Development and validation of an individualized risk prediction model for oropharynx cancer in the US population

Background: The incidence of oropharynx cancers has increased substantially in the United States. However, risk stratification tools for the identification of high-risk individuals do not exist. In this study, an individualized risk prediction model was developed and validated for oropharynx cancers in the US population. Methods: A synthetic, US population–based case-control study was conducted. Oropharynx cancer cases diagnosed at Ohio State University (n = 241) were propensity-weighted to represent oropharynx cancers occurring annually in the United States during 2009-2014 (n = 12,656). Controls (n = 9327) included participants in the National Health and Nutrition Examination Survey (2009-2014) and represented the annual US population aged 30 to 69 years (n = 154,532,508). The individualized 1-year absolute risk of oropharynx cancer was estimated with weighted logistic regression. Results: The risk prediction model included age, sex, race, smoking, alcohol use, lifetime sexual partners, and oral oncogenic human papillomavirus (HPV) status. The model had good discrimination and calibration in split-sample validation (area under the curve [AUC], 0.94; 95% confidence interval [CI], 0.92-0.97; observed/expected [O/E], 1.01; 95% CI, 0.70-1.32) and external validation (AUC, 0.87; 95% CI, 0.84-0.90; O/E, 1.08; 95% CI, 0.77-1.39). In the US population, 1-year predicted risks of oropharynx cancer were highest for older individuals (21.1/100,000 for 65- to 69-year-olds), men (13.9/100,000), whites (10.4/100,000), smokers (18.0/100,000 for >20 pack-years), heavy alcohol users (18.4/100,000), and those with prevalent oral oncogenic HPV (140.4/100,000). The risk prediction model provided substantial risk stratification, with approximately 77% of all oropharynx cancers and approximately 99% of HPV-positive oropharynx cancers occurring in the 10% of the US population with the highest model-predicted risk. Conclusions: This risk prediction model will enable the efficient design of studies to address the outstanding questions pertaining to the natural history, screening, and secondary prevention of oropharynx cancers.