Characterizing Acute Pulmonary Embolism After Off-Pump Coronary Artery Bypass Surgery Using a Predictive XGBoost...

This study aims to characterize the occurrence of acute pulmonary embolism (APE) after off-pump coronary artery bypass grafting surgery (OPCABG) and develop a predictive model for APE to mitigate adverse events and improve patient prognosis.

We reviewed the clinical records of 15,259 patients who underwent OPCABG at Tianjin Chest Hospital from May 2015 to May 2025. APE was identified in 228 patients (1.49%), and 458 non-APE controls were included after matching gender and age at an approximate 1:2 ratio. We conducted statistical analyses to characterize the clinical features of APE patients and used LASSO and XGBoost algorithms to build a predictive model for APE. SHAP method evaluated influential prognostic features.

After OPCABG, 49.1% of patients with APE experienced hypoxia and 33.3% had shortness of breath, while 28.5% were asymptomatic. APE was most frequently located in the right superior lobar pulmonary artery (58.8%). Bilateral pulmonary embolism occurred in 29.8% of patients, with 6.14% exhibiting bilateral pulmo-aortic involvement (central PE). APE patients demonstrated a higher prevalence of post-operative elevated D-dimer, deep vein thrombosis, pleural effusion, anemia, and hypoalbuminemia. Using LASSO regression, 36 pre- and post-operative variables were selected for model development. The XGBoost model achieved 91.3% sensitivity (95% CI: 82.5%-98.2%) and 94.6% specificity (95% CI: 89.1%-98.9%) in validation dataset, yielding a superior AUC of 97.6%. The SHAP analysis highlighted the contribution of postoperative D-dimer, deep vein thrombosis, the start time of low molecular weight heparin, and preoperative uric acid. While the model demonstrated satisfactory performance in our single-center validation cohort, further external validation in a multicenter setting is essential prior to its broader clinical implementation.

Despite adequate antiplatelet and anticoagulant therapy in clinical practice, APE occurred in 1-2% of patients. We developed a robust predictive model for APE following OPCABG to identify at-risk patients, optimize outcomes, and ultimately reduce the associated economic burden.