Deciphering site-specific histopathological parameters with potential clinical value in head and neck squamous cell carcinomas.

In the diagnostic setting of head and neck squamous cell carcinoma (HNSCC), there is an unmet need for robust histological prognosticators, since grading alone is considered to be of disputed clinical relevance. In this context, the present study assessed site-specific and tumor compartment characteristics as potential histological risk factors in HNSCC. Morphological and immunophenotypic (such as CD8 and PD-L1) characteristics of tumor cells, the immune microenvironment and the invasive margin (IM) were assessed in 248 patients with HNSCC who were followed up for >20 years, to determine their site specificity and impact on the overall survival of patients. Laryngeal and hypopharyngeal carcinomas were characterized by keratinization, cell cannibalism and anaplastic features; oropharyngeal carcinoma was characterized by a high grade and necrosis; and oral carcinomas was characterized by keratin pearls, eosinophil infiltrates and perineural invasion (P<0.05). CD8 distribution homogeneity, tumor center and/or IM perineural invasion, and sparse lymphocytic host response in the IM (LHR-IM) were unfavorable prognosticators (P<0.05). Among these parameters, only LHR-IM and perineural invasion throughout the tumor independently predicted an unfavorable prognosis, along with disease stage. Grade, keratinization, anaplastic features, cell cannibalism, necrosis-related features, eosinophils, worst pattern of invasion (the most aggressive pattern of tumor cell proliferation found at the invasive front), tumor lymphocytic and CD8+ T-cell infiltrates, and primary site were not associated with prognosis in the present study. In conclusion, perineural invasion and LHR-IM were confirmed as histological risk factors in HNSCC, which should be included in pathology reports. To revise the assessment of HNSCC tumor grade, the aforementioned site-specific histological characteristics may be used in deep learning algorithms.