Infants who demonstrate delays in multiple domains between 9 and 12 months are significantly more likely to still show persistent challenges at follow-up, according to findings from a large population-based study in Israel.

Researchers compared real world outcomes data to predictions made by an artificial intelligence (AI) model.

Clinicians have long believed children with delays across multiple developmental domains are more likely to experience delays that last past age 1 and require intervention than those who miss milestones in just one area, said Yair Sadaka, MD, head of the Child Development and Rehabilitation Department at Israel’s Ministry of Health, who led the study.

“This was something clinicians had known in the past, it was common sense, but it was not validated yet with scientific models. Now it has been,” Sadaka said.

The research, published in JAMA Network Open, analyzed data from 37,760 infants (55% boys) who underwent routine developmental screening between 2014 and 2022 as part of Israel’s well-baby program. The infants had at least one developmental delay detected when they were between 9 and 12 months of age, documented in the national health record.

Clinicians assessed infants using the Tipat Halav Israeli Surveillance (THIS) developmental scale, which was created using data from more than 4 million developmental assessments conducted during well-child visits at clinics run by the government. An estimated 97% of children in the country receive care through this system, and all medical records are centralized.

The THIS scale includes 59 milestones in four domains: gross motor, fine motor, language, and personal-social. Like the milestone checklist from the US CDC, the THIS scale includes milestones children typically reach from infancy to age 5. THIS includes achievement rates — the percentage of kids nationally that reach a certain milestone by a certain age — ranging from 75% to 95%.

At the time of diagnosis, each delay was assigned a score between zero and four based on the milestone achievement rates. The researchers created machine learning models using this data, along with demographic and birth information.

Another analysis was run using simpler, rule-based methods to flag infants who were at risk for persistent delays, defined as those who did not meet milestones in two or three domains.

Researchers followed the children’s developmental assessments through 24 months to determine which early delays persisted.

For about 8800 children — 25% — delays persisted at follow-up. Of these, 40% failed to attain a language-social milestone, such as understanding simple instructions. About 18% lagged in a gross motor skill, such as climbing stairs without assistance. More than 56% had missed a milestone related to fine motor skills, such as feeding themselves.

If an infant had not reached a milestone by the time 95% of their peers had, they were more likely to show developmental delays at follow-up than infants who had failed to reach a milestone in the 75th percentile.

Babies who missed milestones in more than one area were also more likely to have ongoing developmental issues.

The prediction models achieved good accuracy (area under the receiver operating characteristic curve, approximately 0.75), while simpler decision rules provided a feasible and more interpretable alternative in contexts where advanced modeling is impractical, the study found.

Understanding which infant needs intervention immediately, and who may just require a “wait and see” approach — a follow-up visit and education for parents — can reduce strain on the healthcare system, Sadaka said. However, while the study established that two or more delays in milestones and persistent delays are correlated, clinicians should still go through a formal diagnosis process, he and his colleagues wrote.

Ann Reynolds, MD, a professor of pediatrics and section head of Developmental Pediatrics at the University of Colorado Medical School in Aurora, Colorado, said using a predictive AI model would be helpful in her practice.

Getting to a specialist for further care “is hard. I love the idea of telling patients, ‘Here are some things you could do, and then I’ll see you back in 2 months,” said Reynolds, who was not involved with the study.

Building a predictive model in the US, however, could be challenging. Although the American Academy of Pediatrics recommends milestone screening of infants at regular intervals, no national system exists to track this data, Reynolds said. Clinicians also display wide variation in screening. Research has also shown many physicians do not follow the group’s recommendations or use its developmental tools.

Some caregivers also do not bring their children in for well-child visits, said Toni Whitaker, MD, division chief of Developmental Pediatrics at the University of Tennessee Health Science Center in Memphis, Tennessee.

“Many children aren’t seen for well-care visits or by pediatricians,” Whitaker, who was not involved in the study, said.

Compared with infants and toddlers with private insurance, those who have public health insurance have a 20% higher risk for missing well-child visits, research has shown. Other studies have found 30-50% of children overall miss well visits.

Reynolds said she would like to see analysis of the THIS data of delays that occur after age 2. Delays after this age are common, particularly among children with autism , she said.

“Development gets more complex as kids age,” she said. “With autism, there can be a regression or plateau in the second year of life.” 

This study received funding from Horizon Europe and UK Research and Innovation. Sadaka reported being a shareholder in LinkCaring LTD. No other disclosures were reported.

Kaitlin Sullivan is a freelance reporter.