A Comparative Analysis of the ADOS-G and ADOS-2 Algorithms: Preliminary Findings.

Fleury et al. (2018) looked at eight past studies that compared the old ADOS-G algorithm with the newer ADOS-2. They asked: does ADOS-2 catch more true cases and send fewer typical kids to mis-diagnosis? The team pulled every paper that gave hard numbers for sensitivity and specificity.

The numbers were all over the map. Some studies said ADOS-2 was better; others said it missed more kids or over-called autism. When the eight studies were lined up, no clear winner appeared. The authors label the evidence 'mixed' and urge caution.

Bennett et al. (2008) saw the same mess earlier. In Hispanic preschoolers the revised algorithm (the heart of ADOS-2) lowered specificity in modules 2 and 3, flagging too many kids with mild social delays. That single-study red flag now shows up in the 2018 big-picture review.

Hong et al. (2021) looks like a contradiction at first glance. Their 2021 replication found excellent accuracy for the Toddler Module cut-offs. The fight disappears when you notice age: toddlers have clearer symptom profiles, while school-age kids with language or cultural differences trip up modules 2 and 3.

de Bildt et al. (2011) and Hus et al. (2014) add another layer. They showed that calibrated severity scores help compare kids across time, but module 2 still needs more work. Fleury et al. (2018) folds that warning into the wider claim: newer algorithms are not automatically better.

Before you write 'meets ADOS-2 criteria' in your report, check which module and which algorithm you used. If the child is Hispanic, bilingual, or has only mild social delays, get extra parent data or run the ADI-R to guard against false positives. Stay skeptical of cut-off labels and document your clinical override.