Results
Patient Characteristics
The baseline patient clinical characteristics are described inTable 1 . The median age of the patient population was 58 years (ranging from 38 to 81 years). Most of the patients were male (162, 97.6%). Of these 166 patients, 66.9% had a history of smoking, and 57.2% had a history of drinking. There were 7 patients (4.2%) in pT1 stage, 14 patients (8.4%) in pT2 stage, 91 patients (54.8%) in pT3 stage, and 54 patients (37.9%) in pT4. Furthermore, 7.2% of patients did not have pathological nodal involvement, while others had (46 for pN1, 88 for pN2, and 20 for pN3). More than half of the patients (123 patients, 74.1%) were diagnosed as stage IV at pathological diagnosis. The study included 135 (81.3%) patients with pyriform sinus carcinoma, 6 (3.6%) with postcricoid region carcinoma, and 25 (15.1%) with posterior pharyngeal carcinoma.
Patients had a median number of 15 (range from 1 to 71) resected lymph nodes, a median PLN number of 2 (range from 0 to 14), a median LODDS of -0.845 (range from -1.92 to 0.73) and the median LNR of 0.1 (range from 0 to 1). SIRI median value was 1.46 (range from 0 to 11.2). Moreover, 9.6% of patients (n=16) had pathology suggestive of the involvement of the thyroid gland. Most patients (n=118, 71.1%) received adjuvant therapy postoperatively.
Nodal status and cutoff value
We used X-tile to define the optimal cut-off points for LNR and LODDS. We extracted the points with the highest Chi-squared value and the lowest P-value. Taking DFS as the dependent variable, two categories were obtained for the LODDS values: patients with LODDS values lesser than -0.91 (n = 69, 41.5%), and patients with LODDS values exceeding -0.91 (n = 97, 58.5%). Patients with LODDS ≥ -0.91 faced a 7.22 times higher risk of recurrence compared with those with LODDS <-0.91 (P < 0.001). The LNR values were divided into two categories using DFS as the dependent variable: patients with LNR values less than 0.11 (n = 91, 54.8%) and patients with LNR values more than 0.11 (n = 75, 45.2%). Considering the patients with LNR < 0.11 as the reference, patients with LNR ≥0.11 had a 6.63 times higher risk of recurrence (P <0.001).
Survival analysis
Using ROC curves, we compared the prognostic value of LODDS with that of LNR and pN. As shown in Fig. 1 and Table 2 the ROC curves corresponding to each parameter analyzed, taking OS as the dependent variable, pN displayed the highest area under the curve (AUC) (0.72, 95% CI: 0.64-0.8). While using DFS as the dependent variable, LNR demonstrated the highest AUC (0.712, 95% CI: 0.63-0.8). The results of the ROC analysis comparing the sensitivity and specificity of predicting survival indicated that LNR was the best predictor of DFS, outperforming pN and LODDS.
Using the Kaplan–Meier methodology, the estimated probability of DFS was found to be 78.7% at one year, 67.7% at three years, and 47.3% at five years. On the other hand, the estimated probability of OS was 80.7% at one year, 49.9% at three years, and 40.8% at five years. Regarding the treatment failure pattern, the rate of local failure was 8.4% (14/166), regional failure was 19.9% (33/166), combined local and regional failure was 7.8% (13/166), and distant failure was 20.5% (34/166) (Fig. 2, 3) .
In univariate analysis, the significant risk factors for OS were high pT classification (p=0.002), high LNR (p=0.031), margin (p<0.001), staging (p=0.047) and large tumor size (p=0.003) (Table 2). The significant risk factors for DFS were high SIRI value (p=0.019), high LNR (p<0.001), high LODDS (p<0.001), high pN classification (p=0.002), and staging (p=0.002) (Table 3 ). In multivariate analysis, we found that LNR, pT classification, and surgical margins were significantly correlated with OS (P < 0.05). As shown in Table 4 , LNR was significantly correlated with patient DFS (P = 0.001).