Integrating radiomics and real-world data to predict immune checkpoint inhibitor efficacy in advanced non-small-cell lung cancer

Background: Immunotherapy (IO) revolutionized the prognosis of patients with non-small-cell lung cancer (NSCLC). However, identifying optimal candidates for this treatment remains challenging. Based on previous studies suggesting the potential power of radiomics in predicting clinical outcomes in different clinical settings, we aimed to assess its capability in predicting IO efficacy in advanced NSCLC patients. Materials and methods: A total of 375 advanced NSCLC patients treated with IO-based regimens from April 2013 to May 2022 were enrolled. Primary lung lesions were segmented and radiomic features extracted. Using clinical benefit rate and overall survival status at 6 and 24 months (OS6 and OS24) as endpoints, machine learning classifiers were trained and then evaluated on a test set. Results: Model achieving the highest performance predicting long-term survival (OS24) reached an accuracy of 0.71 and area under the curve of 0.79 on the test set, using the combination of radiomic features and real-world data (RWD) as input. Combining radiomics with RWD consistently allowed to outperform predictions obtained using the current standard predictive biomarker, i.e. programmed death-ligand 1 expression, for most of the outcomes. Conclusions: We explored a radiomics-based signature with potential utility in predicting the prognosis of NSCLC patients undergoing IO. Further validation is required to confirm its clinical applicability and to support oncologists in making prognostic assessments.