South Korea’s AI model predicts emergency visits for diabetes patients, aiming to reduce hospitalisations

South Korea’s National Institute of Health has developed an AI system that forecasts emergency room visits among type 2 diabetes patients, potentially enabling earlier intervention and reducing unnecessary hospitalisations.

South Korea’s National Institute of Health has developed an artificial intelligence model that can predict which people with type 2 diabetes are more likely to end up in the emergency room, in a bid to catch serious complications earlier and reduce avoidable hospital visits. According to the Ministry of Health and Welfare’s Korea Disease Control and Prevention Agency, the system was trained on real-world electronic medical records from 220,720 patients treated at five medical institutions between 2008 and 2022.

The institute said 22.6% of the patients in the study, or 49,770 people, visited an emergency department at least once within a year before or after their diabetes diagnosis. Those who did so were generally older and had worse blood sugar control and kidney function than those who did not. They were also about twice as likely to be using insulin or diuretics, and had higher rates of hypertension and cerebrovascular disease.

Researchers tested models using 55 routine clinical variables, including blood pressure, blood test results, kidney function and prescription history. The best performer, a CatBoost model, achieved an accuracy rate of 87%, outperforming a traditional statistical model that scored 73%. A separate study summary indexed by the Korea Citation Index also shows the growing use of machine learning to forecast emergency-room visits in younger patients, underlining the wider interest in prediction tools based on routine medical data.

The National Institute of Health said the model could help doctors identify high-risk patients earlier and strengthen prevention and management in outpatient care. The broader research, published using data from five institutions, suggests such machine learning systems may help reduce emergency-room use linked to acute diabetes complications if they are integrated into everyday clinical practice.