Is There a Female Protective Effect Against Attention-Deficit/Hyperactivity Disorder? Evidence From Two Representative Twin Samples

a PDF file of an that a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. ABSTRACT Objective: Attention-deficit/hyperactivity disorder (ADHD) is more frequent in males than females. The “female protective effect” posits that females undergo greater exposure to etiological factors than males in order to develop ADHD, leading to the prediction that relatives of females with ADHD will display more ADHD behaviors. We thus tested whether co-twins of females displaying extreme ADHD traits would display more ADHD traits than co-twins of males displaying extreme ADHD traits. Method: Parents of approximately 7,000 pairs of non-identical twins in Sweden, and around 4,000 pairs of twins in England and Wales, completed dimensional assessments of ADHD traits. Probands were selected on the basis of scoring within the highest 10% of the distribution in each sample. Dimensional scores of co-twins of probands, as well as the categorical recurrence rate, were investigated by proband sex. Results: Co-twins of female probands displayed higher mean ADHD trait scores ( (cid:1)̅ =0.62-0.79) than co-twins of male probands ( (cid:1)̅ =0.38-0.55) in both samples. This trend was significant in the Swedish sample (p<.01) and when the two samples were merged into a single, larger sample (p<.001). When the samples were merged, there was also a significant association between proband sex and co-twin’s categorical status, with more co-twins of female probands also being probands than co-twins of male probands. Conclusion: These findings support a female protective effect against ADHD behaviors, suggesting that females require greater exposure to genetic and environmental factors associated with ADHD in order to develop the condition. Data were collected from participants in two representative, community-based twin studies. The Child and Adolescent Twin Study in Sweden (CATSS) is a study of twins born in Sweden since 1992. Initially, the twins were contacted in connection with their ninth birthdays 15 . For the present study, data were collected from twins participating in CATSS when they were aged 9. The second sample comprised participants in the Twins Early Development Study (TEDS); TEDS is a sample of twins born in England and Wales between 1994 and 1996 16 . Data for the present study were collected from TEDS participants when twins were aged 8 years. TEDS and CATSS are representative of the populations of England and Wales, and Sweden, respectively 15-16 . Our findings provide a platform for future research into the genetic basis of ADHD to build upon. While twin studies of ADHD have consistently supported its high heritability 5-9 , elucidating the precise genetic mechanisms underpinning ADHD has proven elusive 22 . The female protective effect model provides an opportunity to raise further research questions in such research. For example, genes can be divided into high-impact and low-impact sets 23 . One possibility is that females with ADHD are more likely to inherit higher impact genes associated with ADHD, which are rarer. To illustrate, ASD also more commonly affects males than females 24 , and recent twin and family studies support a female protective effect against ASD 12,25 . A genetic study then indicated that females with ASD displayed a higher degree of larger copy number variants, which were more likely to be maternally inherited 11 . Similar studies of ADHD may well prove useful in furthering our understanding of the etiology of ADHD.


INTRODUCTION
Attention-deficit/hyperactivity disorder (ADHD) is a neurodevelopmental condition characterized either by excessive hyperactivity and impulsivity, inattentiveness, or a M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 2 combination of these symptoms 1 . Epidemiological studies suggest that, overall, ADHD affects between 5-7% of the population 2 . Notably, ADHD appears to be substantially more common in males than females. A study of ten European countries, for instance, indicated that males with ADHD outnumbered females with ADHD by ratios of two to one to sixteen to one 3 . The excess of males with ADHD has been further confirmed by meta-analyses, with four times as many males than females thought to be affected 2,4 .
A number of twin studies have established that ADHD is among the most heritable of neuropsychiatric conditions [5][6][7][8][9] . The high heritability of ADHD does not vary markedly whether it is conceptualized as a categorical, diagnosed condition [5][6] or treated as a continuous trait in the general population [7][8][9] , thus indicating that severe forms of ADHD may be linked genetically with milder, subclinical traits of ADHD in the general population. Such studies, however, have yet to shed light on the reasons why ADHD appears to be so much more common in males than females.
One possible explanation for the sharp sex discrepancy in ADHD prevalence is a putative "female protective effect" model. Under this model, females would be predicted to require greater exposure to genetic and environmental factors associated with ADHD than males in order to display sufficient ADHD behaviors to warrant a diagnosis, thus meaning that fewer females would be expected to be diagnosed with ADHD than males [10][11] . As such, one would expect more causal factors to be present in the families of females with ADHD, leading to the prediction that ADHD and ADHD behaviors will be more prevalent in the relatives of females with ADHD. The female protective effect is presently receiving considerable attention in relation to autism spectrum disorders (ASD), which are similarly male-biased conditions. In one study, for example, the fraternal co-twins of females displaying a high degree of autistic traits displayed more autistic traits than did co-twins of M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 3 males with high degrees of autistic traits, and were also more likely to display high scores themselves 12 .
Very few studies have tested for the existence of a female protective effect against ADHD. A recent Swedish investigation suggested that merely having a female co-twin is associated with displaying a greater degree of ADHD traits than having a male co-twin, although this study did not take account of the degree to which the index twin displayed ADHD symptoms 13 . In one study of a US-based twin sample, traits of ADHD were examined in the co-twins of individuals displaying a high degree of traits of ADHD. Co-twins of females displaying extreme ADHD traits displayed significantly greater ADHD-like behaviors than the co-twins of males displaying extreme traits of ADHD 14 . Of note, however, the effect was not present for the co-twins of the most severely affected twins, perhaps owing to the small effect size and lower number of twins displaying the very highest scores.
As a consequence, we aimed to test for the existence of a female protective effect against ADHD behaviors in two independent, large-scale European twin samples. We first tested whether the co-twins of females displaying extreme degrees of ADHD traits would exhibit more continuous ADHD traits than the co-twins of high-scoring males. Second, we sought to test whether high-scoring female twins were more likely to have a high-scoring cotwin than high-scoring male twins. The one previous study documenting this effect reported that the effect size was small 14 ; thus, we not only aimed to test for the female protective effect against ADHD in our two samples independently, but also pooled the two samples to increase power. We expected, in light of existing evidence, to find evidence of a female protective effect against ADHD behaviors [13][14] .

Participants
Data were collected from participants in two representative, community-based twin studies. The Child and Adolescent Twin Study in Sweden (CATSS) is a study of twins born in Sweden since 1992. Initially, the twins were contacted in connection with their ninth birthdays 15 . For the present study, data were collected from twins participating in CATSS when they were aged 9. The second sample comprised participants in the Twins Early Development Study (TEDS); TEDS is a sample of twins born in England and Wales between 1994 and 1996 16 . Data for the present study were collected from TEDS participants when twins were aged 8 years. TEDS and CATSS are representative of the populations of England and Wales, and Sweden, respectively [15][16] .
Both CATSS and TEDS comprise both monozygotic (MZ) and dizygotic (DZ) twins, although only DZ twins were included in this study owing to the fact that the genetic resemblance of two DZ twins within a pair is the same as the resemblance between two singleton siblings (approximately 50% of their segregating DNA code on average). Both same-sex and opposite-sex DZ twins were included. In CATSS, families of 6,817 pairs of DZ twins returned data, while 4,309 participating families in TEDS returned data. In CATSS, exclusions were conducted for known brain injuries and chromosomal syndromes (n=113),

Measures
In CATSS, the ADHD modules of the Autism-Tics, ADHD, and Other Comorbidities inventory (A-TAC) 17 were administered to parents of the twins over the telephone. There are two ADHD modules, assessing hyperactivity/impulsivity and inattentiveness, comprising a total of 19 items that correspond closely to DSM-IV criteria for ADHD 1 . Each item comprised a question, answered "yes" (for a score of 1), "yes, to some extent" (for a score of 0.5), or "no" (for a score of 0). Thus, the maximum possible score was 18. In the sample of DZ twins used in the present study, the A-TAC ADHD module had strong internal consistency (α = .92). A prior study reported strong construct validity for the scale, with 92% sensitivity and 75% specificity for detecting ADHD 18 . The A-TAC also comprises two subscales, assessing ADHD subtypes: Hyperactivity/Impulsivity (10 items) and Inattention (9 items).
Parents of twins participating in TEDS completed the ADHD subscale of the Conners' Parent Rating Scale-Revised (Conners ADHD) 19 . The measure was mailed out to parents of the twins, who completed and returned it. Like the A-TAC, the Conners ADHD measure comprises 18 items that are closely linked with the DSM-IV criteria for ADHD 1 .
Each item comprised a statement in response to which the parents rated, on a 0-3 scale, the extent to which each item was true of their children. The maximum possible score was 54. In the present study, the Conners ADHD showed strong internal consistency (α = .91).
Previously, individuals with ADHD have been shown to score more highly on the measure than controls 19 , supporting its construct validity. As with the A-TAC, Conners ADHD comprises Hyperactivity/Impulsivity and Inattention subscales (9 items each).

Proband Selection
In both samples, one twin was randomly selected as the "index twin." All other twins were co-twins. Probands were selected as the index twins scoring within the highest 10% of the A-TAC and Conners ADHD distributions, with such a cut-off designed to maximize M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 6 statistical power while capturing severe-enough cases. Thus, probands in CATSS were selected on the basis of A-TAC scores of 6.5 or more, while TEDS probands were defined as index twins scoring at least 23 on the Conners ADHD. Subsequently, analyses were repeated using more conservative cut-offs of 9.5 on the A-TAC and 28 on the Conners ADHD. These cut-offs were designed to capture the highest scoring 5% of each sample, thus testing for a female protective effect in relation to even more extreme scores. Due to the lack of sexspecific diagnostic criteria for ADHD, the same cut-offs were used to select probands, regardless of sex. The number of probands, split by sex, is given in Table 1.

Statistical Analysis
To test whether co-twins of female probands would display higher ADHD trait scores than co-twins of male probands, 3x2 between-subjects analysis of variance (ANOVA) was employed. Proband status of the index twin (male proband, female proband, or control) was the grouping variable, with co-twins' ADHD trait scores acting as the outcome variable. An omnibus test initially compared scores across co-twins of male probands, female probands, and controls, before planned comparisons compared the scores of co-twins of male and female probands. Individual p-values were adjusted for multiple comparisons within each sample using the Bonferroni correction. All reported p-values are adjusted in this manner.
Effect sizes were summarized using Cohen's d. Effect sizes were interpreted using Cohen's criteria 20 , with d of .20-.49 considered a small effect, .50-.79 medium, and greater than .80 large.
To test whether sex of the probands was associated with whether or not their co-twin would also be a proband, categorical analyses were employed. Using the above identified cutoffs, co-twins were classified as either "affected" (i.e., scoring above a given cut-off) or "unaffected" (i.e., scoring below a given cut-off). Chi-square tests of association were then used to test whether co-twin status was significantly associated with proband sex. Effect sizes were summarized using odds ratios (ORs).
Analyses were first conducted separately in CATSS and TEDS. To bolster statistical power, a third set of analyses was performed on the two samples combined. The Conners ADHD and A-TAC were both heavily, positively skewed and were therefore log transformed prior to analysis (see Table 2). All co-twin scores used in analyses were standardized by sex of the co-twin, thus ensuring that co-twin sex was controlled for and allowing easier comparability of findings across samples. All analyses were performed in R 21 .
Post hoc analyses subsequently tested for a female protective effect against specific ADHD behaviors. All of the analyses detailed above were repeated on the Hyperactivity/Impulsivity and Inattention subscales of the A-TAC and Conners ADHD.
Since these analyses were post hoc, p-values were not adjusted for multiple comparisons.

RESULTS
Descriptive statistics for the A-TAC and Conners ADHD are given in Table 2.

Analysis of Continuous Scores
Mean standardized scores of co-twins of male probands, female probands, and controls are all shown in Figure 1 for the analyses using the 10% cut-offs. In CATSS, scores differed significantly across the three groups (F 2,6688 =79. 35 Merging the two samples produced the same pattern; mean ADHD trait scores differed significantly in the three groups (F 2,10731 =175.90, p<.001), with co-twins of female probands showing the highest ADHD trait scores ‫̅ݔ(‬ =0.69), followed by co-twins of male probands ‫̅ݔ(‬ =0.45), and controls ‫̅ݔ(‬ =-0.06). The planned contrast confirmed that mean ADHD trait scores were significantly higher for co-twins of female probands than co-twins of male probands (t 10731 =-3.73, p<.001, d=.07).
All mean ADHD trait scores for co-twins of probands selected under the more severe, 5% cut-offs are given in Table 3. Merging the two samples produced the same pattern of results as the 10% cut-off. Index twin status exacted a significant main effect on the mean ADHD trait scores of co-twins (F 2,10731 =95.90, p<.001), with co-twins of female probands displaying the highest ADHD trait scores ‫̅ݔ(‬ =0.73), followed by co-twins of male probands ‫̅ݔ(‬ =0.51) and controls ‫̅ݔ(‬ =-0.03). Mean ADHD trait scores were significantly elevated in cotwins of female probands compared with co-twins of male probands (t 10731 =-2.38, p<.05, d=.05).
Using a 5% cut-off, the same pattern emerged in each individual sample. In CATSS, mean A-TAC scores differed significantly across co-twins of male probands ‫̅ݔ(‬ =0.43), female probands ‫̅ݔ(‬ =0.69), and controls ‫̅ݔ(‬ =-0.03), F 2,6688 =46.39, p<.01; however, mean A-TAC scores for co-twins of female probands were not significantly higher than mean A-TAC scores for co-twins of male probands (t 6688 =-2.10, p=. 16 Conners ADHD scores for co-twins of female probands were not significantly higher than mean scores for co-twins of male probands (t 4040 =-1.10, p=.27, d=.03). Table 4 shows the number of affected and unaffected co-twins by proband sex for each sample and cut-off. Using a 10% cut-off to select probands, the association between proband sex and co-twin status was only significant when CATSS and TEDS were merged In CATSS, a greater proportion of co-twins of female probands scoring above the 10% also scored above the cut-off (15% of co-twins of female probands compared with 9% of co-twins of male probands. The association was small and non-significant, however

Figures 1b and 1c show the mean scores of co-twins of probands on the A-TAC and
Conners ADHD subscales (Hyperactivity/Impulsivity and Inattention), with probands scoring within the highest 10% of the subscales. In CATSS, TEDS, and the combined cohorts, cotwins of female probands displayed the highest mean Hyperactivity/Impulsivity scores, followed by co-twins of male probands and co-twins of controls. In all analyses, the mean scores of co-twins of female probands were significantly higher than those of the other two groups (p<.05).
Inattention followed the same pattern, as shown in Figure 1c. Co-twins of female probands displayed the highest mean Inattention score, followed by co-twins of male probands and co-twins of controls. The mean scores of co-twins of female probands were significantly higher than both other groups of co-twins in all three analyses (p<.05).
These results are shown in full in tables S1-S4, available online.

DISCUSSION
This investigation sought to test whether a female protective effect can account for the substantially elevated prevalence of ADHD in males relative to females [2][3][4] . The results of this study lend partial credence to a female protective effect hypothesis for ADHD. In line with the results of an existing US study 14 and our hypotheses, there was some evidence to indicate that the co-twins of females displaying an extreme degree of characteristic ADHD behaviors displayed more such behaviors themselves than did the co-twins of males showing an extreme degree of ADHD traits. Further, co-twins of females with particularly high ADHD trait scores were more likely to display an extreme degree of ADHD behaviors than were the co-twins of males. As such, these findings tentatively indicate that a female protective effect could be a potentially viable model to aid understanding the development of ADHD.

M A N U S C R I P T
A C C E P T E D ACCEPTED MANUSCRIPT 11 Our findings provide a platform for future research into the genetic basis of ADHD to build upon. While twin studies of ADHD have consistently supported its high heritability [5][6][7][8][9] , elucidating the precise genetic mechanisms underpinning ADHD has proven elusive 22 . The female protective effect model provides an opportunity to raise further research questions in such research. For example, genes can be divided into high-impact and low-impact sets 23 .
One possibility is that females with ADHD are more likely to inherit higher impact genes associated with ADHD, which are rarer. To illustrate, ASD also more commonly affects males than females 24 , and recent twin and family studies support a female protective effect against ASD 12,25 . A genetic study then indicated that females with ASD displayed a higher degree of larger copy number variants, which were more likely to be maternally inherited 11 .
Similar studies of ADHD may well prove useful in furthering our understanding of the etiology of ADHD.
Indeed, while our study did not investigate any specific etiological mechanisms associated with ADHD, our findings suggest that investigating the degree of exposure to etiological factors associated with ADHD in males and females with the condition may be a worthwhile future research direction. While the above example mentioned larger, rarer copy number variants, one might also test whether females with ADHD exhibit a greater number of smaller, common genetic variants. Indeed, in using polygenic scores, which have yielded useful insights in the genetic architecture of ADHD 26 , one could investigate whether females with ADHD display a greater degree of common genetic variants associated with ADHD than males with ADHD 27 .
One could also extend this to causal environmental factors. While twin studies indicate that genetic factors seem to outweigh environmental factors in the etiology of ADHD 5-9 , research has implicated certain environmental exposures with ADHD. For instance, lower birth weight is thought to be a causal environmental factor in ADHD [28][29] . It

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
12 may be that females with ADHD undergo greater exposure to such factors compared with males; for instance, could females with ADHD display an even lower birth weight than males with ADHD?
The presence of a female protective effect against ADHD behaviors also has implications for clinical practice. If clinicians take account of family history when diagnosing ADHD, it may be beneficial to also account for the sex of any previously affected relatives, under the assumption that relatives of females with ADHD are more likely to exhibit ADHD symptoms than relatives of males with ADHD. The caveat to this assertion, however, is that our findings are based only on twin data. The female protective effect against ADHD needs to be replicated in alternative, non-twin samples before such a conclusion can be decisively drawn. For instance, a recent study of ASD found that siblings of female non-twins with ASD were more likely to have ASD than siblings of male non-twins 25 . Such studies of non-twin relatives are now needed in relation to ADHD.
It does need to be noted that the overall sizes of the effects reported here, where significant, were weak. The effect reported in this paper is less than half the size of the female protective effect in relation to autism reported in a similar study 12 . Indeed, significant findings only emerged for the more severe cut-off of 5% when the two samples used were merged to create a larger sample. The small effect size seen here is consistent with that reported previously 14 , and so it is quite clear that subsequent studies testing the female protective effect model of ADHD are going to need to depend on large samples.
The small female protective effect seen here does, nevertheless, stress the need not to discount alternative explanations for the increased number of males with ADHD relative to females. There is very limited research considering phenotypic differences between males and females with ADHD. For instance, one study investigated sex differences in ADHD across ten European countries, and reported that females with ADHD displayed more Two further possibilities cannot be discounted from our study. While our findings lend support to the notion of a female protective effect against ADHD, it is not mutually exclusive to the hypothesis that males have more risk factors for ADHD. It is also in theory possible that rater contrast effects drove the higher scores seen in co-twins of female probands. Rater contrast effects refer to the scenario whereby parent ratings of one twin are influenced by how they view their co-twin 32 . To create the observed pattern of results, parents would need to have shown a stronger rater contrast effect on the co-twins (who were both male and female) of male probands than of female probands. In twin analyses of the A-TAC and Conners ADHD scale, rater contrast effects have been modeled and shown to be modest 9,33 , suggesting that rater contrast effects are unlikely to be an adequate explanation for our findings.
In addition to the caveat of the small effect size, our study did have further limitations that need taking into account. Proband status was ascertained through use of dimensional questionnaire measures, as opposed to in-depth assessments of ADHD. The use of this approach would, however, have come at the cost of the large sample size. As alluded to above, only twins were used in this study. While we removed MZ twins to ensure that the genetic relatedness of the relatives in our sample was similar to fraternal siblings, it is important to know whether these findings extend to non-twin relatives in future. In defense of our use of a twin sample, on the other hand, there is evidence to indicate that ADHD traits are not elevated in twins relatives to singletons 34 . Finally, we did not examine the dimensions of hyperactivity/impulsivity and inattentiveness separately. This was intentional; we already had