This meta-analysis updated evidence regarding sensory over-responsivity (SOR), under-responsivity (SUR) and seeking symptoms in individuals with autism spectrum disorders (ASDs) relative to typical controls and those with other conditions. Fifty-five questionnaire studies included 4606 individuals with ASD. Moderators tested were age, IQ, male ratio, matching group, and self-report. Compared to typical controls, effect size was large and significant for SOR, SUR, and Seeking but heterogeneous. For Seeking, age, IQ and self-report were significant moderators. Compared with developmental disorders (DDs) groups, effect size was significantly positive for SOR and Seeking; whereas compared with other clinical groups, only SOR was significant. These findings highlight the core nature of sensory symptoms in ASD and particularly SOR. Explanatory factors are yet to be revealed.

Keywords: Autism spectrum disorders; Sensory symptoms; Sensory profile; Meta-analysis

Although sensory symptoms have been associated with autism spectrum disorder (ASD) since autism was first described (Kanner [64]), they were not considered a core component (DSM-IV, American Psychiatric Association [2]; DSM-IV TR, American Psychiatric Association [3]) until 2013, despite research reporting that 45% to 95% of individuals with ASD demonstrated sensory symptoms (Tomcheck and Dunn [119]; Baranek et al. [9]) and almost all first-hand accounts of ASD included sensory issues (Chamak et al. [19]; Grandin [50]). Over the past two decades, professionals in numerous disciplines have increasingly recognized the sensory features of ASD (DuBois et al. [32]). These features often have been referred to as impairments in sensory modulation in which an individual has difficulty regulating and organizing the type and intensity of behavioral responses to sensory inputs to match environmental demands (Miller et al. [89]). These can manifest in response to touch, sight, sound, taste, smell and movement, with many individuals presenting several types of symptoms. Sensory features can be classified into three patterns known as, sensory over-responsivity (SOR), sensory under-responsivity (SUR), and sensation seeking (Miller et al. [89]) with many individuals with ASD showing more than one sensory pattern.

In 2009, we published a meta-analysis examining symptoms of sensory modulation in individuals with ASD (Ben-Sasson et al. [14]). While several authors published literature reviews or critical reviews on the topic prior to that time (e.g. Rogers and Ozonoff [104]), this was the first meta-analysis of the subject. It examined the magnitude of sensory symptoms overall and specifically for each of the three sensory modulation patterns, SOR, SUR, and Seeking in individuals with ASD. It investigated whether these findings were consistent across studies and could be explained by various moderators. We identified 14 studies that used parent questionnaires to compare sensory modulation in ASD versus typically developing (TD) and/or other developmental disorders (DDs) groups. We found strong evidence that, compared to TD groups, children with ASD displayed atypical patterns of SOR, SUR, and Seeking. Of the ASD–TD comparisons, mean effect size was high for all types of sensory scores (d = 0.82–2.14), but larger for SUR. SOR and Seeking symptoms were highest in the 6–9 years old group and decreased thereafter. The three sensory patterns showed different relations relative to severity of ASD and relative to whether they were compared to a mentally versus chronologically age matched control group supporting their differentiation. The distinct relations of these three sensory patterns with developmental, psychological, and maturational indices in ASD was also supported in more recent reviews (Glod et al. [48]; Schauder and Bennetto [110]). While the 2009 review contributed to our understanding of sensory features, findings were limited by the small number of available studies, with only four studies including a DD comparison group.

Since our 2009 paper, there has been increasing recognition of the prominence of sensory features of ASD. In 2013, DSM-5 added both sensory hypo-reactivity (referred to as SUR hereafter) and sensory hyper-reactivity (referred to as SOR) to the diagnostic criteria for ASD (American Psychiatric Association [4]). Sensation seeking was already captured in DSM-IV within the rubric of unusual sensory interests. Since the DSM-5 publication, research on sensory features in ASD has skyrocketed, as shown by a simple PubMed search (October 2018) using the terms sensory AND autism, limited to humans (see Fig. 1). Several systematic reviews about sensory features and ASD have been published recently, mostly highlighting correlates of sensory symptoms such as adaptive behavior and attention impairments (Dellapiazza et al. [31]), psychological symptoms (i.e., emotional, affective and behavioral symptoms; Glod et al. [48]) and participation in daily functioning (Ismael et al. [61]). Additional recent literature reviews focused on bridging inter-disciplinary methodologies towards sensory symptoms in ASD at the assessment and neurobiology levels. This includes a literature review by Hazen et al. ([56]), a review integrating symptom and neural literature in this area (Schauder and Bennetto [110]), and a scoping review about measures of sensory symptoms in adolescents and adults (DuBois et al. [32]). None of the recent reviews used a meta-analysis design.

Graph: Fig. 1 The number of articles between 1963 and 2017 with autism and sensory (filtered by humans) in PubMed

There has been increasing acknowledgement that these sensory symptoms are neurobiologically grounded since our 2009 publication, with individuals with ASD showing different neural processing compared to peers (Cascio et al. [17]; Chang et al. [20]; Green et al. [55], [54]; Hazen et al. [56]; Madsen et al. [80]; Marco et al. [81]). Review of fMRI research points to abnormalities in the activation patterns across sensory brain areas, and MEG evidence indicates the presence of maturational and lateralization atypicalities in sensory processing in ASD (Schauder and Bennetto [110]). Nonetheless, the correspondence between observable sensory symptoms and neurobiological atypicalities is yet to be determined.

Evidence attests to the association of sensory symptoms with developmental and other facets of the disorder, as well as the likelihood of co-morbid conditions. Sensory symptoms are associated with compromised adaptive performance, participation (Dellapiazza et al. [31]; Tomchek et al. [120]) and performance of activities of daily living (Dunn et al. [38]; Ismael et al. [61]), including eating (Cermak et al. [18]; Mazurek et al. [86]; Zobel-Lachiusa et al. [137]) and sleeping (Mazurek and Petroski [85]). Sensory symptoms have been associated with greater severity of restricted and repetitive behaviors (Chen et al. [21]; Gabriels et al. [46]), and reduced social functioning (Glod et al. [48]; Hilton et al. [59]). They are also associated with increased affective symptoms (Ben-Sasson et al. [13]; Glod et al. [48]), particularly anxiety (Green et al. [52]; Lane et al. [73]). In addition, sensory symptoms have been shown to interplay with communication and language deficits (Glod et al. [48]; Tomchek et al. [120]) and attention competencies (Dellapiazza et al. [31]; Glod et al. [48]). Further, sensory symptoms negatively impact family life (Bagby et al. [7]; Kirby et al. [71]). It is possible that the inherent involvement of sensory processing in every interaction and learning process during development explains its pervasive impact upon functional impairment and well-being. Given that adaptive living skills contribute to a person's ability to participate successfully in the environment, it is critical to fully understand sensory processing and make it a high priority for research and practice.

Sensory symptoms and specifically SOR also appear early in other types of developmental and psychopathological conditions such as ADHD (Cheung and Siu [22]), intellectual disability (Gal [47]), Fragile X (Baranek et al. [10]), anxiety disorders (Conelea et al. [26]), and obsessive compulsive disorder (OCD; Lewin et al. [75]). This raises questions regarding their distinct manifestation in ASD, both in magnitude and nature. Aside from their unique co-occurrence with social–communication symptoms in ASD, it is unclear how much these symptoms are related to developmental level. There is evidence associating higher mental capacity with fewer sensory symptoms in young children with ASD (Baranek et al. [9]) as well as adults (Kargas et al. [65]). In our 2009 review there was not sufficient IQ data to test this moderator across studies and sensory patterns. We hypothesized that there would be fewer significant and smaller effect sizes between ASD and other clinical groups compared to TD groups, as well as moderating effects related to developmental level.

There is conflicting evidence regarding the course of sensory symptoms in ASD throughout the life span. Age related changes in sensory symptoms are important indicators for determining their role in early identification, maturational mechanisms and for allocating services addressing sensory features. Some studies found a reduction in sensory symptoms in adolescents and adults compared with children with ASD (Kern et al. [70]; Leekam et al. [74]), some showed an increase in symptoms with age (Talay-Ongan and Woods [114]), while others showed stability throughout childhood (McCormick et al. [87]; Rogers et al. [103]). In our previous meta-analysis, we found a non-linear course of symptom development, the 6–9-year-old age group presented with greater SOR and Seeking effect sizes compared to younger and older age groups. Note that in 2009 there were many more samples with mixed age groups (e.g., sample of children and adults) to rely upon, hence the recent evidence is of value for examining distinct age related changes. Sensory symptoms may decrease with age due to neuro-hormonal changes and/or the establishment of coping strategies. The inconsistency in findings calls for testing the moderating effects of age in explaining differences between groups.

Gender may also play a moderating role as recent research suggests gender differences in severity of repetitive and restricted behaviors (Antezana et al. [5]; Van Wijngaarden-Cremers et al. [123]). In a meta-analysis, females with ASD over the age of 6 years compared to males showed a lower severity in the repetitive and restricted domain (Van Wijngaarden-Cremers et al. [123]). Other evidence in adolescents with ASD points to an increased likelihood of specific types of repetitive and restricted behaviors in males and others in females. Given the DSM-5 inclusion of sensory symptoms in ASD diagnotic criteria, we tested for gender differences in sensory symptoms.

In a recent scoping review focused on 66 studies of sensory processing of adolescents and adults with ASD, DuBois et al. ([32]) reported that proxy reporting was the most prevalent (78.7%) method of measurement. The Adolescent/Adult Sensory Profile was the most frequently used measure (71.1%) and was used as self-report and adapted for proxy report. While observation tools and physiological measurement were included in a few of the recent studies, overall there was too much variability in the conceptualization of the target behaviors for the quantitative demands of a meta-analysis. Thus, we decided to continue limiting the current meta-analysis to questionnaire-based methodology.

To summarize, 10 years have passed since the publication of the first meta-analysis examining the magnitude of sensory symptoms in individuals with ASD. During this time, sensory symptoms have become part of the diagnostic criteria for ASD and the number of new studies including clinical groups with and without ASD has increased tremendously (DuBois et al. [32]; Glod et al. [48]; Schauder and Bennetto [110]). Further, the new DSM-5 ASD criteria groups all ASD subtypes together potentially impacting the nature of samples studied. As such, it is critical to examine current research and update the meta-analysis to understand the factors that moderate sensory processing in individuals with ASD more fully. Some of our previous moderators were no longer relevant such as percent of autism diagnosis in sample (e.g., autism category omitted from DSM-5) and some became relevant with more studies reporting them (e.g., male ratio, IQ). The objectives of the meta-analysis were: (1) to characterize the magnitude of sensory symptoms (i.e., SOR, SUR, and Seeking) in ASD compared with typical and clinical samples and (2) to identify the contribution of demographic and methodological moderators to the variability in findings across studies. We hypothesized that there would be fewer significant and smaller effect sizes between ASD and other clinical groups compared to effect sizes relative to TD groups. Furthermore, we expected to find moderating effects related to developmental level.

This study adhered to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA-P: Moher et al. [91], [92]) including the careful definition of selection criteria and data points of interest.

The inclusion criteria for the meta-analysis were studies in which the investigators: (a) used or administered a sensory processing questionnaire, (b) compared an ASD group to another group, (c) measured at least one of the three sensory patterns SOR, SUR or Seeking distinctly and across several sensory modalities, and (d) provided sensory scores from which we could gather means and SDs for all groups either in the published paper or the researchers provided them upon request. Excluded were studies (a) of animals, (b) evaluating sensory symptoms only in family members of individuals with ASD, (c) measuring sensory traits only in the general population, (d) with single case research design, and (e) reporting of sensory scores which mix SOR, SUR and/or Seeking. The current meta-analysis relied for the most upon similar search procedures, research inclusion criteria and consisted of the 14 studies included originally in the 2009 meta-analysis. Note that self-report studies were included in the current meta-analysis as they were not available in 2009.

Three search methods were used and further described below: (a) database search, (b) reference list review, and (c) direct contact with authors to identify grey literature. To minimize risk of publication bias the search was expanded beyond journal publications to theses and unpublished data. The literature search was conducted by a research assistant under the guidance of a librarian with expertise in meta analyses. The first author verified search results independently. Any questions or conflicts regarding study eligibility were discussed and resolved among the authors of the paper. See Fig. 2 for the filtering process of articles.

Graph: Fig. 2 Flow of article search

• The database search involved the following keywords entered for identifying relevant published studies:
• Population term autism OR Asperger OR pervasive developmental disorder OR autistic disorder OR autism spectrum disorder OR autistic syndrome disorder.
• Sensory term sensory OR reactivity OR response OR auditory OR tactile OR vestibular OR oral OR hypo-responsivity OR hyper-responsivity OR hyper-reactivity OR hyper-responsive OR hypo-responsive OR seeking.
• Descriptor term processing OR integration OR modulation OR regulation OR profile OR symptom OR unusual OR hypo OR hyper OR over OR under.
• The databases searched were EBSCO (which combines Medline/PubMed and CINAHL), Web of Science, PsychInfo and ProQuest Dissertations and Theses. The search was restricted to studies published in English, with humans (PsychoInfo), and June 2007 to June 2018. This resulted in 4351 articles which were imported into Covidence data management system (Veritas Health Innovation [124]) for filtering to select qualifying, non-duplicate references. In cases of duplicate authors, they were contacted to ensure that the samples did not overlap between papers. In case of overlap, only the study with the largest quantity of statistical information needed for this meta-analysis was included.
• Reference lists of literature reviews and relevant papers were screened for additional potential papers. Two papers included in the meta-analysis were identified through this process (Matsushima and Kato [82]; Mattard-Labrecque et al. [83]).
• Authors from main labs who study sensory processing in ASD were approached for any relevant unpublished data they had to minimize publication bias. This resulted in three additional studies (Katz-Zetler et al. [67]; Lane [72]; Tavassoli [115]). We also contacted authors of studies that were missing means and/or SDs for the sensory questionnaires. This enabled the inclusion of 15 studies of the 41, which would not have been included without author communication (see Table 1).

Study characteristics

iFor these studies raw data was sent by authors for sensory scores and/or demographic information iiThe reason for matching data is that it reflects the whole sample, no specific data for the separate groups was given for these variables (only in sensory variables) iiiDemographics are based on a larger sample size than the sensory data available ivOne subgroup or one wave was selected for the meta-analysis based on sensory scores available or age group with the least studies vIQ = DQ viIQ computed from MA by authors or by us viiIQ based on VMA viiiLangQ based on expressive and receptive Mullen DQs ixLevel of functioning 1 = high functioning and 2 = low functioning. This was based on description of inclusion criteria and/or self-report measure applied and not on mean IQ scores 1Mullen Scales of Early Learning ([93]), 2Bayley Scales of Infant Development-Second Edition ([11]), 3Wechsler Adult Intelligence Scale-Third Edition ([130]), 4Wechsler Intelligence Scale for Children-Third Edition ([129]), 5Wechsler Intelligence Scale for Children-Fourth Edition ([132]), 6Differential Ability Scales-II (Elliot [40]), 7Leiter International Performance Scale-Revised (Roid and Miller [106]), 8Wechsler Preschool and Primary Scale of Intelligence-Third Edition ([131]), 9Wechsler Preschool and Primary Scale of Intelligence-Revised ([128]), 10Vineland Adaptive Behavior Scales (Sparrow et al. [112]), 11Wechsler Abbreviated Scale of Intelligence ([133]), 12British Picture Vocabulary Scale (Dunn et al. [36]), 13Preschool Language Scale. 4 (Zimmerman et al. [136]), 14Wechsler Intelligence Scale for Children-unspecified, 15Tanaka–Binet Intelligence Scale-V (Matsubara et al. [84]), 16Kyoto Scale of Psychological Development (Ikuzawa et al. [62]), 17The Raven's Standard or Coloured progressive matrices ([99], [100]), 18Parent Estimated Developmental Age (PEDA; Ausderau et al. [6]), 19Kaufman's Brief Intelligence Test-Second edition (Kaufman and Kaufman [68]), 20Stanford–Binet intelligence scales. 5 (Roid [105]) aSensory Profile (Dunn and Westman [37]; Dunn [34]) bSensory Sensitivity Questionnaire-Revised (Talay-Ongan and Wood [114]) cInfant/Toddler Sensory Profile (Dunn [34]) dShort Sensory Profile (McIntosh et al. [88]) eSensory Experiences Questionnaire: Version 1 (Baranek et al. [9]) fSensory Sensitivity Survey (60 items from The Sensory Profile, Dunn [33]; Liss et al. [77]) gChinese Sensory Profile (Tseng and Cheng [121]) hAdolescent/Adult Sensory Profile (Brown and Dunn [15]) iJapanese Sensory Inventory-Revised (Ota et al. [96]) jThe Sensory Processing Scales Inventory: Sensory Over-Responsivity subscale (Miller and Schoen [90]) kSensory Experiences Questionnaire: Version 3.0 (Baranek [8], Unpublished manuscript) lSensory Perception Quotient (Tavassoli et al. [116], [118]) mSensory Reactivity in Autism Spectrum (Elwin et al. [42]) nComposite based on some of the Short Sensory Profile factors and the Sensory Over-Responsivity subscale oThe Sensory profile-2 (Dunn [35])

The meta-analysis included 55 studies, 14 from the 2009 meta-analysis and 41 new studies. Table 1 presents the detailed features of the studies and their samples.

The 2009 meta-analysis coding guidelines were adapted to meet changes in diagnostic procedures and the availability of more descriptive information in papers (e.g., ratio of Caucasian ethnicity). Two independent researchers (the first and fourth authors) reviewed the coding guideline for extracting background information and sensory data from the publications, as well as coding examples from the 2009 coding. Both researchers coded the same four papers and reached 99.95% agreement. The researchers further refined the guidelines and added variable categories as needed (e.g., new types of sensory questionnaires not studied in 2009). The papers were coded for: group size, publication year, type of comparison group, type of group matching, mean age, age group, mental level, male ratio, Caucasian ratio, type of ASD, ASD diagnostic method and professional diagnosis, recruitment setting, type of mental test, type of sensory test, self-report, version of Sensory Profile, Short Sensory Profile (SSP), sensory quadrants reported, direction of the sensory scores (+ 1 for studies in which higher scores indicated higher severity and − 1 vice versa). The remaining papers were coded separately by each of the researchers. Periodic review of coding (for approximately 30% of papers) and discussions of outstanding examples in studies were regularly conducted.

For each study (and group) an effect size was computed for the three sensory patterns, SOR, SUR and/or Seeking. As such, studies that reported sensory pattern subcomponents only (e.g., for SOR they reported sensory avoiding and sensory sensitivity scores) required the computation of one effect size for each pattern. See "Appendix" for a conceptual comparison of scores between measures. This "Appendix" details the types of scores for which one effect size was computed per pattern. In addition, when mean sensory scores were reported for each ASD subgroup (e.g., autism and PDD-NOS) as opposed to the total ASD sample, computation was weighted by the size of each subgroup.

The SSP does not provide a distinct Seeking score; hence, studies using this measure do not have a Seeking effect size. The SSP has a low energy score that was used to compute the SUR effect size. In 2009, we did not include this score as it has limited items which are not cross-modality. However, in the current review the additional studies using the SSP enabled us to test the moderating effect of using the SSP (most popular measure) upon the SUR effect size results. In SSP studies (k = 4) for which we had the raw data, SOR, SUR, and Seeking scores were computed directly to include a wider range of items under these patterns (personal communication Dunn August [34]).

For studies using any version of the Sensory Profile that reported quadrant scores in addition to factor scores or instead of factor scores, we relied on the quadrant scores. This was because they are based on a wider range of items. For these studies, the SOR score was computed as a sum of Avoiding and Sensitivity quadrants following the manual guideline (Dunn [34]).

Cohen's d effect size was computed i.e. the standardized mean difference using the pooled standard deviation of the ASD and control groups with a bias correction (Hedges and Olkin [57]). In two studies (Dunn [34]; Tomcheck and Dunn [119]), only effect sizes were reported and not means and SDs. Therefore, we relied on the effect sizes reported. If an eta-squared was the type of effect size reported, it was transformed into a d value (Cooper [27]). When several comparisons were reported leading to several effect sizes of the same study, we averaged them. Since the correlations of these effect sizes were not reported, we estimated the effect size variance by the average of their estimated variances. The justification for this is that all effect sizes should have similar asymptotic variance (i.e. 1/n1 + 1/n2) and they may be highly correlated.

The following variables were tested as moderators; significant moderators were reported in results:

• Demographic variables for ASD group mean age in years, age group (five groups), ratio of males, IQ scores, level of functioning (high vs. low), type of group matching (only chronological age vs. only mental age).
• Methodological variables publication year, Sensory Profile based measure, SSP used, self-report.

Some moderators overlap conceptually. They were coded to enable quantification for studies that had one overlapping moderator and not the other. For example, the categorical moderator of level of functioning overlaps with the continuous IQ variable. Studies that did not report IQ scores for the ASD group but stated that the ASD group included only a certain level of functioning could be coded for the categorical variable. Self-report measures can be only applied in studies of adolescents/adults who are cognitively able to complete them. As such, this variable also captures mental level and is not solely an indication of biases of caregiver versus self-report. The moderation of self-report was tested only among studies with individuals for whom self-report measures were available (≥ 9 years).

Each study was included once in each of the nine models. Therefore, if a study had two DD groups, the larger was included in the meta-analysis. The only exception was in the case of testing the moderating effect of the type of group matching on ASD and TD effect sizes. Here, an additional group of 99 TD children from the same study (Ben-Sasson et al. [12]) matched on mental age was included, given the few mental age-matched studies. All effect size calculations and moderator testing were done with the "metafor" package (Viechtbauer [125]) in the software (R Core Team [101]).

Changes in diagnostic practices (e.g., DSM-5 subtyping does not include autism separately hence coding for percentage of autism diagnosis was no longer relevant) as well as more detailed descriptive information reported in research (e.g., with regards to raw IQ values) offered an opportunity for fine grained coding which in 2009 was not possible. The only identical moderator tested in both meta analyses was type of matching, i.e., chronological versus mental age matching. Similar moderators but defined differently between meta-analyses were age group which we were able to expand to a wider age range in the current analysis, percentage of autism diagnosis in 2009 and level of functioning in this analysis. New moderators available only in the current analysis were: IQ score, male ratio, Sensory Profile based measure, SSP used, self-report.

The 55 studies included in the meta-analysis included 4606 individuals with ASD, 5508 TD (k = 47 [with k denoting the number of studies and n the number of participants], 376 with DD [k = 11]), and 399 with other clinical conditions (k = 7, 5 ADHD comparison groups, 1 Learning Difficulties, and 1 SPD). Nine of the 47 studies, had another comparison group besides the typical group. For five studies this was a DD group and for four another clinical group. Studies were published from 1998 to June 2018 (2009 studies years of publication were 1998–2007), with 21 of the 55 studies published after DSM-5 publication in 2013. See Table 1 for features of the individual studies.

For 48 studies for which chronological age was reported, the average age of the ASD group was 10.85 years (SD = 9.26), of the typical group 10.9 (SD = 9.67), of the DD group 6.59 (SD = 3.98) and of the mixed clinical group 8.90 (SD = 0.91). ASD age group data was available for all 55 studies: 10.9% (6 of 55) of the studies included children under the age of 3 years, 12.7% (7 of 55) included children 3–6 years, 40.0% (22 of 55) included children 6–9 years old, 21.8% (12 of 55) were 9–18, and 14.3% (8 of 55) were above 18 years of age. For 26 of 55 studies with IQ information for the ASD group, the mean IQ for the ASD group was 83.95 (SD = 18.73). For the 15 of 47 studies with IQ for the TD group, the mean was 110 (SD = 5.24), for 9 of 11 studies with IQ data for the DD group, the average was 65.77 (SD = 8.39). For the two of seven studies with IQ data for the clinical group, the mean was 73.65 (SD = 40.8). Most studies included higher functioning individuals with ASD (70.6% 27 of 34 studies based on study inclusion criteria or mean IQ). The ASD samples from the 14 studies analyzed in 2009 versus the 41 new ones, were younger (95.8 vs. 139.5 months respectively) and had lower average IQ scores (66.97 vs. 83.95 respectively).

Forty-nine studies (of 55 studies, 89.1%) reported the gender of participants. The percentage of males in the ASD group ranged from 36.6 to 100%, with an average of 81.4% as compared to 69.6% in typical group, 75.3% in DD group and 78.7% in the clinical group. The study participants were predominantly Caucasian (for k = 20, M = 72.6%). Three studies with large Asian samples appear in the current meta-analysis, representing a population that was not previously studied (Cheung and Siu [22]; Matsushima and Kato [82]; Tseng and Cheng [121]; n = 189 with ASD compared to none in the 2009 meta-analysis), and a Saudi Arabian sample not previously represented (Al-Heizan et al. [1]).

Fifty-five studies reported the type of matching between groups. For studies compared to a typical group, most (k = 33 of 47) matched on chronological age, five on mental age, and nine on both. For studies with a DD comparison group, most (k = 8 of 11) matched on mental and chronological age, two matched only on chronological age and three only on mental age. Of studies with a different clinical comparison group, six matched only on chronological age and one on both chronological and mental age.

The most common diagnostic method reported was meeting DSM or ICD criteria (k = 30 of 50 that reported diagnostic method) with or without another method. Other diagnostic criteria were ADOS or ADOS-G (k = 21), having a current or previous clinical diagnosis of ASD (k = 18), ADI or ADI-R (k = 12), and CARS (k = 4). A total of 24 studies required meeting at least two diagnostic methods. Nine studies of the 21 which required meeting ADOS cutoff required also meeting ADI cutoff for ASD. Five of these nine studies also required meeting DSM/ICD criteria. In one study, parents completed the sensory questionnaire prior to receiving the child's diagnosis (Wiggins et al. [134]).

Most studies described the professional required to diagnose ASD (k = 36 of 55). Of these, 26 professionals who diagnosed were an MD and/or psychologist (k = 26 of the 36 studies), and 10 did not specify the profession required for diagnosing but stated that they were trained professionals.

Forty-one studies (of 55, 74.55%) used some version of the Sensory Profile or an adapted version of the Sensory Profile, of which 16 (of 41) were the SSP, and for 19 (of 41) we had quadrant scores (i.e., seeking, low registration, avoidance, sensitivity). Three of the 55 studies used the Sensory Experience Questionnaire, three of 55 used different versions of the SensOR scale (known today as SP3D). The remainder of measures appeared once. Nine studies relied on self-report (of 16 studies that could potentially use self-report based on their mean age of at least 11 years).

Average effect sizes across studies were large and significant for all types of sensory scores, SOR 1.28 (k = 47, 95% CI = [1.11–1.45]), SUR 1.38 (k = 43, 95% CI = [1.17–1.59]), and Seeking 0.66 (k = 30, 95% CI = [0.25–1.07]: see Figs. 3, 4, 5). Forty-five of the 47 studies with SOR scores were positive and significant (d = 0.47–2.57; with 84.4% with large effect sizes greater than 0.8). Two non-significant studies were Gonthier et al. ([49]) with a negative effect size and Van Etten et al. ([122]) with a positive effect size. For SUR, 41 of the 43 studies were positive and significant (d = 0.43–2.82; with 85.4% greater than 0.8). One study had a significant negative SUR effect size (− 0.89, 95% CI = [− 1.13 to 0.66]). Only 1 study was not significant (d = 0.06, 95% CI = [− 0.47 to 0.58]). For Seeking, 20 out of 30 studies had positive and significant effect sizes (d = 0.30–2.15; with 75% greater than 0.80). Seven studies had significant negative Seeking effect sizes. Three studies had a non-significant positive effect size with a CI that crossed '0'.

Graph: Fig. 3 Effect sizes (d) and 95% CI for SOR between ASD and typically developing groups

Graph: Fig. 4 Effect sizes (d) and 95% CI for SUR between ASD and typically developing groups

Graph: Fig. 5 Effect sizes (d) and 95% CI for Seeking between ASD and typically developing groups

While aggregated average effect sizes were significant, they were not homogeneous (SOR Qw[df = 46] = 486.43, SUR Qw[df = 42] = 724.07, Seeking Qw[df = 29] = 1146.32, p <.0001 for all). This suggests that studies varied in the magnitude of difference between groups. Therefore, potential moderators were tested. Demographic and methodological moderators tested did not significantly explain the heterogeneity in effect sizes for SOR or SUR models.

For Seeking effect sizes, several moderators significantly explained differences between ASD and TD groups. Effect sizes significantly differed between age groups (Qw[df = 3] = 10.51, p =.02), Fig. 6 shows that effect size increased up to Group 3, ages 6–9 years and decreased thereafter (see Fig. 6). IQ significantly moderated Seeking (Qm[df = 1] = 4.71, estimate = − 0.04, p = 0.03), such that the lower the IQ, the larger the effect size. Self-report was also a significant moderator (Qm[df = 1] = 9.16, estimate = − 1.98, p =.003). Those who self-reported, had smaller effect sizes than those whose caregivers reported for them.

Graph: Fig. 6 Seeking effect size compared to TD by age groups

Average effect size for the comparison of ASD and DD groups was significant for SOR 0.54 (k = 11, 95% CI = [0.30–0.78]; see Table 2). Significant positive effect sizes were found in six of the 11 studies (54%), ranging from 0.58 to 1.06. Average Seeking effect size was significant and positive but low, 0.49 (95% CI = [0.25–0.73]), with three of the six studies showing significant positive effect sizes. Average SUR effect size between ASD and DD groups was not significant 0.22 (k = 11, 95% CI = [− 0.14 to 0.58]).

Effect size of sensory symptoms of ASD compared to DD groups

Further testing showed significant heterogeneity for SOR (Qw[df = 10] = 28.96, p =.0013) but not for Seeking (Qw[df = 5] = 9.50, p =.09). No significant moderators explained the SOR heterogeneity.

The average effect size for the seven studies comparing SOR of ASD to another non-DD clinical group (i.e., ADHD, SMD, LD) was significant, 0.52 (95% CI = [0.19–0.85]; see Table 3). Four of the studies showed significant positive ds from 0.33 to 1.58. Testing showed significant heterogeneity among the SOR effect sizes (Qw[df = 6] = 22.93, p = 0.0008); however, moderators with sufficient data were not significant. Average effect size was not significant for SUR, 0.31 (95% CI = [− 0.16 to 0.76]), and for Seeking − 0.10 (95% CI = [− 0.41 to 0.21]).

Effect size of sensory symptoms of ASD compared to other clinical groups

Since the 2009 meta-analysis and the inclusion of sensory symptoms in the DSM-5 ASD diagnostic criteria, there has been a dramatic increase in research in this area. In addition to our goal to aggregate a larger body of evidence, the current meta-analysis was able to expand inquiry to moderators for which previously we had limited data such as age, male ratio, other clinical group comparisons, and type of questionnaire. The current meta-analysis focused on 55 studies comparing individuals with ASD to typical and clinical control groups using sensory processing questionnaires. Particular growth was observed in studying older age groups, groups with higher functioning individuals, samples from Asian countries, and inclusion of clinical control groups other than DD/ID. Findings from this meta-analysis strongly support the inclusion of sensory symptoms as part of the diagnostic criteria of ASD, given the significant differences among most sensory comparisons and their robustness resulting from their independence of the demographic and methodological variations among studies. The main findings showed how SOR uniquely differentiates individuals with ASD, no matter what the comparison group and with no age, developmental, or methodological explanations. While SUR distinguished ASD from TD groups, differences diminished when related to other clinical groups. Seeking was elevated in ASD compared to TD and DD groups and was moderated by age, IQ and self-report. This meta-analysis indicates the need to examine different types of sensory patterns separately and to investigate factors underlying variability in their expression.

SOR differences were the most prominent, as shown by their consistently positive differences across individual studies and significance across all types of control groups. The percentage of individual studies with significant effect sizes was highest among comparisons to a TD group (95.74% for TD, 54.55% for DD, and 57% for clinical comparison). TD comparison studies showed significant heterogeneity; however, none of the moderators tested could account for this. In our previous meta-analysis, SOR was moderated by age, type of group matching, and whether over 80% of the sample included people diagnosed with autism. It is plausible that the current moderators estimating level of functioning and development lost their power as a result of the addition of older and higher functioning samples in this meta-analysis relative to the previous report. While at the individual study level moderators such as IQ were significantly correlated with sensory symptoms (e.g., Crane et al. [29]; Stewart et al. [113]; Watson et al. [127]), this relation was masked at the aggregated level. Future research is needed to identify moderators for SOR. Potential candidate moderators to test in future reviews are those with established correlational evidence, such as psychopathology symptoms, including anxiety (Green and Ben-Sasson [51]), gastrointestinal problems (Mazurek et al. [86]), and sleep disturbances (Mazurek and Petroski [85]).

SUR symptoms, such as delayed reaction to pain and temperature, have long been associated with ASD (Rogers and Ozonoff [104]). In the 2009 meta-analysis, the SUR TD model showed a larger effect size (d = 2.02 in 2009 vs. d = 1.38 in current study) and was moderated by age and type of matching. While most studies in the current meta-analysis showed positive effects, a larger, Asian study yielded a significant, negative effect size, which lowered the average SUR d (Cheung and Siu [22]). Potential cultural factors in threshold for under-reactivity may have created bias in this study. It has been proposed that SUR characterizes ASD earlier rather than later in development (Schauder and Bennetto [110]). SUR may have decreased in magnitude given the extensive increase in the proportion of older age-groups in the current meta-analysis. There may be methodological explanations for the differences in SUR findings. Note that in 2009, SUR of the SSP studies was not analyzed due to the limited number of items in this scale, resulting in very few studies included in the previously tested model. In the current review, SUR of the SSP was included given the large number of SSP studies identified as well as several raw datasets which became available and enabled a richer computation of SUR and Seeking for SSP studies. Even though SSP is very limited in measuring SUR, its use in a study did not appear to moderate SUR differences.

Surprisingly, SUR average effect size was positive and significant relative to TD controls but not relative to DD or other clinical groups. The individual DD comparisons indicate a diversity in the direction of the effect size, with two studies showing a negative effect size for very young children (McCormick et al. [87]; Rogers et al. [103]). The lack of significant SUR effects is surprising as many of the sensory questionnaires ask within their SUR scale about social under-response which is characteristic of ASD. Hence, one would expect that SUR would be consistently different across comparisons. Although developmental moderators did not explain the SUR TD model, the lack of significant comparisons with DD supports the role of development in the manifestation of SUR symptoms in ASD.

Atypical sensory seeking difference, while smaller in magnitude relative to the other sensory patterns, was the only pattern with significant moderating factors. Average Seeking effect size was significant when compared with TD and DD groups (ASD group on average showed greater Seeking) but not relative to other clinical groups. Nonetheless, there was great variability in the direction of the effect size and in its significance among the individual studies. Seven of the 30 studies with Seeking effect sizes, had significantly negative effect sizes (ASD group shows less Seeking), six of which included adolescents/adults, and five of the six relied on the AASP. Three studies were not significant and the rest (k = 20) were significant and positive. The ASD versus TD Seeking effect size was moderated by age, IQ and self-report. There was a non-linear relation with age group, with increases through the 6–9 years' age group and a decrease thereafter. This is consistent with the non-linear relation we reported in 2009 for Seeking. Looking at individual studies, we saw stability in studies with a young narrow age range (McCormick et al. [87]) versus a decrease in sensory symptoms when looking across a wider age range (3–56 years, Kern et al. [70]; 3–11 years, Leekam et al. [74]). It seems that in older samples, the atypicality is in the low level of seeking, which may be an outcome of avoidance. Higher IQ was associated with a smaller Seeking effect size. In addition, there was a smaller effect size for studies using a self-report measure; probably a factor of higher IQ. The great heterogeneity in effect sizes among studies for Sensory Seeking might be the reason it is the only pattern explained by age and IQ. Alternatively, the inconsistent conceptualization of Seeking across studies might lead to this variability (Schauder and Bennetto [110]). Part of this inconsistency is that Seeking items in young children's scales refer to intense sensory exploration, while in adult questionnaires, such as the AASP, they refer to personal sensory preferences, which do not capture the repetitive circumscribed, idiosyncratic nature of Seeking in ASD (Elwin et al. [41]).

We approached the investigation of the role of developmental level in explaining sensory symptoms in ASD from different angles to capture as many studies as possible and to obtain a comprehensive definition of this construct. This included coding IQ, level of functioning of ASD group, type of matching (CA- vs. MA-based), and separating effect sizes compared to typical versus DD groups. Nonetheless, the various developmental level variables were not significant, except in the Seeking TD model. This is consistent with findings of lack of relation between sensory symptoms and cognitive abilities (O'Donnell et al. [95]). Note that in the 2009 meta-analysis, the moderation of developmental level was estimated more crudely by the type of matching and the dichotomous severity of autism variable (> 80% autism diagnosis). In the current meta-analysis, there was a trend toward a relation with development level seen by lower average d's and fewer significant d's when ASD was compared to DD and other clinical groups versus TD comparisons. In addition, SUR was only significant for TD comparison, suggesting it is a symptom mostly accounted for by developmental level.

Conflicting findings for the association between sensory symptoms and adaptive behaviors (i.e., proxy of developmental level) were also reported in a recent review of correlational research by Dellapiazza et al. ([31]). The diversity in the operational definition of sensory patterns and of developmental level among studies requires caution in generating far-reaching conclusions concerning the contribution of developmental level to the severity and type of sensory symptoms.

Compared to other clinical groups that did not include a DD/ID, the SOR average d was the only significant pattern. Nonetheless, the magnitude of d and the number of significant individual d's was lower relative to the TD and DD SOR models. At the individual study level, three (Cheung and Siu [22]; Clince et al. [25]; Little et al. [79]) of the five studies comparing ASD to an ADHD group had non-significant SOR d. Further understanding of sensory differences between ASD and ADHD is needed. These results regarding non-DD clinical disorders are based on a very few studies (k = 7) which looked at clinical comparisons other than DD (k = 5 ADHD, 1 SPD, 1 LD). None of the studies compared ASD to a psychopathological group with anxiety disorder or OCD, which also present extreme sensory symptoms. There was also no language disorder or social–communication disorder control group, conditions important to study, particularly when thinking of the strong social components of tactile and auditory stimuli.

Several limitations within this body of literature should be acknowledged when interpreting the findings. Several studies were excluded because they did not publish means or SDs for the sensory scores or did not report scores reflecting one or more of the three distinct sensory pattern scores. For example, the SSP does not yield a distinct Seeking score, hence if authors did not send us means or raw data so we could compute distinct SUR and Seeking scores their research could not be included in the SUR and Seeking models. This may have created bias among SSP studies. While many authors sent us these scores, others did not or used other questionnaires that do not readily provide these type of scores (e.g., Sensory Processing Measure, Parham and Ecker [97]). The publication of the Sensory Profile 2 versions will reduce this limitation by providing the same three pattern scores across age groups.

ASD severity was not examined directly in this meta-analysis due to the studies relying on different metrics, different versions of the DSM, or not reporting severity at all. Studies varied in the homogeneity of the ASD group, with some purposefully studying a unique type of ASD, such as only HFASD (Hilton et al. [59]), LFASD (Gonthier et al. [49]), or those with ASD and ADHD (Mattard-Labrecque et al. [83]). Our previous moderator testing of percent of autism diagnosis was no longer relevant given changes in DSM-5 classification. We partially captured ASD severity by examining developmental level.

It is worth noting that IQ was measured in different ways across studies. Only non-verbal testing was used in some studies, while full-scale IQ was used in others. Crane et al. ([29]) demonstrate how different types of IQ scores yield different correlations with SOR, SUR and Seeking. Setting unified guidelines for phenotyping in research would strengthen the ability to conduct wide-scale, cross-site research, as well as large meta-analyses.

The age range within some samples was very wide, with some studies including both adolescents and adults (e.g., Kern et al. [69]). The skewed distribution of age as a continuous variable supported the analysis of age groups. Age groups were defined by the mid-age range of the group, however at sometimes this could not fully capture the heterogeneous age range in the study.

We were limited in our ability to examine the effects of gender upon sensory symptoms, hence its non-significant moderation across models was not surprising. This was due to the narrow range and variance of the male ratio leading to a non-normal distribution among the compared studies. To fully examine this effect more sensory data for females with ASD is needed.

From a measurement perspective, most measures are relying upon parental reports about the frequency of the behavior or the number of symptoms. A few studies included self-reports, which differ in that they provide a first-hand account from individuals with high abilities. In addition, current healthcare models (e.g., ICF) stress the need to assess limitations based on the condition's impact upon performing activities and participation. The degree of interference/functional impairment associated with sensory symptoms in ASD need to be measured.

While conceptually, we aimed to examine comparable constructs (see "Appendix"), variability in the actual measurement of the construct may have biased results beyond what we could identify in this meta-analysis. First, some of the studies modified the use of standard sensory questionnaires. This included self-report rephrased for caregiver report for individuals who are low functioning and living in a care center (Gonthier et al. [49]), reducing the number of SP items for the Chinese SP to fit culturally (Cheung and Siu [22]), or using the SSP for a younger age range (Wiggins et al. [134]). Second, the number of items and types of modalities covered by each questionnaire varied between measures. For instance, SUR in the SSP was based on two modality-specific items compared to 11 items across modalities in the ITSP. Third, even for the same measure, scores may have reflected a different set of items, such as, researchers who used the SSP but added a few factors from the SP (O'Brien et al. [94]). The recent publication of one scoring system across versions of Sensory Profile will assist in comparing results across studies.

Finally, for studies with multiple scores, we averaged their effect size into a composite effect size. Nonetheless, in the absence of information regarding the shared covariance between the subscores, there is no correct way to estimate the standard error of this composite effect size. Although the method we used is common practice in meta-analytic research (Lipsey and Wilson [76]), its limitations should be acknowledged.

At the review level we aimed to reduce risk bias through systematic search methods, inter-rater reliability testing, however, there may be literature missed such as non-English reports. In addition, given the back and forth nature of the decision making process for paper extraction we were unable to purely differentiate title/abstract based filtering versus full-text filtering of studies. Also note that one reason for research exclusion was selected for each paper based on predetermined hierarchy of reasons, but many papers qualified for multiple reasons.

The aggregated results from this review are of relevance for theory, assessment and intervention. First, our findings suggest that sensory symptoms in ASD require clinical attention throughout the lifespan and across levels of functioning in terms of professional awareness of their impact and allocation of resources for the matter. SOR in particular can assist in diagnosis as it was the sensory pattern with the most consistent distinction of ASD group across comparisons. Nonetheless, diagnostic procedures must examine the co-occurrence of SOR with other ASD symptoms as of its own it cannot be used for early identification as noted by the presence of SOR in other clinical groups. Second, there is a need to recognize the poor differential property of SUR and Seeking between ASD and other clinical conditions as suggested by the smaller magnitude of group differences in these scores as well as their inconsistent effect size significance. The nature of the sensory features and/or the co-occurrence of sensory symptoms with social–communication deficits may be what signifies ASD. Third, from a measurement perspective, while the Sensory Profile questionnaires are the most commonly used tool internationally there is need for evaluating the functional impairment associated with each pattern of sensory symptoms. In addition, clinicians should aim to gain a 'pure perceptual' measurement of sensory experiences which is independent of affective response such as anxiety and stress. Adding observational evaluation methods to the profiling of sensory symptoms is critical for overcoming report biases. Finally, SOR, SUR and Seeking symptoms require separate acknowledgement given their different pattern of results in the tested models. However, given the commonality of more than one pattern in an individual (Ben-Sasson et al. [13]), we need to examine implications of these combinations.

This large-scale study attests to the atypical nature of sensory symptoms in ASD, with consistently atypical SOR symptoms. Since 2009, additional studies have been published, including studies with older, higher functioning, and Asian samples, as well as clinical comparison groups. Although the Sensory Profile remains the most common measure, new measures are being introduced to the field, some of which are perception-focused (e.g., SPQ, Tavassoli et al. [116], [118]). As opposed to looking at sensory symptoms as one construct or by modality, the varying patterns of differences for SOR, SUR and Seeking call for separate measurement and research of these patterns, as noted in the conclusion of the systematic review by Glod, et al. ([48]). Seeking differences relative to TD groups were the only pattern explained by age and developmental level. Sensory differences across studies and types of comparison groups often showed significant heterogeneity; nonetheless, most of the moderators we tested couldn't explain this diversity. There may be other dimensions not captured in the body of literature reviewed which can account for such variability and assist in identifying a clinical profile associated with sensory symptoms in ASD. We believe new discoveries in this area will be made possible by expanding methods of measurement and through investigation of effective interventions.

This work has been supported by a National Institute of Psychobiology Dylan Tauber Young Investigator Grant awarded to Dr. A. Ben-Sasson (Grant Number 204-177-18b).

We thank all the researchers who sent us the data needed for their samples to be included in the meta analysis.

AB, EG, and SC conceived of the study and its design. AB designed and coordinated the study, performed literature search, data extraction, coding, analyses, and wrote the manuscript; EG reviewed coding and data interpretation. RF conducted statistical analysis; NK was involved in data extraction and coding; SC was involved in data analsis planning, data interpretation and writing the manuscript. All authors approved the manuscript.

The authors declare they have no conflict of interests.

All procedures performed in this meta-analysis were in accordance with the Ethical Standards of the Institutional Research Committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

See Table 4.

Comparison of sensory scores across studies

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.