The goal of this research is to develop accurate and scalable machine learning risk algorithms to detect risk of suicide attempt and death from electronic health records and a matching clinical decision support tool optimized for use by military primary care. The study will evaluate the effectiveness of the tool to detect risk of suicide attempts and death among active duty military personnel.

Over the past several decades, despite an increase in scientific efforts to better understand and prevent suicide, the suicide rate in the United States (U.S.) has remained largely unchanged. Remarkably, some segments of the U.S. population, such as military service members, have actually demonstrated an increase in the rate of suicide deaths. These data suggest that traditional methods of identifying service members at risk for suicide, such as through clinician-delivered suicide risk assessments via question-and-answer format, may not be sufficient to yield a measurable decrease in the suicide rate. 

Researchers from Florida State University believe they have a solution to optimize suicide risk detection: the use of complex statistical algorithms, such as machine learning. Machine learning, a form of artificial intelligence, is widely used in non-mental health settings. Email platforms, for instance, utilize machine learning principles to sort email into “spam” or “important”. In this scenario, an array of variables—for example, the sender, content of the subject line, word choice, organization of words, etcetera—are utilized by machine learning algorithms to make decisions. A similar approach can be applied to electronic medical records within large healthcare systems, such as the U.S. military, to optimize the detection of which patients are at increased risk for suicide. 

With funding from the Department of Defense (DoD), through the Military Suicide Research Consortium (MSRC), Assistant Professor Jessica Ribeiro, Ph.D. is leading a team of researchers to develop and evaluate a machine learning-based clinical decision support tool to accurately detect suicide risk within military primary care settings. Ribeiro and her team plan to conduct a three-stage study. To begin, Ribeiro and her colleagues will adapt machine learning algorithms utilized in civilian samples for use in military healthcare settings. Next, Ribeiro and her team will work closely with military stakeholders to identify barriers and facilitators to the implementation of machine learning-based suicide risk detection. As part of this stage, the researchers will develop a clinical decision support tool tailored to military primary care. The clinical decision support tool will provide clinicians with actionable steps to take in the event the machine learning algorithm flags their patient as at elevated risk for suicide. In the last stage of the study, utilizing a randomized clinical trial design, Ribeiro and her team will evaluate the effectiveness of the clinical decision support tool in military primary care practice. 

Military primary care settings represent a particularly important venue in which to develop and test a machine learning algorithm and clinical decision support tool. For one, healthcare is readily available in the U.S. military with no financial barriers to service members. Further, a robust body of research has demonstrated that individuals at risk for suicide are more likely to present to their primary care physician than a mental health care provider, representing a major opportunity for suicide risk detection and intervention. 

If the results of this study are positive, machine learning might prove to be an effective tool in the detection—and, in turn, mitigation—of suicide risk among U.S. military service members. One advantage of machine learning, especially for the detection of suicide risk in large healthcare systems, Ribeiro says, is its potential for scalability. By contrast with traditional approaches that require considerable time and personnel effort and expertise, machine learning offers a quick, efficient, and relatively low-cost alternative that, Ribeiro says, appears to also be considerably more accurate regarding suicide risk detection. Within large health care systems, such as the U.S. military, the application of machine learning has immense potential to create system-wide standards in the identification of service members at increased risk for suicide.