Shape invariant mixture model for clustering non-linear longitudinal growth trajectories

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Growth Modeling

Background:

• Clustering individual trajectories in longitudinal studies is crucial for understanding developmental patterns.
• Non-linear growth, sparse data, and irregular time points present significant modeling challenges.
• Existing methods may not adequately capture complex trajectory shapes in biological data.

Purpose of the Study:

• To propose a novel shape invariant growth mixture model for clustering non-linear longitudinal trajectories.
• To address challenges posed by sparse and irregularly timed data.
• To apply the model to hormone profiles in pregnant women.

Main Methods:

• Developed a shape invariant growth mixture model.
• Employed Bayesian inference using Markov Chain Monte Carlo (MCMC) for parameter estimation.
• Conducted simulation studies to compare the proposed model with traditional growth mixture models and functional clustering.

Main Results:

• The proposed model demonstrated superior performance in clustering non-linear trajectories compared to existing methods in simulation studies.
• Analysis of pregnant women's hormone profiles revealed distinct non-linear growth patterns.
• The model effectively handles sparse and irregularly sampled longitudinal data.

Conclusions:

• The shape invariant growth mixture model provides a robust framework for clustering complex, non-linear growth trajectories.
• This approach enhances the analysis of longitudinal biological data, particularly when measurements are sparse and irregular.
• The model offers valuable insights into developmental processes, as demonstrated in the study of maternal hormone profiles.