Childhood obesity prediction from nationwide health records

Rapid rise in childhood obesity prevalence worldwide and its major implications for health warrant the development of better prevention strategies. Here, we analyzed electronic health records of children from Israel’s largest healthcare provider from 2002 to 2018 and developed a model for predicting children’s obesity. Data included demographics, anthropometric measurements, medications, diagnoses, and laboratory tests of children and their families. Obesity was defined as body mass index (BMI) ≥95th percentile for age and sex. To identify the most critical time-window in which the largest annual increases in BMI percentile occurs during early childhood, we first analyzed childhood BMI trajectories among 382,132 adolescents. Next, we devised a prediction model targeted to identify children at high risk of obesity prior to this time-window, predicting obesity at 5-6 years of age based on data from the first 2 years of life of 136,196 children. Retrospective BMI analysis revealed that among obese adolescents, the greatest acceleration in BMI percentiles occured between 2-4 years of age. Our model, validated temporally and geographically, accurately predicted obesity at 5-6 years old (auROC of 0.804). Discrimination results on subpopulations demonstrated its robustness across the pediatric population. The model’s most influential features included anthropometric measurements of the child and family. Other impactful features included ancestry and pregnancy glucose. Antibiotics exposure in utero and during the first 2 years of life had no observed impact on obesity prediction. Our model, targeted to identify children prior to the most critical time-window of BMI acceleration, may allow a more accurate identification and implementation of early prevention strategies for children at high risk of obesity and can be readily embedded into healthcare systems.