Characterizing subdomains of insistence on sameness in autistic youth

Insistence on sameness (IS) encompasses a range of behavioral patterns, including resistance to change, routines, and ritualized behaviors, that can be present across social and non‐social contexts. Given the breadth of behaviors encompassed by IS, it is important to determine whether this domain is best conceptualized and measured as uni‐ or a multi‐dimensional construct. Therefore, the current study aimed to characterize the structure of IS and explore potentially distinct of patterns of associations between identified IS factors and relevant correlates, including age, sex, IQ, anxiety, social abilities, emotional and behavioral dysregulation, and sensory hypersensitivity. Exploratory graph analysis was conducted using the dimensional assessment of restricted and repetitive behaviors to examine the structure of IS in a sample 1892 autistic youth (Mage = 10.82, SDage = 4.14; range: 3–18 years; 420 females) recruited from the Simons Foundation Powering Autism Research for Knowledge cohort. Three distinct IS subdomains labeled as IS‐Ritualistic/sameness, IS‐Routines, and IS‐Others (referring to IS behaviors during interactions with others) were identified. Generalized additive models demonstrated that each of the IS subdomains showed a unique pattern of association with key variables. More specifically, while sensory hypersensitivity was significantly associated with IS‐Ritualistic/sameness and IS‐Routines, it was not associated with IS‐Others. Further, while emotional dysregulation was a unique predictor of IS‐Ritualistic/sameness (but not IS‐Routines or IS‐Others), social interaction abilities were a unique predictor of IS‐Routines (but not IS‐Ritualistic/sameness or IS‐Others). Current findings provide preliminary evidence that the IS may encompass several distinct subdomains.


INTRODUCTION
Insistence on sameness (IS) is a complex and heterogenous subdomain of restricted and repetitive behaviors (RRB) that encompasses a diverse range of behavioral and cognitive patterns exhibited across social and nonsocial contexts, including, but not limited to, inflexible adherence to routines, ritualized patterns of behaviors, and resistance to change (American Psychological Association, 2013).IS has consistently emerged as a unitary factor across factor analytic studies of general autism assessment measures and dedicated RRB instruments (Bishop et al., 2013;Honey et al., 2012;Lidstone et al., 2014;Tao et al., 2016).Given the breadth and diversity of behaviors within this domain, it has been suggested that existing measures might undersample IS complexity and that, rather than being a unidimensional construct, IS may be multifaceted, encompassing several distinct subdomains (Lecavalier et al., 2020;Leekam et al., 2011;Uljarevi c, Spackman, et al., 2022).Thus, further work is needed to characterize the structure of the IS domain.
IS factors identified across different studies and instruments have encompassed behaviors as diverse as rituals, routines, resistance to change, and sometimes even compulsions and/or circumscribed interests.Some measures have attempted to draw conceptual distinctions between potential subdomains.For instance, the Repetitive Behavior Scale-Revised (RBS-R; Bodfish et al., 2000) originally aimed to differentiate between ritualistic behaviors, defined as the tendency to perform activities in a specific manner, and sameness behaviors, defined as a preference for environments to remain the same.However, majority of subsequent factor analytic studies failed to support the noted distinction and found that items across these two hypothesized constructs load onto a single factor (Bishop et al., 2013;Russell et al., 2019;Uljarevi c, Spackman, et al., 2022).Additionally, some previous studies have attempted to distinguish between rituals and routines, whereby routines are defined as behaviors performed in the same sequence each time to achieve an instrumental goal, and rituals are behaviors with a more symbolic meaning or affective component (Boyce et al., 1983).However, empirical support for this noted distinction is currently lacking.Finally, although IS is traditionally conceptualized within non-social contexts, recent work has emphasized that it also occurs in the context of social interactions (Lecavalier et al., 2020;Manor-Binyamini & Schreiber-Divon, 2019;Strang et al., 2017), an aspect that is not captured by most current measures of IS.
Characterizing the structure of the IS domain is important for advancing our understanding of correlates and mechanisms.For instance, although IS has been theorized to serve as means for reducing anxiety by constraining unpredictability (Laing et al., 2009;Leekam et al., 2011;Uljarevi c et al., 2017), recent small scale studies have suggested that autistic individuals often report elevated anxiety in response to certain IS behaviors, especially when these behaviors are interrupted or not completed, and that levels of distress significantly vary between individuals and contexts (Jaffey & Ashwin, 2022;Rodgers et al., 2016).Thus, it is important to clarify whether the association between IS and anxiety is unform across different manifestations of IS, or if specific behaviors may be more strongly linked to anxiety than others.Further, if IS is multidimensional, collapsing potentially distinct subdomains within a single total score may obscure specific patterns of association with other clinical, demographic, and cognitive correlates and result in findings that are inconsistent and lack generalizability.Indeed, literature exploring IS association with age, sex, IQ, social functioning, self-regulation, and sensory features has reported largely inconsistent findings (Baribeau et al., 2021;Black et al., 2017;Evans et al., 1997Evans et al., , 2017;;Factor et al., 2016;Frazier et al., 2014;Gotham et al., 2013;Lam et al., 2008;Lidstone et al., 2014;Richler et al., 2010;Uljarevi c et al., 2021;Uljarevi c, Frazier, et al., 2022;Uljarevi c, Spackman, et al., 2022).
In summary, despite tentative evidence for the multifaceted nature of IS, factor analytical investigations using comprehensive measures of IS are still lacking.Therefore, the current study employed a novel, comprehensive measure of RRBs to characterize the structure of IS.Further, we aimed to explore patterns of associations between potential IS factors and demographic, cognitive and clinical correlates previously suggested to play an important role in the expression of IS.These included age, sex, IQ, anxiety, social difficulties, and different aspects of selfregulation, and sensory hypersensitivity.Given the lack of previous factor analytic studies that explore the factor structure of IS, it was not possible to form specific hypotheses.Nevertheless, it was expected that IS within social contexts would form a distinct factor.

Participants
Parents of autistic youth were recruited through the Simons Foundation Powering Autism Research for Knowledge (SPARK) cohort (SPARK Consortium, 2018).In line with previous SPARK studies (Fombonne et al., 2022;Uljarevi c, Frazier, et al., 2022), only individuals with a parent-reported professional autism diagnosis scoring above the Social Communication Questionnaire (SCQ, Rutter et al., 2003) cut-off were included.The total sample included 1892 youth with autism (M age = 10.82,SD age = 4.14; range: 3-18 years; 420 females; Table S1).This research study was approved by the Stanford University Institutional Review Board (IRB #53669).

Measures
IS was assessed using Insistence on Sameness items from the Dimensional Assessment of Restricted and Repetitive Behaviors (DARB; Uljarevi c, Frazier, et al., 2022), a comprehensive informant-rated RRB questionnaire.For a full list of IS items see Table S2.In addition, parents also completed a battery of online parent-report questionnaires summarized in Table 1.As standard measures of self-regulation were not available, the SDQ dysregulation profile score was used to index emotion dysregulation (Deutz et al., 2018), whereas the SDQ hyperactivity subscale served as an indirect indicator of behavioral dysregulation.Parent-reported cognitive ability scores were available for 958 children and was coded on 10 levels (IQ = ≤24, 25-39, 40-54, 55-69, 70-79, 80-89, 90-109, 110-119, 120-129, and ≥130).Validity studies of the SPARK dataset have shown this to be a strong proxy estimate of IQ (Shu et al., 2022).

Statistical analysis
Prior to exploring the structure of IS, the dataset was randomly divided into an exploratory (n = 971) and confirmatory (n = 921) subsample.Exploratory graph analysis (EGA; Golino et al., 2020) using the graphical least absolute shrinkage and selection operator (GLASSO; Friedman et al., 2008) and the Walktrap community detection algorithm was conducted on the exploratory subsample in R using the EGAnet package.Reliability and item stability were investigated using bootstrapped EGA with 2000 bootstrap samples (Christensen, et al., 2020).
Following the EGA, the fit of the derived IS model was tested in the confirmatory subsample using the confirmatory application of Exploratory Structural Equation Modeling (ESEM; Asparouhov & Muthén, 2009) with target rotation and weighted least squares means estimator.Model fit was evaluated using the Comparative Fit Index (CFI); the Tucker-Lewis Index (TLI); the Root Mean Square Error of Approximation (RMSEA) and the Standardized Root Mean Square Residual (SRMR).CFI and TLI values ≥0.90 and RMSEA and SRMR values of <0.08 indicated adequate fit (Bentler, 1998).An additional bi-factor model was run to model the potential presence of a general underlying IS factor and further evaluate the distinctiveness of separate, correlated IS factors.Finally, generalized additive models (GAMs) were used to assess non-linear relationships between IS subdomains and key variables.

RESULTS
Of the 1892 autistic youth, 98% presented with at least one IS behavior.The most commonly endorsed items were bedtime routines (66%) and gets stuck on an idea (65%).Endorsement frequency across all items can be seen in Figure 1.
EGA suggested the three-factor model as optimal (Figure S1).Identified factors were interpreted as IS-Ritualistic/Sameness (10 items), IS-Routines (6 items), and IS-Others (7 items).The IS-Ritualistic/Sameness factor included items related to insistence that objects in one's environment are arranged in the "correct" manner or following certain rules, as well as items related to repeating certain activities or "rituals" over and over until an intrinsic sense of things being "just right" is achieved.The factor labeled IS-Routines captured behaviors performed in the same order or sequence each time, typically to achieve an instrumental goal.Finally, the IS-Others factor captured sameness behaviors in one's interactions with others, including insistence that other people also follow specific routines or rituals (See Table S2 for item list).Descriptive statistics are provided in Table 2. Bootstrapped EGA showed strong structural consistency, with the three-factor solution identified in 71% of replications and a mean item stability of 0.97 (Figure S2).Twenty-nine percent of replications favored a four-factor solution with two strongly correlated items (items 20 and 21) sometimes forming their own factor.However, as these two items showed acceptable weighted topological overlap (wTO = 0.280) and were conceptually distinct, both were retained in the final solution.Confirmatory ESEM confirmed the threefactor structure showed good fit (CFI = 0.96; TLI = 0.95; RMSEA = 0.06; SRMR = 0.03).Each subdomain F I G U R E 1 Percentage response distribution for all IS Items.For ease of interpretation, items in this figure are divided into the subdomains identified below using EGA.showed strong reliability (IS-Ritualistic/sameness: ω = 0.89; IS-Routines: ω = 0.87; IS-Others: ω = 0.83).As identified factors were significantly inter-related (IS-Ritu alistic/sameness-IS-Routines = 0.76; IS-Ritualistic/sameness-IS-Otherss = 0.57; IS-Routines-IS-Others = 0.62), the presence of the general IS factor was modeled.The bifactor model showed a slightly better fit (CFI = 0.98; TLI = 0.96; RMSEA = 0.05; SRMR = 0.03), however, all items continued to load significantly onto their individual factor after accounting for the presence of the general IS factor, providing further support for the validity of the described correlated, three-factor structure (Table S2).
Total scores for each IS subdomain were calculated by summing individual item scores for items loading onto the respective EGA factor.For each of the IS subdomains, GAMs were fitted with age, IQ, sex, social and communication abilities, anxiety, emotional and behavioral dysregulation, and hypersensitivity as predictor variables.Full model parameters are presented in Table 3 and significant relationships are depicted graphically in Figure 2. Anxiety was the strongest unique predictor of all three IS subdomains.While hypersensitivity was significantly associated with IS-Ritualistic/sameness and IS-Routines, it was not associated with IS-Others.Social interaction difficulties were a unique predictor of IS-Routines (but not IS-Ritualistic/sameness or IS-Others).Being male was a significant predictor of higher IS-Routines and IS-Others, but not of IS-Ritualistic/  which point this association became negative (see supplementary material S1).

DISCUSSION
The primary aim of this article was to characterize the structure of the IS domain.A three-factor structure encompassing IS-Ritualistic/sameness, IS-Routines, and IS in interactions with others (IS-Others) emerged in the exploratory and demonstrated excellent fit in the confirmatory subsamples.This study is the first large-scale, quantitative study to identify routines as a unique subdomain of IS which is most likely due to DARB having more comprehensive item coverage than previous measures.The ritualistic/sameness factor encompassed (i) items related to needing things in one's environment to be in the "right" place, and (ii) items capturing actions that are repeated over and over to achieve an intrinsic sense of things being "just right."On the other hand, the routines factor encompassed actions performed in the same way each time to achieve an instrumental goal.Interestingly, the item "Does your child insist that particular activities have to take place at the exact same time of the day regardless of the situation or context?" loaded onto the routines.It is possible that despite routines initially developing to achieve an instrumental goal, these behaviors may sometimes be overgeneralized and performed in the absence of an instrumental goal to satisfy an intrinsic feeling or need.Therefore, it will be important for future research to further clarify and characterize the distinction between rituals and routines.Finally, although recent work has emphasized that IS occurs in social interactions (Lecavalier et al., 2020;Manor-Binyamini & Schreiber-Divon, 2019;Strang et al., 2017), our study and a study by Strang et al. (2017) are the first to identify IS in social contexts as a distinct IS subdomain.The items loading onto this factor include both behaviors performed by autistic individuals themselves when interacting with others (e.g., asking repetitive questions) and behaviors they would like others to perform when interacting with them (e.g., insisting others respond is a certain way during social interactions).While it is important to note that the IS-Others subdomain has some conceptual overlap with stereotyped language domains identified in previous literature, items in this subdomain have a strong ritualistic/IS component.Therefore, it will be important for future studies to further address potential overlaps and distinctions between IS and other RRB subdomains, as well as to replicate and extend our understanding of the novel IS-Others subdomain.
The secondary aim of the study was to further examine the validity of the identified subdomains by exploring their associations with several theoretically driven correlates.Although several of the associations trended in the same direction for each of the subdomains, in particular for IS-Ritualistic/sameness and IS-Routines, no two subdomains had the same pattern of significant predictors.Anxiety was uniformly the strongest predictor of all three subdomains, with higher anxiety predicting higher IS.This finding is in line with previous studies that have consistently demonstrated a strong relationship between IS and anxiety (Baribeau et al., 2021;Black et al., 2017;Gotham et al., 2013).It has been theorized that IS behaviors may serve as a means of anxiety reduction, either by constraining unpredictability, preventing exposure to adverse stimuli, or as means of self-regulation (Evans et al., 1997;Hwang et al., 2020;Uljarevi c et al., 2017).Although more research is needed to examine this, it is possible that different IS subdomains may reduce anxiety through different pathways or mechanisms.For example, sensory hypersensitivity was the second strongest predictor of IS-Ritualistic/sameness and the third strongest predictor (after anxiety and age) of IS-Routines, but did not predict IS-Others.One possible explanation is that while rituals and routines may function to constrain or limit adverse sensory experiences (Black et al., 2017;Leekam et al., 2011), IS in social interactions may reduce anxiety through a different pathway or serve another function altogether.However, despite correlational findings reported in this study and elsewhere (e.g., Uljarevi c et al., 2017), it is important to note that evidence more directly speaking to the causative nature and directionality of these effects are currently missing; thus, it will be crucial for future research to utilize specific methods, including treatment-oriented research targeting different mechanisms, to address these gaps.
While higher emotional dysregulation was a significant predictor of lower IS-Ritualistic/sameness, it was not significantly associated with IS-Routines or IS-Others.Behavioral dysregulation did not predict any IS subdomains, although its relationship with routines was approaching significance.Although preliminary, these findings may suggest that specific aspects of selfregulation might play different roles in the expression of specific IS subdomains.Furthermore, although the direction of this association was negative, such that higher emotional dysregulation predicted lower IS behaviors, further analyses revealed that this association was positive prior to anxiety being introduced into the model, at which point emotional dysregulation instead became a negative predictor of IS-Ritualistic/sameness.One possible explanation for this finding might be that, for autistic individuals with high anxiety, IS behaviors may be an important method of self-regulation, meaning individuals with high anxiety who do not use these behaviors to self-regulate may be more likely to show higher levels of emotional dysregulation.However, it is crucial to emphasize the preliminary nature of these findings and highlight the need to replicate and extend current findings by using more comprehensive measures of different aspects of self-regulation, including different facets of executive functioning and emotion regulation strategies.Finally, social interaction difficulties indexed by the SCQ predicted IS-Routines, but not IS-Ritualistic/sameness or IS-Others.The SCQ is a broad index of social difficulties that does not capture fine-grained individual differences across specific facets of social functioning such as social motivation and social cognition.Thus, further research using more comprehensive measures of social functioning is needed to provide a better characterization of the nature of associations between distinct aspects of social processing and IS subdomains.
The findings reported here should be considered in light of several limitations.First, the study relied on parent-reports of diagnosis and IQ.As such, it was not possible to independently confirm diagnoses and it will be important for future research to replicate findings reported here by utilizing gold standard diagnostic instruments and dedicated cognitive assessments.Second, this study also relied on informant-report measures of IS and other clinical variables which has been shown to be an efficient means of measuring autism characteristics in samples including non-verbal participants (Irwin et al., 2012), but may be subject to parent bias.Further, we acknowledge the importance of incorporating selfreport measures into autism research.Finally, SDQ dysregulation and hyperactivity scores were used as proxies of emotional and behavioral dysregulation.Even though this approach has been used previously (e.g., Deutz et al., 2018) and hyperactivity and behavioral dysregulation constructs are strongly linked (Barkley, 1997;Granziera et al., 2021), scores used in this study nevertheless provide a limited index of emotional and behavioral dysregulation.Thus, future studies will need to replicate and extend findings reported here using dedicated parentreport (e.g., Behavioral Rating Inventory of Executive Functioning [Gioia et al., 2000]) and experimental (e.g., Go/No-Go Task [Verbruggen & Logan, 2008]) measures self-regulation and executive functioning.
While significantly more research is needed to support the validity and utility of IS subdomains identified here, this study represents an important step toward better characterizing specific manifestations of IS across different settings.Identified IS subdomains, IS-Ritualistic/ sameness, IS-Routines, and IS-Others, each showed unique patterns of association with key demographic, cognitive and clinical correlates.These findings offer preliminary evidence that these subdomains represent distinct but related constructs, potentially underpinned by distinct mechanisms, and highlights the importance of improving our characterization and measurement of the heterogenous IS domain.It will be important for future to replicate subdomains identified here, in particular the novel IS-Others subdomain, and to further characterize their clinical correlates and continuities and discontinues in underpinning mechanisms.
sameness.Higher emotional dysregulation was a unique predictor of lower IS-Ritualistic/sameness, but not IS-Routines or IS-Others.Supplemental analysis exploring the direction of this association revealed that emotional dysregulation was a positive predictor of IS-Ritualistic/ sameness prior to anxiety being included in the model, at F I G U R E 2 Effects of significant predictors on IS subdomains.Y-axis values reflect predicted IS scores for each x-axis value, when holding all other variables constant.The range of each y-axis has been chosen to improve readability and may not necessarily start at 0.