Predicting multi-modal symptom trajectories across 7 years in veterans with chronic posttraumatic stress

Benjamin Pierce, Thomas Kirsh, Erich Kummerfeld, Adam R. Ferguson, Thomas C. Neylan, Beth E. Cohen, Sisi Ma, Jessica L. NielsonmedRxiv 2020 · DOI

Abstract

Veterans are disproportionately affected by post-traumatic stress (PTS) and its associated impacts on health and psychosocial functioning. While many improve over time, a subset experiences severe, persistent, and far-reaching impairment. This study identifies a "worst-faring" subgroup of veterans using seven years of longitudinal data and characterizes the specific predictors—ranging from social isolation to systemic inflammation—that distinguish them from their peers.

Introduction: Beyond the Symptom Checklist

Post-traumatic stress (PTS) research has historically focused on specific psychiatric symptoms like flashbacks, hypervigilance, and nightmares. However, for many veterans, PTS does not exist in a vacuum. It is often intertwined with physical health decline and social erosion. Traditional "siloed" approaches to studying veteran health often fail to capture the full picture of how these domains interact over time.

The Mind Your Heart (MYH) study sought to address this gap by moving toward a multidimensional, longitudinal perspective. By tracking 747 veterans over seven years, researchers aimed to identify those at the highest risk for long-term decline and determine which factors—beyond standard PTS scores—predict a chronic, debilitating course.

Methodology: A Multi-Stage Machine Learning Pipeline

To handle the complexity of seven years of biometric and behavioral data, the researchers employed a sophisticated four-stage analytic workflow:

  1. Dimension Reduction: Using Nonlinear Principal Component Analysis (NLPCA) and Confirmatory Factor Analysis (CFA), the team condensed nine univariate measures (including depression, physical activity, and alcohol use) into three stable "multivariate dimensions": Distress/Impairment, Distress/Activity, and Alcohol Use Concerns.
  2. Latent Growth Mixture Modeling (GMM): This technique allowed researchers to identify different "trajectory classes," recognizing that veterans do not follow a single path of recovery or decline.
  3. K-means Clustering: This step corroborated the GMM results, ensuring the identified "worst-faring" groups were statistically robust across different methods.
  4. Random Forest Classification (RFC-FS): Finally, a machine learning classifier was used to determine which baseline variables most accurately predicted membership in the high-risk group.

The "Worst-Faring" Subgroup: A Profile of Deterioration

The most significant finding was the identification of a specific subgroup—comprising approximately 10% of the cohort—that exhibited a "multidimensional profile of deterioration." Unlike other veterans who remained stable or improved, this group showed worsening outcomes across almost every metric: mental health, physical vitality, and social engagement.

Interestingly, the study found that Alcohol Use Concerns largely followed an independent trajectory. For older veterans, problematic drinking appeared to be a distinct issue that did not necessarily correlate with the broader "Distress/Impairment" dimension, suggesting that substance use interventions may need to be decoupled from general PTS treatment in this population.

The Biopsychosocial Signature of Chronic Distress

What distinguishes the veterans who struggle the most? The research identified a unique "biopsychosocial signature" for the worst-faring group, characterized by:

  • Behavioral Markers: High levels of avoidance and anhedonia (the inability to experience pleasure).
  • Interpersonal Factors: Significant social isolation, coupled with high levels of cynicism, anger, and hostility toward others.
  • Biological Indicators: Elevated markers of systemic inflammation, specifically C-reactive protein (CRP) and Interleukins (IL-6 and IL-1β), along with higher Body Mass Index (BMI).

The Random Forest model highlighted social isolation and depression as the strongest predictors, but biological markers of immune response emerged as salient predictors in over 50% of the assessment points. This suggests that chronic distress in veterans is a whole-body phenomenon, where mental health struggles are reflected in—and perhaps exacerbated by—physiological inflammation.

Clinical Implications and Future Directions

The findings suggest that for the most vulnerable veterans, treating PTS symptoms alone may be insufficient. The "worst-faring" group requires a multidisciplinary approach that targets the interconnected nature of their distress.

Effective interventions might need to simultaneously address:

  • Systemic Inflammation: Through nutritional counseling, exercise, or medical management.
  • Social Connectivity: Moving beyond individual therapy to address social isolation and interpersonal hostility.
  • Transdiagnostic Drivers: Focusing on avoidance and anhedonia as core mechanisms that keep veterans trapped in a cycle of decline.

While the study's sample was predominantly older male veterans from the Vietnam and Gulf War eras, the stability of the identified dimensions over seven years provides a robust framework for future research. The next logical step is testing multidisciplinary treatment packages designed specifically for this high-risk 10%, ensuring that those most in need do not fall through the cracks of a siloed healthcare system.

Summary updated

Abstract

Abstract Background Veterans are disproportionately affected by symptoms of post-traumatic stress (PTS) and associated poor health and psychosocial functioning. While most improve over time, others experience severe and persistent concerns. The ability to predict this latter group is critical for early intervention. Characterizing this subgroup has proven difficult, with most studies focusing on PTS and neglect a wider assessment of veterans’ wellbeing. Consequently, little is known about veterans who experience chronic symptoms and far-reaching impairment. Method The present study uses dimension reduction, growth mixture modeling, and clustering methods to identify veterans with the worst-faring trajectories of psychiatric symptoms, health, and psychosocial functioning, using data from the seven-year Mind Your Heart study (MYH) of people receiving Veterans Affairs services (n = 747). Random forest classification and feature selection were then used to examine predictors that distinguish the worst-fairing veteran group from others in the cohort. Results The combined analyses revealed a subgroup of veterans with severe and diverse symptoms across psychiatric domains, impairment in multiple facets of living, and poor health with deterioration over seven years. This subgroup was distinguished by transdiagnostic symptom severity and greater social isolation, avoidance, anhedonia, cynicism, anger/hostility, and immune response and inflammation. Conclusions Veterans whose distress spans multiple domains appear to be more broadly impaired, socially isolated, cynical or angry/hostile to others, and show elevated immunoreactivity and inflammation. Care for this population should be informed by a multidisciplinary approach that is conscious of veterans’ mental and physical health, and interpersonal needs.
BibTeX
@misc{pierce2020predicting,
  author = {Benjamin Pierce and Thomas Kirsh and Erich Kummerfeld and Adam R. Ferguson and Thomas C. Neylan and Beth E. Cohen and Sisi Ma and Jessica L. Nielson},
  title = {{Predicting multi-modal symptom trajectories across 7 years in veterans with chronic posttraumatic stress}},
  year = {2020},
  doi = {10.1101/2020.10.08.20185272},
  howpublished = {medRxiv},
}
Figure