Optimum Grouping of Children Based on Early Childhood Growth Patterns

• Activity Number: 520
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Md Jobayer Hossain*+ and Mingxing Gong and Tim Bunnell and Samuel S. Gidding and Tim Wysocki
• Companies: Nemours Biomedical Research and University of Delaware and A.I. duPont Children Hospital and A.I. duPont Children Hospital and Nemours Children Clinic
• Keywords: growth patterns ; BMI z score ; cluster analysis ; group ; principal component analysis ; mixed effect model

Abstract

Early childhood growth patterns are potential determinants of body composition and body weight at later ages. Weight-for-length (for age < 2 years) and body mass index (BMI) (for age = 2 years) are commonly used measures of body compositions and weight. Standardized scores of these two measures are weight-for-length z- score and BMI z-score. For simplicity, we termed both as bmiz score. We planned to group 3365 children based on the patterns of temporal changes in the trajectories of bmiz score of their first five years of life. We collected (or interpolated when data for a time point was not available) bmiz score at every six-month interval of age (a total of 11 time points) for each child. We performed several suitable cluster analyses of (i) raw bmiz score of different time points (treating bmiz at each time point as a separate variable), (ii) PC scores out of principal component analysis of bmiz score, (iii) random coefficients (that specify the individual growth model) of mixed effects model, and (iv) coefficients of auto regressive model for each patient. In this research, we are attempting to evaluate the performance of these cluster analyses.