Click here to

Session: Personalized Treatment via Outcome Modeling [Return to Session]

Outcome Prediction for Metastatic Lung Cancer Patients After Radiotherapy and Immunotherapy Using Patient, Tumor, Treatment, and Immunologic Factors

Y KIM1*, I Chamseddine2, C Grassberger2, Y Cho3, W Sung1, (1) Department of Biomedical Engineering and Department of Biomedicine & Health Science, College of Medicine, The Catholic University of Korea, Seoul, South Korea (2) Department of Radiation Oncology, Massachusetts General Hospital/Harvard Medical School, United States (3) Department of Radiation Oncology, Yonsei Cancer Center, Yonsei University College of Medicine, Seoul, Republic of Korea


TH-D-BRC-6 (Thursday, 7/14/2022) 11:00 AM - 12:00 PM [Eastern Time (GMT-4)]

Ballroom C

Purpose: This study aims to predict overall survival (OS) and progression free survival (PFS) in metastatic lung cancer patients treated with immune checkpoint inhibitors (ICI) and radiation therapy (RT) and to explore the importance of RT-related prognostic factors and irradiated tumor burden.

Methods: We used data from stage 4 lung cancer patients (n=79) treated with ICI and preceding RT between 2015 and 2018. The model included 14 patient-, tumor-, and RT-related features as well as total lymphocyte counts (TLC) at baseline and before ICI. To quantify the scope of radiotherapy, we define irradiated tumor burden (ITB) as the irradiated volume fraction of total visible tumor. Log-rank tests for OS and PFS were performed to test correlation to ITB. We developed a Cox proportional hazards model (Cox) and—to capture feature interactions—a random survival forest (RSF), and quantified feature interactions using Friedman’s H-statistics. We repeated the nested cross-validation 100 times to mitigate possible model sensitivity to the training data, and evaluated final models using Harrell’s concordance index (c-index).

Results: Patients with ITB>50% had significantly better prognosis than <50% (median OS: 15.7 vs 9.1 months, p=0.0012; median PFS: 13.5 vs 7.5 months, p<0.001). ITB was significant for predicting OS, and the RSF showed significantly better performance than the Cox (RSF c-index=0.76 90%CI [0.75-0.77]; Cox c-index=0.71 90%CI [0.70-0.72]; t-test p<0.001), due to strong interactions among predictors. PFS prediction was similar, c-index of 0.74 for RSF, though the dose per fraction became one of the most important features. In addition, TLC before immunotherapy, metastasis extent, and planning target volume were the most important for both endpoints.

Conclusion: ITB and lymphocyte counts have important roles in predicting outcome, and could be used to improve patient selection in this setting. RSF outperforms Cox due to strong interactions between clinical and treatment variables.

Funding Support, Disclosures, and Conflict of Interest: This work was supported by grants from the National Research Foundation of Korea (NRF, No. 2021R1C1C1005930) funded by the Korea government (MSIT).


Radiation Therapy, Lung, Modeling


TH- Response Assessment: Modeling: Machine Learning

Contact Email