AI Predicts 'Hypotension' Occurrence During Hemodialysis in Advance

Seoul National University Hospital Professor Han Seung-seok's Team and Seoul National University Graduate School of Convergence Science and Technology Professor Kwak No-jun's Team
Development of a Hypotension Prediction Model During Hemodialysis

[Asia Economy Reporter Kim Ji-hee] Professor Han Seung-seok's team from the Department of Nephrology at Seoul National University Hospital and Professor Kwak No-jun's team from the Graduate School of Convergence Science and Technology at Seoul National University announced on the 15th that they have developed a model that predicts hypotension during hemodialysis using artificial intelligence technology.

The research team analyzed 261,647 hemodialysis sessions performed on 9,292 patients at Seoul National University Hospital. They collected various data on hemodialysis patients, including basic information such as the patient's gender and age, pre-dialysis systolic and diastolic blood pressure, vascular access, and anticoagulants. Hypotension during hemodialysis was observed in approximately 27,971 of the total 260,000 cases.

A common side effect during hemodialysis is the occurrence of hypotension. It is known that one in five hemodialysis patients experiences hypotension during dialysis. Patients usually complain of nausea, vomiting, and convulsions, and in severe cases, it can lead to other problems such as cardiac ischemia. Although it occurs frequently, predicting it during dialysis is difficult, exposing many patients to risk.

The research team proposed a "real-time prediction model" that predicts hypotension within one hour during hemodialysis. The entire dataset was randomly divided for model development, validation, and testing. As a result, the prediction model showed excellent predictive ability with a score of 0.94 (closer to 1 means better predictive ability).

Previously, predicting the occurrence of hypotension was very difficult because blood pressure fluctuates frequently during dialysis and there are various causes that can induce hypotension. The factor that most influenced the predictive performance of the model designed by the research team was the real-time changing blood pressure data. By having artificial intelligence pre-learn a vast amount of blood pressure data, it detected complex blood pressure changes and improved the prediction rate.

In practice, Professor Han Seung-seok's team at the Department of Nephrology, Seoul National University Hospital, is managing the risk of hypotension during hemodialysis by introducing a "patient-customized hemodialysis process."

Professor Han stated, "The 5-year mortality rate of hemodialysis patients is as high as 40%, and hypotension during hemodialysis is most closely related to the risk of death. Accurately predicting and preventing hypotension during hemodialysis is the first step toward improving patient survival rates and quality of life."

Meanwhile, the results of this study were published in the recent issue of the "American Journal of Clinical Nephrology."

This content was produced with the assistance of AI translation services.