Facilitating Autism Diagnosis

Psychiatric Times, Vol 38, Issue 10,

Autism spectrum disorder often evades formal diagnosis until a child is 4 years or older, but these artificial intelligence and telehealth systems may be able to help facilitate diagnosis.

SPECIAL REPORT: CHILD & ADOLESCENT PSYCHIATRY

Clinicians generally screen children for autism spectrum disorder (ASD) twice, at ages 18 months and 2 years, during routine health maintenance exams. Unfortunately, this neurodevelopmental condition often evades formal diagnosis until a child is 4 years or older.1,2 To expedite diagnosis, investigators are exploring a variety of artificial intelligence (AI)–based diagnostic systems as well as telehealth- and sensor-based technologies.

Diagnostic Issues

ASD affects 1 in 59 children in the United States.3 In many circumstances, a child suspected of having ASD is referred to a multidisciplinary diagnostic team composed of multiple medical and nonmedical specialists. Thus, there are often lengthy wait times for an evaluation. If a child does not meet threshold criteria, the assessment is considered inconclusive and parents must wait until the child is older for repeat assessment.

Many factors may delay the diagnosis. Individuals in poor communities and members of racial minority groups often have limited access to the services.4 In addition, only 60% of pediatricians nationwide screen children for developmental delays despite American Academy of Pediatrics (AAP) recommendations to perform screens at 18- and 24-month wellness checks. Also, screening may not result in appropriate follow-up care. In one study, for example, only two-thirds of those who failed screening were referred for a diagnostic ASD evaluation.5,6 The screening tools themselves present challenges. For instance, it has been shown that the Modified Checklist for Autism in Toddlers – Revised with Follow-up (M-CHAT-R/F) has a sensitivity as low as 39% in detecting children with ASD.7 Lastly, the COVID-19 pandemic has resulted in significant delays in evaluations.

Many tools are available to screen for ASD, including the Ages and Stages Questionnaires, the Communication and Symbolic Behavior Scales, the Parents Evaluation of Developmental Status, the Screening Tool for Autism in Toddlers and Young Children, and the aforementioned M-CHAT-R/F (Table). Pediatricians often use M-CHAT-R/F at ages 18 months and 24 months; it queries parents regarding their child’s perception of others, use of gestures, and interactive eye contact, as well as vocal communication and their ability to interact with parents and children. The AAP recommends referral for a diagnostic evaluation when children screen positive or if the pediatrician is suspicious of the diagnosis.

The DSM-5 consolidated previous categories of ASD into 1. To meet ASD criteria, a child must have 3 of 3 symptoms relating to social communication and interaction, plus 2 of 4 symptoms relating to repetitive or restricted behaviors (Figure).8

This recharacterization and simplification of the elements needed for ASD diagnosis has facilitated its diagnosis by specialists. Many use a chart similar to the Figure as an intake screening form before proceeding with a diagnostic evaluation.

The Autism Diagnostic Observation Schedule (ADOS) test and/or the Autism Diagnostic Interview – Revised (ADI-R) can also be used to evaluate for a suspected diagnosis. The ADOS is an observational play and activity assessment that takes up to an hour to complete, while the ADI-R is a 93-point questionnaire that may take several hours to complete.9 Many developmental pediatricians and child psychologists utilize their own screening methods rather than performing a full ADOS or ADI-R, sometimes using such screening tools as the Social Responsive Scale and the Childhood Autism Rating Scale. Most importantly, although these tools help with diagnosis, ASD is a clinical diagnosis best rendered by clinicians with the training and expertise to do so.

Technologies Using Sensors

Children affected with ASD are a heterogeneous group, with varying degrees of functional limitations. Many have comorbidities that include seizures, attention-deficit/hyperactivity disorder, oppositional defiant disorders, sleep disorders, speech delay, and others. Researchers have long looked to sensor-based technologies, many employing AI algorithms, to screen for children with ASD. AI is particularly useful for identifying patterns within data, and it therefore can be advantageous in identifying markers associated with an ASD diagnosis. AI algorithms are only as good as the training datasets inputted into a diagnostic system, but they do have the potential to improve medical care.10

Although a complete discussion of sensor technologies is beyond the scope of this article, investigators have had some promising results using sensors to analyze facial expressions, vocalizations, touch sensitivity, eye tracking, movements, and interactions with robots—all of which may help identify children with ASD. Unfortunately, to date, none of these technologies have been proven to be sensitive enough to be used diagnostically.11

DTx-Based Tools

Developed over several years with National Institutes of Health grant funding, the Naturalistic Observation Diagnostic Assessment (NODA) was developed by Behavior Imaging Solutions. This is a digital therapeutic (DTx) that can facilitate the diagnosis of ASD, and it has 2 components. Parents use a smartphone-based application to complete a developmental questionnaire and to record and upload 10-minute videos of their child. NODA scenarios include the child playing alone, the child playing with others, a family mealtime, and a behavior of parental concern.

The web-based portal allows the clinician to analyze submitted videos for features of ASD. The doctor then completes a DSM-5 ASD checklist and considers the diagnosis. The results of several studies have indicated that the BSI system is easy to use and renders diagnoses comparable with those produced by a traditional ADOS evaluation.12,13 For example, a 2017 study compared the diagnostic accuracy of the NODA system with in-person evaluations by experienced ASD diagnosticians for 40 families who were seeking an ASD evaluation and 11 families with normally developing children. The diagnostic clinicians were blinded as to which group they were evaluating and used ADOS, ADI-R, or other diagnostic tools to render a determination of ASD status for each child. Sensitivity between NODA and the in-person exam was 85%, and the specificity was 94%.14

The developer is currently in the process of integrating AI into NODA, using vision algorithms to tag video frames containing suspect diagnostic features. This AI is expected to be incorporated into the portal over the next 2 years and will speed the diagnostic process.

As with many DTx, the ease of use and availability of this system can shorten wait times to diagnosis considerably. Best of all, the evaluation can be done remotely, even in a pandemic, and the specialist can render a diagnosis expediting appropriate interventions for the child with ASD.

A Promising ASD Diagnostic

Another possibility is using a multimodular AI-based system to diagnose ASD.15,16 The prescription-only system developed by Cognoa, called Canvas Dx, was granted breakthrough device status via a preliminary submission to the US Food and Drug Administration (FDA), and it recently received FDA authorization.17

Cognoa AI software was trained on datasets that were compiled from ADOS and ADI-R score sheets of children aged between 18 and 84 months, supplied by numerous ASD evaluation and treatment centers. Cognoa’s system includes: 1) a parental questionnaire; 2) a questionnaire filled out by the patient’s pediatrician; and 3) an analysis of 2 to 3 uploaded short videos, each 1 to 2 minutes long, of the child during mealtime or play at home. The videos are analyzed and scored by trained analysts for features of ASD who respond to a behavior questionnaire. AI algorithms analyze the questionnaires and video report and render a decision regarding whether the child has or does not have ASD, or if the valuation is inconclusive.


...the system could identify children with ASD with sensitivity and specificity as high as 90% and 83%, respectively.


Study results validated Cognoa’s system; 375 sample patients evaluated over 2 years, indicating that the system could identify children with ASD with sensitivity and specificity as high as 90% and 83%, respectively.18

More recently, the company completed a double-blind clinical trial at 14 sites around the United States, utilizing an improved algorithm to gather data. The trial involved 425 participants aged between 18 and 72 months whose parents or doctors expressed concern about their development but had not previously been evaluated for ASD. Each child was assessed twice: once using Cognoa’s diagnostic, and once by a specialist clinician based on DSM-5 criteria. The study ran from July 2019 through May 2020, so some of the children were evaluated remotely during the pandemic via telemedicine. According to the company, the tool performed equally well when administered remotely and showed that Cognoa’s diagnostic is highly accurate across males and females as well as ethnic and racial backgrounds.

Concluding Thoughts

The diagnosis of ASD remains a challenge, and recent advances in technology may expedite the identification of affected children so the child can receive early interventions. Looking ahead, various smart device-based applications are either in development or will soon be available to assist parents in socializing ASD children. Progress is being made, and soon parents and providers will leverage technologic advances to better identify ASD.

Dr Schuman is clinical assistant professor of pediatrics at Geisel School of Medicine at Dartmouth University and an editorial advisory board member of Contemporary Pediatrics, a sister publication of Psychiatric TimesTM.

References

1. Baio J, Wiggins L, Christensen DL, et al. Prevalence of autism spectrum disorder among children aged 8 years – Autism and Developmental Disabilities Monitoring Network, 11 sites, United States, 2014. MMWR Surveill Summ. 2018;67(6):1-23. Published corrections appear in MMWR Morb Mortal Wkly Rep. 2018;67(19):564 and 2018;67(45):1280.

2. Øien RA, Schjølberg S, Volkmar FR, et al. Clinical features of children with autism who passed 18-month screening. Pediatrics. 2018;141(6):e20173596.

3. Hyman SL, Levy SE, Myers SM; Council on Children With Disabilities, Section on Developmental and Behavioral Pediatrics. Executive summary: identification, evaluation, and management of children with autism spectrum disorder. Pediatrics. 2020;145(1):e20193448.

4. Siu AL, Bibbins-Domingo K, Grossman DC, et al; US Preventive Services Task Force (USPSTF). Screening for autism spectrum disorder in young children: US Preventive Services Task Force Recommendation Statement. JAMA. 2016;315(7):691-696.

5. Monteiro SA, Dempsey J, Berry LN, et al. Screening and referral practices for autism spectrum disorder in primary pediatric care. Pediatrics. 2019;144(4):e20183326.

6. Chlebowski C, Robins DL, Barton ML, Fein D. Large-scale use of the modified checklist for autism in low-risk toddlers. Pediatrics. 2013;131(4);e1121-e1127.

7. Guthrie W, Wallis K, Bennett A, et al. Accuracy of autism screening in a large pediatric network. Pediatrics. 2019;144(4):e20183963.

8. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition: DSM-5. American Psychiatric Association; 2013.

9. Le Couteur A, Haden G, Hammal D, McConachie H. Diagnosing autism spectrum disorders in pre-school children using two standardized assessment instruments: the ADI-R and the ADOS. J Autism Dev Disord. 2008;38(2):362-372.

10. Schuman AJ. AI in pediatrics: past, present and future. Contemporary Pediatrics. 2019;36(5).

11. Cabibihan J-J, Javed H, Aldosari M, et al. Sensing technologies for autism spectrum disorder screening and intervention. Sensors. 2017;17(1):46.

12. Nazneen N, Rozga A, Smith CJ, et al. A novel system for supporting autism diagnosis using home videos: iterative development and evaluation of system design. JMIR Mhealth Uhealth. 2015;3(2):e68.

13. Oberleitner R, Laxminarayan S, Suri J, et al. The potential of a store and forward tele-behavioral platform for effective treatment and research of autism. Conf Proc IEEE Eng Med Biol Soc. 2004;2004:3294-3296.

14. Smith CJ, Rozga A, Matthews N, et al. Investigating the accuracy of a novel telehealth diagnostic approach for autism spectrum disorder. Psychol Assess. 2017;29(3):245-252.

15. Wall DP, Dally RL, Luyster R, et al. Use of artificial intelligence to shorten the behavioral diagnosis of autism. PLoS One. 2012;7(8):e43855.

16. Abbas H, Garberson F, Glover E, Wall DP. Machine learning approach for early detection of autism by combining questionnaire and home video screening. J Am Med Informatics Assoc. 2018;25(8):1000-1007.

17. AI-driven autism diagnosis aid receives FDA authorization. News release. Cognoa; August 18, 2021. Accessed September 2, 2021. https://cognoa.com/ai-driven-autism-diagnosis-aid-receives-fda-authorization/

18. Abbas H, Garberson F, Liu-Mayo S, et al. Multi-modular AI approach to streamline autism diagnosis in young children. Sci Rep. 2020;10(1):5014.

19. Centers for Disease Control and Prevention. Screening and diagnosis of autism spectrum disorder for healthcare providers. Accessed September 2, 2021. https://www.cdc.gov/ncbddd/autism/hcp-screening.html