Machine learning-enhanced preoperative assessment of spontaneous ventilation risk in diverse video-assisted thoracoscopic surgeries.

Objectives: Despite its widespread use, spontaneous ventilation can lead to intraoperative hypoxemia or hypercapnia. In this retrospective study, we developed and validated a preoperative clinical data-based model to predict intraoperative hypoxemia or hypercapnia risk during spontaneous ventilation video-assisted thoracoscopic surgery. Methods: We enrolled patients who underwent spontaneous ventilation video-assisted thoracoscopic surgery at the National Respiratory Medicine Center of First Affiliated Hospital of Guangzhou Medical University between 2011 and 2023. Four machine learning algorithms-light gradient boosting machine, category boosting, extreme gradient boosting, and random forest-were trained and combined into a weighted ensemble to predict intraoperative hypoxemia (peripheral oxygen saturation <94%) or hypercapnia (partial pressure of end-tidal carbon dioxide >80 mm Hg). Results: Among 6590 included patients, 12.6% and 3.8% developed intraoperative hypoxemia and severe hypercapnia, respectively. The final risk assessment model achieved an area under the receiver operating characteristic curve of 0.72 (95% CI, 0.68-0.76) and an area under the precision-recall curve of 0.35 (95% CI, 0.29-0.42). The statistical analysis of the data and the risk interpretation from the ensemble machine learning algorithm consistently identified body mass index (23.6 [21.6-25.5] vs 22.2 [20.4-24.0] kg/m Conclusions: The predictive model based on preoperative clinical characteristics effectively assessed intraoperative hypoxemia and severe hypercapnia risk in diverse spontaneous ventilation video-assisted thoracoscopic surgery procedures. This model can systematically guide preoperative evaluation to select appropriate patients and improve the intraoperative safety and surgical outcomes of spontaneous ventilation video-assisted thoracoscopic surgery.