Introduction
Attention to faces and eyes gives insight into the emotional and mental states of others (Baron-Cohen et al.,
2001; Peterson & Eckstein,
2012). Following cues such as gaze direction can help determine the focus and intentions of another person and infer meaning and context (Frischen et al.,
2007; Senju et al.,
2008). Modulation of attention to socially relevant stimuli is an essential element for engaging in successful social interaction (Freeth & Bugembe,
2019).
A core diagnostic feature of autism spectrum disorder (ASD) is impaired social interaction. There is extensive evidence that individuals with ASD show decreased attention to socially informative elements of visual scenes (e.g., people, faces, and gestures) compared with typically developing (TD) peers (Chita-Tegmark,
2016; Dawson et al.,
1998; Frazier et al.,
2017; Guillon et al.,
2014; Klin et al.,
2002). The cause for this decreased attention is likely multifactorial, particularly given the heterogeneity of clinical presentation of ASD. Researchers have postulated a number of contributing factors, ranging from biases in social perception, issues of motivation or salience (i.e., interest or preference), to perceptual or cognitive biases (Del Bianco et al.,
2018; Hessels et al.,
2018).
Eye-tracking studies are typically used to examine differences in gaze patterns between individuals with ASD and their neurotypical peers, or other developmental and psychiatric disorders. The main focus of these studies has been to assist with early detection and to understand early development. There are some indications that limited social attention may be predictive of later diagnosis and symptom severity in ASD (Campbell et al.,
2014; Chawarska et al.,
2013; Jones & Klin,
2013). A large meta-analysis of gaze differences between ASD and control participants in eye-tracking studies found that the overall effect sizes remained consistent across ages, indicating that social attention differences develop early and remain consistent throughout adolescence and adulthood (Frazier et al.,
2017). However, studies vary widely in rigor, nature and complexity of visual stimuli. The meta-analysis revealed that methodological issues [for example, region-of-interest (ROI) definition, and differences in the type of stimuli used, such as complexity of social interaction] influenced effect size of the apparent deficit, regardless of age (Frazier et al.,
2017). Other studies have indicated that results may be influenced by contextual factors and stimulus design, and that, in some contexts, individuals with ASD may have similar viewing patterns to those without ASD (Chevallier et al.,
2015; Guillon et al.,
2014; Hanley et al.,
2015; Kwon et al.,
2019). A recent comparative study looked specifically at developmental changes in attention to social interaction in a large sample of ASD and TD individuals across a range of paradigms that had previously demonstrated between group differences in social attention (Fujioka et al.,
2020). They found that whole group differences in viewing patterns were not replicated in all paradigms. Specifically, the static faces with moving or non-moving mouth showed no diagnostic group differences in attention to eyes, but did show an interaction between group and age, indicating that there are qualitative differences in allocation of social attention across ASD development (Fujioka et al.,
2020).
There is an increasing interest in the use of eye-tracking measures of social attention as potential biomarkers in clinical trials, as indicators of the presence or/and severity of ASD symptoms (Murias et al.,
2018). In order for eye-tracking measures of social attention to be useful as biomarkers for stratification, diagnosis or change in symptom severity, they must be reliable and interpretable across the lifespan of ASD. One way to test social attention for its potential as a valid biomarker of ASD is to develop and use the same paradigm consistently within different age groups. For example, several research groups report on a set of dynamic stimuli where context was manipulated to include four experimental conditions comprising activity without speech, viewer-directed speech, joint attention, and moving non-social object (a toy), against a background of non-social potential distractor stimuli (Campbell et al.,
2014; Chawarska et al.,
2012,
2013; Wang et al.,
2018). Use of different experimental conditions enabled investigation of contextual factors that modulate differences in allocation of attention within the groups of toddlers with or without ASD. Infants aged 6 months who were later diagnosed with ASD showed limited spontaneous attention to social scenes, particularly the actress’ body and the actress’ face, compared to other infants who did not receive the diagnosis later. For infants, the findings were across all contexts or experimental conditions (Chawarska et al.,
2013). In contrast, in toddlers with ASD, in experimental conditions without eye contact and speech, attention distribution was similar to that of TD and developmentally delayed (DD) control participants. Only the dyadic bid condition, where the actor engaged in direct speech and eye contact, resulted in differences between the diagnostic groups. Toddlers with ASD showed less time looking at this scene overall, and specifically less time looking at faces and mouths compared to TD and DD controls (Chawarska et al.,
2012). Using the same data and a different analytic approach (High-Cohesion Time Frames) to quantify atypical gaze, the ASD group gaze was found to be the least typical during the dyadic bid condition, and further that the atypicality was associated with more severe ASD symptoms (Wang et al.,
2018). Similarly, in relation to symptom severity, increased attention to eyes and mouths at 2 years of age was shown to be positively related to functional outcome in the ASD group 1 year later (Campbell et al.,
2014).
In this study, we employed an alternative version of the Chawarska’s paradigm (Plesa Skwerer et al.,
2019) for the first time in older children and adults with ASD. The primary goal was to examine whether older children and adults with ASD differed from neurotypical peers in allocation of visual attention when viewing dynamic social videos. Specifically, we were interested in whether there was a potential modulation of those differences in gaze duration and quality by context. Based on previous findings in toddlers, we hypothesized that children and adults with ASD would spend less time looking at faces and face core features than the TD controls, with this difference being more prominent in the context with viewer-directed speech and eye contact. We also tested for relationships between the level of visual attention allocated to different experimental conditions or context and severity of ASD symptoms, as captured by several social behavior rating scales. Our hypothesis was that severity of ASD symptoms would correlate with time spent looking at faces.
Discussion
This study examined spontaneous visual attention to dynamic stimuli across specific experimental conditions in participants with and without ASD, aged 6–63 years. Attention to the screen, measured by %
Valid Time, was above 85% on average in both the ASD and TD groups, suggesting good task attention overall. There was no significant difference between the two groups for attention to the screen between the four experimental conditions, and both groups showed similar patterns of visual attention allocation across ROIs in the four conditions. Specifically, in relation to social attention, there was no significant difference in average looking time at faces between the groups in the
Dyadic Bid condition, contrary to what was previously reported in toddlers with ASD (Chawarska et al.,
2012). Both ASD and TD groups allocated more visual attention to faces during this condition. However, unlike the findings in toddlers, significant differences in attention to faces were seen in the
Activity and the
Joint Attention experimental conditions. This indicates that while social attention is reduced in older children and adults with ASD, compared to TD groups, there are qualitative differences in the allocation of visual resources across the developmental trajectory of ASD. Differences in experimental conditions, or context, do not impact allocation of social attention in the same way across different age groups in ASD.
We also replicated findings that within general viewing of the face the ASD group paid less attention to eyes and mouths. However, similarly, this difference was only statistically significant in one specific experimental condition (Activity condition), where the ASD group paid less attention to the mouth area.
Consideration of the context in which patterns of eye gaze differ across age may help with understanding the mechanisms and developmental trajectory of social attention in ASD. The
Activity condition was distributed throughout the duration of the video stimuli, but the vast majority of activity time was in the first portion of the videos. Individuals with ASD did attend to the faces during this time, but the TD group attended longer. It may be that the result of competing non-social factors, which were novel at the beginning of the videos, was that the ASD group did not prioritize attention to faces at the point at which they were first exploring the visual scene. This would be consistent with a previous report (Kwon et al.,
2019) that hypothesized that the presence of distractors rather than attention to faces was a driving factor moderating differences in viewing. The authors suggest that competition between faces and external distractors might be a more robust measure than attention to faces itself.
Salience of non-social stimuli may also explain the differences seen in the Joint Attention condition. This experimental condition always preceded the Moving Toy condition, in which the actor was looking at a static toy while the same toy as previously was moving. Therefore, the novelty of the toy moving in the Joint Attention condition might have been more salient and attracted more attention from individuals with ASD. In the second condition with the moving toy, the actors’ face may have not had the same competition for salience and therefore no differences were observed.
Unlike the studies in toddlers (Campbell et al.,
2014; Chawarska et al.,
2012), the
Dyadic Bid condition did not reveal any differences in attention to faces. This may be a result of developmental changes in one or both groups, either with experience or training or both. For example, improvements in processing of social information in TD individuals may mean that less attention is given to faces, as pertinent information can be garnered in a shorter time (i.e., they infer more information from faces more quickly). In contrast, some individuals with ASD may have learned or been specifically taught the relevance of attention to looking at a person who is talking to them, thus resulting in longer gaze time. Alternatively, increased efficiency of social attention in the TD group may not be picked up when using ROIs to analyze the eye-tracking signal. This hypothesis requires additional eye-tracking features to determine whether focus on the amount of viewing time might have masked other differences potentially existing between the two groups. Some evidence comes from a time-course analysis study that showed no difference in viewing time for faces between the ASD and TD groups but found that the TD group was quicker to look first and then look less as time went on (Freeth et al.,
2010). Other studies with older individuals with ASD also indicate that although total looking time may be the same, differences may be found in the timing and slower response to socially informative elements in the TD group (Caruana et al.,
2018; Fletcher-Watson et al.,
2009; Frost-Karlsson et al.,
2019). In terms of impact on social functioning, less urgency to look to faces might mean that important information is missed.
The
Activity condition revealed more looking at the mouth by the TD group, which was also the reverse of the results observed in toddlers, where mouth viewing has been related to poor outcomes in the ASD group (Chawarska et al.,
2012). Here, however, in older children and adults, mouth viewing may be an effective strategy that has developed in TD individuals—scanning of face and attention to mouth in anticipation of verbal communication. In contrast, studies comparing ASD and TD adults in both real-life or online Skype situations have found increased attention to mouths by ASD participants in certain contexts, for example when discussion relates to emotional factors (Hutchins & Brien,
2016). The relationship between attention, as manifested in
% looking time, to the core features of faces is complex in ASD. A multitude of bottom-up and top-down processes may potentially influence the allocation of social attention in both groups, and it is likely that these become more sensitive to modulation by context with development and learning between toddlerhood and older child and adulthood.
Previous studies reported significant relationships between severity of ASD symptoms, as captured by a variety of behavior rating scales, and eye-tracking based measurements (Frazier et al.,
2018; Murias et al.,
2018). We hypothesized that similar relationships could have been established in our data. However, only a few relationships, not many more than would be expected by chance, were identified. A recent comparative study also did not report substantial relationships between behavior rating scales and eye-tracking behavior (Fujioka et al.,
2020). This lack of consistent finding of relationships may be due to, as the authors propose, the scale measures capturing a wide range of social deficits, of which eye looking or contact is only one. Further, it could be that allocation of social attention, measured by eye-tracking, is a different construct to that captured in behavioral report and provides something additional to assessment of social attention. In the context of biosensors such as eye-tracking for use in clinical trials it is important to compare how both direct and behavioral report measures change over time to determine whether one may be more sensitive as an outcome measure than the other, and also whether short-term biosensor changes may lead to longer term changes in observable behavior as reported in scales.
The purpose of this study was to investigate the use of a paradigm that has been well established in one age group (toddlers) in the literature (Chawarska et al.,
2012,
2013,
2016; Wang et al.,
2018) in a group of older individuals with ASD in order to explore its potential as a biomarker in clinical trials. Often the paradigms designed to measure social attention typically minimize the number of examined parameters in order to facilitate comparison between the ASD and TD groups. As a consequence, a possible limitation is that this increasingly results in a situation when a tested paradigm does not resemble a real-life social interaction. In particular, the paradigm tested in the current study was developed to be suitable for younger children. Although we were able to establish persistent differences in social attention between an older group of TD controls and ASD individuals, the tested paradigm in older children and adults may be less reflective of a real-life social interaction than it may be for toddlers (Macari et al.,
2021). Therefore, the differences observed may be anomalies of the paradigm rather than typical viewing patterns of either group in real-life (Grossman et al.,
2019; Hanley et al.,
2015; Risko et al.,
2016). Furthermore, by combining the response to both naturalistic and experimentally controlled or probe paradigms might lead to more helpful characterizations of social attention in ASD that could help with subtyping for clinical trials, similar to the approach adopted for the development of the Autism Risk Index (Frazier et al.,
2018). In addition, albeit helpful and important in replication of other eye-tracking studies, the use of ROIs may mask other differences in features that may be more consistent across ASD development, could be more closely related to ASD symptoms, or more sensitive to change. Such additional features could reflect scan-path length and recursion during exploration of a visual scene as well as fixation rate (Heaton & Freeth,
2016), duration and frequency of saccades (Vabalas & Freeth,
2016), and variability of gaze patterns (Avni et al.,
2020). Future studies should include these features to ensure that differences and similarities in viewing patterns are captured and characterized more fully.
In conclusion, the present findings support the general observation that eye-tracking studies using ROI can demonstrate differences in allocation of visual resources to social scenes in individuals with ASD compared to TD groups. However, the differences in allocation are context specific, and, depending on the experimental condition, there may be no differences between the groups, or the results could differ by age of the participants. Studies such as this can contribute to our understanding of the developmental trajectory of social attention in ASD, as well as support the process of identification of biomarkers for clinical trials. For utility of eye-tracking and social attention as a potential biomarker we need to select paradigms that include context and probes that have been shown to detect differences in the target age group. In addition, we should consider features of eye-tracking beyond ROI that may capture more information about the qualitative differences in response to social stimuli.
Acknowledgments
This study was funded by Janssen Research & Development, LLC, USA The authors thank the study participants and the following investigators for their participation in this study: Arizona: Christopher J. Smith, PhD; California: Bennett Leventhal, MD and Robert Hendren; Connecticut (at the time of study conduct): Frederick Shic, PhD; Massachusetts: Jean Frazier, MD; New Jersey: Yvette Janvier, MD; New York: Russell Tobe, MD; North Carolina: Geraldine Dawson, PhD; Pennsylvania: Judith S. Miller, PhD; Washington: Bryan King, MD. The authors also thank the following Johnson & Johnson colleagues for their involvement in the study: Seth Ness, Matthew Boice, and Andrew Skalkin. Priya Ganpathy, MPharm CMPP (SIRO Clinpharm Pvt. Ltd.) assisted with preparing the manuscript for journal submission and Ellen Baum, PhD (Janssen Global Services, LLC) provided additional editorial support.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.