Study Reveals Massive Underreporting of Self-Harm in Veterans
A groundbreaking study led by researchers at The University of New Mexico School of Medicine has uncovered a significant gap in how the Veterans Health Administration (VHA) tracks self-harm among veterans. Using a novel machine learning method, the team analyzed electronic health records of over 1.3 million patients and found that standard diagnosis codes capture only about one-fourth of clinically documented self-harm history. This discovery has profound implications for mental health service planning, clinical awareness, and research accuracy.
The study, published in the Journal of Medical Internet Research, highlights a common problem in health systems: important mental health history is often present in medical records but buried in unstructured text, making it difficult to find. Diagnosis codes, which clinicians and researchers rely on to search for and count conditions, frequently miss these critical details. As a result, the true prevalence of self-harm among veterans may be vastly underestimated.
Methodology: Machine Learning to the Rescue
The research team employed a novel machine learning method previously developed by members of the group. This approach was designed to sift through vast amounts of unstructured clinical notes and identify mentions of self-harm that are not captured by standard diagnostic codes. After expert chart review and statistical calibration, the researchers estimated that documented self-harm was present in about 7.9% of the veterans seen by VHA clinicians—more than four times the 1.85% visible through diagnosis codes alone.
This gap is not just a statistical curiosity; it has real-world consequences. Missed history can affect clinical awareness, skew research findings, and lead to inadequate planning for mental health services. For instance, problem lists—notations that providers compile to flag important conditions for clinical teams—also showed a visibility gap. Among veterans with a diagnosis code for self-harm, only 22.6% had self-harm or a history of self-harm listed on their VHA problem list. This means that even when self-harm is documented in codes, it is often not consistently maintained in problem lists, further obscuring the patient's history.
Implications for Mental Health Care
Dr. Christophe Lambert, Ph.D., professor and interim chief of the Division of Translational Informatics in the UNM School of Medicine's Department of Internal Medicine and the study's corresponding author, emphasized the importance of accurate measurement. "For research and planning, if we only count what is easy to see in diagnosis codes, we may substantially underestimate the need for mental health services," he said. "Better measurement can help health systems plan better, help researchers study care more accurately, and eventually help clinicians know when a patient may need a closer look."
The findings suggest that health systems like the VHA could benefit from integrating machine learning tools into their data analysis pipelines. By automatically scanning clinical notes for mentions of self-harm, these tools could provide a more complete picture of patients' mental health histories. This, in turn, could lead to better risk assessment, more targeted interventions, and improved outcomes for veterans struggling with mental health issues.
Challenges and Future Directions
While the machine learning method shows great promise, the researchers caution that it is not a silver bullet. The algorithm requires careful calibration and expert review to ensure accuracy. Moreover, the study focused on the VHA, which serves a unique population; the findings may not be directly generalizable to other health systems. However, the underlying issue—that diagnosis codes often miss important clinical information—is widespread, suggesting that similar approaches could be valuable in other contexts.
Future research should explore how to integrate such machine learning tools into routine clinical workflows without overburdening providers. Additionally, efforts to improve the consistency and completeness of problem lists could help bridge the gap. Ultimately, the goal is to ensure that every veteran's mental health history is accurately captured and available to guide their care.
Conclusion
This study underscores the power of machine learning to uncover hidden insights in electronic health records. By revealing that self-harm history is vastly undercounted in veterans, it calls for a rethinking of how health systems track and respond to mental health needs. As Dr. Lambert noted, better measurement is the first step toward better care. With tools like this novel machine learning method, health systems can move closer to that goal, ensuring that no veteran's history of self-harm goes unnoticed.
This article is based on reporting by Medical Xpress. Read the original article.
Originally published on medicalxpress.com
