Park, A. H., Patel, H., Mirabelli, J., Eder, S. J., Steyrl, D., Lueger-Schuster, B., Scharnowski, F., O’Connor, C., Martin, P., Lanius, R. A., McKinnon, M. C., & Nicholson, A. A. (2025). Machine learning models predict PTSD severity and functional impairment: A personalized medicine approach for uncovering complex associations among heterogeneous symptom profiles. Psychological Trauma: Theory, Research, Practice, and Policy, 17(2), 372–386. https://doi.org/10.1037/tra0001602

Key Takeaways

Focus: The study explored the potential of machine learning models to predict PTSD symptom severity and functional impairment in a real-world treatment-seeking population using self-report clinical data.
Aims: The research aimed to determine if machine learning models could effectively predict PTSD-related illness and identify key predictors of symptom severity and functional impairment.
Findings: The study found that machine learning models could accurately predict PTSD symptom severity and functional impairment, with anxiety, dissociation, depression, negative trauma-related beliefs, and emotion dysregulation being the most significant predictors.
Implications: The findings suggest that machine learning models can inform personalized medicine approaches to maximize trauma recovery in real-world inpatient populations, highlighting the importance of targeting specific symptoms for effective trauma-related interventions.

Rationale

Posttraumatic stress disorder (PTSD) is a complex and heterogeneous disorder with a wide range of symptom presentations.

Traditional statistical methods often struggle to capture the complex relationships among PTSD symptoms and their associated factors.

Machine learning techniques offer a promising approach to overcome these limitations by allowing for a more accurate representation of complex nonlinear relations among PTSD symptoms and the disorder’s correlates.

Previous research has identified several factors associated with PTSD severity and functional impairment, including dissociation, emotion dysregulation, depression, anxiety, and cognitive dysfunction.

However, most studies have focused on the associations across variables in isolation, making it difficult to account for the complex relationships among relevant clinical variables.

This study aimed to address this gap by employing machine learning models to predict PTSD symptom severity and functional impairment in a real-world clinical sample.

The study hypothesized that the models would confirm features that have been identified in the literature as contributing to the severity of PTSD symptoms and that PTSD symptom severity and cognitive dysfunction would contribute to predicting functional impairment.

Method

Procedure

The study used secondary clinical data collected from adults admitted to an inpatient unit for PTSD in Canada between 2017 and 2019.

The participants completed a battery of self-report questionnaires at admission, including measures of trauma-related symptoms, other mental health symptoms, functional impairment, and demographic information.

Two nonlinear machine learning models (extremely randomized trees) were trained to predict PTSD symptom severity and functional impairment.

Model performance was assessed based on predictions in novel subsets of patients.

Sample

The sample consisted of 393 adults (196 female) aged 18-80 (M = 43.7 + 9.9) seeking inpatient treatment for PTSD.

Participants were required to be 18 years or older, able to participate in group therapy, able to tolerate their emotions and manage basic personal safety, and willing to abstain from all substances during the program.

Individuals were excluded if they had placed exclusions on using their personal health information for research purposes.

Measures

The study used several self-report measures, including:

PTSD Symptom Severity: The PTSD Checklist for DSM-5 (PCL-5)
Functional Impairment: The World Health Organization Disability Assessment Schedule 2.0 (WHODAS 2.0)
Other Mental Health Symptoms: The Depression, Anxiety, and Stress Scale (DASS-21), the Multiscale Dissociation Inventory (MDI), and the Difficulties in Emotion Regulation Scale (DERS)
Trauma-Related Guilt and Shame: The Trauma-Related Guilt Inventory (TRGI) and the Trauma-Related Shame Inventory (TRSI)
Negative Trauma-Related Cognitions: The Posttraumatic Cognitions Inventory (PTCI)
Cognitive Dysfunction: The Cognitive Failures Questionnaire (CFQ)

Statistical Measures

The study employed machine learning regression analyses using the Extremely Randomized Trees (ERT) algorithm. Model performance was assessed using a nested cross-validation procedure.

The contribution of each input feature was assessed using permutation feature importance and model-based feature importance.

Results

Hypothesis 1: Prediction of PTSD Symptom Severity

The machine learning model accurately predicted PTSD symptom severity, explaining approximately 43% of the variance (R2 avg = .43, R2 median = .44, p = .001).

The predictors contributing the most to the variance were anxiety (12.6%), dissociation (7.2%), negative trauma-related beliefs about others (6.0%), depression (4.7%), and emotion dysregulation (4.1%).

Hypothesis 2: Prediction of Functional Impairment

The model also effectively predicted functional impairment, accounting for approximately 32% of the variance (R2 avg = .32, R2 median = .33, p = .001).

The most influential predictors were anxiety (8.1%), PTSD symptom severity (6%), cognitive dysfunction (5.8%), dissociation (2.9%), and depression (2.0%).

Insight

This study provides valuable insights into the complex interplay of factors contributing to PTSD severity and functional impairment.

By utilizing machine learning models, the research was able to capture the nonlinear relationships among various clinical variables, going beyond traditional statistical methods that often focus on isolated associations.

The findings confirm the importance of several factors previously identified in the literature, such as anxiety, depression, dissociation, negative trauma-related beliefs, and emotion dysregulation, while also highlighting their relative contributions to PTSD severity and functional impairment.

The study’s findings extend previous research by demonstrating the potential of machine learning models to predict PTSD severity and functional impairment with high accuracy in a real-world clinical setting.

This has significant implications for personalized medicine approaches, as it suggests that clinicians could use machine learning models to identify individuals at risk for more severe PTSD or functional impairment and tailor treatment plans accordingly.

The study also underscores the importance of addressing specific symptoms, such as anxiety, dissociation, and negative trauma-related beliefs, in PTSD treatment.

The findings suggest that targeting these symptoms may lead to more effective interventions and improved outcomes for individuals with PTSD.

Future research could explore the application of machine learning models to predict treatment response in PTSD, identify subgroups of individuals with distinct symptom profiles, and develop personalized treatment recommendations.

Additionally, incorporating other data sources, such as neuroimaging and physiological measures, could further enhance the predictive power of these models.

Implications

The findings of this study have several implications for practitioners and policymakers:

Personalized Treatment: The study highlights the potential of machine learning models to inform personalized medicine approaches for PTSD. Clinicians could use these models to identify individuals at risk for more severe symptoms or functional impairment and tailor treatment plans accordingly. This could lead to more targeted and effective interventions, improving outcomes for individuals with PTSD.
Symptom-Specific Interventions: The study underscores the importance of addressing specific symptoms, such as anxiety, dissociation, and negative trauma-related beliefs, in PTSD treatment. Practitioners should consider incorporating evidence-based interventions that specifically target these symptoms to improve overall treatment effectiveness.
Early Identification and Intervention: The study’s findings suggest that machine learning models could be used to identify individuals at risk for developing PTSD or experiencing more severe symptoms. This could enable early intervention efforts, potentially preventing the development of chronic PTSD and reducing the associated functional impairment.
Policy Recommendations: Policymakers could consider incorporating machine learning models into clinical practice guidelines for PTSD, promoting the use of these models to inform treatment decisions and improve the quality of care for individuals with PTSD. Additionally, policies could support the development and implementation of early intervention programs based on the identification of at-risk individuals using machine learning models.

Strengths

Real-World Setting: The study used data from a real-world clinical sample, increasing the generalizability of the findings to clinical practice.
Machine Learning Approach: The use of machine learning models allowed for the identification of complex nonlinear relationships among clinical variables, providing a more comprehensive understanding of PTSD severity and functional impairment.
Prediction Accuracy: The models demonstrated high accuracy in predicting PTSD symptom severity and functional impairment, suggesting their potential for clinical application.
Personalized Medicine: The findings support the use of machine learning models to inform personalized medicine approaches for PTSD, potentially leading to more targeted and effective interventions.

Limitations

Self-Report Measures: The study relied on self-report measures, which may be subject to response bias and social desirability effects.
Secondary Data: The use of secondary data limited the ability to control for potential confounding variables and collect additional information that may be relevant to PTSD severity and functional impairment.
Sample Characteristics: The sample consisted of individuals seeking inpatient treatment for PTSD, which may not be representative of the broader population of individuals with PTSD.
Limited Data Sources: The study only included self-report clinical data. Incorporating other data sources, such as neuroimaging and physiological measures, could further enhance the predictive power of the models.

Reference