Association of Race/Ethnicity and Social Disadvantage With Autism Prevalence in 7 Million School Children in England

Key Points Question What is the prevalence of autism spectrum disorder (ASD) in the total English state school population, and what are the social determinants associated with ASD status? Findings In this ASD prevalence cohort study of 7 047 238 pupils, national English prevalence was 1.76%, with marked differences according to racial/ethnic group. The highest prevalence was found in Black pupils (2.11%) and the lowest in Roma/Irish Travelers (0.85%), with important variability across geographic areas. Meaning These results show differences in ASD prevalence estimates across racial/ethnic minority groups in England, which could be attributable to diagnostic biases, possible differences in detection and referral, or differential phenotypic prevalence for racial/ethnic minority groups.

A utism spectrum disorder (ASD) is a cluster of neurodevelopmental conditions characterized by persistent difficulties in social communication and interaction, and restricted, repetitive patterns of behavior, interests, or activities across multiple contexts. [1][2][3] The global prevalence of ASD is between 1% and 2% of the population, with a reported male-to-female ratio (MFR) of 3:1 or 2:1. 1,2,4,5 Discussions about a possible increase in ASD prevalence have pointed to its shifting diagnostic features, 6,7 with little research in Black, Asian, and racial/ethnic minority groups. [8][9][10] Accurate estimates of ASD prevalence are vital to planning diagnostic, educational, health, and social care services and identifying possible access barriers to diagnostic pathways and services and inequalities based on social determinants of health, such as socioeconomic disadvantage, race/ethnicity, and or sex.
In this ASD prevalence study, which to our knowledge is the largest to date, we address the aforementioned gaps using the Spring School Census 2017 from the National Pupil Database (NPD) in England, which is an administrative data source. We estimated ASD prevalence in English schools to explore the association between ASD and sociodemographic and socioeconomic factors or characteristics. [11][12][13] We stratified the prevalence estimates by sex; race/ethnicity; Special Educational Needs and Disability (SEND) status, which is an assessment determining whether a pupil requires special educational provision 14 ; and Local Authority Districts (henceforth districts) and then calculated MFRs nationally and by locality. We then sought to identify possible access barriers to SEND services in ASD by stratifying our outcomes by race/ethnicity, the Free School Meals (FSM) program (used as a proxy for socioeconomic disadvantage), and first language spoken (as a measure of social inclusion) in pupils aged 5 to 19 years. We hypothesized that social determinants of health and experiences of multiple socioeconomic disadvantage and identity-or culture-based forms of alienation would influence the likelihood that people in disadvantaged and racial/ethnic minority groups will access ASD services.

The NPD and Data Access
The NPD is a total school population registry collected in England by the Department for Education that maintains counts of all pupils aged 2 to 21 years under State education provision. One of its components is the Pupil Level Annual Schools Census, which is conducted every term. For the purpose of this project, we used the Spring School Census 2017, which was collected January 17, 2017. Data access was granted by the Department for Education in March 2018 and approved by the Ethics Committee from the Department of Psychology, University of Cambridge, Cambridge, United Kingdom. Informed consent was not needed for this publicly available data source. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. Details of the NPD are provided in the eMethods in the Supplement.

Special Educational Needs and Disabilities in England
The care that children and young people receive in English schools varies according to SEND status. Two levels of provision are in place in England: SEND support, which is a schoolspecific learning program given to a pupil; and Education, Health and Care Plans (EHCPs; previously known as statements), which were introduced as part of the Children and Families Act 2014. 14 Currently, the School Census includes 14 support/EHCP categories: (1) specific learning difficulty; (2) moderate learning difficulty; (3) severe learning difficulty; (4) profound and multiple learning difficulties; (5) speech, language, and communication needs; (6) hearing impairment; (7) visual impairment; (8) multisensory impairment; (9) physical disability; (10) ASD; (11) other difficulty/disability; (12) social, emotional, and mental health; (13) SEND support but no specialist assessment; and (14) unclassified. The EHCP needs assessments are usually performed by local authorities with assistance from the National Health Service, Child and Adolescent Mental Health Services, which manages the ASD diagnostic pathway across England. The NPD SEND registry includes as many as 2 SEND categories per pupil.

Operationalizing ASD Status From Administrative Educational Data
To operationalize ASD by using SEND registry data contained in the Spring School Census, we coded ASD if it was a primary or secondary SEND and included (1) an EHCP, which contains F84.0 to F84.9 ASD diagnosis codes from the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, 15 and (2) school support given to pupils by educators. Autism spectrum disorder SEND support therefore includes pupils who require access to additional ASD-specific support beyond the school's differentiated curriculum, which also encompasses pupils awaiting their EHCP. 15 We calculated the prevalence for these 2 different SEND provision categories separately and then created a composite third variable that was coded by merging pupils with an ASD EHCP and those pupils receiving SEND support to calculate the standardized prevalence of ASD status in the English educational system. This variable is the main outcome for our prevalence estimates and, for the purpose of this study, defines pupils as having documented ASD status in English schools. We assume that this composite variable of ASD status is the most representative estimate of formally recognized ASD prevalence in the English educational system.

Sociodemographic Variables and Regional Analysis Units
Individuals were categorized into 3 age groups (5-9, 10-14, and 15-19 years). Sex was binarily coded. The NPD has 8 selfreported major categories of race/ethnicity, including (1) any other racial/ethnic group, (2) Asian, (3) Black, (4) Chinese, (5) mixed, (6) unclassified, (7) White, and (8) Roma/Irish Traveler, all of which are based on the English and Welsh census categories. First language spoken was coded as English or other. The NPD defines this as the language to which the child was exposed during early development and continues to use in the home or in the community regardless of subsequent English proficiency. We used dichotomized lifetime claimed eligibility to the FSM program as a proxy of socioeconomic disadvantage. 16 England has 326 districts, which comprise a level subnational division of England used for the purposes of local government. We included all 326 districts for our analysis. We also created a novel outcome ratio that we have termed the standardized statement-to-support ratio (SSR). The SSR is used to describe the proportion of EHCPs vs support as a standardized rank to assess how districts perform in processing EHCPs. The higher the SSR, the higher the proportion of ASD EHCPs compared with ASD SEND support. Further details are included in the eMethods in the Supplement.

Statistical Analyses, Multiple Imputation for Missing Data, and Sensitivity Analyses
Data were analyzed from August 2, 2018, to January 28, 2020. Raw prevalence estimates for ASD were directly standardized and stratified by age group, sex, and districts using the English 2011 census projections for 2017 as a standard population to calculate national prevalence by race/ethnicity and sex and prevalence across 326 districts by sex, using the World Health Organization standard method. 17-19 Adjusted relative risk (referred to henceforth as adjusted prevalence ratio [aPR]) estimates were obtained using a Poisson regression with robust error variance to evaluate access to services using the following outcome variables: (1) pupils with ASD, (2) pupils with ASD who had SEND support, (3) pupils with ASD who had an EHCP, and (4) pupils with ASD who had a second SEND. In each outcome model, we used the same independent variables of age, sex, race/ethnicity, first language spoken, and socioeconomic disadvantage (through the FSM program) and included them in the same adjusted model comparing all levels against each other and reporting for missing data. After this initial approach and identification of missingness, we optimized model fit while accounting for missing data by using multiple imputation by chained equations in our Poisson regression with robust error variance, which uses a separate conditional distribution for each imputed variable. 20 To avoid bias in the imputation model, we included all variables listed in the main analyses. 20 To control for multiple comparisons, we used a significance level of 2-sided P < .001 for all reported outcomes. We then conducted a 1-way sensitivity analysis com-paring the multiple imputation by chained equations ASD model with the Poisson regression with a robust error variance model and a log-binomial regression. The comparisons were performed against a full data model and a complete analysis model for the variables with missing data, checking for model fit and effect on statistical significance. 20 For the mediation analysis, which was used to evaluate the role that race/ethnicity and socioeconomic disadvantage plays in ASD status, we included a dichotomized ASD status as the outcome, a dichotomized claimed eligibility to the FSM program and a dichotomized variable of language spoken as mediators, and race/ethnicity as our independent variable, coded as 7 dummy variables using White pupils as our comparator group. 21 We applied a planned regression approach to evaluate associations among the dependent, independent, and mediating variables. 21 We included age group and sex as covariates of interest. We then used a weighted least-square means and variance-adjusted model design used for categorical data. In this analysis, SEs for the standardized path coefficients are not computed (eFigure 6 in the Supplement). Consequently, we only report the standardized estimates and the respective bootstrapped 95% CIs with P values.

Descriptive Statistics
Our final sample included 7 047 238 pupils, of whom 50.99% were male and 49.01% were female (mean [SD] age, 10.18 [3.47] years). A total of 119 821 pupils with ASD were identified in our sample, of whom 21 660 (18.08%) were also classified as having a learning difficulty. Among pupils with ASD without a learning difficulty, 65.41% were recorded with a second SEND. With regard to socioeconomic disadvantage, 25.80% of all English school pupils (n = 1 818 195) had ever claimed eligibility for the FSM program. When comparing socioeconomic disadvantages in the school population, we found that pupils with ASD (35.23%), pupils with ASD and a learning difficulty (37.93%), and pupils with other types of SEND (44.61%) all had higher proportions of socioeconomic disadvantage than pupils with no SEND (24.47%). A total of 18.51% of pupils spoke a language other than English first, with other ethnic group (80.47%), Asian (73.77%), Black (43.68%), Chinese (75.89%), and Roma/Irish Traveler (44.65%) having the largest proportion of pupils in this category. All descriptive statistics are presented in eTable 1 in the Supplement.

National Prevalence by Race/Ethnicity, Sex, EHCP, and Support Status
The age-and sex-standardized prevalence of ASD in our national sample was 1.76% (95% CI, 1.75%-1.77%) ( Figure 1 shows heat maps for England and London. For all local prevalence estimates, see eTables 4 and 5 in the Supplement. Figure 2 illustrates the MFR distribution of ASD in England and London specifically.

Poisson Regression Analysis Results
Autism spectrum disorder was almost 4 and a half times more likely in male pupils than female pupils (aPR, 4 . When investigating the relationship between receiving ASD SEND support and race/ ethnicity, non-White pupils were less likely to receive support when compared with White pupils. In our 1-way sensitivity analysis comparing both negative missingness of FSM (using liberal assumptions) and a listwise deletion model of FSM (using conservative assumptions), we observed outcomes similar to those from our multiple imputation by chained equation models (eTables 2 and 3 in the Supplement), increasing confidence in our imputation method. In the crude model, there was no loss due to missing data. The complete case analysis had complete data for 93.81% (n = 6 611 261).

Mediation Analysis
Our mediation analysis compared the association of social disadvantage (using the FSM program as a proxy) with ASD, with White pupils as the reference group. Our results showed an in-direct effect of race/ethnicity with ASD status through FSM. Black pupils and those of mixed race/ethnicity showed notably higher effect sizes (standardized mediation coefficient [SMC], 0.018 [P < .001] with 12.41% of effects mediated and 0.010 [P < .001] with 12.05% of effects mediated, respectively). In other words, 12.41% and 12.05% of the increased ASD prevalence in these groups can be explained by the relative social disadvantage associated with their race/ethnicity when compared with White pupils. In the case of first language spoken as a mediator of ASD status, we obtained inconsistent mediation (SMC range for race/ethnicity categories, −0.007 to 0.001; SMC for first language other than English, −0.132). For all details, see Table 3; for models used, see eFigures 7 and 8 in the Supplement.

Discussion
This study reports formally recorded ASD prevalence in English pupils aged 5 to 19 years using a total school population sample, providing evidence of consistent prevalence differences among different racial/ethnic groups. The reported English standardized prevalence of ASD in our sample was 1.76%

Figure 1. Heat Maps of Autism Spectrum Disorder (ASD) Prevalence by Local Authority District (LAD) in England
Prevalence of ASD, %  Data point sizes represent sample sizes. LAD indicates Local Authority District.

Research Original Investigation
Association of Race/Ethnicity and Social Disadvantage With Autism Prevalence in 7 Million School Children in England issue further, we found that racial/ethnic differences in ASD were mediated through socioeconomic disadvantage. The greatest effect was found among Black pupils (standardized mediation coefficient, 0.018; P < .001), with 12.41% of the increased prevalence of ASD among Black pupils being explained by social disadvantage. The interaction between ASD   status and social disadvantage can be 2-fold: (1) children from socially disadvantaged families may be at higher odds of developing ASD, or (2) having a child with ASD can increase the risk of a family experiencing poverty. The indirect effect of race/ ethnicity on ASD status through first language spoken was inconsistent in our analysis. Although the estimates obtained for race/ethnicity were consistently larger than zero (SMC range, −0.004 to 0.525), this is opposed by the estimate obtained when English was not the first language spoken (SMC, −0.130). These statistics support our hypothesis that although some racial/ethnic groups are more likely to receive an ASD diagnosis or support, this might be counterbalanced by barriers in accessing SEND support in families who speak other languages at home, such as Asian families (73.77% of whom speak other first languages) and Black families (43.68% of whom speak other first languages). This is an important issue considering the social dimensions of the core features of ASD and how they are recognized in bilingual pupils. Other issues shown in our findings are sex differences and the large regional variation in the MFR observed across English districts, with Burnley (12.87: 1), Three Rivers (10.83:1), and Fareham (10.32:1) having the highest MFRs in England. The same can be said to the SSR that we report, with Newham (0.42), Rushcliffe (0.52), and Forest Heath (0.54) having the lowest ratios. These rates indicate approximately 1 pupil with EHCP per 2 pupils with support in these districts. When comparing these rates with the national mean SSR for pupils with ASD, it becomes evident that EHCPs are underprovided in some districts (eTable 4 in the Supplement).

Limitations
This study has a number of limitations. First, the proportion of pupils aged 3 to 18 years attending state-funded schools in England is 93%, with 7% enrolled in independent schools or in alternative arrangements such as home schooling. 22 Another limitation is that pupils from certain racial/ethnic backgrounds (such as Roma/Irish Traveler) tend to leave school prematurely, with some evidence demonstrating that approximately half leave school by 16 years of age. 23 Considering that we used World Health Organization-validated direct standardization methods with British census data, this issue has been addressed to the best of our ability given the resources available. A different hypothesis regarding the low ASD prevalence figures found among the Roma/Irish Traveler community might involve questioning whether diagnostic recognition of ASD symptoms (such as communication and repetitive behaviors) in the English educational system adversely affects this subgroup, and whether it is culturally appropiate. 23-25 Although complex challenges exist in construing prevalence estimates with administrative data, we assessed health service and educational system use, and our findings reflect actual numbers of individuals receiving ser vices in schools by loc ality. 2 6 Furthermore, the NPD does not account for pupils with subclinical ASD or those who do not meet service thresholds to receive support or EHCPs at school. 15 Further research should explore the reasons for this, but more importantly should focus on the dynamic between an ASD diagnosis and the school system.

Conclusions
Our results highlighting higher ASD prevalence in pupils of racial/ethnic minority groups are in line with a broader literature of large population-based studies using health and education registry data that have described an increased ASD prevalence in pupils of racial/ethnic minority and immigrant backgrounds both in European and Nordic countries. 9,27 When comparing our findings with US evidence, they are in line with the findings of a population study by Becerra et al that found higher ASD rates in immigrants with foreign-born mothers, 28 whereas they contrast with the findings of Durkin et al that the Black community and people with a lower socioeconomic status have notably lower ASD rates. 29 Our findings also challenge evidence from the Centers for Disease Control and Prevention-Autism and Developmental Disabilities Monitoring (ADDM) Network in the US that showed wide variation between sites in 2016 30 ; for example, Colorado reported a prevalence of 1.31% (95% CI, 1.21%-1.43%), whereas the prevalence in New Jersey was 3.14% (95% CI, 2.95%-3.33%), and Colorado reported no differences according to race/ethnicity. Higher prevalence in racial/ethnic minority groups has also been reported for other neurodevelopmental conditions, such as schizophrenia, with racial/ethnic minority and immigration status found to be associated with higher risk of developing psychosis. 31 The large genetic overlap between these 2 neurodevelopmental conditions should be taken into consideration when investigating possible links between immigration and racial/ethnic minority and ASD status. [32][33][34][35] We also believe that it is crucial to assess the intersection of ASD policy and public health policy as well as their assumptions about racial/ethnic minority groups, immigrants, and other typically underserved populations. 36 Our results challenge us to better understand which pupils receive an ASD diagnosis, when they receive it, what support is provided to them, and, most importantly, to what degree social determinants of health, immigration, and race/ethnicity affect ASD status. 36