Abstract
<title>Abstract</title> <p>Background Vertebral compression fractures (VCFs) are clinically significant skeletal complications that may occur after radiotherapy in patients with esophageal squamous cell carcinoma (ESCC). This study evaluated the predictive performance of computed tomography (CT) radiomics for newly developed post-radiotherapy thoracic VCFs and determined whether radiomics can provide additional value for individualized risk stratification. Methods This retrospective two-center study included 551 patients with pathologically confirmed ESCC who underwent radiotherapy. Patients from institution 1 were randomly assigned to a training cohort (n = 321) or an internal validation cohort (n = 138), whereas patients from institution 2 were included in an external validation cohort (n = 92). Vertebral CT attenuation and radiomics features were extracted from pretreatment CT images of irradiated thoracic vertebrae. Candidate predictors were identified using univariable Cox regression, followed by Akaike information criterion–based Cox model selection in the training cohort. Six prediction models were developed: vertebral CT attenuation, radiomics, radiotherapy, clinical–radiotherapy, clinical–radiotherapy–CT attenuation, and radiomics-integrated models. Model performance was evaluated using the concordance index (C-index), time-dependent area under the receiver operating characteristic curve (time-dependent AUC), calibration curves, decision curve analysis, net reclassification improvement, and integrated discrimination improvement. Results During a median follow-up of 0.92 years (interquartile range, 0.58–1.50 years), 77 of the included patients developed post-radiotherapy thoracic VCFs. Age, alcohol consumption, total radiotherapy dose, vertebral CT attenuation, and Radscore were retained in the final radiomics-integrated model. The radiomics-integrated model demonstrated the best overall performance, with C-index values of 0.878, 0.825, and 0.759 and 24-month AUCs of 0.895, 0.724, and 0.802 in the training and internal and external validation cohorts, respectively. Calibration and decision curve analyses demonstrated good calibration and clinical utility of the model. Conclusion CT radiomics demonstrated favorable predictive performance for post-radiotherapy VCFs in patients with ESCC and provided additional value for individualized risk stratification. The radiomics-integrated model may aid in identifying patients who require closer imaging surveillance and preventive management.</p>