Hypoxia-related radiomics predict immunotherapy response: A multi-cohort study of NSCLC

Abstract

word count: 398 Word count: 4245 Abstract Checkpoint blockade immunotherapy provides improved long-term survival in a subset of advanced stage nonsmall cell lung cancer (NSCLC) patients. However, highly predictive biomarkers of immunotherapy response are unmet clinical need. In this study, we utilized pre-treatment clinical factors and quantitative image-based biomarkers (radiomics) to identify a parsimonious model that predicts survival outcomes among NSCLC patients treated with immunotherapy. The NSCLC patients treated with single or double agent immunotherapy were included in three different cohorts: Training (N = 180), test (N = 90) and validation (N = 62) cohorts. The models were created based on overall survival (OS) and were additionally assessed for progression-free survival (PFS). Including most predictive radiomic features and clinical covariates, Classification and Regression Tree analysis was applied to stratify patients into survival risk-groups in the training cohort. The risk groups were later generated in the test and validation cohorts. Four independent NSCLC cohorts (total N = 446) were utilized for further validation of the radiomic signature. The biological underpinnings of the most informative radiomics were assessed using gene expression data from a radiogenomics dataset and validated by immunohistochemistry data (IHC). A parsimonious clinical-radiomics model was found to be significantly associated with OS and PFS after stratifying patients into groups of low-, moderate-, high-, and very-high risk of death and progression. This trained model was further tested and validated in two independent cohorts. When the extreme phenotypes were compared, the very-high risk group was found to be associated with extremely poor OS in both the test (hazard ratio [HR] = 5.35, 95% confidence interval [CI]: 2.14 -13.36; 1-year OS = 11.1%) and validation (HR = 13.81, 95% CI: 2.58 -73.93; 1year OS = %47.6) cohorts when compared to the low risk group (HRs = 1.00; 1-year OS = 85.0% & 80.2%).Similar findings were observed for PFS. The final radiomic feature (GLCM inverse difference) was associated with OS in four independent NSCLC cohorts and was found to be positively associated with the hypoxia-related carbonic anhydrase, CAIX, by gene expression profiling and immunohistochemistry. We validated a novel clinical-radiomics model that is associated with OS and PFS among NSCLC patients treated with immunotherapy and identified a highly vulnerable subset of patients that are unlikely to respond to immunotherapy. The most informative radiomic feature was associated with CAIX, a marker of tumor hypoxia, tumor acidosis, and treatment resistance.

Prognostic Value of Deep Learning-Mediated Treatment Monitoring in Lung Cancer Patients Receiving Immunotherapy

“…These results are comparable to state-of-the-art methods, which currently employs laborious and time-consuming segmentation procedures (11,12). While the field of research has been focusing on single-lesion analysis-leveraging different known factors in cancer growth, including vascularity (30), oxygenation (31), and metabolic activity (32)-our approach offers a novel fully automatic procedure which completely eradicates the need of timeconsuming segmentations, and simultaneously offers a way to provide a full picture of the patient status as seen on chest imaging. While this does not preclude the usefulness of the single-lesion approach, it proposes a way for future multi-scale solutions that leverage both single lesion imaging biomarkers as well as whole image approaches that provide general quantitative information about the status of the patient receiving treatment.…”

Identification of Stage IIIC/IV EGFR-Mutated Non-Small Cell Lung Cancer Populations Sensitive to Targeted Therapy Based on a PET/CT Radiomics Risk Model

“…Tunali et al. performed genomic analysis and immunohistochemistry analysis for carbonic anhydrase IX, and demonstrated that CT treatment response biomarkers for patients with lung cancer treated with immunotherapy were strongly associated with hypoxia, a prognostic factor ( 56 ). Ganeshan et al.…”

Prediction of immunotherapy response using deep learning of PET/CT images

“…Jiang et al(12) utilized PET/CT radiomics of 399 NSCLC patients from a single institute to generate a classifier model with an AUC of 0.86. For the studies utilizing radiomics to predict immunotherapy treatment response, Tunali et al built parsimonious classifier models with pre-treatment CT radiomic features combined with clinical covariates to predict hype-progression and progressive disease phenotypes with AUCs of 0.80-0.87(22), which were also highly correlated with PFS and OS of immunotherapy(23). Trebeschi et al(24) developed a CT-based radiomic signature that significantly discriminated progressive disease from stable and responsive disease (AUC=0.83) among NSCLC patients treated with immunotherapy.…”

“…When the DLS was combined with clinical covariates and tested in two cohorts for clinical utility by identifying patients most like to benefit to immunotherapy, we found high C-indices of 0.83-0.86 for predicting DCB, but somewhat attenuated C-indices of 0.72-0.78 for the DLS to predict PFS and OS. These models also demonstrated good performance in the external VA cohort.While others have demonstrated the utility of radiomics as a non-invasive approach to predict PD-L1 expression(11,12) or predict lung cancer immunotherapy treatment response(22)(23)(24)(25)(26) , the current work is the first to develop a PD-L1 radiomic signature and then to use this for response prediction. With respect to PD-L1 expression prediction, Patil et al (11) utilized CT images from 166 early stage NSCLC patients from a single institution to develop and validate a machine learned predictor of PD-L1 status with an AUC of 0.73.…”