In this retrospective review, the authors examined demographic/clinical characteristics and overall survival in patients with squamous cell carcinoma of the oropharynx at a tertiary cancer center, and they report the characteristics that influenced any observed survival trends over time.
The study included 3891 newly diagnosed, previously untreated patients who presented at the authors' institution between 1955 and 2004.
Over time, patients presented at younger ages and were more likely to have base of tongue or tonsil tumors and to be never-smokers or former smokers. Patients who were diagnosed between 1995 and 2004 were almost half as likely to die as those who were diagnosed before 1995 (hazard ratio, 0.6; 95% confidence interval, 0.6-0.8). In both multivariable and recursive partitioning survival analyses, the TNM staging system predicted the survival of patients who received treatment before 1995 but did not predict the survival patients treated during the period from 1995 to 2004.
The incidence of cancer at most head and neck sites has been decreasing in the United States,1-6 but the incidence of squamous cell carcinoma of the oropharynx (SCCOP) has been increasing, especially among middle-aged white men.2, 4, 5 These trends also have been observed in Europe and Canada,7-10 and it is suspected that they are caused by a decrease in smoking prevalence11 and an increase in human papillomavirus (HPV) infections of the oropharynx.4
The strong association between SCCOP and HPV infection is now well established.12-14 Furthermore, in both the United States and Europe, the proportion of archived SCCOP tumor specimens that are positive for HPV DNA has increased from the 1970s to the present, suggesting an epidemic of HPV-associated SCCOP.7-10, 15
Within the National Cancer Data Base for the years 1985 to 2001, an increase in the use of chemoradiation for SCCOP was observed that suggested an acceptance of nonsurgical, multidisciplinary therapy for locoregionally advanced SCCOP.16 Over the same period, significant increases in 5-year survival rates for patients with SCCOP were reported,5 and patients with SCCOP now appear to have better 5-year survival rates than patients who have cancers of the oral cavity or larynx.17 In addition, most case series have demonstrated a favorable impact of HPV infection on the survival of patients with SCCOP.18-20 This has been confirmed in a meta-analysis21 as well as in population-based studies.15, 22 Finally, 1 phase 2 study and 4 phase 3 studies all reported improved overall and progression-free survival for HPV-positive patients.23-27
The first objective of the current study was to examine trends over time in demographic and clinical characteristics and in the overall survival of patients with newly diagnosed SCCOP at a tertiary cancer center to determine how our institutional experience compared with the current national trends. The second objective was to analyze and report the demographic and clinical characteristics (especially disease stage) that influenced the observed trends in overall survival.
MATERIALS AND METHODS
All patients with newly diagnosed, previously untreated SCCOP who presented to The University of Texas M. D. Anderson Cancer Center between 1955 and 2004 were included (N = 3891). For survival analysis, only patients who presented without evidence of distant metastases and who were subsequently treated at The University of Texas M. D. Anderson Cancer Center were included (N = 2299). Patients' medical records and the institutional tumor registry were reviewed for demographic characteristics, including smoking and alcohol exposures; clinical characteristics, including treatment; and overall survival. The sixth edition of the American Joint Committee on Cancer TNM staging system was used to determine disease stage at the time of presentation for all patients. A patient was considered a smoker if they had smoked at least 100 cigarettes in their lifetime and was considered a drinker if they had consumed at least 1 alcoholic beverage per week for at least 1 year. Patients who had not smoked or consumed alcohol in the year before their cancer diagnosis were considered former smokers and former drinkers, respectively. This retrospective review was approved by the Institutional Review Board. Because recurrence of SCCOP is highly correlated with subsequent death and in an effort to circumvent misclassifications and missing documentation of causes of death, we chose overall survival as the most reliable outcome.
Statistical analyses were carried out using the Stata statistical software package (version 11.0; Stata Corporation, College Station, Tex). A P value of .05 was used to determine statistical significance, and all tests were 2-sided. Categorical variables were created for age (young, aged ≤45 years; middle aged, ages 46-55 years; historically typical age at SCCOP diagnosis, ages 56-65 years; or old, aged >65 years), smoking status (never/former/current smoker), pack-years for smokers (1-19, 20-39, or ≥40 pack-years), alcohol drinking status (never/former/current drinker), and decade of diagnosis (1955-1964, 1965-1974, 1975-1984, 1985-1994, or 1995-2004). A nonparametric test for trend was used to detect significant trends over time.28 Chi-square tests and Fisher exact tests were used to detect statistically significant differences between groups. For the survival analyses, overall survival was defined as the time from first appointment to death from any cause. Kaplan-Meier curves and log-rank tests were used to detect statistically significant differences in survival. Cox proportional hazards models were generated to calculate hazard ratios to detect any differences in survival with respect to decade of diagnosis. We performed recursive partitioning analysis using the survival tree program STREE (Collaborative Center for Statistics in Science, Yale University, New Haven, Conn) to determine the most important factors affecting survival.29 Patients were divided into overall survival risk groups on the basis of median survival (Group 1 [low-risk]: median survival, >90 months; Group 2 [low-intermediate risk]: median survival, 60.1-90 months; Group 3 [high-intermediate risk]: median survival, 30.1-60 months; and Group 4 [high risk]: median survival, ≤30 months).
Trends Over Time in Demographic and Clinical Characteristics
The demographic and clinical characteristics of the 3891 patients with SCCOP who were diagnosed between 1955 and 2004 are presented by decade of diagnosis in Table 1. Over time, patients presented at significantly younger ages: The median age was 60 years during 1955 to 1994, and it was 55 years during 1995 to 2004 (P < .001; Ptrend < .001). Before 1995, less than one-third of patients were never-smokers or former smokers; whereas, during 1995 to 2004, almost two-thirds of patients were never-smokers or former smokers (Ptrend < .001). Furthermore, among smokers, the intensity of smoking also decreased over time (Ptrend < .001). No obvious changes in alcohol drinking patterns over time were observed. Before 1995, SCCOP cases were distributed relatively evenly between the tonsil, base of tongue, and other sites; whereas, in 1995 and later, over 85% of SCCOP cases were located in either the tonsil or the base of tongue. We also observed significant trends over time for the following: 1) lower T classification (Ptrend < .001), 2) higher lymph node status (Ptrend < .001), and 3) higher TNM stage (Ptrend < .001).
Table 1. Demographic Characteristics of All Patients With Incident Squamous Cell Carcinoma of the Oropharynx by Decade of Presentation, n = 3891
No. of Patients (%)
1955-1964, n = 473
1965-1974, n = 666
1975-1984, n = 657
1985-1994, n = 850
1995-2004, n = 1245
P for Trend
Pack-years for smokers
Alcohol drinking status
Base of tongue
Lymph node status
Survival Over Time
Overall survival of the 2299 patients with SCCOP who presented without distant metastases and received treatment at The University of Texas M. D. Anderson Cancer Center according to decade of diagnosis is presented in Figure 1. The most significantly improved survival was observed in the most recent decade (1995-2004 vs 1955-1994; P < .001).
On the basis of these findings, we chose to analyze overall survival of this cohort dichotomized by year of diagnosis (before and since 1995) for the following reasons: 1) Significantly improved overall survival was observed for patients who were treated during and after 1995 (Fig. 1), as stated above; 2) the increase in SCCOP incidence in the United States has been most pronounced since 19952; 3) the population trend of increasing prevalence of HPV infection in archived specimens of SCCOP has been most pronounced in patients who were diagnosed in 1995 and later7-10, 15; and 4) within our cohort, we observed that the most dramatic shifts in age at presentation, smoking status, tumor site distribution, and stage occurred since 1995.
Survival Over Time by Demographic and Clinical Characteristics
To better understand the factors that influence the survival gains observed in during 1995 to 2004, we focused on the subgroup of patients with SCCOP who did not have distant metastases at diagnosis and who were treated at our institution. We compared demographic and clinical characteristics between patients who were diagnosed during 1955 to 1994 and those who were diagnosed during 1995 to 2004, and we examined differences in overall survival between periods within various subgroups (Table 2). For patients who received treatment at our institution and remained alive, the median follow-up was 120 months (120 months for 1955-1994 and 101 months for 1995-2004).
Table 2. Demographic and Clinical Characteristics by Treatment Period and Relation to Overall Survival in Patients With Incident Squamous Cell Carcinoma of the Oropharynx Without Distant Metastases Who Received Treatment at The University of Texas M. D. Anderson Cancer Center, n = 2299
Treatment Period: No. (%)
Survival Analysis With the Earlier Cohort as the Referent Category
Similar to what we observed in the entire cohort, those diagnosed during 1995 to 2004 were more likely to be younger men who were never-smokers or former smokers than patients who were diagnosed before 1995 (Table 2). We observed that patients who were diagnosed during 1995 to 2004 were less likely to die within 10 years of their diagnosis regardless of age or sex than patients who were diagnosed during 1955 to 1994. White patients who were diagnosed during 1995 to 2004 had a 40% decreased risk of death compared with patients who were diagnosed earlier; however, patients of other ethnicities had no apparent survival advantage associated with more a recent diagnosis. Regardless of smoking or drinking status, survival improved over time. Never-smokers had the greatest improvement in survival, and smokers with the highest pack-years did not have a demonstrable improvement in survival.
Similar to what was observed in the entire cohort, patients who received treatment during 1995 to 2004 were more likely to have tonsil or base of tongue primary tumors and T1-T2, N1-N3 tumors (Table 2). Major shifts in the treatment of SCCOP also were observed: During 1995 to 2004 compared with earlier time periods, patients less frequently underwent surgery and more frequently received a combination of radiation and chemotherapy. Typically, patients with stage I and II disease either underwent surgery alone or received radiotherapy alone, whereas those with advanced stage disease were more likely to receive combined therapy (typically surgery plus radiation before 1995 and radiation plus concurrent chemotherapy after 1994). Although it was not a part of the current review, intensity-modulated radiotherapy was introduced at our institution in late 1999, but it did not become our standard practice until 2004.
The risk of death was decreased for patients who received treatment during 1995 to 2004 regardless of cancer site, and the risk was particularly decreased (40%) for those with tonsil or base of tongue cancers. No survival improvement over time was observed among patients who underwent surgery alone; however, among patients who received combined radiation and chemotherapy, the risk of death was reduced by 60% during 1995 to 2004 compared with the risk during 1955 to 1994.
Improved overall survival for patients who received treatment during 1995 to 2004 was observed in some staging subgroups, but not in others (Table 2). No clear improvements in overall survival were observed for patients with T4 tumors, patients without lymph node metastases (N0), or patients with the most advanced lymph node metastases (N3). The most dramatic observation was that, although there was no change in overall survival for patients with stage I and II SCCOP (the 5-year survival rates were 52% for patients diagnosed before 1995 and 56% for those diagnosed during 1995-2004; log-rank P = .718), the risk of death fell 50% for those with stage III and IV SCCOP (Table 2, Fig. 2). Among patients who were diagnosed before 1995 (Fig. 2, top), as expected, those with stage I and II disease had the best overall survival, and those with stage IV disease had the worst overall survival. However, among patients who were diagnosed in 1995 or later (Fig. 2, bottom), those with stage III disease fared significantly better than those with stage I and II disease (5-year survival rate, 68% vs 56%; log-rank P = .011), and those with stage IV disease fared slightly better than those with stage I and II disease (5-year survival rate, 61% vs 56%; log-rank P = .115).
Factors That Influenced Survival by Treatment Period
Patients who were diagnosed during 1995 to 2004 had a 40% decreased risk of death compared with patients who were diagnosed before 1995 (Table 2). To determine whether the factors affecting the overall survival of patients with SCCOP before and since 1995 were different, we created Cox proportional hazards models (Table 3). Treatment type did not appear to have a major impact on survival either before or since 1995. For the earlier cohort, the most important independent factors affecting survival were T classification, N classification, TNM stage, age, smoking status, and tumor site. It is noteworthy that, for the most recent decade, neither N classification nor TNM stage appeared to affect survival, although other factors common to HPV-associated SCCOP were associated with improved survival (younger age, white race, less smoking exposure, tonsil or base of tongue tumor site, and low T classification) (Table 3).
Table 3. Impact of Clinical and Pathologic Factors on Overall Survival by Period of Treatment, n = 2299
The referent for each variable is the second category in parentheses.
Estimates for each covariate were adjusted for all other covariates except pack-years and TNM stage.
Age (≥56 y vs <56 y)
Sex (men vs women)
Race(nonwhite vs white)
Smoking (ever-smoker vs never-smoker)
Pack-years of smoking (≥10 vs <10)
Alcohol (ever-drinker vs never-drinker)
Site(other vs tonsil or base of tongue)
Treatment (no surgery vs surgery)
Treatment (no chemotherapy vs chemotherapy)
Treatment (no radiation vs radiation)
Treatment (1 modality vs multimodality)
Tumor classification (T3-T4 vs T1-T2)
Lymph node status (N1-N3 vs N0)
TNM stage(III-IV vs I-II)
Recursive Partitioning Analysis
Recursive partitioning analysis identified different prognostic factors among patients who were diagnosed in the 2 time periods (Fig. 3, top). For patients who were diagnosed during 1955 to 1994, T and N classifications were the most important factors affecting survival (Fig. 3, top left). For patients who were diagnosed in 1995 and later, smoking status was most important. T classification, age, and tumor site also were important (Fig. 3, top right). Among smokers with T1 and T2 primary tumors (approximately one-third of patients diagnosed in 1995 and later) (Fig. 3, top right), a complete discrepancy regarding the intended prognostic value of the N classification was observed: Patients with lymph node metastases had a better prognosis than similar patients lacking lymph node metastases (log-rank P < .001). Although the results are not provided here, among smokers with advanced primary tumors (another 35% of patients) (Fig. 3, top right), patients with lymph node metastases also had a (nonsignificantly) better prognosis than patients without lymph node metastases. In summary, among patients who were diagnosed before 1995, the various stage categories appeared to be central to prognostication; whereas, among patients who were diagnosed in 1995 and later, variables that were likely to represent surrogates for HPV positivity (such as nonsmoking, T1-T2 tumors, and base of tongue or tonsil site)6 appeared to be central to prognostication. There was a very small subgroup of patients who were diagnosed before 1995 that had an excellent prognosis (Risk Group 1; N = 26) (Fig. 3, left), and the characteristics of this subgroup (never-smokers aged <56 years with small primary tumors) may suggest an early cohort of patients with HPV-associated SCCOP. We did not have HPV data available for the majority of patients in this cohort; however, of the patients in the 1995 to 2004 cohort for which we these data were available, 61 of 82 patients (74%) were positive for HPV by polymerase chain reaction (PCR) analysis.
For patients who were diagnosed before 1995, with the exception of Risk Group 1, survival stratification on the basis of recursive partitioning analysis (Fig. 3, left) was not substantially better than traditional TNM staging (Fig. 2, top). However, for patients who were diagnosed during 1995 to 2004, survival stratification on the basis of recursive partitioning analysis (Fig. 3, right) was much better than traditional TNM staging (Fig. 2, bottom). Compared with patients in the lowest risk group (Group 1), those in the low-intermediate risk group (Group 2) had 3 times the risk of death (hazard ratio, 3.2; 95% confidence interval, 2.2-4.7), those in the high-intermediate risk group (Group 3) had 6 times the risk of death (hazard ratio, 6.4; 95% confidence interval, 4.5-9.3), and those in the highest risk group (Group 4) had 10 times the risk of death (hazard ratio, 10.2; 95% confidence interval, 7.1-14.8). The difference in the risk of death was statistically significant between all risk groups (P < .001 between any 2 groups).
The current review of 1 institution's experience demonstrates clear changes over time in demographic and clinical characteristics of SCCOP that were particularly pronounced during 1995 and later. These findings are consistent with the decrease in the prevalence of smoking in the United States along with an increase in the proportion of HPV-positive SCCOP and Surveillance, Epidemiology, and End Results data indicating that the majority of patients with SCCOP are being diagnosed with late-stage disease—particularly disease associated with lymph node metastases.4-6
We observed significantly better overall survival among patients who were diagnosed with SCCOP in the most recent decade than among those who were diagnosed before 1995. This improvement in survival was observed in most subgroups, and the magnitude of the effect was greater in subgroups that were likely to be HPV positive and was less or zero in subgroups that were likely to be HPV negative. For instance, no such survival improvement was observed for nonwhite patients, heavy smokers, or patients with T4 primary tumors, N0 or N3 disease, or TNM stage I or II disease. In analyses of the factors that contributed to survival both before and since 1995, traditional staging variables accounted for most of the stratification in survival before 1995, whereas smoking history accounted for much of the stratification in survival since 1995. Because lower smoking exposures among patients with SCCOP can be a surrogate for HPV status, and because patients with HPV-associated SCCOP have improved survival outcomes, the current findings further suggest the emergence of an HPV-associated SCCOP cohort at our institution. This interpretation is further supported by the national pattern of increasing incidence of SCCOP coupled with falling mortality, a pattern indicative of a change in tumor biology or risk factors rather than improvements in the management of the disease.30
For the most recent decade, the current TNM staging system for SCCOP has not provided an effective prognostic stratification of patients, but smoking status and other surrogates for HPV status have. Unfortunately, the existing clinical trials in which HPV status was documented cannot be used to adequately assess TNM prognostic stratification in a recent cohort of patients with SCCOP, because enrollment was limited to patients with TNM stage III and IV disease, and patients who had N0 disease were either excluded and/or grouped together with patients who had with N1 and N2a disease.23-27 Furthermore, 2 of those trials included only relatively small subsets of patients with SCCOP within a mixed head and neck cancer trial, precluding detailed TNM stage or N classification analyses.23, 26 However, case series, population-based studies, and clinical trials all support HPV status as an important, independent prognostic factor for SCCOP.15, 18-20, 22, 23, 25-27 Additional literature supports smoking as an important prognostic factor,31, 32 and it was demonstrated recently that a combination of smoking and HPV status further refined the prognostic significance of HPV status: HPV-positive nonsmokers had an exceptionally good prognosis, HPV-negative patients had a poor prognosis, and HPV-positive ever-smokers had an intermediate prognosis.24 Consequently, we interpret the findings presented here and the existing literature as supporting the incorporation of HPV status and possibly smoking into the current TNM staging system.
Our study has several limitations. First, HPV status was not available for this retrospective cohort. However, we can confirm that, consistent with existing series,15, 18, 33 a high proportion of patients with SCCOP in the 1995 to 2004 cohort (74%; 61 of 82 patients) were associated with HPV type 16 by PCR. Furthermore, according to data from ongoing molecular epidemiologic studies through 2011, there is a decreasing proportion of HPV-positive tumors with worsening survival risk group as identified by recursive partitioning analysis (Fig. 3, right) (trend test: P = .003, P = .053, and P = .092 for p16 immunohistochemistry, HPV type 16 PCR, and HPV in situ hybridization, respectively). Second, comparisons of treatment in a retrospective review may be biased by factors that influence treatment assignments, although multivariable adjustment was performed to limit such biases. Third, we considered only all-cause mortality. Increased risk of comorbidities in smokers and older individuals may have biased the results, because the earlier cohort had a greater proportion of smokers and older patients. However, we observed improved outcomes in all age groups as well as when we adjusted the analysis for age. Fourth, improvements in imaging evaluation over time contribute to stage drift but do not account for the observed findings in multivariable or recursive partitioning analyses for the most recent decade. Finally, several prognostic variables, such as comorbidity, treatment breaks, surgical margin status, extracapsular spread, etc, were not available for inclusion in our analysis.
In conclusion, the survival of patients with SCCOP at our institution has improved markedly over the past 50 years, most dramatically since 1995. Although changes in treatment have occurred, they were not the major contributor to the observed improved survival. Instead, changes in clinical characteristics were the most important influence on survival and are likely a result of the changing etiology of SCCOP. The important prognostic factors accounting for the observed improvements in SCCOP survival over time, in part, are surrogates for HPV status and reflect an emerging epidemic of HPV-associated SCCOP. Our observations indicate that the current TNM staging system for oropharyngeal cancer has limited utility and encourage us to advocate for the incorporation of HPV status and perhaps smoking into the staging system for SCCOP.
We thank Stephanie Deming for editing the article.
This research was supported in part by The University of Texas M. D. Anderson Cancer Center's Support Grant CA016672 from the National Institutes of Health and by a cancer prevention fellowship for Kristina Dahlstrom supported by National Cancer Institute grant R25T CA57730 (Shine Chang, PhD, principal investigator).