Clinical alerts driven by AI can help doctors identify patients at risk for suicide, potentially improving prevention efforts in routine medical settings, a new study has shown.
The study investigated whether an AI system could effectively prompt doctors in three neurology clinics at Vanderbilt University Medical Center (VUMC) to screen patients for suicide risk during regular clinic visits.
It compared two approaches — automatic pop-up alerts that interrupted the doctor’s workflow versus a more passive system that simply displayed risk information in the patient’s electronic chart.
Results revealed that the interruptive alerts were far more effective, leading doctors to conduct suicide risk assessments in connection with 42 per cent of screening alerts, compared to just 4 per cent with the passive system.
The team, led by Colin Walsh, associate professor of Biomedical Informatics, Medicine and Psychiatry, tested their AI system called the Vanderbilt Suicide Attempt and Ideation Likelihood model (VSAIL).
Most people who die by suicide have seen a healthcare provider in the year before their death, often for reasons unrelated to mental health,” Walsh said.
“But universal screening isn’t practical in every setting. We developed VSAIL to help identify high-risk patients and prompt focused screening conversations.”
The VSAIL model, which Walsh’s team developed at Vanderbilt, analyses routine information from electronic health records to calculate a patient’s 30-day risk of suicide attempt. In earlier prospective testing, where VUMC patient records were flagged but no alerts were fired, the model proved effective at identifying high-risk patients, with one in 23 individuals flagged by the system later reporting suicidal thoughts.
In the new study, when patients identified as high-risk by VSAIL came for appointments at Vanderbilt’s neurology clinics, their doctors received on a randomised basis either the interruptive or non-interruptive alerts. The research focused on neurology clinics because certain neurological conditions are associated with increased suicide risk.
The researchers suggested that similar systems could be tested in other medical settings.
“The automated system flagged only about 8 per cent of all patient visits for screening,” Walsh said.
“This selective approach makes it more feasible for busy clinics to implement suicide prevention efforts.”
The study involved 7,732 patient visits over six months, prompting 596 total screening alerts.
During the 30-day follow-up period, in a review of VUMC health records, no patients in either randomised alert group were found to have experienced episodes of suicidal ideation or attempted suicide.
While the interruptive alerts were more effective at prompting screenings, they could potentially contribute to “alert fatigue” — when doctors become overwhelmed by frequent automated notifications. The researchers noted that future studies should examine this concern.
“Healthcare systems need to balance the effectiveness of interruptive alerts against their potential downsides,” Walsh said.
But these results suggest that automated risk detection combined with well-designed alerts could help us identify more patients who need suicide prevention services.
