Methods for comparing groups of growth curves when patterns of observation times vary among subjects

A variety of statistical techniques is available if growth is to be compared for two groups followed up over a certain period. In practice straightforward application of such techniques may be hampered if patterns of observation times vary among subjects. Different approaches to deal with such variation are illustrated by analysing data from infants (0-18 months) living in Lesotho, Southern Africa. Data comprises 236 individual series of multiple measurements on five anthropometric variables. Individual patterns of observation times differ because of variation in a) age at entry, b) length of follow-up, and c) observation intervals (nominally 4 weeks). First an ad hoc approach is followed by taking a few summary measures for each individual series, e.g. 'growth velocity'. By contrast two other approaches hinge on preliminary data transformation required to make standard techniques applicable. The pros and cons of the different approaches are discussed.