Brief Report: Machine Learning for Estimating Prognosis of Children with Autism Receiving Early Behavioral Intervention-A Proof of Concept.

The team built five computer models to guess how much 3- to young learners with autism would improve after early ABA.

They fed each model the same intake data: ADOS scores, Vineland scores, and child details.

Then they tested which model made the best forecasts about later adaptive skills and autism symptoms.

Every model beat random guessing.

The best one got closer to the real outcomes than flipping a coin would.

This shows computers can give families a clearer picture of what to expect from therapy.

Kremkow et al. (2022) already showed tablet games can spot autism risk in toddlers. Préfontaine et al. (2024) moves the same idea forward — instead of just flagging who has autism, it predicts how far each child will go.

Bailey et al. (2021) used ML to sharpen QABF results. Préfontaine et al. (2024) uses the same trick, but for long-term prognosis instead of short-term function.

Lanovaz et al. (2020) proved ML can read single-case graphs better than the naked eye. The new study widens the lens from one child’s daily data to many children’s yearly outcomes.

Rivard et al. (2023) found that grouping kids by developmental profiles also predicts adaptive gains. Préfontaine et al. (2024) keeps the same goal but swaps hand-made groups for computer-built models.

You can start giving parents data-driven hope. Ask your intake team to save clean ADOS and Vineland scores. When easy ML tools hit the clinic, you will be ready to plug them in and set realistic goals from day one.