Diagnosis model for assessing chronic thromboembolic pulmonary hypertension in high-altitude pulmonary embolism patients: a machine learning approach.
Patients with pulmonary embolism (PE) at high altitude face an increased risk of developing chronic thromboembolic pulmonary hypertension (CTEPH). This study aims to establish a diagnosis model of CTEPH patients at high altitude to optimize early screening.
A retrospective cohort of CTEPH and PE patients was rigorously selected through inclusion/exclusion criteria. Clinical data encompassing biochemical profiles, echocardiography, and CT angiography (CTA) were collected, yielding 103 candidate variables. Feature parameters were screened using the Boruta algorithm, followed by predictive model development with seven machine learning architectures. The optimal model was identified based on area under the curve (AUC). The optimal Random Forest model was subsequently interpreted through Shapley Additive Explanations (SHAP) to quantify feature contributions.
Among 57 PE patients, 44% met echocardiographic criteria for pulmonary hypertension following PE. Diameter of right atrium, diameter of right ventricle, Vessel-Grade (of embolization) and Sup-inferior (superior or inferior of embolization) were key identified predictors. Random Forests model had the highest AUC of 0.842. Enlarged right heart, embolization of small vessels and superior pulmonary artery embolism increased the risk of CTEPH, while normal right heart structure and isolated inferior pulmonary embolism reduced it.
The Random Forests model demonstrated potential for detecting CTEPH in PE patients, enabling early and rapid pulmonary hypertension assessment.
A retrospective cohort of CTEPH and PE patients was rigorously selected through inclusion/exclusion criteria. Clinical data encompassing biochemical profiles, echocardiography, and CT angiography (CTA) were collected, yielding 103 candidate variables. Feature parameters were screened using the Boruta algorithm, followed by predictive model development with seven machine learning architectures. The optimal model was identified based on area under the curve (AUC). The optimal Random Forest model was subsequently interpreted through Shapley Additive Explanations (SHAP) to quantify feature contributions.
Among 57 PE patients, 44% met echocardiographic criteria for pulmonary hypertension following PE. Diameter of right atrium, diameter of right ventricle, Vessel-Grade (of embolization) and Sup-inferior (superior or inferior of embolization) were key identified predictors. Random Forests model had the highest AUC of 0.842. Enlarged right heart, embolization of small vessels and superior pulmonary artery embolism increased the risk of CTEPH, while normal right heart structure and isolated inferior pulmonary embolism reduced it.
The Random Forests model demonstrated potential for detecting CTEPH in PE patients, enabling early and rapid pulmonary hypertension assessment.
Authors
Fan Fan, Ma Ma, Zhang Zhang, Yang Yang, Zhakeer Zhakeer, Huang Huang, Yu Yu, Zeng Zeng, Mi Mi
View on Pubmed