Multimodal Biomarkers For Oropharyngeal Cancer

摘要

Abstract Head and neck cancers are the fifth most common cancer type in the United States, with an overall survival rate lower than 50%. Although the incidence of other sub-sites of head and neck cancer has decreased steadily in past decades, the number of oropharyngeal squamous cell carcinoma (OPSCC) cases has increased significantly. Most OPSCC patients receive standard cancer therapy.4 However, the clinical outcomes vary significantly and are difficult to predict. Predicting early in treatment whether a tumor is likely to respond to treatment is one of the most difficult yet important tasks in providing individualized cancer care. Human papillomavirus (HPV) is a known driving oncogenic factor in oropharyngeal cancer, as well as a significant prognostic biomarker for patient survival. Retrospective studies conducted by the International Head and Neck Cancer Epidemiology Consortium (INHANCE) have demonstrated that clinical biomarkers have prognostic value in helping stratify OPSCC patients into groups with differing risks of death or disease progression. However, HPV-positive oropharyngeal cancer patients have similar rates of metastatic spread to HPV-negative patients. The same is true for patient groups stratified with other clinical biomarkers. More robust prognostic biomarkers are needed to accurately stratify patients for optimally effective treatment. MicroRNAs (miRNAs) are a family of small non-coding RNA molecules that collectively control the expression of thousands of protein-coding genes. Multiple studies indicate that miRNAs are promising cancer biomarkers and play critical regulatory roles in oropharyngeal cancer. Imaging features extracted from medical images are an exciting new class of cancer biomarkers for characterizing tumor habitats. For several tumor sites, imaging biomarkers have shown promise in accurately separating favorable and unfavorable prognosis patients. However, current efforts to utilize high-dimensional multimodal biomarkers for treatment outcome prediction have been compromised by small patient numbers relative to the feature space dimensionality; feature redundancy, heterogeneity, and uncertainty; and patient cohorts with unbalanced outcomes. The correlation, independence, and complementary nature of multimodal biomarkers (imaging, miRNA, HPV, clinical, and histopathologic biomarkers) remains unexplored as well. The major goal of this research is to develop a multimodal biomarker-based model that can reliably predict subsets of OPSCC patients with low and high risks for treatment failure. The model will serve as a clinical decision-making tool. Specifically, we propose a novel principle and systematic machine learning-based strategy to effectively identify and seamlessly combine prognostic information carried by multimodal biomarkers. Aim 1: Identify prognostic multimodal biomarkers, given OPSCC patient data. Aim 2: Develop and test a comprehensive multimodal biomarker-based model for predicting OPSCC treatment outcomes. Aim 3: Assess the clinical benefit of the model for OPSCC patient stratification and individualized treatment.