A predictive model for treatment efficacy in RAS wild-type advanced colorectal cancer: development and external validation for EGFR inhibitor plus anti-angiogenic therapy based on a retrospective cohort
摘要
Efficacy of EGFR inhibitor-anti-angiogenic combination therapy varies substantially in RAS wild-type advanced colorectal cancer (CRC), and current clinical guidelines lack individualized predictive tools to identify optimal candidates and tailor regimens. This study aimed to develop and externally validate a multi-dimensional efficacy prediction model to address this unmet clinical need. This retrospective multi-center study included 600 RAS wild-type advanced CRC patients (development cohort: 420 patients from two centers; external validation cohort: 180 patients from an independent center) treated with EGFR inhibitors (cetuximab/panitumumab) plus anti-angiogenic agents (bevacizumab/fruquintinib/regorafenib) between 2018 and 2021. Candidate variables encompassed clinical, laboratory, radiomic, and biological indices. Radiomic features were screened via ANOVA-dimensionality reduction, and LASSO-Cox regression was used for variable selection and nomogram construction (with shrinkage calibration to reduce overfitting). Model performance was evaluated by discrimination (AUC), calibration (Hosmer–Lemeshow test), and clinical utility (decision curve analysis [DCA]); the overfitting risk was assessed by calculating events per variable (EPV), and model stability was verified by multi-step internal validation (tenfold cross-validation, bootstrap resampling) and subgroup/risk stratification analyses. The final nomogram integrated five core predictors: vascular density, neutrophil-to-lymphocyte ratio (NLR), carcinoembryonic antigen (CEA), metastatic sites, and ECOG score. The model exhibited moderate discrimination with clinical practical value (development cohort AUC = 0.641, 95%CI 0.588–0.691; validation cohort AUC = 0.532, 95%CI 0.445–0.617), which is consistent with the performance of most multi-dimensional clinical prediction models for advanced colorectal cancer reported in similar studies (AUC range 0.58–0.68). Meanwhile, the model showed excellent calibration in the external validation set (H–L χ2 = 1.12, p = 0.572), indicating a high consistency between the predicted PFS probability and the actual clinical outcome. The limited discrimination of the model is mainly due to the inherent biological heterogeneity of advanced colorectal cancer and the lack of dynamic monitoring indicators (e.g., circulating tumor DNA) in the study variables. Risk stratification identified low (0.7%), intermediate (57.3%), and high-risk (42.0%) groups with significantly distinct progression-free survival (PFS) and overall survival (OS) (all log-rank p < 0.001). High-risk patients who switched regimens achieved longer median PFS (11.1 vs. 5.9 months, p < 0.001). DCA confirmed superior net benefit over “treat all/none” strategies, and the model outperformed guidelines (median PFS: 9.3 months [both recommended] vs. 6.7 months [guideline-only], p < 0.05). Key biomarkers (vascular density, tumor mutational burden) correlated with treatment response and risk stratification, providing biological rationale. This externally validated nomogram integrating five readily available clinical and laboratory indicators can realize individualized PFS prediction and risk stratification for RAS wild-type advanced CRC patients receiving EGFR inhibitor-anti-angiogenic combination therapy, and provide preliminary reference for clinical regimen adjustment of high-risk patients. As a supplementary tool to current clinical guidelines, the model can partially address the problem of clinical response heterogeneity in combination therapy and provide simple decision support for clinicians in primary and secondary hospitals with limited detection conditions. However, the model has certain limitations in long-term prognostic prediction and needs to be further optimized and validated in larger, multi-center prospective cohorts before it can be translated into clinical practice of precision oncology.