Latent Trajectory Modeling for Discovering Longitudinal Subgroups

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 269 papers74 conceptsfrozen 2026-10-04

In many scientific fields, the "average" is a lie. If half of a patient population recovers rapidly from a disease while the other half deteriorates, the statistical average suggests a population that is slowly, uniformly improving. This "one-size-fits-all" perspective obscures the very heterogeneity that clinicians and researchers need to understand.

To solve this, researchers use a family of techniques known as Latent Trajectory Modeling—including Latent Class Growth Analysis (LCGA) and Growth Mixture Modeling (GMM). These methods move beyond population-wide averages to identify unobserved (latent) subgroups that share similar patterns of change over time. By clustering individuals based on the shape of their progression, these models reveal the distinct "stories" hidden within longitudinal data.

The Fallacy of the Static Snapshot

Traditional analysis often relies on static snapshots—measuring a variable at a single point in time. However, as Chifa Ma and colleagues demonstrated in a 10-year study of diabetes, the pattern of change (e.g., an unstable or increasing HbA1c trajectory) is a much more powerful predictor of mortality than a single blood sugar measurement Effect of Hemoglobin A1c Trajectories on Future Outcomes in a 10-Year Cohort With Type 2 Diabetes Mellitus.

Similarly, in acute bipolar mania, Ilya Lipkovich and John P. Houston used hierarchical clustering to move beyond binary "responder vs. non-responder" labels. They identified four distinct recovery shapes, showing that a patient's clinical features at baseline can predict whether they are on a path toward rapid recovery or a quick relapse Identifying patterns in treatment response profiles in acute bipolar mania: a cluster analysis approach.

How Latent Trajectories Work

The core mechanism involves fitting finite mixture models to time-series data. Instead of assuming every individual is a variation of a single mean, the model assumes the population is composed of several distinct, unobserved classes.

  1. Latent Class Growth Analysis (LCGA): As described by Bengt Muthén, LCGA assumes that all individual variation is captured by the differences between groups. Every member of a specific class is modeled as following the exact same trajectory Longitudinal modeling of population heterogeneity: Methodological challenges to the analysis of empirically derived criminal trajectory profiles.
  2. Growth Mixture Modeling (GMM): This is a more flexible evolution of LCGA. It allows for variation within the groups, acknowledging that while people in a "deteriorating" group share a trend, they don't all follow identical paths.
  3. Group-Based Trajectory Modeling (GBTM): This approach, used by Jinqiang Zhuang to study sepsis, identifies subphenotypes based on physiological markers like blood lactate. By mapping these trajectories, clinicians can identify high-risk patients whose lactate levels remain persistently high, requiring more aggressive intervention Lactate trajectories and outcomes in patients with sepsis in the intensive care unit: group-based trajectory modeling.

Predicting Outcomes Through Temporal Shapes

The power of this approach lies in its predictive utility. By identifying which "latent class" an individual belongs to early on, researchers can forecast long-term outcomes with higher accuracy than traditional risk factors.

For example, Mary D. Sammel and Naisyin Wang used Bayesian joint modeling to identify subgroups in hormone trajectories. They found that the timing of shifts in follicle-stimulating hormone (FSH) could predict the age of menopause 20% more accurately than standard clinical markers Bayesian estimation of associations between identified longitudinal hormone subgroups and age at final menstrual period.

In geriatrics, Jialu Yang found that a rapidly worsening ability to perform daily tasks was a significantly stronger predictor of Mild Cognitive Impairment than a person's baseline ability level Instrumental activities of daily living trajectories and risk of mild cognitive impairment among Chinese older adults.

Beyond Medicine: Social and Biological Networks

The application of latent trajectories extends into social dynamics and complex biological ecosystems. Felipe Montes and colleagues combined Latent Transition Analysis (LTA) with network modeling to show how anti-smoking norms spread through adolescent friendship groups. LTA allowed them to track how individuals move between "latent states" (e.g., from pro-smoking to anti-smoking) over time Exploring the mechanistic pathways of how social network influences social norms in adolescent smoking prevention interventions.

In the study of the infant microbiome, Ziliang Zhu and Tengfei Li used Gaussian Mixture Models to cluster children based on the growth and decline trajectories of specific gut bacteria, linking these microbial "growth curves" to cognitive milestones Gut microbiota maturation and early behavioral and cognitive development.

The Danger of Reification

While these models are powerful, Bengt Muthén cautions against "reification"—the tendency to treat statistical clusters as if they are immutable, real-world categories. Different modeling choices (like LCGA vs. GMM) can yield different "stories" from the same data. The goal of latent trajectory modeling is not just to find clusters, but to provide a more nuanced, dynamic lens through which to view human development, disease progression, and social change.

Papers in this version

Showing 60 of 69 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 74 papers
  2. Version 1 · 2026-10-04 · 69 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_group_based_trajectory_modeling_ff239b7c,
  title = {group-based trajectory modeling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/group-based-trajectory-modeling/rdr_ff239b7c33de4d0c90ef4073a2b510bd},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_latent_trajectory_modeling_for_discoveri_ff239b7c_v1,
  title = {Latent Trajectory Modeling for Discovering Longitudinal Subgroups},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/latent-trajectory-modeling-for-discovering-longitudinal-subgroups/rdr_ff239b7c33de4d0c90ef4073a2b510bd/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}