Beyond pass-fail: Examining the potential utility of two thresholds in the autism screening process.

The prediction model was built and tested in the same 171-child sample using only one random split, so external validation is limited. The sample included only English-speaking children in Illinois who were already enrolled in early intervention, so results may not apply to other languages, cultures, or children not referred through early intervention. The specific reason for referral to the diagnostic evaluation was not recorded, so the sample may not represent all referred children. The same developmental therapist administered both the STAT and the ADOS-2, which may have influenced ADOS-2 scoring. The STAT is a Level 2 screener, not a diagnostic measure, and it does not include restricted and repetitive behaviors, so a high probability score is not a diagnosis. The two-threshold logistic result applied to 54% of the validation sample, not all 171 children, and the original and 7-item two-threshold results applied to smaller parts of the entire sample. One STAT imitation item behaved unexpectedly in the logistic model, and the authors said this may reflect instability because STAT items are highly correlated.

Do not treat a high STAT probability as an autism diagnosis or as a triage rule for most referred toddlers. The logistic two-threshold result covered 54% of the validation sample, not all 171 children, and the STAT is a screening tool that does not include restricted and repetitive behaviors.