As the number of high-risk pediatric patients rises alongside more births to older mothers and multiple pregnancies, an artificial intelligence (AI) model has been developed that can monitor arterial carbon dioxide partial pressure (PaCO₂) during surgery using biosignal data.
Seoul St. Mary's Hospital of the Catholic University of Korea announced on the 21st that a research team consisting of Professor Kim Hyun-ho of the Department of Pediatrics, Park Ju-hyun, a resident in the Department of Anesthesiology and Pain Medicine at Seoul Asan Medical Center, and Cho Chae-eun, a medical student at Korea University College of Medicine, developed an AI-based non-invasive surgical monitoring solution to reduce unnecessary invasive procedures in pediatric patients and improve postoperative outcomes. The findings were published in a recent issue of the international journal Anesthesiology.
With advances in medical technology improving the survival rate of premature infants born before 32 weeks of gestation, the number of high-risk pediatric patients requiring surgery and anesthesia for congenital heart disease, digestive disorders, and neurosurgical conditions is increasing. Accordingly, the importance of precisely monitoring biosignals for safe anesthesia and surgery is growing further. In particular, arterial carbon dioxide partial pressure is considered a key indicator for identifying respiratory abnormalities and acid-base imbalances that can occur during anesthesia in real time, and for evaluating appropriate mechanical ventilation settings and ventilation status during surgery.
The method of inserting an arterial catheter into the radial or femoral artery to draw blood and analyze it with a blood gas analyzer offers the highest accuracy. However, because children have thin blood vessels, the risk of complications such as vascular damage or blood flow disorders during catheter insertion is significantly higher. Above all, there was a considerable concern that repeated blood sampling in pediatric patients who are already experiencing bleeding from surgery could cause anemia. To compensate for this, end-tidal carbon dioxide (EtCO₂), which can be measured non-invasively, is sometimes used, but it differed from the actual arterial carbon dioxide partial pressure.
To overcome these limitations, the research team developed a model that combines pediatric patients' end-tidal carbon dioxide and collected clinical information with AI to estimate arterial carbon dioxide partial pressure more accurately. After obtaining 8,853 pairs of end-tidal carbon dioxide and arterial carbon dioxide partial pressure data from 3,586 pediatric patients registered in an open operating-room biosignal database built by researchers in the Department of Anesthesiology and Pain Medicine at Seoul National University Hospital, the team applied SHAP (Shapley Additive Explanations) analysis to select 10 key variables. These were end-tidal carbon dioxide, body temperature, minute ventilation, age, fraction of inspired oxygen (FiO₂), presence of heart disease, lung compliance, preoperative hemoglobin, body mass index, and mean arterial pressure. When these were trained on a machine learning algorithm, the model recorded a mean absolute error (MAE) of 2.73 mmHg, about 23% lower than the conventional method using end-tidal carbon dioxide alone (average error of 3.56 mmHg), showing meaningful performance improvement.
The research team also confirmed the model's reproducibility in external validation using recent patient data from two institutions, Chungnam National University Hospital and Seoul National University Hospital. In the first study to estimate arterial carbon dioxide partial pressure in pediatric patients using various biosignal data obtainable during surgery, the team demonstrated high accuracy. It is expected to be particularly useful in patients requiring precise carbon dioxide management, such as those undergoing pediatric neurosurgery.
Professor Kim Hyun-ho said, "Although the model developed this time cannot completely replace arterial blood gas testing and requires further validation through prospective clinical trials, it can be a useful auxiliary tool in situations where arterial catheter insertion is difficult or blood gas testing cannot be performed frequently."
