A Radiomic Prognostic Index Refines Risk Models for Distant Metastasis in HPV-Related Oropharyngeal Cancer

Abstract

Distant metastasis (DM) is the main cause of treatment failure in human papillomavirus (HPV)-related oropharyngeal cancers (OPC). However, predicting DM remains challenging. We hypothesize that quantitative imaging (i.e. radiomic) features may help identify primary tumor features associated with higher risk of DM. We retrieved radiotherapy planning CT scans for 280 UICC/AJCC 8th ed. stage I-III p16-positive OPC patients treated with definitive radiotherapy or chemoradiotherapy at a single institution between 2005 and 2010. Gross tumor volumes (GTVs) were previously contoured by each treating radiation oncologist. A radiomic prognostic index (RPI) was derived based on four validated prognostic radiomic features from tumor intensity, shape, texture, and wavelet feature groups. RPIs were extracted from each GTV based on published algorithms. We used univariate and multivariate analysis with Cox proportional hazards models to identify predictors of DM. Feature selection for multivariate models was performed using recursive partitioning analysis (RPA) and validated. Receiver-operating characteristic (ROC) curves were used to compare models. Median follow-up was five years. There were a total of 34 DM events. Factors associated with increased risk of DM included UICC/AJCC 8th ed. stage III (T4 or N3) (73/280; p<0.001), large GTV (140/280; p=0.048), and high risk RPI (25/280; p<0.001). High risk RPI remained significant (p<0.001) on multivariate analysis. RPA confirmed RPI as a key predictor segregating p16-positive OPCs into low (n=255) and high (n=25) risks of DM with 3-year distant control rates of 91% (0.88-0.95) and 50% (0.32-0.78) respectively. Area under the ROC curve for the RPI-based DM model was 0.72 (0.62-0.82) compared with 0.65 (0.56-0.74) for the conventional stage-based DM model. In subgroup analyses, the RPI consistently stratified patients for DM risk, particularly for those cohorts with greater risk, such as UICC/AJCC 8th ed. stage III cancers. The four-feature RPI can be used alone or in conjunction with other clinical or treatment characteristics to classify HPV-related OPC patients by DM risk. The RPI may therefore allow for greater precision in selection of patients for systemic therapies for HPV-related OPC.