Differentiating Transient From Persistent Developmental Delays in a Nationwide Infant Cohort

Early childhood developmental surveillance is critical for identifying children at risk of developmental delays and ensuring timely intervention. Current well-child surveillance policies are not well designed to distinguish transient delays from persistent ones, potentially leading to unnecessary referrals or missed opportunities for early support.

To evaluate whether the Tipat Halav Israeli Surveillance (THIS) developmental scale can be leveraged for distinguishing transient from persistent developmental delays and thus augment the precision of early childhood surveillance policies.

This retrospective cohort study used national data from maternal child health clinics in Israel that perform routine developmental surveillance of children aged 0 to 6 years, serving approximately 65% of Israeli children. Participants were individuals born at 37 weeks' gestation or later in the years 2014 to 2022 and assessed between January 1, 2014, and December 31, 2022, who failed to attain at least 1 developmental milestone between age 9 and 12 months based on their electronic health records. Analysis was done between July 2024 and April 2025.

Age-appropriate milestone attainment results at ages 9 to 12 months, along with demographic and birth-related covariates.

The primary outcome was ongoing milestone attainment failure at age 12 to 24 months. Accuracy of machine learning models trained to predict the outcome was measured by the area under the receiver operating characteristic curve, while for simpler decision rules, sensitivity and specificity were computed.

In a cohort of 529 797 infants born at 37 weeks' gestation or later and assessed at age 9 to 12 months, developmental delays were observed in 37 760 (7.1%), among whom 20 862 (55.2%) were male and median gestational age was 39 weeks (IQR, 38-40 weeks). A total of 35 163 (93.1%) were assessed again in their subsequent year of life, among whom persistent delay was suggested for 8802 (25.0%). Machine learning models distinguished transient from persistent delays, with areas under the receiver operating characteristic curve ranging from 0.71 (95% CI, 0.70-0.72) to 0.77 (95% CI, 0.74-0.79). Simply counting the number of developmental domains in which unmet milestones occurred showed good fidelity as well (eg, sensitivity of 0.48 and specificity of 0.81 to identify persistent gross motor delay after early milestone failure in that domain plus 1 other).

This cohort study of infants showing early signs of developmental delay suggests that transient delays can be effectively distinguished from persistent ones using routine well-child surveillance data and also provides evidence for the validity of counting domains in which milestones are not attained, which is often used in practice. This suggests an accessible approach for developmental surveillance that may improve outcomes for children worldwide.