Machine learning prediction of postoperative pulmonary embolism: a multicenter external validation study highlighting inflammatory response and intraoperative hemodynamics - Report - MDSpire
Advertisement
Machine learning prediction of postoperative pulmonary embolism: a multicenter external validation study highlighting inflammatory response and intraoperative hemodynamics
Clinical Report: Utilizing Machine Learning to Predict Postoperative PE
Overview
This study developed and validated a machine learning model to predict postoperative pulmonary embolism (PE) across multiple surgical centers. The model demonstrated strong predictive performance, particularly in identifying key clinical determinants associated with PE risk.
Background
Postoperative pulmonary embolism is a rare but serious complication that can lead to significant morbidity and mortality. Accurate prediction and early identification of high-risk patients are crucial for improving surgical outcomes and patient safety. This study addresses the challenge of heterogeneous perioperative risk factors by utilizing machine learning techniques to enhance risk stratification.
Data Highlights
Metric
Value
Patients with PE
48 (1.38%)
AUC (External Validation)
0.925 (95% CI 0.877–0.972)
Key Findings
The final model included predictors such as age, BMI, malignancy history, and intraoperative tachycardia.
XGBoost provided the best performance with strong discrimination and calibration.
SHAP analysis identified key factors influencing PE risk, including surgery duration and CRP levels.
The model achieved an AUC of 0.925 in the external validation cohort.
1.38% of patients developed postoperative PE in the study population.
Clinical Implications
The machine learning model can aid clinicians in identifying patients at high risk for postoperative PE, allowing for targeted interventions and enhanced perioperative management. Understanding the key predictors can inform clinical decision-making and improve patient outcomes.
Conclusion
The study successfully developed a robust machine learning model for predicting postoperative PE, highlighting the importance of specific clinical factors in risk assessment. This model has the potential to improve perioperative care and patient safety.
