A Latent Trajectory Analysis for Multivariate Outcomes with Mixed-Scale: Application to Alzheimer's Disease Neuroimaging Initiative

Abstract

Latent trajectory analysis is a statistical method for explaining heterogeneity by partitioning patients into homogeneous subgroups based on similarities in outcome variables. In the context of clinical work, patients often do not follow the same course of illness or treatment response, and traditional analyses often average across patients, masking important subgroups. This work proposes a novel latent trajectory model for multivariate longitudinal outcomes with mixed types and uses the expectation-maximization algorithm as an estimation strategy. The proposed model can identify individual-level, time specific latent class memberships and a latent trajectory membership that describes how the latent class memberships change over time. By capturing these dynamic changes, we can highlight patients at higher risk of poor outcomes, reveal early indicators of improvement or decline, and ultimately support more individualized treatment planning. We present an application of our methodology to the Alzheimer's Disease Neuroimaging Initiative (ADNI), a longitudinal, multi-center, observational study to validate biomarkers for Alzheimer disease (AD) clinical trials.