Background
Differences in behavioral responses to sensory inputs from the environment have been associated with autism spectrum disorder (hereafter “autism”)
1 since the first clinical descriptions of the condition [
6,
7]. Sensory phenotypes are present across multiple modalities (e.g., auditory, visual, tactile) and include differences in sensory reactivity and modulation, multisensory integration, and certain aspects of perception [
8‐
14]. With regard to sensory reactivity, these features are frequently parsed into three specific behavioral “response patterns”:
hyperreactivity (HYPER; i.e., excessive and/or defensive reactions to stimuli that most individuals find innocuous),
hyporeactivity (HYPO; i.e., diminished or absent responses to sensory stimuli that most individuals would respond to), and
sensory seeking (SEEK; i.e., unusually strong fascination with or craving of sensory stimulation, often accompanied by repeatedly seeking out specific sensory inputs [
15‐
17]). Notably, these response patterns are not mutually exclusive, and many individuals express behaviors characteristic of multiple sensory response patterns, even within the same modality [
9,
18,
19]. Sensory reactivity differences are extremely common in autistic individuals: the point prevalence of a child displaying differences in any of the three response patterns (i.e., HYPER, HYPO, or SEEK in any modality) was recently estimated to be 74% using large-scale population-based data from the Autism and Developmental Disabilities Monitoring Network [
20], and 70.9–88.3% of autistic youth in two large samples (from the United States and Australia, respectively) were determined to have sensory reactivity differences of at least “mild” severity [
21,
22].
Although sensory reactivity differences are prevalent in many childhood-onset neurodevelopmental and neuropsychiatric conditions (e.g., attention deficit hyperactivity disorder [ADHD], anxiety, obsessive–compulsive disorder, Tourette syndrome, Williams syndrome [
23‐
28]), and all of these clinical groups can be differentiated from neurotypical controls in terms of sensory reactivity differences (see also [
29]), a recent meta-analysis suggests that autistic individuals demonstrate higher average levels of HYPER (with findings mixed and inconclusive for HYPO and SEEK) when compared to individuals with other clinical conditions [
30]. Moreover, many qualitative and quantitative studies have linked specific sensory features of autism to functional impairment, reduced activity participation, and lower quality of life (e.g. [
31‐
41]), further emphasizing the importance of research into the sensory aspects of the autism phenotype. However, it is worth noting that not all sensory features of autism are inherently impairing or pathological, and some (particularly within the SEEK domain) are viewed positively by autistic people themselves [
42‐
50].
Although recognition of the sensory features of autism has grown noticeably in recent years [
30], relatively little published research in this area has evaluated structural relationships between different domains of sensory reactivity or tested the validity of existing theoretical subdimensions to describe this aspect of the autism phenotype (e.g. [
51‐
56]). The majority of studies examining sensory features in autism have utilized caregiver-report questionnaires such as the Sensory Profile (SP [
57,
58]), the Sensory Experiences Questionnaire (SEQ [
19,
59,
60]), and the Sensory Processing-3 Dimensions: Inventory (SP-3D:I [
61,
62]), which contain a mixture of HYPER, HYPO, and SEEK items split among sensory modalities. Though combinations of all three response patterns and the five classical sensory modalities (vision, audition, olfaction, gustation, and touch) are typically represented on most sensory reactivity questionnaires, the number of items tapping each subconstruct can vary substantially, and additional sensory modalities (e.g., vestibular sense, facets of somatosensation such as pain/temperature and proprioception, interoception) may or may not be included as well. Notably, these measures are most often scored by generating supra-modal (i.e., combining multiple sensory modalities) HYPER, HYPO, and SEEK “response pattern scores” that aggregate items within a single response pattern across all assessed sensory modalities.
Although the aforementioned supra-modal constructs are consistent with the major conceptual models of sensory features [
63‐
65], empirical support for the practice of combining responses to stimuli across multiple sensory modalities into a single “overall response pattern [HYPER, HYPO, or SEEK]” construct (as would be operationalized by a total score on all HYPER items, for instance; see [
66]) is relatively limited. When examining the factor structures of existing sensory questionnaires, models that consider supra-modal response pattern factors in isolation tend to be inadequate, demonstrating very poor overall fit to empirical data (e.g. [
67]). Thus, in order to successfully explain the factor structure of the HYPER, HYPO, and SEEK constructs, previous studies have needed to utilize more complex models that include not only
supra-modal response pattern factors but also
modality-specific response pattern factors that account for the additional shared (co)variance between items within a given sensory modality (e.g. [
51,
54,
68]). Notably, these models represent
bifactor structures (i.e., two-level structures with orthogonal “general” and “specific” factors contributing to each item response), with variance attributable to both supra-modal constructs (e.g., HYPER, HYPO, SEEK) and modality-specific constructs (e.g., Vision, Audition, Olfaction) [
69,
70]. Given this division of variance between levels, summed supra-modal response pattern scores may only be clearly interpretable as measures of HYPER, HYPO, or SEEK if the strength of the supra-modal factor is much stronger than the modality-specific factors [
70‐
73]. However, there has been a dearth of psychometric work using bifactor indices to examine the interpretability of these supra-modal sensory constructs in the autistic population to date (though see [
68]); thus, their construct validity in this population remains unclear.
In contrast to studying HYPER, HYPO, and SEEK at the supra-modal level, a minority of studies (e.g. [
27,
74‐
81]) have investigated these sensory constructs in a
modality-specific manner by calculating response pattern scores that are limited to a single sensory modality (e.g., “Auditory HYPER,” which reflects the sum score of only the HYPER items within the Auditory modality). As psychophysical and neural measures of sensory function (e.g., detection thresholds, psychometric function parameters, evoked potential amplitudes) are frequently limited to a single sensory modality, some researchers theorize that the modality-specific subconstructs represented by these measures will correlate more strongly with sensory reactivity measures that are limited to that same sensory modality rather than collapsed across modalities (e.g., visual evoked potential amplitudes may be expected to correlate moreso with a measure of Visual HYPER than with general HYPER). To our knowledge, studies to date have not formally tested these hypotheses to determine whether or not modality-specific response pattern scores demonstrate any empirical advantages over conceptually broader supra-modal response pattern scores when correlated with psychophysical or neurophysiological measures of sensory function.
Determining the most appropriate “level of analysis” (supra-modal versus modality-specific versus some combination of the two) for these sensory constructs has major implications for other areas of sensory autism research, as this decision will impact whether modality-specific or supra-modal sensory constructs are assessed by diagnostic/phenotyping instruments, targeted by clinical interventions, correlated with other individual differences, explained with neuroscientific or psychological models (e.g., multiple forms of sensory reactivity having a shared underlying mechanism or cause versus separate mechanisms), and even incorporated into the diagnostic criteria for autism. Thus, additional research is needed to more conclusively determine whether sensory reactivity differences in autism are most appropriately studied at the level of a response pattern score (HYPER, HYPO, or SEEK) combined across modalities (e.g. [
30,
66,
82‐
85]), at the level of modality-specific response pattern scores (e.g. [
75,
77‐
80,
86]), or some combination of the two (e.g., interpreting both types of scores; favoring one level of analysis at different points in a study based on the research question or the specific construct(s) being studied).
Purpose
To address this critical gap in research on sensory features in autism, the present study sought to quantitatively investigate the latent structure of caregiver-reported sensory features across a large and heterogeneous group of autistic children. By pooling data from multiple independent research groups and the National Database for Autism Research (NDAR [
87]), we compiled a cohort of several thousand autistic children to be analyzed within the methodological framework of integrative data analysis (IDA [
88,
89]). The IDA approach has recently gained popularity within autism research, going beyond small sample studies to yield insights about the latent structure of core and associated autism features [
56,
90‐
95], the psychometric properties of widely-used measures [
56,
92,
96‐
101], and the associations between autism features and other related clinical and demographic variables [
93,
99,
102‐
104]. However, many of these studies have not explicitly quantified the degree to which effects of interest vary across pooled datasets (i.e., effect size heterogeneity), arguably a major strength of IDA methodology (see [
103] for a notable exception). Utilizing modern psychometric techniques such as item response theory (IRT [
105,
106]) and bifactor modeling [
69,
70,
107], the current IDA sought to rigorously evaluate the latent structure of sensory features in autism across multiple measures. Our specific aims were to:
-
derive psychometrically sound metrics of HYPER, HYPO, and SEEK within modalities;
-
derive psychometrically sound supra-modal metrics of HYPER, HYPO, and SEEK;
-
use bifactor models and indices to evaluate whether modality-specific HYPER, HYPO, and SEEK metrics (e.g., Auditory HYPER, Tactile HYPO, Visual SEEK) provide added-value to the supra-modal metrics; and
-
estimate meta-analytic associations between psychometrically derived sensory constructs and other clinical and demographic variables (as well as the degree of heterogeneity in those associations) in the autistic population.
Discussion
Despite the recent elevation of sensory reactivity differences to the status of a diagnostic criterion for autism [
15,
16], there has been relatively little empirical work examining the underlying latent structure of these core sensory features within the autistic population [see also
51. By analyzing caregiver-reported sensory reactivity differences in a heterogeneous cross-sectional sample of nearly 4000 autistic children, the current study sought to investigate the hierarchical structure of sensory hyperreactivity (HYPER), hyporeactivity (HYPO), and sensory seeking (SEEK) across the full spectrum of children captured under the label of autism spectrum disorder. Utilizing modern psychometric techniques, we developed structural models of HYPER, HYPO, and SEEK in individual sensory modalities, subsequently testing whether each construct is most appropriately studied at the level of a single supra-modal sensory response pattern (e.g., an overall SEEK score) or separately for each modality within the sensory response patterns (e.g., separate scores for Visual SEEK, Auditory SEEK, and Tactile SEEK). Of the three sensory response patterns included within current autism diagnostic criteria [
15,
16], only HYPER demonstrated unambiguous evidence of an interpretable supra-modal construct, whereas supra-modal HYPO scores as currently operationalized were found to have limited construct validity. The evidence for supra-modal SEEK scores was more ambiguous, as we were unable to generate an adequately-fitting bifactor model of SEEK that included all relevant sensory modalities, but once modalities measured with only single items (i.e., Auditory, Olfactory, and Gustatory) were removed, the model containing the four remaining modalities (Visual, Tactile, Oral Tactile, and Movement) demonstrated an adequately fitting latent structure and acceptable general factor saturation. Although our findings did not conclusively support the construct validity of a SEEK composite that includes all seven standard modalities (i.e., those operationalized by the SP or SEQ SEEK response pattern scores), the more limited “General SEEK'' construct described here (consisting of only Visual, Tactile, Oral Tactile, and Movement items) may be a useful supra-modal aspect of the sensory autism phenotype if replicated in future studies. Additionally, irrespective of the construct validity of supra-modal scores, nearly all modality-specific sensory subconstructs demonstrated added value over and above their respective response pattern scores, indicating that modality-specific HYPER, HYPO, and SEEK scores (e.g., Visual HYPER, Visual HYPO) are able explain additional individual differences in sensory reactivity to a greater degree than a single supra-modal HYPER, HYPO, or SEEK score. These findings have implications for researchers interested in characterizing, explaining, or intervening on sensory reactivity in autistic individuals, as they suggest that some of the supra-modal response pattern scores commonly used in these areas may have previously unrealized psychometric limitations. HYPER, HYPO or SEEK scores that are limited to one modality (i.e., single-modality subconstruct scores) could potentially prove advantageous in some contexts, although additional research is necessary to determine the extent to which these measures demonstrate broader construct validity and practical utility over established supra-modal HYPER, HYPO, or SEEK scores.
The current study also provided a wealth of information about the measurement of sensory reactivity in autistic youth based on caregiver-report questionnaires. Using the most frequently employed caregiver-report sensory measures in the autism literature (the SP and SEQ), we attempted to generate unidimensional scales to operationalize each combination of modality × response pattern as its own unique sensory subconstruct. However, based on a priori psychometric criteria, we were unable to generate acceptable unidimensional scales for three of five HYPO modalities (Visual, Olfactory, Gustatory) and three of seven SEEK modalities (Auditory, Olfactory, Gustatory), necessitating the use of single-item indicators of these subconstructs (and one doublet) in later structural analyses. Moreover, within the HYPO domain, the two subconstructs that did produce sufficiently reliable scales reflected fairly specific subsets of the total item pool (e.g., HYPO to pain and temperature rather than all somatosensory stimuli), suggesting that they did not fully operationalize the “Auditory HYPO” and “Tactile HYPO” constructs that we had originally intended to measure. Thus, despite broadband sensory reactivity measures such as the SP, SEQ, and SP-3D:I typically including HYPER, HYPO, and SEEK items for each sensory modality, a sizable minority of “modality × response pattern” subconstructs demonstrated inadequate construct validity within this large autistic sample. In the current study, it is unclear whether this finding stems from instrument-specific measurement issues (i.e., inadequate construct coverage within the specific questionnaires from which items were drawn) as opposed to more general issues regarding the theoretical definition of the construct or its ability to be reliably operationalized as a set of observer-reported questionnaire items. As both sets of issues are likely to contribute in different cases, we suggest some sensory subconstructs (particularly within the HYPO and SEEK domains) are (a) underrepresented in existing questionnaires (i.e., more questions are needed), (b) poorly defined (e.g., Visual HYPO items have unclear relations with specific aspects of visual perception), (c) difficult for caregivers to report on reliably (e.g., few indicators of Gustatory SEEK are present in most children, limiting the pool of potential items available to capture this construct), and/or (d) of potentially limited theoretical relevance when predicting clinical outcomes (e.g., Olfactory HYPO may be unlikely to substantially influence the expression of other core features of autism). Future work should attempt to evaluate which, if any, of these poorly operationalized sensory subconstructs are relevant to autism research and clinical practice and if so, how they can be reliably measured.
As the majority of work examining sensory constructs in both autism and other clinical populations has utilized supra-modal scores from the SP, SEQ, SP-3D:I, or similar measures (e.g. [
29,
30,
82,
149,
150]), our findings signal the need for sensory autism research to broaden the ways in which sensory reactivity differences are characterized, potentially shifting away from the field’s nearly exclusive reliance upon supra-modal HYPER, HYPO, and SEEK scores for this purpose. Single-modality measures of HYPER, HYPO, and/or SEEK represent a viable alternative method of assessing these constructs and may be particularly useful when substantive research hypotheses include associations with other sensory constructs in a single sensory modality (e.g., tactile detection thresholds, visual evoked potentials; see [
151] for a recent example). Moreover, there is a great need to develop more comprehensive measures of modality-specific HYPER, HYPO, and SEEK subconstructs, either by expanding upon the item banks employed in the current study, adapting questionnaires used in other fields (e.g. [
152]), or developing and validating entirely novel measures (e.g. [
86]). By more densely sampling each subconstruct of interest, these measures could ostensibly increase the reliability, validity, conceptual breadth, and clinical utility of modality-specific response pattern scales compared to the short and relatively general item pools currently included in longer broadband sensory measures. Importantly, we are not suggesting that researchers entirely abandon the study of supra-modal sensory constructs—particularly HYPER, for which we have found some empirical support for the supra-modal response pattern—as there is certainly value in the investigation of these higher-order constructs as well. Rather, in future studies where both supra-modal and modality-specific subconstruct scores could feasibly be interpreted, we strongly recommend that researchers use contextual factors to determine which “level of analysis” is most appropriate or informative to answer the substantive research question(s) at hand.
For researchers who do choose to characterize sensory reactivity at the single-modality subconstruct level going forward (e.g., examining only Auditory HYPER or Auditory HYPO in the context of an auditory neuroscience study), it is notable that the “level of analysis” chosen to study a problem will likely frame the ways in which sensory features of autism are conceptualized and studied more broadly as an aspect of autism’s heterogeneity (e.g. [
10,
153,
154]). In particular, when developing clinical interventions for sensory reactivity in autism, a focus on modality-specific sensory subconstruct outcomes may motivate clinicians and researchers to investigate the efficacy of intervention modalities that are more focused on specific subconstructs rather than sensory reactivity in general (e.g., the use of sound generators to treat hyperacusis, a specific type of Auditory HYPER [
155]). Personalized interventions that seek to assess an autistic child’s specific pattern of sensory reactivity differences and ameliorate challenges associated with each domain could also be assessed within this framework, using modality-specific assessments of each sensory response pattern (e.g., an outcome measure specifically focused on Visual HYPER or Pain/Temperature HYPO) to monitor the effectiveness of each putative “active ingredient” of the intervention. A shift in measurement practices will also allow researchers to associate these single-modality behavioral subconstructs with psychophysical and/or neurophysiological measures within a given modality (e.g. [
75,
76]), informing theories of the neurocognitive underpinnings of certain types of sensory reactivity in autism (e.g. [
41,
156,
157]). Though we do not claim a single-modality perspective to be advantageous in all cases or for all research questions (particularly for those focused on “real-world” multisensory contexts), we believe that a greater diversity of theoretical approaches and frameworks within sensory autism research is needed to make optimal progress towards improving the lives of autistic people within this line of work.
In addition to our examination of the latent structure of sensory reactivity and assessment of evidence to support each “level of analysis,” we also employed random-effects integrative data analysis to estimate the meta-analytic associations between all sufficiently interpretable sensory subconstructs (i.e., General HYPER, six HYPER subconstructs, two HYPO subconstructs, and four SEEK subconstructs) and relevant demographic and clinical correlates (i.e., age, sex, cognitive abilities, adaptive functioning, core autism features, and co-occurring psychiatric symptoms). In general, the majority of associations were modest in size (with a few exceptions, e.g., large correlations between RBS-R RSM/Movement SEEK and RBS-R RSC/General HYPER), and Bayes factor tests indicated that many of the observed effects (particularly cross-sectional associations with age, sex, adaptive functioning, and cognitive abilities), were small enough to be practically equivalent to zero. Notably, none of the assessed sensory variables were significantly associated with sex or cognitive abilities as continuously quantified, although intellectual disability status was associated with moderately higher levels of Oral Tactile SEEK (i.e., seeking out the sensations of non-food objects in one’s mouth, not necessarily for consumption). Significant negative associations were also observed between certain HYPO/SEEK scores and adaptive behavior scores (with Pain/Temperature HYPO demonstrating the most consistent associations); however, these effects were relatively small in magnitude (|r| values < 0.199).
In line with the classification of sensory reactivity differences as a core diagnostic criterion for autism classified under restricted/repetitive behaviors and interests, most sensory subconstructs correlated moderately with one or more of the RBS-R subscales (i.e., RSM [lower-order repetitive behaviors] and/or RSC [higher-order repetitive behaviors]). Notably, the largest summary effect observed in the current study was the correlation between Movement SEEK and the RBS-R RSM subscale, although this was likely driven to some extent by overlapping item content (e.g., both SEQ 2.1 Item 27/SEQ 3.0 Item 76 and RBS-R item 4 contain jumping and spinning in circles as exemplars). Nevertheless, multiple RBS-R subscales demonstrated practically meaningful positive correlations with the majority of sensory constructs considered in the current study. Associated psychiatric symptoms also demonstrated small to moderate correlations with many HYPER subconstructs and several modality-specific HYPO and SEEK subconstructs, suggesting that outside of other core autism features, sensory reactivity (particularly HYPER) is most robustly related to transdiagnostic psychiatric symptomatology, particularly in the externalizing domain. Notably, as the sensory subconstructs of Tactile and Movement SEEK demonstrated practically significant positive correlations with externalizing symptoms and features of ADHD but not internalizing symptoms, these two domains may be reflective of an underlying liability for dysregulated or impulsive behavior (see also [
81]). Multiple domains of SEEK also showed negative correlations with age, potentially suggesting that these traits decrease over time as children develop increased capacity to regulate their motor impulses with age (e.g. [
158,
159]). Although cross-sectional correlations such as those explored in the current study are insufficient to determine causal relationships between sensory reactivity and other clinical constructs [
160], the present findings can nevertheless be useful in generating hypotheses for future targeted investigations of the causal interplay between sensory constructs and other core/associated features of autism.
Despite only two HYPO subconstructs (Speech HYPO and Pain/Temperature HYPO) being considered within the analysis of meta-analytic correlates, it is notable that these two domains of sensory reactivity diverged strongly in terms of their correlations with non-sensory variables. Speech HYPO demonstrated practically significant positive correlations with all core autism features (i.e., SRS, RBS-R subscales) and domains of psychiatric symptoms, whereas none of these domains were associated with Pain/Temperature HYPO. Notably, a child not responding to their name or other speech stimuli is frequently considered a core feature of autism outside of the sensory domain, conceptualized as a failure to orient attention to socially salient stimuli (e.g. [
161‐
163]). Thus, it is notable that observed Speech HYPO feasibly be present in the absence of underlying differences in sensory reactivity (e.g., due to differences in broader social or attentional processes). Future studies, particularly those that include multi-method assessments of both social-communicative and sensory factors, may be necessary to determine whether the underlying causes of Speech HYPO are indeed sensory in nature, thereby investigating the appropriateness of classifying this subconstruct as a sensory reactivity difference.
The Pain/Temperature HYPO subconstruct was significantly negatively associated with multiple domains of the VABS, though no correlations with core autism features nor psychiatric symptoms reached the threshold for practical significance. There was also a modestly increased level of Pain/Temperature HYPO in autistic individuals with a categorical label of intellectual disability, although this result fell slightly below our threshold for practical significance. Though these results seemingly indicate that insensitivity to pain and temperature covary with reduced adaptive functioning in the autistic population, we strongly caution against overinterpretation of these findings due to the substantial limitations of quantifying response to pain in autism based on solely reports from caregivers [
164,
165]. Although a co-occurring diagnosis of intellectual disability or more significant impairments in adaptive behavior may be more common in individuals with additional rare neurological conditions that truly include insensitivity to pain as a symptom (e.g., congenital insensitivity to pain with anhidrosis [
166]), it is also quite possible that proxy reporters such as caregivers underestimate the pain or discomfort of autistic children who are not able to communicate their internal states in typical ways [
165]. With the recent development of methods to better capture the internal pain experiences of autistic individuals with intellectual disability and/or limited language [
167], additional work is greatly needed to determine whether caregiver reports of seeming insensitivity to pain correspond with self-reports of pain experience in this population, providing more conclusive evidence for or against the claim that autistic individuals with more significant adaptive/functional impairments are truly less sensitive to pain and temperature than autistic individuals who are more cognitively- and adaptively-able (vs this difference being driven by atypical
communication of pain or distress).
Overall, the findings of the current study with regard to studied HYPO subconstructs suggest that Speech HYPO and Pain/Temperature HYPO represent theoretically distinct aspects of the autism phenotype with completely non-overlapping significant correlates and divergent future directions relevant to construct validation. Therefore, for applied researchers hoping to investigate these aspects of the autism phenotype using caregiver-report questionnaires, we strongly recommend that these two HYPO subconstructs in particular be studied at the single-modality level, as the nuanced associations between modality-specific variables and external correlates may be obscured by the use of supra-modal HYPO scores that combine subconstructs into a single variable when assessing individual differences. Notably, it is currently unclear whether these HYPO subconstructs demonstrate equally divergent patterns of external correlations when measured using other techniques (e.g., clinician observation [
84,
85]), and this remains an important avenue for future research.
Strengths and limitations
The current study had a number of strengths, including its very large sample size, representation of autistic children and adolescents across a wide range of ages and developmental levels, sensory phenotyping of many combinations of modalities and response patterns with widely used caregiver-report measures, and state-of-the-art statistical approaches that allowed for the pooling of partially overlapping sensory item scores and evaluation of between-dataset effect heterogeneity. However, it was not without limitations. Most notably, the studies that comprised the dataset utilized vastly different methods; each had substantially different inclusion/exclusion criteria, geographic locations, and assessment batteries. To allow for maximal pooling of similar data across studies, we combined measures of the same construct (e.g., different versions of the same questionnaire, standard scores on different measures of an ostensibly similar construct such as FSIQ or internalizing symptoms) into single variables, potentially introducing additional heterogeneity due to noninvariance between the different measures or measure versions. For sensory constructs, this pooling was also done at the item level to allow for different versions of the same measure (i.e., SP1 and SP2, SEQ-2.1 and SEQ-3.0) to be calibrated on the same latent scale using IRT. Though many items on different questionnaire versions were extremely similar, version-specific changes in anchor wording, item stems, or order effects could theoretically have resulted in noninvariance of the homologous items, again increasing overall heterogeneity. Nevertheless, the random-effects IDA model utilized in the current study allowed for the heterogeneity of each effect to be quantified (i.e., using ICCs and prediction intervals), thereby helping to contextualize both the population summary effect and the range of possible effects observable under different study conditions. The measurement of many sensory subconstructs was also a limitation, as despite the large initial item pool, a number of subconstructs had relatively few initial indicators and were, therefore, difficult or impossible to form into viable unidimensional scales from the start. For constructs that could not be operationalized in the current study using a psychometrically adequate unidimensional scale, we opted to use an ad-hoc single item indicator (or in one case, a doublet) such that these subconstructs would still remain in each supra-modal bifactor model. However, it is unclear whether the use of single-item indicators partially contributed to the psychometric inadequacy of the higher-order HYPO and SEEK constructs, and future studies in which all modality-specific subconstructs are adequately captured are necessary to rule out poor subconstruct measurement as a potential cause of supra-modal construct invalidity for sensory response pattern scores.
Another major limitation of the current study was the fact that multisensory items were removed from the questionnaires before psychometric analyses were undertaken. Although this choice greatly simplified the bifactor models and their computation due to the lack of specific-factor cross-loadings, it is notable that “real-world” sensory experiences are inherently multisensory in nature [
168]. By removing items containing multiple sensory modalities, we may have inadvertently excluded a number of relevant sensory behaviors in real-world contexts from the measurement models, limiting the content validity of the supra-modal constructs operationalized by the general HYPER, HYPO, and SEEK factors. Though it remains unknown whether these items would have been retained in our models based on psychometric criteria or excluded due to misfit, future studies are warranted to investigate the utility and properties of sensory reactivity bifactor (or indeed more complex hierarchically structured) models that include multisensory items in addition to single-modality subconstructs.
As an additional limitation, the questionnaires used in the current study were all based on caregiver-report of a child’s behavior; even in cases where autistic individuals were capable of self-reporting on their own sensory experiences (or provided such data), this information was not included in the current investigation. As sensory percepts are fundamentally subjective experiences, reports solely based on the observations of untrained proxy reporters (i.e., caregivers) may be capturing only the most extreme and/or distressing sensory reactivity differences, potentially also introducing confounding according to the child’s language or communication ability (see also [
169]). Moreover, it is quite possible that our conclusions regarding the inadequacy of supra-modal HYPO and SEEK scores (and/or the adequacy of supra-modal HYPER scores) are limited to caregiver-reported sensory measures, and additional work is needed to test the appropriateness of such scores using other measurement methods, including self-report (e.g. [
170,
171]), clinician-rated behavioral observation (e.g. [
84,
85,
172,
173]), and parent/caregiver interview (e.g. [
84]) tools. Ideally, further studies of autistic youth and adults capable of self-report should attempt to utilize multimodal sensory measurements that include both self-
and informant-reports simultaneously (e.g., [
174]; see also [
84] for a measure combining clinician observation with caregiver interview), therefore allowing both an individual’s internal experience and observed behavior to contribute to their ratings of sensory reactivity. For the subpopulation of autistic individuals who cannot reliably report on their own experiences (e.g., very young children, individuals with severe/profound intellectual disabilities, many of the individuals labelled with so-called “profound autism” [
175,
176]), multimodal measures of sensory features remain just as important, despite self-reports being inaccessible, and we strongly encourage researchers to consider alternative ways to augment parent or caregiver-reported sensory questionnaires when examining sensory differences in this particular segment of the autistic population (e.g., [neuro]physiologic measures, behavioral observations, clinician ratings, parent/caregiver interviews, cognitively accessible psychophysical tasks [
177‐
181]).
Considering the statistical limitations of the study, it is worth noting that all associations between caregiver-reported sensory reactivity differences and clinical/demographic variables were estimated using models that did not control for other relevant demographic or clinical variables (e.g., age, sex, IQ/DQ, or language level). Thus, the meta-analytic correlation estimates in our current study may overestimate the strength of hypothesized effects due to the presence of (often substantial) residual confounding [
160]. On the other hand, the current study only examined unconditional, linear associations between variables; therefore, it is also possible that the strength of any conditional and/or nonlinear relationship was underestimated. Future work should attempt to quantify such effects of various sensory predictors on relevant clinical outcomes over and above other potentially confounding variables (e.g. [
103]). Lastly, it is notable that the current investigation relied solely on cross-sectional data, limiting our ability to draw conclusions regarding potential developmental trends in sensory features or the predictive validity of sensory reactivity for other relevant outcomes. Although some studies have begun to demonstrate the predictive utility of sensory reactivity in autistic children and other populations such as infants at elevated likelihood to develop autism (e.g. [
182‐
187]), these studies have largely used supra-modal response pattern scores; therefore, additional large-scale, longitudinal studies are necessary to determine which single-modality sensory subconstructs (or combinations thereof) can be utilized as clinically-relevant predictors of core and frequently co-occurring features of autism.