Abstract
Recently, log odds of positive lymph nodes (LODDS) was proven a better prediction of outcomes than other methods in gastric cancer, pancreatic cancer, and colon cancer. However, the validity is not yet tested in oral cavity squamous cell carcinoma (OSCC). We conducted a retrospective study to compare the predictive ability of LODDS, traditional pN classification and lymph node ratio (rN) in OSCC patients.In total, 347 OSCC patients receiving surgery with or without adjuvant therapy at the time of diagnosis between 2004 and 2013 were identified from the cancer registry database of the Dalin Tzu Chi Hospital. Cox proportional hazards models were used to compare the disease-specific survival (DSS) rates for pN, rN, and LODDS after adjusting for possible confounding risk factors. The discriminatory ability of different classification systems was evaluated using the adjusted hazard ratio and Akaike information criterion (AIC) by multivariate regression model. The prediction accuracy of the model was assessed by Harrell's c-statistic.The 347 OSCC patients had a mean age of 57 years old. Among them, 322 patients (92.8%) were male and 189 patients (54.5%) were in stages III to IV. LODDS showed better discriminatory ability for patients with <5 pathological cervical metastatic nodes and those with rN < 0.2. The hypothetical T-LODDS-M staging system had higher linear trend Chi-square, lower AIC, and higher prediction accuracy compared with the American Joint Committee on Cancer (AJCC) TNM, or hypothetical T-rN-M system. After adjusting for other factors, the LODDS unfavorable group had the highest adjusted hazard ratio (HR, 5.42; 95% confidence interval [CI], 3.19-9.12) and LODDS-based model lowest AIC of 704, comparing with pN and rN-based model. The LODDS-based system had the highest prediction accuracy for 3-year DSS (Harrell's c-statistic, 0.803).In our series, LODDS shows great promise as a prognostic tool for OSCC. Compared with the AJCC pN classification and the rN classification, LODDS can stratify OSCC patients and help to identify high-risk patients missed by the other systems.