Targeted use of growth mixture modeling: a learning perspective

Area of Science:

• Statistical Learning
• Biostatistics
• Psychometrics

Background:

• Growth mixture modeling (GMM) identifies latent subpopulations with heterogeneous outcome trajectories.
• Conventional GMM uses empirical model fitting for candidate model generation (unsupervised learning).

Purpose of the Study:

• To propose a novel statistical learning approach integrating unsupervised and supervised learning for GMM.
• To enhance the utility of latent trajectory classes as prediction targets in clinical prognostic models.

Main Methods:

• Utilized conventional GMM for unsupervised candidate model generation via empirical model fitting.
• Evaluated candidate GMM models using supervised learning: prediction of trajectory types by baseline features.
• Applied the approach to data from the Longitudinal Assessment of Manic Symptoms study.

Main Results:

• Demonstrated a new direction for GMM application in statistical learning.
• Showcased the predictive validity of latent trajectory classes using clinical and demographic baseline features.
• Illustrated the utility of GMM-derived trajectory classes as outcomes in prognostic models.

Conclusions:

• The proposed approach effectively integrates unsupervised and supervised learning for GMM.
• Latent trajectory classes derived from GMM can serve as valuable prediction targets in clinical settings.
• This methodology offers a robust framework for developing more accurate clinical prognostic models.