Development and validation of an explainable machine learning model for risk stratification in patients with clinically suspected acute pulmonary embolism: retrospective study.

BACKGROUND

Acute Pulmonary Embolism (APE) is a life-threatening cardiovascular emergency. Computed Tomography Pulmonary Angiography (CTPA) serves as the diagnostic gold standard but is associated with risks of overuse and radiation exposure.

OBJECTIVE

To develop and validate an explainable machine learning model for predicting the likelihood of APE in clinically suspected patients, aiming to optimize CTPA decision-making.

METHODS

This study retrospectively enrolled 649 patients who underwent CTPA for suspected APE based on clinical symptoms, signs, or elevated D-dimer levels. Seventy-seven clinical variables were collected. Least Absolute Shrinkage and Selection Operator (LASSO) regression was used for initial feature screening, followed by multivariate logistic regression to identify independent predictors. Eight machine learning algorithms were compared to select the optimal model. The Shapley Additive exPlanations (SHAP) framework was applied for model interpretability analysis.

RESULTS

The model demonstrated stable performance in both training and validation sets (training set AUC: 0.865±0.003; validation set AUC: 0.843±0.038). In the independent test set, the logistic regression model achieved the best performance (AUC = 0.793). Thirteen key predictors were identified, including age, lactate, fibrin degradation products, and others. SHAP analysis visually illustrated the contribution of each feature to risk prediction.

CONCLUSIONS

The developed APE risk prediction model, based on logistic regression and SHAP interpretability, exhibits good discriminative ability and transparency. It shows potential for individualized risk stratification and optimization of CTPA utilization in clinical decision-making.