Machine learning applications related to suicide in military and Veterans: A scoping literature review

Yuhan Zhang, Yishu Wei, Yanshan Wang, Yunyu Xiao, Ronald K. Poropatich, Gretchen L. Haas, Yiye Zhang, Chunhua Weng, Jinze Liu, Lisa A. Brenner, James M. Bjork, Yifan Peng

TL;DR

This paper reviews the current landscape of machine learning (ML) applications for predicting suicide risk in military personnel and veterans. It highlights ML's potential to identify complex risk factors and improve early detection, but also points to significant gaps in data sources, methodological rigor, and clinical integration.

What Problem Does This Paper Address?

Suicide is a critical public health issue, particularly within military and veteran communities, with rates often exceeding those in the general population. Early detection and prediction of suicidal ideation, attempts, and mortality are crucial for effective prevention. While traditional statistical methods have been used, the complexity and sheer volume of data available in military health systems present an opportunity for machine learning (ML) to offer more nuanced and accurate predictions. However, the application of ML in this domain is fragmented, and a comprehensive overview of existing research, identified risk factors, and future directions is needed.

Methods

This study employed a scoping literature review following the PRISMA protocol to systematically identify and summarize research on ML applications for suicide risk assessment and prediction in military and veteran populations. A keyword search was conducted across multiple databases (PubMed, IEEE, ACM, Google Scholar). Inclusion criteria focused on peer-reviewed English articles using ML techniques to predict suicide-related outcomes in active military personnel or veterans. Thirty-two articles met the criteria. The review involved descriptive analysis of study characteristics, data sources, ML techniques, and identified risk factors.

Datasets

The paper does not introduce new datasets but rather analyzes datasets used in the reviewed studies. These datasets primarily include:

  • Surveys and Questionnaires: Collecting data on suicidal ideation, substance misuse, PTSD symptoms, well-being, and life events.
  • Electronic Health Records (EHRs): Including data from the Veterans Health Administration (VHA) Corporate Data Warehouse (CDW).
  • Army Study to Assess Risk and Resilience in Service-members (Army STARRS): Utilizing administrative data from the Historical Administrative Data System (HADS).
  • Social Media Data: One study used social media posts.
  • Audio Recordings: One study used audio recordings of conversations between military couples.

The sample sizes varied widely, from 124 to over 1.2 million participants.

Key Insights and Ideas

  • ML's Superiority in Data Handling: ML models can process vast datasets and identify complex, non-linear relationships that traditional methods might miss, leading to improved prediction accuracy for suicide-related outcomes.
  • Diverse Risk Factors Identified: ML has confirmed and identified a broad spectrum of risk factors, including psychological (depression, anxiety, PTSD), biological, and sociocultural elements, underscoring the multifaceted nature of suicide risk.
  • Sex-Specific Differences Emerge: While many risk factors are common, ML analyses have highlighted distinct patterns for males and females, with alcohol misuse and sexual abuse being more significant for females, and traumatic brain injury (TBI) more prominent for males.
  • Potential for Real-Time Monitoring: ML's ability to adapt risk scores in real-time, especially with continuous data streams from wearable technology, offers a significant advantage for timely intervention during critical periods.
  • Clinical Integration is Lacking: A major gap exists in connecting ML findings to clinical rationales and practical implementation, hindering the translation of research into actionable prevention strategies.

Key Results

  • Predictive Accuracy: While specific metrics varied, ML models generally demonstrated reasonable predictive accuracy, with AUC values often around 0.80-0.85.
  • Identified Risk Factors: Key predictors across studies included mental health issues (depression, anxiety, PTSD), history of suicidal ideation/attempts, physical health problems, substance abuse, and demographic factors.
  • Data Modalities: EHRs and surveys were the most common data sources, but emerging modalities like social media and audio recordings show promise.
  • Intervention Strategies: Studies suggest a dual approach: brief, broad interventions for lower-risk populations (requiring high recall) and intensive, targeted interventions for high-risk groups (requiring high precision).

Limitations and Future Work

The authors identify several limitations and suggest future research directions:

  • Limited Data Sources: Reliance on surveys and EHRs can be expanded by incorporating diverse data types like social media content and wearable technology for real-time physiological and behavioral data.
  • Methodological Enhancements: Current ML methodologies often stem from the pre-deep learning era. Future work should explore advanced deep learning techniques, transformers, and longitudinal modeling for better causal inference.
  • Performance Metrics: Many studies overlooked crucial metrics like positive predictive value (PPV) and negative predictive value (NPV), which are vital for policy decisions. Survival-specific metrics like c-index were also underutilized.
  • Clinical Rationales: A significant gap exists in linking ML findings to clinical rationales, hindering practical application.
  • Diversity and Cultural Representation: Most studies focus on U.S. samples, leading to underrepresentation of international military populations and racial minorities. Cross-cultural comparisons are needed.
  • Specific Subgroups: More detailed analyses are required for specific subgroups, such as individuals with active suicidal ideation or those with PTSD, and to ensure sufficient sample sizes for female service members.
  • Ethical Considerations: Bias in ML models, particularly concerning racial groups, and the need for transparency through data and code sharing for external validation are critical ethical concerns.

Summary updated

Abstract

Suicide remains one of the main preventable causes of death among active service members and veterans. Early detection and prediction are crucial in suicide prevention. Machine learning techniques have yielded promising results in this area recently. This study aims to assess and summarize current research and provides a comprehensive review regarding the application of machine learning techniques in assessing and predicting suicidal ideation, attempts, and mortality among members of military and veteran populations. A keyword search using PubMed, IEEE, ACM, and Google Scholar was conducted, and the PRISMA protocol was adopted for relevant study selection. Thirty-two articles met the inclusion criteria. These studies consistently identified risk factors relevant to mental health issues such as depression, post-traumatic stress disorder (PTSD), suicidal ideation, prior attempts, physical health problems, and demographic characteristics. Machine learning models applied in this area have demonstrated reasonable predictive accuracy. However, additional research gaps still exist. First, many studies have overlooked metrics that distinguish between false positives and negatives, such as positive predictive value and negative predictive value, which are crucial in the context of suicide prevention policies. Second, more dedicated approaches to handling survival and longitudinal data should be explored. Lastly, most studies focused on machine learning methods, with limited discussion of their connection to clinical rationales. In summary, machine learning analyses have identified a wide range of risk factors associated with suicide in military populations. The diversity and complexity of these factors also demonstrates that effective prevention strategies must be comprehensive and flexible.
BibTeX
@article{zhang2025machine,
  author = {Yuhan Zhang and Yishu Wei and Yanshan Wang and Yunyu Xiao and Ronald K. Poropatich and Gretchen L. Haas and Yiye Zhang and Chunhua Weng and Jinze Liu and Lisa A. Brenner and James M. Bjork and Yifan Peng},
  title = {{Machine learning applications related to suicide in military and Veterans: A scoping literature review}},
  journal = {J. Biomed. Informatics},
  year = {2025},
  url = {https://openreview.net/forum?id=SWBp0FLi8i},
}
Figure