Educational and employment outcomes associated with childhood traumatic brain injury in Scotland: A population-based record-linkage cohort study

In the United Kingdom, 280 children per 100,000 are admitted to hospital annually for traumatic brain injury (TBI) [1]. TBI is a leading cause of death and disability in childhood and increases the risk of psychiatric disorders [2–7], suicide [8], criminal behaviour, and incarceration in later life [9,10]. Childhood is an important period for brain development and learning and, while 80{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c} to 90{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c} of childhood TBI cases are mild [11], moderate and severe injuries can result in physical, psychosocial, cognitive, emotional, and behavioural deficits that impair normal development and the ability to participate in home, school, and community activities [12–15].

A meta-analysis of 28 studies with follow-up between 6 months and 2 years reported no significant long-term effects on cognitive outcomes following mild TBI, but persistent deficits in intellectual ability, memory, and attention following moderate and severe TBI, as well as slower rates of cognitive development and deterioration in executive function [16]. The deficits resulting from childhood TBI can require lifelong support [11,16–19], reducing the quality of life of both the affected individual and their caregivers [20–22]. The aim of this study was to use record linkage of Scottish administrative data to investigate a range of educational outcomes following hospitalised TBI in childhood. We hypothesise that children who have been hospitalised for TBI will have poorer educational outcomes compared to their peers.

This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 STROBE Checklist). While we did not publish an analysis plan, our analyses were planned before the research team accessed any data. Data were extracted in June 2015 as part of a wider linkage project [23] and this specific study commenced in February 2021 utilising the same data. A Scotland-wide cohort was constructed by linking, at an individual level, 2 health sector databases held by Public Health Scotland to 3 education sector databases held by the Scottish Exchange of Educational Data (ScotXed), using a two-step process involving probabilistic and deterministic matching. This linkage methodology has been described in detail elsewhere [23] and shown to be 99{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c} accurate for singleton births [24]. The cohort comprised singleton children who were born in Scotland and attended a Scottish school, aged 4 to 18 years, at some point between 2009 and 2013. Inclusion was restricted to singleton children because of difficulties linking same sex multiple births to the correct education records.

The Scottish Morbidity Record 02 (SMR02) collects maternal, obstetric, and neonatal data on all pregnancy-related admissions. SMR01 records acute hospital admissions, including dates of admission and discharge and International Classification of Diseases (ICD-10) diagnostic codes. The annual School Pupil Census collects data at the start of each school year from all local authority maintained primary, secondary, and special schools in Scotland. The data collected include demographic information, stage of education, and record of special educational need (SEN). Attendance data are recorded biennially on school absences (due to illness or prior arrangement) and school exclusions. They are collected prospectively then appended to the School Pupil Census at the end of the school year. The Scottish Qualifications Authority (SQA) collects grades attained in national examinations undertaken over the last 3 years of school attendance. The School Leaver Destination Survey is self-completed by 99{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c} of school leavers 6 months after leaving school [25]. Destination was classified as being in paid or voluntary employment, training, further or higher education, or being unemployed.

In the Scottish education system, children attend 7 years of primary school education (P1-P7) from ages 4/5 through 10/11 followed by 4 or 5 mandatory years of secondary education (S1-S4/S5) from ages 11/12 until their 16th birthday. Children can leave school when they turn 16 but many go on to complete the last 2 years of secondary school (S5 and S6). The outcomes analysed were: record of all-cause SEN, cause-specific SEN, school absences, school exclusions, low attainment in examinations, and unemployment. There is a statutory requirement for schools to record SEN, defined as a requirement for additional support, beyond that normally provided to children of the same age. All-cause SEN was defined as SEN due to learning disability; learning difficulty; sensory impairment; physical or motor impairment; communication problems; autistic spectrum disorder; physical health problems; mental health problems; and/or social, emotional, or behavioural difficulties. A child can have more than 1 cause recorded. Overall level of attainment over the last 3 years of secondary school (S4-S6) was derived from the total number of examination grades attained at each level of the Scottish Credit and Qualifications Framework (SCQF) [26]. Different levels of exam and grade combination correspond to different SCQF levels. The overall scoring system takes account of external examinations at different levels that can either be taken chronologically in the same children or at the same age by 2 different children. These exams are normally taken between 14 and 18 years of age. For this study, overall attainment was dichotomised as low attainment (≥1 awards corresponding to SCQF level 2, ≥5 awards corresponding to SCQF level 3, ≥2 but ≤7 awards corresponding to SCQF level 4, or >0 but ≤4 awards corresponding to SCQF level 5) or high attainment (>7 awards corresponding to SCQF level 4, ≥5 awards corresponding to SCQF level 5, ≥3 awards corresponding to SCQF level 6, or ≥1 award corresponding to SCQF level 7). Destination 6 months after leaving school was collapsed into a dichotomous variable: in education/employment/training or unemployed. The attainment and unemployment analyses were restricted to school leavers. Attainment analyses were additionally restricted to children who had either sat, or were of an age eligible to sit, exams across all 3 school stages S4, S5, and S6 during the study period. These methods, which ensured that potential follow-up time to attain academic awards was comparable across different children, are described elsewhere [23]. The number of records and pupils excluded at each stage of data cleaning and used to investigate each outcome are detailed in S1 Fig.

For each outcome, 3 models were constructed: univariate, partially adjusted (adjusted for sociodemographic covariates), and fully adjusted (adjusted for sociodemographic and maternity covariates). We adjusted for maternity covariates because previous studies have demonstrated an association between maternity factors and long-term cognitive sequelae including SEN [27–29]. Pupil sociodemographic covariates comprised: age (derived from date of birth and date of School Pupil Census), sex, ethnic group, and current area deprivation quintile category. Deprivation quintile category was based on the Scottish Index of Multiple Deprivation (SIMD), derived from the postcode of residence recorded on the School Pupil Census. SIMD is a composite measure covering 7 domains: income, employment, education, health, access to services, crime, and housing [30]. Maternity covariates comprised: maternal age, marital status, smoking status, parity, mode of delivery, gestational age at delivery, sex- and gestation-specific birth weight centile, and 5-min Apgar score.

Children had serial measurements of SEN, absenteeism, and exclusion over more than 1 School Pupil Census, violating the assumption of independent observations. Therefore, these outcomes were analysed using generalised estimating equation (GEE) models to take account of correlations between measures recorded for the same pupil across different census years [31]. All-cause SEN and cause-specific SEN were binary outcomes investigated using GEE models with a binomial distribution and logit link function. School absences and exclusions were count outcomes investigated using GEE models with a negative binomial distribution and log-link function with the number of possible attendances included as an offset to adjust for different exposure times. Correlation structures were chosen based on the quasi-likelihood under the independence model criterion (QIC) values [32]. The correlation structure with the lowest QIC was deemed the most appropriate.

Children previously hospitalised for TBI were significantly more likely to be white and male and were born in more deprived areas (Table 1). Their mothers were younger, less likely to be married, and more likely to have smoked during the current pregnancy and be parous (given birth previously). They were less likely to have had an operative delivery but were more likely to have been born preterm and had a lower birth weight for gestational age and sex. Demographic variables that contained the largest percentages of missing data included marital status (6.8{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c}) and maternal smoking (11.4{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c}). However, missing values for these variables were analysed as “unknown” categories and included in all analyses to minimise loss of records. The remaining variables were ordinal categories therefore including missing data as unknown categories did not make sense. However, the percentage of missing data within these variables was less than or equal to 0.2{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c}, except for parity (0.5{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c}), Apgar score (1.0{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c}), and ethnicity (1.4{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c}). Therefore, we did not deem multiple imputation to be necessary and instead used complete case analyses.

An independent correlation structure was deemed most appropriate for the various GEE models. On univariate analysis, children with a history of hospitalised TBI were significantly more likely to have a record of SEN than their peers (OR 1.62, CI 1.50 to 1.75, p < 0.001). The effect size was attenuated, but remained significant, after adjusting for sociodemographic (OR 1.35, CI 1.25 to 1.47, p < 0.001) and maternity (OR 1.28, CI 1.18 to 1.39, p < 0.001) factors. Previous TBI hospitalisation was significantly associated with all causes of SEN except for learning difficulty, autism spectrum disorder, and mental health problems, and the association with SEN attributable to communication problems was of borderline significance. The effect sizes were largest for physical health problems and physical motor impairment (Table 2).

It has previously been reported that 39{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c} of children were enrolled in special education 1 year following hospitalisation for TBI [33] and that 50{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c} to 79{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c} had either failed a grade or been placed in special education 2 or more years after TBI [34,35]. Previous studies have suggested that the risk of SEN was greater among children who had severe TBI [33,35–40], which is consistent with our observation of stronger associations with adverse outcomes when hospital admissions for concussion were excluded. Our findings help to address the paucity of evidence on the association between childhood TBI and absenteeism and exclusion. Behavioural problems are well-recognised sequelae of childhood TBI [37,41–44], and a single UK study of 525 children admitted to hospital for TBI reported a dose-relationship between injury severity and parent-reported school absences and exclusions over 3-year follow-up [38]. Sociodemographic and maternity factors play an important role in predicting behavioural problems independent of childhood TBI. Indeed, previous studies have reported associations between maternity factors and educational outcomes including SEN [27–29]. Our study demonstrated that the observed associations between hospitalisation for TBI and later educational outcomes remained even after adjusting for maternity and sociodemographic factors.

Current evidence does not suggest that mild TBI in childhood results in significant or lasting reductions in school performance, but children with a moderate or severe TBI achieve significantly lower scores in the reading, mathematical, and language domains of achievement tests [35]. There is consistent evidence that deficits in academic achievement persist beyond the first-year post injury [43,45–48], and that, 2 years after severe TBI, children are 18-fold more likely to have an unfavourable academic performance than their peers [35].

Severe TBI in childhood is also associated with poorer academic performance in adulthood, compared with moderate or mild TBI or orthopaedic injuries [19,49,50]. Studies on the association between childhood TBI and employment are few in number and have reported inconsistent findings [19,49–51], but 2 studies have demonstrated a significant association between history of childhood TBI and receipt of government benefits [49,50].

Most previous studies of childhood TBI and educational outcomes have been small-scale. Indeed, only 2 previous studies have used administrative data. Gabbe and colleagues used the Welsh Electronic Cohort for Children to study 90,611 children born between 1998 and 2001 [52]. They reported that the 290 children previously hospitalised for TBI performed less well in their Key Stage 1 National Curriculum assessment, administered at 5 to 7 years of age, than their peers. Sariaslan and colleagues used a Swedish cohort of children born between 1973 and 1985, to compare 104,290 adults who sustained a TBI prior to 25 years of age with their 68,268 unaffected siblings [50]. Previous TBI was associated with not achieving secondary school qualifications, and receipt of means-tested welfare benefits and disability pension, as well as psychiatric disorders and premature mortality [50]. Jackson and colleagues analysed data from the USA National Survey of Children’s Health, on 15,402 preschool children (3 to 5 years) and demonstrated reduced school readiness among the 252 with a history of TBI [53].

The strengths of our study include its large, unselected sample size, recruitment of children from both mainstream and special schools, and the ability to investigate 6 educational outcomes in the same cohort. We were able to adjust for a range of sociodemographic and maternity confounders but, as with all observational studies, residual confounding is possible. Since both exposure and outcome data were collected routinely by the health and education sectors, recall or reporting bias was avoided. Association does not necessarily infer causality. However, the likelihood of the associations being causal was strengthened by the increased effect estimates observed when less severe traumatic brain injuries were excluded. Since SEN was recorded at school, for TBIs occurring before school age it was impossible to determine whether the underlying reason for subsequent SEN was already present before the TBI. Additionally, it is conceivable that, in some cases where the SEN was genuinely diagnosed after the TBI, the underlying difficulties necessitating SEN may have been present for a period beforehand and possibly before the TBI. Therefore, for the specific outcome of SEN reverse causation may be a limitation. It is also possible that not all children are formally identified as having SEN post TBI despite experiencing difficulties. Children born in private hospitals, privately educated, or home-schooled could not be included in the study. However, only 4{e538325c9cf657983df5f7d849dafd1e35f75768f2b9bd53b354eb0ae408bb3c} of Scottish children attend private schools [54], and home-schooling and private maternity care are very uncommon in Scotland. Because of the need to link birth and education records, we could not include children born outside of Scotland or who emigrated from Scotland before starting school. Finally, the large sample size contributed to small p-values (<0.001). Some of the findings, therefore, while highly statistically significant, may be less clinically significant since they represented modest differences in real terms, and this should be considered during interpretation.

Previous studies have demonstrated heterogeneity in outcome and recovery trajectory for childhood sufferers of TBI [16,17,55]. Our analyses could not investigate age at injury or time since injury which are further limitations and areas for potential future analyses. Future studies utilising population-level data sources present an opportunity for further investigation into the relationships between childhood TBI and educational outcomes accounting for additional factors not included in this study.