Abstract 3634: PET/CT imaging prediction of response to checkpoint blockade in advanced non-small cell lung cancer patients

Abstract

Abstract Immune checkpoint blockade with monoclonal antibodies directed at the inhibitory immune receptors PD-1 and PD-L1 has emerged as a one of the most successful treatment approaches in advanced Non-small Cell Lung Cancer (NSCLC) patients. However, given its adverse effects, high cost and only approximately 20% response rate, accurate response predication is import to identify patients most likely to respond. In previous work, we and others have shown that tumor acidosis, generated from high rates of glucose metabolism, was a potent inhibitor of anti-tumor immunity and that neutralization of acid improved response to checkpoint blockade. In this study, we investigated the value of pre-therapy PET/CT images on response prediction to test the hypothesis that the most metabolically active tumors would have the poorest response to checkpoint blockade. 64 patients were retrospectively curated with pathology verified NSCLC and who did not receive intervening treatments between the PET/CT imaging and initiation of anti-PD1 therapy. Responders (long-term and ongoing responders, n=22) and non-responders (non-responders and transient responders, n=42) were classified according to RECIST criteria. These patients were randomly divided into two groups (n=32 each) for training and testing, respectively. From their images, we constructed 5 different fusion images comprised of PET and CT data. We then extracted 455 quantitative features from these fusion images, as well as the original PET/CT images. We then generated an additional 975 fusion features calculated from the above 455 features, as well as some clinical features (location, age, sex, and histologic type). From these, the most informative features were selected with Receiver Operating Characteristic Curve (ROC) analysis, and these were analyzed with a combined 2-layer cascade support vector machine (cSVM) classification model to relate the features to prediction of objective response. From this we determined that the most informative features were fusion features. With the cSVM, the training set achieved accuracy and area under the ROC curve (AUROC) for classifying response of 100% and 1.00, respectively. Using this model on the test set achieved accuracy and AUROC of 89.66% and 0.86, respectively. We conclude from that PET/CT fusion images may be useful prior to initiation of therapy to predict response to checkpoint blockade in advanced non-small cell lung cancer patients. Citation Format: Wei Mu, Jin Qi, Hong Lu, Mathew Schabath, Yoganand Balagurunathan, Ilke Tunali, Shari Pilon-Thomas, Robert J. Gillies. PET/CT imaging prediction of response to checkpoint blockade in advanced non-small cell lung cancer patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 3634.