<\/p>\n
An exterior view of Parkland Hospital in Dallas. An AI model is helping to predict the risk of violent patient behavior to improve the safety of hospital staff.<\/p>\n
Smiley N. Pool\/Dallas Morning News<\/p>\n
Once upon a time it would have sounded like science fiction, but North Texas scientists have developed an AI tool to help Parkland Health hospital staff identify patients at risk of becoming violent. This predictive model can help protect healthcare workers at a moment when violent incidents have become a major concern for hospitals all over the country.\u00a0<\/p>\n
In 2024, 8 out of 10<\/a> nurses reported experiencing violence at their workplace in the preceding year. This affects staff retention.\u00a0<\/p>\n Needless to say, the cost for hospitals is significant, and exposure to violence is even fueling healthcare workers\u2019 strikes in some states. A picketing nurse at a Michigan hospital told NPR this month that her colleagues were getting shoved, pushed, scratched and even bitten<\/a> by their patients.\u00a0<\/p>\n Article continues below this ad<\/p>\n Hospitals\u2019 approach to this rising violence has been mostly reactive, relying on personal alarms and video surveillance. But Parkland Health is taking a step further by using this AI tool.<\/p>\n Developed by the Parkland Center for Clinical Innovation and first introduced in 2025, the Workplace Safety Model identifies patients unlikely to pose a risk \u2014 which is most of them \u2014 allowing the hospital staff to focus on the roughly 30% flagged as higher risk.\u00a0<\/p>\n In validating the tool, the scientists found that among a group of patients identified as high risk, 77% were involved in a violent incident, PCCI staff told us.<\/p>\n Scientists are careful not to profile incoming patients, said Steve Miff, PCCI president and CEO. Race, ethnicity and religion are not considered. The AI risk assessment is a moment-in-time analysis.<\/p>\n \u201cEvery time a person comes in, [he or she] gets a new risk evaluation,\u201d Miff said. \u201cWe build this in a way that it predicts an instance, not a person.\u201d\u00a0<\/p>\n Article continues below this ad<\/p>\n Factors PCCI uses in the model include severe infection, if the patient arrived by ambulance or on their own, or what medications they are receiving, for example.<\/p>\n One key factor is whether the patient has a smoking history, since Parkland is a tobacco-free campus. If a patient can\u2019t smoke, that may lead them to be more irritable.\u00a0<\/p>\n