Objective: Among military personnel posttraumatic stress disorder is strongly associated with non-specific health symptoms and can have poor treatment outcomes. This study aimed to use machine learning to identify and describe clusters of self-report health symptoms and examine their association with probable PTSD, other psychopathology, traumatic deployment exposures, and demographic factors. Method: Data were from a large sample of military personnel who deployed to the Middle East (n = 12,566) between 2001 and 2009. Participants completed self-report measures including health symptoms and deployment trauma checklists, and several mental health symptom scales. The data driven machine learning technique of self-organised maps identified health symptom clusters and logistic regression examined their correlates. Results: Two clusters differentiated by number and severity of health symptoms were identified: a small ‘high health symptom cluster’ (HHSC; n = 366) and a large ‘low health symptom cluster’ (LHSC; n = 12,200). The HHSC had significantly higher proportions of (Gates et al., 2012 ) scaled scores indicative of PTSD (69% compared with 2% of LHSC members), Unwin et al. (1999a)  scores on other psychological scales that were indicative of psychopathology, and (Graham et al., n.d. ) deployment trauma. HHSC members with probable PTSD had a stronger relationship with subjective (OR 1.25; 95% CI 1.12, 1.40) and environmental (OR 1.08; 95% CI 1.03, 1.13) traumatic deployment exposures than LHSC members with probable PTSD. Conclusion: These findings highlights that health symptoms are not rare in military veterans, and that PTSD is strongly associated with health symptoms. Results suggest that there may be subtypes of PTSD, differentiated by health symptoms.
- Health symptoms
- Machine learning
- Posttraumatic stress disorder
- Self-organised maps