Development of a multi-feature predictive model for risk stratification in stage IB-IIA non-small cell lung cancer: a multicenter analysis

Abstract

Objective

This study aimed to develop a comprehensive risk stratification model for stage IB-IIA non-small cell lung cancer (NSCLC) by integrating clinicopathological data with pre-treatment CT imaging.

Methods

This retrospective study included three independent cohorts of patients with stage IB-IIA NSCLC for model development and validation (Training: n = 370; Internal validation: n = 120; External validation: n = 70). Disease-free survival (DFS) was the primary endpoint. Radiomics features were extracted from both tumoral and peritumoral regions of CT images to construct a radiomics model. A ResNet50-based deep learning architecture was adopted to develop a deep learning model using CT imaging data. Logistic regression was used to identify significant clinicopathological factors. These components were integrated into a multi-feature combined model (CRD model) that utilized clinicopathological, radiomics, and deep learning features for DFS prediction. Model interpretability was assessed using the SHapley Additive exPlanations (SHAP) method.

Results

The combined CRD model demonstrated superior performance in predicting DFS, achieving areas under the curve (AUC) of 0.865, 0.798, and 0.803 in the training, internal validation, and external validation cohorts, respectively. Patients were stratified into high- and low-risk groups using the CRD model, and in the external validation cohort, the hazard ratio (HR) for high-risk patients was 17.509, with a C-index of 0.73. SHAP analysis revealed that radiomics features contributed most significantly to the performance of the CRD model.

Conclusions

The multi-feature combined model effectively predicts DFS and identifies high-risk patients with stage IB-IIA NSCLC. It could facilitate personalized postoperative treatment strategies, improving patient outcomes.