M. B. Saad1, E. Showkatian1, Q. Al-Tashi1, M. Aminu1, X. Xu1,2, M. Qayati Mohamed1, M. Salehjahromi1, S. J. Sujit1, S. H. Lin2, Z. Liao2, S. Gandhi2, D. Qian2, D. A. Jaffray1,3, C. Chung2,3, N. Vokes4, J. Zhang4, J. Heymach4, J. Wu1,3, and J. Y. Chang2; 1Department of Imaging Physics, The University of Texas MD Anderson Cancer Center, Houston, TX, 2Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, 3Institute of Data Science in Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, 4Department of Thoracic/Head and Neck Medical Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX
Purpose/Objective(s): Our recent phase 2 randomized clinical trial (I-SABR, NCT03110978) demonstrated improved event-free survival (EFS) from combining stereotactic ablative radiotherapy (SABR) with immune checkpoint inhibitor therapy for early-stage non-small cell lung cancer (NSCLC) relative to SABR alone. However, not every patient benefits from immunotherapy. We report here a secondary analysis in which clinical-radiomics, with machine learning, was developed into a model to identify patients who would or would not benefit from immunotherapy. Materials/
Methods: Subjects were 141 patients with early-stage NSCLC enrolled in the I-SABR trial, 101 in the discovery and 40 in the validation cohort. We used the discovery cohort to develop the I-SABR-SELECT framework to model treatment outcomes and inform patient selection for combining immunotherapy with SABR. We extracted radiomics patterns characterizing the tumor/peritumoral and lung regions and the angiogenesis network surrounding the tumor. Radiomics features were harmonized, qualified, and integrated with clinical factors for downstream selection to mitigate model overfitting. A random survival forest algorithm was applied to model relationships between patient characteristics and treatment outcome separately for I-SABR and SABR-only. Counterfactual reasoning was used to assess treatment effects and optimize selection. The model was evaluated separately in the discovery and validation cohorts and in an independent group of patients treated on the STARS trial of SABR for early-stage NSCLC. Results: Overall, the model recommended that 46 of the 141 (33%) patients enrolled in I-SABR switch treatments (34 of 75 [45%] in the SABR-only arm and 12 of 66 [18%] in the I-SABR arm). Patients treated according to this recommendation achieved significantly improved EFS in both arms during model discovery and validation. Stratified by this recommendation, patients who received I-SABR showed an EFS interval 1.1 to 1.6 times longer than those who did not receive immunotherapy. Notably, patients who were treated according to the I-SABR-SELECT recommendation had improved EFS (hazard ratio 22.8, p<0.001) compared with matched counterparts who did not receive the model-recommended treatment. Conversely, when the model recommended SABR-only, no difference in EFS was observed between patients given SABR-only vs those given I-SABR. In the benefit stratum by the model, the average immunotherapy effect was more than two-fold greater than in the randomized trial. Having worse performance status, a less complex angiogenesis network, and larger tumors were associated with more benefit from combining immunotherapy with SABR. Conclusion: I-SABR-SELECT provides an individualized approach for guiding who needs immunotherapy combined with SABR for early-stage NSCLC.
