The ADHD deficit in school performance across sex and parental education: A prospective sibling‐comparison register study of 344,152 Norwegian adolescents

Abstract Background Attention‐Deficit Hyperactivity Disorder (ADHD) is associated with impaired school performance, but the impact of ADHD may vary across sex, family background, and school subjects. By using prospective population‐wide register data, we describe impairment in academic performance related to ADHD across different school subjects and investigate how this impairment differ across sex and parental education. Methods We examined grades and Grade Point Averages (GPA) at age ∼16 among 344,152 Norwegian children born between 1997 and 2002. We linked grades with diagnoses from publicly funded general practitioners and with demographic information. Associations between ADHD diagnosed between age 10 and 16 and school performance were estimated with linear models, including sibling‐models which control for unobserved variables shared within families. Results Children with ADHD (4.0%) had −1.11 standard deviations lower GPAs compared to children without ADHD. This difference remained substantial after adjusting for demographic factors (−0.87), comorbid mental disorders (−0.82), early school performance (−0.54), and when comparing full siblings (−0.60). The relative ADHD deficit was 22% larger for girls than for boys and 39% larger for children with highly educated parents than for children of parents without completed high school, but the absolute deficit was smaller. Conclusion The ADHD deficit in school performance was large and not easily attributable to other factors. Because the ADHD deficit was large in all school subjects, interventions should ideally address factors that affect school performance broadly, although targeting theoretical subjects specifically may be most effective given limited resources.

Attention-Deficit Hyperactivity Disorder is substantially more prevalent among boys than girls and among children of parents with less education or less income (Kinge et al., 2021;Larsson et al., 2014;Russell et al., 2016;Torvik et al., 2020). School performance also varies with sex and social background, like ADHD does. Girls generally outperform boys (OECD, 2015;Voyer & Voyer, 2014), and children with highly educated parents or from high income households generally outperform less advantaged children (OECD, 2016;Sirin, 2005;Statistics Norway, 2020).
Different school performance in groups with different prevalence of ADHD raises the question of how the impact of ADHD may differ across groups. To the best of our knowledge, no study has investigated how the impact of ADHD varies by social background, and the studies on sex differences have been small and underpowered (e.g., DuPaul et al., 2006;Gershon, 2002). Furthermore, although ADHD has been linked with both reading disabilities and arithmetic difficulties (Taanila et al., 2012), little is known about to what extent ADHD leads to general deficits in school performance versus deficits specific to some school subjects. By investigating relative performance in different school subjects, we can point out where children are most affected by ADHD and hence where to target potential interventions.
The association between ADHD and school performance could be confounded by several factors that influence both (Polderman et al., 2010). First, family characteristics such as parental education are robustly associated with both school performance and ADHD (Kinge et al., 2021;Sirin, 2005). We account for this by comparing full siblings, which controls for unobserved variables shared within families, such as social, geographic, and parental characteristics, as well as half of the genetic risk (Taylor, 2021). Second, children with ADHD often have comorbid mental disorders, which are also associated with lower academic performance (Fröjd et al., 2008;Lawrence et al., 2019). By adjusting for comorbid mental disorders, we test whether reduced school performance is specific to ADHD.
Third, children may be diagnosed with ADHD because they perform poorly in school, rather than the other way around. We therefore control for early school performance to see how ADHD is associated with progress.
To summarize, the aims of this study are to (1) accurately describe the ADHD deficit in school performance, (2) describe how the ADHD deficit in school performance varies across sex and parental education, and (3) compare the ADHD deficit in different school subjects.
We do this using register data on all Norwegian children with appropriate statistical controls, including sibling-comparisons, which should result in precise and representative analyses.

Sample
This study comprises all Norwegian inhabitants born between 1997 and 2002 that were alive and living in Norway between age 10 and 16 (N = 359,492). The Norwegian national population register includes personal identification numbers of all inhabitants, which allowed us to link data from separate sources. We linked data from publicly funded health services with data on school performance, parental education, and parentage from Statistics Norway. Norwegian school is compulsory up to 10th grade (age 16), meaning the resulting grades and GPAs are largely representative of the population. We identified 344,523 (95.7%) individuals who had GPA registered within 1 year of normed time (age 16) which we included in the main analyses (see Figure S1). Only an additional 14,969 (4.3%) individuals did not have a GPA registered within normed time, which we included in sensitivity analyses. Relatedness data were used to identify 145,051 full siblings nested in 69,765 sibships. The dataset was constructed to maximize the number of children with health data (born 1997 and later) and school performance data (born 2002 and before).

Measures
Exposure: ADHD Due to the subsidized nature of the Norwegian healthcare system, general practitioners send reimbursement claims to the govern- is not easily attributable to other factors, which suggest that interventions must target ADHD symptoms directly independent of sex, parental education, early school performance, and other psychiatric disorders � The relative ADHD deficit is larger for girls and for children of highly educated parents, but the absolute deficit is smaller (i.e., girls with ADHD still outperform boys with ADHD, and children with ADHD and highly educated parents still outperform children with ADHD and less educated parents with ADHD) � The ADHD deficit is large in all school subjects, meaning interventions should target factors that is shared across different school subject, although the potential for improvement appears to be largest in theoretical subjects where the ADHD deficit is largest.
version 2 (ICPC-2, WONCA, 2005). Just like ICD-10 (WHO, 1992), ICPC-2 does not differentiate between different kinds of attention disorders such as ADD and ADHD, but instead uses the term Hyperkinetic Disorder (code P81). Henceforth, we treat ADHD and hyperkinetic disorder as synonymous. We defined individuals as diagnosed with ADHD if they, between the age of 10 and 16, had at least one contact with the primary care system registered with code P81.
Outcome: School performance At the end of 10th grade, children -normally 16 years old -are graded based on their whole-year performance and take exams.
Grades are integers on a scale from one to six and the grade point average (GPA) are the mean of all grades, including final exams (M = 4.12, SD = 0.83). The higher the grade, the better the performance. GPA was standardized to z-scores (M = 0, SD = 1) before analyses, meaning all coefficients can be interpreted as standard deviations. For individual school subjects, we limited our analyses to grades that most pupils share (e.g., excluding electives, see Table S1).
For these grades, we standardized using the combined mean and standard deviation of all grades irrespective of school subject (M = 4.17, SD = 1.08).
We also used scores from standardized national tests in mathematics and reading, which pupils take in fifth, eight, and ninth grade (approx. 10, 13, and 14 years old, respectively). These were also standardized to z-scores.

Parental education and other demographic information
Statistics Norway provided information on parental education at the time of the child's graduation (i.e., age 16), which we coded into four categories reflecting the highest achieved education of either parent: Master's degree or equivalent (n = 53,648), Bachelor's degree or equivalent (n = 130,155), high school (n = 116,920), or not completed high school (n = 30,860), and a fifth category for children with missing information (n = 12,569). Statistics Norway also provided information on sex, parentage (for within-family analyses, see below), birth month, and parity (i.e., maternal birth order).

Comorbid psychiatric disorders
Comorbid psychiatric disorders were defined in a similar way to ADHD: at least one contact with the primary care system registered with diagnostic codes for psychiatric disorders between the ages of 10 and 16. For each individual, we counted the total number of unique psychological diagnostic codes (P70 and above, excluding P81) from the ICPC-2, and entered the count as a categorical variable (truncated at three). In an alternative model, we used separate indicator variables for all individual disorders with a prevalence above 0.1% (see Table S2).

Statistical analyses
To estimate the ADHD deficit, we calculated bivariate and adjusted associations between ADHD and GPA with linear regression models.
The covariates included in the adjusted models were sex, parental education, birth year, parity (truncated at five), and birth month. All of these were treated as categorical variables. The adjusted model only included covariates where reverse causation (and hence potential collider bias) is impossible or unlikely. Early school performance and comorbid mental disorders were therefore included in separate models. We used standardized test scores from fifth grade as a measure of early school performance. This variable had 28,765 missing observations, which were not included in the analyses using this variable. To investigate how the ADHD deficit varied across sex and parental education, we allowed ADHD to interact with sex and parental education in separate models. For comparison, we also estimated a regression model with covariates only (i.e., without ADHD).
The models were then re-estimated in within-family models (random-intercept multilevel models with parents as grouping variable) where we compared full siblings with and without ADHD. Only children with full siblings in the sample were included in these models (N = 145,051). Of the 69,765 sibships, 3952 (5.7%) included siblings discordant on ADHD. Covariates that vary within families (sex, parity, birth year, birth month) were included. As most full siblings will have equally educated parents, only the interaction between ADHD and sex was estimated with a within-family model.
Finally, the adjusted and within-family models with and without interactions were re-estimated with each individual school subject as the dependent variable.
We performed sensitivity analyses with several alternative outcomes. First, we re-estimated all models with standardized tests from eighth and ninth grade as dependent variables. Second, we included the 14,969 individuals who did not have registered GPA within normed time and used logistic regressions to re-run the adjusted models with missing GPA as the dependent variable.

RESULTS
Of the 344,152 included children (51% boys), 13,800 (4.01%) were diagnosed with ADHD (see Figure 1). The prevalence was more than twice as high among boys (5.54%) than among girls (2.42%). The prevalence was more than three times higher among children of parents where neither had completed high school (6.87%) compared with children where at least one parent had a Master's degree or equivalent (1.91%).  Figure 2). The ADHD deficit was substantially larger than deficits associated with any other diagnoses (see Table S3-S4).

The ADHD deficit in GPA
The effects of covariates on GPA were only negligibly attenuated compared to a model where ADHD was not included (see Table S3-S4).
For example, in a model without ADHD, girls had on average 0.52 (0.52, 0.53) standard deviations higher GPA than boys, whereas in a model with ADHD, the difference was 0.50 (0.49, 0.50).

The ADHD deficit across sex and parental education
The ADHD deficit was −0.17 (−0.20, −0.14) standard deviations larger for girls than for boys. As seen in Figure Table S4).
Similarly, the ADHD deficit was substantially greater among children of more educated parents, although it did not differ between the two highest education levels as illustrated by the overlapping confidence intervals in Figure  Neither the ADHD deficit, general sex differences, nor parental educational differences are constant across school subjects, which must be considered when interpreting the relative deficits. Figure 4 presents expected grades (z-scores) for individuals with and without ADHD across sex or parental education (see also Figure S2-S3). In some school subjects, the sex difference is almost as large as the ADHD deficit. Thus, in school subjects with large sex differences such as Norwegian (i.e., language skills), boys without ADHD barely outperformed girls with ADHD, despite the ADHD deficit being larger  ) showing the Attention-Deficit Hyperactivity Disorder (ADHD) deficit on GPA (z-scores) bivariate and adjusted for sex, parental education, birth year, parity, and birth month, and additionally adjusted for early school performance and comorbid disorders. The lower panel shows the ADHD deficit by sex and parental education. As most full siblings will have equally educated parents, we did not include parental education in the within-family models (hence no interaction between parental education and ADHD in the within-family model) for girls. Because grade differences by parental education are larger than sex differences, this pattern becomes more pronounced for parental education. Children with ADHD and highly educated parents perform better in most school subjects than children without ADHD but with parents who had not completed high school.

Sensitivity analyses
All analyses were repeated with standardized tests from eighth and ninth grade as outcome variables, which yielded similar results (Tables S10-S17 and Figures S4-S7). Missing GPA was also analyzed with logistic regressions, again with similar results. Children with ADHD were 4.28 (4.08, 4.49) times more likely than unaffected children to not have GPA registered, adjusting for covariates (see Table S18 and Figure S8). There were no statistically significant interaction effects in this set of analyses.

DISCUSSION
Using large and representative register data, we have found: (1) that the ADHD deficit in school performance is large and evident even when comparing siblings and when adjusting for comorbid disorders and early school performance, (2) that the relative ADHD deficit is somewhat larger for girls and children of highly educated parents but the absolute deficit is smaller, and (3) the ADHD deficit is largest in theoretical school subjects, but still substantial in all school subjects.
In the sensitivity analyses, we also found that ADHD increased the likelihood of not being registered with a GPA, meaning the deficit on GPA is a conservative estimate of the association between ADHD and poor school performance.
Even though children diagnosed with ADHD had substantially lower GPAs than other children, with an overall deficit of −1.11 standard deviations, this was partly due to family confounding. When comparing same-sex siblings and therefore controlling for unobserved family characteristics and genetic similarity, the ADHD deficit was reduced to −0.60 standard deviations. The ADHD deficit remain large after adjusting for potential confounders, but it would nevertheless be premature to conclude that the relationship is causal. A second caveat to causal interpretations is reverse causation: ADHD and school performance pose a chicken and egg problem in that children must show age-inappropriate levels of inattention and/or hyperactivity in at least two contexts to be eligible for ADHD diagnosis (APA, 2013;WHO, 1992). For children, school will likely be one of these contexts, meaning that many may have received an ADHD diagnosis because they struggled in school.
While our data does not permit us to completely discount reverse causation, controlling for early school performance should give an indication: If poor early school performance was the primary reason children were diagnosed with ADHD, then the association between ADHD and later school performance should disappear or be substantially reduced when adjusting for early school performance.
Instead, we found that among children with similar performance in fifth grade, those with ADHD still had on average −0.54 standard deviations lower GPAs at the end of 10th grade compared to those without ADHD. In other words, ADHD is not only associated with poor school performance, but also with worsening school performance relative to peers. This is not consistent with poor early school performance being the primary reason children are diagnosed with ADHD.
Most of the ADHD deficit remained, but it was attenuated by about a third from −0.86 down to −0.54. It can be tempting to interpret this to mean that a third of the ADHD deficit is due to selection bias. We caution against this interpretation, as we cannot disentangle the degree to which the attenuation is caused by selection bias (poor early school performance → ADHD) or mediation (ADHD → poor early school performance), and adjusting for mediators will underestimate the effect.
A third caveat is that ADHD often covaries with other mental disorders, such as depression and anxiety, which might confound the association between ADHD and school performance. We found that adjusting for other mental disorders had only a negligible impact on the ADHD deficit, meaning the ADHD deficit is generated independently of other mental disorders. This mirrors earlier research finding that ADHD impacts school performance independently of comorbid conduct disorders (Daley & Birchwood, 2010). Overall, we find that the ADHD deficit is not easily attributed to other factors, and that ADHD therefore reflects an independent risk factor of poor school performance.
The effect of covariates on GPA remained similar in models with and without ADHD, suggesting their effect on GPA is not partly mediated by ADHD diagnoses. Differences in school performance across, e.g., sex and parental education must therefore be explained by other factors.
The relative ADHD deficit was larger in groups with lower prevalence of ADHD: It was 22% larger for girls than for boys, and 39% larger for children with highly educated parents compared with children of parents who did not complete high school. Similar findings were found for individual school subjects. There are at least two ways of interpreting the larger relative deficit in low-prevalence groups: First, the threshold for receiving an ADHD diagnosis may be higher for girls and for children of highly educated parents, which would result in the average diagnosed case being more severe and consequently would impact school performance more. Whether there are true differences or merely different thresholds for receiving a diagnosis remain debated (e.g., Slobodin & Davidovitch, 2019), but one representative study found sex differences to result from true differences in mean and variance of ADHD symptom severity, not selection bias (Arnett et al., 2015). The second and in our opinion more likely interpretation is that girls and children of highly educated parents appear more impacted by ADHD because their undiagnosed peers have relatively higher grades. In fact, the differences in school performance associated with parental education are often bigger than the relative ADHD deficits, meaning that ADHD-affected children of highly educated parents on average outperform children without ADHD with parents who did not complete high school (see Figure 4). Likewise, in some school subjects (e.g., Norwegian), girls with ADHD perform similarly to boys without ADHD. In terms of absolute performance, then, girls and children of highly educated parents are less impacted by ADHD.
We found that the ADHD deficit was large in all school subjects.
This mirrors Jangmo et al. (2019), who reported similar findings from Swedish registers (see Appendix S1 for a more detailed comparison).
We also investigated interactions and found it particularly large between ADHD and parental education in Mathematics. The interaction between sex and ADHD were mostly similar across school subjects, despite varying sex differences. Large ADHD deficits in all school subjects suggest that the effect of ADHD is largely mediated through general factors shared across school subjects. We do not know what these are, but they could relate to classroom size, organisation of homework, or emotion regulation (e.g., Daley & Birchwood, 2010;Rushton et al., 2020). Potential interventions should therefore target general factors that impact school performance in a way that is shared across school subjects. However, if limited resources forces interventions to target specific school subjects, then the potential for improvement appears to be largest in theoretical subjects such as Mathematics or Science, where the ADHD deficit was largest.

Strengths and limitations
Population-wide register studies like this have numerous strengths (Thygesen & Ersbøll, 2014). First, unlike clinic-referred samples and cohort samples, this study does not suffer from non-random attrition and is much less affected by selection bias. In addition to unrepresentative results, selection bias can systematically bias estimates when the investigated variables are associated with likelihood of participation (i.e., collider bias: Munafò et al., 2018). The subsidized and equal-access nature of the Norwegian healthcare system means cases are unlikely to go unregistered, and only a few children (4.3%) did not have GPA registered (which we included in sensitivity analyses), meaning this study captures a representative picture of the association between ADHD and school performance. Second, register studies have very large sample sizes, which results in narrow confidence intervals and consequently high statistical power. Even large cohort studies can have few participants satisfying several criteria, such as being girls with ADHD and highly educated parents, which would result in low power and large confidence intervals even if the original cohort sample was large (Button et al., 2013). The difference in clinical consequences between the ends of large confidence intervals can be considerable, and small but meaningful differences may go undetected (Funder & Ozer, 2019;Götz et al., 2021;