Early prediction of immune checkpoint inhibitor-related pneumonitis in advanced non-small cell lung cancer based on primary tumor delta-radiomics features.

Objective: To investigate the effectiveness of predicting immune checkpoint inhibitor-related pneumonitis (ICIP) in patients with advanced non-small cell lung cancer (NSCLC) using Delta radiomics features derived from pre- and post-treatment enhanced CT images. Methods: This single-center retrospective study extracted radiomics features of primary tumors from baseline enhanced CT images and from enhanced CT images obtained after the first to third treatment cycles in patients with stage IIIB-IV NSCLC receiving immune checkpoint inhibitors (ICIs). Differences between features were calculated as Delta features. Feature selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO) regression. Prediction models were developed using Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Extreme Gradient Boosting (XGBoost) algorithms. These models were compared and further integrated with a clinical feature-based model incorporating a history of interstitial lung disease, absolute lymphocyte count, and neutrophil/lymphocyte ratio. Model performance was assessed using five-fold cross-validation. Results: A total of 131 patients were included, among whom 46 (35.1%) developed ICIP, including 8 patients (17.4%) with grade 3-5 ICIP. From 2153 initial features, 22 key Delta radiomics features were selected for model construction. The Delta radiomics model based on the LR algorithm showed the best performance in both the training and validation sets, with AUCs of 0.92 (95% CI: 0.88-0.97) and 0.85 (95% CI: 0.78-0.92), respectively. After integration with the clinical model, the performance of the combined model was further improved in the training set, achieving an AUC of 0.94 (95% CI: 0.90-0.98), while the validation set AUC was 0.86 (95% CI: 0.79-0.93). Although the difference in AUC between the combined model and the LR model in the validation set was not statistically significant (DeLong test, P = 0.4691). Calibration curves and decision curve analysis indicated good calibration and favorable clinical utility. Conclusion: This preliminary model offers a potential imaging-based biomarker for early risk stratification of any-grade ICIP in patients with advanced NSCLC. Performance specifically for high-grade (grade 3-5) ICIP could not be evaluated due to the limited number of such events. External validation in independent cohorts is required before clinical application.