Machine learning–based construction of a clinical prediction model for hypercapnia during one-lung ventilation for lung surgery

Abstract

Abstract In this study, we developed a clinical prediction model for hypercapnia during one-lung ventilation for lung surgery by machine learning. We analyzed the cases and intraoperative blood gases of 348 patients who had undergone lung surgery at Jiangxi Cancer Hospital from November 2019 to June 2021. We analyzed the factors that independently influence hypercapnia during one-lung ventilation for lung surgery by selecting the best variables through a combination of random forest and logistic regression stepwise selection (Step AIC). Thereafter, we used these factors to construct logistic regression models and a nomogram. Receiver operating characteristic curves were used to measure the predictive accuracy of the nomogram and its component variables, and the predictive probabilities of the nomogram were compared and calibrated by calibration curves. We used bootstrap to verify the internal validation method to judge the reliability of the model, and we employed decision curve analysis (DCA) for clinical decision analysis. The independent influencing factors for hypercapnia during one-lung ventilation for lung surgery were age, gender, and one-lung ventilation position. We established the hypercapnia during one-lung ventilation for lung surgery logistic regression model: −5.421 + 0.047 × age + 1.8 × gender (=1) + 0.625 × one-lung ventilation position (=1). The prediction accuracy probability of the nomogram is 0.7457 (95% confidence interval [0.6916, 0.7998]). The prediction model showed good agreement between the calibration curve and the ideal predicted value, and bootstrap internal validation showed the area under the curve was 0.745 and the C-index was 0.742. DCA indicated that the model has some clinical value. In this study, three independent influences on hypercapnia during one-lung ventilation were established. We constructed an individualized model for predicting hypercapnia during one-lung ventilation for pulmonary surgery, as well as the first internally validated predictive model and nomogram for hypercapnia during one-lung ventilation for pulmonary surgery, both of which have good predictive and calibration properties and can provide some clinical guidance value.