How ADOS-2 Scores Become a Diagnosis

What gets coded, and how the codes are organized

Nothing is scored while you watch. During the session the examiner runs activities and takes notes; codes are assigned afterward, from those notes and often from video, once the whole hour can be weighed together. Each coded behavior receives a number from 0 to 3, where 0 means no evidence of atypicality in that behavior and 3 means it was markedly present.

The examiner is not scoring in front of you. The room was the sampling. The numbers were built later.

Two mechanics explain why a parent's own arithmetic never matches the report. Only a portion of what gets coded enters the algorithm at all — much is recorded because it shapes the written description, not because it becomes part of a total. And within the algorithm a code of 3 counts as 2, so the highest available code and the heaviest scoring weight are not the same thing.

The algorithm items then sort into two domains. Social Affect covers communication and reciprocal social interaction: gesture, eye contact, shared enjoyment, how a conversation is carried, how much insight someone shows into relationships. Restricted and Repetitive Behavior covers stereotyped movement, repetitive use of objects, unusual sensory interest, and highly circumscribed interests. Added together, they give the algorithm total.

• The total leans social. Social Affect contributes many more items, so the sum is weighted toward the social side almost by construction.
• Totals do not travel between modules. Each module has its own item set and algorithm, which is the entire reason the comparison score exists.
• The two domains echo the diagnostic criteria, so an evaluator moves between score sheet and diagnostic language without translating.

How the total becomes a classification

The algorithm total is set against cutoffs printed in the protocol for the module used. Most modules carry two thresholds and yield one of three outcomes: autism, autism spectrum, or non-spectrum. Those cutoff values differ by module and by algorithm version and are not reproduced here — what a family needs from the report is which of the three it states, and which module produced it.

The word autism appearing as an instrument classification is the most misread line in these documents. It is a statement about where one observed hour fell relative to a threshold. It is not a DSM diagnosis, and the gap between "autism" and "autism spectrum" is not a severity distinction in any clinical sense. The clinician writing the report makes the diagnosis, and may depart from the instrument in either direction and say why.

None of it stands alone. Identification is a two-step sequence, a screen and then a comprehensive diagnostic evaluation, and an observational classification belongs to the second step rather than replacing it 1Ref 1Centers for Disease Control and Prevention (2024).Screening and Diagnosis of Autism Spectrum Disorder.That identification proceeds in two steps — developmental screening followed by a comprehensive diagnostic evaluation — so an instrument classification is a component of the second step rather than a substitute for it.. Outputs also come in several shapes: one large community surveillance program classifies a child as high likelihood when at least three key behaviors are atypical, with no number attached at all. That program reported a positive predictive value of 83% and a negative predictive value of 99% across roughly 13,500 children 2Ref 2Barbaro J, Sadka N, Gilbert M, et al. (2022).Diagnostic Accuracy of the Social Attention and Communication Surveillance–Revised With Preschool Tool for Early Autism Detection in Very Young Children.That a community surveillance program using age-specific checklists of early social-communication behaviors classifies a child as high likelihood when at least three key items are atypical, producing a category rather than a score, and that this approach reported PPV 83% and NPV 99% in a prospective study of roughly 13,500 children..

The comparison score, and what it does not measure

Alongside the raw total, most modules yield a comparison score on a scale of one to ten. It is derived from the algorithm total but adjusted for age and language level, which a raw total cannot do for itself. Its purpose is comparability — it lets a result from one module be read beside a result from another, and lets a person be compared with themselves across years.

The comparison score is not an IQ, not a percentile against the general population, and not a DSM-5 support level.

That last point causes real confusion, because research papers call the same figure a calibrated severity score. The name is unfortunate. Autism support levels — the DSM-5 autism severity levels written as level 1, 2, or 3 — are assigned by a clinician weighing how much support a person actually needs across daily life. That support needs classification is a judgment about a life, not the output of a test.

Reports often attach a descriptive word to the number; that wording, and what the clinician meant by it, is the part worth reading closely. The reason a number is used at all is that autistic traits do not distribute as a clean yes or no. When a 25-item caregiver questionnaire scoring each item from 0 to 4 was given to an unselected sample of 754 toddlers, the totals came out close to normally distributed, averaging 26.7 on a 0-to-100 range, while a group of autistic children averaged 51.8 — overlapping distributions rather than two separate populations 3Ref 3Allison C, Baron-Cohen S, Wheelwright S, et al. (2008).The Q-CHAT (Quantitative CHecklist for Autism in Toddlers): A Normally Distributed Quantitative Measure of Autistic Traits at 18–24 Months of Age: Preliminary Report.That a 25-item caregiver questionnaire scored 0 to 4 per item over a 0-to-100 total, with higher scores indicating more autistic traits, was near-normally distributed in an unselected sample of 754 toddlers (mean 26.7) while an autistic group averaged 51.8 — and that its authors explicitly published no cutoff, deferring threshold testing to later work.. Its authors published no cutoff at all, saying threshold testing lay beyond that report's scope 3Ref 3Allison C, Baron-Cohen S, Wheelwright S, et al. (2008).The Q-CHAT (Quantitative CHecklist for Autism in Toddlers): A Normally Distributed Quantitative Measure of Autistic Traits at 18–24 Months of Age: Preliminary Report.That a 25-item caregiver questionnaire scored 0 to 4 per item over a 0-to-100 total, with higher scores indicating more autistic traits, was near-normally distributed in an unselected sample of 754 toddlers (mean 26.7) while an autistic group averaged 51.8 — and that its authors explicitly published no cutoff, deferring threshold testing to later work.. A number on a continuum tells you where an observation fell, not which side of a line a person lives on.

Why the number is not the verdict, in either direction

Two failure modes matter, and a report worth trusting names both. A score below the cutoff does not rule autism out, and a score above it does not establish autism. An observational instrument samples roughly an hour, one-to-one, in a quiet room, with a skilled stranger, no competing demands, and none of the unstructured social pressure of a playground or a staff meeting. For many autistic people that is the most favorable social environment they meet all year.

The low-score direction has well-known causes. Camouflaging and rehearsed social behavior — described most often in autistic girls and women and in fluent adults — can carry someone through a structured hour. A child can have a genuinely good day, and someone who has had therapy has practiced the very behaviors being sampled.

A score that misses the cutoff while an experienced clinician still has concerns is common and well recognized, not a contradiction that has to be resolved by discarding one of the two.

This is why the caregiver interview carries independent weight rather than serving as background. The ADI-R is a semistructured, investigator-based diagnostic interview conducted with a caregiver, tied to established diagnostic criteria and usable from a mental age of roughly eighteen months into adulthood 4Ref 4Lord C, Rutter M, Le Couteur A. (1994).Autism Diagnostic Interview-Revised: a revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders.That the ADI-R is a semistructured, investigator-based diagnostic interview administered to caregivers, explicitly linked to established diagnostic criteria and applicable from a mental age of roughly 18 months into adulthood — the developmental-history component that sits alongside direct observation.. It reaches backwards into developmental history, which no hour of live observation can see. Federal clinical guidance says the same thing from the other end: no laboratory test, developmental history set alongside observed behavior, and often several disciplines contributing 5Ref 5Centers for Disease Control and Prevention (2024).Clinical Testing and Diagnosis for Autism Spectrum Disorder.That there is no laboratory test for autism, that the conclusion rests on developmental history together with observed behavior, and that a comprehensive evaluation may involve developmental pediatricians, child psychologists or psychiatrists, or neurologists..

Running the other way, a total above a cutoff is not self-interpreting either. Language disorder, intellectual disability, ADHD, anxiety, selective mutism, hearing loss, and the effects of trauma can all raise the same items, because the items describe behavior rather than its cause. Sorting that out is the clinician's work.

The score does not decide what services a child gets

This surprises families more than anything else on this page. Early-intervention services for children under three and school-based services from three onward operate under their own eligibility rules, and a family can enter both without waiting for a completed medical diagnosis 6Ref 6Centers for Disease Control and Prevention (2024).Accessing Services for Autism Spectrum Disorder.That families can access early intervention under Part C for children under three and school services under Part B from age three without waiting for a completed medical diagnosis.. A pending evaluation, a long waitlist, or a report that landed ambiguously does not close those doors.

The two systems use different documents and different tests of eligibility, which is why the distinction between ifsp vs iep is worth learning before you need it. An IDEA Part C IFSP is written for a child under three around family-centred goals and is often delivered at home; the school plan that follows works from educational need and is delivered in a classroom. The handover happens around the third birthday, and families who understand the sequence in advance are the ones who do not lose months at exactly that transition.

A school evaluation is also not a medical evaluation. School eligibility answers whether a child needs specialized instruction; it does not produce a diagnosis and generally cannot stand in for one with an insurer. Families commonly run both tracks at once.

Reading the report without letting the number flatten the child

The scores usually sit in a table near the end of the document, sometimes tucked into an appendix, and they occupy perhaps a tenth of a page. The narrative is the report. What the clinician observed, under what conditions, what they made of it, and what they recommend is the part that schools, therapists, and every later clinician will actually act on — and it is the part that deserves the slow read.

Questions that make a feedback session useful rather than a recitation of numbers:

• What did you see that mattered most, and when in the session?
• What would you have needed to see to land somewhere different?

Making sense of the evaluation report is a task in its own right, and a good clinic expects these questions. If a report arrived with no feedback appointment, asking for one is ordinary — and understanding the report before it circulates to a school district is worth the wait. If a second evaluation is under discussion, autism evaluation cost differs substantially by setting.

One last thing, said plainly. Nobody who loves a child reads a number about them neutrally. The comparison score is a research-derived index built so that clinicians could compare observations that were not otherwise comparable. It was never designed to describe a person, and it does not.