Prognostic Involvement of Lymph Node Density in Oral Squamous Cell Carcinoma. A New Predictive Model

Lymph node density (LND) has been reported to be a significant predictor of survival in oral squamous cell carcinoma (OSCC). The aim of this study was to analyze LND as a prognostic factor in OSCC and create a predictive model that determines the probability of death in these patients.

A retrospective cohort study was carried out with a cohort of patients who underwent cervical dissection and primary resection of OSCC between 1980 and 2020. The primary predictor variable in this study was LND, which is defined as the number of positive lymph nodes divided by the total number of lymph nodes removed. The cutoff values for prediction of disease-specific survival (DSS) were calculated by receiver operating characteristic curve analysis, which determined the best cutoff value was 0.07. Patients were divided into binary subgroups (low and high risk) using the best cutoff value of LND. The outcome variable was DSS, defined as the duration from the date of diagnosis to death due to OSCC and not due to other causes or secondary tumors. Other variables were type of neck dissection, postsurgical treatment, surgical margin, pathological T category stage, pathological N category (pN) stage, extranodal extension, perineural invasion, bone invasion, and presence of recurrence. A predictive model (score) was generated by selecting variables using a log-rank test and by using the Cox proportional-hazards regression (multivariate analysis).

The sample consisted of 368 patients, 252 (68.5%) male and 116 (31.5%) female patients, with a mean age of 60.3 years. According to the LND cutoff value, there were 289 patients with a low LND (≤0.07) and 79 with a high LND (>0.07). The univariate analysis showed LND as a significant predictor of DSS at 5 years (67.1% in LND ≤ 0.07 vs 32.9% in LND > 0.07; P < .001). The Cox multivariate analysis identified LND (hazard ratio [HR] = 27.2; 95% confidence interval [CI], 3.18-231; P = .002), recurrence (HR = 4.45; 95% CI, 2.3-8.4; P < .001), and type of treatment (HR = 0.52; 95% CI, 0.34-0.81; P < .001) as independent predictive factors for DSS. In the predictive model, the presence of recurrence was the most important factor with 8 points, whereas LND >0.07 contributed only 1 point; however, the 2 categories resulting from this limit were statistically significant.

Our study demonstrates that LND is an additional prognostic factor in patients with a pN+ disease. In addition, our predictive model could be useful in the therapeutic algorithm of OSCC patients, as it can predict the probability of death in these patients.