Predicting Immunotherapy Response in Stage III-IV Non-Small Cell Lung Cancer Using Integrated Radiomics and Clinical Features

Na Geng¹Zhijun Li²Yuxiang Li³Langlang Deng¹Haitao Ma^1*

• ¹The Fourth Affiliated Hospital of Soochow University, Suzhou, China
• ²The Second Affiliated Hospital of Bengbu Medical College, Bengbu, China
• ³Changzhou First People's Hospital, Changzhou, China

To develop a combined predictive model based on CT radiomics and clinical features and evaluate its diagnostic value for predicting the efficacy prognosis of immunotherapy in stage III-IV non-small cell lung cancer (NSCLC).A retrospective analysis was conducted on 106 patients with stage IIIa-IVB NSCLC who underwent immunotherapy at the Second Affiliated Hospital of Soochow University between December 2018 and December 2023 . Patients were divided into two groups based on whether their progression-free survival (PFS) exceeded 12 months.The cohort was randomly split into a training set (75 patients) and a validation set (31 patients) in a 7:3 ratio. Additionally, 34 patients enrolled from another center were collected as an externalValidtion test cohort.Clinical and imaging data were collected, and independent predictive factors were identified through univariate and multivariate logistic regression analysis to construct a clinical feature model. Radiomic features were extracted from contrast-enhanced chest CT images, and LASSO algorithm along with Pearson correlation coefficients were applied to select optimal features and calculate a radiomics score. A combined predictive model integrating clinical independent predictors and radiomic features was developed and visualized as a nomogram.Model performance was assessed by subject work characteristics (ROC) curves and area under the curve (AUC). Clinical utility was assessed via decision curve analysis (DCA), and calibration curves were used to evaluate the nomogram's predictive accuracy.Tumor location was an independent predictor of immunotherapy efficacy and formed the clinical model. Twelve contrast-enhanced CT radiomic features comprised the radiomics model. The combined model (clinical + radiomic) demonstrated superior diagnostic performance: Internal training set AUCs (clinical: 0.705, radiomics: 0.835, combined: 0.896); In t e r n a l validation set AUCs (clinical: 0.691, radiomics: 0.833, combined: 0.863);External validation set AUCs(clinical: 0.653, radiomics: 0.831, combined: 0.884).The combined model's AUC was significantly higher than either submodel alone in both sets.DCA confirmed its highest net clinical benefit, and calibration curves indicated good accuracy.This study developed a predictive model based on clinical and radiomic features for assessing immunotherapy efficacy in NSCLC. The model demonstrated excellent performance, suggesting its potential as a clinical decisionsupport tool for prognosis prediction and treatment planning in NSCLC immunotherapy.