A recent study developed a machine learning model that accurately predicts the risk of posttraumatic stress disorder (PTSD) before military deployment. By analyzing pre-deployment assessment data from 4,771 soldiers, the model utilized 58 core predictors and achieved an area under the curve of 0.74. Identifying one-third of participants at the highest risk accounting for 62.4% of PTSD cases, this research supports pre-deployment PTSD risk stratification for veterans to improve early intervention and prevention strategies.
A recent study published in JAMA Network Open highlights the development of a machine learning (ML) model capable of accurately predicting the risk of posttraumatic stress disorder (PTSD) before military deployment in the United States. Military service often exposes individuals to life-threatening and traumatic events that can result in PTSD, a condition with severe implications for their well-being. Given that veterans are more prone to experiencing post-traumatic stress compared to civilians, accurately predicting PTSD risk has become a crucial area of research.
In this particular study, researchers collected pre-deployment assessment data from 4,771 soldiers belonging to three US Army brigade combat teams, approximately one to two months before their deployment to Afghanistan. Follow-up assessments were conducted three to nine months after deployment. Leveraging this data, the research team developed multiple ML models to forecast PTSD, utilizing up to 801 predictors obtained from pre-deployment assessments. Through cross-validation, they identified the optimal model.
The performance of the final model was evaluated using metrics such as the area under the receiver operating characteristics curve and expected calibration error in a separate cohort. Of the veterans included in the study, around 15.4 percent (746 individuals) met the criteria for PTSD after deployment.
The optimal ML model, employing a gradient-boosting algorithm, utilized 58 core predictors to assess the risk of PTSD. The model achieved an area under the curve of 0.74 and a low expected calibration error of 0.032. These core predictors encompassed 17 distinct domains, including factors such as stressful experiences, social network, substance use, childhood or adolescence, unit experiences, health, injuries, irritability or anger, personality, emotional problems, resilience, treatment, anxiety, attention or concentration, family history, mood, and religion.
Interestingly, the study revealed that one-third of the participants at the highest risk accounted for 62.4 percent of PTSD cases. Based on these findings, the researchers suggested that it is feasible to stratify pre-deployment PTSD risk for US veterans. Such an approach could facilitate the development of early intervention and prevention strategies, ultimately improving outcomes and quality of life.
This research effort aligns with the broader interest in leveraging artificial intelligence (AI) to enhance PTSD care. For instance, the US Defense Health Agency (DHA) recently announced its collaboration with AiCure, an AI and data analytics company based in New York. The partnership aims to optimize treatments and enhance patient support by evaluating the effectiveness of therapies for service members and veterans, including those with PTSD. AiCure’s tools will also be employed by DHA to support its PTSD adaptive platform trial (APT), which seeks to advance precision medicine approaches for prescribing PTSD treatments and informing future therapeutic strategies.
