The (a)typical burden of COVID-19 pandemic scenario in Autism Spectrum Disorder

Psychological and mental health consequences of large-scale anti-contagion policies are assuming strong relevance in the COVID-19 pandemic. We proposed a specific focus on a large sample of children with Autism Spectrum Disorder (ASD), developing an ad hoc instrument to investigate changes occurred in specific (sub-)domains during a period of national lockdown (Italy). Our questionnaire, named AutiStress, is both context-specific (being set in the COVID-19 pandemic scenario) and condition-specific (being structured taking into account the autistic functioning peculiarities in the paediatric age). An age- and gender-matched group of neurotypical (TD) controls was also provided. As expected, the severe lockdown policies had a general negative impact both on ASD and TD children, reflecting the obvious burden of the pandemic situation. However, our findings also indicate that children with ASD experienced more positive changes than TD ones. Noteworthy, we report a thought-provoking double dissociation in the context-specific predictor (i.e., accessibility to private outdoor spaces), indicating that it impacts differently on the two groups. Focusing on the ASD group, results suggest a condition-specific impact of the COVID-19 pandemic on core autistic (sub-)domains. Taken together, our data call for a multi-layered, context- and condition-specific analysis of the pandemic burden beyond any oversimplification.

AutiStress was developed to investigate changes occurred in specific domains and sub-domains (also defined as types-of-change) following the severe anti-contagion movement restrictions policies. It is context-specific being set in the COVID-19 pandemic scenario, and condition-specific being structured taking into account the autistic functioning peculiarities in the paediatric age. Thus, we differentiated core Vs. non-core autistic features (both in domains and in sub-domains) according to the idea that ASD is not a condition primarily characterized by phenotypical "social" manifestations. AutiStress was distributed using Google Forms between 3rd and 29th April, 2020. After a preliminary description of the aims of the study, we collected general basic socio-demographic information. Notably, we asked about the house characteristics considering that accessibility to outdoor spaces may be a critical factor during a period characterized by severe movement restrictions. To maximize statistical power and to avoid misattributions due to ambiguous situations, we merged the four choices using a handy dichotomization between apartment regardless of the presence of the balcony (hereafter, apartment) and home with access to private outdoor spaces (hereafter, private garden).
We structured our requests into seven domains referring both to core autistic features (Sensory Interests; Repetitive Behaviours) and non-core ones (Mood; Play; Eating Behaviour; Circadian Rhythm Sleep; Bowel and Bladder Control). For each domain, parents were asked to report-if any-children's behavioural changes as compared to pre-pandemic situation (e.g., "Did your child show any change in his/her mood? [yes/no]"). Following a negative answer, the online form moved to the subsequent domain. On the contrary, if parents replied positively in that domain they were asked to rate on a 5-point Likert scale how frequently several domain-related types-of-change occurred. Being a questionnaire that synergistically combined the efforts of clinicians engaged in remote support protocols and researchers, certain AutiStress requests were primarily driven by clinical purposes (notably to provide prompt and comprehensive monitoring for potential behavioural changes, even in less expected spheres). This was the case for example of the Bowel and Bladder Control domain, that was not extensively explored in our analyses due to very limited frequencies (12% in ASD group, 3% in TD group). Similarly, we simplified the Sensory Interests domain focusing exclusively on the main question (related to presence or absence of any change concerning sensory interests), without furtherly analysing sub-domains (i.e., each sense).

Statistical analysis. Given the different nature of dependent variables between domains (dichotomous)
and sub-domains/types-of-change (Likert scale), we tested our predictors of interest with different statistical models in Jamovi.
Domains. Generalized linear models were implemented to test main and interaction effects 29 . Group (ASD Vs. TD) and House-Characteristics (apartment Vs. private garden) entered in the model as dichotomous predictors, and Age as continuous covariate. The Group*House-Characteristics interaction also entered in the model. Moreover, for each sample we tested the effects of House-Characteristics and Age, in order to further assess the specific impact of these predictors within-group. In addition, being available children with ASD's cognitive functioning measures from pre-pandemic clinical records, in the ASD group we ran the analysis also considering the Cognitive Level as a covariate. This predictor was dichotomized into 'intellectual-disability' Vs. 'no-intellectualdisability' , to overcome variability in instruments used to measure cognitive functioning.

Sub-domains (types-of-change).
Analyses on the sub-domains were performed only on the residual sample of children reporting changes in the correspondent domain. Multiple linear regression models were implemented to test potential change in each sub-domain. Except for the Group*House-Characteristics interaction, which was not included here due to the small number of responses for some combinations of the levels of the two factors, for the analysis of the sub-domains we used the same predictors set that we used for the analysis of the domains. If the assumption of normality (K-S test) of the residuals was violated, then we dichotomized the continuous dependent variable into presence (i.e., from "1" to "4") or absence (i.e., "0") of change. Being sub-domains generally characterized by low frequencies in the right side of the Likert scale (i.e., responses "3-4"), we collapsed responses from "1" to "4". Then a generalized linear model was used. We anticipate that, due to their peculiar nature, the Mood sub-domains were analysed somewhat differently from the other sub-domains ("Changes in non-core autistic features" section).
As we observed missing values, we also tested the hypothesis that missing data were randomly distributed (Missing Completely At Random-MCAR Analysis). If missing values were detected within at least one subdomain, we ran the MCAR analysis within the correspondent domain. Control analyses confirmed that data are missing completely at random (Circadian Rhythm Sleep: p = 0.09; Eating Behaviour: p = 0.15; Play: p = 0.14; Mood: p = 0.62). No missing values were found in Repetitive Behaviour domain.

Results
Results are presented according to the condition-specific and context-specific hallmarks of AutiStress. Thus, we focus on the core Vs. non-core autistic features distinction, making reference both to specific core autistic domains and sub-domains. In addition, we focus on the House-Characteristics predictor to consider peculiarities of the COVID-19 pandemic scenario. Nevertheless, detailed significant results that do not fall into this conceptual and clinically-driven framework are reported in Tables S2, S3, S4.
Changes in non-core autistic features. Mood domain represents an unspecific and non-core autistic feature that may reliably help in characterizing the burden of the pandemic scenario both for ASD and TD children. We did not find Group  www.nature.com/scientificreports/ LRT-χ 2 = 0.08, p = 0.78) (Fig. 1a,b), nor any other significant effect of House-Characteristics (LRT-χ 2 = 0.10, p = 0.75), Age (LRT-χ 2 = 1.86, p = 0.17) or Group*House-Characteristics (LRT-χ 2 = 0.62, p = 0.43). Concerning the within-group analyses, no significant effects of any predictor were found neither in the ASD nor in the TD group (ASD group: House-Characteristics LRT-χ 2 = 0.75, p = 0.39; Age LRT-χ 2 = 0.51, p = 0.48; Cognitive Level LRT-χ 2 = 0.12, p = 0.73; TD group: House-Characteristics LRT-χ 2 = 0.05, p = 0.83; Age LRT-χ 2 = 1.86, p = 0.17). Considering their peculiar nature, the Mood sub-domains were analysed differently from the other subdomains. Specifically, instead of analysing separately the types-of-change, we first grouped them based on their positive Vs. negative value (Table S1). In other words, we averaged all the intra-subjects negative and all the intra-subjects positive mood changes. This allowed us to obtain valuable general information about positive and negative mood changes, independently of the specific nature of these changes. Because neither mean positive nor mean negative mood changes were normally distributed (ps < 0.01), we used Wilcoxon Paired Tests to compare ranks of mean positive Vs. mean negative differences in mood changes within each group. The results showed that both ASD and TD children reported more mean negative than mean positive changes in mood (ASD[N = 84]: 1.79 ± 0.86 > 1.30 ± 0.87, W = 2213, p = 0.005; TD[N = 40]: 1.50 ± 0.67 > 0.97 ± 0.96, W = 595, p = 0.004) (Fig. 1c).
In order to test the possible effect of the Group predictor on mean negative and mean positive mood changes, we used two separate Mann Whitney U-tests. If the U-test indicated a statistically significant effect of Group on the mean positive (mean negative) change, then we conducted a regression analysis with predictors Group, House-Characteristics, and Age on each of the positive (negative) type-of-change. Specifically, we used a generalized linear model if the assumption of normality of the residuals was violated and a simple linear model otherwise. In other words, analyses on the positive (negative) types-of-change were performed only if the Group predictor had a statistically significant effect on mean positive (mean negative) mood changes. In line with this post-hoc logic, we adjusted the significance threshold according to Bonferroni correction (Positive mood changes significant threshold: p = 0.05/3 = 0.017; negative mood changes significant threshold: p = 0.05/6 = 0.008). Note that, although these post-hoc analyses were focused on the possible effects of the Group predictor on each typeof-change, we also entered House-Characteristics and Age in the models in order to control for possible spurious correlation effects.
No The possible effects of the House-Characteristics predictor (private garden Vs. apartment) and of the Age predictor on the mean positive and mean negative mood changes were separately analysed using the same logic as for the Group predictor. The effects of both predictors were tested separately on the TD and ASD group. Consistently with the choice of using the Mann Whitney U-test, the Age predictor was dichotomized (i.e., children ' < 6 y.o. ' and ' ≥ 6 y.o. '). As for the House-Characteristics factor, in the TD group we found a significant effect in the Among the types-of-change more specifically reflecting core autistic characteristics, we found that ASD children were reported to show more changes than TD ones in the food selectivity (ASD:  As expected, the number of TD children showing any changes in the autistic core (sub-)domains was basically negligible. Thus, we proposed a specific intra-group focus on the ASD sample. This choice was also justified by the fact that these (sub-)domains refer to aspects taken into account during clinical rehabilitative protocols, therefore children with ASD's parents are familiar with them. In contrast, they may appear less transparent for neurotypical children's parents. Although we clinically hypothesized a potential beneficial effect of accessing to private outdoor spaces on core autistic domains, the effect of House-Characteristics (apartment Vs. private garden) did not attain (Repetitive Behaviours, apartment: 32% [ (Fig. 3a).
Among the types-of-change specifically reflecting core autistic features, the House-Characteristics predictor (apartment Vs private garden) similarly impacted on self-stimulation (apartment[N = 34]: 2.59 ± 0.86; private garden[N = 13]: 1.69 ± 1.11; β = 0.66; p = 0.029), denoting again that children without access to private garden were reported to show more changes (Fig. 3b). Notably, we also found a significant Age effect on self-stimulation (β = 0.36; p = 0.008), indicating that older children were reported to show more changes in this sub-domain.

Discussion
Governments worldwide are trying to face this pandemic providing health policies that balance distinct aspects such as epidemiological situation, socioeconomic and biopolitics factors 10,11,31 . Although expression "unprecedented times" is becoming a sort of cliché in the COVID-19 era 32 , are we really transferring this idea and the burden of its implications in the mental health clinical settings? Are we really dealing with the lockdown effects on individuals with ASD taking into account the unique scenario of COVID-19 pandemic? AutiStress tries to tackle this point balancing peculiarities of the current situation, the body of knowledge on ASD (≈ autistic functioning), and notably the interaction of these aspects. Our first result concerns the Mood domain, that refers to an unspecific and non-core autistic feature. As expected, the severe lockdown policies in Italy had a general negative impact both on children with ASD and neurotypicals controls. This generic and unsurprising result reflects the obvious burden of the pandemic situation, and it was a widely predictable effect of unpleasant (but undelayable) movement restrictions. Notably, we found that mean negative mood changes were higher than mean positive ones in each group (see also Tables S5,  S6). This suggests that the effect was not condition-specific for the ASD group whereas it is shared also by our neurotypical pediatric sample, in agreement also with previous reports on clinical/non-clinical populations 4-6 . However, and importantly considering AutiStress aims, children with ASD reported more positive mood changes than TD children (Fig. 1c); notably, they have been reported to show more changes in the more calm, in the more helpful, and in the more cooperating behaviours sub-domains. Although the mainstream view tended to focus on the presumed amplified burden in ASD 25,27,28 , the possibility that our clinical sample may experience also positive changes during the COVID-19 pandemic cannot be considered totally unexpected. As outlined in the Introduction, children with ASD may have benefited from reduced stressful situations (e.g., public transports, mandatory school activities) and from new more manageable home-based routines. This may be particularly plausible in our ASD sample, considering that all children pursued readapted remote clinical intervention provided also during the more critical period. This may have guaranteed at least basic continuity of care, notably supporting parents in structuring and managing daily activities, and it is consistent with other recent reports describing beneficial effects of remote/telehealth programs in ASD [33][34][35] . Our data also suggest an interesting double dissociation in modulating positive mood changes between the two groups (Fig. 2). Age seems to explain positive changes in the ASD group, with younger children reported to show more changes in the more calm, in the more helpful, and in the more cooperating behaviours sub-domains. These findings may partially reflect a sort of "protective" role of age, being younger children naturally less aware of the situation, and also the new home-based routines established by caregivers supported by our clinical staff (see also 36 ). In contrast, Age does not seem to play a critical role in explaining positive changes in the TD group. Intriguingly, an additional dissociation emerged for the House-Characteristics predictor that significantly predict positive changes in the TD group. Children with access to private outdoor spaces were reported to show more changes in the more calm and in the more cooperating behaviours sub-domains. In contrast, House-Characteristics does not seem to play a role in explaining positive changes in the ASD group. Taken together, results on the Mood domain depict a scenario largely consistent with the previous literature, with obvious negative effects of the severe anti-contagion policies impacting both on ASD and TD children. However, the double dissociation on positive changes may assume strong theoretical and clinical significance taking into account the autistic functioning peculiarities, and notably the distinction between core Vs. non-core autistic features.
A very timely report faced the challenge of setting up remotely readapted protocols 37 . This report has the merit of pushing readers in considering the peculiar core of the autistic phenotype, and nicely fits with the AutiStress purpose of differentiating changes in core Vs. non-core autistic features. Thus, our second major finding concerns a condition-related impact of severe anti-contagion policies on specific core autistic features. We found that children with ASD reported changes more often than TD controls in the Sensory Interests and Repetitive Behaviours domains. In addition, children with ASD reported significantly more changes than controls in food selectivity and in difficulties with transitions, that represent two prototypical sub-domains related to the autistic functioning (i.e., inflexibility, rigidity). From the one side, it is not surprising that neurotypical children basically did not present any changes in these aspects, and likely these requests may have appeared less transparent for neurotypical children's parents. However, percentages and rough proportion in the ASD sample clearly indicate relevant changes compared to pre-pandemic period in these condition-specific aspects. Intriguingly, a subtler and deeper analysis of our ASD sample suggests an additional thought-provoking dissociation with the neurotypical control group. Considering ASD children reporting changes in at-least-one of the autistic core domains (Sensory Interests and/or Repetitive Behaviours), a significant effect of House-Characteristics emerged. Moreover, House-Characteristics significantly impacts on a sub-domain distinctive of the autistic functioning such as self-stimulation (Fig. 3). Results on House-Characteristics in the ASD sample hint at refined and clinically insightful perspective-shift. Our suggestion is that a context-specific critical variable may have a relevant impact on condition-specific (sub-)domains (i.e., core autistic features for the ASD group), whereas it would impact on a more general and unspecific domain such as Mood in neurotypical controls. A provocative hypothesis is that Mood domain for TD children corresponds to core autistic domains for the ASD group, as it was the "core neurotypical domain". Although this remains a rather speculative hypothesis, our results clearly indicate a different effect of context-specific protective factors, and implicitly call for a multi-layered, condition-specific and less simplistic analyses of complex phenomena. Having access to private outdoor places reasonably represents a critical variable during severe lockdown. Potential mental health and well-being benefits of time spent outside in natural environments have been largely hypothesized in the literature 38 , notably in coping with strict movement restrictions 39 . Countries such as Italy and Spain promoted very restrictive lockdown policies in which people basically were not allowed to leave their homes except for healthy (e.g., brief outside walks were permitted in the Lombardy region for people with severe mental health disorders or intellectual disabilities, provided that www.nature.com/scientificreports/ the caregiver presented a written certificate signed by the clinical mental health specialist, see 37 ), emergency or essential job reasons, for buying food and medicines, or-in certain cases-walking the dog. Even in a scenario of national lockdown, other countries permitted at least limited access to nature or public outdoor spaces for recreational activities (e.g., France, UK). Thus, accessibility to private outdoor spaces (e.g., private garden) is likely more critical in areas, such as Italy or Spain, in which not even limited access to public park or greenspace was permitted. This hypothesis not only drove our choices in AutiStress of controlling house characteristics, but it also nicely fits with the findings recently reported in a large sample of 3403 individuals from Spain 39 . The authors tested the potential resilient effect of the residual contact with the nature, and results suggest that having access to private outdoor spaces and window views of natural features (as a proxy of indirect contact with nature, see 40 ) can act as a protective factor against negative consequences of severe stay-at-home restrictions. Without underestimating or even neglecting the role of other critical moderators (e.g., socio-demographic factors, personal circumstances), accessibility to private outdoor places may effectively represent a relevant factor in this pandemic. Accordingly, also our results indicate that private garden availability plays a role in coping the strict movement restrictions policies. More interestingly for our aims, we showed that children with ASD and neurotypical controls seem to be impacted differently from this modulator factor, as if it acted primarily on each individual's more "sensible" spheres. Here again, no doubt that House-Characteristics is not the only ingredient modulating individual reaction to the pandemic. For children with ASD a primary role was also played by familiar coping strategies, eventually supported by remote programs guaranteeing continuity of cares. Finally, a methodological point deserves to be underlined. It would be virtually meaningless considering each (sub-)domain per se, without a within-subject pre-/post-pandemic comparison. Two distinct approaches may solve this concern. The first one consists in the comparison of results on standardized questionnaires used during periodic clinical follow-up (as proxy of pre-pandemic timepoint) with a new re-evaluation ideally administered just before the relaxing of restrictions (as proxy of post-lockdown timepoint) [41][42][43] . This strategy has the undeniable advantage of using standardized instruments, however these instruments are not context-specific and usually not even condition-specific. A second alternative approach consists in ad hoc questionnaires. AutiStress clearly accompanied the respondents to not consider absolute rate for each (sub-)domain, whereas we drove them in focusing on potential changes in relation to pre-pandemic period. Noteworthy, AutiStress is context-specific: it is tailored to the COVID-19 pandemic situation (e.g. effects of enforced stay-at-home orders), and this inspired us in using ad hoc predictor (e.g., House-Characteristics) that would not be so befitting in other emergency contexts such as wars or natural disasters areas. AutiStress is also condition-specific: it is based on a renovated clinical framework that considers sensory processing and behavioural rigidity as the key components of ASD (≈ autistic functioning) that, in turn, inspired us in differentiating core Vs. non-core autistic features. Finally, asking for changes we implicitly "controlled" for a sort of individual threshold, and this helps to explain the modest impact of ASD children's cognitive level in our results. At first glance this result may appear weird, being clinically well-established that certain components are normally more pronounced in ASD children with lower cognitive abilities (e.g., self-stimulation for which-indeed-results indicate a non-significant trend). Thus, we speculate that the emphasis on reporting changes may explain the absence of any significant results of the cognitive level (domains, all ps > 0.22; sub-domains, all ps > 0.11).
We intended AutiStress both as a case-specific instrument and a potentially replicable tool, as long as specific (but finally minimal) adaptations are provided. AutiStress strengthens the idea that we would benefit from an approach that considers seriously the autistic functioning and its learning/thinking/acting peculiarities, since extraordinary situations may selectively impact on them. This does not imply that more classical works focusing on general mental health symptoms regardless of the specificities of the clinical population are less informative, in light for example of the clear benefit of having access to standardized instruments. We simply state that they are not enough, and we should promote convergent/combined approaches that consider both perspectives. Similarly, facing an "unprecedented" emergency scenario we need peculiar contextual constraints (e.g., access to outdoor space). How we act and interact in the world is a puzzle even during golden age periods. Promoting the best possible equilibrium between safe and effective social dynamics during this pandemic time seems to be a matter of titanic importance both for neurotypical and autistic population. We are all main characters of this challenge.