Pairwise approach for analysis and reporting of child's free sugars intake from a birth cohort study.

Abstract

The prospective cohort design is an important research design, but a common challenge is missing data. The purpose of this study is to compare three approaches to managing missing data, the pairwise (n=1386 children), the partial or modified pairwise (n=1019) and the listwise (n=546), to characterize the trajectories of children's free sugars intake (FSI) across early childhood. By applying the Group-based Trajectory Model Technique to three waves of data collected from a prospective cohort study of South Australian children, this study examined the three approaches in managing missing data to validate and discuss children's FSI trajectories. Each approach identified three distinct trajectories of child's FSI from 1 to 5 years of age: (1) 'low and fast increasing', (2) 'moderate and increasing' and (3) 'high and increasing'. The trajectory memberships were consistent across the three approaches, and were for the pairwise scenario (1) 15.1%, (2) 68.3% and (3) 16.6%; the partial or modified pairwise (1) 15.9%, (2) 64.1% and (3) 20.0%; and the listwise (1) 14.9%, (2) 64.9% and (3) 20.2% of children. Given the comparability of the findings across the analytical approaches and the samples' characteristics between baseline and across different data collection waves, it is recommended that the pairwise approach be used in future analyses to optimize the sample size and statistical power when examining the relationship between FSI in the first years of life and health outcome such as dental caries.