A South Korean government research institute said Aug. 11 that it has developed an artificial-intelligence model capable of predicting which patients with Type 2 diabetes face a higher risk of visiting an emergency department, using real-world medical records.
The model was developed using a multicenter diabetes database established with support from the National Institute of Health, part of South Korea’s Korea Disease Control and Prevention Agency, beginning in 2022. Researchers analyzed medical records from 220,720 people with Type 2 diabetes treated at five medical institutions between 2008 and 2022.
Diabetes is typically managed through routine outpatient care. But severe hypoglycemia, dangerously high blood sugar and diabetic ketoacidosis can trigger emergency visits and raise the risk of hospitalization and death.
Among the patients studied, 49,770, or 22.6%, visited an emergency department at least once within one year before or after their diabetes diagnosis.
Those who visited an emergency department tended to be older and had poorer blood-sugar control and kidney function than patients who did not. They were also about twice as likely to be taking insulin or diuretics and had higher rates of hypertension and cerebrovascular disease.
The researchers trained and compared several AI models using 55 clinical variables routinely collected during medical care. The top-performing model was better than conventional statistical methods at identifying patients at elevated risk of an emergency-department visit.
Diastolic blood pressure was the strongest predictor, followed by serum creatinine and systolic blood pressure.
The findings are notable because the three leading predictors are generally factors that can be addressed through medication and lifestyle changes. Identifying patients at higher risk during routine care could give doctors an opportunity to intervene before an acute complication sends them to the emergency department.
The research points to a potential shift in diabetes care, from treating patients after an emergency occurs to using predictive tools to identify those at risk before complications become severe.
The National Institute of Health said it expects the model eventually could be deployed in primary-care settings, allowing doctors to identify patients at elevated risk earlier and take steps to prevent and manage complications.
The researchers cautioned that the model is intended as a tool to support clinical decision-making rather than replace physicians. Its broader use will depend on further validation and implementation in routine medical practice.
