Head and Neck Cancer

PV QA 2 - Poster Viewing Q&A 2

MO_34_2810 - Predicting PD-L1 Expression using Radiomics in Oropharyngeal Cancer Patients treated with Definitive Radiation

Monday, October 22
10:45 AM - 12:15 PM
Location: Innovation Hub, Exhibit Hall 3

Predicting PD-L1 Expression using Radiomics in Oropharyngeal Cancer Patients treated with Definitive Radiation
R. Rahman1, V. Sridharan2, G. J. Hanna3, N. Chau3, J. Lorch3, J. Kass4, D. Annino4, L. Goguen4, R. Uppaluri4, R. I. Haddad3, R. B. Tishler5, D. N. Margalit5, J. D. Schoenfeld5, and R. Y. Huang6; 1Harvard Radiation Oncology Program, Boston, MA, 2Harvard Medical School, Boston, MA, 3Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA, 4Department of Surgery/Otolaryngology, Brigham & Women's Hospital and Dana-Farber Cancer Institute, Boston, MA, 5Brigham and Women's Hospital and Dana-Farber Cancer Institute, Boston, MA, 6Department of Radiology, Brigham and Women's Hospital and Dana-Farber Cancer Institute, Boston, MA

Purpose/Objective(s): Immunotherapy directed against the programmed death (PD)-1 receptor is approved for the treatment of advanced squamous cell carcinoma of the head and neck (SCCHN), and is being actively explored in the locoregionally advanced setting in combination with radiation. PD-L1 expression is a biomarker that can potentially enrich for responding patients; therefore, accurately determining PD-L1 expression may have predictive value and be of clinical use. The current pilot study employed a machine-learning algorithm to generate models predictive of tumor PD-L1 expression in oropharyngeal SCCHN patients.

Materials/Methods: We retrospectively evaluated patients with oropharyngeal SCCHN treated at our institution with definitive radiation or chemoradiation between 2004-2014 with a diagnostic contrast-enhanced neck CT completed prior to radiation therapy with evaluable gross tumor. Tumor PD-L1 expression was determined by immunohistochemical staining, and primary and nodal tumor volumes were manually contoured for extraction of features encompassing shape, texture, and intensity. Using a support vector machines (SVM) algorithm, we used non-redundant features to generate a predictive model with ten-fold cross-validation for low vs. high tumor PD-L1 expression with dichotomization by the median value in the cohort for each parameter. Individual features and model performance were evaluated with receiver operator characteristic area under the curve (AUC) analysis.

Results: We identified 47 oropharyngeal patents of median age 57 (range 43-71); 75% of patients were AJCC version 7 clinical stage IVA. Of 41 patients with known HPV status, 37 (90%) were HPV-positive. Forty-four (94%) patients received concurrent or induction chemotherapy. The median PD-L1 expression expressed as percentage of tumor cell was 14.5% (IQR 1.8%-34.6%). The median PD-L1 H-score average incorporating PD-L1 staining intensity was 33.8 (IQR 5.6-73.7). We extracted a total of 130 CT-based features. Model performance was improved when both primary and nodal volumes were included in tandem. Using the SVM algorithm, a predictive model was generated that predicted PD-L1 expression as percentage of tumor cells (AUC = 0.72, 95% CI 0.66-0.78) with a median sensitivity and specificity of 0.74 and 0.75, respectively. A predictive model was also generated for PD-L1 H-score incorporating PD-L1 staining intensity (AUC = 0.77, 95% 0.70-0.81) with a median sensitivity and specificity of 0.75 and 0.78, respectively.

Conclusion: Using a machine-learning algorithm, we constructed models for prediction of tumor PD-L1 expression in locoregionally advanced oropharyngeal SCCHN patients treated with definitive radiation or chemoradiation. Further validation of our predictive models is warranted.

Author Disclosure: R. Rahman: None. V. Sridharan: None. G.J. Hanna: None. N. Chau: Research Grant; Merck. J. Lorch: Research Grant; Millennium, BMS, Bayer, Novartis. J. Kass: None. R. Uppaluri: Advisory Board; Merck. R.I. Haddad: Research Grant; Merck, Celgene, Pfizer, BMS, Astra Zeneca. Consultant; Merck, Celgene, Pfizer, BMS, Astra Zeneca, Genzyme, Eisai. CHAIR THYROID PANEL; NCCN. R.B. Tishler: Advisory Board; EMD Serrono, Izun Pharmaceuticals. D.N. Margalit: Research Grant; NCCN. J.D. Schoenfeld: Research Grant; Merck, BMS. Consultant; Tilos. Advisory Board; Nanobiotix, AstraZeneca, Debiopharm, BMS. Travel Expenses; BMS. Translational PI; NCI Match Subprotocol Z1D.

Send Email for Rifaquat Rahman


Assets

MO_34_2810 - Predicting PD-L1 Expression using Radiomics in Oropharyngeal Cancer Patients treated with Definitive Radiation



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Predicting PD-L1 Expression using Radiomics in Oropharyngeal Cancer Patients treated with Definitive Radiation