Radiomic Biomarkers in Oropharyngeal Carcinoma (OPC-Radiomics)

Summary:

Oropharynx cancer is increasing in frequency in North America. More refined risk stratification and treatment personalization strategies are needed to improve cure rates and limit toxicities. Radomic features have shown promise for prognostication in oropharynx cancer based on datasets including hundreds of patients. Clinical outcomes data for oropharynx cancer patients treated at Princess Margaret Cancer Centre is of the highest quality, collected prospectively at the point of care. This data has resulted in the recent AJCC/UICC 8th edition staging system for oropharynx cancer (first published by O'Sullivan et al. in Lancet Oncology, 2016 [PMID 26936027]). By making publicly available the published data from our institution, which has now been published twice (see next question), we will empower researchers to perform meta-analysis and validation exercises of prognostic radiomic models. Ultimately this will lead to more robust models and bring us closer to clinical implementation and impact for our patients.

Purpose:

Distant metastasis (DM) is the main cause of death for patients with human papillomavirus (HPV)-related oropharyngeal cancers (OPC); yet there are few reliable predictors of DM in this disease. The role of quantitative imaging (i.e. radiomic) analysis was examined to determine whether there are primary tumor features discernible on imaging studies that associate with a higher risk of developing DM.  This data collection consists of oropharynx squamous cell carcinoma.

Methods:

Radiotherapy planning CT scans were retrieved for all non-metastatic p16-positive OPC patients treated with radiotherapy or chemoradiotherapy at a single institution between 2005 and 2010. Radiomic biomarkers were derived from each gross tumor volume (GTV). Biomarkers included four representative radiomic features from tumor first order statistics, shape, texture, and wavelet groups as well as a combined four-feature signature. Univariable Cox proportional hazards models for DM risk were identified. Discriminative performance of prognostic univariable and multivariable models was compared using the concordance index (C-index). Subgroup analyses were performed.

Results:

There were 300 HPV-related OPC patients who were eligible for the analysis. A total of 36 DM events occurred within a median follow-up of five years. On univariable analysis, top results included the four representative radiomic features (p<0.001), the radiomic signature (p<0.001), tumor stage (p<0.001), tumor diameter (p<0.001), and tumor volume (p<0.001). C-indices of the radiomic features (0.670-0.686), radiomic signature (0.670), stage (0.633), and tumor size metrics (0.653-0.674) demonstrated moderate discrimination of DM risk. Combined clinical-radiomic models yielded significantly improved performance (0.701-0.714; p<0.05). In subgroup analyses, the radiomic biomarkers consistently stratified patients for DM risk, particularly for those cohorts with greater risks (0.663-0.796), such as patients with stage III disease.

Conclusions:

Radiomic biomarkers appear to classify DM risk for patients with non-metastatic HPV-related OPC. Radiomic biomarkers could be used either alone or with other clinical characteristics in assignment of DM risk in future HPV-related OPC clinical trials.

• 300 patients in this dataset are also a subset of 542 patients from Princess Margaret Cancer Centre published in 2015: External validation of a prognostic CT-based radiomic signature in oropharyngeal squamous cell carcinoma. PMID: 26264429 DOI: 10.3109/0284186X.2015.1061214,
• This dataset is also described in Radiomic Biomarkers to Refine Risk Models for Distant Metastasis in HPV-related Oropharyngeal Carcinoma. PMID: 29506884 DOI: 10.1016/j.ijrobp.2018.01.057

For scientific inquiries about this dataset,

please contact: Dr. Scott Bratman Scott.Bratman@rmp.uhn.ca

Acknowledgements

We would like to acknowledge the individuals and institutions that have provided data for this collection: