Characterizing Individual and Population-Level Heterogeneity in Infant BMI Trajectories Using Flexible and Functional Modeling Approaches
Infant body mass index (BMI) follows a rapid and nonlinear trajectory, yet substantial inter-individual variation complicates accurate characterization of growth dynamics. Here we systematically compared individual- and group-based approaches for modeling longitudinal BMI trajectories in 2,114 healthy infants. Among linear spline, natural cubic spline and fractional polynomial mixed-effects models and SITAR, natural cubic splines provided the best fit while preserving individual variation in trajectory shape. Across all models, males reached their BMI peak earlier than females. To characterize population-level heterogeneity, we further compared latent class mixed models with functional principal component analysis (FPCA) followed by Gaussian mixture clustering. Whereas latent class modeling produced numerous small classes, FPCA identified three interpretable trajectory groups in both sexes, including a rare subgroup characterized by higher BMI and delayed or unestimable peak timing. Maternal gestational diabetes was associated with earlier BMI peak timing in females and was enriched in the rare female trajectory subgroup. Together, these results support combining flexible mixed-effects modeling with functional trajectory analysis to characterize both individual growth dynamics and population-level heterogeneity during infancy.