FORECASTING BODY MASS INDEX DISTRIBUTIONS AMONG THE US CHILDREN USING A LONGITUDINAL DATASET

2013 SMDM Abstract - David Daeho Kim

Purpose:
To forecast Body Mass Index (BMI) distributions among children and adolescents in the United States (U.S.) using a longitudinal dataset

Methods:
We developed a statistical model to generate transition probabilities between BMI categories, and then using the estimated transition probabilities predicted the future BMI distributions among children aged 6 to 17.  We defined four different BMI categories based on age-adjusted BMI percentiles distributions and Center for Disease Control recommendations: Underweight, Normal, Overweight and Obese.  In a nationally representative longitudinal dataset, Early Childhood Longitudinal Program (ECLS) that followed children at kindergarten in 1999 to 8^th grade, the measured children's weight and height were reported five time periods.  We used four BMI categories in (N+1)th period as ordered outcomes to run an ordered logistic regression, and four BMI categories in Nth period, age, gender, race, and interaction terms are used for covariates. The transition probabilities from each of four BMI categories in the previous period to four BMI categories in the next period were estimated using recycled predictions.

Results:

In the spring of 1999, among kindergarten-aged children, there are 3.7% of underweight, 70.3% of normal, 14.6% of overweight, and 11.4% of obese children.  By the time of reaching 8^th grade in the spring of 2007, there are 2.3% of underweight, 62.0% of normal, 16.9% of overweight, and 18.8% of obese children.  Over an 8-year period of following, in average 83% of children who were obese in the previous period had remained in the same category, while 14.4% and 2.7% had moved to overweight and normal category respectively. Overall trends between 1999 and 2007 showed that children with underweight had been stable at 2-3% prevalence, while children with obesity and overweight increased until age 11, then the prevalence seemed to reach a plateau.

Conclusion:  

Based on the transition probabilities, the estimated prevalence of obesity (18.8%) and overweight (16.9%) corresponds with the published prevalence of obesity and overweight in 2007 among children and adolescent.  The transition probabilities estimated using the longitudinal dataset would be valuable to forecast the future trends of BMI distribution and to predict the critical age for childhood obesity.