We performed an observational study on a cohort of 68 patients diagnosed with OSCC who were surgically treated at the Maxillofacial Surgical Unit of the Santiago Teaching Hospital (Hospital Clinico Universitario de Santiago), Galicia, Spain, between January 2001 and March 2010. The inclusion criteria was established according to the surgical treatment in compliance with standard procedures, including resection of the primary tumor; radical, selective ipsilateral or bilateral removal of the regional lymph glands; clinical and pathological data and the availability of sufficient paraffin-embedded material to construct matrices. The clinical and pathological variables of each case included age, gender, tumor location, tumor stage, smoking habits, drinking habits, recurrence, dysplasia in the adjacent margin and vital status (death, by any cause) until February 2011.

The sample consisted of 35 men (51.5%) and 33 women (48.5%), with an age range from 41 to 96 years (average age, 67±13.08 years). Tumors were classified according to tumor stage at the time of diagnosis in accordance with the 7th Edition of the AJCC Cancer Staging Manual by the American Joint Committee on Cancer (19). Informed consent for use of the samples and data analysis were obtained from each patient or caretaker.

Hematoxylin and eosin-stained (H&E) slides were available for review in all cases. Paraffin blocks were selected on the basis of the availability of suitable formalin-fixed, paraffin-embedded tissue (at least 1-mm thick). For characterization of immunohistochemical protein expression, we constructed five different tissue arrays containing representative areas of every tumor. After microscopic evaluation, two areas of each tumor were selected, avoiding necrosis and keratin pools. Each tissue array was assembled as previously described (20). Briefly, two 1.5-mm-diameter cylinders of tissue were obtained from representative areas of each archival paraffin block and arrayed into a new recipient paraffin block with a custom-built precision instrument (Beecher Instruments, Silver Spring, MD, USA). These tissues were fixed in 4% buffered formalin and were paraffin-embedded according to routine procedures. Areas chosen for the cylinder core were representative of the tumors. In addition, normal tonsil samples were placed adjacent to the tumoral tissues to serve as internal controls and to ensure the quality of staining in the slides. Initial sections were stained with hematoxylin and eosin to verify histopathologic findings.

Tissue sections (3-μm) from the TMA blocks were sectioned and applied to special immunohistochemistry coated slides (Dako, Glostrup, Denmark). Immunohistochemical analysis for c-myc was performed using a monoclonal antibody (clone 9E11; Novocastra, Newcastle, UK) with a concentration 1:100, according to the manufacturer’s instructions. In brief, antigen retrieval was performed for 10 min at 95–99°C in a water bath, with a citrate buffer, pH 9.0. After blocking endogenous peroxidase activity, the slides were incubated for 30 min with the c-myc antibody. A secondary antibody reagent (polymer-based goat anti-mouse antibody fragment conjugated to horseradish peroxidase) was applied for 30 min. After applying the chromogenic visualization step using the 3,3′-diaminobenzidine chromogen, slides were counterstained with hematoxylin. Negative controls were performed using the negative reagent control (isotype control antibody). Dysplasia was graded as mild, moderate, severe and carcinoma ‘in situ’, according to the criteria of the WHO (21).

Cases showing cytoplasmic or nuclear positivity for c-myc were considered as positive. Slides were digitized using an automated slide scanner to produce high-resolution images for visual quantitative analysis using an ACIS^® III automatic image analysis system (Dako). The digital images of the slides were captured by the ACIS scanner at low power magnification, and the entire slide images were viewed on a monitor. The regions with the highest immune percentage of positive cells were selected for further automatic scoring. A minimum of three of these areas containing only tumor cells was selected manually for quantitative evaluation. The final score was the average result of the different areas measured. To confirm the accuracy of the measurements, selected areas from 10 of the specimens were measured three times.

Qualitative variables are expressed as frequencies and percentages; quantitative variables are expressed as means (standard deviation) and ranges. The χ² test or Fisher’s exact test was used as required to compare the qualitative variables. ANOVA or the Kruskal-Wallis test was used to contrast quantitative and qualitative variables. An association was considered statistically significant when the P-value was ≤0.05. The Kaplan-Meier method/estimator and the Cox regression model were used for evaluating survival in the study sample.

Initially, we carried out a univariate analysis in which we studied the Kaplan-Meier curves of each of the categorical/qualitative variables to assess if we detected differences in the survival between the different groups or levels of these variables. Subsequently, we adjusted the Cox regression univariate models to assess the influence of each of the analyzed variables on the survival prognosis.

A time-dependent multivariate Cox regression mode was fitted to jointly evaluate the possible risk factors in terms of survival. The recurrence of the patients was considered as the time-dependent covariate. The predictive multivariate analysis/model was performed using a stepwise procedure in terms of the best (the lowest) Akaike Information Criterion (AIC) value. The magnitude of the association between covariates and survival was evaluated through hazard ratios (HR), together with their corresponding 95% confidence intervals (CI). Taking censored survival times into account, the log hazard ratios of the Cox model were used as criterion variables Y, to construct time-dependent ROC curves (22).

All statistical analyses were conducted using SPSS Statistics 17.0 and R 2.15.0 (R Development Core Team, 2012), using the survival package (for fitting parametric Cox regression models), the survivalROC package (for computing time-dependent ROC curves) and the censboot function (from the bootstrap package) for bootstrapping survival models with censored observations. The function stepAIC (from the MASS package) was also used for obtaining the multivariate model with the best AIC. All these packages are freely available at http://www.R-project.org.

The main clinical characteristics of the 68 patients selected for the present study are documented in Table I. Tumors were classified into initial stage (stages I and II), which accounted for 32 cases (47.8%), and advanced stage (stages III and IV), which amounted to 36 cases (52.9%). More men were diagnosed in early tumor stages [21 (65.6%)], whereas women were found more frequently in advanced stages at the time of initial diagnosis [22 (61.1%)] (χ²=4.848, P<0.05).

In terms of the relationship with exposure to tobacco, no statistically significant differences in terms of staging at the time of diagnosis (χ²=1.227, P=0.541) were noted. In regards to alcohol, of the 22 cases exposed to the carcinogen, 16 (72.7%) were diagnosed in early stages (χ²=11.338, P<0.01).

The existence of dysplasia in the adjacent margin was positive in 42.6% of cases; 34.4% of the tumors in initial stages and 50% of the tumors in advanced stages (χ²=2.176, P=0.337). The existence of CIS was slightly higher in advanced stage tumors [9 (25%)] in contrast to 21.9% found in initial cases, although the differences were not statistically significant.

Immunohistochemistry showed both nuclear and cytoplasmic staining in neoplastic cells, while in many cases, the intensity was higher in the cytoplasm than that in the nuclei (Fig. 1).

The average expression of c-myc (N=53) was 50.32 (SD, 26.05) with a range from 6.60 to 99.48; similar for the initial stages [mean 50.77 (25.94)] and advanced stages [49.94 (26.59)] (F=0.013; P=0.909). Women showed higher levels [56.47 (25.58)] than men [44.40 (25.56)] although there were no statistically significant differences (F=2.951; P=0.877). The patients showed virtually similar values of c-myc in terms of the degree of tumor differentiation (Kruskal-Wallis χ²=0.062; P=0.970). Non-smoking patients had higher levels of c-myc [56.15 (5.00)], showing statistically significant differences (Kruskal-Wallis χ²=5.975; P=0.05). We found no statistically significant relationship between the quantitative expression of c-myc and any other clinical or pathological parameters (Table I).

The average follow-up was 33.5 months (CI, 28.23–40.38). Of the 36 (52.9%) patients who died, 24 (66.7%) were in the advanced stage group at the time of diagnosis (χ²=5.785, P<0.05). The average survival of the sample was 50.52 months (95% CI, 41.07–59.96); according to the Kaplan-Meier curve, the cases at the initial stages had a higher survival rate (40.72 months) than those in advanced stages (28.6 months) and the differences were statistically significant (P<0.01) (Fig. 2A). Recurrence occurred in 39 cases (57.4%) during the follow-up period, regardless of the tumor stage at the time of diagnosis (χ²=0.30, P=1.000). We detected statistically significant differences in the survival of patients with recurrence in contrast with those who did not relapse, showing a lower overall survival rate in the recurrence group (P<0.01) (Fig. 2B). Survival was also statistically significantly reduced in patients with any degree of dysplasia in the adjacent margin, in contrast with those who did not have dysplasia (P<0.05).

The Cox univariate regression analysis verified that the effect of the value of c-myc alone was not statistically significant (P=0.735). In a multivariate Cox analysis, however, the multivariate with the lower ACI obtained included the covariates: recurrence, c-myc, gender, location, differentiation, stage, size, alcohol consumption, dysplasia in the adjacent margin, interaction between c-myc and recurrence, and the interaction between gender and stage (ACI=155.0655). All of the variables were statistically significant, except for gender, stage and dysplasia in the adjacent margin. Gender and stage were not significant, but their relevant interactions were (Table II). Using this model, we observed that patients with recurrence had a 8303.37-fold higher risk than those without recurrence (P<0.01; HR, 8303.37; 95% CI, 34.30–2010000). For each unit of increase of c-myc, the risk increased by 1.15 (P<0.001; HR, 1.150; 95% CI, 1062–1245). Patients with tumors located in the gum presented a 278.86-fold higher risk (P<0.001; HR, 278.86; 95% CI, 13.76–5653), followed by tumors in the tongue, with a risk of 16.24 (P<0.05; HR, 16.24; 95% CI, 1.37–193) when compared to tumors located in the buccal mucosa. Patients with moderate differentiation presented a risk 8.87-fold higher (P<0.01; HR, 8.87; 95% CI, 2.01–39.13) than those with well-differentiated tumors. We found the same results for poorly differentiated tumors, which showed a risk 7.48-fold higher (P<0.05; HR, 7.48; 95% CI, 1.06–52.73). In patients who were former drinkers, the risk decreased by 0.02 (P<0.001; HR, 0.02; 95% CI, 0.02–0.19).

Temporary AUC showed an approximate mean value of 0.9 for the follow-up between 35 months and the end of the follow-up period, which indicates a good discrimination capacity of the Cox model (Fig. 3).