Modeling sparse longitudinal data in early neurodevelopment

Functional maturation in visual pathways predicts attention to the eyes in infant rhesus macaques: Effects of social status

“…This study joins the body of literature showcasing the utility and advantage of using nonparametric modeling approaches, in particular FPCA, to map developmental change and test for associations between dynamic neurobehavioral processes ( Jones and Klin, 2013 , Dai et al, 2019b , Dai et al, 2019a , Chen et al, 2021 ). Here, we identify developmental milestones in trajectories of social visual attention and functional brain development in a data-driven manner.…”

“…To achieve this aim, we implemented a non-parametric curve-fitting approach specifically designed for sparse longitudinal data. Functional Principal Components Analysis (FPCA) incorporates functional data analysis and empirical dynamics methods to offer unique advantages for modeling sparse longitudinal data ( Yao et al, 2005 , Chen et al, 2021 ). First, FPCA is designed to handle datasets with non-uniform sampling and missing data points, a common reality in longitudinal developmental data.…”

Network evolution of regional brain volumes in young children reflects neurocognitive scores and mother’s education

“…All data were drawn from an ongoing RESONANCE longitudinal study of healthy and neurotypical brain and cognitive development from early childhood to preadolescence, based at Brown University in Providence, RI, USA. RESONANCE 39 , 40 was designed as an accelerated longitudinal study of a large community cohort of healthy children, with around half of the cohort enrolled between the ages of 2 and 8 months, and the remaining children between the ages of two and four years. Children in this study are typically enrolled between birth and 2 years of age, and then followed with repeated study visits and assessments at 6 or 12-month increments depending on child age.…”

Trajectories of brain volumes in young children are associated with maternal education

“…The most prominent distinguishable sub‐volumes of brain that can be extracted from MRI scans are labeled as cerebrospinal fluid (CSF), grey matter (GM), and white matter (WM). It is customary to adjust the observed volumes of CSF, WM, and GM by total brain volume, which leads to recordings of proportional cerebrospinal fluid volumes (pCSF), proportional grey matter (pGM), and proportional white matter (pWM) (Chen et al, 2021 ). These are compositional data as the fractions sum to 1 and are nonnegative (Aitchison, 1982 ).…”

“…To handle longitudinal data without considering the compositional nature of observations, the most commonly used approaches are based on mixed effects modeling, where one uses fixed effects for population effects and random effects for individual differences (Bernal‐Rusiel, Greve, et al, 2013 ; Bernal‐Rusiel, Reuter, et al, 2013 ; Lindstrom & Bates, 1990 ; Pinheiro & Bates, 2006 ; Sanford et al, 2018 ). An alternative and often preferable approach is functional data analysis Chen et al ( 2021 ) where random effects are included in the form of functional principal component scores. An extension of mixed effects modeling to longitudinal compositional data was developed in Chen and Li ( 2016 ) with a focus on microbiome data.…”