Development and validation of A CCTA-based risk prediction model for major adverse cardiovascular events in esophageal cancer patients receiving radiotherapy.
Authors
Affiliations (6)
Affiliations (6)
- School of Medicine, Chongqing University, Chongqing, China; Department of Radiology, Chongqing University Cancer Hospital & Chongqing Cancer Institute & Chongqing Cancer Hospital, Chongqing, China.
- Department of Radiology, Chongqing University Cancer Hospital & Chongqing Cancer Institute & Chongqing Cancer Hospital, Chongqing, China.
- Department of Radiology, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, Fujian Province, China.
- Deepwise Artificial Intelligence (AI) Lab, Deepwise Inc., Beijing 100000, China.
- Department of Radiology, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, Fujian Province, China. Electronic address: [email protected].
- Department of Radiology, Chongqing University Cancer Hospital & Chongqing Cancer Institute & Chongqing Cancer Hospital, Chongqing, China. Electronic address: [email protected].
Abstract
Major adverse cardiovascular events (MACEs) remain a significant concern in esophageal cancer (EC) patients receiving radiotherapy (RT). This study aimed to develop and validate a CCTA-based model for predicting MACEs in this population. 322 and 216 patients with EC at thoracic middle or lower segment from hospital 1 were randomly divided into the training and internal validation cohorts, while 227 patients from hospital 2 were assigned to the external validation cohort. Pericoronary adipose tissue (PCAT) radiomics features were selected by the least absolute shrinkage and selection operator Cox regression (Lasso-Cox) and Max-Relevance and Min-Redundancy (mRMR). Radiomics model was constructed and compared using seven machine-learning classifiers. A nomogram for predicting MACEs was developed with multivariable Cox regression analysis. Predictive performance of models was evaluated by C-index, and feature importance was interpreted using SHapley Additive exPlanations (SHAP) analysis. The median follow-up was 31 months (IQR, 25-36 months), during which 139 of 765 (18.2 %) patients experienced MACEs. The eXtreme gradient boosting (XGBoost) was used to construct radiomics model. A nomogram incorporating the PCAT radiomics signature, age, mean dose of left circumflex artery (LCX), and fat attenuation index of LCX achieved a moderate to strong predictive capacity across the training, internal, and external validation cohorts (C-index = 0.855, 0.839, and 0.845, respectively). SHAP analysis revealed that the PCAT radiomics signature was the most important predictor of MACEs. A nomogram combining clinical risk factors, CCTA-derived parameters, and PCAT radiomics signature can predict MACEs in patients with EC receiving radiotherapy.