A study from Vanderbilt University Medical Center reveals that artificial intelligence (AI) alerts can significantly improve the identification of patients at risk of suicide, enhancing prevention efforts in routine medical settings.
Led by Colin Walsh, an associate professor of Biomedical Informatics, Medicine, and Psychiatry, the research tested the Vanderbilt Suicide Attempt and Ideation Likelihood (VSAIL) system in three neurology clinics at the center. The study, published in JAMA Network Open, compared two approaches: interruptive alerts that disrupt the physician's workflow and a passive system that displays risk information in the patient's electronic health record.
Results showed that interruptive alerts were far more effective, prompting doctors to assess suicide risk in 42% of cases compared to just 4% with the passive system. Walsh noted, “Most individuals who die by suicide had contact with a healthcare provider in the past year, often for non-specific health reasons.”
Suicide rates in the U.S. have risen over the past generation, with 14.2 deaths per 100,000 people annually, making it the 11th leading cause of death. Research indicates that 77% of those who die by suicide consulted a primary care provider in the previous year.
The VSAIL model analyzes current data from electronic health records to estimate the risk of suicide attempts within the next 30 days. In previous tests, this model, which flagged patients without sending alerts, showed promising results: one in 23 marked patients reported suicidal thoughts later.
During the current study, when patients flagged as high-risk by VSAIL had appointments, their doctors received either intrusive or passive alerts at random. The focus on neurology clinics was due to the association of certain neurological conditions with higher suicide risk.
Walsh mentioned that the automated system flagged only about 8% of all patient visits for screening, indicating that targeted approaches can facilitate suicide prevention even in busy clinics. The study analyzed 7,732 medical visits over six months, generating 596 screening alerts. No episodes of suicidal thoughts or attempts were recorded in the following month among patients in both alert groups.
While intrusive alerts increased screening frequency, researchers highlighted the risk of alert fatigue, where doctors may become overwhelmed by excessive notifications. Future studies will explore this phenomenon further.
“Healthcare systems must balance the efficiency of intrusive alerts with potential negative effects,” Walsh stated. “Our findings suggest that automated risk identification, along with well-designed alerts, can help identify more patients in need of suicide prevention services.”