Evaluation of a records-review surveillance system used to determine the prevalence of autism spectrum disorders.

Avchen et al. (2011) checked how well a records-review system spots autism. They read school and clinic files instead of testing kids in person.

The goal was to see if paper records give a true count of ASD in the area.

The system was good at ruling kids in; almost every record labeled ASD truly was ASD. But it missed about four in ten children who really had autism.

This means the method gives a low, cautious number, not the full picture.

Prigge et al. (2013) extend the same idea. They added early-intervention and university charts and found every toddler the 2011 method had missed. The two papers agree: more record sources catch more cases.

Towle et al. (2009) did an earlier, smaller test using only early-intervention charts. Their count matched published rates, hinting that single-source reviews already run low—exactly what Nonkin later proved.

Lotfizadeh et al. (2020) look similar on paper but show a negative twist. Billing-code algorithms also miss lots of kids and wrongly label some non-ASD children. Together the studies warn: any paper-only hunt will under-count unless you add extra data sources.

If you help with district or state prevalence reports, treat records-review numbers as a floor, not a ceiling. Add early-intervention files and university eval logs before you lock the final count. And always tell stakeholders the real total is likely higher.