|Home | About | Journals | Submit | Contact Us | Français|
Restricted and repetitive behaviors (RRBs) have long been considered one of the core characteristics of autism. RRBs include a very broad category of behaviors such as preoccupation with restricted patterns of interest (e.g. having very specific knowledge about vacuum cleaners), adherence to specific, nonfunctional routines (e.g. insisting on taking a certain route to school), repetitive motor manners (e.g., hand flapping), and preoccupation with parts of objects (e.g. peering at the wheels of toy cars while spinning them). Most research on RRBs has used caregiver reports either through interviews or questionnaires; thus, the purpose of this study was to use clinicians’ observations of RRBs, made during the Autism Diagnostic Observation Schedule (ADOS: Lord, Rutter, DiLavore & Risi, 2000) to discover how RRBs change over time in very young children who may have ASD and what other factors are related to having RRBs. The ADOS is a 45 minute long, semi-structured, standardized assessment of communication, social interaction and play, which was administered to 121 children with autism, 71 with pervasive developmental disorders-not otherwise specified (PDD-NOS), 90 with a nonspectrum disorder, and 173 children who were typically developing. Even during a relatively short-term observation in the context of an office visit, we found that RRBs occurred more frequently and were more severe in young children with autism and PDD-NOS diagnoses than children in other groups. Diagnostic group differences also emerged in the associations between RRB scores and participant characteristics (e.g. age, NVIQ scores, etc). We also examined different subtypes of RRBs and their associations with NVIQ, age, diagnosis, and gender.
Restricted and repetitive behaviors (RRBs) observed during the Autism Diagnostic Observation Schedule (ADOS: Lord, Rutter, DiLavore & Risi, 2000) were examined in a longitudinal dataset of 455 toddlers and preschoolers (age 8–56 months) with clinical diagnosis of Autism Spectrum Disorders (ASD; autism, n = 121 and pervasive developmental disorders-not otherwise specified (PDD-NOS), n = 71), a nonspectrum disorder (NS; n = 90), or typical development (TD; n = 173). Even in the relatively brief semi-structured observations, Generalized Estimating Equations (GEE) analyses of the severity and prevalence of RRBs differentiated children with ASD from those with NS and TD across all ages. RRB total scores on the ADOS were stable over time for children with ASD and NS; however, typically developing preschoolers showed lower RRB scores than typically developing toddlers. Nonverbal IQ (NVIQ) was more strongly related to the prevalence of RRBs in older children with PDD-NOS, NS and TD than younger children under 2 years and those with autism. Item analyses revealed different relationships between individual items and NVIQ, age, diagnosis, and gender. These findings are discussed in terms of their implications for the etiology and treatment of RRBs as well as for the framework of ASD diagnostic criteria in future diagnostic systems.
Restricted and repetitive behaviors (RRBs) have long been considered one of the core characteristics of autism (Kanner, 1943). According to the Diagnostic and Statistical Manual (DSM-IV; American Psychiatric Association, 1994), RRBs include a very broad category of behaviors such as preoccupation with one or more restricted patterns of interest (e.g. having very specific knowledge about vacuum cleaners), adherence to specific, nonfunctional routines (e.g. insisting on taking a certain route to school), repetitive motor manners (e.g., hand flapping), and preoccupation with parts of objects (e.g. peering at the wheels of toy cars while spinning them).
In the past, RRBs were thought to be rare in preschoolers or toddlers with autism (Charman & Baird, 2002; Stone, Lee, & Weiss, 1999; Ventola et al., 2006). This assumption has been challenged in recent studies that reported the presence of RRBs in preschoolers, toddlers, and even infants as young as 8 months later diagnosed with autism (Watson et al., 2007; Richler, Bishop, Kleinke & Lord, 2007). However, at young ages, RRBs are not unique to children with autism spectrum disorders (ASD) but are also present in children with nonspectrum disorders (NS), such as intellectual disabilities and language disorders as well as in typical development (TD) (Thelen, 1979; Evans et al., 1997; Sallustro and Constance, 1978). Even though RRBs are not unique to ASD, RRBs are described by caregivers as more prevalent and severe in very young children later diagnosed with ASD than in children later diagnosed with NS or found to be typically developing (Watson et al., 2007; Richler et al., 2007).
Examining early diagnostic differences in RRBs has important implications for the revision of the ASD criteria in future diagnostic systems (e.g. Diagnostic and Statistical Manual-V, Internal Classification of Diseases-11). Given previous findings (Charman & Baird, 2002; Stone, Lee, & Weiss, 1999; Ventola et al., 2006), one question is whether RRBs should be deliberately excluded as a requirement for the diagnosis of ASD in very young children. According to the current DSM-IV (APA, 1994), Pervasive Developmental Disorders-Not Otherwise Specified (PDD-NOS), a milder form of ASD, includes children who have social deficits that are similar to those in autism and difficulties in either communication or RRBs or both (but at a mild level). Thus, in the current system, children without any RRBs can be diagnosed with PDD-NOS. However, previous studies have found that RRBs consistently occurred more in young children with PDD-NOS as well as autism compared to children with NS or TD (Watson et al., 2007; Richler et al., 2007). In the current proposal for DSM V (www.apa.org), the current distinction between autism and PDD-NOS will be eliminated, creating one general category of ASD, within which different dimensions (e.g. social-affect, RRBs) may be quantified. If this is the case, all children with ASD could be required to have evidence of some kind of RRBs.
Because results of the past literature in this area have been less clear, further investigation of the possibility of subgroups within ASD is needed. For example, children classified with PDD-NOS have been found to show less severe levels of RRBs than those with autism (Loh et al., 2007; Georgiades et al., 2007). Yet research on PDD-NOS has suggested that there is no single behavior or factor that differentiates it from autism or Asperger syndrome. The way the terms, autism, Asperger syndrome, and PDD-NOS are used colloquially, it often appears as if these are discrete groups that can be clearly differentiated by the presence (vs. absence) of particular symptoms including the entire domain of RRBs. However, the evidence to date suggests that these subgroups within ASD share common symptoms, such as RRBs, with differences among them primarily lying in the severity of such symptoms.
Compared to a number of studies examining RRBs based on parent reports, only a few studies have examined RRBs using observational assessments. Because many observations occur within a brief time period, they may not provide an optimal opportunity for the assessment of RRBs. For this reason, Lord et al. (2000) initially excluded RRB items from the diagnostic algorithm in the Autism Diagnostic Observation Schedule – Generic (ADOS-G). However, because the inclusion of RRBs resulted in stronger predictive validity, recently revised algorithms for the ADOS-G do contain RRB items, (Gotham, Risi, Pickles, & Lord, 2007). DiLavore, Lord, and Rutter (1995) also included RRB items in the Pre-Linguistic (PL)-ADOS diagnostic algorithm because RRB scores significantly increased the ability to classify young children with autism, developmental delays, and TD and to predict diagnostic outcome years later (Lord et al., 2006).
Research using parent reports of RRBs in young children showed most subtypes of RRBs, but not all, were stable or increased over time (Richler, Huerta, Bishop, & Lord, 2010; Lord et al., 2006; Moore & Goodson, 2003). Using longitudinal data based on the RRB items in the Autism Diagnostic Interview-Revised (ADI-R; Rutter, LeCouteur, & Lord, 2003), Richler et al. (2010) found that RRBs in children with ASD either remained relatively high or increased over time from the age of 2 to 9. Lord et al. (2006) also found that RRB scores at age 2 predicted RRB scores at age 9 for children with autism and that their RRBs were stable over time. In another study, Moore & Goodson (2003) found that ADI-R scores for unusual preoccupations, compulsions and rituals, hand and finger mannerisms, and repetitive use of objects increased between 2 and 4–5 years while scores for complex mannerisms decreased over the same time.
Other clinical features such as children’s age and intellectual functioning and the interaction between those features have been found to be associated with RRBs measured by the ADI-R (Bishop, Richler, and Lord, 2006). Bishop et al. (2006) found an interaction effect of age and intellectual levels on RRBs among children with ASD under 12 years old; NVIQ was more closely associated with the presence of RRBs at older ages compared to younger ages. More recently, motor stereotypies in individuals with ASD and intellectual disability were found to show less improvement over time compared to individuals with ASD with higher intellectual functioning (Esbensen, Seltzer, Lam, & Bodfish, 2009). These findings raise the possibility that level of cognitive functioning is associated with changes over time in RRBs; however, these findings are based on parent reports of relatively older children and adults with ASD. Thus, longitudinal data based on toddlers and preschoolers provides important additional information as to whether this claim would apply to younger children.
Associations between RRBs and other clinical features of ASD can also vary depending on the specific RRBs. For instance, Turner (1999) suggested that “lower order” RRBs (e.g. unusual sensory interests) are negatively correlated with IQ and “higher order” RRBs (e.g. compulsions/rituals) are positively correlated with IQ. Using data obtained from children from 2 to 11 years old, Militerni et al. (2002) found that sensory behaviors were more prevalent in children with lower IQ scores, whereas complex motoric sequences were more prevalent in those with higher IQ scores. All of these findings highlight the importance of distinguishing different types of RRBs in relation to their associations with age and IQ scores.
Previous studies have also examined heterogeneity in RRBs using factor analyses of ADI-R items and found support for two different RRB factors-repetitive sensory-motor behaviors (RSMB; including items such as hand/finger mannerisms, unusual sensory interests, repetitive use of objects, complex mannerisms, etc.) and insistence on sameness (IS; difficulties with change in routine, compulsions/rituals, unusual attachment to objects, etc.) (Cuccaro et al., 2003; Szatmari et al., 2006; Bishop et al., 2006; Richler et al., 2010), or three factors -circumscribed interests in addition to the two factors previously mentioned (Lam, Bodfish & Piven, 2008). These factors differed in their associations with cognitive levels and age. For example, Richler et al. (2010) found that RSMB were associated with NVIQ and stable over time whereas IS was relatively independent of NVIQ and closely associated with age. It is important to note that none of these studies using a factor analytic approach have used observational data with toddlers and preschoolers, partly due to the limited number of items capturing RRBs in some available observational measures such as the ADOS. In addition, for ADOS items in modules appropriate to very young children, RRBs that have a quality of insistence on sameness (e.g. lining things up or placing objects in a particular ways) are not distinguished from other repetitive activities (e.g. spinning a top; opening and closing a door) so that the distinction between RSMB and IS behaviors cannot be made. However, even though a factor analytic approach is not feasible with a limited number of items in certain measures, heterogeneity in RRBs can be examined through observing differences in the associations between individual RRB item scores and other clinical features such as age and IQ.
The present paper adds to the emerging literature on RRBs by contributing longitudinal data from toddlers and preschoolers (from 8 to 56 months old) with autism, PDD-NOS, NS, and TD using scores from the Autism Diagnostic Observation Schedule (ADOS). RRBs that are coded in the ADOS are presented in Table 1. These items result in a single RRB total score on the ADOS diagnostic algorithms. Using the observational data, we hypothesized that:
Data for this study were primarily obtained from three research projects, First Words and Toddlers (FW/T), Early Diagnosis of Autism (EDX), and Word Learning (WL) at the University of Michigan Autism and Communication Disorders Center (UMACC). Remaining participants were seen through clinic evaluations at UMACC. Children in the FW/T and WL projects entered the study at around 12 months of age and were assessed every several months or every 6 months (based on availability for repeated assessments primarily due to geographic locations) with the Toddler module of the ADOS (Lord, Luyster, Gotham, & Guthrie, in press). Children in the EDX project were assessed at ages 2 and 3 with the Pre Linguistic-ADOS (PL-ADOS; DiLavore et al., 1995), the previous version of the ADOS Module 1 and the ADOS-Toddler Module.
Participants were 347 males (76%) and 108 females (24%). The sample consisted of 121 children with autism, 71 children with PDD-NOS, 90 children with NS, and 173 children with TD based on best estimate diagnoses (See measures). The NS group consisted of children with a history of developmental delay including language impairment and/or intellectual disability without a diagnosis of ASD. Their diagnoses were primarily expressive language delay, mild intellectual disability, and nonspecific developmental delay. With regard to ethnicity, 73% of children were Caucasian, 20 % were African American, 4 % were multiracial, 1% were Asian, and 2% of parents did not provide information regarding ethnicity. Of the participants, 60 % were recruited from the EDX study, 14 % from the FW/T study, 25 % from the WL study, and 1% from the clinic at UMACC. Of the 455 children, 394 children were assessed at least twice. As a result, 635 observations of children 8–56 months of age, with a mean age of 28.37 months (SD = 0.42) were included in the present study. Of 635 cases (which includes the multiple observations), 501 cases were males and 134 cases were females. In regards to each diagnostic group, the percentage of males ranged from 66 to 89 % (TD, 66%; Autism, 86%; PDD-NOS, 89%; NS 73%). The ASD groups (autism and PDD-NOS) had significantly higher ratios of males to females than TD and NS groups (F = 11.7, p < .001). Nonverbal IQ scores ranged from 13 to 155 with scores in the different diagnostic groups as follows: Autism, M = 62.94, SD = 1.32; PDD-NOS, M = 72.5, SD = 2.12; NS, M = 77.18, SD = 2.26; TD, M = 113.11, SD = 1.3. No significant difference between males and females emerged for NIVQ scores; however, NVIQ scores differed significantly by diagnosis (F = 211.2, p < .001). In addition, even when children with TD were excluded, there were significant age differences by diagnosis (F = 129.5, p < .001), which was one of the reasons why we divided children into different cohorts (See Age Cohorts). Furthermore, males were significantly older than females (Males, M = 29.1, SD = 10.8; Females, M = 25.2, SD = 10.2; F = 14.807, p < .001). Because there were significant diagnostic group differences for gender, NVIQ, and age, we controlled for these factors in all of the analyses as covariates.
Because we were interested in looking at effects of age on RRBs, when sample size permitted, toddlers and preschoolers were further divided into even smaller chronological age cohorts. Thus, there were a total 6 age cohorts for each diagnostic group (Table 2). There were 27 cases under 12 months old of age and all of them were in the TD group with mental ages over 12 months. Even though they were younger than 12 months, we included these children to take advantage of the available data given the limited number of children in the TD group. When statistics were performed, children under 25 months (cohort 1 and 2) in the autism and PDD-NOS groups and 43–56 month-olds in the NS group were combined to obtain sample sizes large enough for the analyses. Each cohort included no more than one observation per child. Data for the TD group were only available from 8 to 30 months. NVIQ scores were also examined across different cohorts as follows: cohort 1, M = 85.63, SD = 20.12, range = 27–141; Cohort 2, M = 82.31, SD = 18.17, range = 47–128; cohort 3, M = 71.14 SD = 21.55, range = 13–123; cohort 4, M = 64.71, SD = 20.91, range 22–118; cohort 5, M = 66.65, SD = 22.62, range = 17–132; cohort 6, M = 59.19, SD = 21.57, range = 19–120. The NVIQ means were higher for younger cohorts than older cohorts even when typically developing children were excluded (F = 14.26, p < .001) even though the standard deviations and ranges for NVIQ scores were similar across the cohorts. This was why we controlled for NVIQ in all of the analyses along with gender and age as covariates.
A standard assessment battery was administered for each child after the IRB was approved and informed consent was obtained. The battery included the ADOS-Tor PL-ADOS, the ADI-R, and the Mullen Scales of Early Learning (MSEL; Mullen, 1995) or the Bayley Scales of Infant Development (BSID; Bayley, 1993). Testing was usually completed within a period of 2 weeks.
The ADOS is a semi-structured, standardized assessment of communication, social interaction and play for children who may have ASD. Children in this study were administered either the Pre-Linguistic ADOS (PL-ADOS; DiLavore et al., 1995) or an experimental version intended for children under 36 months of age (ADOS – Toddler Module; Lord et al, in press; Luyster et al., 2009). A standardized diagnostic algorithm can be calculated for each version, and established cut-off scores based on algorithm totals are used to differentiate children with autism, ASD, and NS or TD. For the present study, the focus was on the domain of RRBs.
Scores on the ADOS-T and PL-ADOS items range from 0 to 3. A score of 0 indicates that the particular behavior is not present, and ratings from 1 to 3 vary in severity based on both frequency of the behaviors measured and their interference with other behaviors. A higher score indicates more severe abnormality. The revised algorithms of the ADOS-Module 1 are different for children who use no words or some words during the observation (Gotham et al., 2007).
For the RRB algorithm totals, we used the no words algorithm for 235 cases (37% of the entire cases) when the children used no or fewer than 5 spontaneous words/word approximations during the ADOS. The some words algorithm was used for the rest of cases, all of whom used more than 5 different words. Children who used simple phrases who had been given the PL-ADOS, were also included here. The items in the algorithms were: stereotyped language, intonation of vocalizations, sensory interests, hand and finger mannerisms, complex mannerisms, and repetitive behaviors (See Table 1 for specific examples for each item). Stereotyped language and intonation of vocalizations were included in these algorithm subtotals on the basis of factor analyses that indicated they most often grouped with other RRBs (Luyster et al., 2009). In both no words and some words algorithms, scores of 3 on the ADOS protocols are converted to 2, and the highest of score of either hand and finger mannerisms or complex mannerisms is selected and then combined with the other three items. The no words algorithm included all the items mentioned above except stereotyped language. The some words algorithm included all the items except intonation of vocalizations. As a result, the maximum score for the RRB total was 8 (when a child received a “2” on either hand and finger mannerisms and/or complex mannerisms and on each of the three other items) and the minimum was 0.
Ratings for two items in the ADOS-T were modified to match the same items on the PL-ADOS; for sensory interests, a score of 2 was converted to 1 because the score for a 2 in the ADOS-T was equivalent of a rating of 1 in the PL-ADOS item; in the same way, ratings of 1 and 2 were converted to 0 and 1 respectively for complex mannerisms. The present study used the RRB totals from the ADOS Module 1 algorithms since we were able to maximize the number of identical items across the ADOS-T and PL-ADOS using the Module 1 algorithms. However, because of a concern that unusual intonation and stereotyped phrases are not typically considered RRBs, though they had loaded onto factors with RRBs, analyses were also carried out on raw RRB total scores without these items. This also could address another concern of no words and some words algorithms having a slightly different composition of RRB totals when the algorithm RRB totals are used (because stereotyped language is substituted by unusual intonation in the no words algorithm) because the raw RRB totals consist of the same items across children who received no words and some words algorithms. For the raw RRB totals, we added the scores of sensory interests, hand and finger mannerisms, complex mannerisms, and repetitive behaviors, which ranged from 0 to 8.
The MSEL is a developmental test intended for children from birth to 68 months. The MSEL was designed to assess children from birth through 68 months old. Because the MSEL does not provide separate VIQ and NVIQ scores, ratios of the two verbal subtests divided chronological age and multiplied by 100 and ratios of the two nonverbal subtests divided by chronological age and multiplied by 100 were used (Richler et al., 2007). NVIQ scores were used as a measure of the child’s overall intellectual ability because they tend to be more stable over time among children with autism than verbal or full scale IQ scores (Howlin, Goode, Hutton, & Rutter, 2004). The MSEL was administered to all children except 60 cases in the TD group seen as part of the EDX study.
The BSID was used for 60 out of 188 cases in the TD group. The BSID was designed to measure the developmental functioning of infants and toddlers. Because the BSID does not provide separate NVIQ and VIQ scores, full scale IQs on the mental scale were used as a proxy for NVIQ scores.
For children in the FW/T project, all available data, including research diagnosis history over the most recent months and chart notes, were used by two examiners to generate consensus best estimate “working diagnoses.” The most weight was given to most recent diagnosis and “blind diagnoses” made by an examiner not familiar with the child (See Luyster et al., 2009). For children in the EDX study, an experienced clinical researcher used the ADOS and ADI-R and observations during the full assessment to generate independent best estimate diagnoses of autism, PDD-NOS, or a nonspectrum disorder (See Lord et al., 2006). All of the examiners were trained to meet the standard requirements for research reliability on both instruments. Training involved 5 days of didactic lectures and hands-on practice in small groups with the new examiner then practicing partial and then full administrations until he or she was reliable in terms of exact agreement on 80% of items for 3 consecutive scorings including one scoring of a standard videotape and one administration. Consensus coding was conducted approximately every fifth administration to ensure maintenance of reliability. All of the ADOS administrators were blind to the diagnostic status of the children at each evaluation except 66 children from the FW/T study. These children were seen, on an alternating basis, by a combination of a familiar clinician and a new clinician who was blind to their previous performance and tentative diagnosis (See Luyster et al., 2009).
First, with SPSS 16.0, a Generalized Estimating Equations (GEE) logistic regression for repeated measures was used to examine differences in the prevalence of RRBs (percentage of children who showed at least one RRB indicated by getting a score of more than 0 on any of 6 items) among different diagnostic groups. Another GEE analysis was used to assess differences in the severity of RRBs (RRB total scores on the algorithms) among different diagnostic groups.
To examine the predictability of RRB totals, a hierarchical regression analysis was carried out using children’s earlier RRB totals (from 8 to 30 months) as one of the predictors for the same children’s later RRB totals (from 31 to 56 months). For this analysis, a subsample of 72 children at least a RRB total score of one for either cohort 1, 2, or 3 (Time 1) and one score for either cohort 4, 5, or 6 (Time 2) were selected. Step #1 included NVIQ scores, age, gender, and diagnosis at Time 2. Step #2 included RRB totals at Time 1 predicting RRB totals at Time 2. The stability of RRBs was also examined by performing another GEE analysis to see if RRB totals were independent of age for children with ASD.
To examine how the association between RRB totals and NVIQ varied by age, a GEE model was performed for each age cohort independently. For all of these GEE models, rate ratios (RR) were calculated for each covariate in the model. Additionally, we also performed separate analyses for each diagnostic group independently. For the TD group, a hierarchical regression analysis was performed because all data were cross-sectional, and for the rest of the groups, GEE analyses were performed because of repeated measures.
Last, based on the hypothesis that relationships between the prevalence of RRBs and diagnosis, age, and NVIQ might differ for each type of RRB, a GEE logistic regression for each individual item score was performed, and odds ratios (OR) were calculated for each covariate in the model. One of the results from the item analyses yielded a significant association between RRBs and gender (See Results); therefore, gender was included as one of the covariates in all of the analyses mentioned above.
Not surprisingly, a GEE logistic regression indicated that there was a significant main effect of diagnosis (χ2 = 53.34, p < .001) on RRB prevalence while controlling for age, NVIQ, and gender. As shown in Table 3, RRBs in children with autism and PDD-NOS were significantly more prevalent than those in children with NS and TD at all ages (pairwise comparisons with p < .05). Almost all of children with ASD scored more than 0 on at least one of the RRB items. Prevalence rates of at least one RRB ranged from 96–100% and 90–97% by age cohorts for autism and PDD-NOS groups respectively; See Table 3.
Even though RRBs in children with ASD were significantly more prevalent than in children with NS and TD, RRBs were relatively common in the NS group as well. When prevalence was compared across all of the diagnostic groups, children with NS showed positive scores (score of 1, 2, or 3 vs. 0) on one RRB item on average (mean (M) = 1.01, standard deviation (SD) = 1.04). The TD group showed positive scores on fewer than one item (M = 0.71, SD = 0.81). Children with autism and PDD-NOS showed positive scores on more than two RRB items on average (Autism, M = 2.83, SD = 1.02; PDD-NOS, M = 2.15, SD = 1.17).
When the severity of the RRBs in the four diagnostic groups was compared across age groups, the GEE analysis indicated that there was a significant main effect of diagnosis (χ2 = 97.24, p < .001) while controlling for NVIQ, age, and gender. As hypothesized, consistent with the findings from the prevalence analyses, both groups with autism and PDD-NOS showed significantly higher RRB totals than children with NS and TD at all ages for which they were available (See Figure 1). When the raw RRB total scores without language items were used for these analyses, the results remained the same; there was a main effect of diagnosis (χ2 = 49.03, p < .001) while controlling for NVIQ, age, and gender.
Because an interaction effect of diagnosis and age (χ2 = 38.8, p < .01) emerged as well, we performed more analyses to examine the differences within the ASD groups (autism vs. PDD-NOS) for each age cohort. For this analysis, cohort 1 and 2 were combined to achieve sample sizes large enough for statistical analyses. Though the autism and PDD-NOS groups showed the biggest difference in RRB totals for cohort 1 (12–18 months) originally, when the difference was measured with cohort 1 and 2 (12–24 months) combined as one cohort, the difference was no longer significant. However, the differences in RRB totals between autism and PDD-NOS were significant for the remaining cohorts (25–56 months) confirming the age by diagnosis interaction (pairwise comparisons with p < .05). However, when the raw totals were used without language items, the differences between these subgroups were no longer significant though the raw RRB totals for children with autism were still higher than those for children with PDD-NOS across age groups.
Across all ages, RRB totals for children with autism (M = 4.39; SD = 1.77) were almost 3 times higher (RR = 2.93, p < .01) than RRB totals for children with NS (M = 1.40, SD = 1.5); RRB totals for children with PDD-NOS (M = 3.03, SD = 1.84) were more than two times higher (RR = 2.26, p < .01) than RRB totals for children with nonspectrum disorders.
As hypothesized, unstandardized regression coefficients indicated that RRB totals at Time 1 (12–30 months) were a significant predictor of RRB totals at Time 2 (31–56 months) for children with autism, PDD-NOS, and NS while controlling for NVIQ scores, diagnosis, age, and gender at Time 2 (β = .61, p < .001). The model with all the predictors including the earlier RRB scores accounted for 46% of the variance in the later RRB totals (R2adj. = .46, p < .001). When the raw RRB totals without language items were used, results were identical.
We expected that age would be independent of RRB scores because ADOS algorithm items were originally chosen specifically to distinguish ASD from other diagnoses at different language levels and age. As expected, the severity of RRBs for the autism and PDD-NOS groups was independent of age. RRB totals were similar for all ages in both groups, suggesting the stability of RRB scores over time. The NS group showed a similar pattern as well. However, age was significant for the TD group indicating that RRBs in children with TD became less severe with increasing age (β = −2.81, p < .05).
As hypothesized, lower NVIQ scores predicted higher RRB scores for all of diagnostic groups combined across all ages (χ2 = 24.86, p < .01). To examine how the association between RRBs and NVIQ differed by age, we performed separate analyses for three age groups: children under 25 months, 25–36 months, and 37–56 months. When the relationship between RRBs and NVIQ scores was examined by these groups, as predicted, the relationship was not significant for children under 25 months but it was for children over or equal to 25 months (χ2 = 7.55, p < .01 for 25–36 months; χ2 = 13.32, p < .01 for 37–56 months).
To our surprise, we also found that the impact of NVIQ on the RRB totals differed by diagnosis when each diagnostic group was analyzed separately. Interestingly, NVIQ was not a significant predictor of RRBs for children with autism at any age. However, NVIQ was a significant predictor of RRBs for children with PDD-NOS (χ2 = 6.53, p < .05), NS (χ2 = 12.12, p < .01), and TD (β = −.17, p < .05) over or equal to 25 months, but not for those under 25 months. Thus, children with higher NVIQ scores showed lower RRB scores for all children except for children with autism and children under 25 months in all diagnostic groups. In addition, the associations between the raw RRB totals without language items and NVIQ scores by different age groups were consistent with the results mentioned above; raw RRB totals were not significantly related to NVIQ scores for the youngest cohorts, but were significantly related to NVIQ for the older cohorts (χ2 = 19.99, p < .001 for those from 25 to 36 months, χ2 = 38.18, p < .001 for those from 37 to 56 months). The associations between the raw RRB totals and NVIQ scores by diagnosis were nearly identical to the results using the algorithm RRB scores with significant associations emerging only between the raw RRB totals without language items and NVIQ scores for children with PDD-NOS (χ2 = 15.03, p < .001), NS (χ2 = 11.41 p < .01) and TD (β = −.2, p < .05) and not for children with autism.
GEE logistic regression was run for each item separately to examine the heterogeneity in the associations between the prevalence of each RRB item with NVIQ and age while controlling for diagnosis and gender. For the analysis of stereotyped language, only those children who received “Some Words” algorithms (thus, using phrase speech during the ADOS) were included.
Gender was a significant predictor for repetitive behaviors; girls were more likely to show these behaviors than boys. (OR = 0.7, p < .01). No other gender effects or interactions with gender were significant.
As expected, there were differences in the associations between each RRB and age. Age was not associated with the prevalence of intonation of vocalizations, sensory interests, hand and finger mannerisms, and complex mannerisms across any diagnosis, suggesting stability over time (See Figure 2). However, age was a significant predictor of prevalence rates for repetitive behaviors and stereotyped language; the prevalence rates of these items for the children with autism, PDD-NOS, and NS increased as age increased, while the rates of the TD group decreased over time (See Figure 2 and Table 4).
As hypothesized, the association between RRBs and NVIQ also differed by types of RRBs. NVIQ was a significant predictor for sensory interests, hand and finger mannerisms and complex mannerisms (See Table 4). These three items were significantly more prevalent in children with lower NVIQ scores than children with higher NVIQ scores. For example, for each additional 1 point of NVIQ, the odds of showing sensory interests across cohorts diminished by 3% (OR = 0.97, p < .01). The prevalence rates of repetitive behaviors and intonation of vocalizations were independent of NVIQ. Even within the group of children who had phrases, stereotyped language was significantly more prevalent in children with higher NVIQ scores than in children with lower NVIQs.
Consistent with the results from RRB domain analyses, individual RRB item analyses showed that each type of RRBs for children with ASD were more prevalent than for children with NS and TD for all items except stereotyped language, which was independent of diagnosis (See Table 4). There were diagnostic differences in the prevalence rates of all the other items with the autism group showing the highest rates consistently across all ages (see Figure 2). Sensory interests, hand and finger mannerisms, complex mannerisms, and repetitive behaviors were significantly more prevalent in children with ASD than children with NS and TD in all ages. For example, compared to the NS group, the autism group was nearly 13 times more likely (OR = 12.6, p < .001), and the children with PDD-NOS were about 4 times more likely to show repetitive behaviors (OR = 3.95, p <.001). The autism group was nearly 11 times more likely (OR = 10.8, p <.001), and the PDD-NOS group 4 times more likely than the NS group to show sensory interests (OR = 4.2, p <.001).
Using longitudinal data from brief observations, we were able to find stable diagnostic differences in both prevalence and severity of RRBs measured during the ADOS among 655 observations of children (from 455 toddlers and preschoolers ranging from 8 to 56 months of age) with autism, PDD-NOS, NS and TD. The diagnostic differences in RRBs found in the current study suggest the importance of RRBs to early diagnosis of ASD, in line with previous studies that have shown similar results using parental interviews and questionnaires (Watson et al., 2007, Richler et al. 2007; Lord et al, 2006). The results of this study also extend the findings of Morgan, Wetherby & Barber (2008) who reported diagnostic differences in RRBs among children with ASD, developmental delays, and typical development from 18 to 24 months based on systematic observations.
The present study adds to the growing body of literature showing that semi-structured observations which occur in a brief time period can successfully provide an opportunity for the assessment of RRBs. In fact, the diagnostic differences in RRBs found in the present study using the ADOS highlight the importance of using observational data with very young children because of evidence that parents may not notice RRBs in very young children (Chawarska, Klin, Paul, & Volkmar, 2007). While parents might have more opportunities to observe RRBs in their children, they may also have more difficulties in judging the abnormality of such behaviors compared to clinicians. In fact, though our results indicated that most children with ASD showed RRBs during the ADOS, the prevalence analyses revealed that there were 11 cases in the PDD-NOS group who scored 0 on the RRB totals in the ADOS. However, it was found that all children with autism and those with PDD-NOS had at least one RRB at the time of assessment when RRB scores from both the ADI-R and ADOS, which were administered within the same week, were taken into account. All of the children with ASD who scored 0 on the RRB totals in the ADOS scored more than 0 on at least one of the current RRB item in the ADI-R.
Importantly, our findings with very young children suggest that when data from observations and parent interviews are combined, RRBs are almost always present in ASD, including in children given PDD-NOS diagnoses. This supports the idea of having RRBs as a requirement for the diagnosis of broader ASD (including PDD-NOS) in new diagnostic frameworks under development when information from parent report and observation are both taken into account.
The results of the present study on the differences in the prevalence and severity of RRBs between the ASD (autism and PDD-NOS) and NS groups are of practical significance. Even though RRBs in the ASD groups were consistently more prevalent and higher in severity than the NS and TD groups consistent with past studies (Bodfish, Symons, Parker, & Lewis, 2000; Watson et al., 2007; Richler et al., 2007), RRBs were relatively common in the NS group. On average, children with NS showed at least one type of RRBs during an ADOS session while ASD groups showed 2–3 types of RRBs on average. Although current diagnostic guidelines, such as DSM-IV (APA, 1994), require the presence of a single RRB for a diagnosis of ASD, clinicians should be aware that the presence of any single RRB alone does not distinguish children with ASD from those with other non-spectrum disorders. Nevertheless, though the appearance of RRBs is not unique to ASD, particularly in very young children, the presence of a RRB coupled with social and communication deficits increases the likelihood of a stable diagnosis of ASD over years to come (Lord et al., 2006).
In the present study, the likelihood of having an RRB was the same for both autism and PDD-NOS groups; though the rated severity of these RRBs was higher in children over age two with autism than PDD-NOS diagnoses. This is consistent with the past literature that has shown that no single behavior or factor differentiated PDD-NOS from autism or Asperger’s syndrome (Klin et al., 2005). This is also in line with previous research using the ADI-R indicating that individuals with autism had significantly higher scores than those with PDD-NOS in the repetitive sensory-motor behaviors factor (Georgiades et al. 2007). These findings support the concept of a single category ASD differentiated by severity, rather than separate subgroups of ASD such as PDD-NOS or autism..
Consistent with previous findings on the trajectory of RRBs (Richler et al., 2010), the severity of RRBs was independent of age such that the RRB totals were stable over time in both ASD and NS groups. Even though these results were expected because ADOS items were originally selected to be differentially diagnostic at different language levels and ages, it is interesting that RRB totals for typically developing children were still associated with age such that their scores decreased over time, showing that differentiating children with ASD from typically developing children becomes easier as they get older. Early RRB scores in children with ASD and NS under 30 months predicted their later RRB scores indicating that RRBs that are severe in toddler years often remain severe over the course of development into preschool years.
The current study showed that IQ is more closely related to the manifestation of RRBs in older children with PDD-NOS, NS and TD than very young or more severely affected children. Lower NVIQ scores were not associated with higher RRB scores in children with autism (regardless of their age) and children under 25 months (regardless of their diagnosis). This might be because of differences in the NVIQ distributions by cohort. In the present study, the older cohorts had the fewer children with lower NVIQ scores than younger cohorts which might have minimized the effect of NVIQ scores on RRB totals for younger cohorts. However, another possibility is that older children with higher IQs may have more interests and abilities that foster participation in less repetitive activities based on similar findings by Bishop et al. (2007). In toddler years, these options may be limited since these very young children have not yet acquired the chance to develop creative play and more productive activities regardless of the level of their cognitive functioning. As children enter preschool years, they would be more likely to be exposed to novel environments and activities that would facilitate the development of more elaborate play. In the same way, since children with autism have more severe levels of impairment in their social and communication functioning compared to those with PDD-NOS, a milder form of ASD, they may have fewer opportunities to develop more productive, non-repetitive activities through social interactions. These findings have important implications for treatment. One of the goals of early intervention may be to alter the course of developmental trajectories of RRBs by providing alternative behaviors that are equally motivating.
As we hypothesized, different patterns of association between the prevalence of RRBs and child characteristics were observed. Interestingly, three items, sensory interests, hand and finger mannerisms, and complex mannerisms showed similar patterns of associations with NVIQ, age, and diagnosis, distinct from the other RRB items. Surprisingly, the prevalence of these three items was associated with NVIQ but was stable over time, which is similar to findings from parent reports on repetitive sensory-motor behaviors (RSMB) in past studies (Richler et al., 2010; Hus et al., 2007; Lam, Bodfish, & Piven, 2008). Turner (1999) described RSMBs as “low-order” behaviors related to developmental delays, which was also true for these three items that were more prevalent in children with lower NVIQ scores. Other items, stereotyped language, intonation of vocalizations, and repetitive behaviors differed in terms of their associations with NVIQ, age, diagnosis, and gender. Repetitive behaviors and intonation of vocalizations were independent of NVIQ, and stereotyped language and repetitive behaviors became more prevalent with increasing age, which suggests that these items in the ADOS might capture more “higher-order” behaviors (Turner, 1999). This can be because, unlike in the ADI-R and more advanced Modules of the ADOS where the item, repetitive behaviors, is distinguished from compulsions and rituals, repetitive behaviors in the earlier Modules of the ADOS encompass a very broad range of behaviors including both Repetitive Sensory-Motor and Insistence of Sameness behaviors (e.g. repetitive nonfunctional use of toys; insistence on unusual routines). These results support the idea of different classes of RRBs and heterogeneity among different types of RRBs in associations with intellectual abilities and age (Turner, 1999). The results also support that examining RRB subtypes and their relationships with other characteristics (e.g. NVIQ, age, gender, etc) can help us to better identify specific types of RRBs predictive of diagnosis. Stereotyped language did not differentiate between diagnostic groups likely because of the still limited language levels of the children with ASD. A main effect of gender emerged for repetitive behaviors. However, since there was no significant effects of gender on NVIQ scores and on the other RRB subtypes as well as on the RRB total scores, the possibility of gender differences in subtypes of RRBs needs further exploration in future studies.
The present study assessed children with TD only up to 30 months in comparison to the other groups of children who were assessed up to 56 months. Had the TD group been followed at older ages, we would have been able to compare the rest of the groups to the TD group from age 31 to 56 months. Furthermore, children in the present study were divided into 6 cohorts, and the time interval for each cohort was about 5 months. Since rapid developmental changes occur during toddler and preschool years, it will be important for further studies to examine RRBs in those early years with shorter time intervals such as 1 or to 2 months to capture the rapid developmental changes in more detail.
One other limitation was that, in order to maintain a sufficiently large sample, we combined samples who received two different algorithms even though there was a slight difference in the composition of the RRB totals between these two algorithms (intonation of vocalizations was substituted for stereotyped language for the no words algorithm). In addition, one could argue that the language items can be considered not as central as the other items to the concept of RRBs. To address these concerns, we performed the same set of analyses using the raw RRB totals without language items for the severity of RRBs. However, when the raw totals were used, all of the results remained the same although the differences in the RRB raw totals between subgroups of ASD were not significant anymore. This confirms our belief that RRB algorithm totals can validly represent the severity of RRBs to test our hypotheses. On the other hand, it will be interesting for the future research to examine nonverbal and verbal samples separately if possible to investigate the role of the child’s verbal level on RRBs in regards to items related to language.
Furthermore, not all aspects of RRBs can be assessed in brief observations. Distinctions between insistence on sameness and other RRBs was not possible in these young because frequency, content or quality of a behavior were not coded sufficiently specifically. Parent and other caregiver (e.g. teacher; therapist) reports remain critical in order to capture broader aspects of RRBs. Semi-structured observations with very young children suspected of having ASDs can provide more information than even we originally assumed, but still must be complemented by detailed information from people who know them well.
This heterogeneity in RRBs found even in these very young children during relatively brief observations in the current study is consistent with past studies based on parent interview, suggesting researchers should attend to emerging differences among RRB items that could easily be missed by grouping multiple items under one single domain (Lam, Bodfish, & Piven, 2008; Cuccaro et al, 2003; Szatmari et al., 2006; South et al, 2005). Further evidence about the heterogeneity in RRBs and their developmental trajectories may hold important clues for etiology, pathological mechanisms and treatment of RRBs.
We gratefully acknowledge the help of Somer Bishop, Kaite Gotham, Whitney Guthrie, Rhiannon Luyster, and Jen Richler as well as the families who participated in this research. We also acknowledge the supports from NRSA, NIMH, Autism Speaks, and Simons Foundation.
Grant Sponsor: National Institute of Mental Health (NIMH), National Institute of Child Health and Human Development (NICHD), Department of Education (DOE), & Simons Foundation
Grant Number: R01 MH 066496, R01 MH081873, H324C030112
Disclosure: C. Lord receives royalties for the ADOS; profits related to this study were donated to charity.