A review of the DSM-III-R criteria for autistic disorder.

Szatmari (1992) read every paper on the brand-new DSM-III-R autism rules. The goal was to see how well the criteria caught real cases and how often they flagged the wrong kids.

The review looked at sensitivity (does it find kids who truly have autism?) and specificity (does it skip kids who do not?).

The DSM-III-R rules were great at catching autism but poor at ruling out other problems. In plain words, the criteria acted more like a wide-net screener than a final test.

High sensitivity, low specificity means you get few false negatives but many false positives.

Later DSM editions swung the pendulum the other way. Whitehouse et al. (2014) and Heald et al. (2020) showed DSM-5 cuts autism diagnoses by roughly one-third, mostly by dropping PDD-NOS cases. This looks like a contradiction—first too loose, now too tight—but the difference is method: DSM-5 added stricter social-communication requirements and collapsed subtypes.

Maddox et al. (2015) found DSM-5 criteria fit the data better even while shrinking the diagnosed group, proving the trade-off can be worth it.

Peters et al. (2020) added that toddlers with milder traits are the ones most likely to lose the label under DSM-5, echoing P’s warning that criteria shape who gets help.

If you still reference DSM-III-R for historical data, remember it over-counts. Use it only as a screener, then back up with current tools. When you switch to DSM-5, expect some kids to exit the spectrum; plan reassessments and document need for services either way.