Primary school attainment outcomes in children with neurodisability: Protocol for a population-based cohort study using linked education and hospital data from England

Patient and public involvement

Prior to developing this protocol, independent meetings were conducted with patient, pupil, and public engagement groups to understand their views on the use of linked health, education, and social care data for research. These groups included the Young Person’s Advisory Group (YPAG), Council for Disabled Children’s Group (FLARE), and the Great Ormond Street National Children’s Bureau Families Research Advisory Group (FRAG). On 14^th November 2020, FLARE were introduced to the ECHILD database and felt strongly about the need for better awareness about administrative data to better support those similar to them. A positive attitude was expressed towards the design and conduct of the HOPE study and the importance of covering the whole population to investigate the interrelated areas of health and education was emphasised. After a meeting with the YPAG on the 27^th November 2021, we learned that exams and performance in school was a particular point of stress for children with health conditions, forming a key outcome measure in this study. Additional meetings with these groups further underlined the importance of continued recruitment of the public to embed children and young people’s voices in the way our findings are interpreted and applied. Using these interactions, we formed our research question which was presented to the HOPE study steering committee. They continue to review and advise the presentation and dissemination of study findings. Key learnings from past public engagements can be found here.

Ethics and dissemination

Permissions to use linked, de-identified data from Hospital Episode Statistics and the National Pupil Database were granted by the Department of Education (DR200604.02B) and NHS Digital (DARS-NIC-381972). Ethical approval for the ECHILD project was granted by the National Research Ethics Service (17/LO/1494), NHS Health Research Authority Research Ethics Committee (20/EE/0180), and UCL Great Ormond Street Institute of Child Health’s Joint Research and Development Office (20PE06).

Access to the ECHILD database is approved by the ECHILD team (ich.echild@ucl.ac.uk) and data can only be used within the Office for National Statistics Secure Research Service by approved researchers.

Findings will be disseminated to diverse stakeholders (academics, relevant government departments, service users and providers) through seminars, workshops, and peer-reviewed publications. Methods and code will be published to enable others to reproduce and extend our analyses using ECHILD.

Study design

This is an observational population-based cohort study using linked health and education records.

Dataset and linkage

We will use the pseudonymised ECHILD database which links administrative data from the National Pupil Database (NPD) and Hospital Episode Statistics (HES). The linkage rate between HES and NPD is high and improved with time (94–98%)^9,10. ECHILD contains data on approximately 14.7 million children and young people aged 0–24 in England who were born between 1^st September 1995 and 31^st August 2020^11,12. Linking health and education records at the individual level allows for the construction of longitudinal histories to explore how diagnosis of a particular health phenotype impacts an individual’s trajectory of health and education.

Education data from the NPD contains information on children attending state-funded schools in England for academic terms from 2001/2 onwards^11,12. Information on the registration, attainment, exclusions, and absences of these children are included, as well as pupil-level details on the school, local authority, ethnicity, gender, index of multiple deprivation, English as their first language, free school meal status, social care related data and SEN status. Anonymised Pupil Matching Reference (aPMR) numbers are produced by the Department for Education (DfE) to link education records for the same pupil across their schooling years.

Health data from HES contains records of healthcare contacts with all NHS funded hospitals in England, including details on admitted patient care, outpatient appointments, accident and emergency utilisation, and critical care from April 1997 onwards: individuals are linked through different healthcare contacts recorded in HES using an encrypted pseudonymised patient identifier^11,12. HES records contain basic demographic information (e.g., sex, ethnicity, area of residence) and clinical information based on International Classification of Diseases 10^th Revision (ICD-10) diagnostic codes and Office of Population Censuses and Surveys Classification of Interventions and Procedures 4^th Revision (OPCS-4) procedure codes. Information on birth characteristics such as delivery date, gestational age, birthweight, and maternal age are also included. As most secondary care in England occurs in NHS or NHS-funded hospitals, nearly 97% of all children born in NHS-funded hospitals in England have a birth record in HES, and an estimated 98–99% of secondary care contacts are recorded in this dataset^11–13. From 1^st January 1998 onwards, HES records are linked to Office for National Statistics (ONS) mortality data covering details on mortality causes and timing of deaths.

HES and NPD records are deterministically linked by NHS England through an algorithm which uses real-world identifiers (including name, date of birth, sex, and postcode). For each matched pair of identifiers from education and health, a pseudonymised child identifier is attached. This is created specifically for ECHILD, and researchers are not provided access to personal identifiable information.

Study population

The study population includes all singleton children born in an NHS-funded hospitals between 1^st September 2003 and 31^st August 2008 (academic years 2003/4 to 2007/8) who were linked to NPD and recorded as enrolled in reception (age 4/5) at state-funded mainstream schools, academies, special schools, and alternative provision. Children who appear in the January (Spring) census of reception will be included in the study. Children will be excluded if they do not subsequently appear in the January census for year two (aged 6/7) and year six (aged 10/11). This will be done to keep a constant cohort of children who appear in the census of each year of interest (academic years in which children are examined by key stage assessments in primary school). Those excluded may not appear in the subsequent census due to emigration, transition non-state funded school, or death. Children who are two years or more outside of their expected age for their school year will be excluded. At each stage, we will enumerate the number of pupils excluded and compare included and excluded children to ascertain whether there are potential issues of selection bias.

This population was chosen as these children would be expected to have completed primary school (end of year six, aged 10/11) by the 31^st August 2019 (Figure 1). This was the last academic year of follow-up before the COVID-19 pandemic, which caused disruption to school attendance, children’s learning, and completion of primary school assessments¹⁴. The January school census is used to capture our study population as it is used for the allocation of school funding, so is assumed to be the most complete¹⁵.

Figure 1. Expected age at each year of primary school by birth year and follow-up year.

R = reception; Y = year; birth and follow-up year defined according to the academic calendar (e.g. 2003/4 includes 1^st September 2003 to 31^st August 2004, inclusive).

The study population will be followed-up at three time points: term 3 of reception (EYFSP assessment), term 3 of year 2 (KS1 assessment), and term three of year six (KS2 assessment).

Key variables

Exposure: Neurodisability. Children’s hospital admission and mortality records from birth up to the 31^st August of year six (age 10/11) will be used to identify children with neurodisability based on ICD-10 diagnostic codes and OPCS-4 procedural codes. The codes used to indicate cases of neurodisability were derived from previously published papers and collated in collaboration with clinicians. The methods used to identify children with neurodisability in HES according to this code list are published elsewhere¹², and code lists will be made available in an online repository.

We will derive three exposure groups:

1) Children who had an indicator of neurodisability first recorded before the start of primary school (before 1^st September of reception).

2) Children who had an indicator of neurodisability first recorded during primary school (between 1^st September of reception and 31^st August of year six, inclusive).

3) Children with no indicator of neurodisability before the end of primary school.

By creating an exposure group of children who receive a neurodisability-indicative code after the start of primary school (group 2), we aim to overcome the complex thresholds of receiving a neurodisability-indicative code in HES. This allows us to capture children with a delayed recording of neurodisability in HES due to social barriers to access health services, conditions that may not be recognised or admitted to hospital until later, or those who may have another serious morbidity (i.e. neurodisability is not the first intervention factor that is recorded).

Our definition of neurodisability follows the consensus definition proposed by Morris and colleagues:¹.

“Neurodisability describes a group of congenital or acquired long-term conditions that are attributed to impairment of the brain and/or neuromuscular system and create functional limitations. A specific diagnosis may not be identified. Conditions may vary over time, occur alone or in combination, and include a broad range of severity and complexity. The impact may include difficulties with movement, cognition, hearing and vision, communication, emotion, and behaviour.”

Thus, conditions considered to be a neurodisability include neurodevelopmental disorders (e.g., learning disability, ASD, ADHD), neurological disorders (e.g., cerebral palsy, epilepsy), genetic conditions likely to affect learning (e.g., Down syndrome, sex chromosome anomalies), central nervous system anomalies of the brain or spinal cord (e.g., spinal bifida), and other conditions affecting the brain (e.g., paediatric stroke, hydrocephalus, perinatal brain injury). Specific conditions will be decided through consultation with expert clinicians and preliminary data analysis to identify conditions that are likely to require a hospital admission. Our definition of neurodisability excludes traumatic brain injuries and other acquired injuries to the head, as it is difficult to assess the severity of injury and HES does not record functional limitations that would discriminate between brain injuries with no subsequent neurodisability.

Outcome: Educational attainment. For each time point (reception, year two, year six) we will present three outcomes: a) the proportion of children who did not complete the assessment expected to be taken in that year (i.e. appear in the Census but did not have an assessment record) and of those who did have an assessment, b) the proportion of children who did reach nationally expected levels as defined in line with DfE standards; and c) the cohort-specific standardised test scores calculated using the mean and standard deviation of all pupil scores in a given academic year. These outcomes will be compared between our three exposure groups (neurodisability indicated before primary school, during primary school, or no neurodisability indicated).

Reception: Early Years Foundation Stage Profile (EYFSP)

The EYFSP is based on teacher assessments of children’s development across multiple areas of learning in the final term before year one. The EYFSP involves the teacher assessing whether children in the summer term of reception have reached a ‘Good Level of Development’ or not. To be classified as reaching GLD, teacher reports must indicate whether a child is either reaching ‘expected’ or ‘exceeding’ levels across all 17 ‘early learning goals’ across 7 ‘prime areas of learning’. These include personal, social, and emotional development, communication and language, physical development, expressive art and design, understanding the world, mathematics, and literacy¹⁵.

Year Two and Year Six: Key Stage Assessments

We will derive standardised attainment measures using recorded scores from national tests in reading and maths at the end of year two (aged 7, KS1) and at the end of year six (aged 11, KS2). At KS1, tests are marked by teachers at the child’s school. The reading assessment is scored out of 40 points and the maths assessment is scored out of 60 points. These raw scores are scaled by the DfE to account for potential small differences in difficulty from year to year. Pupils must score at least 100 scaled points on reading and maths assessments to achieve the minimum expected level¹⁶. At KS2, assessments are marked externally. Raw scores on the reading assessment have a maximum of 50 points and raw scores on the maths assessment have a maximum of 110 points¹⁵. These raw scores are also scaled by the DfE, where a scaled score of 100 points is required to meet the minimum expected level of achievement in reading and maths. We will calculate standardised scores using the mean and SD of the raw scores of all pupils in a given academic year to account for time-varying changes in the recording of educational outcomes and the Flynn effect (observed rise in standardised IQ scores over time)¹⁷.

Additional variables

Results will be presented stratified by sex at birth recorded in HES and school year (the academic year, which runs from 1^st September to 31^st August the following year). School year will be determined based on the school year indicator recorded in each January school census. We will assume that children with missing data on school year are in the expected school year given their date of birth. As we expect children with a record of neurodisability to be more likely to require additional support, we will report the proportion of children in the neurodisability exposure groups who receive different levels of SEN provision. Levels of SEN provision range from SEN support, Education, Health, and Care Plan (EHCP) to special/alternative school attendance¹¹.

In secondary analyses, we will describe educational outcomes of children with neurodisability stratified by five socioeconomic and demographic characteristics, measured at school entry:

1) Income deprivation quintile affecting children index (IDACI) at the time of reception entry. This measures the proportion of children under the age of 16 within a lower super output area who are living in an income-deprived household¹⁸.

2) Recorded entitlement to free school meals (yes/no)

3) Modal recording of ethnicity across school censuses (Asian or Chinese, White, Black, Mixed, any other ethnic group, or unclassified)

4) Government Office Region of residence associated with the pupil’s residential address (North East, North West, Yorkshire and the Humber, East Midlands, West Midlands, East of England, London, South East, South West, or missing).

5) Relative month of birth: the youngest children in each school year (born in August) have lower average attainment compared to older peers (the August-September gap)¹⁹.

If data are missing in reception, we will use the earliest complete recording available in any subsequent January school census under the assumption that these characteristics are unlikely to change during primary school.

Bias

Misclassification may occur in identifying children with neurodisability. Conditions that are largely managed in primary care may contribute to misclassification, as children with these conditions are less likely to have a hospital admission because their conditions are largely treated and managed outside of hospital settings. As children with milder forms of neurodisability may be misclassified into the unaffected group, this may reduce the relative and absolute difference between children with and without a record of neurodisability in primary school. Therefore, any estimates are likely to be conservative. Additionally, we understand that the unaffected comparison group may experience health conditions that do not fall under our neurodisability code list. We recognise that these unaccounted health conditions may have an unmeasured influence on attainment outcomes.

Descriptive analysis

Derivation of the analysis sample will be presented in a CONSORT style flowchart, detailing the number of included and excluded children at each stage. For each of the three exposure groups, the proportion of children who were not assessed at each timepoint, the proportion of children who did reach nationally expected levels, and average standardised scores for the EYFSP, KS1, and KS2 assessments will be derived. We will describe follow-up and loss to follow-up in children with and without neurodisability and describe the percentage of children lost to follow-up before the end of primary school (age 10/11) due to death (ascertained through linkage to ONS mortality data).

We will additionally describe the proportion of children in each subgroup of neurodisability who receive SEN provision and the proportion of children in alternative provision or special schools who achieve the expected level of achievement at each time point.

Statistical analysis

Standardised attainment scores for the three exposure groups will be plotted in the summer term of three school years when the exams occur (EYFSP in reception, KS1 in year two, KS2 in year six). A linear mixed-effects model will be fitted to compare longitudinal data between the two broader neurodisability groups (neurodisability code recorded before reception or between reception and year six) and the unaffected group. Fixed effects will be included for group membership and time. Interaction terms between group and time will allow us to test whether educational attainment trajectories differ over time between groups. Random effects will be included for intercept and slope to account for individual variability at the start of follow up and the rate of change over time. To investigate the effect of sociodemographic variables and other potential confounders between neurodisability and attainment outcomes, we have created a directed acyclic graph (DAG, Figure 2) using the open-source software DAGitty version 3.0.

Figure 2. A DAG representing the minimal sufficient adjustment sets containing covariates and mediators created using DAGitty version 3.0.

Note: Ethnicity, IDACI, relative month of birth, and sex are the selected covariates for estimating the total effect of neurodisability (exposure) on educational attainment (outcome). Health complications, absenteeism, and SEN are included as mediators, and will not be controlled for as they lie on the causal pathway.

Sensitivity analysis

To check the robustness of the results, we will conduct the following sensitivity analyses: Additional information on the characteristics of children in the primary school cohort will be provided to compare all children born in English NHS-funded hospitals between 1st September 2003 and 31st August 2008. We will also explore whether results differ when further stratifying by academic year of birth, to check whether there are year-to-year improvements in coding and diagnosis of neurodisability across cohorts which have substantially affected our findings.

The DAG will guide our selection of variable adjustment set to reduce the risk of mediating away any true effects and overadjustment. A mutually adjusted model will be fitted to assess the associations of the confounding variables with educational outcomes. This will adjust for the effects of other variables in the model and allow for the examination of how each variable independently contributes to the outcome while controlling for potential confounding by other variables. We will be cautious when interpreting the effects of the covariates due to the Table 2 Fallacy (when effect estimates of secondary exposures are presented in the same manner as the primary exposure estimated from the same model)²⁰.