A total of 256 patients (179 men and 77 women; aged from 26 to 75 years, with a mean of 55.4 years) with resectable stage III and IVA OSCC (T1-2N1-2M0 or T3-4N0-2M0) using the Tumor-Node-Metastasis staging system (20), who came from a prospective, randomized, phase 3 trial at Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine (Shanghai, China; registration ID: NCT01542931), were enrolled in the present study. The aim of the trial was to test the hypothesis that TPF induction chemotherapy administered prior to surgery and post-operative radiotherapy improves survival in patients with locally advanced OSCC. The experimental group received TPF induction chemotherapy followed by surgery and post-operative radiotherapy; the control group underwent surgery and post-operative radiotherapy; and there were 128 patients in each group. The detailed treatment protocol was as previously described (5). In patients assigned to the experimental group, the palpable edges of the primary lesion (both the longest and shortest axis) were marked prior induction chemotherapy by at least four points that were 0.5 cm away from the lesion. Chemotherapy consisted of docetaxel (75 mg/m²), cisplatin (75 mg/m²) and 5-fluorouracil (750 mg/m²/day) as a 120-h infusion for 5 days. Induction chemotherapy was administered every 3 weeks for 2 cycles. Surgery was performed ≥2 weeks after completion of induction chemotherapy. Radical resection of the primary lesion and full neck dissection with appropriate reconstruction was performed. The safety margins of the primary lesion were 1.5 cm away from the palpable margins; for patients who received induction chemotherapy, the safety margins were 1.0 cm away from the marks that had been placed prior to induction chemotherapy. Frozen sections of the surgical margins, including one or more of the anterior, posterior, upper and lower margins, which are at the mucosal surfaces and bottom margins, which are the deepest muscle layers, towards the muscle layers of the surgical bed, which is the remaining surface following tumor removal, were removed during the surgery. If cancer cells were identified in the margins, wider resection for at least 1.0 cm away from the positive margins was performed as possible as the surgeons can. Frozen section of the new surgical margins would be performed again to ensure the margins to be negative. Post-operative radiotherapy was initiated 4–6 weeks after surgery. Standard conformal or intensity-modulated radiotherapy was utilized, at a dose of 1.8–2 Gy/day, 5 days/week for 6 weeks, totaling 54–60 Gy. In patients with high-risk features (defined if two or more regional lymph nodes were involved, if there was an extracapsular spread of disease, or a microscopically involved mucosal margin of resection), a total dose of 66 Gy was recommended.

Formalin-fixed (at room temperature for 12–18 h for 5 µm sections) and paraffin-embedded biopsies were collected for immunohistochemical staining against GDF15. The methodology and assessment were as described previously (19). Following deparaffinization with xylene, the sections (5 µm) were rehydrated using an ethanol series of 100, 95, 85 and 5% ethanol, then distilled water. Prior to incubation with antibody solutions, the sections were heated by a water bath at 98°C with 0.01M citrate buffer solution (pH=6.0) for 20 min for antigen retrieval and cooled at room temperature, then washed with phosphate buffer solution 3 times for 5 min each. Then, the sections were incubated with the primary rabbit polyclonal antibody against GDF15 (cat no. ab82569; dilution, 1:100; Abcam, Cambridge, UK) overnight at 4°C and the secondary goat polyclonal to rabbit IgG antibody for 1 h at room temperature (cat no. ab150077; dilution, 1:3,000; Abcam), visualized using a 3,3′-diaminobenzidine detection kit (cat no. K067311; Dako; Agilent Technologies, Inc., Santa Clara, CA, USA). Positive staining for GDF15 expression was observed in the cytoplasm. Two pathologists performed blind examination using a light microscope at a magnification of ×200. The GDF15 expression level was determined using the immunoreactive score (IRS) system, including a proportion score (PS) and an intensity score (IS). The PS was calculated as the percentage ratio of positive GDF15-stained tumor cells to the total number of tumor cells, classified as follows: 0, 0%; 1, 1–10%; 2, 11–50%; 3, 51–80%; and 4, >80%. The IS was calculated as the staining intensity by visual assessment and was scored as follows: 0, negative; 1+, weak; 2+, moderate; and 3+, strong. The final GDF15 expression score was calculated using the PS and the IS (IRS=PSxIS), which ranged between 0 and 12. GDF15 expression was classified as low when IRS≤3 and as high when IRS≥4 (Fig. 1) (19,21).

Following initial treatment, patients were monitored every 3 months in the first 2 years, every 6 months in the subsequent 3–5 years, and once a year thereafter until mortality or data censoring. OS was counted from the date of random assignment to the date of mortality. Disease-free survival (DFS), locoregional recurrence-free survival (LRFS) and distant metastasis-free survival (DMFS) were counted from the date of random assignment to the date of recurrence, locoregional recurrence, distant metastasis or mortality, respectively.

For descriptive analysis, categorical data are expressed as the number and percentage. Survival analyses was conducted using the Kaplan-Meier method, followed by the log-rank test. Hazard ratios (HR) were calculated using the Cox proportional hazards model. Subgroup survival analyses according to the baseline characteristics (including the following subgroups: Sex, male and female; age, <60 years and ≥60 years; site, tongue and non-tongue; T stage, cT1/2N1-2M0 and cT3/4N0-2M0; N stage, cT3/T4N0M0 and cT1-4N1-2M0; clinical stage, clinical III and IV; pathological differentiation grade, well and moderately/poorly (22); smoke status, never smokers and current/former smokers; alcohol use, positive and negative use) were performed in patients with low or high GDF15 expression. All hypothesis-generating tests were two-sided and P<0.05 was considered to indicate a statistically significant difference. Data analyses were performed using the statistical software SPSS 18.0 for Windows (SPSS, Inc., Chicago, IL, USA).

Among the 256 patients, pretreatment biopsy samples were available from 230 patients for assessing GDF15 expression, including 104 patients in the experiment group and 126 patients in the control group. The difference in distribution of baseline characteristics, including sex, age, primary tumor site, clinical Tumor-Node-Metastasis stage or pathological differentiation grade between the patients with low and high GDF15 expression was not statistically significant (P>0.05) (19). At the time of data cut-off in December 2014, the median follow-up period was 67 months. Although patients in the experimental group had a slight survival advantage in OS, DFS, LRFS and DMFS, this difference was not significant (Fig. 2).

Among the 230 patients with OSCC with biopsy samples, low GDF15 expression was observed in 68 patients (31 in the experimental group and 37 in the control group) and high GDF15 expression in 162 patients (73 in the experimental group and 89 in the control group), with similar distribution of GDF15 expression between the two groups (P=0.942).

In the patients with low GDF15 expression, the 5-year OS, DFS, LRFS and DMFS rates were 73.4, 64.5, 66.0 and 73.4%, respectively, the locoregional recurrence rate was 27.9% and the distant metastasis rate was 5.9%. In the patients with high GDF15 expression, the 5-year OS, DFS, LRFS and DMFS rates were 57.7, 49.2, 51.5 and 56.6%, respectively, the locoregional recurrence rate was 39.5% and the distant metastasis rate was 9.3%

Patients with low GDF15 expression exhibited significantly better long-term outcomes, with regards to DFS (P=0.033), LRFS (P=0.043) and DMFS (P=0.038) rates, compared with those with high GDF15 expression. Although the difference in the OS rate was not significant, there was a trend towards a better OS rate (P=0.059) for patients with low GDF15 expression than those with high GDF15 expression (Fig. 3).

Analysis of the impact of baseline characteristics on the time-to-event end points was performed using a univariate Cox proportional hazards model according to the method used previously (19,23). Risk factors for OS, DFS, LRFS and DMFS included GDF15 expression (low vs. high), lymph node status (cN- vs. cN+) and clinical stage (stage III vs. stage IV). The risk factors for GDF15 expression and clinical stage were used in subsequent multivariate Cox model analysis, lymph node status was not included due to the direct association between clinical stage and lymph node status. GDF15 expression and clinical stage were independent risk factors for OS (P=0.022 and P<0.001), DFS (P=0.01 and P<0.001), LRFS (P=0.012 and P<0.001) and DMFS (P=0.014 and P<0.001) rates.

In order to analyze whether the patients with low or high GDF15 expression may benefit from TPF induction chemotherapy, survival analysis was performed to determine the difference in the outcomes of patients with different expression levels of GDF15 and with differing characteristics. In general, the patients did not benefit from TPF induction chemotherapy, neither those with low GDF15 expression nor those with high GDF15 expression (Fig. 4). Subgroup survival analysis according to baseline characteristics revealed that only the patients with high GDF15 expression and cN0 benefited from TPF induction chemotherapy with respect to OS (HR=0.233; 95% CI=0.068–0.795; P=0.02), DFS (HR=0.296; 95% CI=0.111–0.785; P=0.014), LRFS (HR=0.347; 95% CI=0.129–0.929; P=0.035) and DMFS (HR=0.212; 95% CI=0.062–0.72; P=0.013; Fig. 5); while the patients with low GDF15 expression and cN0 did not benefit from TPF induction chemotherapy. No significant difference was identified between any other subgroups.