Predictive and prognostic value of CT based radiomics signature in locally advanced head and neck cancers patients treated with concurrent chemoradiotherapy or bioradiotherapy and its added value to Human Papillomavirus status

Objectives: To explore prognostic and predictive value of radiomics in patients with locally advanced head and neck squamous cell carcinomas (LAHNSCC) treated with concurrent chemoradiotherapy (CRT) or bioradiotherapy (BRT). Materials and Methods: Data of 120 patients (CRT vs. BRT matched 2: 1) were retrospectively analyzed. A total of 544 radiomics features of the primary tumor were extracted from radiotherapy planning computed tomography scans. Cox proportional hazards models were used to examine the association between survival and radiomics features with false discovery rate correction. The discriminatory performance was evaluated using receiver operating characteristic curve analysis. Results: Multivariate analysis showed a 24-feature based signature significantly predicted for OS (HR = 0.3, P = 0.02) and progression-free survival (PFS) (HR = 0.3, P = 0.01). Combining the radiomics signature with p16 status showed a significant improvement of prognostic performance compared with p16 (AUC = 0.78 vs. AUC = 0.64 at 5 years, P = 0.01) or radiomics signature (AUC = 0.78 vs. AUC = 0.67, P = 0.01) alone. When patients were stratified according to this combination, OS and PFS were significantly different according to the 4 sub-types (p16+ with low/high signature score; p16-with low/high signature score) (P < 0.001). Patients with high signature score significantly benefited from CRT (vs. BRT) in terms of OS (P = 0.004), while no benefit from CRT in patients with low signature score. Conclusion: Our analysis suggests an added value of radiomics features as prognostic and predictive biomarker in HNSCC treated with CRT/BRT. Moreover, the radiomics signature provided additional information to HPV/p16 status to further stratify patients. External validation of such findings is mandatory given the risk of overfitting. (C) 2017 Elsevier Ltd. All rights reserved.