Association of Human Papillomavirus Status at Head and Neck Carcinoma Subsites With Overall Survival

Importance Data are limited on the prognostic value of human papillomavirus (HPV) status for head and neck carcinoma subsites. Objective To determine whether HPV positivity at each head and neck subsite is associated with improved overall survival. Design, Setting, and Participants This retrospective population-based cohort study used the National Cancer Database to identify patients diagnosed with head and neck squamous cell carcinomas from January 1, 2010, to December 31, 2014. Patients were classified according to the location of their primary malignancy into 1 of the 6 main subsites of the upper aerodigestive tract: oral cavity, oropharynx, nasopharynx, hypopharynx, larynx, and sinonasal tract. Patients were also classified by their HPV status. Data collection for this study took place from January 1, 2010, to December 31, 2014. Data analysis was conducted from August 1, 2017, to September 30, 2017. Main Outcomes and Measures The difference in 5-year overall survival between patients with HPV-positive status and those with HPV-negative status in various head and neck carcinoma subsites; the role of HPV status in an unadjusted Cox multivariate regression model. Results Of the 175 223 total number of patients identified (129 634 [74.0%] male; 45 589 [26.0%] female; mean [SD] age, 63.1 [11.9] years), 133 273 (76.1%) were ineligible and 41 950 (23.9%) were included in the sample. This sample included 16 644 patients (39.7%) with HPV-positive tumors and 25 306 (60.3%) with HPV-negative tumors. Patients with an HPV-positive status were more likely to be younger, be white, be male, present with local T category tumors, and have poor differentiation on histologic examination. HPV-positive status was associated with survival at 4 tumor subsites: oral cavity (hazard ratio [HR], 0.76; 95% CI, 0.66-0.87), oropharynx (HR, 0.44; 95% CI, 0.41-0.47), hypopharynx (HR, 0.59; 95% CI, 0.45-0.77), and larynx (HR, 0.71; 95% CI, 0.59-0.85). The HPV status was the greatest factor in survival outcome between the HPV-positive and -negative cohorts at the oropharynx subsite (77.6% vs 50.7%; survival difference, 26.9%; 95% CI, 25.6%-28.2%) and hypopharynx subsites (52.2% vs 28.8%; survival difference, 23.4%; 95% CI, 17.5%-29.3%). For the nasopharynx (HR, 1.03; 95% CI, 0.75-1.42) and sinonasal tract (HR, 0.63; 95% CI, 0.39-1.01) subsites, HPV-positive status was not an independent prognostic factor. Conclusions and Relevance Human papillomavirus positivity was associated with improved survival in 4 subsites (oropharynx, hypopharynx, oral cavity, and larynx), and the largest survival difference was noted in the oropharynx and hypopharynx subsites. In the nasopharynx and sinonasal tract subsites, HPV positivity had no association with overall survival. Given these results, routine testing for HPV at the oropharynx, hypopharynx, oral cavity, and larynx subsites may be warranted.

H ead and neck cancers are the sixth most common solid cancer worldwide, with more than 60 000 new cases per year. 1 Human papillomavirus (HPV) infection is now accepted to be a previously unrecognized cause of head and neck squamous cell carcinoma (HNSCC). 2 In the case of oropharyngeal squamous cell carcinoma (OPSCC), there has been as much as a 225% increase in HPV-positive cancers between 1988 and 2004, 3 and up to 70% of new cases are caused by HPV. 2,4,5 In general, patients with HPV-positive OPSCC use less tobacco and alcohol and are more likely to be younger than their counterparts who are negative for HPV. 3,5 HPV-positive status is associated with a significant beneficial impact on prognosis, 6-8 with 1 study reporting a 25% increase in survival at 3 years. 9 HPV-positive OPSCC responds more positively to radiotherapy, which may be associated with defects in double-strand break repair. [10][11][12] This improvement has led to calls for deintensified treatments, 13 which are currently being investigated. 14-16 However, investigations into non-OPSCC subsites, such as the hypopharynx, nasopharynx, oral cavity, larynx, and sinonasal cavity, are relatively scarce.
The literature on HPV in non-OPSCC subsites is controversial. Studies revealed that HPV is present in these subsites, albeit estimated to be 5 times less prevalent in non-OPSCC than OPSCC. 17-21 A 2016 study that compared the gene expression and DNA methylation profiles of HPV in non-OPSCC subsites with those in OPSCC subsites found them to be identical, leading to the conclusion that HPV can drive carcinogenesis in non-OPSCC. 22 The same study concluded that HPV-driven non-OPSCC has a distinct tumor microenvironment compared with HPV-driven OPSCC. Few studies have looked at the role of HPV at each individual non-OPSCC subsite. Tumors of some subsites, particularly nasopharyngeal tumors, are rare; thus, accurately characterizing the prognostic role of HPV has been difficult.
The purpose of this study was to identify the prognostic role of HPV in all HNSCC subsites. Its results will elucidate HPV's value and importance as a prognostic tool at other subsites, which may help inform treatment decisions and reduce the future burden of HNSCC.

Data
Data on a large sample of patients diagnosed with HNSCC were extracted from the National Cancer Database (NCDB). The NCDB is a joint project of the Commission on Cancer and the American Cancer Society 23 that represents more than 70% of incidences of cancer in the United States. Cases in the NCDB are recorded by more than 1500 accredited hospitals in the United States and Puerto Rico. This study was exempt from review by the Yale University Human Research Protection Program because it used a preexisting, deidentified public database. Patient informed consent was not necessary. Data collection for this study took place from January 1, 2010, to December 31, 2014. Data analysis was conducted from August 1, 2017, to September 30, 2017.

Patient Population
Our study population comprised 41 950 patients in the NCDB whose primary malignancy was diagnosed as HNSCC be- We categorized HPV status as negative, positive for lowrisk HPV types, positive for high-risk HPV types (HPV-16 and/or HPV-18), or unknown. Patients were classified as HPVpositive if they tested positive for high-risk HPV types or as HPV-negative if they received a negative HPV test result. Patients were excluded if they had low-risk HPV types or unknown HPV status.
We examined patient demographic and tumor data, including age at diagnosis, race/ethnicity, Charleson/Deyo comorbidity score (score range: 0-2, with the highest score indicating a patient with 2 or more comorbidities), primary tumor site, TNM classification by the American Joint Commission on Cancer and the International Union Against Cancer, tumor grade, primary treatment type, insurance status, median income quartiles, treatment facility type and location, and rural or urban classification of primary county of residence. Patients were excluded if they were younger than 18 years, their TNM classification was unknown, or their primary treatment type was unknown. Primary treatment types were as follows: no treatment; radiation only; chemotherapy only; surgery only; radiation and chemotherapy; surgery and radiation; and surgery, radiation, and chemotherapy.

Statistical Analysis
Data analyses were performed using SPSS, version 19.0 (IBM Corp). To compare the distribution of characteristics between patients with HPV-positive status and those with HPV-

Key Points
Question What is the prognostic role of the human papillomavirus (HPV) at each unique head and neck carcinoma subsite?
Findings In this population-based cohort study involving 41 950 patients with head and neck squamous cell carcinoma, human papillomavirus-positive status was associated with improved overall survival from tumors at oral cavity, oropharynx, hypopharynx, and larynx subsites but not at the nasopharynx and sinonasal tract subsites.
Meaning Routine testing for human papillomavirus may be warranted in tumors originating from the oral cavity, oropharynx, hypopharynx, and larynx subsites.
negative status, we used 2-sample t tests and χ 2 tests. The comparison of mean age at diagnosis was analyzed using a 2-sample t test. The proportional distribution of race/ethnicity, primary tumor site, T and N classification, lymph node metastasis, primary treatment type, insurance status, median income quartiles, treatment facility type and location, and rural or urban classification of patient's primary country of residence was determined using χ 2 tests. Survival analysis was performed using the Kaplan-Meier (KM) method. An unadjusted Cox proportional hazards regression model was used for multivariable survival analysis. Age, sex, race/ethnicity, TNM classification, Charleson/Deyo score, HPV status, primary treatment type, insurance status, and median income were entered a priori into the model. A 2-sided P < .05 was considered to be statistically significant. Effect-size measures and 95% CIs around the effect-size measures were included to provide estimates of the precision of observed effect size and whether the data were compatible with clinically meaningful differences.

Association of HPV Status With Survival
To determine the association of HPV status with survival among patients with HNSCC, we performed 3 analyses: (1) 5-year unadjusted survival rate, (2) KM survival curve, and (3) unadjusted Cox proportional hazards regression. Subsites in which HPV positivity was found to be associated with improved outcome in the Cox model were further classified as having strong or moderate association on the basis of the difference in 5-year unadjusted survival rates between the HPV-positive and HPV-negative cohorts. A difference greater than 20% survival was classified as strong, and a difference less than 20% was classified as moderate. Subsites in which HPV positivity was found to have no association with improved outcome in the Cox model were classified as having no association.

Results
We identified a total of 175 223 patients (129 634 Figure 1). Of this total, 133 273 patients (76.1%) were ineligible and 41 950 (23.9%) were included in the sample. This sample included 16 644 patients (39.7%) with HPV-positive tumors and 25 306 (60.3%) with HPV-negative tumors). Baseline patient, hospital, clinical, and treatment characteristics by each subsite are shown in eTables 1-6 in the Supplement. In general, patients in the HPV-positive cohort were more likely than their HPV-negative counterparts to be white, be younger, be male, present with local T category tumors, and have poor differentiation on histologic examination.
On multivariate analysis, after accounting for age, sex, race/ ethnicity, Charleson/Deyo score, insurance status, median income, T and N classification, and primary treatment type, we observed that HPV-positive status remained an independent prognostic factor for the oral cavity (hazard ratio Other factors associated with survival at each subsite are shown in eTables 7-12 in the Supplement. Receiving any treatment other than chemotherapy alone was associated with improved survival in 5 of the 6 subsites. The HRs ranged between 0.07 and 0.7, and the 95% CIs did not include unity (1.0) when comparing treatment groups with the baseline no treatment. Receiving chemotherapy alone did not affect survival in 5 of the 6 subsites; the HRs ranged between 0.8 and 1.1, and the 95% CI included unity. Having a score of 2 on the Charlson/ Deyo scale was associated with worse survival at all subsites; the HRs ranged between 1.4 and 2.1, and the 95% CIs were larger than unity.

Discussion
To our knowledge, this study was the largest and most comprehensive retrospective study examining the role of HPV and its association with overall survival at all head and neck subsites. We used a combination of survival differences between the HPVpositive and the HPV-negative cohorts and multivariate models to develop 3 categories for measuring this association: strong, moderate, and no association. We found HPV to have a strong association with overall survival in the oropharynx and hypopharynx subsites, moderate association with improved survival in the  Unadjusted hazard ratios (HRs) for HPV status and its association with overall survival are shown for each subsite. HPV-positive status is compared with baseline HPV status.

Research Original Investigation
Association of HPV Status at Head and Neck Carcinoma Subsites With Overall Survival oral cavity and larynx subsites, and no association in the nasopharynx and sinonasal tract subsites. These results suggest that a variance in the magnitude of survival benefit existed between the subsites, providing a foundation for further study. Why HPV plays a bigger prognostic role in the oropharynx and hypopharynx than in the oral cavity and larynx is unknown, although perhaps it is because the anatomy and function of each subsite differ substantially. This theory may partly explain the similarity in the role of HPV between the adjacent oropharynx and hypopharynx. Preclinical studies alluded to differences in the microtumor environment between OPSCC and non-OPSCC, 22 which may explain the contrast seen between the oral cavity and larynx as well as the oropharynx and hypopharynx subsites.
Recent studies have found that mutations in TRAF3 (OMIM 601896) and CYLD (OMIM 605018) occur only in HPVassociated HNSCC 24 and correlate with survival. 25,26 The absence of viral genome integration is also associated with improved survival 27 and was predicted by mutations in TRAF3 or CYLD. 26 Together, these data suggest that HPV carcinogenesis can occur through HPV integration or through maintenance of the HPV episome 26 and that tumors lacking HPV integration have improved survival. Future studies are required to determine whether laryngeal and oral cavity subsites are more likely to lack mutations in TRAF3 or CYLD and have HPV integration, which could explain why HPV is not associated with as large a survival advantage in these subsites.
The sinonasal tract is unique because it may be at lower risk of exposure to HPV. It is hypothesized that oral HPV infection is transferred by oral sexual contact. 28 However, whether high-risk sexual behavior also affects cancers of the sinonasal tract is not known. A histological analysis of 131 sinonasal carcinomas found high-risk HPV DNA in 21% of tumors. 29 Of interest, although nonkeratinizing squamous cell carcinoma was found to be the most common histological type, the study also reported multiple tumors that were basaloid, papillary, and adenosquamous variants, and some contained features of a salivary gland neoplasm. This study, in combination with our results, suggests that sinonasal carcinomas confer distinct biological and clinical characteristics worthy of further investigation.
The prognostic role of HPV in oropharynx cancers is well established, 9 but a body of conflicting evidence regarding its role in other sites of the head and neck is now emerging. [30][31][32][33][34][35][36] Most studies of the prognostic role of HPV at each non-OPSCC subsite had small sample sizes and varied results. 37-39 Many trials have reported the strong association of p16 with improved progression survival, overall survival, and relapse-free survival in oral cavity, hypopharynx, and larynx cancers. 30-33 Some studies have grouped together all the associations by non-OPSCC subtypes, as opposed to delineating the associations by subsite. The results of such studies range from minimal change 34,35 to a substantial increase in survival. 36 Our data are supported by a recent study by Ko and colleagues 36 that examined the role of HPV at non-OP subsites (oral cavity, hypopharynx, and larynx). The investigators aggregated the 3 subsites and examined the 2 cohorts on the basis of disease staging (I and II; III and IV). Favorable prognosis was iden-tified in both groups for patients with HPV-positive status. However, the investigators did not specifically examine the role of HPV at each subsite by running a multivariate analysis for each subsite cohort, although KM studies were done by subsite. Our study more thoroughly examined the association of HPV because we isolated patient cohorts by subsite to determine the role of HPV. In this way, we were able to exclude the associations of interactions between the primary location of the tumor and HPV status with overall survival.
Our study and many others 30-33 have found an association between improved outcomes and HPV-positive non-OPSCC, but 2 studies offer evidence to the contrary. A recent 2-institution pooled analysis found no survival advantage for patients with larynx, oral cavity, and nasopharynx cancers. 34 Another study comparing advanced p16 and non-p16 tumors in the larynx with hypopharynx tumors demonstrated no outcome differences. 35 These conflicting results may be attributable to a difference in patient population (median age, sex distribution, race/ethnicity, and inclusion criteria) between the aforementioned studies and our own. In addition, the utilization of p16 as a surrogate for HPV status is another factor that differed from our study.
The role of HPV in the nasopharynx is still controversial. One study involving 90 patients (9 HPV-positive patients) found survival benefit with HPV-positive tumors, 40 but another recent study with 125 patients (13 HPV-positive patients) found no survival benefit with HPV-positive tumors. 34 One case series of 45 cases found that HPV-positive nasopharyngeal tumors may represent primary oropharyngeal tumors with extension to the nasopharynx site. 41 This is one of the largest studies examining the role of HPV in nasopharyngeal cancers. Although we found no survival benefit associated with HPV-positive nasopharyngeal tumor status, because of the retrospective and database-use design of our study, we were unable to determine the level of primary site misclassification. Historically, the role of the Epstein-Barr virus has been well characterized in the pathogenesis of nasopharynx tumors. The role of Epstein-Barr virus and its interaction with HPV were outside the scope of our study and not captured by the NCDB; data suggest, however, that Epstein-Barr virus-associated nasopharyngeal cancers have improved prognosis compared with virus-negative tumors, 42 which could confound the analysis of the HPV-negative status factor.

Limitations
The NCDB has well-documented limitations as a source of data. 43 We were limited by the variables captured by the NCDB, such as known risk factors including alcohol and tobacco use. We were also unable to determine the type of testing used for HPV status (eg, polymerase chain reaction, in situ hybridization for HPV DNA vs p16), as this may vary by each reporting institution and agency. Selection bias may exist in our data because routine HPV testing at non-OPSCC subsites is not the current standard of care. Furthermore, the source of the sample may not necessarily be the primary site. Our retrospective study focused on overall survival and not cancer-specific survival; thus, we were unable to distinguish between deaths due to head and neck cancer and deaths due to other causes. The rate of misclassification was likely low because of the nature of the registry of the data; however, any misclassification was likely to have been evenly distributed across the cohorts.

Conclusions
We identified a variance in the role of HPV and its association with outcomes in HNSCC. Although HPV-positive status was associated with improved disease survival outcomes for 4 subsites, it played the greatest prognostic role at the oropharynx and hypopharynx subsites. Human papillomavirus does not appear to affect the prognosis for the nasopharynx and sinonasal tract subsites. Given these results, we recommend routine testing for HPV status in HNSCC at the oropharynx, hypopharynx, oral cavity, and larynx subsites.