Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 203 - Contemporary Machine Learning
Type: Contributed
Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #312612
Title: Post-Hoc Mixture Models of the Best Linear Unbiased Predictors from Linear Mixed Effects Models to Classify Longitudinal Data with Haphazardly Spaced Intervals: A Simulation Study
Author(s): Md Hossain* and Benjamin E Leiby
Companies: Nemours Biomedical Research, A.I. DuPont Children's Hospital and Division of Biostatistics
Keywords: mixture model; mixed effects model; subgroups; classification; eigenvalue; decomposition

In precision medicine, vast amounts of naturally occurring data are used to derive evidence-based knowledge. Recognizing distinct trajectories, identifying potential predictors, and making data-driven decisions are areas of growing interest. Mixture-based mixed effects models, implemented mainly in R and Mplus are useful for this. While theoretically sound, these approaches often fail to fully address the problem due to computational complexities. Linear mixed effects models ballpark curves of irregularly spaced longitudinal data with optimum precision. Random effects (REs), assumed to be distributed as multivariate normal (MVN), vary across individuals accounting for sources of heterogeneity. To identify presence of subgroups, we apply post-hoc Gaussian finite mixtures on empirical BLUPs assuming a mixture of MVN of REs to classify trajectories. Resulting ellipsoids can vary by center or by features- size, volume and orientation that are determined by eigenvalue decomposition of covariance matrix. This study used simulations to extend previous application of real data to evaluate the classification performance of this method compared with that of existing methods.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program