Abstract
This commentary rounds off a collection of papers focusing on statistical methods for analysing growth data. In two papers, Anderson and colleagues discuss growth trajectory models in early life, using data on height and weight from the HBGDki initiative, while two papers from Ohuma and Altman review methods for centile construction, with data from the INTERGROWTH-21(st) project used to provide worked examples of centiles for birthweight and fetal head circumference. Anderson et al focus on four growth trajectory models: quadratic Laird-Ware, SITAR, brokenstick, and FACE, where the latter two fit better than the former two applied to length data in individuals. On this basis, they recommend brokenstick and FACE for future work. However, they do not discuss the timescale on which the growth models assess growth faltering nor the relevance of this timescale to later health outcome. Models that best detect short-term fluctuations in growth (brokenstick and FACE) may not necessarily be best at predicting later outcome. It is premature to exclude the quadratic Laird-Ware or SITAR models, which give a parsimonious summary of growth in individuals over a longer timescale. Ohuma and Altman highlight the poor quality of reporting in fetal centile studies, and they provide recommendations for good practice. Their birthweight centiles example illustrates both the power of the GAMLSS software and its capacity for misuse. The longitudinal fetal head circumference centiles are biased such that 5% of infants are below the 3(rd) centile and 5% above the 97(th) .