Validation of the Histologic Risk Model in a New Cohort of Patients With Head and Neck Squamous Cell Carcinoma

Background: Half of the patients with head and neck squamous cell carcinoma (HNSCC) can be expected to fail therapy, indicating that more aggressive treatment is warranted for this group. We have developed a novel risk model that can become a basis for developing new treatment paradigms. Here we report on the performance of our model in a new multicenter cohort. Design: Eligible patients from 3 institutions (Montefiore Medical Center, University of Manitoba, and New York University Medical Center) were identified and pathology slides from their resection specimens were reviewed by Margaret Brandwein-Gensler; risk category was assigned as previously published. Kaplan-Meier analysis was performed for disease progression and survival. Cox proportional hazards regression was performed, adjusted for potential confounders. A teaching module was also developed; attending pathologists were asked to score coded slides after a lecture and multiheaded microscope teaching session. Agreement was assessed by calculating Cohen unweighted kappa coefficients. Result: The validation cohort consisted of 305 patients, from the above institutions, with 311 primary HNSCC of the oral cavity, oropharynx, and larynx. The median follow-up period for all patients was 27 months. Risk category predicts time to disease progression (P = 0.0005), locoregional recurrence (P = 0.013), and overall survival (P = 0.0000) by Kaplan-Meier analysis. High-risk status is significantly associated with decreased time to disease progression, adjusted for clinical confounders (P = 0.015, hazard ratio 2.32, 95% confidence interval 1.18-4.58) compared with collapsed intermediate and low-risk groups. We also demonstrate substantial interrater agreement (kappa = 0.64), and very good rater agreement when compared with the standard (kappa = 0.87). Conclusions: We demonstrate significant predictive performance of the risk model in a new cohort of patients with primary HNSCC, adjusted for confounders. Our training experience also supports the feasibility of adapting the risk model in clinical practice.