Clinical-radiomics nomogram integrating lymph node radiomic features to predict immunotherapy response in advanced biliary tract cancers.
Immunotherapy is considered a promising treatment approach for advanced biliary tract cancers (BTCs), but only a small number of patients can respond to immunotherapy. This study aimed to develop and validate a clinical-radiomics nomogram integrating radiomic features from lymph nodes (LNs) for predicting immunotherapy efficacy in advanced BTCs.
A total of 258 patients with advanced BTCs were enrolled, comprising 206 patients in the retrospective cohort and 52 patients in the prospective cohort. Radiomic features were extracted from the LNs. The maximum relevance and minimum redundancy and least absolute shrinkage and selection operator were used to develop the radiomics signature (Rad-score). Univariate analysis and multivariate logistic regression (LR) were used to construct the clinical model. A clinical-radiomics nomogram was constructed using LR. The performance of all the models was analyzed using receiver operating characteristic curves.
Nine radiomic features were combined to construct the Rad-score. The nomogram incorporated the six clinical parameters and the Rad-score, and achieved the best discriminative ability with the areas under the curve (AUCs) of 0.899, 0.843 and 0.874 in the training, validation and testing cohorts. The clinical model showed better predictive performance than the Rad-score with the AUCs of 0.834, 0.878 and 0.740 in the training, validation and testing cohorts. The calibration curve and Brier score indicated the goodness-of-fit of the nomogram. Patients with higher nomogram scores had better overall survival (OS) and progression-free survival (PFS) in comparison to those with low scores.
The clinical-radiomics nomogram showed promising performance for predicting the response to immunotherapy in patients with advanced BTCs.
A total of 258 patients with advanced BTCs were enrolled, comprising 206 patients in the retrospective cohort and 52 patients in the prospective cohort. Radiomic features were extracted from the LNs. The maximum relevance and minimum redundancy and least absolute shrinkage and selection operator were used to develop the radiomics signature (Rad-score). Univariate analysis and multivariate logistic regression (LR) were used to construct the clinical model. A clinical-radiomics nomogram was constructed using LR. The performance of all the models was analyzed using receiver operating characteristic curves.
Nine radiomic features were combined to construct the Rad-score. The nomogram incorporated the six clinical parameters and the Rad-score, and achieved the best discriminative ability with the areas under the curve (AUCs) of 0.899, 0.843 and 0.874 in the training, validation and testing cohorts. The clinical model showed better predictive performance than the Rad-score with the AUCs of 0.834, 0.878 and 0.740 in the training, validation and testing cohorts. The calibration curve and Brier score indicated the goodness-of-fit of the nomogram. Patients with higher nomogram scores had better overall survival (OS) and progression-free survival (PFS) in comparison to those with low scores.
The clinical-radiomics nomogram showed promising performance for predicting the response to immunotherapy in patients with advanced BTCs.