Abstract.Background: Head and neck squamous cell carcinoma (HNSCC) is a debilitating and deadly disease largely due to late stage diagnosis. Prior work indicates that soluble CD44 (solCD44) and total protein may be useful diagnostic markers for HNSCC. In this study we combine the markers solCD44, IL-8, HA, and total protein with demographic and risk factor data to derive a multivariate logistic model that improves HNSCC detection as compared to our previous data using biomarkers alone. Methods: We performed the solCD44, IL-8, HA, and total protein assays on oral rinses from 40 HNSCC patients and 39 controls using ELISA assays. Controls had benign diseases of the upper aerodigestive tract and a history of tobacco or alcohol use. All subjects completed a questionnaire including demographic and risk factor data. Results: Depending on cancer subsite, differences between cases and controls were found for all markers. A multivariate logistic model including solCD44, total protein and variables related to smoking, oral health and education offered a significant improvement over the univariate models with an AUC of 0.853. Sensitivity ranged from 75-82.5% and specificity from 69.2-82.1% depending on predictive probability cut points. Conclusion: A multivariate model, including simple and inexpensive molecular tests in combination with risk factors, results in a promising tool for distinguishing HNSCC patients from controls. Impact: In this case-control study, the resulting observations led to an unprecedented multivariate model that distinguished HNSCC cases from controls with better accuracy than the current gold standard which includes oral examination followed by tissue biopsy. Since the components are simple, noninvasive, and inexpensive to obtain, this model combining biomarkers, risk factor and demographic data serves as a promising prototype for future cancer detection tests.

Head and neck cancer is the sixth most common cancer worldwide. Each year 50,000 individuals in the United States and 500,000 worldwide are diagnosed with head and neck squamous cell carcinoma (HNSCC). The majority of HNSCC is identified in advanced stage, when cure rates are low, largely because there is no effective screening program available. Furthermore, minority patients and those of low socioeconomic status (SES) suffer disproportionately from this disease. Building on prior work, this study examines whether the combination of soluble CD44 (solCD44), hyaluronic acid (HA), IL-8 and total protein holds promise for early detection of HNSCC. In our previous work, we determined that a solCD44 oral rinse test distinguished HNSCC cases from benign disease (BD) controls with 62% sensitivity and 88% specificity. In this study we determined if sensitivity and specificity are improved when the model includes a panel of markers consisting of solCD44, IL-8, HA and total protein as well as demographic and risk factor data. We performed the solCD44, IL-8, HA and total protein assays on oral rinses from 40 HNSCC patients and 39 BD controls using ELISA and ELISA-like assays. Controls had benign diseases of the upper aerodigestive tract and gave a history of tobacco and/or alcohol use. All subjects completed a questionnaire including demographic and risk factor data. All marker levels with the exception of total protein correlated with site of disease, with higher levels found in oral cavity and oropharynx compared to larynx and hypopharynx. Differences in marker levels between cases and controls were statistically significant for solCD44 and total protein (P =0.009 and P =0.008, respectively) when sites where grouped together. Univariate logistic models for each marker comparing cancer patients versus BD controls and bivariate logistic models for each marker and each covariate were computed in preliminary analyses. Significant interactions found included that between log2solCD44 and smoking history and between log2total protein and number of teeth removed. This indicates that smoking is an effect modifier for log2solCD44, and teeth removal is an effect modifier for log2total protein. Among never smokers, mean log2solCD44 was significantly higher in cases than in BD controls. With respect to log2total protein, the mean was significantly higher in cases than in controls among subjects having had five or fewer teeth removed. A multivariate logistic model offered a significant improvement over the univariate models with an AUC of 0.853. Sensitivity ranged from 75-82.5% and specificity from 69.2-82.1% depending on predictive probability cut points. We hypothesize that solCD44 and total protein levels reflect early tumorigenesis closely associated with or caused by tobacco exposure and poor oral health. This model, incorporating biomarkers and risk factors, should be investigated further because it is simple and inexpensive to perform and includes effect modifiers such as tobacco use and poor oral health, which are common challenges in low income populations. Citation Information: Cancer Epidemiol Biomarkers Prev 2010;19(10 Suppl):B109.