Prediction Model for Child Obesity
Joseph (Yossi) Levi, Ph.D, Senior biostatistician & data scientist
Background
Nearly 20% of the age-5 children in Israel are overweight or obese, and 40% of them will turn into obese adults. These data invoke the need to identify children with high risk for becoming obese and apply an appropriate intervention program.
Challenge
Based on data of ~60000 children who were born during 2009-2014, we attempted to predict obesity at age 5 based on data collected up to age of two years. The main parameter we focused on was the Positive Prediction Value (PPV), a measure that weights the model's sensitivity and specificity/recall by the prevalence of obesity in the population.
Results
We found that the main predictors for obesity at age 5 is the child's weight and BMI at age of 18 months. Additional important predictors are gender, weight at 12 months, and weight gain from birth to 4 months, and from 12 to 18 months. Surprisingly, the contribution of "suspected" predictors such as socio-economic status, population sector and residence in the periphery was negligible. The model achieved PPV of nearly 60% and AUC of 0.72.
This model will determine intervention populations, for which we will offer (through the parents) to join an ongoing consulting program for obesity prevention.
The study was conducted in collaboration with SFI Israel, an affiliate of Social Finance Global Network
