Multi-level modelling of longitudinal child growth data from the Birth-to-Twenty Cohort: a comparison of growth models

Background: Different structural and non-structural models have been used to describe human growth patterns. However, few studies have compared the fitness of these models in an African transitioning population. Aim: To find model(s) that best describe the growth pattern from birth to early childhood using mixed effect modelling. Subjects and methods: The study compared the fitness of four structural (Berkey-Reed, Count, Jenss-Bayley and the adapted Jenss-Bayley) and two non-structural (2nd and 3rd order Polynomial) models. The models were fitted to physical growth data from an urban African setting from birth to 10 years using a multi-level modelling technique. The goodness-of-fit of the models was examined using median and maximum absolute residuals, Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). Results: There were variations in how the different models fitted to the data at different measurement occasions. The Jenss-Bayley and the polynomial models did not fit well to growth measurements in the early years, with very high or very low percentage of positive residuals. The Berkey-Reed model fitted consistently well over the study period. Conclusion: The Berkey-Reed model, previously used and fitted well to infancy growth data, has been shown to also fit well beyond infancy into childhood.