Growth models used for describing the dynamics of body weight and height generally consider each trait independently. We proposed modeling height and weight trajectories jointly with a nonlinear heteroscedastic mixed model based on the Jenss-Bayley growth function with correlated individual random effects and using Bayesian inference techniques. Simulations showed that our model provides good estimates of the growth parameters. We illustrated how it can be used to assess the associations between maternal smoking during pregnancy, an early-life factor potentially involved in prenatal programming of obesity, and children's growth from birth to 5 years of age. We used real data from the EDEN study, a large French mother-child cohort study with a high number of height and weight measurements (a total of approximately 30 000 measurements for each of the 2 traits across the 1666 children). Our results supported the existence of a relationship between maternal smoking during pregnancy and growth from birth to 5 years of age. Children from mothers who smoked throughout pregnancy were shown to display a higher body mass index from the first few months of life onwards compared to children from nonsmokers. At 5 years of age, their mean body mass index was 0.21 kg/m2 higher than unexposed children. It was mainly explained by the fact that these children tended to be smaller at birth but rapidly exceeded the weight of children from nonsmokers postnatally.
Keywords: Bayesian modeling; joint model; multivariate; nonlinear modeling; postnatal growth trajectories.
Copyright © 2017 John Wiley & Sons, Ltd.