Evaluation of radiomics as a predictor of tumor hypoxia and response to anti-PD-1 mab treatment (IO) in recurrent/metastatic HNSCC patients (R/M).
6045 Background: There is a great need for non-invasive predictors of the tumor microenvironment and the efficacy of anti-PD-1 mAb treatment (IO) in R/M HNSCC patients. We previously showed that lower tumor hypoxia was associated with increased efficacy with IO ( Journal of Clinical Oncol. 38, no. 15_suppl (May 20, 2020) 6546) and now we evaluate the predictive value of radiomics in this same patient cohort. Methods: We studied radiomic signatures in a cohort of 36 patients with R/M HNSCC treated with IO. Treatment response was evaluated using RECIST 1.1. Patients were categorized as: Responders (R) ie CR, PR, SD and non-Responders (NR) i.e PD. As per our previous analysis (ref above) hypoxia was evaluated on archival FFPE samples via immunofluorescent imaging and defined by the ratio of percent area (% CAIX) / the mean intensity (Int) of carbonic anhydrase IX in tumor (%CAIX/Int). ImageJ software was used to determine %CAIX and Int. Feature extraction was performed on the pre-immunotherapy baseline CT scans. The lesions were segmented using 3D slicer v4.10.2 to create a volume of interest (VOI) for radiomic texture analysis (TA). A total of 400 features (10 histogram-based and 390 second-order texture features) were calculated from each extracted volume of interest (VOI). Radiomic features were obtained using a feature selection approach based on Least Absolute Shrinkage and Selection Operator (LASSO). Selected features were used to build a classification model, using XGboost, for prediction of tumor response to immunotherapy. Cross-validation was performed using the Leave One Out Cross Validation (LOOCV) approach for the XGBoost method to evaluate the robustness of the estimates and calculated accuracy, sensitivity, specificity and p-value. Results: Our patient cohort had a median age of 59, 69% male, 58% smokers. 61% received IO for platinum failure, 39% frontline. Primary site included 39% OC, 22% OPC (38% HPV positive), 17% Larynx, 5% hypopharynx, and 17% other. Radiomics applied to the primary HNSCC tumor highly predicted tumor hypoxia status with a sensitivity, specificity, and accuracy of 78%, 83%, and 81%, respectively, p = 0.0001. To predict response, we applied radiomics to both the primary HNSCC tumor and pathological lymph nodes; radiomics was also able to predict whether a patient would be a responder (N = 8) versus a non-responder (N = 28) to IO based on the pre-immunotherapy baseline CT scan. The sensitivity, specificity, and accuracy were 93%, 88%, and 92%, respectively, p = 0.02. Conclusions: Even in a small cohort, radiomics could predict response to IO and tumor hypoxia in R/M HNSCC patients. To our knowledge this is the first evaluation of this kind in this patient population. Further evaluation of radiomics as a predictor of efficacy with IO and the tumor microenvironment is warranted.