Integrated Liquid Biopsy Model for Predicting Metastasis and Guiding PD-1 Therapy in Esophageal Squamous Cell Carcinoma

Hailong Pang¹Shenshen Hao²Shuiwang Hu³Xiaojuan Zhou^4*

• ¹General Hospital of Pingmei Shenma Medical Group, Pingdingshan, China
• ²Qingdao University Qingdao Medical College, Qingdao, China
• ³Southern Medical University, Guangzhou, China
• ⁴Ganzhou Cancer Hospital, Ganzhou city, China

Objective: This study endeavors to develop and validate an integrated biomarker signature (IBS) grounded in serum carbohydrate antigen 72-4 (CA72-4), vascular endothelial growth factor-C (VEGF-C), and the pepsinogen I/II ratio (PGI/PGII). A practical IBS model will be constructed to substantially enhance the accuracy of lymph node metastasis (LNM) risk stratification following surgery for esophageal squamous cell carcinoma (ESCC). This model is anticipated to refine prognostic assessments for patients and identify novel research avenues within the field, thereby providing guidance for future investigations. Methods: A prospective three-cohort design was adopted, encompassing a training cohort of 220 patients, a temporal validation cohort of 138 patients, and a regional external validation cohort of 94 patients. This design was selected for its robustness in ensuring the validity and reliability of the findings. A total of 452 patients with esophageal squamous cell carcinoma (ESCC) who underwent R0 resection were enrolled between March 2022 and June 2024. The predictive model was constructed using the XGBoost algorithm combined with Shapley Additive exPlanations (SHAP) for interpretability. Model performance was evaluated via receiver operating characteristic (ROC) curves, decision curve analysis, and net reclassification index. Underlying biological mechanisms were explored using the 10x Visium spatial transcriptomics platform. Results: The IBS model demonstrated superior discriminative ability in the training cohort (n=220; AUC=0.936, 95% CI: 0.908-0.964) compared to the AJCC 9th edition staging system (ΔAUC=0.154, P<0.001). Performance was maintained in validation cohorts. Spatial transcriptomics revealed distinct biological correlates: In the CA72-4 high group (>12.4 U/mL), the density of Podoplanin cells at the invasion front increased 2.1-fold (P = 0.003). VEGF-C and CXCL12 exhibited significant spatial co-expression (Spearman r=0.73, P<0.001), suggesting a role in lymphatic angiotactic migration. In the PGI/PGII low ratio group (≤4.5), regulatory T cell (Treg) enrichment areas expanded 1.8-fold (Cibersortx, P=0.008). High-risk patients (IBS > 0.5) receiving PD-1 inhibitor plus chemotherapy achieved a pathological complete response (pCR) rate of 28.6%, representing a 115% increase over conventional therapy (P=0.009). Conclusion: The validated IBS model provides high-precision prediction of postoperative LNM risk in ESCC. It offers a novel framework ("tumor antigen burden - lymphangiogenesis - immune microenvironment") for improving patient risk stratification.