Byrd JK, Wilhoit CST, Fordham MT, Reeves TD, McRackan TR, Nguyen SA, Sutkowski N, Gillespie MB. Predicting HPV Status in Head and Neck CancerThe Predictive Value of Sociodemographic and Disease Characteristics. Arch Otolaryngol Head Neck Surg. 2012;138(12):1155-1159. doi:10.1001/jamaoto.2013.850
Author Affiliations: Department of Otolaryngology, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania (Dr Byrd); Departments of Otolaryngology–Head and Neck Surgery (Ms Wilhoit and Drs Fordham, Reeves, Nguyen, and Gillespie) and Microbiology and Immunology (Dr Sutkowski), Medical University of South Carolina, Charleston; and Department of Otolaryngology, Vanderbilt University Medical Center, Nashville, Tennessee (Dr McRackan).
Objective To determine whether the human papillomavirus (HPV) status of head and neck squamous cell carcinomas (HNSCCs) can be reliably predicted based on sociodemographic and disease characteristics alone.
Design A retrospective medical chart review of clinical and pathologic features.
Setting Tertiary academic medical center.
Patients We studied patients treated for HNSCC who were tested for markers of HPV or had tissue available for testing between 2006 and 2010.
Main Outcome Measures Four otolaryngology–head and neck surgery trainees were given the database of patient clinical and pathologic features and asked to predict the HPV status for each patient. The trainees' responses were scored for accuracy, positive and negative predictive value, and interrater agreement. Multiple linear regression analyses were performed to determine predictors of HPV positivity.
Results A total of 174 patients meeting inclusion criteria were identified, 95 of whom were determined to have HPV-positive tumor tissue. Residents were able to accurately predict HPV status in 110 to 125 patients (63%-72%), with positive predictive values of 76% to 84% and negative predictive values of 61% to 69%. The only variables significantly related to HPV status were male sex (P = .01) and oropharyngeal subsite (P = .02). Only 4 patients had a “typical” HPV-positive profile.
Conclusions Knowledge of cancer stage, primary site, basaloid features, tumor differentiation, and presence of cystic neck disease and patient age, race, and smoking status did not allow accurate predictions of HPV status in many patients. Clinical testing of tumor tissue remains essential for a diagnosis of HPV-positive disease.
Over the past 25 years, human papillomavirus (HPV) has been increasingly associated with head and neck squamous cell carcinoma (HNSCC). Recent reports describe HPV-positive (HPV+) head and neck cancer as being distinctly different from HPV-negative (HPV−) disease with regard to its epidemiologic and molecular profile, and there are considerable data that associate HPV+ HNSCC with improved treatment response and survival.1 As a result, more clinicians are interested in the HPV status of their patients with head and neck cancer.
Much has been published in the literature strongly associating a number of sociodemographic factors with HPV+ HNSCC, most frequently young age, limited or no prior tobacco history, increased oral sex exposure, higher socioeconomic status, male sex, and white race.2 Many clinical and pathologic factors are also associated with HPV+ HNSCC, including oropharyngeal or unknown primary tumor; presence of cystic neck metastasis3; Rb downregulation and p16 overexpression4; and poor tumor differentiation, basaloid features, and nonkeratinizing cell types on histopathologic analysis.5
To our knowledge, there have been no studies reporting the HPV testing practices in the United States, but it is not routine for all specimens at our institution to be tested. Rather, testing is performed based on clinician or pathologist suspicion according to clinical or pathologic features present in each case. This practice raises the question of whether these features truly are sufficient to guide HPV testing or even to serve as a surrogate for laboratory testing. However, in 2008, D’Souza et al6 attempted to use a learning machine to accurately predict HPV status based on age, tobacco use, sex, race, and number of oral sexual partners, but they were unable to create a predictive model that was accurate enough for clinical use. We devised an experiment to see if a combining classic HPV-related sociodemographic factors with clinical and pathologic features would allow clinicians to accurately predict HPV status.
A review of a HPV research database and a surgical pathology clinical database was conducted to identify all patients with HNSCC seen at the Medical University of South Carolina between 2006 and 2010 whose tumor tissue had either been (1) tested for HPV or the HPV marker p16 at the time of diagnosis or (2) banked with signed consent for use in research. Only patients with oral cavity, oropharynx, hypopharynx, larynx, nasopharynx, and unknown primary tumors were included. Any patients without surgical pathology reports in their medical record were excluded.
Medical records of patients fitting the above criteria were entered into a database. If an actual HPV status was not available, p16 overexpression determined by immunoperoxidase staining was used as a surrogate for HPV status. In other patients who had signed consent forms but whose tissue samples had not undergone HPV in situ hybridization (ISH) or p16 ISH, tumor tissue was analyzed for HPV using reverse transcriptase–polymerase chain reaction (RT-PCR) analysis using E6 and E7 sense and anti-sense primers. Numbers of copies of genomic DNA per cell were determined based on quantitative RT-PCR dCt (delta cycle threshold) values between GAPDH and E6 or E7, and on the assumption of 2 GAPDH genes per cell.
The medical records of the patients included in this study were reviewed to abstract sociodemographic and clinical features of interest. Sociodemographic variables recorded included age, sex, race, history of tobacco use, and history of alcohol abuse. Age was characterized as either older than 55 years or 55 years or younger. Patients were classified as never smokers, previous smokers, or smokers at the time of diagnosis. Alcohol history was similarly classified as never drinkers, previous drinkers, or drinkers at the time diagnosis.
Clinical and pathologic features captured were primary tumor site; tumor differentiation (well differentiated, moderately differentiated, or poorly differentiated); tumor keratinization (nonkeratinizing vs keratinizing); visualization of grossly papillomatous disease; and the presence or absence of basaloid cell features. In addition to the final pathology report, the operative and radiology reports were investigated for the presence of cystic neck disease.
Three postgraduate year (PGY)-4 and one PGY-5 otolaryngology–head and neck surgery trainees familiar with HPV-related HNSSC were then presented with the database after patient de-identification and asked to predict HPV status of each patient. The residents had clinical experience with patients with head and neck cancer at high-volume tertiary care facilities and were educated about HPV-related HNSCC via didactics, journal clubs, and tumor boards.
All analyses were performed with Sigma Stat 3.5, SPSS 15.0, and Sample Power 2.0 (SPSS). Categorical variables are presented herein as percentages, and continuous variables are presented as means (SDs). All continuous variables were assessed for normality using the Kolmogorov-Smirnov test. Comparisons of outcomes were performed using the χ2 or Fisher exact test (categorical variables). Kappa analysis was used to determine interrater agreement among the residents. A finding of P < .05 was considered indicative of a statistically significant difference for all tests.
From reviewing multidisciplinary tumor board records, we determined that 1228 patients with head and neck cancer were seen at the Medical University of South Carolina, 388 of which had oropharyngeal primary tumors (31.6%). A total of 174 patients with tumors tested for HPV were identified (14.2%), and of these 95 were considered to be HPV+, and 79 were considered HPV−. Sixty-six patients were tested for p16 by immunoperoxidase staining, 31 for HPV by ISH, and the remaining 77 patients were tested for HPV by RT-PCR. The HPV results and p16 results were discordant in 2 patients. In these instances the p16 result was disregarded, and the HPV result was used for the determination of HPV status.
The patient characteristics captured in the database are listed in Table 1. On univariate analysis, the HPV+ and HPV− patient groups were significantly different with regard to race, sex, primary tumor site, T stage, N stage, tobacco use history, and cystic neck disease.
Given this data, the trainees were able to accurately predict HPV status in 63% to 72% of the cases (Table 2). Positive predictive values ranged from 76% to 84% (mean, 79.5%), and negative predictive values ranged from 56.4% to 68.0% (mean 62.3%). Interrater agreement by kappa analysis showed moderate strength of agreement among the 4 residents (kappa range, 0.330-0.561).
On multiple linear regression analysis, only sex (male) and site (oropharynx) were significant predictors of HPV status (Table 3). None of the HPV+ patients in this study met all the criteria that are commonly reported in the literature. Only 4 of 95 HPV+ patients met the “typical” profile: 55 years or younger, oropharynx primary tumor, never-smoker status, and the presence of cystic neck disease.
Differentiating between HPV+ and HPV− head and neck tumors has become of clinical interest for patient counseling, given the potential survival advantage of HPV+ disease.1 Furthermore, there are current proposals for future clinical trials that may ultimately lead to less toxic, de-escalated, or targeted therapies for HPV+ patients. As the body of literature describing HPV+ HNSCC has grown, clinicians have become more familiar with its commonly associated clinical features. An accurate clinical predictive model would serve to help clinicians decide which patients to test for HPV, or if accurate enough, even obviate the need for HPV testing.
Despite the addition of clinicopathological and demographic features, our results were very similar to those of D’Souza et al6 in that our attempts at predicting HPV status were moderately successful. The features of our study population did not match the profile of the typical patient with HPV+ HNSCC that has been promulgated, making their usefulness in predicting HPV status problematic for some patients. Although the typical profile is useful to describe broad features commonly found in the entire population of patients with HPV+ HNSCC, the profile is less often found to hold true for the individual patient, thereby lessening its clinical utility.
One of the most frequently described characteristics assigned to HPV+ HNSCC is young age; however, there is no true consensus on what age is a significant cutoff in this setting. The upper limit of the category in the Surveillance, Epidemiology and End Results database is 44 years,7 yet numerous articles describe “young” patients as anywhere from younger than 40 years to older than 60 years.8,9 With an age cutoff of greater or less than 55 years, our study failed to put a significantly different number of patients in the young category than in the older category. As it stands, using age to predict HPV status is difficult because many patients in both categories were outside any of the reported age ranges.
Another frequently reported characteristic of HPV+ disease is the presence of bulky cystic neck disease. In this study, based on cases where data were available, only 21 of 95 patients with HPV+ disease had cystic neck disease, while two-thirds of patients with HPV+ disease had noncystic neck disease. In addition, again based on cases where data were available, 25% of patients with HPV− had cystic neck disease. Our data therefore are discordant from the widely reported rates published by Goldenberg et al.3
Many sources list a negative tobacco history as a common trait of patients with HPV+ HNSCC.10 Although the number of smokers has decreased significantly in recent years, 28% of our patients with HPV+ disease were current tobacco users, and an additional 34% had used tobacco in the past. This point has also been illustrated in a study by Maxwell et al,11 who reported that only 32% of their patients with HPV+ disease had no history of tobacco use. The inherent imprecision of the “former tobacco user” category, however, is recognized, since tobacco use can range from light social use of only a few years' duration to heavy use over a many years. In addition, there is a very high incidence of endemic tobacco use in patients seen at our institution, and therefore the proportion of HPV+ disease is low. Predictions based on clinicopathologic data may be more accurate in areas with less tobacco use.
In our patient population, 13 of the 93 HPV+ cancers with known subsites were nonoropharyngeal cancers. Although oropharyngeal cancers are the head and neck cancers most commonly associated with HPV, larynx cancers were the first head and neck tumors reported to be HPV+ in 1981.12 Since that time, the focus has moved to the oropharynx owing to the high percentage of HPV+ oropharyngeal squamous cell carcinoma, which may approach 80%. It is becoming increasingly apparent, however, that there are clear cases of nasopharynx, larynx, and hypopharynx primary tumors that are HPV+.13- 15
A weakness of this study is the heterogeneity of the available data in our retrospective review. Not all data were available for all patients, and several different techniques were used to test for HPV. At our institution, p16 testing is the only test performed by the surgical pathology department; HPV ISH was performed at an outside facility if allowed by the individual patient's insurance plan, and HPV RT-PCR was performed on fresh frozen banked tumor specimens in a basic science research laboratory with validated HPV-testing experience. These findings emphasize our point that HPV testing should be performed routinely on all patients with HNSCC and that it should also be routinely covered by insurance plans.
The testing methods in our study introduce an additional testing bias. In our institution, testing is not routinely performed on all patients with HNSCC but is performed when clinical or pathologic features are not typical for tobacco-related cancer. Therefore, the patients tested at the time of biopsy or surgery may inherently have atypical features for HNSCC. However, since the clinical and sociodemographic profiles are shown to be somewhat deficient in their predictive capacity, it becomes evident that many patients may be missed without routine testing. Lack of knowledge of actual HPV status may result in deficient patient counseling with regard to risk factors, may confound clinical trial outcomes, and may result in a greater uncertainty with regard to treatment planning.
A potential criticism of our study is the use of residents as evaluators. However, all the residents were senior-level residents at tertiary care programs with high-volume cancer services. The residents received regular training on oropharyngeal carcinoma and HPV in the forms of lectures, journal clubs, and weekly multidisciplinary head and neck tumor boards. The sociodemographic and histopathologic features reviewed are well established in the literature, which the residents were familiar with, and therefore do not reflect the opinions and biases of their attending staff. On the contrary, the predictions of residents is more likely to be based on the recognized factors associated with HPV in the literature and less likely to be biased by having individually treated numerous patients with head and neck cancer. In this way, the prediction of the residents will be a better reflection of the otolaryngologist in practice who is familiar with the literature but may treat oropharyngeal cancer only sporadically. This study demonstrates to such individuals that tissue testing for HPV provides information that cannot be predicted on clinical, sociodemographic, and histopathologic factors alone.
If clinicians were to adopt a predictive model to replace routine laboratory testing, positive and negative predictive values should approach the gold standard of approximately 95%. The fact that residents were able to identify a significant percentage of patients correctly suggests that it may be possible to create a more accurate model in the future, as our knowledge of HPV-related cancer increases.
In conclusion, HPV+ HNSCCs are believed to represent a different clinical entity than HPV− HNSCCs. Although a typical patient profile has been popularized, predictions of HPV status based on clinicopathologic features are less accurate than laboratory testing. We therefore advocate routine HPV testing in all patients with head and neck cancer.
Correspondence: J. Kenneth Byrd, MD, Department of Otolaryngology, University of Pittsburgh Medical Center, 200 Lothrop St, Ste 500, Pittsburgh, PA 15213 (firstname.lastname@example.org).
Submitted for Publication: May 7, 2012; final revision received September 10, 2012; accepted September 18. 2012.
Author Contributions: Dr Gillespie had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Byrd, McRackan, Nguyen, Sutkowski, and Gillespie. Acquisition of data: Wilhoit, Fordham, Reeves, McRackan, Sutkowski, and Gillespie. Analysis and interpretation of data: Byrd, Wilhoit, Nguyen, and Gillespie. Drafting of the manuscript: Byrd, Wilhoit, McRackan, and Gillespie. Critical revision of the manuscript for important intellectual content: Fordham, Reeves, McRackan, Nguyen, Sutkowski, and Gillespie. Statistical analysis: Wilhoit, McRackan, Nguyen, and Gillespie. Obtained funding: Sutkowski and Gillespie. Administrative, technical, and material support: Wilhoit, Fordham, Reeves, and Gillespie. Study supervision: Byrd, McRackan, Sutkowski, and Gillespie.
Conflict of Interest Disclosures: None reported.
Previous Presentation: This research was presented as a poster presentation at the Combined Otolaryngology Spring Meeting; April 18-22, 2012; San Diego, California.
This article was corrected for errors on February 8, 2013.