What is nonlinear trajectory modelling with splines?

Speaker(s):

Ahmed Elhakeem, University of Bristol

Abstract:

Many developmental processes display non-linear patterns of change with age/time, which makes it important to accurately model the shape of their trajectory. Splines are highly flexible functions (formed by sets of connected piecewise polynomials) that can be used to model most complex nonlinear trajectories. In this session, I will introduce the concept of splines, focusing on three commonly used types: linear and natural cubic regression splines, and penalized regression splines. Using a synthetic cohort dataset and R code (to be provided on GitHub), I will then demonstrate how the standard linear mixed effects model - common method for handling correlated repeated measurements when analyzing longitudinal trajectories - can be easily extended to include spline functions (for age) to a model nonlinear trajectory for a repeated continuous outcome (early life body mass index). I will also introduce and demonstrate a spline-based nonlinear mixed effects model (SITAR) that is useful for modelling pubertal growth.