Sociodemographic variation in the timing of recorded special educational needs provision in primary school in England amongst children with cerebral palsy: a staggered cohort study using the ECHILD database

Patient and public involvement

This study is part of a wider programme of work called Health Outcomes for People in Education (HOPE)¹⁰. At the outset of HOPE, we consulted with stakeholders, including parent/carers of children who have applied for SEN provision. These consultations revealed frustration at the variation in timing and provision of SEN services, particularly for parents and carers who disproportionately lacked resources to advocate for their child(ren). These discussions led us to design this study to investigate inequities in the timing of SEN provision. Presentation and dissemination of the study's findings will be informed by the HOPE study steering committee, which includes parents of children with disabilities and learning difficulties.

Study design and setting

Series of staggered cohorts constructed using linked de-identified child-level administrative health and educational records from England.

Data sources and linkage

This study will use the ‘Education and Child Health Insights from Linked Data’ (ECHILD) database¹⁴, which is comprised from de-identified population-based linked administrative hospital and school/social care records for children and young people in England. The health records used in this study are derived from Hospital Episode Statistics Admitted Patient Care (HES APC), which contains records of all National Health Service (NHS)-funded inpatient hospital activity in England, including records for approximately 97% of births in England¹⁵. ECHILD also contains other HES datasets containing contacts with hospital outpatient and accident emergency departments. Educational records are derived from the National Pupil Database (NPD), which contains pupil- and school-level information¹⁶. The primary source of information in NPD is the termly School Census (collected in October, January and May each academic year), which contains information on pupil enrolment, absences and exclusions and SEN provision from state-funded mainstream schools, special schools and some alternative provision. Nursery schools, which specialise in education for children aged between 3 and 5 years old, are included in the School Census if attached to a state-funded school. Other NPD modules include the early years census, exam attainment, alternative provision census, pupil referral unit census (collected until 2012/13, after which it was merged with the School Census) and children’s social care. Individual records in HES and NPD are linked by NHS England using deterministic algorithms based on name, date of birth, sex and postcode¹⁴.

Population and follow-up

Our population will consist of singleton children with hospital-record defined cerebral palsy born in an NHS hospital in England between 1 September 2003 and 31 August 2013 and recorded as entering a state-funded mainstream or special primary school or alternative provision in England in the January census of nursery year 2, reception, or year 1. The expected age at entry for each of these school years ranges from 3 to 5 years (see Figure 1). Births will be identified in HES APC using a combination of diagnostic and procedure codes, healthcare resource group codes and administrative variables (as outlined in Zylbersztejn et al.)¹⁷. We will exclude multiple births (defined in HES APC data) and non-matches in the recording of year and month of birth in HES APC and NPD to minimise the risk of including erroneous links¹⁷. Cerebral palsy will be defined by any hospital admission before entry into each staggered school cohort with an International Classification of Diseases 10^th Revision (ICD-10) code G80.*, including all subtypes, in any of the 20 diagnoses fields¹⁸.

Figure 1.

Expected age of pupils at entry into each year of primary school, by birth year and follow-up year; light shading represents staggered cohort entry years; N = nursery year 2; R = reception; Y = year; ^adefined according to the academic calendar (e.g. 2003/04 includes 1 September 2003 to 31 August 2004, inclusive); ^bPupils without a school year listed in NPD will be assigned a school year based on their age; ^cevents are measured at the January census each year; ^dPolicy change prompting SEN provision changes: School Action/School Action Plus changes to SEN support; statements of SEN transition to EHCPs

We will create three cohorts, corresponding to different points of entry into state-funded primary school: nursery year 2 (hereafter referred to as ‘nursery’; pupils expected to be age 3 years by start of school year), reception (pupils expected to be age 4 years by start of school year); and year 1 (pupils expected to be age 5 years by start of school year), Figure 2. In England, school attendance is compulsory from the age of 5 years, which means that, depending on their date of birth, children are required to enter school at some point between reception and year 1. For each pupil, start of follow up will be set as the day before their first appearance in a January census (in nursery, reception or year 1). Pupils will be followed up until the first chronological event of: first recorded SEN provision (if using survival analyses, see statistical analysis section), the day after the January census in year 6 (the last year of primary school, pupils aged 10/11 years), loss to follow up, or end of study (the day after the January census in 2019). Loss to follow up is defined by non-appearance in any NPD school census following entry into the study. Apart from death, reasons for non-appearance are not recorded in this study, but may include transfer to a non-state funded school, off-rolling (where pupils are illegally excluded from school)¹⁹, emigration and death.

Figure 2. Proposed staggered cohort study design; N =nursery year 2, R = reception, Y = school year.

Ethics and consent

Permissions to use linked, de-identified data from Hospital Episode Statistics and the National Public Database were granted by DfE (DR200604.02B; 05/08/2020) and NHS Digital (DARS-NIC-381972-Q5F0V-v0.5; 03/09/2020). 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 (20PE06/20PE16). Consent from patients is not required here.

Outcome

The outcome is recorded SEN provision defined at January census from school entry to year 6 (ages 3/4 to 10/11 years) of the early years census, School Census, alternative provision census and pupil referral unit census. The School Census is also collected in November and March; however, we use the January census only as the information recorded in this census is used to determine funding (and is therefore more likely to contain complete information). We define two categories of SEN provision: SEN support (including previous versions called School Action and School Action Plus), which is arranged by the school and includes classroom based support such as different educational materials or small group support; and Education, Health and Care Plans (EHCPs; including statements of SEN), which are 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. SEN provision will be classified as a binary (no SEN provision or any SEN provision) and categorical variable (no SEN provision, SEN support only or an EHCP) in this study.

Exposures

We will present the results by the following sociodemographic characteristics of the pupil, which have been selected due to their documented association with SEN provision from peer reviewed and policy evaluation literature^2,4,20. In this study, these characteristics are included as indicators of social strata that, through structural inequities in power, money and resources, shape access (and barriers) to timely interventions, such as SEN provision^21,22. The following characteristics are measured in NPD at the pupil’s first appearance in a January school census: region of residence (East of England, East Midlands, London, Midlands, North East, North West, South East, South West, West Midlands, Yorkshire and the Humber); income deprivation affecting children index (IDACI) quintile based on residential address; free school meal (FSM) eligibility (yes or no); parent submitted gender (female or male); and English as an additional language (yes, no or unknown). Notably, English as an additional language does not indicate a student’s proficiency in England, and we use this variable as a proxy for minority ethnic groups, where their cultural language is other than English²³. We also include child’s racial-ethnic group, which is defined as their latest non-missing recording of ethnicity across NPD censuses. We use racial-ethnic group as a proxy for lived experience of minority ethnic individuals in a White British country. As such, we aim to conduct analysis at the highest level of detail possible; however, grouping ethnicity variables may be necessary due to small numbers, restrictions in answer options provided and/or documentation error.

Covariates

We include the following covariates from HES APC birth records: sex (female or male) as assigned by physician at birth; year of birth, defined according to the academic calendar (e.g. 2003/04 includes 1 September 2003 to 31 August 2004, inclusive); and month of birth. We will define a series of comorbidities using diagnoses in hospital admission records from birth until study entry, based on the Hardelid chronic conditions list²⁴. We will classify local authority based on the home address of each pupil at first appearance in an NPD census. We will create a variable “time since school entry”, calculated as the day before a pupil’s first appearance in a January census minus their entry date into school. This is to account for differences (mainly pre-year 1) in the date a child begins school and their first possibility of having SEN assigned/recorded (i.e. the January census). We will also generate “age at first cerebral palsy recording”, calculated as the date of the pupil’s first recording of cerebral palsy in HES APC minus the date of their birth admission.

Bias

Table 1 describes our strategies to address potential sources of bias in this study.

Statistical analysis

We will firstly estimate the prevalence of cerebral palsy in our study and externally validate this using national estimates from other data sources¹³. We will plot the prevalence by year of birth and by age of first recording in HES APC records to assess potential biases in our study introduced by using hospital admission records to define cerebral palsy. We will then describe the characteristics of our study population, including missing data, in numbers and percentages. The association between exposures/covariates and the probability of missingness in at least one study variable will be explored using univariable logistic regression. These findings will be used to decide how to treat missing data in this study.

We will describe transitions between no SEN provision, SEN support and EHCPs for pupils as they go through each year of school using a transition matrix and alluvial plots. We will use this information to decide on the precise number of transitions to be modelled in this study. Due to the high support needs of the population, we expect that there will be three transitions at most (study entry to SEN support, study entry to EHCP and SEN support to EHCP, Figure 3). In all cases, we will firstly model time-to-any SEN provision (either SEN support or EHCP) using single-event survival analysis (Figure 3A), which we describe below. These methods will be extended to competing risk survival analysis and multi-state modelling (Figure 3B and 3C, respectively) based on the results of this exploratory step.

Figure 3A–C. Examples of transitions to model in this study.

We will use Cox proportional hazards regression models separately for each exposure of interest (region, IDACI quintile, FSM eligibility, gender and ethnic group) to quantify differences in time-to-first SEN provision between groups. We will consider alternative statistical models if the proportional hazards assumption (tested using Schoenfeld residuals)²⁵ is found to be violated. We will include co-morbidities, sex, year of birth, month of birth and time since school entry to increase precision in estimates, given the documented association between these covariates and the allocation of SEN provision, using stratified Cox models if proportionality for these variables is violated^2,4,20. Sex will not be included in the model with gender as the exposure given the high collinearity between these two variables. We will include age at first cerebral palsy recording and subtypes of cerebral palsy (denoted by asterisks in Table 2) as additional covariates to capture severity of cerebral palsy. We will also examine whether there is evidence of effect modification by including interaction terms between cerebral palsy subtypes and each exposure. To account for geographical clustering of SEN provision, analyses will be clustered by local authority of the child’s address at entry into school using the marginal (robust variance estimator) approach.

We will stratify analyses by cohort and then examine whether estimates can be averaged using a meta-analytical approach, including visualisation via forest plots. The results will be reported in terms of adjusted hazard ratios (aHRs) for each year after school entry, representing the cumulative “relative risk” of recorded SEN provision for each value of each exposure compared with its baseline. For example, we will present the aHR for children eligible for FSMs compared with those not eligible for FSMs after 1 to 7 years. We will graphically present unadjusted cumulative incidence curves of first recorded SEN provision by region of residence, IDACI quintile, FSM eligibility, ethnic group and gender to depict the cumulative probability of an initial recording of any SEN provision. We will also present cumulative incidence curves adjusted for the covariates outlined above using the method outlined by Hernán²⁶.

Statistical analyses will be conducted using Stata v17 in the Office for National Statistics Secure Research Service (ONS SRS). R is a suggested opensource alternative. The code and algorithms used in this study will be made publicly available on GitHub when the full study is published.

Sensitivity analyses

We will conduct two sensitivity analyses to check robustness in our case definition of cerebral palsy using hospital admission records. Whilst cerebral palsy is generally detected by age two/three years²⁷, diagnosis occurs in specialist paediatric settings and the first recording of cerebral palsy in hospital admission records may happen at a later age. In our first sensitivity analyses, we will alter our population to include children who have cerebral palsy first recorded in HES APC records any time before January census in year 6 (ages 10/11 years). To reduce the possibility of including cases where SEN provision may lead to cerebral palsy diagnosis, we will exclude children who have a record of SEN provision before their first recorded cerebral palsy in HES APC. Secondly, we will repeat the main analyses on a population of children defined using a wider definition of cerebral palsies (as has been applied elsewhere)¹⁸. We will include children with hospital-record defined cerebral palsy (G80.*), hemiplegia (G81.*), paraplegia and tetraplegia (G82.0-Q82.5), diplegia (G83.0) and/or monoplegia (G83.1-Q83.3), see Table 2¹⁸. In a third sensitivity analysis, we will examine the extent to which SEN provision may have been assigned prior to school start (as defined in our study). Of the children in the cohort with records in the early year's census or at age 2 in the School Census, we will describe the number and proportion with a recording of SEN support or EHCP before the school start date assigned in our study. We do not use these entry points into education in the main analysis because of variation in the uptake, and data collection, of early years’ placements²⁸.