Prognostic Performance of a New Staging Category to Improve Discrimination of Disease-Specific Survival in Nonmetastatic Oral Cancer

Key Points

Question  Does a new staging category that incorporates the neutrophil to lymphocyte ratio and histopathologic features improve survival estimations compared with routinely used pathologic TNM staging?

Findings  In this retrospective cohort study of 396 patients with newly diagnosed oral squamous cell carcinoma after major surgery, the new staging category provided better monotonicity and better discriminatory ability for 5-year disease-specific survival.

Meaning  This new staging category could help to identify high-risk patients for more intense adjuvant therapy.

Importance  Inflammatory status is associated with outcome in oral squamous cell carcinoma (OSCC). Combining the preoperative neutrophil to lymphocyte ratio (NLR) and histopathologic features may provide clinicians with more exact information regarding the prognosis of OSCC.

Objective  To compare the prognostic performance of the routinely used pathologic TNM staging with a new staging category that incorporates the NLR and histopathologic features.

Design, Setting, and Participants  This retrospective cohort study included 396 patients with newly diagnosed OSCC who underwent major surgery at a medical center from January 1, 2006, through December 31, 2013. Follow-up was completed on October 31, 2015, and data analysis was performed from January 1, 2016, through April 30, 2016.

Main Outcomes and Measures  The multivariate Cox proportional hazards regression model was used to determine the clinical or pathologic factors associated with 5-year disease-specific survival (DSS), and these factors were assigned integer points to create a new staging category. The monotonicity and discriminatory ability of the pathologic TNM staging and new staging category were evaluated with the linear trend χ² test, Akaike information criterion, and Harrell C statistic.

Results  In total, 396 patients who underwent major surgery with curative intent for OSCC with or without adjuvant therapy were included in this study (mean [SD] age, 53 [11] years; 367 men [92.7%] and 29 women [7.3%]). Perineural invasion (adjusted hazard ratio [aHR], 1.74; 95% CI, 1.23-2.46), high NLR (aHR, 1.60; 95% CI, 1.11-2.30), advanced pT (T3 + T4) classification (aHR, 1.59; 95% CI, 1.13-2.25), and advanced pN (N2) classification (aHR, 3.96; 95% CI, 2.78-5.63) were independent prognostic survival factors. The β coefficients from the Cox proportional hazards regression model were used to develop an integer-based weighted point system (perineural invasion, score of 1; NLR, score of 1; advanced pT, score of 1; and advanced pN, score of 3). The summations of these risk scores were stratified for the new staging category as follows: new stage I, score of 0; new stage II, score of 1; new stage III, score of 2 or 3; and new stage IV, score of 4 to 6. Compared with the American Joint Committee on Cancer staging category, this new staging category provided better monotonicity with a higher linear trend χ² value (106 vs 49), better discriminatory ability with smaller Akaike information criterion (1497 vs 1533), and greater Harrell C statistic (0.73 vs 0.69) for 5-year DSS. The results remained robust after adjusting other risk factors.

Conclusions and Relevance  In this study, new staging category had better DSS discriminatory ability and could help to identify high-risk patients for intense adjuvant therapy.

Oral squamous cell carcinoma (OSCC) is one of the most frequent malignant tumors worldwide, with an increasing incidence in Taiwan and some areca quid–use areas.¹ Despite improved therapeutic strategies, such as diagnostic workup, surgical techniques, chemotherapy, and radiotherapy, outcomes for patients with OSCC remain unchanged during the past decade.^2,3 Prognosis of patients with OSCC and treatment considerations depend mostly on the American Joint Committee on Cancer (AJCC) TNM classification. However, the current TNM staging is not precise enough to estimate outcomes because it focuses on only the tumor itself. Patients with the same stage often have different outcomes. Therefore, it is important to improve and adjust the present TNM staging system to identify patients with poor prognosis or at high risk of disease recurrence and death.⁴

A previous study⁵ found that inflammatory status has a prognostic role in cancer pathogenesis and disease progression. The neutrophil to lymphocyte ratio (NLR) is a systemic inflammatory response marker that has been studied as an independent prognostic factor in head and neck cancers.^6-8 The NLR can be measured conveniently, is cost-effective and easily reproducible, may allow widespread clinical use, and can contribute to prognosis estimation. Therefore, as opposed to the TNM staging system that is solely based on anatomical information and does not reflect the biological heterogeneity of cancers, adding an inflammatory biomarker into the TNM classification may complement prognosis estimation for patients with OSCC.

The prognostic values of the NLR have been assessed in patients with OSCC, but a previous study⁷ considered only serum inflammatory biomarkers. As previously known, traditional histopathologic prognostic factors, including tumor thickness, grade, extracapsular spread, margin status, and perineural invasion (PNI), have been routinely used to estimate outcomes for patients with OSCC in the literature.^9-11 Combining NLR and pathologic tumor characteristics may provide clinicians more exact information regarding the prognosis of OSCC. Given this background, we revised the routinely used TNM staging system by incorporating histopathologic features and the NLR to produce a new staging category. The aim of this study was not only to evaluate factors associated with OSCC survival but also to assess whether this new staging category with the addition of the NLR and histopathologic features to the current TNM system–based prediction models improved the discrimination of 5-year disease-specific survival (DSS).

This study was reviewed and approved by the institutional review board of Kaohsiung Veterans General Hospital in Taiwan. The requirement for informed consent was waived because all identifying information was removed from the data set before analysis.

The data for this study were collected from the cancer registry data set from the Kaohsiung Veterans General Hospital Cancer Center from January 1, 2006, through December 31, 2013. Electronic medical records and a cancer registry data set were retrospectively reviewed. The follow-up deadline was October 31, 2015, for survivors. Data analysis was performed from January 1, 2016, through April 30, 2016. The records of all 396 patients with newly diagnosed OSCC who underwent radical surgery with or without adjuvant therapy were identified for this study. Exclusion criteria included the following: patients who had a history of cancer, had chemotherapy or radiotherapy as their initial treatment, or did not have preoperative laboratory data. The cancer registry data set provided the following: the date of diagnosis, sites of the primary tumor, age, sex, personal habits (smoking, alcohol, and chewed betel quid), margin status (positive or negative), degree of differentiation (well, moderately, or poorly differentiated), PNI, lymphovascular permeation, adjuvant treatment (eg, chemotherapy or radiotherapy), cause of death, and clinical and pathologic TNM stage. All cases were staged according to the AJCC stage classification system updated in 2009 (seventh edition).⁴ The clinical end point was 5-year DSS rate. Deaths because of cancer were recorded as events, and deaths secondary to other causes were censored.

Routine complete blood cell counts and differential cell counts were routinely collected as part of a preoperative protocol. The analysis was performed within 30 days of the surgery, and if multiple values existed for a patient, sample values closest to the date of resection were selected. Patients with any signs of infection or who were administered corticosteroids or immunosuppressive medications were excluded from the analysis. The NLR was defined as the absolute neutrophil count divided by the absolute lymphocyte count. A consensus of the NLR cutoff value for OSCC remains to be determined. In this analysis, we tested the use of different sets of cutoff points, and we adapted the cutoff points (median, 2.37) instead of using the receiver operating characteristic curve cutoff value of 3.41 to define high and low NLR groups because of its better discriminability (Table 1).

All recorded variables were considered in a univariate Cox proportional hazards regression analysis for estimating hazard ratios (HRs) with 95% CIs. Those with statistical significance at P < .10 were entered into a multivariate Cox proportional hazards regression model. Then, all significant factors from the multivariate analysis were used to provide the most precise estimates of the coefficients for the selected model and to develop an integer scoring system by following methods used in the Framingham study.¹² The β coefficients from the Cox proportional hazards regression model were used to develop an integer-based weighted point system for stratifying 5-year DSS. The referent for each variable was assigned a value of 0, and the coefficients for the others were adjusted proportionally, rounding to the nearest integer. Individual scores were assigned by summing the individual risk factor points (PNI, 1 point; NLR, 1 point; advanced pT, 1 point; nd advanced pN, 3 points). The total scores were stratified for the new staging category as follows: new stage I, score of 0; new stage II, score of 1; new stage III, score of 2 or 3; and new stage IV, score of 4 to 6.

Continuous variables were analyzed with 1-way analysis of variance, and categorical variables were compared by Pearson χ² or Fisher exact test. Cumulative 5-year DSS rates for the AJCC staging system and new staging category were analyzed using the Kaplan-Meier method. Survival curves were measured from the time of diagnosis using disease-specific mortality as the primary event variable. Proportional hazards assumption was met for all variables. The Cox proportional hazards regression model was used to compare the 5-year DSS rates for these 2 stage systems after adjusting for clinicopathologic factors. Monotonicity was also assessed with linear trend χ² test, and a higher value indicated a better monotonic trend.^13,14 Discriminability of a gradient evaluation for the AJCC staging system and new staging category were assessed with the Akaike information criterion (AIC)¹⁵ and Harrell C statistic. In addition, a model with a lower AIC was preferred to reduce the risk of overfitting. The Harrell C statistic indicated model prediction as follows: 0.5, equal chance; 0.7 to 0.8, acceptable; 0.8 to 0.9, excellent; and 0.9 to 1, outstanding. SPSS statistical software, version 15 (SPSS Inc), was used for analysis, P < .05 was used to determine statistical significance, and all tests were 2-sided.

In total, 396 patients who underwent major surgery with curative intent for oral cancer with or without adjuvant therapy were included in this study (mean [SD] age, 53 [11] years; 367 men [92.7%] and 29 women [7.3%]). Demographic characteristics of the cohort are summarized in Table 1. Oral tongue and buccal cancers were the 2 most common sites and were observed in 330 patients (83.3%). After primary surgery, 41 patients (10.4%) had positive surgical margins, 104 patients (26.3%) had PNI, 29 patients (7.3%) had lymphovascular invasion, and 206 patients (52.1%) received adjuvant therapy. The number of DSS events in this study population was 141, and the median follow-up time was 2.4 years (range, 1-9 years).

In addition, we explored the effect of clinical features on primary outcomes. Univariate Cox proportional hazards regression analysis with respect to DSS revealed that male sex, advanced pT (T3 + T4), advanced pN (N2), tumor thickness greater than 5 mm, poor differentiation, PNI, lymphovascular invasion, and an NLR higher than a median of 2.73 were significant factors, with HRs given in Table 1. Multivariate analysis with stepwise selection revealed that 4 factors, PNI (adjusted HR [aHR], 1.74; 95% CI, 1.23-2.46), a high NLR (aHR, 1.60; 95% CI, 1.11-2.30), advanced pT (T3 + T4) classification (aHR, 1.59; 95% CI, 1.13-2.25), and advanced pN (N2) classification (aHR, 3.96; 95% CI, 2.78-5.63), remained significant factors (Table 2). The summations of these risk scores were stratified for the new staging category as follows: new stage I, score of 0; new stage II, score of 1; new stage III, score of 2 or 3; and new stage IV, score of 4 to 6.

The Kaplan-Meier curves regarding AJCC TNM staging and the new staging category are shown in the Figure. The 5-year DSS using TNM staging was 91.3% (95% CI, 98.9%-83.7%) at stage I, 75.8% (95% CI, 85.2%-66.4%) at stage II, 60.9% (95% CI, 77.2%-44.6%) at stage III, and 40.5% (95% CI, 48.3%-32.7%) at stage IV. For the new staging category, 96 patients were in the new staging category I with 83.7% (95% CI, 91.7%-75.7%) 5-year DSS, 116 patients were in the new staging category II with 73.1% (95% CI, 82.9%-63.3%) 5-year DSS, 97 patients were in the new staging category III with 56.0% (95% CI, 67.6%-44.4%) 5-year DSS, and 87 patients were in the new staging category IV with 19.7% (95% CI, 28.5%-10.9%) 5-year DSS.

For the AJCC TNM staging and new staging category, performance is detailed in Table 3. The new staging category had better monotonicity, with a higher linear trend χ² value than the AJCC TNM staging (106 vs 49). Compared with the AJCC TNM staging, the new staging category also had better discriminatory ability for 5-year DSS, with a smaller AIC (1497 vs 1533) and greater Harrell C statistic (0.73 vs 0.69) (Table 3). This finding implies that the new staging category that added the NLR and histopathologic features provided a better classification system for OSCC than the current AJCC TNM staging. Furthermore, the results remained robust even with adjustment for other factors (Table 3). In multivariate analysis, the new staging category–based model performed better than the AJCC TNM-based model, with a higher liner trend χ² (55 vs 29), lower AIC (1514 vs 1530), and higher Harrell C statistic (0.754 vs 0.745).

In this study, we confirmed that a high preoperative NLR was significantly associated with poor DSS in patients with OSCC, and we attempted to use and test the prognostic utility of combining the NLR, histopathologic features, and pathologic T and N classifications to create a new staging category. The present TNM staging is deficient in estimating survival outcomes in OSCC, and incorporation of the NLR and PNI into a prognostic model based on the TNM classification significantly improved risk reclassification for disease-specific survival. Thus, this new staging category could help to stratify high-risk patients who may benefit from more intense adjuvant therapy.

This study had several strengths. To our knowledge, this is the first study to merge both the NLR and histopathologic features into a new risk category for patients with OSCC. Second, pathologic TNM staging does not consider other clinically or pathologically relevant factors, such as inflammatory status and differentiation or PNI profile for a tumor. This new risk category improves discriminability and monotonicity relative to the current pathologic TNM staging because it has the strengths of the NLR and histopathologic features and eliminates drawbacks with pathologic TNM staging. Third, our study provided a new staging category that is easy to use for patients with early- or advanced-stage OSCC. For patients with OSCC undergoing surgical intervention, risk stratification considering tumor size, location, involved lymph nodes, histopathologic features, and inflammatory status is sufficient to estimate survival.

Despite recent improvements in the identification of genetic and molecular factors in OSCC,^16,17 the most widely used routine prognostic assessment of OSCC still relies on traditional histopathologic features, such as tumor size, involved lymph node status, extracapsular spread, surgical margin, differentiation, and PNI.⁹ However, there is increasing evidence regarding the involvement of inflammation and activation of the immune system in carcinogenesis, tumor growth, invasion, and metastasis.^5,18,19 This systemic inflammatory response leads to changes in relative levels of circulating leukocytes, including neutrophils, macrophages, eosinophils, and lymphocytes. Neutrophils in the tumor microenvironment stimulate tumor development by producing cytokines and chemokines, such as vascular endothelial growth factor, interleukin 6, and tumor necrosis factor.²⁰ Lymphocytes are also crucial components in the adaptive immune system, and infiltrating lymphocytes have been reported to indicate the generation of an effective antitumor cellular immune response. Furthermore, an in vitro study²¹ found that the cytolytic activity of lymphocytes and natural killer cells was suppressed by neutrophils. Thus, a low peripheral lymphocyte level may reflect a poorer lymphocyte-mediated immune response against a tumor and suggests a worse prognosis.²² Therefore, in this study, we selected the most commonly used inflammatory marker NLR, which explained the balance between protumor inflammatory status and antitumor immune response.

Zahorec²³ was the first to propose the NLR as a rapid and simple index to assess systemic inflammatory response in 90 oncologic patients in an intensive care unit. In contrast to other inflammatory markers, it is inexpensive and easily accessible. It also comprises components within the routine full blood cell count assay and can be performed easily before major surgery. A higher NLR is related to poor prognosis in many cancers, such as ovarian cancer,²² several digestive system cancers,^24,25 lung cancer,²⁶ and prostate cancer.²⁷ To our knowledge, evidence regarding the use of the NLR as an outcome estimator for patients with head and neck cancer is limited. Haddad et al²⁸ found that an NLR of 5 or greater was a prognostic factor for locally advanced head and neck cancer for 2-year overall survival (P = .02) and metastasis-free survival (P = .08). Perisanidis et al⁷ also reported that an elevated NLR was significant, with a shorter DSS in patients with oral cancer who were administered preoperative chemoradiotherapy (HR, 10.37; 95% CI, 1.28-84.08; P = .03). In addition, a consensus regarding the NLR cutoff value remains to be determined. In our study, NLR (median value of NLR, >2.37) was significantly associated with poor outcome, similar to previous reports.^7,8 In the multivariate Cox proportional hazards regression analysis given in Table 2, positive PNI (aHR, 1.74; 95% CI, 1.23-2.46), high preoperative NLR (aHR, 1.60; 95% CI, 1.11-2.30), advanced pathologic T classification (aHR, 1.59; 95% CI, 1.13-2.25), and advanced pathological N classification (aHR, 3.96; 95% CI, 2.78-5.63) were independently associated with a poor 5-year DSS.

Currently, management of OSCC is still based on the pathologic TNM staging of patient specimens. However, cancer staging based on the TNM system is considered imperfect in estimating survival, and large numbers of histopathologic, immunohistochemical, and molecular biomarkers have been reported in the literature.²⁹ Better prognostic tools that combine important clinical and pathologic factors to estimate outcome are needed. The present study assesses the effectiveness of tumor characteristics and the NLR in estimating the survival of patients with OSCC after major surgery. The purpose was to group patients into a high-risk category who would benefit from more intense therapy and a low-risk category in which surgery would be efficacious. We divided patients into 4 new staging categories, and the 5-year DSS for each new category was 81.7% in category I, 72.2% in category II, 53.2% in category III, and 18.5% in category IV. Compared with the current TNM classification, this new risk category provides a better monotonicity discriminatory ability for 5-year DSS, with a higher linear trend χ² value (106 vs 49), smaller AIC (1497 vs 1533), and greater Harrell C statistic (0.73 vs 0.69). After adjusting for other factors, the new staging category–based model performed better than the AJCC TNM-based model. Therefore, this new risk category could be used as an alternative to the AJCC TNM classification for patients with OSCC.

Our study had several limitations. First, the study is limited by its retrospective nature and single-center focus. Second, application of this stage grouping requires neck dissection for pathologic N stage; therefore, we did not include patients with OSCC who did not undergo a neck dissection. Third, no patient in our study had N3 or M1 disease because several researchers recommended against surgical intervention for these patients, especially for patients with N3 disease because of poor survival and high comorbidity.^30,31 Future researchers should consider recruitment of patients with stage N3 cancer. Fourth, in this study, we did not select the patients with at least 5-year follow-up because of limited cases. We chose another method, and all patients with OSCC who met the inclusion criteria were included. The patients with no events during the observation period were recorded as censored, which can be a source of bias of the estimates. Fifth, racial variations are known to affect the NLR value.³² Our results are applicable to Asian patients and necessitate validation in Western countries.

The present study found that a high NLR and PNI were independent prognostic factors for patients with OSCC after major surgery and that integration of the NLR and PNI into a pathologic TNM classification–based group can complement and improve the accuracy of survival estimation. This new staging category could be used to further stratify patients with OSCC and help identify low-risk patients for intensive follow-up and high-risk patients for adoption of personal adjuvant therapy.

Corresponding Author: Ching-Chieh Yang, MD, MS, Department of Radiation Oncology, Chi-Mei Medical Center, No. 901, Zhonghua Rd, Yongkang District, Tainan City 710, Taiwan (970816@mail.chimei.org.tw).

Accepted for Publication: October 10, 2016.

Published Online: January 26, 2017. doi:10.1001/jamaoto.2016.3802

Author Contributions: Drs Lee and Yang had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Lee, Huang, T.-S.Chang, Yang.

Acquisition, analysis, or interpretation of data: Lee, Huang, K.-P. Chang, Lin, Su, Chen, Yang.

Drafting of the manuscript: Lee, Huang, T.-S.Chang, Yang.

Critical revision of the manuscript for important intellectual content: Lee, K.-P. Chang, Lin, Su, Chen, Yang.

Statistical analysis: Lee, Huang.

Administrative, technical, or material support: Lee, K.-P. Chang, Chi, Lin, T.-S. Chang, Su, Yang.

Study supervision: Lee, Huang, Chen.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.

Funding/Support: This work was supported by grant VGHKS16-CT3-05 from the Kaohsiung Veterans General Hospital (Dr Lee).

Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and the decision to submit the manuscript for publication.

Additional Contributions: The staff of the Cancer Center of Kaohsiung Veterans General Hospital prepared the data.