AI system helps identify suicide risk during routine medical visits
Vanderbilt researchers demonstrate effectiveness of automated alerts in prompting suicide risk screening, with interruptive alerts proving significantly more successful than passive notifications.
A groundbreaking study from Vanderbilt University Medical Center reveals that artificial intelligence-driven clinical alerts can effectively help doctors identify patients at risk for suicide during routine medical visits. The research, published in JAMA Network Open on 3 January 2025, demonstrates how automated screening prompts can significantly increase suicide risk assessment rates in clinical settings.
The study evaluated the Vanderbilt Suicide Attempt and Ideation Likelihood (VSAIL) model, which analyses electronic health records to calculate a patient’s 30-day suicide attempt risk. The system was tested in three neurology clinics, comparing two different alert approaches: interruptive pop-up alerts versus passive information display in patient charts.
The research team, led by Dr Colin Walsh, found that interruptive alerts were substantially more effective, resulting in suicide risk assessments in 42% of cases compared to just 4% with the passive system. The study involved 7,732 patient visits over six months, with the AI system flagging approximately 8% of visits for screening.
“Most people who die by suicide have seen a health care provider in the year before their death, often for reasons unrelated to mental health,” Walsh noted in the press release. “But universal screening isn’t practical in every setting. We developed VSAIL to help identify high-risk patients and prompt focused screening conversations.”
Clinical impact and implementation challenges
In their paper, the authors highlight the significance of their findings for clinical practice: “This RCT builds on understanding of effectiveness of forms of CDS to drive clinical decisions. Interruptive CDS tends to be more effective in prompting behaviour. This finding has been shown in diverse settings including medication management for heart failure, contact isolation practices in the emergency department, and Patient Health Questionnaire-9 administration in primary care clinics.”
Balance between effectiveness and alert fatigue
The researchers note that while clinicians generally prefer passive, non-interruptive clinical decision support, they acknowledge these might not be seen or used as often as interruptive prompts. This presents an important balance between effectiveness and user preference.
While the interruptive alerts proved more effective, the researchers acknowledged potential concerns about “alert fatigue” – when doctors become overwhelmed by frequent automated notifications. The authors suggest that future research should examine this aspect carefully to find the optimal balance between alert effectiveness and clinical workflow disruption.
Future implications and research directions
The authors emphasize the broader implications of their work: “As a single-centre RCT, this study has implications for trial design using artificial intelligence-driven tools, preventive workflows in ambulatory settings, and CDS. Suicide remains a stigmatized area with risk management concerns, but this study provides evidence how trial designs might rigorously test novel approaches without sacrificing equipoise.”
The researchers conclude that larger-scale trials comparing this type of clinical decision support system with standard care are warranted to measure effectiveness in reducing suicidal self-harm. They emphasise the need for iterative improvement of the system design and testing in diverse clinical settings.
The study represents a significant step forward in using artificial intelligence to support suicide prevention in healthcare settings, particularly in identifying high-risk patients who might otherwise go unnoticed during routine medical visits.
Reference:
Walsh, C. G., Ripperger, M. A., Novak, L., et. al. (January 3, 2025). Risk Model–Guided Clinical Decision Support for Suicide Screening: A Randomized Clinical Trial. JAMA Network Open, 8(1), e2452371. doi: https://doi.org/10.1001/jamanetworkopen.2024.52371
