Using Pathomics-Based Model for Predicting Positive Surgical Margins in Patients with Esophageal Squamous Cell Carcinoma: A Comparative Study of Decision Tree and Nomogram

Ze Tang,&ast; Shiyun Feng,&ast; Qing Liu, Yunze Ban, Yan Zhang Department of Thoracic Surgery, The First Hospital of Jilin University, Changchun, 130021, People’s Republic of China&ast;These authors contributed equally to this workCorrespondence: Yan Zhang, Email [email protected]: Esophageal squamous cell carcinoma (ESCC) has a high incidence and mortality rate. Postoperative positive surgical margins (PSM) often correlate with poor prognosis. This study aims to develop and validate a predictive model for PSM positivity in ESCC patients, with the potential to guide preoperative planning and improve patient outcomes.Methods: We conducted a retrospective analysis of 1776 patients who underwent esophageal cancer surgery at the First Affiliated Hospital of Jilin University between January 2015 and December 2023. Patients with visible residual tumors (R2) or microscopic residual tumors (R1) at the surgical margins were classified as having PSM. High-dimensional pathological features were extracted from digital pathological sections using CellProfiler software. The selected features were used to develop a predictive model based on decision trees and generalized linear regression, and the model was validated in an independent cohort. Clinically significant pathological factors (P < 0.05) were included in multivariate logistic regression for further validation. The model’s performance was assessed using calibration curves and receiver operating characteristic (ROC) curves, generated with the Bootstrap method. Decision curve analysis (DCA) was employed to evaluate the clinical utility of the predictive model.Results: A total of 229 patients (12.89%) were diagnosed with PSM. Logistic regression analysis identified multifocal lesions, vascular invasion, and pathomics-based features as independent predictors of PSM. The predictive model, represented by a decision tree, demonstrated good discrimination with an area under the ROC curve of 0.899 (95% CI: 0.842– 0.956, P < 0.001), and a strong calibration curve between the predicted probability and the actual probability. Additionally, the nomogram demonstrated slightly inferior discrimination with an area under the ROC curve of 0.803 (95% CI: 0.734– 0.872, P < 0.001) in the training cohort.Conclusion: Our study successfully established and validated a pathology-based predictive model for PSM risk, which could enhance preoperative evaluation and inform treatment strategies for ESCC.Keywords: esophageal squamous cell carcinoma, surgical margin, machine learning, pathomics, prediction model