A Five-Item Screener Caught 91% of Autistic Toddlers in 10 Minutes.

Screening instruments get abandoned for a reason that has nothing to do with accuracy: nobody has forty minutes. A team at Kaohsiung Medical University tried the obvious response — cut the instrument down and see what survives.

What they did

The Screening Tool for Autism in Two-Year-Olds (STAT) is a play-based Level 2 screener for toddlers under 36 months. The researchers built a brief form in two studies:

• Study 1 used two independent samples of 40 toddlers each — 20 autistic, 20 with developmental delays — and chi-square tests to find the discriminating items. Five survived: Turn-taking, Bubbles, Snack, Balloon, and Bag of Toys. A cutoff score of 3 gave the best sensitivity-specificity trade-off.
• Study 2 validated the five-item form on 251 toddlers — 81 autism, 55 mild-autism, 115 developmental delays.

The numbers

• Sensitivity for autism: 0.91 — 74 of 81.
• Sensitivity for mild autism: 0.91 — 50 of 55. That the brief form held its sensitivity in the milder group is the more surprising result; short screeners usually lose the subtler presentations first.
• Specificity: 0.81 — 93 of 115 toddlers with developmental delays correctly identified as not autistic.
• Administration time: 5 to 10 minutes, down from an instrument the authors describe as “often limited by administration time in clinical settings.”

Read the specificity honestly. At 0.81, roughly one in five toddlers with developmental delay and no autism screens positive. In a Level 2 setting that is a referral for fuller assessment rather than a diagnosis, which is what the tool is for — but it is not a number to describe to a parent as a result.

The limits, in the authors' own terms

They are explicit about the scope. This is a Level 2 tool: for toddlers “referred for developmental concerns,” not a universal screen of an unselected population. Both samples were Taiwanese, and the authors write that “further validation in diverse populations and longitudinal follow-up is warranted to establish generalizability.”

The design carries the usual constraint of item-reduction work: the five items were chosen on one set of samples and validated on another from the same clinical population, so how they perform elsewhere is untested. And the study reports screening accuracy against existing diagnostic classification, not outcomes — nothing here says a child screened this way does better later.

Why it matters here

Eligibility is starting to hang on the date of diagnosis rather than the date of referral. West Virginia's Medicaid ABA chapter, which we cover on September 16, makes the benefit available only to members diagnosed with autism spectrum disorder before their eighth birthday — a child identified at nine is outside it permanently, whatever their presentation.

Where a rule like that is in force, the machinery that produces early diagnoses stops being somebody else's problem. A five-minute instrument at the front of a developmental clinic is upstream of a referral you may receive years later, or never.

What you must know or do

• If you take referrals from developmental clinics or early intervention, ask which Level 2 screener they use and how long it takes. Administration time is the reason screeners go unused, and it is a question you can ask a referral partner without stepping into their clinical decisions.
• Do not carry these numbers into a parent conversation as a result. A positive at cutoff 3 is a reason for fuller assessment. One in five toddlers with developmental delay and no autism will screen positive on it.
• Check the validation population before adopting anything. The STAT-BF has been tested in Taiwan, in a clinic-referred sample. The authors say generalizability is not yet established, and that sentence is theirs, not a hedge added here.