Infant growth modelling and assessment of growth

Abstract

Assessment of factors influencing infant growth is best performed using a modelling approach; however this is difficult due to the high initial rate of growth and wide variability. The aim is to obtain a model which produces a good fit to the data with a minimum number of parameters. A number of parametric models have been used, motivated mainly by ability to fit data, rather than biological considerations. Biologically it is unlikely that growth can be modelled by a simple function, so a semi-parametric model appears more appropriate and may produce more interpretable parameters. A semi-parametric model is based around a flexible shape which is common to all subjects, combined with parameters that transform the curve for individual subjects, with only few models of this type available. When fitting models, a mixed effects model approach is preferred, rather than fitting subjects individually. Covariates may be included in the models as either timeindependent or time-dependent covariates, but interpretation may be difficult for time-dependent covariates. The models are compared using data on weight and length in the first 2 years of life. A semi-parametric model, the shape invariant model, had similar fit to the Jenss-Bayley model, but with more easily interpretable parameters. A quartic (fourth degree) polynomial did have a superior fit but at the expense of a larger number of parameters, and possible overfitting.