Autistic Characteristics in Youth with Anorexia Nervosa Before and After Treatment

Abstract

Objective.

Cognitive characteristics common to autistic individuals are often seen in adults with anorexia nervosa (AN), raising the question of whether autistic people and people with AN may share an endophenotype. We need to examine autistic characteristics during the early stages of AN to accurately parse true symptom co-occurrence from behavioral alterations due to prolonged illness.

Methods.

We conducted a post-hoc analysis examining autistic characteristics in 59 youth with AN. Adolescents and parents participating in a randomized-clinical trial for AN completed questionnaires probing autistic characteristics at baseline and treatment end. We categorized participants as above or below cut-offs of clinical indicators of autism using the Autism Probability Index (API) and the Autism Spectrum Quotient-10.

Results.

Rates of high autistic characteristics ranged between 0-36% depending on the instrument used and how the data was obtained (i.e., by informant report or self-report). Paternal report of autistic characteristics differed across treatment completers versus non completers and maternal report indicated lower weight gain for those with elevated characteristics.

Conclusions.

Low rates of autism and fluctuations in autistic features during treatment underscore the importance of longitudinal examinations of autistic characteristics in adolescents with AN. Future studies need to replicate findings in a larger adolescent sample.

Anorexia nervosa (AN) is a severe psychiatric illness that often onsets in adolescence (Swanson et al., 2011). Characteristics consistent with autism spectrum disorder, a neurodevelopmental disorder characterized by impairments in social communication and interaction as well as presence of repetitive or restrictive behaviors and interests (American Psychiatric Association, 2013), are often seen in those with longer duration of AN (Saure et al., 2020). This co-occurrence suggests that autistic features or autism may predict a severe and enduring course of AN (Nielsen et al., 2015). Previous research highlights overlaps between cognitive correlates of AN and autism, such as difficulties with theory of mind (Sedgewick et al., 2019) and set-shifting inefficiencies, i.e., the ability to transfer attention to new tasks, situations, or objects (Holliday et al., 2005). This overlap leads to a hypothesis of shared endophenotype or genetic risk (Boltri & Sapuppo, 2021; Kinnaird & Tchanturia, 2021; Rhind et al., 2014; Westwood, Eisler, et al., 2016; Westwood, Stahl, et al., 2016; Zucker et al., 2007).

Recent research indicates a positive association between autistic characteristics and eating disorder psychopathology in those between the ages of 18 and 55 (Tchanturia et al., 2019). A recent systematic review suggests that between 8-25% of adults with AN exhibit characteristics of or meet full criteria for autism (Boltri & Sapuppo, 2021). These rates appear lower in youth (Boltri & Sapuppo, 2021), a highly relevant difference, as autism onsets in childhood and AN often onsets in adolescence, with the exception of a recent study examining autistic features in children with AN, and finding autistic features present in 16.3% of participants (15 out of 92) using measures including the Autism Spectrum Quotient Child’s Version (AQC), standardized body mass index, and Children’s Eating Attitudes Test (ChEAT26) (Inoue et al., 2021). Twin studies have found higher rates of autism in AN probands when compared to people without AN; however, these rates do not exceed those observed in other psychiatric samples (Koch et al., 2015; Rhind et al., 2014). Autistic characteristics are also low in parents of youth with AN (Rhind et al., 2014). Thus, we raise the question as to whether higher rates of autistic characteristics observed in adult AN samples are a true diagnostic co-occurrence, or rather, behaviors akin to autistic characteristics may be a scar of the illness due to the impact of malnutrition on social cognition and executive functioning during a key developmental period. We must examine autistic characteristics in early stages of AN, and throughout the treatment process, to better understand whether autistic characteristics are present and a marker of poor AN prognosis or instead a potential sequelae of AN.

As a step in this direction, we examined the frequency of higher scores on measures of autistic characteristics in youth with AN participating in a randomized clinical trial. We took an exploratory approach to determine: the number of youth with increased likelihood of autistic characteristics at baseline, whether completion of treatment impacted questionnaire scores indicating autistic characteristics and likelihood of autism, and whether those who had scores above measure cutoffs of autistic characteristics at baseline had poorer treatment outcome than those with scores below cutoffs.

Methods

Participants

The sample comprised of 59 youth diagnosed with AN (93% restrictive type and 7% binge-purge type) and their parents. Participants were enrolled in a randomized clinical trial examining Cognitive Remediation Therapy as an adjunct for either parents or adolescents to Family Based Treatment for AN alone (Timko et al., 2021). Eligibility criteria are described in full in Timko and colleagues (Timko et al., 2021) but included adolescents aged 12-18 years with a diagnosis of AN and both biological parents willing to participate. Exclusion criteria included history of brain injury, current use of atypical anti-psychotic medication, or other diagnoses which could impact executive functioning (e.g., autism) for either parents or adolescent.

At baseline, adolescents were aged 15.39 years (SD = 1.69), with 51 assigned female and 8 assigned male at birth. The sample was overwhelmingly white (N = 55) and non-Hispanic (N = 58). Average BMI z-score at baseline was −1.38 (SD = 1.29), with 46% of sample meeting criteria for severe malnutrition. Developmental weight suppression (Singh et al., 2021) ranged from 0.01 to 5.73 (M = 2.16, SD =1.20) with adolescents averaging 83% of expected body weight at baseline. Average duration of illness was 1.03 years at presentation (SD = 1.22).

Measures

Eating Disorder Examination – Questionnaire (EDE-Q, Fairburn & Beglin, 2011).

This self-report version of the Eating Disorder Examination includes the same subscales (restraint, eating, shape, and weight concern). We found excellent internal consistency (a = .96) for scores at baseline.

Autism Spectrum Quotient (AQ-10, Allison et al., 2012).

The AQ is a 10-item questionnaire examining autistic characteristics. The measure was completed by parents for adolescents 12-15 years. Adolescents aged 16-18 completed a self-report version. Scores of 6 or higher (noted as “high” scores) indicate possible presence of autistic characteristics and suggestion that referral for autism diagnostic evaluation is warranted. The AQ-10 possesses good sensitivity and specificity within community samples for both informant and self-report versions (Allison et al., 2012; Westwood, Eisler, et al., 2016).

Behavior Assessment System for Children-3 Parent and Child Version (BASC-3, Reynolds, C.R. & Kamphaus, R.W., 2015).

This self and caregiver report system is a well validated assessment of behavioral and emotional problems in youth. Recent studies found this measure to sensitively and accurately distinguish autistic characteristics from other disorders, including ADHD (Zhou et al., 2020). Specifically for this study, we used the parent-reported Autism Probability Index (API) comprising items from several BASC scales (e.g., atypicality, social skills, and withdrawal). As the API is not included in the self-report, for adolescent data we examined the Possible Diagnostic Indicator for autism. T-scores on the API above 60 are considered “at-risk” and above 70 in the “clinical” range, suggesting responses similar to autistic people in the community. To be consistent with the terminology used in the BASC-3, the terms “normal,” “at-risk,” and “clinical” will be used here to describe the T-scores.

Procedures

Adolescents presenting for outpatient/inpatient eating disorder assessment/treatment at a specialist eating disorder program at a U.S. academic medical center between August 2019 and December 2020 were screened for eligibility. Families were randomized into one of three conditions as described by Timko and colleagues (Timko et al., 2021). The current study focuses on data from baseline and end of treatment (EOT), six months after baseline. We found no differences in variables of interest among treatment arms, hence data were collapsed across groups. All procedures were approved by the institution’s ethics review board and both parents and adolescents gave informed consent/assent.

Statistical analyses.

For all analyses, we dichotomously categorized participants on clinical indicators of autism using the API (“normal” vs. “at risk”/clinical) and AQ-10 (≥6<). We used chi square analyses to examine links between elevations on the API and AQ-10 with markers of poorer treatment outcomes, including treatment dropout and remission status at EOT. Remission criteria included % of target body weight (as determined by historical growth curve (Peebles et al., 2017) and self-reported eating disorder psychopathology [recovered for boys EDE-Q <0.61(Mond et al., 2014), for girls EDE-Q < 1.84 (Carter et al., 2001)]. Participants were categorized as not recovered (<90% target body weight), partially recovered (90-94.9% target body weight), or fully recovered (>95% target body weight, EDE-Q within normal range). We used Mann-Whitney U tests and independent samples t-tests to examine group differences on treatment outcomes (EDE-Q and BMI z-score) at EOT.

Results

Of 773 adolescents (79% female at birth) screened for eligibility for the primary study, only nine adolescents (1.16%) were excluded for a suspected or documented autism diagnosis. Note, only 273 screened were eligible for the original study and 59 were enrolled.

Autism Spectrum Quotient.

Only one adolescent self-reported a AQ-10 score above cutoff at baseline; three adolescents reported high AQ-10 scores at EOT. According to maternal-report, 7 (23%) youth were above the AQ-10 cut-off at baseline, with only 3 (16%) above the cut-off at EOT. Paternal report identified 5 (17%) youth above this cut-off at baseline, with only 2 (11%) at EOT (See Table 1). We found no significant difference between AQ-10 scores at baseline and EOT for maternal or paternal report (p > .05). For analysis, we categorized adolescents as scoring above the cut-off if they scored above 6 on self-report (youth age 16 and over) or if one parent (for youth under 16) score was 6 or higher.

Table 1.

Summary of AQ Scores (dichotomized low/high) at Baseline and End of Treatment

Baseline	EOT	Adolescent (N= 10)	Maternal Report (N = 16)	Paternal Report (N = 15)
Low	Low	7 (70%)	12 (75%)	11 (73.33%)
Low	High	2 (20%)	1 (6.25%)	1 (6.67%)
High	Low	0	1 (6.25%)	2 (13.33%)
High	High	1 (10%)	2 (12.5%)	1 (6.67%)

Note. Data includes youth for whom we have both baseline and EOT data

Autism Probability Index (API).

No youth endorsed sufficient characteristics on the BASC to suggest a possible diagnostic indicator of autism at baseline or EOT. In contrast, maternal report identified 21 (35.6%) of youth in the “at-risk” range on the API and one (1.7%) in the clinical range. Per paternal report at baseline, 22 (37.3%) adolescents were in the “at-risk” range and five (8.5%) were in the clinical range. No adolescents were in the clinical range at EOT per maternal or paternal report (see Table 2). At EOT, mothers indicated 11 adolescents as “at-risk” for autism, whilst fathers reported 12 (30%) in the “at-risk” range. Paired t-tests comparing parent rating API scores at baseline and EOT indicated a significant reduction in scores of autistic characteristics [mothers: t(39) = 2.30, p = .027, Hedge’s g = .36; fathers: t(39) = 2.83, p < .01, Hedge’s g = .44).

Table 2.

Summary of Autism Probability Index Category at Baseline and End of Treatment

Baseline	EOT	Maternal Report (N=39)	Paternal Report (N = 40)
Normal/Normal	Normal	20 (51.28%)	20 (50%)
Normal/At Risk	At Risk	6 (15.38%)	3 (7.5%)
Normal	Clinical	0	0
At Risk	Normal	7 (17.95%)	6 (15%)
At Risk	At Risk	6 (15.38%)	8 (20%)
At Risk	Clinical	0	0
Clinical	Normal	0	2 (5%)
Clinical	At Risk	0	1 (2.5%)
Clinical	Clinical	0	0

Note. Data includes youth for whom we have both baseline and EOT data

Relationship of increased likelihood of autism to demographic and clinical presentation

There were no sex differences in the likelihood of being in the at risk/clinical group on the AQ-10 (c²(1) = 2.35, p = .13) or API at baseline via maternal or paternal report (c²(1) = 0.00, p = .98 and (c²(1) = 1.28, p = .26, respectively). Using the AQ-10, participants did not differ in age (t(36) = 0.97, p = .34, Hedge’s g = .34) or length of illness (U(N_HighAQ = 11, N_lowAQ = 26) = 96.00, z = −1.57, p = .12). Similarly, with the API, participants did not differ in age [maternal report: t(56) = −0.24, p = .81, Hedge’s g = .07; paternal report: U(N_APIatrisk = 11, N_APIlow = 28) = 134.00, z = −0.62, p = .53] or length of illness [maternal report: U(N_APIatrisk = 36, N_APIlow = 21) = 299.50, z = −1.30, p = .19; paternal report: U(N_APIatrisk = 11, N_APIlow = 28) = 121.00, z = −1.03, p = .32].

Relationship of increased likelihood of autism to outcome

Scoring above the cut-off on the AQ-10 was not significantly linked to likelihood of drop out (χ 2(1) = 0.00, p = .98). Similarly scoring above cutoff on baseline maternal API was not significant with treatment dropout (χ²(2) = 0.93, p = .63). However, paternal report of autism risk on the API differed across treatment completers versus non completers (χ ²(1) = 5.25, p = .02), where those in the “at-risk”/clinical group were equally likely to remain in treatment or drop out, whilst those in the “normal” group were more likely to remain in treatment. Neither AQ-10 (χ ²(1) = 0.70, p = .40) nor API [maternal: χ ²(1) = 0.01, p = .92; paternal: χ ²(1) = 1.05, p = .31) were related to the need for higher level of care during the study.

Any AQ-10 elevation was not associated with weight gain at 4 weeks, t(26) = −0.37, p = .72, Hedge’s g = −.14. Higher likelihood for autism per the API at baseline was not significantly related to weight gain at 4 weeks via paternal report (t(35) = 1.51, p = .14, Hedge’s g = .57); however, maternal report did show lower weight gain for those in the API “at risk” group (2.47kg versus 4.25kg, t(53) = 2.65, p = .01, Hedge’s g = .73).

There was no significant difference in global EDE-Q scores at EOT for youth with high AQ-10 scores (U(N_HighAQ = 8, N_lowAQ = 20) = 79.00, z = −0.05, p = .96). This was consistent with baseline API [maternal: U(N_APIatrisk = 16, N_APIlow = 24) = 211.50, z = 0.54, p = .59; paternal: U(N_APIatrisk = 22, N_APIlow = 7) = 84.00, z = 0.36, p = .72]. We did not find a significant difference on EOT BMI-z score between those with high AQ-10, t(35) = −0.56, p = .58, Hedge’s g = −.20, or API, maternal: t(40) = 1.38, p = .18, Hedge’s g = .45; paternal: t(24) = 0.44, p = .66, Hedge’s g = .26 or AQ-10 (t(34) = −1.37, p = .18) scores at baseline.

Discussion

This was an exploratory post-hoc study examining the rates of youth with autistic characteristics in a sample of youth with AN participating in a clinical trial. Overall, we found low rates of co-occurring autistic characteristics suggestive of possible autism diagnosis in our screening sample (patients presenting for assessment/treatment of a restrictive eating disorder). Despite specifically excluding youth with a formal, or suspected, autism diagnosis, about one in three participants showed signs of being “at risk” of autism by scoring above 60 on the API, and five adolescents exhibited clinically significant characteristics by scoring above a 70 on the API. Youth deemed “at risk” or in the clinical range did not appear to differ on demographic or clinical presentation variables when compared to those without autistic characteristics. We did find some differences in clinical course between those “at-risk”/ “not at risk” based on the API, specifically, paternal reported elevations on the API appeared to predict poorer outcomes such as weight gain at four weeks and treatment completion. Further, autistic characteristics appeared to reduce in some youth with treatment, whereas others increased or maintained scores at EOT.

Although we must temper clinical implications of our findings given study limitations, there are several important inferences. Firstly, rates of autism diagnosis in our predominantly female screening sample were consistent with females in a general community population (Shenouda et al., 2022). Yet, even in a sample specifically intended to exclude autistic youth, we observed many youth meeting thresholds on screening instruments. This finding may reflect that autism is often underdiagnosed in girls or it may reflect the overlap between cognitive and behavioral characteristics of AN and autistic characteristics as assessed by screening items on the AQ and API. It could also reflect the social and cognitive challenges resulting from malnutrition. These findings highlight the need for future studies to use comprehensive diagnostic tools in their study of autistic characteristics in AN samples.

Importantly, our divergent findings across parent report on the API and outcomes were unexpected (i.e., maternal report differed across those achieving early weight gain, but paternal report differed across drop out vs. treatment completers). This divergence is particularly important given the difference between parental report of autistic characteristics and historical overreliance on maternal report. Although we made no hypotheses regarding this, prior research has seen associations between data from mothers (but not fathers) and early weight gain outcomes, which may speak to the greater role mothers often play in the refeeding process. And although the impact of autistic characteristics on dropout has not previously been examined, our results may relate to research finding associations between paternal treatment participation and end of treatment outcomes. Both findings emphasize the importance of collecting paternal data, when present/available, and examining it separately from maternal data. Youth-focused research with a parental collateral report often collapses parent data without recognizing the different perspective parents may have on the presence or severity of various emotional and behavioral characteristics in youth.

Finally, we observed fluctuations in autistic characteristics from baseline to EOT. These fluctuations highlight the importance of longitudinal examinations in AN research. Given that autism onsets during early childhood and is present across the life course we do not propose that these results are suggestive of a new diagnosis of autism spectrum disorder. Rather, these findings may reflect improvements in socioemotional and cognitive functioning following treatment or reintegration into school or social situations during FBT. Increases in autistic characteristics may reflect the social difficulties that can result from this readjustment following a medical absence. Future studies should aim to replicate these findings in a larger sample to examine factors mediating change in autism symptomatology.

Limitations and future directions

Limitations of this report include our small sample size and lack of racial and ethnic diversity within our sample, limiting the external validity of our results. As this was an exploratory study, we did have a number of analyses, and the significant findings need to be interpreted in light of this and the focus should be on the effect sizes found. Additionally, the two questionnaires have differing frameworks to quantify autistic characteristics and behavioral and emotional problems in youth, resulting in varied self and parent reported results. It is also unclear if the cutoffs provided for the general population are accurate in adolescents with AN. Finally, given that this was a post-hoc analysis, we did not include a diagnostic interview which would provide diagnostic validation of the autistic characteristics reported in the AQ and API. Future research should include a diagnostic valuation to examine autistic characteristics in more diverse samples with AN prior to, during, and after treatment to better understand the impact of co- occurrence on presentation and outcomes.