Predicting Mortality in Pulmonary Embolism: A Machine Learning Approach with External Validation in COVID-19...

Area of Science:

• Cardiology
• Pulmonology
• Infectious Diseases
• Data Science

Background:

• Pulmonary embolism (PE) is a common complication of SARS-CoV-2 infection, leading to significant early mortality.
• Accurate risk stratification for PE in the emergency department (ED) is challenging, especially with concurrent COVID-19.
• The performance of existing PE prognostic tools in COVID-19 patients is not well understood.

Purpose of the Study:

• To evaluate the effectiveness of machine learning (ML) models compared to the Pulmonary Embolism Severity Index (PESI) for predicting in-hospital mortality in acute PE patients.
• To assess the performance of these models in patients with and without COVID-19.

Main Methods:

• Retrospective, single-centre study of 538 acute PE patients admitted through the ED.
• Utilized univariate analysis and machine learning models (XGBoost, Random Forest, SVM) to assess mortality risk.
• External validation on COVID-19 patients was performed.

Main Results:

• Machine learning models (XGBoost, RF) demonstrated superior discrimination compared to PESI in non-COVID PE patients (AUC 0.864, 0.834 vs. 0.725).
• Performance of ML models and PESI significantly decreased in COVID-19 patients (XGBoost AUC 0.635, RF 0.614, PESI 0.584).
• Sepsis, PESI class V, higher neutrophil count, platelet count, and NT-proBNP were associated with mortality in univariate analysis.

Conclusions:

• Machine learning models using routinely available ED variables improve in-hospital mortality prediction for non-COVID PE compared to PESI.
• The generalizability of current ML models is limited in COVID-19 patients, necessitating COVID-specific refinements.
• Prospective, multicenter validation is required to confirm these findings and develop more accurate prognostic tools for PE in the context of COVID-19.