Traj2: A Native Macro Library for Single and Joint Two-Outcome Group-Based Trajectory Modeling in SAS
Year: 2024
Type: preprint
Abstract: The Group-Based Trajectory Modeling repository, developed by the Community Health and Aging Outcomes (CHAO) Lab, Institute for Health, Health Care Policy and Aging Research, Rutgers, The State University of New Jersey, focuses on identifying and analyzing latent trajectory groups in longitudinal health data. This code enables researchers to understand individual and joint trajectories, uncover patterns, and assess dependencies between multiple outcomes. Purpose: To implement a SAS native macro for group-based trajectory modeling with longitudinal data using the NLMIXED procedure as an alternative to Bobby Jones' PROC TRAJ (which relies on dynamic link libraries written in C). Because the macros contain no externally compiled components, they run as-is inside the CMS Virtual Research Data Center and other secure custodial environments where PROC TRAJ cannot be installed. Outputs: 1. Single-Outcome Model: Fits separate trajectory models for two continuous outcomes over multiple time points using a maximum likelihood estimation approach. 2. Multi-Trajectory Model: Combines the two single-outcome models into a unified model to analyze joint trajectories, capturing dependencies between the outcomes. 3. Ordinal-Probit Model: Fits trajectory models for bounded ordered scales such as ADL, frailty indices and cognitive screening scores, using a class-specific cumulative probit whose thresholds are built as cumulative sums of positive increments, with the first threshold fixed at 0. 4. Visualization: Generates visualizations comparing observed vs. predicted trajectories for latent classes and provides averaged posterior probabilities for class membership. 5. Replication materials: The Validation folder contains the complete code reproducing every table and figure in the accompanying Computer Methods and Programs in Biomedicine and MethodsX papers, for both the continuous and ordinal outcome families.
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W6887646778
Cites:
Cited by:
Related to: