Predicting low cognitive ability at age 5 years using perinatal data and machine learning.

Area of Science:

• Developmental Pediatrics
• Machine Learning in Healthcare
• Population Health

Background:

• Lack of early, accurate, and scalable methods to identify infants at high risk for poor cognitive outcomes.
• Need for predictive tools to enable timely interventions and support for at-risk children.

Purpose of the Study:

• To develop an explainable predictive model using machine learning and population-based cohort data.
• To identify infants at high risk of poor cognitive outcomes in childhood for early intervention.

Main Methods:

• Utilized data from 8858 participants in the nationally representative Growing Up in Ireland cohort.
• Collected maternal, infant, and socioeconomic characteristics at 9 months; cognitive ability measured at age 5 years.
• Employed data preprocessing, synthetic minority oversampling, feature selection, and a random forest model with hyperparameter tuning via ten-fold cross-validated grid search.

Main Results:

• A random forest model with 15 easily collected features achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.77 for predicting low cognitive ability at age 5.
• The model identified 72% of infants with low cognitive ability, with a specificity of 66%.

Conclusions:

• The developed model represents a first step towards early, individual risk stratification for cognitive development.
• Further performance improvements are necessary for implementation as a population-level screening tool.
• Highlights the potential of machine learning for identifying at-risk infants in the perinatal period.