Classification of Distinct Trajectories in Longitudinal Data with Irregular Spaced Intervals: Heterogeneous Linear Mixed Model Vs Mixture Modeling of BLUPs from Linear Mixed Model

• Activity Number: 466 - Personalized/Precision Medicine I
• Type: Contributed
• Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Md Jobayer Hossain* and Benjamin E. Leiby
• Companies: Nemours children Healthcare Systems and Thomas Jefferson University
• Keywords: Classification; irregular spaced time; trajectories; linear mixed model ; heterogeneous; mixture distribution of random effects

Abstract

Classifying heterogeneous trajectories into distinct groups has gained huge appeal in this precision medicine era when massive amounts of naturally occurring longitudinal data can be used to derive evidence-based knowledge for facilitating precise diagnosis, prevention, and tailored treatment. Methods on the topic are sparse. Mixed models with a mixture distribution of random effects (HLME) classify trajectories through estimating profiles and computing posterior probabilities of belonging to a class. Two forms of the method are implemented in R and M-plus among standard statistical software. The method becomes computationally complex with increased data size and level of unbalancedness. It also requires specification of the number of classes prior to analysis. Likelihood function may have multiple local maxima. We introduced a flexible approach that applies mixture models to empirical BLUPs from linear/ spline mixed models. The method reduces drawbacks of HLMEs and works well on large datasets. This study compares the classification ability of two methods using real and simulated datasets of complex temporal curves and identifies situations when one method outperforms the other.