Seoul St. Mary's Hospital team develops AI model to reduce blood-draw burden in pediatric surgery

The Catholic University of Korea Seoul St. Mary's Hospital announced Tuesday that a research team led by Kim Hyun-ho, a professor of pediatrics at St. Nicholas Children's Hospital, has developed an AI model that estimates arterial blood carbon dioxide partial pressure (PaCO₂) in pediatric patients using intraoperative vital-sign data.

The study's co-first authors are Park Ju-hyun, a resident in anesthesiology and pain medicine at Asan Medical Center, and Cho Chae-eun, a student at Korea University School of Medicine.

As South Korea's low birth rate persists and advanced-age pregnancies become more common, the number of high-risk pediatric patients requiring precise vital-sign monitoring during surgery and anesthesia is also rising. The AI model was developed to reduce unnecessary invasive procedures in pediatric patients and improve their postoperative outcomes.

Arterial blood CO₂ partial pressure is a key index for evaluating ventilation status and calibrating mechanical ventilator settings during surgery. It is also essential for detecting respiratory abnormalities and acid-base imbalances that can arise under anesthesia in real time.

The most accurate measurement method involves inserting an arterial catheter into the radial or femoral artery and analyzing the drawn blood with a blood gas analyzer. In children, however, narrow blood vessels make catheter insertion significantly more likely to cause complications — such as vascular injury or impaired blood flow — than in adults.

Repeated blood draws also risk causing anemia in pediatric patients who are already losing blood during surgery, leaving clinicians struggling to strike the right balance between intraoperative safety and postoperative recovery.

End-tidal carbon dioxide (EtCO₂), a non-invasive alternative, has been used to address this challenge, but ventilation-perfusion mismatches and other factors can cause readings to diverge from actual arterial blood CO₂ partial pressure.

To overcome these limitations, the research team developed a model that combines EtCO₂ readings with clinical data and AI to estimate arterial blood CO₂ partial pressure more accurately.

The team retrospectively analyzed 8,853 paired EtCO₂ and PaCO₂ measurements from 3,586 pediatric patients registered in VitalDB, an open surgical vital-sign database built by researchers in the department of anesthesiology and pain medicine at Seoul National University Hospital.

Ten key variables were ultimately selected: EtCO₂, body temperature, minute ventilation, age, fraction of inspired oxygen (FiO₂), presence of cardiac disease, lung compliance, preoperative hemoglobin, body mass index and mean arterial pressure. Training a machine-learning algorithm on these variables confirmed that the model could estimate intraoperative arterial blood CO₂ partial pressure with high accuracy.

The study is the first to estimate PaCO₂ in pediatric patients using a range of intraoperative vital-sign data. The model's reproducibility was also confirmed through external validation using recent patient data from two other university hospitals.

"While the model developed in this study cannot fully replace arterial blood gas analysis, it can serve as a useful supplementary tool in situations where arterial catheter insertion is difficult or frequent blood gas testing is not feasible," Kim said. "We particularly expect it to have clinical utility in patients who require precise CO₂ management, such as those undergoing pediatric neurosurgical procedures."

The research was recently published in Anesthesiology (impact factor 9.4), a leading international journal in the field of anesthesia and pain medicine.