The importance of early diagnosis and treatment

Autism spectrum disorder (ASD) is a neurodevelopmen tal disability diagnosed behaviorally by the presence of early-emerging, persistent deficits in social interaction and communication skills, and by the presence of restricted and repetitive patterns of behavior¹. Early identification and treatment are two of the most important factors promot ing improved lifetime outcomes for children with ASD^2,3. Because up to 80% of parents recognize developmental concerns in their children subsequently diagnosed with ASD by age 2 years⁴, the American Academy of Pediat rics recommends universal screening for ASD at 18 and 24 months⁵. And yet, the median age of ASD diagnosis in the US has been late, between 4 and 5 years of age⁶. This public health challenge moves eligibility and access to intervention services from within the period of maximal neuroplasticity⁷ -the period from birth to 3 years- to a point several years hence, when many years of development have already played a large role in shaping the course of a child’s condition⁸. Age of diagnosis is later still in minority, low-income and rural communities⁹. Several fac tors account for late ages of diagnosis of ASD, including primary care clinicians’ time restrictions, lack of accurate and cost-effective screening and diagnostic tools, limited number of expert clinicians, and a general “wait and see approach” that results in ASD diagnosis being made only when symptomatic presentation is obvious, typically when the child is older¹⁰.

One solution explored in the research literature has been the use of eye-tracking technology to generate quantitative biomarkers that could serve as objective, standardized, quantitative and cost-effective tools for the diagnosis of ASD¹¹. For eye-tracking studies of ASD in infants and toddlers, the focus of research has been on looking behavior in social situations because disruptions in this domain represent one of the core features of the condition⁸. Here we summarize some advances in this area of research.

Social visual engagement in ASD

Social visual engagement (SVE) refers to the way that children look at and learn from the social environment, moment-by-moment, as they interact with others or view scenes of social interaction such as children’s play or people talking⁸. In the past decade, several discoveries have placed SVE at the core of programmatic research aimed to understand the mechanisms underlying the unfolding of autism in infancy and toddlerhood:

SVE emerges early in typical development and is disrupted in ASD: SVE is a foundational mechanism of socialization, attracting babies’ attention to others from the first days and weeks of life¹². Babies’ attention is at tracted to caregivers’ eyes, setting in motion a mutually reinforcing choreography of contingent, back-and-forth social interaction that is critical for the development of speech, language and communication skills. In babies later diagnosed with autism, this response is disrupted from as early as 2-months of life; developmental trajectories of eye looking show that deviations from normative patterns in the first 6 months of life are both predictive of an ASD diagnosis and of levels of autistic social disability at 24 and 36 months¹³.

SVE is under strict genetic control: ASD is one of the most strongly genetic and heritable complex neurodevel opmental conditions. Although several hundred genetic variants have been found to be implicated in ASD, the vast majority of cases result from polygenic combina tions of genes with small effect sizes, resulting in highly complex genetic vulnerabilities. To date, research has not yet revealed how molecular vulnerabilities can lead to the behavioral symptomatology that defines ASD. One approach to this problem has been the stance that rather than leading directly to symptoms, this genetic liability leads to disruptions in normative socialization, and thereof become ASD symptoms, which emerge in the second and third year of life⁸. Critical in this effort has been to establish what behaviors that are core symptoms of ASD are influenced by genetic variation and whether such behaviors are specific to the condition. In a large study involving identical twins (monozygotic or MZ) and fraternal twins (dizygotic or DZ), the former sharing 100% of their DNA whereas the latter sharing 50% of their DNA, typi cally developing MZ toddlers showed 91% agreement in eye looking behaviors relative to only 35% agreement in eye-looking behaviors in typically developing DZ toddlers. This large gap in agreement demonstrates that eye looking behaviors are under strict genetic control¹⁴. Moreover, this study also showed that, when watching naturalistic scenes of social interaction, MZ toddlers are more likely than DZ toddlers to make saccadic eye movements at the same time, in the same direction, and onto the same social content, with within-twinset concordant behavior happening at microscales of tens of milliseconds. Finally, this study also showed that SVE behaviors that are most strongly genetically determined are also the behaviors that segregate toddlers with ASD from typically developing peers, thus indicating that SVE is both genetic in origins and diagnostic for ASD.

Early disruptions of SVE in ASD leads to accumu lation of missed opportunities for social learning: The clinical implications of these findings became clear when space-time distributions of SVE were created, quantify ing moment-by-moment deviations of social attention in individual toddlers with ASD relative to large databases of SVE in typically developing toddlers⁸. When watching com plex scenes of social interaction, groups of children may show dispersed visual focus (when there is no element of great social salience in the scene), or they may show group convergence on a particular element of importance to understand that social situation (these are sometimes called “hot spots of socialization”, because the attention of an entire group of children is entrained to that spot of the screen at that moment in time). The resulting effect on space-time measurements of visual attention is called an ”attentional funnel”, because the group as a whole, during free viewing of the video, momentarily focus on the same spot of the screen at the same time. These attentional fun nels are naturally occurring and data driven, amounting to hundreds of statistically significant attentional space-time convergences, thus representing “experimental presses” against which it is possible to compare the time-varying visual fixations of children with ASD. This line of research has revealed that within a 6-minute free viewing of video depicting social scenes, toddlers with ASD diverge over 500 times from the space-time focus of their typically developing peers. Each one of these divergences rep resents a missed opportunity for social learning, since social looking behavior reflects the child’s social learning from scene content. The social attention of toddlers with ASD does not diverge from normative patterns only during this kind of experiment; it diverges in their real-life social experiences. As a result, toddlers with ASD appear to miss opportunities for social learning many thousands of times in a week of social exposure, accruing these deviations in ever larger numbers in the first three years of life. The accumulating consequences of these deviations is the emergence of social disability in ASD⁸.

In summary, SVE is early emerging in development, is under strict genetic control, and is disrupted in ASD from the first months of life, pointing to a mechanism of social disability that impacts social learning at microscales of tens of milliseconds and that accrues by rates of thousands of times in the first 2-3 years of life. Because SVE behavior segregates toddlers with ASD from their peers, it has di agnostic value and could possibly be used for the purpose of advancing objective diagnostic tools, thus contributing to ongoing efforts to reduce the age of diagnosis of ASD and to increase access to diagnostic services.

Translating measures of social visual engagement into solutions for objective and accessible early diagnosis of ASD

Best practices in the diagnosis of ASD in young children require experienced clinicians who administer standard ized procedures for measuring autistic symptoms (such as the Autism Diagnostic Observation Schedule, 2nd edition, or ADOS-2¹⁵) and for measuring developmental skills such as verbal and nonverbal learning ability (such as the Mullen Scales of Early Learning¹⁶). Together with medical, family and developmental histories, these clini cians use the totality of this clinical information in order to make the diagnosis. The great challenge in the field is that experienced clinicians are scarce, and because this process may take several hours and is very costly, large portions of the population wait many months, at times years, to access diagnostic services, and this challenge is particularly acute among low-resource families in the US10. Therefore, there are several efforts underway to use objective, performance-based procedures to develop tools that will increase access to diagnosis by circumventing the need for experienced clinicians conducting lengthy and costly clinical procedures¹⁷. The following is such an effort that capitalizes on the findings of research on social visual engagement summarized above.

A social visual engagement (SVE) diagnostic classifier: Using moment-by-moment measures of SVE, our group has been able to assemble large datasets of social looking behavior to complex social scenes in typically developing toddlers. These datasets become data-driven norms against which one can compare the SVE measures for a single tod dler with ASD. Because, as noted, the normative datasets contain many hundreds of statistically significant attentional funnels (moment-by-moment space-time normative conver gences in SVE), this procedure can quantify a single child’s divergences from these normative patterns many hundreds of times over the period of 10-minute free viewing of complex social scenes (such as in the case of videos of toddlers play ing and interacting). This level of quantification allows us to test whether the sum of these divergences for a single toddler with ASD -a measure of “relative entrainment” to normative SVE- could perform the function of a diagnostic classifier, in essence matching a child’s ASD or non-ASD diagnosis given by experienced clinicians.

Our group has conducted two large trials to date. The first was a feasibility trial involving 1,089 toddlers including both children with clinician-based diagnoses of ASD and of non-ASD (this group includes primarily children with non-autistic developmental delays as well as children without a diagnosis). Two-thirds of this sample’s data were used to develop a computation model for the diagnostic classifier (the discovery cohort), whereas one-third of this sample’s data were used for independent testing (the replication cohort). In this replication testing, by comparing outputs of the eye-tracking procedure to best-practice clinician-based diagnoses, one can measure the accuracy of the diag nostic classifier. Encouraging results in this study led to a multi-site pivotal trial of this diagnostic procedure involving 505 toddlers. The pivotal trial followed the Standards for Reporting of Diagnostic Accuracy Studies (STARD18), which provide best-practice parameters to improve the completeness and transparency of reporting of studies of diagnostic accuracy. This study was pre-registered in the US repository of clinical trials (accessed at www.clinicaltri als.gov, protocol NCT03469986). Results for this second, multi-site replication cohort, reproduced the encouraging results of the feasibility study, and led to a clearance is sued by the (US) Food & Drug Administration (FDA) for this tool and procedure to be used in the clinical realm.

Social visual engagement (SVE) indices of severity: While a diagnosis is critical for eligibility for, and access to, early treatment and intervention services, treatment planning for a given child also require measurements of disabilities and abilities, such as levels of autistic social disability and levels of verbal and nonverbal learning, since appropriate treatment programs need to not only address a child’s challenges (social disability) but also to build on the child’s assets (their ability to learn via verbal and nonverbal means)^2,3. Because SVE is one of the foundations of language acquisition and learning from others¹², our group explored the possibility of mining the SVE datasets generated by the feasibility trial described above for the development of eye-tracking-based proxies to measures of social disability, and of verbal and non verbal ability resulting from clinician-based assessments using the ADOS-2 and the Mullen. In other words, we used the normative SVE datasets to identify the moments of the video that best predicted social function and verbal and nonverbal ability (each one modeled separately) in these typically developing toddlers. Once the computation model was built, we tested its ability to proxy ADOS-2 and Mullen scores in toddlers with ASD. Results of the replica tion sample in the feasibility trials revealed that the eye-tracking-based indices of social disability and verbal and nonverbal ability could capture over 70% of the variance in scores of the ADOS-2 and the verbal and nonverbal subtests of the Mullen, respectively. In effect, therefore, these eye-tracking indices were good approximations of the clinical measures and could be used to replace the clinical procedures. These results were replicated in the pivotal, multi-site trial described above.

The future of biomarker-based tools for the early diagnosis of ASD

Results of the trials reviewed briefly here offer a science-based solution to increasing access to early diagnostic services while maintaining high quality parameters for the assessment of young children. The community uptake of this and other biomarker-based solutions, however, will depend on our collective success in overcoming systemic barriers that continue to delay diagnosis and make it less accessible, including the ways in which developmental concerns are identified (and often missed) in young children at a population level, and inadequate levels of reimbursement for early detection and diagnosis. To overcome these barriers, however, there will be a need for community-wide awareness and understanding of the profoundly deleterious public health implications of late diagnosis to the almost 90,000 children born in the US every year who will have ASD.