Abstract:
|
Clustering trajectories of disease progression can help in understanding the variability of disease pathways. Recently, a latent class mixed effect models (LCMM) approach was developed and has been applied to a wide range of progressive diseases to discover the underlying patterns of deterioration in function. Often, biological and medical data have complicated structure and substantial noise. Therefore, we use simulation studies to explore the ability of LCMMs to accurately classify individuals for a wide range of noise and variability in trajectories and to provide guidelines for model specification when using this technique. Datasets were simulated from 24 scenarios covering different data structures and variability levels. We find that LCMMs can reliably recover trajectory subgroups and model parameters even for datasets that contain unbalanced subgroups and in which subjects are followed irregularly with different and short follow up time, if the ratio of the between-individual and total variance is relatively small. When the variability is high in the data, LCMMs have difficulty classifying individuals into the correct subgroups and thus should be applied with caution.
|