Children don’t fit in boxes. Their data shouldn’t, either

From speech to social and fine motor skills, children's developmental milestones unfurl uniquely.

"Every child is different," said Julia Anglen Bauer, a UIC professor of epidemiology and biostatistics. "As a mother of a 3- and 5-year-old, I see how unique children's developmental paths can be. As a researcher, it's important to understand variance among children so we can provide the best possible support."

Bauer and her team used artificial intelligence to organize developmental data from preschoolers in the New Hampshire Birth Cohort Study, an extensive, long-running study of families in rural communities. Their findings, published in the journal BMJ Public Health, show that even in a healthy population, children can be grouped into distinct developmental profiles spanning cognition, motor skills, social functioning and behavior.

Most existing research on child development focuses on atypical development (e.g., children born preterm or exposed to illicit drugs during pregnancy), leaving more typical variations understudied. Especially overlooked are the roughly 12 million children living in rural U.S. communities, Bauer said, despite this population's reduced access to health care and fewer economic opportunities compared with city-dwelling children.

Bauer's team analyzed more than 18,000 data points from 235 children, collected when the children were 3 and 5 years old. These included standardized cognitive and motor tests, parent-reported behavior and social responsiveness and measures of parent-child attachment. Then they used artificial intelligence to cluster the data into easy-to-visualize profiles.

Results revealed six distinct behavioral profiles: highest overall scores, worst overall scores, greatest behavioral/social improvement, slight improvement, average scores and highest adaptability. Several correlations emerged. For example, children with the lowest overall scores had lower maternal attachment and more relational frustration. Children with the best overall scores and highest adaptability, on the other hand, were likelier to be girls with mothers who had high IQs.

"Our study shows you don't have to put children into a diagnostic box to learn something important," Bauer said. "By describing subtle gradients in healthy kids, we can identify early risks and resilience factors that shape school readiness and lifelong health."

Bauer envisions broad applications for this method, from childhood health to chronic disease research.

"Artificial intelligence offers a new approach for understanding population health," she said. "Instead of sorting people into rigid categories, we can explore the full spectrum of variation and see how factors like environment, sex and family dynamics come together."

Bauer said collaboration between scientists and clinicians is crucial to turning data analysis into action. She hopes to apply AI to other complex data, like diseases, health and environmental exposures among different populations - her research specialty. Future studies will also include more diverse populations.