Lymph node ratio is associated with adverse clinicopathological features and is a crucial nodal parameter for oral and oropharyngeal cancer

The lymph node ratio(LNR) has been described as a novel predictor of the survival of patients with oral and oropharyngeal squamous cell carcinoma(O/OPSCC). The purpose of this study was to evaluate whether LNR is better at predicting survival and the need for adjuvant treatment than traditional tumour-nodal-metastasis(TNM) staging. Eight hundred nine patients with O/OPSCC and positive lymph node disease were retrospectively enrolled in this study. LNR equal to 0.075 is the best cut-off value for stratifying 5-year disease-free survival(DFS). High LNR is closely associated with more advanced T stage, higher N stage, more severe pathological grade, the presence of diffuse infiltration and extracapsular spread(ECS). LNR is better for evaluating prognosis than the pathological N stage. Patients with high LNR coupled with high number of positive lymph nodes who received adjuvant concurrent chemo-radiotherapy(CCRT) had a better 5-year DFS than patients who received surgery alone. Multivariate analyses revealed that T stage, ECS and LNR are independent prognostic factors of 5-year DFS and disease-specific survival(DSS). Therefore, high LNR is closely correlated with adverse parameters that markedly hinder prognosis. LNR is superior to traditional TNM staging for the evaluation of prognosis,and the combination of the LNR with the number of positive lymph nodes can predict the benefits of adjuvant CCRT.

is based on the simplistic concept of an orderly progression of the primary tumour located in surrounding tissues to lymphatic and vascular metastasis 9 . Due to the complex heterogeneity of O/OPSCC diseases, many other prognostic parameters should be investigated to acquire the best clinical evidence and to meet the demand for personalized diagnosis and treatment 13 .
Based on the results from our previous studies, the LNY alone is not a better predictive factor for long-term survival 14,15 . In addition, the prognostic value of other nodal parameters, such as the LNR, ECS and the number of positive nodes, in a northern Chinese population compared with the TNM stage remains unclear. The purpose of this study was to explore whether the LNR is a more valid predictive factor for treatment choice, tumour relapse and long-term survival than the TNM staging system and other parameters, including ECS and the number of positive nodes. In addition, if the LNR is a better predictive factor, we also sought to determine the best LNR cut-off value for classifying patients with nodal positive (pN+) O/OPSCC into high-or low-risk populations to aid surgeons in adopting personalized diagnoses and treatment strategies.
The 809 patients underwent 981 lateral neck dissections, and 172 patients underwent bilateral neck dissection. The largest proportion of patients was treated with comprehensive neck dissection (n = 452, 55.9%), followed by selective neck dissection (n = 357, 44.1%). The lymph node status of 50.7% of patients was pN2b, followed by 42.0% of patients with pN1, and 7.3% of patients with pN2c. The mean and median LNY values in each lateral lymph node were 27.3 and 25.0, respectively (range: . The mean number of positive lymph nodes was 2.8 (range: 1-55). In pN + patients, the majority of cases were single lymph node metastasis (43.4%), followed by two lymph node metastases (21.1%). The mean and range of the LNR were 10% and 1.5-77.4%, respectively. Three hundred eighty-two successive cases were available for the analysis of the extracapsular spread (ECS) status. The ECS status was positive in 152 cases and negative in 230 cases. One hundred eleven (13.7%) patients received surgery alone, 540 (66.7%) patients received adjuvant radiotherapy, 114 (14.2%) patients received concurrent chemo-radiotherapy, and data for 44 (5.4%) patients were missing (Table 1).

Cut-off values for nodal parameters.
The patients were divided into high-or low-risk subgroups using the best cut-off values for the nodal parameters. For continuous variables (including the LNR and the number of positive lymph nodes), the cut-off values for predicting 5-year disease-free survival (DFS) were first determined based on an analysis of the receiver operating characteristic (ROC) curves. Subsequently, all patients were allocated into the high or low subgroups according to the cut-off value (the continuous variables were replaced with categorical variables). The differences in 5-year DFS between the high-risk and low-risk subgroups were then analysed using Kaplan-Meier method and log-rank tests, as described by Liao et al. 7 . In addition, the cut-off values for categorical variables (including ECS and the traditional N stage) were determined based on the ECS status (ECS − or ECS + ) and N stage (N1, N2b and N2c). According to the analysis of the ROC curves, the best cut-off values for the LNR and the number of positive lymph nodes were 0.075 (area under the curve (AUC): 0.592, sensitivity: 58.0%, specificity: 58.4%, P = 0.0001, Fig. 1A) and 2 (AUC: 0.593, sensitivity: 42.9%, specificity: 73.3%, P = 0.0001, Fig. 1B), respectively. All patients were divided into different subgroups according to their cut-off values. Patients with primary tumours located in the oral cavity have a greater proportion of a high LNR than patients with primary tumours located in the oropharynx subsite (P < 0.001, Table 1).

Subsite analysis.
According to the Kaplan-Meier analysis, patients with stage III oral cancer have a similar 5-year disease-specific survival (DSS) compared with patients with oropharyngeal cancer (70.7% vs. 73%, P = 0.755). In addition, patients with stage IV oral cancer have a similar 5-year DSS compared with patients with oropharyngeal cancer (50.2% vs. 40.4%, P = 0.121). In this study, only 13 cases of all oropharyngeal cancers were analysed for p16 (INK4A) expression through immunohistochemistry and human papillomavirus 16/18 (HPV16/18) expression using a specific in situ hybridization protocol. Only 2 cases were identified as HPV-positive, and the remaining 11 cases were HPV-negative.
A high LNR is closely associated with many poor prognostic factors for O/OPSCC. According to the chi-square test, a high LNR was closely associated with a more advanced T stage (P < 0.001), a higher N stage (P < 0.001), a more severe pathological grade (P = 0.005), and the presence of diffuse infiltration (P = 0.007) and ECS (P < 0.001). Therefore, in this study, patients with O/OPSCC and a high LNR underwent more adjuvant RT or CCRT than those with a low LNR (P < 0.001).         Interestingly, patients with both a high LNR and a high number of positive lymph nodes (LNR > 0.075 and number of positive lymph nodes > 2) who received surgery and CCRT had a markedly better 5-year DFS than patients who received surgery alone (49.0% vs. 28.6%, P = 0.036). In comparison, patients who received surgery and RT had a similar 5-year DFS to patients who received surgery alone (32.0% vs. 28.6%, P = 0.086; Fig. 5A). Moreover, based on a further analysis of the ROC curves, the combination of the LNR and the number of positive lymph nodes has a better prognostic value for 5-year DFS than the LNR alone (AUC: 0.600, sensitivity: 51.4%, specificity: 68.7%, P = 0.0001, Fig. 5B).

LNR and ECS
Cox multivariate regression analysis. According to the results of the univariate analyses, the T stage

Discussion
Although lymph node metastasis is the most important prognostic factor for O/OPSCC, the precise risk stratification of pN + patients using traditional TNM stages is inadequate 16 . Based on the results from our previous studies, the traditional TNM stage might not be the best prognostic factor in the presence of ECS, adjuvant treatments and even some important biomarkers 17,18 . The LNR has recently been described as a potential predictor of survival and the need for adjuvant treatment in patients with O/OPSCC. However, its predictive value compared with ECS, the number of positive lymph nodes and N stage requires further testing 10,12 . In this study, we first determined the best cut-off value for the LNR. In addition, we then compared the values among different nodal parameters, including LNR, N stage, ECS and the number of positive lymph nodes, in predicting DFS, DSS and treatment choice. Based on our results, an LNR equal to 0.075 was the best cut-off value for dividing pN + patients with O/ OPSCC into low-and high-risk subgroups according to the 5-year DFS. The use of an LNR of 0.075 as the cut-off value was similar to the results reported by Ebrahimi et al. 19 and Patel et al. 20 . In this study, we did not observe a significant difference in the LNR between patients who received selective neck dissection (SND) and comprehensive neck dissection (CND). Therefore, bias from different surgical neck dissection methods was excluded.
A high LNR was closely associated with many adverse clinicopathological factors, including advanced T and N stages, a severe pathological grade, and the presence of diffuse infiltration and ECS. These adverse factors have been widely accepted as negative prognostic parameters and poor pathological characteristics. Although the predictive value of the LNR has never been considered when scheduling treatment strategies, most patients with O/OPSCC and a high LNR underwent more adjuvant RT or CCRT after the primary tumours were resected. Unfortunately, patients with a high LNR still exhibited very poor 5-year DFS (approximately 14% decrease) and DSS (approximately 20% decrease) compared with patients with a low LNR.
In this study, the traditional N stage did not serve as an independent predictive factor for 5-year DFS and DSS. Moreover, no significant differences in 5-year DFS and DSS were observed between pN2b and pN2c patients. Thus, we were unable to easily perform further risk stratification for patients with positive lymph nodal diseases. Based on the results of the comparison analysis, the LNR and ECS have stronger prognostic value than the traditional N stage, according to Cox proportional hazards regression models. If patients were divided into LNR subgroups, the pN stage (pN1 or pN2) would lose its efficacy to accurately predict patient prognoses. Thus, the LNR might be used as an alternative staging system because it is superior to the TNM pathological nodal staging system for predicting recurrence risk and survival after surgery, corroborating the conclusions reported by Ebrahimi et al. 21 .
The importance of the ECS status in determining the prognosis of patients with O/OPSCC has been widely accepted 22 . In this study, the LNR influenced the ECS prognosis (ECS − or ECS) of patients, and conversely, the ECS status was used to classify patients with or without a high LNR. In this study, the LNR was closely associated with the ECS status, and a subsequent multivariate analysis also revealed that both LNR and ECS were independent predictive factors for 5-year DFS and DSS. Therefore, we speculate that the LNR and ECS have similar predictive values in patients with O/OPSCC. Many published reports have verified that patients with a high LNR could benefit more from adjuvant CCRT than from RT or surgery alone for the treatment of solid tumours 10,23 . In this study, patients with a high LNR had received more rounds of postoperative adjuvant RT or CCRT; however, these patients still exhibited a worse prognosis than patients with a low LNR. Interestingly, patients with a high LNR who underwent adjuvant RT or CCRT had a better long-term survival than patients who underwent surgery alone. LNR alone not only stratified high-risk patients according to prognosis but also identified patients who may benefit from adjuvant treatment. When determining the best adjuvant treatment strategy for only patients with a high LNR, CCRT was not superior to RT because only a 5% improvement in 5-year DSS was observed. Amazingly, when the LNR and number of positive lymph nodes were combined, patients with an LNR > 0.075 and more than 2 positive lymph nodes and who underwent only adjuvant CCRT exhibited improvements in 5-year DFS. The conclusion was meaningful because the National Comprehensive Cancer Network (NCCN) only describes the presence of ECS and/or positive margins in head and neck cancer as an indication for CCRT. A further prospective clinical trial analysing whether the combination of the LNR and the number of positive lymph nodes can be used to identify whether the benefits of adjuvant CCRT will be helpful for validating this conclusion.
In this study, the clinical T stage coupled with the LNR and ECS were independent prognostic factors for pN + patients with O/OPSCC. The T stage has a role in determining prognosis because all patients were selected based on positive lymph node status. Notably, the LNR status is positively correlated with the T stage, further indicating that a high LNR is a high-risk prognostic parameter.
The incidence of HPV-related oropharynx carcinoma is continuously increasing worldwide, as approximately 60% of US patients with oropharyngeal cancer are HPV-positive. HPV-positive patients are now recognized as a distinct subgroup of patients with oropharyngeal carcinoma and are associated with an improved response to treatment and better prognosis than HPV-negative patients 24 . However, our previous studies have shown that HPV-positive oropharyngeal carcinoma is rare in China (less than 20%) 25,26 . In this study, tumours located in the oropharynx have a lower LNR than tumours located in the oral cavity. According to the results of the Kaplan-Meier analysis, oropharyngeal cancer has a similar prognosis as oral cancer. The results were not consistent with reports from Europe and the USA, where most cases of oropharyngeal cancers are related to HPV infection and often have a later N stage and better prognosis. The main cause for this difference was that most Chinese patients with oropharyngeal cancer in the study sample were HPV-negative. Therefore, oropharynx and oral squamous cell carcinoma cases in China were pooled in this study, and the results should be more reasonable than the results from Europe and the USA.
This retrospective study has inherent limitations. Data were not available for some important baseline factors, including the number of ECS, the depth of tumour invasion and tumour thickness. Another limitation was that SCieNtifiC REPORTS | 7: 6708 | DOI:10.1038/s41598-017-07134-7 the HPV status was not analysed in the majority of cases of oropharyngeal cancers and the choice of adjuvant therapy did not refer to the HPV status. These limitations will be considered in future studies. In this study, we excluded patients with an LNY of less than 10 to reduce the statistical bias from unqualified cases, as reported by Prabhu et al. 10 . Furthermore, patients with pN3 disease were also excluded from this study because of the presence of uncountable matted lymph nodes, as we previously reported 14 . Based on these considerations, the design of our study was rigorous.
Cancer is a complex disease, and its development is driven by inherent oncological molecular features. In the future, the identification of crucial gene signatures will represent a greater advantage in personalized medicine compared with traditional clinicopathological factors 27 . Based on results from our recent studies, several potential biomarkers, such as cyclin D1, enhancer of zeste homologue 2 and glycerol-3-phosphate dehydrogenase 1-like gene, can assist in guiding adjuvant therapy decisions and stratifying high-risk prognostic populations and might have a more accurate efficiency than clinical the TNM stage 18,25,28,29 . The concept of precision medicine based on translational research and novel pathological features is the most promising breakthrough that will further improve the quality of life and long-term survival of patients with head and neck cancer.
In conclusion, an LNR equal to 0.075 is the best cut-off value for predicting 5-year DFS. A high LNR is closely correlated with adverse parameters that markedly hinder favourable prognoses. The LNR and ECS are superior to traditional TNM staging, and the combination of the LNR with the number of positive lymph nodes predicts the benefits of adjuvant CCRT and improves the power of prognostic judgement.

Patients and Methods
Patients. This study was conducted in full compliance with ethical principles, including the World Medical Association Declaration of Helsinki (2002 version), and with the approval of the Institutional Review Board of the Beijing Stomatological Hospital of Capital Medical University. Due to the retrospective nature of this study, exemption was granted for obtaining written informed consent from the subjects. The investigators designed and implemented a retrospective cohort study to address the research aim. The study population comprised all patients who were treated in the Department of Oral and Maxillofacial-Head and Neck Oncology, Beijing Stomatological Hospital, Capital Medical University, and were pathologically diagnosed with O/OPSCC between January 2000 and August 2015. For inclusion in the study sample, the patients must fulfil the following criteria: (1) the patients underwent neck dissection (level I-III, level I-IV or level I-V) with pathological lymph node metastasis; (2) the LNY and the number of positive lymph nodes were available; (3) primary tumours were located in the tongue, lower or upper gingiva, buccal mucosa, floor of the mouth, oropharynx and hard palate; and (4) there was no evidence of distant metastasis. Patients were excluded if they had received preoperative chemotherapy/radiotherapy, had unresectable disease, had an LNY less than 10, or has a lack of adequate information to determine the LNY 14 . Patients with clinical N3 disease were also excluded due to the difficulty in distinguishing a largely single metastatic lymph node from uncountable matted lymph nodes, as reported in our recently published article 15 . Variables. The predictor variable was the LNR, which was calculated by dividing the number of positive lymph nodes by the number of lymph nodes on the side of the neck characterized by metastatic disease, regardless of whether the patients had undergone bilateral neck dissection. The parameters used to assess the main outcomes were 5-year DFS and 5-year DSS. DFS was calculated as the length of time from diagnosis until the first documented recurrence or death. DSS was calculated as the time from the first operation to the time of death or last follow-up; patients who died from causes other than OSCC were defined as survivors at the time of death. Other variables included demographic (age and sex), anatomical (T stage, growth pattern, and sites), habitual (tobacco and alcohol use) and pathological (LNY, number of positive lymph nodes, pN status, ECS and grade) variables.
Treatment and neck specimens. All patients were initially treated with surgery. The surgical procedure was selected by the surgeon according to the tumour site and local practice. Standard surgery, including radical tumour resection, neck dissection and the reconstruction of tissue defects (as necessary), was performed. Local excision of the primary site was performed with a minimum margin of 15 mm.
Neck dissection specimens were resected en bloc, and each nodal level was placed in a separate group. The surgical specimens were carefully palpated, and all identified lymph nodes were counted, sectioned at 2-to 3-mm intervals, and embedded in paraffin. Standard haematoxylin and eosin (H&E) staining was performed. All lymph node-related data were derived from original pathology reports. The seventh edition of the TNM staging system for OSCC was used for nodal staging 30 .
Postoperative RT to the neck was advised for all patients. A conventional radiotherapy regimen, which consisted of five 200-cGy fractions per week administered from Monday to Friday, was followed. The total dose for the primary tumour bed and involved neck nodes was >6000 cGy. CCRT with cisplatin (30 mg/m 2 weekly) was recommended for patients with multiple pathological nodal metastases and/or ECS.
Data analyses. The cut-off date for all surviving patients was November 2016. Descriptive statistics were summarized as frequencies, percentages, and means ± standard deviations. ROC curves, the AUC, sensitivity, specificity and 95% CIs were calculated to determine which LNR best defined the different risk groups of patients with O/OPSCC. ROC curve analyses were performed using MedCalc software, version 10.4.7.0 for Windows (MedCalc, Ostend, Belgium). All patients were classified into binary subgroups using the best LNR value and the number of positive lymph nodes as the cut-off points, which were directly calculated using MedCalc software.
The Kaplan-Meier method was used to derive estimates of the 5-year DSS. Statistical significance was determined using the log-rank test. A Cox proportional hazards model was used to adjust for the effects of other