Preoperative Multivariable Model for Risk Stratification of Hypoxemia During One-Lung Ventilation

Abstract

BACKGROUND: Hypoxemia occurs with relative frequency during one-lung ventilation (OLV) despite advances in airway management. Lung perfusion scans are thought to be one of the most accurate methods to predict hypoxemia during OLV, but their complexity and costs are well-known limitations. There is a lack of preoperative stratification models to estimate the risk of intraoperative hypoxemia among patients undergoing thoracic surgery. Our primary objective was to develop a risk stratification model for hypoxemia during OLV based on preoperative clinical variables. METHODS: This is a single-center, retrospective cohort study including 3228 patients who underwent lung resections with OLV from 2017 to 2022, at a tertiary academic health care center in the United States. Vital signs and ventilator settings were retrieved minute by minute. Intraoperative hypoxemia was defined as an episode of oxygen desaturation (Spo 2 <90%) for at least 5 minutes. Demographic and clinical characteristics were included in a stepwise logistic regression, which was used for the selection of predictors of the risk score model. All patients included in this cohort underwent elective lung surgery in lateral decubitus position, with double lumen tube and placement confirmation with fiberoptic bronchoscopy. Our model was validated internally using area under the receiver operating curves (AUC) with bootstrapping correction. RESULTS: The incidence of hypoxemia during OLV was 8.9% (95% confidence interval [CI], 8.0–10.0). Multivariable logistic regression identified 9 risk factors with their corresponding scoring: preoperative Spo 2 <92% (15 points), hemoglobin <10 g/dL (6 points), age >60 years old (4 points), male sex (4 points), body mass index >30 kg/m2 (8 points), diabetes mellitus (4 points), congestive heart failure (7 points), hypertension (3 points), and right-sided surgery (3 points). The AUC of the model after bootstrap correction was 0.708 (95% CI, 0.676–0.74). Based on the highest Youden index, the optimal score for predicting intraoperative hypoxemia was 13. The risk of hypoxemia increased from 4.7% in the first quartile of scores (0–13 points), to 32% in the third quartile (27–39 points), and 83.3% in the fourth quartile (>39 points). At scores of 20 or greater, the specificity of the model exceeded 90% and reached a positive predictive value of 80%. CONCLUSIONS: The risk of hypoxemia during OLV can be stratified preoperatively using accessible clinical variables. Our risk model is well calibrated but showed moderate discrimination for predicting intraoperative hypoxemia. The accuracy of preoperative models for risk stratification of hypoxemia during OLV should be explored in prospective studies.