3676 - Peritumoral Vascular Morphology Associates with T Stage and Predicts Immunotherapy Response in Stage IV NSCLC

Purpose/Objective(s):

PD-1/PD-L1 inhibitors have transformed stage IV NSCLC therapy, but responses are heterogeneous and predictive biomarkers are limited. Tumor vascular architecture may reflect tumor invasiveness and the immune microenvironment. This study aimed to evaluate whether peritumoral vascular morphology is associated with T stage and predicts first-line immunotherapy outcomes.

Materials/Methods:

We retrospectively analyzed 89 IV NSCLC patients receiving first-line anti–PD-1/PD-L1 therapy with pre-treatment contrast-enhanced CT. Tumor and vessel segmentation, skeletonization, and feature extraction generated 91 vascular features (61 morphological, 30 spatial), including vessel density, tortuosity, orientation, and histogram heterogeneity. Global associations between clinical variables (TNM stage, PG-SGA score, .etc) and vascular features were evaluated using Spearman correlation, ANOVA, Kruskal–Wallis, or Chi-square tests with FDR <0.05. Significant clinical variables were further examined via univariate screening and multivariate machine learning models (Random Forest for feature importance and Lasso regression for multicollinearity control) to create a consolidated feature set. PCA reduced dimensionality for survival analysis. Cox proportional hazards models and Kaplan–Meier curves were applied to evaluate progression-free survival (PFS) and overall survival (OS), with median or optimized thresholds used for patient stratification.

Results:

In global association analysis, T stage was the only clinical variable significantly associated with vascular features (FDR <0.05). Vessel-to-tumor volume ratio showed the strongest negative correlation (? = –0.452, FDR = 0.021). Tortuosity metrics, feeding vessel density, and histogram heterogeneity increased with advancing T stage, indicating progressive vascular remodeling and loss of structural regularity. PCA reduced the 91 features to 23 representative vascular features for survival analysis. Multivariable Cox models identified protective factors: number of immunotherapy sessions (OS & PFS, HR = 0.32, p <0.001) and histogram_bin8 (OS, HR = 0.56, p = 0.023); risk factors: PG-SGA score, XY_Kurtosis, and Rt_Kurtosis (PFS, HR >1, p <0.05). Kaplan–Meier curves using optimized cut-offs confirmed significant stratification of PFS and OS by these vascular features, demonstrating predictive value independent of clinical variables. Notably, features capturing tortuosity and spatial orientation of vessels showed the strongest associations with treatment response, suggesting a functional link between vascular architecture and immunotherapy efficacy.

Conclusion:

Peritumoral vascular morphology systematically remodels with T stage and independently predicts outcomes of first-line immunotherapy in stage IV NSCLC. Quantitative vascular features derived from routine CT provide reproducible, biologically grounded biomarkers for patient stratification and precision immunotherapy.