Cluster Analysis of Clinical Features of Children Suspected to Have Neurodevelopmental Disorders.

The team looked at 194 kids who were being checked for neurodevelopmental disorders. They fed 25 test scores into a computer. The scores came from IQ tests, language tests, and behavior checklists. The computer grouped the kids into clusters based on how similar their scores were.

No labels like "autism" or "ADHD" were used. The math let the patterns speak for themselves.

Three clear profiles popped out. One group had strong language but weak social skills. Another showed low IQ across the board. The third mixed high non-verbal IQ with big attention problems.

Each cluster looked different on paper and in real life. The method worked even though every child started with the same vague referral question.

Némorin et al. (2025) did almost the same thing with the kids who already had an autism diagnosis. They also found clusters, but got four instead of three. The extra group in Harmony’s study was made of kids with mild symptoms and strong daily living skills. The difference makes sense: Harmony only studied children with ASD, while Mélina cast a wider net.

Laugeson et al. (2014) showed that IQ alone drives a lot of adaptive behavior scores. Mélina’s clusters line up with that finding. The low-IQ cluster scored worst on daily living skills, just as A et al. predicted.

Bachrach et al. (2026) used cognitive and irritability scores to guess which preschool class a child would enter. Mélina’s profiles could give placement teams even richer data, since they fold in language and attention too.

Instead of waiting for a firm diagnosis, you can run the same short battery of tests and see which profile fits. Then pick targets that match the child’s real pattern. A kid in the low-language cluster might start with social stories and PECS. A kid in the high-IQ, poor-attention cluster might begin with self-monitoring and token boards.