Abstract
BACKGROUND: Adequate ventilation in mechanically ventilated patients is contingent upon the monitoring of the arterial partial pressure of carbon dioxide (PaCO2) during general anesthesia. Despite its significance, continuous monitoring remains challenging due to the imprecision of noninvasive estimations and the invasive nature of traditional methods such as arterial blood gas analysis. OBJECTIVE: This study aimed to develop a machine learning model to continuously estimate PaCO2 in mechanically ventilated patients, with the goal of potentially improving intraoperative monitoring accuracy under general anesthesia. METHODS: This retrospective study used the VitalDB dataset from Seoul National University Hospital, comprising records of 6388 noncardiac surgery patients between August 2016 and June 2017. After applying inclusion and exclusion criteria, data from 2304 surgical cases (4651 PaCO2 measurement event points) were analyzed. The CatBoost regressor model was trained to predict PaCO2 using noninvasive physiological parameters and clinical information. The model's performance was evaluated using nested cross-validation across hypocapnic (<35 mm Hg), normocapnic (35-45 mm Hg), and hypercapnic (>45 mm Hg) subgroups and compared to conventional estimation methods based on end-tidal carbon dioxide (ETCO2). RESULTS: The developed model demonstrated superior overall performance compared to traditional estimations. It achieved a mean absolute error of 2.38 mm Hg and an average intraclass correlation coefficient of 0.87. Furthermore, 90.02% of the model's estimations fell within the clinically highly acceptable range (error<±5 mm Hg) while only 1.20% of errors exceeded ±10 mm Hg. Performance improvements were observed across all PaCO2 subgroups. CONCLUSIONS: The developed model provides more accurate and reliable estimates of PaCO2 than traditional ETCO2-based methods. This approach shows potential for facilitating real-time monitoring and timely clinical interventions. This study demonstrated the potential of artificial intelligence to enhance continuous monitoring of PaCO2; however, further validation, including prospective studies assessing clinical impact, is necessary.