Early detection and neoadjuvant efficacy prediction for esophageal cancer using cfDNA methylation-based liquid-biopsy assay.

Abstract

4070 Background: Esophageal cancer (EC) is a major malignancy of the upper gastrointestinal tract globally. Early detection and timely therapeutic intervention are pivotal in improving patient outcomes. However, current diagnostic methods often fail to detect EC at an early stage and lack the ability to predict treatment response. There is an urgent need for non-invasive, sensitive, and specific biomarkers to enhance early detection and guide personalized treatment strategies. This study aims to develop a tool to detect EC and predict the neoadjuvant efficacy. Methods: This is a prospective multicenter diagnostic study. From July 2023 to October 2024, a total of 236 esophageal cancer cases (stage I: 19.9%, stage II: 24.6%, stage III: 39.4%, stage IV: 16.1%), 31 chronic esophagitis cases, and 441 healthy controls were enrolled from multiple centers. Methylation features and fragmentomic characteristics derived from methylation sequencing data were integrated to develop a gradient-boosted tree model. A nested cross-validation framework was employed to ensure robustness and reliability. Additionally, predictive models for therapeutic responses to neoadjuvant treatment were constructed. Results: The detection model achieved an area under the curve (AUC) of 0.954（95% CI: 0.936-0.971. At a specificity of 97.9% (95% CI: 96.1%-99.0%), the overall sensitivity reached 84.7% (95% CI: 79.5%-89.1%) , with stage-specific sensitivities of 69.5% for early-stage (I/II) and 97% for advanced-stage (III/IV) disease. The detection model maintained robust performance across various clinicopathological parameters, including differentiation grade, neural invasion, vascular invasion, tumor count, and tumor location, with no significant differences in subgroup performance. Among the cohort, 44 patients underwent neoadjuvant therapy, with 90.9% (40/44) receiving immunochemotherapy. The major pathological response (MPR) rate was 52.3% (23/44) and the pathological complete response (pCR) rate was 15.9% (7/44). No clinical features were found to correlate with MPR or pCR rates. Differential methylation profiles between MPR/pCR and non-MPR/pCR patients were analyzed to construct predictive models for neoadjuvant therapy outcomes. Using logistic regression and leave-one-out cross-validation, the MPR prediction model achieved an accuracy of 86.3%, while the pCR prediction model demonstrated an accuracy of 90.9%. Conclusions: Our cfDNA-methylation based assay demonstrated high performance in early EC detection and promising value in predicting neoadjuvant therapy responses. This non-invasive approach has the potential to revolutionize EC management by enabling earlier diagnosis and personalized treatment strategies.