Human Papillomavirus Infections are Common and Predict Mortality in a Retrospective Cohort Study of Taiwanese Patients With Oral Cavity Cancer

Ethics Statement

The local Institutional Review Board at Chang Gung Memorial Hospital approved the study (No. 99-0112B), which followed the tenets of the Helsinki Declaration. Written informed consent was obtained from each patient.

Participants

This study was designed as a retrospective research. In total, 1002consecutive Taiwanese OCC patients were enrolled between 2004 and 2011. The inclusion criteria were as follows: newly histology-proven OCC; previously untreated tumor scheduled for radical surgery (either with or without neck dissection); absence of suspected distant metastases detected by imaging; and willingness to undergo imaging-guided biopsy or exploratory surgery if necessary. The exclusion criteria included the refusal or inability to undergo radical surgery of OCC.

Procedures

All participants underwent an extensive presurgical evaluation.^12,15 Cancer staging was performed according to the 2002 American Joint Committee on Cancer (AJCC) 6th edition staging criteria.¹⁷ All patients underwent radical excision of the primary tumor with ≥1 cm gross safety margins. Patients with advanced-stage cancer and/or close margins ≤4 mm underwent adjuvant radiotherapy or concomitant chemoradiotherapy, as reported elsewhere.¹² The following variables were collected in all participants: sex, age at disease onset, alcohol drinking (ever/never), betel quid chewing (ever/never), cigarette smoking (ever/never), tumor subsite, differentiation, pathological T-status, pathological N-status, pathological stage, extracapsular spread (ECS), level IV/V metastases, treatment modality, and patient status at the last follow-up.

HPV Typing

Formalin-fixed, paraffin-embedded tumor samples collected during radical surgery were prepared for DNA extraction as described elsewhere.¹² DNA was extracted using a Lab Turbo 48 automatic nucleic acid extraction system and a Lab Turbo Virus Mini Kit LVN500 (Taigen, Taipei, Taiwan). We utilized a commercially available HPV L1 gene polymerase chain reaction assay (EasyChip HPV Blot genotyping assay, King Car Ltd, Ilan, Taiwan) to screen for HPV infections. The assay underwent strict quality-check procedures,^18,19 and it has been successfully utilized in previous OCC studies.^{11–13,15,16} A 192-bp DNA fragment was produced after amplification of HPV DNA using the MY11/biotinylated GP6+ L1 consensus primer system. Samples that were β-globin-negative were repeatedly extracted. Briefly, the resultant amplimers (15 μL) were hybridized to a nylon membrane that was coated with 39 HPV type-specific oligonucleotide probes.

Definitions of HPV Infections

The prevalence of genotype-specific HPV infections was reported both individually and in distinct groups (all types, oncogenic types, and nononcogenic types). The category “any HPV infection” was defined as a positive test result for 1 (single) or more than one (multiple) of the 39 distinct HPV genotypes included in the assay. Patients with both oncogenic and nononcogenic HPV types were classified as having an “oncogenic HPV infection”. Because more than half of oncogenic HPV infections are caused by HPV16,^6,10–14 oncogenic HPV infections were further divided into 2 subgroups: “HPV16 infections” and “other oncogenic HPV infections”.

Statistical Analysis

The main endpoint was OS (defined as the time period between primary surgery and death by any cause). Follow-up visits were continued until April 30, 2014. Patients without a documented event were censored at time of last follow-up. To investigate the temporal trends of HPV infections and their impact on 5-year OS, patients were divided into 2 cohorts according to calendar periods: “2004 cohort” (2004–2007; n = 466) and “2008 cohort” (2008–2011; n = 536). Continuous variables were categorized using published thresholds, laboratory norms, and quartiles. Data were compared using the Mann–Whitney U test and the chi-square test.

A prognostic scoring system was devised based on clinicopathological factors identified as significant for OS by multivariate analysis in the 2004 cohort. Internal validity was examined using the 2008 cohort. Survival curves were plotted using the Kaplan–Meier method and compared with the log-rank test. We used Cox proportional-hazard regression analysis with a bootstrap approach (200 runs) to identify the variables associated with survival and to estimate unadjusted and adjusted hazard ratios (HRs) and their corresponding 95% confidence intervals (95% CIs). Dichotomized variables were considered for further analysis only when they were of prognostic significance in univariate analysis. Correlations between variables were assessed with the Spearman correlation. All variables that showed significant association with OS on univariate analysis and variables of interest were consequently included in multivariate stepwise Cox proportional-hazard regression models were fitted using a backward selection procedure, with a significance level for retention of any given variable into the model of 0.05. Independent factors associated with OS in the final model were included in the prognostic index.

We assigned a weighted risk score to each factor based on the range of their corresponding HRs. The total risk score was then calculated by the sum of the ratings of individual factors. The discrimination ability of the developed models was evaluated using the Harrell c-statistic. Models are typically considered acceptable when c-values are between 0.7 and 0.8, excellent when between0.8 and 0.9, and outstanding when ≥0.9.²⁰ All calculations were performed using the SPSS 23.0 statistical package for Windows (SPSS Inc., Chicago, IL), the R software (Version 3.2.2; R Foundation for Statistical Computing, http://www.r-project.org/), and GraphPad Prism 6.0 for Windows (GraphPad Software, Inc., San Diego, CA). Two-tailed P values <0.05 were considered statistically significant.

Patient Characteristics and Clinical Outcomes

Sixty-four females and 938 males who were scheduled to receive primary surgery at our hospital were enrolled. In Taiwan, the incidence of OCC has been found to be markedly lower in females than in males (2.23/10⁶ vs 23.67/10⁶, respectively [2004]; 2.71/10⁶ vs 25.52/10⁶, respectively [2011]).^8,9 The discrepancy of tobacco and betel quid use may account for the different incidence of OCC between male Taiwanese and female Taiwanese.²¹ On the basis of the observed sex-related differences in the incidence of OCC in our country, it is not surprising that males outnumbered females in the current study. Table 1 depicts the general characteristics of the study participants. There were 723 (72%) alive patients who completed a minimum follow-up of 5 months (median: 59 mos, interquartile range [IQR]: 37–83 mos). A total of 279 (28%) patients died during the study period (median time to death: 15 mos, IQR: 9–30 mos). The 5-year OS rate in the entire study cohort was 72% (95% CI 70%–75%).

Prevalence and Temporal Trends of HPV Infections

The overall prevalence of any HPV infection in the study cohort was 19%. Notably, 2% of the participants had multiple HPV infections. No statistically significant differences in terms of baseline variables were found between HPV-positive (any HPV infection) and HPV-negative patients (Table 1). More individuals were infected with oncogenic HPV types (16%) than nononcogenic types (4%). The 3 most common genotypes (including single and multiple infections) were HPV16 (8%), HPV18 (5%), and HPV52 (2%) (Table 2). The prevalence of HPV infections in OCC patients showed a decreasing trend across calendar-years for all assays (Fig. 1). We observed a significantly lower prevalence of HPV infections in the 2008 cohort compared with the 2004 cohort (16% vs 23%; P = 0.011). Notably, oncogenic HPV infections were significantly less frequent (12% vs 21%; P < 0.001), whereas nononcogenic HPV types were significantly more common (5% vs 2%; P = 0.022). Similar decreases were observed for the prevalence of both HPV16 (6% vs 10%; P = 0.022) and HPV18 (3% vs 7%; P = 0.007) infections.

Impact of HPV Infections on Survival Rates

The 5-year OS rate was 68% in HPV-positive patients (n = 194) and 73% in HPV-negative patients (n = 808; P = 0.243). Moreover, OCC patients with HPV16 infections (n = 80; 5-year OS, 67%), other oncogenic HPV infections (n = 78; 5-year OS, 78%), or nononcogenic HPV infections (n = 36; 5-year OS, 81%) did not significantly differ from those without HPV infections in terms of OS (P = 0.341).

Prognostic Impact of HPV Infections According to Risky Oral Habits

We then analyzed the prognostic impact of HPV infections according to the presence of risky oral habits (alcohol drinking, betel quid chewing, or cigarette smoking). The study cohort was divided according to the presence or absence of risky oral habits. Interestingly, patients without risky oral habits and HPV infections (n = 12) showed a worse OS (49% vs 80%; P = 0.021) than those with no evidence of HPV infections (n = 56). In contrast, HPV infections were not associated with OS in patients with risky oral habits (HPV-positivity [n = 182] vs HPV-negativity [n = 752]: 70% vs 73%; P = 0.187).

Univariate and Multivariate Analyses

In univariate analysis, we identified the following clinicopathological factors as significantly associated with lower 5-year OS: histological differentiation, pathological tumor status, pathological lymph node status, pathological stage, histological tumor depth, ECS, and level IV/V metastases (2004 cohort; Table 3). No significant associations were found between OS and any HPV infection, HPV16 and other oncogenic HPV infections, HPV16 infection, and HPV18 infection. However, additional univariate analysis with stratified data revealed that, among patients with advanced OCC (pathological stage III/IV), the presence of HPV16 infection (n = 23) was associated with a significantly worse 5-year OS (n = 234; 34% vs 56% in those without HPV16 infection; P = 0.038). These results suggest that HPV16 infection was not a significant prognostic factor in univariate analysis because of confounding effects; therefore, HPV16 infection was entered into the multivariate regression model.

Correlation analyses between clinicopathological variables showed that all of the pathological factors were significantly correlated with each other, whereas HPV16 infection was not associated with any pathological factor. The results of multivariate analyses identified 4 adverse independent prognostic factors for OS (T2, T3, and T4 status; pathological N1 and N2 status; ECS; and HPV16 infection; Table 3).

In the 2008 cohort (Table 3), the same clinicopathological factors (histological differentiation, pathological tumor status, pathological lymph node status, pathological stage, histological tumor depth, ECS, and level IV/V metastases) were identified as being significantly associated with lower 5-year OS in univariate analyses. When these variables and HPV infections were entered as covariates into a multivariate regression model, pathological lymph node status (N1 and N2) and histological tumor depth (9–15 mm and 16–80 mm) were found to independently predict 5-year OS.

Prognostic Scoring System

For translational purposes, we further dichotomized continuous variables according to their optimal cut-off values(age >50 y; tumor depth >8 mm) in the 2004 cohort (model 1; Table 4). Multivariate analysis demonstrated that 4 factors were independent adverse prognostic predictors of 5-year OS: pathological T3/T4 status (HR 2.4 [95% CI 1.7–3.4]); pathological N1/N2 status (HR 2.9 [95% CI 1.8–4.5]); ECS (HR 1.9 [95% CI 1.2–3.0]); and HPV16 infection (HR 1.7 [95% CI 1.0–2.8]). In the 2008 cohort (model 2), 3 factors were independent adverse prognostic predictors for 5-year OS: pathological N1/N2 status (HR 2.4 [95% CI 1.5–3.9]); pathological stage III/IV (HR 2.0 [95% CI 1.0–4.1]); and histological tumor depth >8 mm (HR 2.1 [95% CI 1.3–3.5]). The weighting was based on a simple algorithm assigning the integer value of the corresponding HR to each factor (ie, 2 point for HR 1.5–2.4; 3 points for HR 2.5–3.4).

Specifically, a score of 2 was assigned in presence of pathological T3/T4 status, ECS, or HPV16 infection, and a score of 3 was given for pathological N1/N2 status (model 1). The resulting prognostic score (ranging from 0 to 9) was then applied to stratify the study cohort into 4 distinct prognostic groups, as follows: low (score 0 [40%], intermediate (score 1–3 [31%]), high (score 4–6 [14%]), and very high (score 7–9 [16%]; Table 5) risk. Patients with a low risk had an excellent 5-year OS (89%). However, the 5-year OS rates of patients with intermediate, high, and very high risk were 76%, 50%, and 25%, respectively (P < 0.005 for all comparisons). The c-index was 0.76, suggesting a satisfactory prediction performance.

A score of 2 was assigned in presence of each of the following risk factors: pathological N1/N2 status; pathological stage III/IV; and histological tumor depth >8 mm. The resulting prognostic score (ranging between 0 and 6) was applied to stratify the 2008 study cohort into 4 distinct prognostic groups, as follows: low risk (score 0 [30%]), intermediate risk (score 1–2 [18%]), high risk (score 3–4 [24%]), and very high risk (score 5–6 [28%]) (Table 5). Patients with a low risk had an excellent 5-year OS (95%). The 5-year OS rates of patients with intermediate, high, and very high risk were 83%, 74%, and 54%, respectively (P < 0.005 for all comparisons, the only exception being intermediate vs high [P = 0.152]). The c-index was 0.73 (indicating a satisfactory prediction performance of model 2).

The analysis of OS in the 2008 cohort also demonstrated the internal validity of the model 1 system for the main endpoint. The 5-year OS rates of the 4 risk groups were 92%, 77%, 57%, and 51%, respectively (P < 0.001; Table 5), with a c-statistic of 0.71. Subgroup analyses corroborated the discriminative strength of the prognostic scoring system for all comparisons, the only exception being the very-high–risk group compared with the high-risk group (P = 0.216). As the prevalence of HPV16 infection significantly decreased from 10% (2004 cohort) to 6% (2008 cohort), the proportion of HPV16 infections was similarly reduced from 19% to 10% in the very-high–risk group. The proportion of the very-high–risk group was also reduced from 16% to 13% in model 1, albeit not significantly so (P = 0.121 and 0.461, respectively). As long as risk scores were analogous, similar survival rates were observed. This phenomenon caused a decreased discriminative strength for the comparison between the very high and high-risk groups in the 2008 cohort. The internal validity of model 2 for the main outcome measure was investigated in the 2004 cohort. The 5-year OS rates of the 4 risk groups were 87%, 87%, 64%, and 36%, respectively (P < 0.001; c-statistic = 0.74; Table 5). Only the low and intermediate-risk groups did not differ from each other (P = 0.920). All other comparisons were statistically significant.

Comparison Between Prognostic Scoring System and Sixth Edition AJCC Stage System

The 5-year OS differed significantly according to the traditional AJCC pathological stage in the 2004 (Table 3) and 2008 (Table 5) cohorts, the only exception being a marginal difference between patients with a pathological stage of 2 (85% and 91%, respectively) and those with a pathological stage of 1 (94% and 94%, respectively; P = 0.053 and 0.68, respectively). The c-statistics for pathological stage were 0.72 for the 2004 cohort and 0.70 for the 2008 cohort, respectively. The Kaplan–Meier survival curves for OS according to the pathological stage and the model 1 and model 2 systems in the entire cohort are shown in Figure 2. In the entire study cohort, the c-statistics for pathological stage, model 1, and model 2 were 0.71, 0.73, and 0.72, respectively.