AI Models Identify Patient Groups at Risk of Being Mistreated in Hospital ED

Now, a team of researchers has applied machine learning models to reveal which patient factors may influence triage outcomes—helping hospitals improve decision-making and reduce risks of mistreatment.

In a multinational collaborative study led by the University of Bergen (Bergen, Norway), researchers used artificial intelligence (AI) to assess thousands of patient triage records and identify patterns in undertriage (low-priority assignments for patients who later required intensive care) and overtriage (high-priority assignments for stable patients). Using a metric called SHAP-values, derived from game theory, the model ranked how individual variables contributed to triage outcomes.

The findings, published in the Journal of Medical Internet Research, showed that although incorrect triage was rare—affecting less than 1% of patients—the machine learning approach provided new insights that challenged assumptions. Contrary to earlier findings that emphasized age and gender, the new analysis showed that the clinical referral department and diagnostic codes were stronger predictors of triage inaccuracies in the Bergen dataset.

By comparing AI-driven insights with physician-led assumptions, the study highlights how machine learning can correct biases and reveal hidden influences on clinical decision-making. The researchers emphasize that while AI is not a flawless tool, it can offer valuable new perspectives for improving patient safety and optimizing emergency care systems.

“For optimal usage, appropriate methods must be tailored to the specific research context, and common pitfalls need to be avoided,” said Dr. Sage Wyatt, lead author and researcher at the University of Bergen. “More research is needed in the future about triage systems and new applications of machine learning methods, such as automated triage classification systems.”

Related Links:
University of Bergen