Objective Despite the widespread use of immune checkpoint inhibitors (ICIs) improving survival outcomes in non-small cell lung cancer (NSCLC) patients, immune-related adverse events (irAEs) triggered by ICIs have become a major challenge in clinical practice. This study aims to establish an interpretable machine learning model to predict ICI treatment of irAE risk in advanced NSCLC patients, thereby supporting clinical decision-making and thus improving the safety of immunotherapy. Methods A total of 550 patients were enrolled in the study. The development cohort consisted of 420 patients treated with ICIs from January 2019 to October 2023 and was randomly divided into a training set (n = 295) and a test set (n = 125). In addition, a temporally distinct cohort of 130 patients treated from November 2023 to November 2024 served as the validation set. Nine machine-learning algorithms were trained and evaluated in parallel, and the optimal model was selected based on discrimination, calibration, and clinical utility. Results Among the 550 patients, 361 (65.6%) developed irAEs. Six essential features were chosen including neutrophil count (Neut), lymphocyte count (Lymph), platelet count (Plt), hemoglobin (Hb), Eastern Cooperative Oncology Group Performance Status (ECOG PS), and history of diabetes. Although the neural network (NN) model performed slightly better in the test set, the logistic regression (LR) model offered superior interpretability, and its clinical net benefit was similar to that of the NN model. Therefore, the LR model was ultimately selected as the optimal predictive model (AUC = 0.855 in the test set; 0.801 in the validation set). Conclusion The LR model enables the early identification of patients at high risk of developing irAEs during hospitalization and supports their timely adoption of individualized management measures. The web tool developed based on this model is available at https://lingchun.shinyapps.io/web123/ .
Cao et al. (Fri,) studied this question.