Integration of Risk of Survival Measures Estimated From Pre- and Post-Treatment CT Scans Improves Stratification of Early Stage Non-Small Cell Lung Cancer Patients Treated with Stereotactic Body Radiation Therapy

Abstract

To predict overall survival of early-stage non-small cell lung cancer (ES-NSCLC), we developed a radiomic model to integrate risk of death estimates based on pre- and post- treatment CT scans in patients receiving stereotactic body radiation therapy (SBRT). We hypothesize this will allow us to distinguish patients with various oncologic outcomes. Two cohorts of ES-NSCLC patients treated with SBRT were studied: 100 patients with treatment planning CTs for training, and 60 patients with pre- and post-treatment CTs for testing. The patients had a median follow-up of 2.0 years. Although all patients were treated uniformly (a median dose of 50Gy in 4-5 fractions), they had different primary tumor outcomes. Each tumor was characterized with 680 radiomics features derived from both the original CT scans and wavelet decomposed data. A survival prediction model was built using random survival forests (RSF) to predict overall survival based on the training cohort with RSF parameters optimized with a 3-fold cross-validation strategy. The survival prediction model was then applied to pre- and post-treatment scans of the testing cohort. The pre-treatment risk of death measures were used to estimate the survival prediction performance measured with concordance-index (c-index), and k-means clustering with 2 clusters was applied to the pre-treatment risk of death measures to identify patients with high and low risks. Changes in risk of death measures from the pre-treatment to post-treatment scans were binarized as increased or decreased to characterize treatment effect. Finally, patients were stratified into 4 groups based on the pre-treatment risk of death measures (high vs. low) and change in risk (increased vs. decreased). All group differences were accessed with log rank tests. The RSF model achieved a c-index of 0.654. Significant difference (χ2 = 5.59, p = 0.018) was observed between the subgroups with high and low pre-treatment death risks. The subgroups with increased and decreased risks were significantly different in overall survival (χ2 = 4.25, p = 0.039). The integration of the pre-treatment risk and change in risk stratified the patients into 4 subgroups (high + increased, high + decreased; low + increased; low + decreased) with significant differences (χ2 = 14.44, p = 0.002). Integration of risk of survival measures estimated from pre- and post-treatment CT scans could help distinguish patients with good survival from those who respond less well to SBRT or develop early nodal or distant metastases. Our study may inform clinical decision to identify early stage NSCLC patients who will most benefit from adjuvant treatment after lung SBRT.