Artificial Intelligence Model Predicts Hypotension Risk During Surgery

Tuesday, July 04, 2023

Intraoperative hypotension can occur frequently in patients under general anesthesia, but it affects the incidence of postoperative complications and mortality.

To prevent this, it is important for anesthesiology specialists to observe changes in arterial blood pressure in patients undergoing surgery to identify the risk of hypotension and minimize hypotension. said to have developed

Asan Medical Center Anesthesiology Professor Kim Seong-hoon and Park Yong-seok, National Cancer Center International Cancer Graduate University Cancer AI Digital Health Professor Kim Jun-tae's team developed an artificial intelligence model trained on the arterial blood pressure data of 10,000 surgical patients. As a result, hypotension during surgery It was recently announced that it predicted patients who could develop this with about 91% accuracy.

In particular, the research team provided the basis for judgment on the probability of hypotension in real time, so that medical staff could easily interpret the hypotension prediction process.

Intraoperative hypotension is a situation in which the patient's mean arterial blood pressure remains below 65 mmHg for at least 1 minute, and may occur due to bleeding or drug side effects. When hypotension occurs during surgery, it is very important to minimize hypotension as it can affect the incidence and mortality of complications such as acute kidney injury and myocardial infarction.

In the meantime, artificial intelligence that predicts hypotension during surgery has been studied, but there is a limit to the lack of interpretation of the prediction process.

To overcome these limitations, the team led by Professor Seong-Hoon Kim of the Department of Anesthesiology and Pain Medicine at Asan Medical Center analyzed large-scale data from 10,000 patients, and conducted validation and evaluation of anesthesiology specialists.

The research team first trained the arterial blood pressure data of 10,454 patients who underwent surgery at Asan Medical Center from 2018 to 2021 into an artificial intelligence model to extract arterial blood pressure trends that indicate the amount of blood flowing inside the blood vessels.

After that, the probability of developing hypotension at 10 minutes was predicted by comparing similarities with typical arterial blood pressure trends for each risk of hypotension generated based on large-scale patient data. Furthermore, the odds ratio, which indicates the risk ratio of hypotension according to similarity, was presented to medical staff as a basis for prediction.

The research team demonstrated the generalizability of the artificial intelligence model through internal validation using arterial blood pressure data from 10,454 patients who underwent surgery at Asan Medical Center in Seoul and external validation using data from 3,278 patients registered on the open source datanet. As a result, the prediction accuracy was about 91% for internal verification and about 90% for external verification.

In addition, as a result of evaluating the applicability of the artificial intelligence model to 17 specialists in the department of anesthesiology and pain medicine at Asan Medical Center in Seoul, △clinical accuracy △clinical usefulness △ They scored 24%, 41%, and 26% higher on the intraoperative decision-making intention, respectively.

Seong-Hoon Kim, Professor of Anesthesiology and Pain Medicine at Seoul Asan Medical Center, said, "The results of this study are meaningful in that they can solve unmet medical needs by providing not only the probability of hypotension occurrence but also the basis for judgment to medical staff in real time." We will continue to conduct research to find out if it helps in terms of improvement.”

Joon-Tae Kim, Professor of Cancer AI and Digital Health, National Cancer Center, International Cancer Graduate School, said, “The results of this study are not focused on simple predictive performance improvement, but the fact that this is the first study to evaluate anesthesiologists working in an operating room. It has meaning,” he said.

Meanwhile, the results of this research were recently published in 'IEEE Transactions on Neural Networks and Learning Systems (citation index 14.255)', a renowned academic journal in the field of artificial intelligence. It was conducted with the support of the National Cancer Center Public Interest Red Cancer Research Project.

 

Source: www.amc.seoul.kr