Given the complex nature of autism spectrum disorder (ASD), its diagnosis can be challenging. Unlike conditions where biomarkers such as those found with a blood test can provide a definitive diagnosis, diagnosing ASD requires clinical observation and interpretation of a child’s behavior and developmental history. Screening and diagnostic tools can be utilized to aid the clinician’s assessment, but no tool should be used as a “stand-alone” diagnostic. The clinician must assess each child’s presentation and make a diagnosis based on whether the Diagnostic and Statistical Manual of Mental Disorders – Fifth Edition (DSM-5) criteria for ASD are met (“Screening and Diagnosis of Autism Spectrum Disorder” | CDC; “Diagnostic Criteria | Autism Spectrum Disorder (ASD)” | CDC).
Children with ASD can be identified as toddlers and early intervention can and does influence outcomes. In addition to screening when concerns arise, the American Academy of Pediatrics (AAP) “recommends screening all children for symptoms of ASD” through a combination of standardized autism-specific screening tests at 18 and 24 months of age and general developmental screening at 9, 18, and 30 months of age visits (Hyman, et al 2020).
ASD Screening Tools
A screening tool is used in the general population in individuals without symptoms to identify those at risk. The goal of a screening tool is to enable earlier identification of concern and does not provide a diagnosis (Bovbjerg 2020; “Screening and Diagnosis of Autism Spectrum Disorder” | CDC). For ASD, examples of commonly used screening tools available include the Modified Checklist for Autism in Toddlers (M-CHAT-R/F), Screening Tool for Autism in Toddlers and Young Children (STAT), and Rapid Interactive Screening Test for Autism in Toddlers (RITA-T) (“Screening and Diagnosis of Autism Spectrum Disorder” | CDC; “RITA-T Research” | Boston Children’s Hospital), with M-CHAT-R/F being the most studied and widely used in toddlers in the primary care setting (Hyman, et al 2020).
ASD Diagnostic Tools
A diagnostic tool is used in individuals with symptoms, concerns, or those for whom screening has identified an area of concern to assess the probability or likelihood that an individual may have a particular condition. Examples of ASD diagnostic tools include Autism Diagnostic Observation Schedule-Second Edition (ADOS-2), Autism Diagnostic Interview-Revised (ADI-R), Childhood Autism Rating Scale-Second Edition (CARS-2) (“Screening and Diagnosis of Autism Spectrum Disorder | CDC” n.d.; “(CARSTM2) Childhood Autism Rating ScaleTM, Second Edition” n.d.) and Canvas Dx. These tools have been clinically validated in varying individual populations and most are routinely used by specialist clinicians (Randall et al. 2018; Megerian et al. 2022). To date, Canvas Dx, is the only AI based diagnostic device that has received FDA marketing authorization for ASD (FDA. 2021).
A tool’s performance can be evaluated based on the following characteristics (Bovbjerg 2020):
- Sensitivity – The ability of the tool to correctly identify individuals who truly have ASD
- Specificity – The ability of the tool to correctly identify individuals who truly do not have ASD
- Positive predictive value (PPV) – The likelihood that an individual who receives an ASD positive outcome truly has ASD
- Negative predictive value (NPV) – The likelihood that an individual who receives an ASD negative outcome truly does not have ASD
Sensitivity and specificity are “fixed test characteristics” because they do not change, regardless of disease prevalence – how often the disease may occur in a certain population of individuals. The PPV and NPV do change when disease prevalence changes (Bovbjerg 2020). There can be substantial variation in the sensitivity and specificity of ASD tests, likely due to methodological differences and variations in the clinical characteristics of populations recruited (Randall et al. 2018). Therefore, when assessing the performance of a test in the real world, PPV and NPV should also be considered.
Artificial intelligence (AI) and machine learning (ML) technologies are increasingly being used in healthcare. ML is a form of AI where systems have the ability “to learn” using large amounts of data to improve accuracy. Thus, large amounts of data combined with computational power can help improve clinical efficacy by helping improve the accuracy and efficiency of diagnosis and treatment (Ahuja 2019). Considering the diverse and complex symptom presentation in ASD, AI-based tools designed to aid in diagnosis are an active area of research and growth (Shannon, et al. 2022). ML-based diagnostic devices such as software as medical devices (SaMDs) have the advantage that they have the potential to improve performance with additional data or real-world use (FDA 2019).
When making an ASD diagnosis, best practices require integration of caregiver reports, behavioral observations, standardized assessments of cognitive, language and functional adaptation, and clinical assessment based on the DSM-5 criteria (Randall et al. 2018). Regardless of the tool selected, no single ASD tool should be used as the basis for diagnosis, and it should be used by clinicians to support rather than replace clinical judgment (“Screening and Diagnosis of Autism Spectrum Disorder” | CDC; Elder et al. 2017).
Dr Jennifer Shannon is a board-certified child, adolescent, and adult psychiatrist passionate about healthcare innovation, particularly as it relates to pediatric mental health. She practices in Bellevue, WA and is also Executive Medical Director at Cognoa. For more information, please contact jennifer.shannon@cognoa.com.
References
Ahuja, Abhimanyu S. 2019. “The Impact of Artificial Intelligence in Medicine on the Future Role of the Physician.” Edited by Matteo Lambrughi. PeerJ 7 (October): e7702. https://doi.org/10.7717/peerj.7702.
Bovbjerg, Marit L. 2020. “Screening and Diagnostic Testing,” October. https://open.oregonstate.education/epidemiology/chapter/screening-and-diagnostic-testing/.
“(CARSTM2) Childhood Autism Rating ScaleTM, Second Edition.” n.d. Accessed February 27, 2023. https://www.wpspublish.com/cars-2-childhood-autism-rating-scale-second-edition.html.
“Diagnostic Criteria | Autism Spectrum Disorder (ASD) | NCBDDD | CDC.” n.d. Accessed February 9, 2023. https://www.cdc.gov/ncbddd/autism/hcp-dsm.html.
Elder, Jennifer Harrison, Consuelo Maun Kreider, Susan N Brasher, and Margaret Ansell. 2017. “Clinical Impact of Early Diagnosis of Autism on the Prognosis and Parent–Child Relationships.” Psychology Research and Behavior Management 10 (August): 283–92. https://doi.org/10.2147/PRBM.S117499.
FDA. 2019. “Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) – Discussion Paper and Request for Feedback.” https://www.regulations.gov/document/FDA-2019-N-1185-0001.
FDA. 2021. FDA Authorizes Marketing of Diagnostic Aid for Autism Spectrum Disorder. June 2, 2021. https://www.fda.gov/news-events/press-announcements/fda-authorizes-marketing-diagnostic-aid-autism-spectrum-disorder.
Hyman, Susan L., Susan E. Levy, and Scott M. Myers. 2020. “Identification, Evaluation, and Management of Children with Autism Spectrum Disorder.” Pediatrics 145 (1).
Megerian, J.T., S. Dey, M. Raun, D.L. Coury, M. Lerner, C.J. Nicholls, Kristin Sohl, et al. 2022. “Evaluation of an Artificial Intelligence-Based Medical Device for Diagnosis of Autism Spectrum Disorder.” Nature Partner Journal- Digital Medicine. https://doi.org/10.1038/s41746-022-00598-6.
Randall, Melinda, Kristine J. Egberts, Aarti Samtani, Rob JPM Scholten, Lotty Hooft, Nuala Livingstone, Katy Sterling‐Levis, Susan Woolfenden, and Katrina Williams. 2018. “Diagnostic Tests for Autism Spectrum Disorder (ASD) in Preschool Children.” Cochrane Database of Systematic Reviews, no. 7.
“RITA-T Research | Boston Children’s Hospital.” n.d. Accessed February 27, 2023. https://www.childrenshospital.org/research/labs/rita-t-research.
“Screening and Diagnosis of Autism Spectrum Disorder | CDC.” n.d. Accessed February 9, 2023. https://www.cdc.gov/ncbddd/autism/screening.html.
Shannon, J., C. Salomon, T. Chettiath, H. Abbas, and S. Taraman. 2022. “Autism Spectrum Disorder and the Promise of Artificial Intelligence.” J Child Adolesc Behav 10 (428): 2.
Shannon, J, Sharief Taraman, Dennis P. Wall, Stuart Liu-Mayo, and Carmela Salomon. 2022. “Optimizing a de Novo Artificial Intelligence-Based Medical Device under a Predetermined Change Control Plan: Improved Ability to Detect or Rule out ASD in General Pediatric Settings.” Journal of the American Academy of Child & Adolescent Psychiatry 61 (10, Supplement): S242-243.
