment, social disadvantage, and those who speak a different language are also assigned SEND provision [1]. Countries worldwide report rising demand and increasing costs of SEND provision, reflecting improved survival among children with medical conditions, increasing recognition of neurodevelopmental disorders, and global efforts to expand educational access and provide inclusive education for all [8–10].

This paper provides an overview of the Health Outcomes of young People throughout Education (HOPE) study and synthesises findings of sub-studies conducted by the programme [11]. The aim of the HOPE study was to examine variation in SEND provision in English primary schools and assess its impact on children’s health and educational outcomes articulated across four main questions [11]. We conducted 12 quantitative studies using linked health and education administrative data from the Education and Child Health Insights from Linked Data (ECHILD) database [12] and 10 mixed methods studies. Comprehensive descriptions of methods and results can be found in 30 papers listed in Appendix 4.

We organised the paper as follows. First, we summarise existing evidence on the effectiveness of SEND provision and outline SEND policy and practice in England. Second, we give an overview of the HOPE study questions and methods - including the ECHILD data resource and how we derived cohorts and synthesised findings. Third, we report findings for each of four research questions and present key findings. Fourth, we present our conclusions on variation in SEND provision and its impact on outcomes and discuss strengths and limitations of the study. We end by discussing challenges posed by administrative data relevant globally and the implications of our findings for policy and future research.

In England, SEND provision seeks to support four broad areas of functional needs: communication and interaction, cognition and learning, social, emotional and mental health, and sensory and physical needs [13–15]. These needs often arise from physical or mental health conditions or from social contexts that affect areas of child development [5, 6, 16, 17]. High quality SEND provision has the potential to improve health and education outcomes in childhood, and thereby health and wellbeing in adolescence, with long-lasting impacts on health and economic inequalities in adulthood [6, 18–20]. However, evidence for effective SEND provision is mixed.

On the one hand, an international systematic review of 467 comparative studies of children with known additional needs (including 297 randomised controlled trials; RCTs), reported consistent beneficial effects of different types and durations of SEND activities compared with service as usual or alternative education activities [4, 21]. On average, these benefits equate to 5 to 6 months of improved learning in maths, reading, or related skills. Most studies included in the review were focused on targeted activities for well-defined learning difficulties (e.g. 40% were for dyslexia), rather than evaluating SEND provision across the breadth of usual practice [22]. In addition, the impact of these targeted interventions, assessed under controlled research conditions, may not generalise to the complex, interdependent processes through which SEND provision operates in real-world school settings. Few studies have accounted for the underlying health conditions and health inequalities that shape both the need for, and the outcomes of, SEND provision.

Evidence for the effectiveness of SEND provision as delivered in schools is more limited than the evidence base for targeted interventions. SEND provision as a service involves a variety of interventions for a wide range of different needs. The delivery of SEND services intersect with socioeconomic position and family factors, health, social care services, and local service capacity. Such complex processes are hard to evaluate. So far, limited evidence of benefit has been reported by numerous observational studies over the past 25 years. As expected, descriptive studies found that children assigned SEND provision had worse outcomes than peers, due to confounding by higher underlying health and education needs [16, 22, 24]. However, 49 studies that attempted to account for confounders using quasi- or natural experimental designs, also found no difference or worse outcomes for children assigned SEND provision [22, 23, 26–30]. Similarly adverse findings were reported in an evaluation of teaching assistants in England, a group who are typically involved in delivering SEND provision [31].

Recent advances in causal methods are presented as a framework for using natural experiments, also called quasi-experimental methods, to evaluate population health interventions [32, 33]. The framework refers to a range of methods to achieve comparability (or exchangeability) between those with and without intervention. Various methods can be used to account for measured confounding, including those within the target trial emulation framework (TTE) [32, 34]. Methods often used to address unmeasured confounding include difference-in-difference and instrumental variable techniques. A meta-analysis of 44 U.S. studies that tried to minimise confounding found eight studies that addressed the bias arising from unmeasured confounding [28]. These studies used policy changes [35, 36], or changes within children’s SEND status over time [37, 38], or differential SEND thresholds across racial groups, in natural experiment approaches [35, 39, 40]. All eight found effects of SEND provision: five reported beneficial impacts, while three reported harmful effects for certain subgroups. For example, two studies found worse outcomes for children who were just below the threshold for SEND provision who received provision due to policy changes [39, 40, 41]. These effects for marginal groups might not be generalisable to all children who need SEND provision.

In summary, uncertainty remains about the benefits of SEND provision as delivered in routine practice globally, as well as in England. At the same time, there is strong evidence that targeted activities, delivered as key elements of the SEND process, are effective, albeit often for narrowly defined needs and outcomes [9]. Policy makers need robust evidence on what types of SEND provision work, for whom, and under what conditions, and on the effectiveness of the overall SEND process, to deliver effective support while containing costs. Applying robust study designs to linked administrative health and education data provides a unique opportunity to fill these evidence gaps.

Despite the uncertain evidence [2, 42], England has seen major policy changes to funding and delivery of SEND provision over recent decades, which offer opportunities for natural experiment designs that address unmeasured confounding. Since 2001 [43], policy has shifted towards inclusion of children receiving SEND provision in mainstream education. The Children and Families Act (2014) [2], and the SEND Code of Practice (2015) [3], aimed to improve SEND joint working with health and social care services and devolved decisions and funding for less intensive SEND provision to schools. However, schools were challenged by a more competitive, attainment-focused, marketisation of education that disincentivised inclusion of children unlikely to perform well in attainment tests. In 2022, a House of Lord’s review of the 2014 Act concluded the government had ‘ultimately failed’ to improve outcomes, due to failure of implementation [44].

Implementation of the 2014 Act coincided with two broader system shifts. Austerity-related cuts to local authority (LA) budgets from 2010 reduced SEND and social care funding, and expansion of academy schools from 2007 drove competition between schools and widened inequalities in SEND provision between schools [8, 45]. Analyses using English primary school education data showed that the primary school a child attends is a stronger predictor of SEND provision over any personal or family characteristics [25].

SEND provision affects a substantial minority of children in England. One-third of all children in state school in Year 1 in 2005/6 had been assigned SEND provision by age 16 [11]. Approximately 30% of all children were assigned SEN Support, which is determined and delivered by schools. Since then, between 3-5% of all children have been assigned an Education, Health and Care Plan (EHCP), which is assessed, approved and part-funded by local authorities (Appendix 1 Table A1, Section 5) [3]. SEN Support typically includes in-class assistance, teaching aides or adaptive learning, while EHCPs provide tailored and personalised support for more complex needs. Approximately half of all children who receive an EHCP attend a special school [3]. Since 2010, the proportion receiving SEN Support has declined, while use of EHCPs has risen steadily since 2016, driving annual SEND expenditure in England above £11 billion in 2025 [8, 42]. Further reforms to the SEND provision system will be published by the Department for Education (DfE) in 2026 [46, 47].

The HOPE study aimed to examine variation in SEND provision in English primary schools and assess its impact on children’s health and educational outcomes [11]. We addressed four main research questions and consulted with young people, parent-carers and professionals throughout the study (Appendix 2):

We used the ECHILD (Education and Child Health Insights from Linked Data) database to address the first three questions and mixed methods studies to address the fourth (Figure 1). Details of data resources are in Appendix 1, Table A1, sections 4-6.

We synthesised findings from the sub-studies addressing the four research questions through an iterative process. First, researchers integrated findings within each individual question. Then, teams working on each question shared their results throughout the study to summarise common and contrasting emerging themes across all four research areas and to identify issues for further discussion with stakeholders (see Appendix 2). We generated conceptual diagrams to map findings between studies (e.g. Figure 7). In the final stage of synthesis, we met public and professional and government stakeholders to discuss the implications of findings (Appendix 2).

The HOPE study used the ECHILD data database for the 12 sub-studies addressed by research questions 1 to 3 listed above. The ECHILD database links routinely collected health and education data in England. For the HOPE study we used ECHILD version 1, which includes all children and young people aged 0-24 years who were born in a state-funded hospital between 1 September 1995 and 31 August 2021 (approximately 14.7 million individuals) [12]. Health and education datasets were linked by National Health Service (NHS) England using a multistep deterministic linkage algorithm [48]. Linkage rates were high and increased overtime (92% of school pupils born in academic year 1990/1991 were linked to a hospital record, compared with 99% of pupils born in 2004/2005) [48].

Health data from the Hospital Episode Statistics (HES) database includes records of all NHS-funded hospital care in England. We created birth admission cohorts from 1 September 2003 onwards as linkage to subsequent admissions, outpatient appointments and accident and emergency attendances was high from this point onward. HES admissions include birth records for approximately 97% of all children who are born in NHS facilities and captures an estimated 98–99% of all secondary care contacts [11, 49, 50]. HES records contain demographic details (e.g., sex, ethnic group, area of residence) and admissions data include clinical information based on International Classification of Diseases, 10th Revision (ICD-10) diagnostic codes and Office of Population Censuses and Surveys, 4th Revision (OPCS-4) procedure codes [50].

Education data from the National Pupil Database (NPD) capture information on children attending state-funded schools in England from the 2001/02 academic year onwards [51]. The NPD includes pupil registration, attainment, exclusions, and absences, alongside individual- and school-level characteristics such as local authority, ethnicity, gender, Index of Multiple Deprivation (IMD), English as a first language, free school meal eligibility, social care involvement, and SEND status [11, 51]. Further details are in Appendix 1, Table A1, Sections 4–6.

Ten birth cohorts were constructed from ECHILD to meet the specific analytic needs of each sub study (Table 1; Appendix 3). Sub studies used longitudinal designs and spanned multiple academic primary school years (eg: 1 September 2003 to 31 August 2004) onwards [11, 51]. Follow-up periods were defined according to each study’s objective, exposure group and educational variables of interest and data availability. For example, data access for cohort 4, which was analysed at the start of the study, was limited to births in 2004/5 [52].

Most analyses using educational outcomes up to the end of primary school (Key Stage 2, aged 10-11 years) restricted birth years to the end of the 2007/8 academic year (31 August 2008). This ensured that all children had finished primary school before the COVID-19 pandemic which disrupted attendance, learning and assessments [53]. The January school census was used to identify children enrolled in each academic year, as this census determines school funding allocations and is therefore considered the most complete record of pupil enrolment [54].

The first step of the HOPE study involved developing high-need health phenotypes to proxy groups of children likely to need SEND provision. This step is critical, as education data captures only the assignment of SEND provision but no indicators for the underlying need for additional support. Using external evidence, clinical input and insights from ECHILD data regarding the reliability of diagnostic coding in administrative data, we chose three groups of health phenotypes which capture populations with different levels of need for SEND provision: neurodisability [55], congenital anomalies [56], and preterm or early term births, defined by week of gestational age [52, 57]. These phenotypes were mainly recorded before school entry and, on average, were expected to have mild to moderate (congenital malformations) or moderate to high (neurodisability) need for SEND provision or a gradient of need by week of gestational age at birth. We described outcomes for each of these phenotypes, including variation between subgroups within each phenotype. Combinations of neurodisability and other high need health phenotypes were described for children with Down syndrome. In addition, we grouped children with any chronic health conditions in analyses of factors that influence SEND provision (Question 2, Table 1 cohorts 6,7).

We assessed health and education outcomes for children with high-need phenotypes and unaffected peers, from school entry to Year 6 (age 11) using cohorts 1 to 4 (Table 1) [52, 58–60]. Outcomes included: mortality before school start, comorbidities, planned and unplanned hospital admissions, unauthorised school absences, and whether a child reached the expected level of development, based on school test scores at ages 5, 7 and 11 (Appendix 1, Table A1. Section 6.6)

Children in each of the three high need health phenotype groups had more unplanned admissions [59, 61], more comorbidity [55, 62], more frequent absences [61], and lower attainment [58, 60, 63, 64], than their unaffected peers. As expected, there was substantial variation between phenotype subgroups (Appendix 4, papers 2-10). We illustrate these patterns by summarising results for children with neurodisability. Of 3.58 million children, 2.4% had a hospital-recorded neurodisability before age 5, rising to 3.6% by age 11 (Cohort 1) [55]. These children were admitted to hospital five to seven times more often than their peers and accounted for 15% of all hospital bed-days during primary school (for the 2.96 million children who had linked education records in primary school in Cohort 2) [61]. We also found 60% higher school absence rates and consistently lower attainment during primary school among children with neurodisability [61, 63]. At age 5, only 30% of children with neurodisability before school entry reached a ‘good level of development’ (GLD), compared with 58% of peers, and fewer than half met expected levels at ages 7 and 11 (Figure 2a) [63]. By the end of primary school (Figure 2a, Key Stage 2), over a third of children with neurodisability were not assessed in national attainment tests, contrasting with 6.4% of unaffected peers enrolled in school [63]. Outcomes varied significantly by neurodisability type and in each phenotype subgroup (Figure 2b) [63].

Children with congenital anomalies comprised 3.5% of 2.35 million children entering primary school (Cohort 3). Attainment varied between subgroups, with the lowest average attainment through primary school occurring among children with chromosomal or neurological congenital anomalies [64]. Average differences between children with and without congenital anomalies were small to moderate [64]. We found lower attainment across all anomaly subgroups for boys than girls [64]. Separate analyses of the whole population (Cohort 4) revealed lower attainment for each week of gestational age at birth before and after 40 weeks of gestation (Figure 3) [52].

See key findings, Table 3.

We planned to use two quantitative approaches to evaluate the impact of SEND provision on health and education, i.e. to perform causal analysis. Both approaches are within the natural experimental framework [32]. The first approach involved assessing whether observed changes in policy, or differences in practices between schools or local authorities, introduced sufficient exogenous variation to enable the use of instrumental variable methods for the estimation of the impact of the new policies (see Appendix 1, Table A1 Section 3 for definitions). These methods can account for unmeasured confounding [32]. The second approach used the target trial emulation (TTE) framework to define causal questions that could replicate an ideal RCT with observational data [70]. The framework requires defining the causal contrasts of interest, the exposure (SEND provision), the relevant outcomes and confounders. Principled methods to deal with measured confounding, such as propensity score-based and g-methods are then adopted (under the additional assumptions of no interference, consistency and positivity). By making all elements of the causal question explicit, the TTE framework helps to minimise biases, for example, due to misalignment of study entry, eligibility and exposure assessment, or to incorrect assignment of later exposure status to an earlier period [71].

Our analyses for question 2 did not reveal any suitable instrumental variables to support the first approach. We therefore evaluated the impact of SEND provision using the TTE framework. We focussed on two sub-phenotypes of children: those with cleft lip with palate and those with cerebral palsy (Cohorts 9 and 10) [72, 73].

We chose cleft lip and/or palate as a phenotype likely to reflect mild to moderate need for SEND provision, because average differences in education and health outcomes between affected children and peers are small (Cohort 9) [72]. The cerebral palsy phenotype was chosen to proxy high need for EHCP provision using specific codes for cerebral palsy (Cohort 10) [73]. For each cohort, we iteratively refined the phenotype to increase similarity in need by excluding children with anomalies affecting other body systems. Outcomes were unplanned hospital admissions, unauthorised school absence in Years 2-6 and attainment in mathematics in Year 2 (age 6/7) and Year 6 (age 11).

We estimated the impact of SEN Support in Year 1 versus none for both phenotypes, and EHCP versus SEN Support in Year 1 (for cerebral palsy) on unplanned admissions, unauthorised absences and key stage test scores through primary school, Other comparisons were not possible because of lack of positivity [72, 73].

We found no evidence that SEN Support in Year 1 compared with none improved rates of unplanned hospital admissions or educational attainment in either the cerebral palsy or the cleft lip and/or palate phenotype subgroups. Findings were similar for EHCP vs SEN Support for children with cerebral palsy, except for reduced unauthorised absences (Table 2) [72, 73]. There was evidence that SEN Support compared to none reduced the rate of unauthorised absences for both phenotype groups (Table 2).

The findings of improved and worse outcomes in Table 2 should be interpreted with caution given the likely impact of residual confounding arising from the coarseness of the measures available in ECHILD. Furthermore, our choice of outcomes was limited by the accuracy and relevance of data available. For example, absences may be less likely recorded as unauthorised for children with an EHCP. Also, other outcomes relevant to learning or socio-emotional skills may not be captured in ECHILD [72, 73].

See key findings, Table 3.

We conducted ten mixed methods sub studies to understand experiences of the SEND process from the perspectives of young people and parent-carers using the service and from professionals contributing to SEND provision in some way [11]. We examined three core stages of the SEND process: noticing the need, SEND assessment, and implementation of provision. We explored outcomes of these processes for the child and family, and wider influences such as service capability and policies (Figure 7). The mixed methods sub studies included qualitative studies (interviews and focus groups), online surveys (with children, young people, parents/carers, and professionals) and document reviews of Ombudsman complaints, Ofsted/CQC inspection reports, Local Offer (LO) websites and grey and peer-reviewed literature on variation in SEND processes and provision at LA and multi-academy trust levels (Appendix 1, Table A1, Section 5). Appendix 5 summarises methods and key findings from each of the 10 sub-studies mapped to stages of the SEND process (identification, assessment, received provision), outcomes, local capability and policy.

Nearly one-third (30%) of children who entered primary school were assigned SEND provision by age 11 in 2014-19. SEND provision was strongly linked to health characteristics such as having a chronic health condition, including neurodisability or a congenital anomaly, or being born too early. However, these characteristics alone were not sensitive indicators of SEND provision. Other factors, such as being male, facing social disadvantage, and a low developmental score at age 5, were also important.

When children were assigned SEND provision was complex. Children with social disadvantage or disability, which is linked to eligibility for state-provided nursery at age 2-3, had higher rates of SEN Support and EHCP provision throughout primary school than their peers. Despite this increased provision, many did not start SEND provision until Year 1 or 2 (age 5–7), possibly due to delayed recognition, lack of school capacity, or changing need as education became more complex. However, in analyses of children with cerebral palsy, we found that those entering state education in Nursery and living in the 20% most income-deprived neighbourhoods were assigned EHCPs later on average than those from less deprived neighbourhoods [67].

Which type of school children attended was associated with their chances of SEND provision. Children attending voluntary, religious schools and academy sponsor-led schools were less likely to be assigned an EHCP compared with community, foundation and academy converter schools [65].

Findings from ten mixed methods studies highlighted variability in services. Some children and families reported meaningful benefits from provision, however, interactions with services, delays and inadequately tailored interventions frequently included negative experiences, some experiencing long term harm. Qualitative studies and surveys revealed unmet need and delays in identification of need, assessment, planning and provision, that particularly affected disadvantaged children who did not have parent-carers who could advocate for them. Parent-carer, child and professional perspectives and document reviews reflected services that lacked capacity, training and specialist expertise. These findings also highlighted a system that interacted poorly with other weakened systems (such as Child and Adolescent Mental Health Services) and lacked national guidance and accountability. These problems were compounded by underfunding, academisation, and a narrow curriculum that was perceived to undermine inclusion of children’s diverse abilities.

We found weak evidence that SEND provision in Year 1 (age 5/6) reduced rates of unauthorised absences, but no evidence for reductions in hospital admissions or improved attainment during primary school. These equivocal results are consistent with 49 natural experiment studies worldwide that evaluated SEND provision as implemented in routine practice (see introduction) [22, 23, 26–30].

These quantitative findings were complemented by qualitative findings in the HOPE study from parent-carers and children using SEND services and professionals. Their reports indicated that timely, tailored provision improved academic engagement, anxiety, and independence for children. However, the stresses of the SEND process, including inappropriate or inadequate provision, were experienced as harmful by children and their families.

Strengths of the HOPE study included exploration of the evaluability of SEND provision using different approaches to causal analyses using rich, longitudinal health and education data for all children in state-funded primary education in England captured in the ECHILD database. The value of combining mixed methods with quantitative methods for policy evaluation has been emphasised by Craig et al [32]. In the HOPE study, mixed methods and stakeholder engagement, alongside the quantitative analyses, were critical for understanding variation in the adequacy of SEND processes and the many elements that were not measured in administrative data.

Limitations centred on a data science problem: inadequate measures of the need for SEND and other confounders, limited information on the nature of the intervention, and insensitive or insufficiently relevant outcomes that might not be measurably changed by SEND provision. A further limitation was that we found no instrumental variables to use in natural experiment designs that address unmeasured confounding, despite major changes in national SEND policy and types of school over the past two decades. These limitations undermined the evaluability of the impact of SEND provision on health and education using quantitative analyses of ECHILD.

Our findings reflect challenges that are relevant to other jurisdictions that use administrative data to evaluate SEND provision. One lesson from the HOPE study is that health phenotypes as recorded in hospital data do not adequately proxy need for SEND provision. Linked hospital and education data lack direct measures of function, disease, severity or complexity of need, and social and behavioural factors linked to high need for SEND provision and worse outcomes [81].

Secondly, SEND provision recorded in ECHILD reflects assignment to the SEND process but not what activities or support were received, at what intensity and when. Young people and parent-carers reported frequent delays in receiving support, watered-down activities or activities the child did not tolerate, due to lack of capacity, skills or ability of staff to adapt activities to the child.

Third, the outcomes measurable in ECHILD - hospital admissions, attainment test results and school absences - are relatively insensitive to positive changes reported in response to SEND provision by parent-carers and young people. They mentioned feeling happier, less anxious or distressed, improved sense of self, behaviour and mental health, and being able to attend school, participate, learn and socialise. Impacts on parent wellbeing were also mentioned, while reduced hospital contacts for the child were not.

Data linkage to relevant information from administrative, cohort, survey, trial or audit data on need for, and content of, SEND provision, and on relevant outcomes, could mitigate these three challenges. Examples exist elsewhere. In Wales, researchers can access anonymised education data linked to primary care and hospital data, further enhanced with survey data on function and wellbeing in the HAPPEN cohort [82].

Fourth, the SEND process may have been harmful for some children and families. Our study spanned a period of national austerity, with cuts to education and related services, rising child poverty and need for SEND provision. These changes occurred in parallel to policy changes that incentivised academic attainment and reduced funding and accountability for progress among children with additional needs. Parents and practitioners reported erosion in the capacity or skills to deliver SEND provision of adequate quality. These problems created distrust between parents and school staff and undermined partnership working between staff, child and parent-carer.

Fifth, notwithstanding these limitations, the HOPE study illustrates the opportunities to inform policy makers, teachers, clinicians and families about children’s trajectories through health and education and into adolescence. Descriptive analyses while being essential to design causal analyses, also provided important information on expected outcomes and variation and inequalities in SEND provision.

Achieving effective and equitable SEND provision is a policy challenge in many countries. SEND provision is an expensive and important service - £11 billion per year in England - with the potential to change the life opportunities of one-third of all children [8]. More effective and efficient services could improve experiences and outcomes for children and their families, limit escalating costs, and yield long-term benefits across public services as better supported, more independent and educated children with additional needs transition to adulthood. Broader policy changes are also needed beyond SEND provision, including broadening the academic curriculum to value and measure other forms of progress, including social and emotional skills. Beyond schools, better integration is needed between health and education services for children with additional health and social needs, and more support for parent-carers to reduce the toll on families.

Further reforms to SEND in the UK and elsewhere [8], need to be based on, and continue to generate, evidence of what works. Development of high quality linked data research infrastructure across education and other public services should be at the forefront of government’s research areas of interest. Government should plan to integrate primary care data into the ECHILD database, to provide more granular data on the health determinants and consequences of education for children with additional health needs.

Despite the hundreds of RCTs of targeted SEND interventions and, with this study, 49 natural experiment studies on the impact of SEND provision, uncertainty remains about what works for whom, in what contexts and how. Achieving effective SEND provision requires randomised trials of alternative approaches to SEND provision in situations where uncertainty exists and comparators are acceptable. For example, the HOPE study found that SEND is provided to few children in the general population in the preschool years (Nursery and Reception class). Although earlier implementation makes intuitive sense, and is provided for some disabilities such as visual impairment, there is limited evidence of benefit from contemporary nursery-based trials [83, 84]. A randomised, staggered trial should be considered to implement policy change in England and, at the same time, generate robust evidence on early versus deferred SEND provision in the early years [85].

Policy makers also need evidence on what good SEND provision and alternative practice options look like to develop meaningful evaluations of impact. Our mixed methods studies revealed a SEND service that few users experienced as being able to meet needs due to lack of funding, staff, skills, interagency support and trust from families [42]. Comparison of such a weakened system of SEND provision against education as usual is unlikely to advance policy.

More and better research is needed to guide development of effective and equitable SEND provision globally. International collaboration could accelerate delivery of research for example, sharing knowledge on administrative data linked across education, health and other sectors to enable parallel studies across diverse needs and contexts relevant to SEND provision. Surveys, cohort studies, and RCTs of SEND provision, or targeted components, in education or healthcare, could be conducted and linked into administrative data. Such linkages would minimise costs of data collection, and if conducted across jurisdictions, could improve comparability and give insights about differing impacts according to context. Sharing of mixed methods approaches would improve understanding of services across jurisdictions.

International research collaboration is needed to build on the findings from the HOPE study through wider integration of natural experiment, randomised and mixed methods studies to improve the effectiveness of SEND provision. However, the challenges facing the HOPE study, are echoed by other population-based, service interventions, such as early home visiting for first-time, teenage mothers and perinatal mental health hubs [86, 87]. The recent framework on use of natural experimental designs [32], needs to be extended to consider how policy makers, funders, and researchers can introduce and document changes in policy and practice and efficiently enhance administrative data to enable robust evaluation of education, health and social interventions in populations.