Online Program Home
My Program

Abstract Details

Activity Number: 71 - Longitudinal/Correlated Data
Type: Contributed
Date/Time: Sunday, July 28, 2019 : 4:00 PM to 5:50 PM
Sponsor: Biometrics Section
Abstract #304888 Presentation
Title: Identification of Disease Subtypes Using Multivariate Longitudinal Data: a Comparison of Growth Curve Mixture Models and a Two-Stage Cluster Analysis Approach
Author(s): Benjamin E. Leiby* and Md Jobayer Hossain and Ayako Shimada
Companies: Thomas Jefferson University and Nemours children Healthcare Systems and Thomas Jefferson University
Keywords: Latent class; Clustering; Longitudinal; Multivariate

In many chronic conditions, the majority of patients will experience stability in their disease characteristics. Models that assume a single population may identify no change over time. The presence of smaller subtypes who have significant change is often known from clinical experience. Early identification of these subtypes is important for better classification of disease and development of interventions. We consider a longitudinal cohort of patients with glaucoma with annual measures of multiple outcomes, some in both eyes. We use growth curve mixture models to identify subtypes of longitudinal trajectories and discuss the challenges of applying multivariate models to data of this type, including relaxing assumptions of conditional independence to allow for correlation among both eyes and equality of variances across latent classes. We also discuss sample size requirements when the subtype is relatively rare. We consider an alternative approach that applies conventional cluster analysis methods to empirical best linear unbiased predictors from mixed effects models. Through data analysis and simulation, we identify the advantages and disadvantages of both approaches.

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

Back to the full JSM 2019 program