A personal introduction to early identification, diagnosis and treatment of autism

In my 35-year career as a clinician and di agnostician, I have had the privilege of being trained and then worked with some luminar ies in diagnostics and clinical care of autistic children such as Donald J. Cohen, Fred R. Volk mar, and Sara S. Sparrow. I have also had the opportunity to build programs of care, science and training in this clinical domain, including one of the largest centers of its kind in the US. I have also been proud of leading clinical and research teams that adhere to the most rigor ous standards of diagnostics, which emphasize the expert use of reference standard diagnostic and developmental tools and the deployment of clinicians with many years of clinical training and experience in the field. Most of all, I have achieved some professional fulfillment for serv ing many families who reached out to us, and we could serve, in the two academic health centers of excellence to which I have been af filiated throughout my professional life. And yet, the public health challenge that this domain of clinical care presents at a societal level is over whelming and universal. It traumatizes families with endless waits and leads to missed opportu nities to optimize the outcomes of the majority of autistic children. Autism is a lifelong condi tion, which can be associated with severe dis abilities and challenges to child and family, and to costly burdens to the healthcare, educational and supports systems. If we believe that the children we do not see and serve are as important as the children that we do see and serve, there is an urgent need to recognize that academic health centers of excellence sometimes operate as an insolated, content with their rigorous standards but virtually inaccessible and irrelevant to the needs of the community at large, wherein most of the over 95 000 children born every year in the US who will have autism are not identified and diagnosed in their first three years of life. In this way, these many thousands of children will fail to benefit from early treatment and in terventions that may afford optimized lifetime outcomes. This fact is the impetus for the work described in this manuscript.

The public health crisis of early identification, diagnosis and treatment of autism

Autism spectrum disorder (hereafter autism) is a neurodevelopmental disability diagnosed behaviorally by the presence of early emerg ing and persistent deficits in social interaction and communication skills, and by the presence of restricted and repetitive patterns of behav ior. Early identification and treatment are two of the most important factors promoting improved lifetime outcomes for children with au tism¹. Because up to 80% of parents recognize developmental concerns in their autistic chil dren by age 2 years, the American Academy of Pediatrics recommends universal screening for autism at 18 and 24 months² during well-child visits. More recently, many health insurance payers have made screening for autism a re quirement for coverage of the well-child visits at these ages. This, in turn, has led to a dramatic upsurge in the uptake of early screening for au tism by primary care pediatricians³. Although the most widespread screening tool for autism may still miss more children than it identifies, (with a rate of up to 60% of false negatives)⁴, for the large number of toddlers who screen posi tive, the main obstacle to accessing services has become the limited access to diagnosticians^5,6. Most diagnostic programs have long waitlists⁷, contributing to the fact that only one in every 5 autistic children in the US is diagnosed before the age of 3 years⁸. Delay in diagnosis delays opportunities to support children and families at early ages, when the brain is still most mal leable, treatment can provide optimal benefits, and before maladaptive behaviors may become entrenched^9,10. With a prevalence rate of 1 in 36, autism is the most common complex neurode velopmental disorder¹¹. Therefore, as a result of this challenge, many thousands of families are deprived from early intervention services every year, including services that should be available to them via federally mandated programs cre ated to support any child with developmental concerns such as autism.

Clinicians’ challenges in early diagnosis of autism

Expert diagnosticians are trained to perform extensive observations of children using stan dardized tests, to elicit and review extensive in formation from parents and other sources, and then to reach a diagnostic assignment based on the totality of information available in a process that may take many hours and sessions. This la borious and costly model limits the number of children clinicians can diagnose, leading to wait lists, typically of many hundreds of families, and delays that may extend to months and years. Yet, most diagnosticians know that a very large percentage of these referrals have autism and, in some programs, the very referral to an autism clinic may carry an autism diagnostic predictive value of up to 80% to 90%. Therefore, maintain ing these children in waitlists is unconscionable, and yet it is a burden experienced by most diag nosticians in the field. In addition, the diagnosis of toddlers can be very challenging, as symp toms are still emerging, developmental changes are occurring rapidly, and clinical tools are sub jective, leading to the fact that in up to 30% of all cases, clinicians feel suboptimal confidence in their diagnosis¹². These various challenges ex perienced by clinicians are exacerbated further by the knowledge that the families most likely to have limited access to their services are those from minority, low-income and rural communi ties¹³, resulting in a system that is, overall, inequitable, unethical, and unacceptable.

Biomarker-based solutions to greater access to early diagnosis of autism

These public health and clinical challenges have accelerated research on biomarker-based solutions intended to make the diagnostic pro cess more efficient and more accessible, while retaining high quality and the assurance of ac curacy. One such solution, called EarliPoint, an eye-tracking based tool for early diagnosis and assessment of autism, was recently cleared by the FDA for broad clinical use. EarliPoint leverag es an objective and quantitative biomarker -so cial visual engagement-or the way children look at and learn about their social environment, which directly reflects core features of emerging social disability⁹. Prior research indicated that dynam ic quantification of social visual engagement has neurobiological and clinical face validity: individual variation in social visual engagement reflects individual genetic variation¹⁴; and au tism-related differences^14,15 in social visual en gagement are developmentally early emerging, and are predictive of later autism diagnosis¹⁵.

Two recent simultaneously published re ports provided clinical validation data for Ear liPoint^16,17. Pooled results across feasibility and pivotal studies involving over 1600 toddlers (age 16-30 months) and two independent replications indicated strong diagnostic performance: accu racy of 84.0%, sensitivity (se) of 80.6%, specific ity (sp) of 87.0%, positive predictive value (PPV) of 84.4%, and negative predictive value (NPV) of 83.7%. EarliPoint also yielded accurate proxies for indices of social disability, (capturing 74.5% of the variance of the Autism Diagnostic Observa tion Schedule, 2nd edition), and for indices of ver bal and nonverbal ability (capturing 79.5% and 69.0%, respectively, of the variance of the verbal and nonverbal age equivalent scores of the Mul len Early Scales of Learning). EarliPoint is intended to provide evidence-based clinical information to the diagnostician, in the form of a quantita tive diagnostic classifier (of autism vs. not-au tism) and quantitative indices of severity, all of which should inform clinical decision-making regarding modality and intensity of treatment and interventions.

Context of use of EarliPoint in early diagnosis and assessment of autism

EarliPoint was not intended to replace clini cians. Autism is a consequential diagnosis, and it is critically important that families have an opportunity to discuss the implications of autism with clinicians, and to avail themselves of supports to translate the results of an evalua tion into a plan for treatment and access to the services and supports they need. EarliPoint was designed to provide a clinician with science-based, objective and quantitative information on a toddler’s diagnostic and developmental sta tus. The procedure takes on average 12 minutes, as a toddler watches in a tablet the video scenes of other children interacting while eye-tracking data are collected at a rate of 120 times per sec ond. Upon completion of the procedure, data are uploaded to a secure server and are processed through a data quality and analysis pipeline. Some 15 minutes later, a clinical report is issued to a clinician portal, containing a diagnostic classifier of autism or not-autism, including the predictive probability of the classifier: a positive predictive value if the diagnosis is autism, or a negative predictive value if the diagnosis is of not-autism. The clinical report also includes the three indices of severity: levels of social disabil ity or autism, and age equivalent levels of verbal and nonverbal ability.

EarliPoint was intended to make the diagnos tic process more efficient and less costly, and, in this way augment capacity of, and increase access to, diagnostic services. Because the tool was tested against expert clinicians conducting rigorous evaluations, while using standardized testing, the goal of EarliPoint is also to make high-quality diagnosis more accessible to fami lies by broadening the pool of clinicians who can offer diagnostic services, from the limited number of highly experienced clinicians to a larger number of less specialized clinicians who can nevertheless support families’ access to the services they need now. In our experience, Earli Point can reduce the number of hours required for diagnostic evaluations by almost a third, and in this way, free clinicians to spend more time with families while also allowing them to aug ment the volume of their services, reduce wait lists and avoid economic losses, which could jeopardize the sustainability of their program.

The successful adoption of EarliPoint requires adjustments of patient flow and procedures to maximize its clinical utility. To optimize the clin ical utility of the tool, these adjustments need to be made in a given context of use or clinic. One current model is to complete some proce dures prior to an in-person visit to the clinic. Specifically, trained clinical staff can perform a diagnostic interview with parents via telehealth, and a trained technician can complete the Ear liPoint protocol at the point of easiest access to the family, in a clinical setting or at home. The diagnostician then reviews the clinical informa tion available -EarliPoint’s results, the diagnos tic interview, and a review of a child’s electronic health record - in preparation for the in-person evaluation. During the in-person evaluation, if the totality of information is clearly consistent with a diagnosis of autism, the clinician can proceed to explore the needs for any behavioral management supports (e.g., how to address a child’s problem behavior) or additional refer rals (e.g., for genetic testing), and move forward with a parent conference in which the focus is on supports the child and family need, and on how to pursue those in their community. If the totality of information is either not entirely consistent with autism, or is consistent with a non-autistic developmental delay, the clinician may choose to conduct additional but abbrevi ated behavioral observations to disambiguate the diagnosis and, typically, to identify the form of non-autistic developmental delay a child may have (e.g., a language, a cognitive, or a global de velopmental delay).

A key consideration in the context of use, or where EarliPoint is being deployed, is the quan titative predictive values associated with the diagnosis it returns. Fundamentally, a diagnos tic test is evaluated by its ability to classify pa tients correctly. While the performance of any diagnostic test derives from statistical mea sures obtained in clinical trials, some of these measures are more important to clinicians than others. Sensitivity (se) and specificity (sp) refer to the intrinsic ability of a test, independent of population context (i.e., independent of where it is used), to correctly identify those with the diagnosis (se) and those without the diagnosis (sp) when compared with the performance of a gold standard such as expert clinicians conduct ing an extensive evaluation. While informative to those interested in overall performance of the test, se and sp do not provide what a clinician needs when using the test to make a diagnosis in the case of an individual child. In such real-world implementation context, it is the Positive Predictive Value (PPV) and Negative Predictive value (NPV) that describe the proportion of pa tients with positive or negative test results who are identified correctly. That is the reason why EarliPoint’s results include a PPV statement (if the diagnosis is autism) and an NPV statement (if the diagnosis is not autism) with the precise probability that the results return for the indi vidual child in the specific clinic where the eval uation is taking place. Of great importance, PPV and NPV are influenced strongly by the preva lence of the autism in the clinic where the test is being used (e.g., the proportion of children evaluated in the clinic who typically receive the diagnosis of autism)¹⁸. Equipped with this quan titative information --for example, that a child meets criteria for autism with 92% of probabil ity-a clinician can decide whether this level of certainty is acceptable or whether addition al information, or testing, are needed. Because these PPV and NPV considerations apply to any diagnostic test in medicine, and because some clinicians might not be aware of these statistical facts, or might not be used to data-driven clini cal decision-making processes, EarliPoint has the potential of aiding clinicians who are less specialized or experienced in autism diagnostics to achieve higher levels of accuracy, approaching the rigor that defines gold standard diagnostic evaluations in autism.

Opportunities and challenges ahead

EarliPoint has the potential to meaningfully address the limited access to high-quality di agnosis and assessment of autism by replacing costly procedures and by providing clinicians with objective, quantitative and accurate clini cal information. In this way, it will create effi ciencies in clinical care, reduce costs, augment volume, and make high-quality diagnosis more available to families. However, there will be chal lenges to its adoption. For example, because cli nicians have essentially been diagnosing autism in the same fashion, subjectively, via behavioral observations, since Leo Kanner first described autism in 1943, it is unlikely that expert clini cians will change their clinical procedures over night by adopting biomarker-based tools. Ear liPoint achieves quantitative results of clinical utility and issues a clinical report in about 25 minutes. This compares with several hours of testing and evaluation required by experienced diagnosticians. Therefore, we expect that in aca demic centers of excellence, confidence in Ear liPoint’s results is likely to first come from the use of EarliPoint in parallel to current diagnos tic models, allowing clinicians to acquire their own sense of trust in the tool. This reluctance, however, is unlikely to be shared by clinicians who work in large, typically public programs of treatment and services, whose ability to serve families depends on diagnosis, and their access to expert diagnosticians is extremely limited.

The contrast of these two attitudes defines the main clinical policy discussion ahead of us: should we maintain our traditional models of autism diagnosis, despite of the fact that most families do not have access to it? Or, should we instead leverage science-based tools to increase objectivity, efficiency and accessibility to diag nosis, even if this is new? The answer to that question is likely to depend on where a given stakeholder stands in the ecosystem of autism diagnosis and care. If your perspective is that of a parent who knows that “earlier is better”, but who also knows that access to diagnosis may take months, maybe years, if ever, the choice should be clear.

Beyond EarliPoint: in search of biomarkers to guide intervention in autism

While EarliPoint is the first diagnostic and as sessment biomarker in autism cleared by the FDA, the field of biomarker research in autism has continued to grow exponentially in the past few years, with, however, limited success. In a recent review¹⁹, evidence for 940 biomarkers was synthesized covering molecular biomark ers (including cytokines, growth factors, mea sures of oxidative stress, neurotransmitters and hormone), neurophysiology (such as EEG- and eye-tracking-based), and neuroimaging (such as functional MRI). The conclusion was that there is currently no response biomarker with suffi cient evidence to inform autism clinical trials. In another wide-ranging systematic review²⁰, extending the focus from autism to neurode velopmental disorders more generally, 780 stud ies covering biochemical, genetics, neuroimaging, neurophysiological and neuropsychological studies were assessed, and no evidence for a validated biomarker was identified that met the authors’ criteria of representing consistent find ings in two or more studies from independent research groups, and sensitivity and specificity exceeding 80%. An additional concern in bio marker-based research, in autism and in other areas of medicine, and particularly about tools or procedures leveraging artificial intelligence (AI) analytic methods, is the methodological pit fall of “overfitting”. This phenomenon refers to the possibility of achieving high accuracy results via machine learning, sometimes with sensitivi ties and specificities that approach perfect per formance, but these results remain specific to a single cohort of participants, with no successful attempt to implement the algorithmic biomark er in an independent cohort of participants. Because of this danger, unless such AI-based biomarkers are successfully replicated in an in dependent cohort, their clinical utility remains unproven, and their potential remains sugges tive but not definitive²¹.

Beyond diagnosis

If only 1 in 5 autistic children is diagnosed with autism before the age of 3 years in the US, the advent of efficient and accessible biomark er-based diagnostic tools is likely to double and triple the number of children diagnosed early in their lives. This will shift the bottleneck to the treatment and support system, which is al ready grossly inadequate to provide services to all those who need it now. However, it would be cynical to propose that the status quo should not be improved because the current system will not have capacity to address increasing needs. There is already accumulating evidence that community-viable modalities of early treat ment such as caregiver-mediated interventions and inclusive therapeutic pre-school settings, can deliver effective results if services are de livered early in a child’s life. And yet, while the evidence for effectiveness already exists, there is a chronic shortage of providers, few incentives for these providers to pursue and stay in these careers, and mandates to cover the costs are un derfunded. It will take societal fortitude to in vest in such a new system of early brain health. While challenges are expected, the community of families impacted by autism deserve no less.