We thank Nancy Bennett for her editorial assistance and Maria Coleman and Dr. Pen-Yuan Chu for their assistance in collecting the data.
The impact of lymph node metastases on prognosis in patients with oral cavity squamous cell carcinoma (OSCC) has been well recognized. However, accurate stratification of risk for recurrence among patients with lymph node metastases is difficult based on the existing staging systems. In the current study, the utility of lymph node density (LND) was evaluated as an alternative method for predicting survival.
Three hundred eighty-six patients who underwent neck dissection were included. The median follow-up was 67 months. Five-year overall survival (OS), disease-specific survival (DSS), and locoregional failure (LRF) rates were calculated using the Kaplan-Meier method. LND (number of positive lymph nodes/total number of excised lymph nodes) and tumor-node-metastasis (TNM) staging variables were subjected to multivariate analysis.
Using the median (LND = 0.06) as the cutoff point, LND was found to be significantly associated with outcome. For patients with LND ≤0.06, the OS was 58 percent versus 28 percent for patients with LND >0.06 (P < .001). Similarly, the DSS for patients with LND ≤0.06 was 65 percent and was 34 percent for those with LND >0.06 (P < .001). On univariate analysis, pathologic T and N classification, extracapsular spread, and LND were found to be significant predictors of outcome (P < .001). However, on multivariate analysis, LND remained the only independent predictor of OS (P = .02; hazards ratio, 2.0), DSS (P = .02; hazards ratio, 2.3), and LRF (P = .005; hazards ratio, 4.1). LND was also found to be the only significant predictor of outcome in patients receiving adjuvant radiotherapy (P < .05). Within individual subgroups of pN1 or pN2 patients, LND reliably stratified patients according to their risk of failure (P < .05).
Squamous cell carcinoma of the oral cavity (OSCC) is one of the common malignant tumors of the head and neck worldwide. The management of OSCC is largely surgical and adjuvant treatment including radiotherapy or chemoradiation is used for patients with advanced stage tumors.1, 2 Because adjuvant therapy may induce severe toxic effects, a significant challenge is to find a reliable method for stratifying patients for the risk of tumor recurrence immediately after surgery.
The conventional tumor-node-metastasis (TNM) staging system quantifies lymph node disease according to the number, size, and laterality of positive cervical lymph nodes.3 This system factors the lymph node status into 6 categories: N0, N1, N2a, N2b, N2c, and N3. The presence of 1 or more positive lymph nodes is a significant predictor of poor outcome. However, modern studies using multivariate analysis report that, among patients with positive neck metastases, lymph node stage does not necessarily predict prognosis, especially after adjuvant radiotherapy.4-8
The extent of neck dissection, surgical technique, and the level of histopathologic scrutiny determine the degree to which the regional lymph nodes are examined for neck metastasis and, hence, the probability of identifying metastasis in lymph nodes at risk.9, 10 Therefore, these factors can be expected to determine the pN status (negative or positive) and the pN stage of the neck disease. Recently, lymph node density (LND) (the number of positive lymph nodes/total number of excised lymph nodes) has emerged as an alternative staging system for predicting survival after surgery for carcinoma of the bladder11 and esophagus.12 In this system, the ratio of positive lymph nodes to the total number of excised lymph nodes was found to be superior to conventional TNM lymph node staging in predicting survival.13, 14 Because limited lymph node dissection may result in pathologic understaging, LND attempts to compensate for this factor by recapitulating 2 pieces of information: the extent of cancer spread to the neck (number of positive lymph nodes) and the extent of surgical lymph node clearance or sampling (total number of lymph nodes removed during surgery). The purpose of this study was to evaluate the utility of LND as a potential prognostic predictor in patients with OSCC.
MATERIALS AND METHODS
Patients and Methods
Our study cohort included 386 patients treated with primary surgery, with or without adjuvant radiotherapy, between 1986 and 1996 for OSCC. During this period, adjuvant chemoradiotherapy was not yet used. Thus, this provides a relatively more uniform cohort of patients than could be expected from more recent years. All patients underwent a standardized modified radical neck dissection involving levels I to IV or I to V as described by the American Head and Neck Society.15 The type of neck dissection was prespecified in all patients before surgery. One hundred ninety-five patients (50.5 percent) died of their disease, 15 (3.9 percent) of them with distant metastases. Table 1 presents demographic and clinical data on these patients. The follow-up interval ranged from 4 to 184 months, with a median of 67 months.
Table 1. Patient Demographics
No. of Patients
+ indicates positive; TNM, tumor-node-metastasis.
Mean age, y
58 ± 14 (range, 14-88)
Floor of mouth
Surgery and adjuvant radiation
Type of neck dissection
Extent of neck dissection
Selective neck dissection
Modified radical neck dissection
Radical neck dissection
Bilateral neck dissection
Overall TNM stage
Follow-up of all patients, mo
Mean: 65 ± 49
Follow-up of N+ patients, mo
Mean: 51 ± 48
Lymph nodes were evaluated for metastasis by pathologists at the Memorial Sloan‒Kettering Cancer Institute. All specimens were re-analyzed and evaluated by a single pathologist (D.L.C.) who was blinded to the pathology report. Overall, 5877 lymph nodes were evaluated, 457 of which were positive.
Specimen dissection and tissue sampling of the primary tumor were performed in accordance with the current guidelines for the histopathologic assessment of head and neck cancer carcinoma.16 Neck dissection specimens were submitted en block with metal tags attached designating the levels. Lymph nodes were detected by palpation. All lymph nodes identified by the pathologist were submitted for analysis. Lymph nodes were defined as aggregates of encapsulated lymphoid tissue of any size, which had a peripheral sinus. Extracapsular spread (ECS) was defined as tumor extension beyond the lymph node capsule with a desmoplastic stromal response. Each lymph node was sectioned every 2 mm, put in a different cassette, and embedded in paraffin. Sectioning was performed at 200-μm intervals into the block. Lymph nodes with ECS were representatively sectioned. There were 95 patients with 173 lymph nodes that had evidence of ECS. Of those, 100 lymph nodes (58 percent) had microscopic extracapsular extension and 73 (42 percent) demonstrated macroscopic extension.3
Five-year overall survival (OS), disease-specific survival (DSS), and locoregional control rates were calculated using the Kaplan-Meier method, and the difference in survival rate was assessed by the log-rank test.17 OS was measured from the date of surgery to the date of death or last follow-up. For DSS, the patients who died from causes other than OSCC were censored at the time of death. The variables that had prognostic potential suggested by univariate analysis were subjected to multivariate analysis with the Cox proportional hazards regression model.18 All statistics were 2-sided (JMP; SAS Institute Inc, Cary, NC). A value of P < .05 was considered to indicate statistical significance. Variables used to stratify lymph node metastases included the total number of lymph nodes dissected, the number of positive lymph nodes, pN stage (pN0, pN1, pN2a, pN2b, pN2c, and pN3), ECS of tumor, and the LND. The sixth edition of the TNM staging system for OSCC was used for staging.3 LND distribution was evaluated as a continuous variable, and the active data plot was best fitted by 4 Gaussian equations using chi-square analysis (Microcal Origin; Microcal Software Inc, Northampton, Mass). For analysis of outcome, an LND cutoff point of ≤ or > the median distribution (LND = 0.06) was used. Other cutoff points tested were the intersection of the first and second Gaussian equations (LND = 0.05), the second and third Gaussian equations (LND = 0.076), or the third and fourth Gaussian equations (LND = 0.1). The 0.06 cutoff was selected because the results of exploratory analysis demonstrated no significant survival advantages over the other ratios, except for risk stratification among a subgroup of patients with pN2 disease and those undergoing therapeutic neck dissections, for whom a cutoff point of 0.1 was found to be a better predictor of outcome. Correlation analysis was performed using Pearson regression coefficient.
The study was approved by the institutional review board committee.
Kaplan-Meier estimates of 5-year OS and DSS rates were 61 percent and 70 percent, respectively. The management and outcome of the entire cohort of 386 patients included in the current study is summarized in Figure 1. On histopathologic examination, 167 (43 percent) of the patients had lymph node-positive disease. The 5-year OS rate for patients with pathologically negative neck lymph nodes was 76 percent and that for patients with positive lymph nodes was 42 percent (P < .0001). The 5-year DSS of patients with pathologically negative neck lymph nodes was 85 percent and that for patients with positive lymph nodes was 50 percent (P < .0001). Figure 2 shows the Kaplan-Meier curves of OS and DSS according to the lymph node (N) status.
Patients were further analyzed on the basis of the pathologic status of their lymph nodes at the time of surgery using the American Joint Committee on Cancer (AJCC) TNM classification system. There were 219 patients with pN0 disease (57 percent), 72 patients with pN1 disease (18.5 percent), 93 patients with pN2 disease (24 percent), and 2 patients with pN3 disease (0.5 percent). The group of patients with N-positive disease (n = 167) was analyzed to identify prognostic predictors that reliably stratify the risk for adverse outcome within this group. Among these patients, the number of lymph nodes removed ranged from 6 to 114 (mean, 35 ± 19 lymph nodes) and the number of positive lymph nodes was between 1 and 22 (mean, 2.7 ± 2.8 lymph nodes). LND was calculated as the ratio of positive lymph nodes to the total number of lymph nodes removed. The distribution of LND among the study population is shown in Figure 3A. The LND frequency distribution could be reliably fit by 4 Gaussian equations (P < .05). The median LND was 0.06. The remainders of the parameters for each of the 4 Gaussian equations are depicted in Figure 3A. The distribution of LND according to pN classification is shown in Figure 3B. In the pN1 group (n = 72), the mean LND was 0.05 ± 0.03 (range, 0.018-0.19); in the pN2 group (n = 93), the mean LND was 0.12 ± 0.09 (range, 0.01-0.5); and in the N3 group (n = 2), the mean LND was 0.19 ± 0.18 (range, 0.06-0.32). Statistical analysis demonstrated a significant difference in LND distribution between the different pN classification groups (P < .0001).
We next investigated whether an LND model can be used to predict patient outcome. On univariate analysis, pathologic T classification, pN classification, ECS, number of positive lymph nodes, and LND were found to be significant predictors of outcome (P < .001). The total number of excised lymph nodes was not found to be associated with survival. An increase in the pN classification was associated with decreased 5-year OS and DSS rates (Fig. 4A and 4B, respectively). Similar to pN classification, LND was also found to be significantly associated with 5-year OS and DSS (Figs. 4C-4F). For patients with LND ≤0.06, the 5-year OS rate was 58 percent, versus 28 percent for patients with LND >0.06 (P < .001), as demonstrated by the Kaplan-Meier curves. Similarly, the 5-year DSS rate was 65 percent for patients with LND ≤0.06, versus 34 percent for patients with LND >0.06 (P < .001). Based on the frequency distribution of LND, analysis was performed with the data set using an LND separation point of 0.05, 0.076, and 0.1. These analyses yielded similar results (P < .001).
To investigate whether LND predicts the prognosis of patients with positive cervical lymph nodes, we first created a multivariate model with all relevant variables except LND and then added LND to the model. The variables compared were T classification, pN classification, overall pathologic stage, ECS, total number of lymph nodes excised, and total number of positive cervical lymph nodes. On the first model (without LND), pN classification was found to be the only significant predictor of OS (P = .03) and DSS (P = .04). However when LND with a separation point of 0.06 (median value) was introduced to the multivariate model, the only significant predictor of outcome was LND (OS, P = .02; DSS, P = .02; and locoregional control, P = .005), whereas the traditional variables were not found to be independent predictors of survival (Tables 2-4). In all analyses (of OS, DSS, and locoregional control), the whole model proportional hazards fits (-log likelihood and P value) were better for the model including LND than the model without LND (Table 5). Similarly, when classification of the neck disease to N1, N2 and N3, was added to the model as an independent variable instead of pN classification, LND remained the only significant predictor of outcome. In addition, when patients with pN2 and pN3 disease were grouped together, LND remained the only significant independent predictor of outcome (P < .05). We also used 3 other separation points in this analysis, as suggested by the Gaussian function fit of the LND distribution: 0.05, 0.076, and 0.1. These analyses yielded similar results; however, the 0.1 cutoff did not reach statistical significance on multivariate analysis for DSS.
Table 2. Multivariate Analysis of Prognostic Factors for Disease-Specific Survival
Table 5. Multivariate Analysis of Prognostic Factors for Distant Metastases Disease-Free Survival
Cox Regression Model
LND indicates lymph node density.
To further assess the ability of LND to predict treatment response in a more homogeneous population and to account for the potential impact of adjuvant treatment, only those patients who receive postoperative radiotherapy were subjected to multivariate analysis (n = 154). Also in this subpopulation, LND was found to be the only independent predictor of outcome on multivariate analysis (P < .05).
The majority of the patients without clinical or radiologic evidence of neck metastases (N-) underwent selective neck dissection involving levels I to IV (80 percent). The remaining patients underwent comprehensive neck dissection. For patients with clinical evidence of lymph node metastases (N+), we performed a modified radical neck dissection involving levels I to V in 84 percent of the patients. Table 6 shows the different levels dissected in patients undergoing elective or therapeutic neck dissection. The multivariate analysis was repeated for elective neck dissections (clinically negative necks) and therapeutic neck dissections (clinically positive necks) separately. LND was found to be the most significant predictor of outcome for patients undergoing elective neck dissection (P < .004). For therapeutic neck dissections, none of the variables reached statistic significant when a LND of 0.06 was used as the cutoff point. However, when a separation point of 0.1 was used, LND remained the only significant predictor of outcome (P < .04). Table 6 shows an increase in the yield of positive lymph nodes and in the overall number of lymph nodes when therapeutic neck dissection was used compared with elective neck dissection (P < .0001), with no change in LND noted.
Table 6. Lymph Node Density in Elective and Therapeutic Neck Dissection Groups
Finally, individual pN subgroups (pN1 alone or pN2 alone) were stratified by LND at a cutoff of 0.06 using Kaplan-Meier analysis and the log-rank test. This analysis demonstrates the ability of LND to predict OS (P < .0002) and DSS (P < .001), even in the subgroup of patients with pN1 disease. Similar results were found for cutoff points of 0.05, 0.076, and 0.1 (P < .05). In the pN2 subgroup, the ability of LND to predict OS and DSS was significant only for a cutoff point of 0.1. The pN3 subgroup consisted of only 2 patients and therefore was not included in this type of analysis. Figure 5 shows the ability of LND to distinguish between low-risk and high-risk patients within individual pN classification subgroups.
Resection of the primary tumor with an appropriate neck dissection is considered the standard of care for patients with OSCC. Analysis of the patterns of failure in patients with oral cancers reveals that approximately 33% of them will fail due to regional metastases.7, 19-21 Some of the risk factors for recurrence include T classification, surgical margin status, depth of invasion, and major nerve invasion.22 In addition, 1 of the most significant prognostic factors in this population is the presence of neck metastasis.19, 23 Lymph node status and the number of positive lymph nodes are primarily based on the lymph node sampling procedure (ie, neck dissection) and secondarily on examination by the pathologist. It was shown that cervical metastases are more likely to be found for a lymph node yield >20.9, 10
Obviously, 1 of the weaknesses of the current study is inconsistency in the analysis of the pathologic specimens. In the current study cohort, the mean number of lymph nodes removed was 35, with a standard deviation of 19 (range, 6-114 lymph nodes). However, only in 35 patients, <20 lymph nodes were found and nearly all of them had undergone selective neck dissection. The previously reported mean lymph node yield in a unilateral radical neck dissection was 21 to 50.10, 24, 25 A study of the National Cancer Institute Surveillance, Epidemiology, and End Results (SEER) registry reported that the mean number of lymph nodes dissected in a neck dissection range from 1 to 97 lymph nodes per side.24 The variations in the number of lymph nodes retrieved from the specimens in the current study are, therefore, similar to other studies. Furthermore, even after we excluded cases with <20 lymph nodes from our analysis, LND remained the only significant independent predictor of outcome.
The value of the number of positive lymph nodes as a predictor of outcome was suggested by Mamelle et al23 for head and neck cancer patients. Recently, it was shown, for carcinoma of the bladder and esophagus, that LND is a superior predictor of outcome compared with conventional lymph node staging.11, 12, 14, 26, 27 Simple to calculate, LND is a ratio of the number of positive lymph nodes divided by the total number of lymph nodes examined by the pathologist. It was postulated that LND may have a greater prognostic value, because it takes into consideration 3 factors: 1) tumor factors (number of positive lymph nodes), 2) treatment factors (number of lymph nodes removed during neck dissection), and 3) staging factors (completeness of the sampling procedure, including those related to the surgeon and pathologist). This model proposes, for example, that a patient with 1 positive lymph node among 20 examined (LND = 0.05) has a better prognosis than a patient with 1 positive lymph node out of 5 excised lymph nodes (LND = 0.2). Although both patients will have the same pN classification, the latter patient is more likely to have positive lymph nodes left behind and therefore would be understaged by the conventional TNM system. Thus, the patient with a higher ratio is expected to fare worse than the patient with a lower ratio, although each has a similar number of positive lymph nodes examined.
In the current study, we have evaluated for the first time, to our knowledge, the value of LND in comparison with the conventional staging system to determine its ability to predict OS, DSS, and locoregional recurrence-free survival in patients undergoing neck dissection. Using multivariate analysis, we found that, in patients with positive cervical lymph nodes, LND is superior to the conventional N classification system (AJCC) in predicting OS, DSS, and locoregional control. Our data also indicated that LND is superior to T classification, ECS, overall stage, and number of positive lymph nodes in predicting survival. Within the subgroups of patients with pN1 or pN2 neck disease, LND reliably distinguished between low-risk and high-risk patients. Most importantly, multivariate analysis also demonstrated that LND is a better predictor than conventional N classification for predicting treatment failure in 3 groups of patients: those undergoing elective neck dissection, those undergoing therapeutic neck dissection, and those receiving adjuvant radiotherapy.
Recent studies have demonstrated slight improvements in 5-year survival rates after adjuvant concurrent chemoradiotherapy over radiotherapy alone for patients with advanced head and neck SCC.1 However, due to the significant morbidity of adding chemotherapy to radiotherapy, considerable controversy remains regarding the pathologic tumor characteristics that predict the need for more aggressive adjuvant treatment. Furthermore, a recent meta-analysis suggested that ECS and microscopically involved surgical margins, not pathologic N classification, may serve as predictors of outcome and, therefore, can potentially help determine the type of adjuvant treatment needed.8 The data from the current study indicate that LND may be useful as an adjunct to the conventional staging system in clinical studies investigating the role of adjuvant therapy after surgery for patients with OSCC.
The LND ratio introduces into the equation the expected variability in the extent of lymph node dissection and may be even more important when surgeons perform varying degrees of lymph node dissection. The value of LND was studied in our institution, in which a standard neck dissection is performed. For patients undergoing very limited neck dissections and with very poor lymph node yields, it is expected that LND will be higher than that reported in the current study. Whether LND will remain a significant predictor of outcome when different techniques or types of neck dissection are used awaits further evaluation.
Collinearity may exist when there are high correlations among sets of independent variables.28 It is also known that, when the collinearity is extreme (correlation coefficiency R >0.85), the numeric accuracy of the multivariate model can be affected.29 Therefore, we studied the correlation between LND and pN classification or the number of positive lymph nodes using regression analysis. In this analysis, the correlations were low (R = 0.4 between LND and pN classification and R = 0.6 between LND and number of positive lymph nodes). These results preclude profound effects on the multivariate model.29 We also found no significant correlation between the total number of excised lymph nodes (positive and negative) and the number of positive lymph node in the specimen (R = 0.3). This precludes the possibility of an inherent bias in the number of lymph nodes removed during surgery (ie, if the surgeon decides on the number of lymph nodes to be resected during surgery, when there are multiple positive lymph nodes).
Although the current study data provide a strong argument in favor of lymph node ratios to stratify risk of disease recurrence, other factors related to lymph node status such as the size and volume of the occupied lymph node, lymph node site, presence of occult micrometastases discovered by molecular methods, and extent of ECS may also be significant predictors of outcome and their interplay needs elaboration. Furthermore, the results of the current study represent a single institutional experience, with a relatively high median number of lymph nodes examined by our pathologist (N = 30). Thus, it does not necessarily comply with more limited dissections, in which a denominator (total number of excised lymph nodes) of <20 is common. The quality of the pathologic analysis may also affect the number of lymph nodes (negative or positive) identified after neck dissection. These variables lend a degree of subjectivity that is inherent to the concept of LND. Thus, until the value of LND is validated in multi-institutional studies, the conventional AJCC TNM staging system should remain in standard use.