Radiomic and clinical predictors of epidermal growth factor receptor mutation in stage IA non-small cell lung cancer

Background Epidermal growth factor receptor (EGFR) mutation status plays a critical role in guiding targeted therapy for non-small cell lung cancer (NSCLC). However, molecular testing in patients with stage IA NSCLC may be limited by insufficient tissue availability, procedural invasiveness, and resource constraints. Therefore, developing a non-invasive approach for EGFR mutation prediction is of substantial clinical interest. This study aimed to develop a computed tomography (CT) radiomics based model integrating clinical variables for non-invasive prediction of EGFR mutation status in stage IA NSCLC patients. Methods A total of 375 patients with stage IA NSCLC who underwent pre-treatment chest CT and EGFR mutation testing were retrospectively enrolled. Tumor volumes of interest (VOIs) were manually segmented on CT images, and radiomic features were extracted using the pyradiomics package. Clinical and radiomic features were selected through a feature selection pipeline, and multiple machine learning algorithms were evaluated for EGFR mutation prediction. Model performance was assessed using the Area Under Curve (AUC). Results Predictive performance varied across feature selection strategies and machine learning algorithms. Among all evaluated combinations, the Linear Regression (LR) model built using the Least Absolute Shrinkage and Selection Operator (LASSO)-30 feature set achieved the best performance, with a test-set AUC of 0.745. In this model, CT radiomic features served as the primary predictive component, while selected clinical variables provided complementary information and modestly improved predictive performance. These findings support the value of integrating radiomic and clinical features for non-invasive EGFR mutation prediction in early-stage NSCLC. Conclusions A CT radiomics based model demonstrated only moderate performance for the non-invasive prediction of EGFR mutation status in patients with stage IA NSCLC. When clinical variables were incorporated, predictive performance improved, suggesting that clinical features provide complementary information beyond radiomics alone. The combined model highlights the added value of integrating CT-derived radiomics with clinical data for more accurate individualized molecular assessment, particularly when tissue-based genotyping is unavailable or limited.