Background:

The incidence of pulmonary embolism (PE) has been on the rise annually. Despite receiving regular sequential anticoagulation therapy, some patients with non-high-risk acute PE (APE) continue to experience residual pulmonary vascular obstruction (RPVO). This study sought to identify the risk factors for RPVO following 3 months of sequential anticoagulation therapy for non-high-risk PE. Machine learning techniques were utilized to construct a clinical prediction model for predicting the occurrence of RPVO.

Methods:

A total of 254 acute non-high-risk PE patients were included in this study, all of whom were admitted to the Third People's Hospital of Yunnan Province between 2020 and 2023. After 3 months of regular anticoagulant treatment, computed tomography pulmonary angiography (CTPA) were reviewed to identify the presence of RPVO. Patients were then categorized into either the thrombolysis group or the thrombosis residue group. Throughout the study period, 49 patients were excluded due to missing data, irregular treatment, or loss to follow-up. Clinical symptoms, physical signs, and laboratory results of 205 PE patients were recorded. Correlation and collinearity analyses were conducted on relevant risk factors, and significance tests were performed. Heat maps illustrating the relationships between influencing factors were generated. Predictors were selected using least absolute shrinkage and selection operator (LASSO) regression, followed by multivariate logistic regression analysis to create a predictive model. Internal validation of the model was also carried out.

Results:

By searching the literature to understand all the clinical indicators that may affect the efficacy of anticoagulation therapy. A total of 205 patients with non-high-risk acute pulmonary thromboembolism were evaluated for various risk factors. Five independent factors were identified by multivariable analysis-age, chronic obstructive pulmonary disease (COPD), acratia, pulmonary systolic blood pressure (PASP), and major arterial embolism-and their P value, odds ratio (OR) and confidence interval (CI) were as follows (P=0.012, OR =1.123; 95% CI 1.026-1.23), (P=0.002, OR =13.30; 95% CI 2.673-66.188), (P=0.001, OR =14.009; 95% CI 2.782-70.547), (P=0.003, OR =1.061; 95% CI 1.020-1.103) and (P<0.001, OR =18.128; 95% CI 3.853-85.293), which may indicate a poor prognosis after standard anticoagulant therapy. A nomogram was constructed using these variables and internally validated. The receiver operating characteristic (ROC) curves of the model demonstrated strong predictive accuracy, with an area under the curve (AUC) of 0.94 (95% CI 0.89-0.96) for the training set and 0.93 (95% CI 0.88-0.95) for the validation set. Calibration curves were utilized to assess the practicality of the nomogram.

Conclusions:

A novel predictive model was developed based on a single-center retrospective study to identify patients with RPVO following anticoagulant therapy for acute non-high-risk PE. This model may aid in the early detection of patients, prompt adjustment of treatment, and ultimately lead to a decrease in adverse outcomes.