Abstract
Purpose
Current staging systems for oral cavity cancers incorporate lymph node (LN) size and laterality, but place less weight on the total number of positive metastatic nodes. We investigated the independent impact of numerical metastatic LN burden on survival.
Methods
Adult patients with oral cavity squamous cell carcinoma undergoing upfront surgical resection for curative intent were identified in the National Cancer Data Base between 2004 and 2013. A neck dissection of a minimum of 10 LNs was required. Multivariable models were constructed to assess the association between the number of metastatic LNs and survival, adjusting for factors such as nodal size, laterality, extranodal extension, margin status, and adjuvant treatment.
Results
Overall, 14,554 patients met inclusion criteria (7,906 N0 patients; 6,648 node-positive patients). Mortality risk escalated continuously with increasing number of metastatic nodes without plateau, with the effect most pronounced with up to four LNs (HR, 1.34; 95% CI, 1.29 to 1.39; P < .001). Extranodal extension (HR, 1.41; 95% CI, 1.20 to 1.65; P < .001) and lower neck involvement (HR, 1.16; 95% CI, 1.06 to 1.27; P < .001) also predicted increased mortality. Increasing number of nodes examined was associated with improved survival, plateauing at 35 LNs (HR, 0.98; 95% CI, 0.98 to 0.99; P < .001). In multivariable models accounting for the number of metastatic nodes, contralateral LN involvement (N2c status) and LN size were not associated with mortality. A novel nodal staging system derived by recursive partitioning analysis exhibited greater concordance than the American Joint Committee on Cancer (8th edition) system.
Conclusion
The number of metastatic nodes is a critical predictor of oral cavity cancer mortality, eclipsing other features such as LN size and contralaterality in prognostic value. More robust incorporation of numerical metastatic LN burden may augment staging and better inform adjuvant treatment decisions.
INTRODUCTION
Regional neck metastasis represents an ominous prognostic factor in head and neck squamous cell carcinoma (HNSCC). The presence of just one metastatic lymph node (LN) commits patients to an advanced-stage disease category and has been shown to confer up to a 50% decrease in overall survival (OS).1 The American Joint Committee on Cancer (AJCC) staging system classically incorporates numerous factors to account for nodal disease, including size, laterality, and number of malignant nodes. Recent changes also factor in extranodal extension (ENE), also known as extracapsular spread.2
The influence of biologic heterogeneity is increasingly recognized, with head and neck staging systems evolving to better approximate the clinical behavior of unique subsites. Both nasopharyngeal and human papillomavirus–positive oropharyngeal carcinoma staging systems highlight changes that reflect their distinct pathogenic underpinnings. By comparison, nodal staging for HNSCC sites more associated with tobacco- and alcohol-mediated carcinogenesis remains broad in scope: N2b status encompasses any number of ipsilateral nodes greater than one, whereas N3 status includes all nodes greater than 6 cm. This area deserves further study, given that it may underperform in certain aspects. For instance, patients with 10 ipsilateral metastatic nodes empirically fare much worse than those with two, yet remarkably, they are staged the same. Nodal staging also remains generalized for oral cavity, larynx, and hypopharynx cancers, which arguably involve disparate prognoses and management.
Recent studies in mucosal head and neck cancers have suggested that the number of positive nodes or the number of nodes examined may convey a better measure of prognosis.3-6 Given the need for more precise staging metrics and treatment stratification, we investigated the impact of quantitative metastatic nodal burden in a large population of patients with oral cavity cancer. We focused on oral cavity cancers because of their surgical treatment paradigm with more complete pathologic nodal data.
METHODS
Data Source
Data were abstracted from the National Cancer Data Base (NCDB), a tumor registry maintained by the American Cancer Society and the Commission on Cancer of the American College of Surgeons. The NCDB captures data from more than 1,500 hospitals for approximately 70% of all patients with cancer treated in the United States. All current NCDB head and neck participant user files were investigated, covering patients treated from 2004 to 2013. This study was deemed exempt by the Cedars-Sinai Medical Center institutional review board.
Patients
All adult patients ≥ 18 years old undergoing upfront surgical resection that included neck dissection for primary oral cavity squamous cell carcinoma (International Classification of Diseases, 9th revision, clinical modification, 0-3 codes 8050-8084) for curative intent were eligible. Specific subsites included oral tongue (C02.0-C02.3), upper/lower gum (C03.0-C03.9), floor of mouth (C04.0-C04.9), hard palate (C05.0), and other parts of the mouth (eg, buccal mucosa, retromolar trigone; C06.0-C06.9). Ambiguous or overlapping sites that could potentially be oropharyngeal in origin (ie, C02.8-C02.9 for tongue/base of tongue, C05.8-C05.9 for hard palate/soft palate) were excluded.
Patients with incomplete staging, treatment, or follow-up data were excluded. Patients with clinical or pathologic distant metastasis were eliminated. Patients with fewer than 10 LNs examined were also omitted to filter out excisional LN biopsies and censure substandard neck dissections that might have artificially undermined survival.
Statistical Analysis
Missing data patterns for the variables with missing values (ie, race, insurance, income, AJCC (7th edition) N classification, LN size, lower LN involvement, ENE, margins, and contralateral LN involvement) were examined using the method proposed by Little.7 Missing rates were 24.2% for ENE, 11.2% for LN size, and 1.2% to 5.5% for other variables. The data were found to be not missing completely at random. To reduce the chance of bias from missing data, missing values were imputed using fully conditional specification implemented by the multivariable imputation by chained equations algorithm under the missing at random assumption.8,9 We generated 15 complete data sets, which were analyzed separately with results combined using the formula given in Rubin.10
The primary outcome was OS calculated from diagnosis to the date of death or censored at last follow-up. Baseline characteristics in patients with AJCC N0 classification versus N-positive classification were compared with Wilcoxon rank-sum test for continuous variables and χ2 test for categorical variables. Median follow-up was calculated using the reverse Kaplan-Meier method.11 Survival functions were estimated by the Kaplan-Meier method and compared using a log-rank test.12 Univariable and multivariable survival analyses were carried out using a Cox proportional hazards model.13 Multivariable analyses were performed using a stepwise variable selection procedure on the basis of Akaike information criterion (AIC).14 Final multivariable models were returned by the lowest AIC value. The proportional hazards assumption was assessed graphically and analytically with scaled Schoenfeld residuals.15 Violation of the proportional hazards assumption was addressed by use of a stratified Cox regression model.
The number of positive metastatic LNs and number of LNs examined were modeled using a restricted cubic spline function allowing for their nonlinear association with OS. The optimal number of knots was chosen based on the lowest AIC. For positive metastatic LNs, three knots were placed at one, two, and seven positive metastatic LNs corresponding to the 55th, 75th, and 95th percentiles, respectively, because of their right-skewed distribution. For the number of LNs examined, three knots were placed at 14, 28, and 57 LNs corresponding to default quantiles for three knots, 10th, 50th, and 90th percentiles, respectively.16 Estimated associations were illustrated with smoothed restricted cubic spline plots of the natural logarithm of adjusted hazard ratios (HRs) versus the number of positive metastatic LNs and number of LNs examined, with 0 and 10 as the reference levels, respectively. HRs were estimated with Cox proportional hazards models stratified on postoperative radiation after adjusting for age, gender, race, insurance status, income, Charlson-Deyo comorbidity index, T classification, number of positive LNs with three knots, number of LNs examined with three knots, lower neck (level 4-5) LN involvement, ENE, margins, and postoperative chemotherapy. Change points in the number of positive metastatic LNs and number of LNs examined were further estimated with a piecewise linear regression model.17
A new N classification system was devised via recursive partitioning analysis (RPA)18,19 using independent nodal predictors of mortality (ie, number of positive LNs [continuous], ENE, and lower LN involvement) in patients with a determinable AJCC (8th edition) stage. A conditional inference tree was estimated by the optimized binary recursive partitioning on the basis of a permutation test with a quadratic form of the standardized log-rank statistic with Bonferroni-adjusted P values for multiple comparisons. The performance of the multivariable models with the proposed N classification system derived from RPA and AJCC (8th edition) N classification were assessed with c-indices.16 Internal validation was performed by estimating and correcting possible optimism in c-indices using the bootstrap method with 1,000 replicates.16,20
Statistical analyses were performed using SAS 9.3 (SAS Institute, Cary, NC) and R package (Version 3.3.2; mice, rms, survival, SiZer, party libraries),21 with two-sided tests and a significance level of .05.
RESULTS
Patient Cohort
Of 85,786 eligible patients with oral cavity cancer, 14,554 met inclusion criteria (Appendix Fig A1; Appendix Table A1, online only). Median OS was 68.3 months (95% CI, 64.4 to 71.7), with a median follow-up of 46.5 months (95% CI, 45.7 to 47.3). The mean number of LNs examined was 32.1 (standard deviation ± 17.4). Among patients with node-positive disease with known data, the mean number of identified positive metastatic nodes was 3.3 (standard deviation ± 4.3), 17.2% had lower neck (level 4-5) involvement, 45.2% demonstrated ENE, and 13.3% harbored contralateral nodal involvement.
Number of Positive Metastatic LNs
In univariable analysis, the number of metastatic LNs strongly predicted for worsening OS; P < .001; Table 1). The estimated 5-year OS rates were 65.3%, 49.9%, 41.1%, 29.7%, 27.5%, 18.5%, and 9.7% for those with zero, one, two, three, four to six, seven to nine, and 10 or more metastatic LNs, respectively (Fig 1A). A similar impact of the number of metastatic LNs was seen in N2b (Fig 1B) and N2c (Fig 1C) subgroups. After adjustment for potential confounders in a multivariable model, the number of positive metastatic LNs remained strongly associated with OS (P < .001). Using a three-knot restricted cubic spline function, mortality risk escalated continuously with increasing number of metastatic nodes without plateau (Fig 2A). Given the nonlinear relationship between mortality and the number of metastatic LNs, a change point at four metastatic LNs was identified. The HR per metastatic LN increased steeply up to four metastatic LNs (HR, 1.34; 95% CI, 1.29 to 1.39; P < .001). Beyond this, the risk of death continued to increase with each additional metastatic LN, albeit more slowly (HR, 1.03; 95% CI, 1.02 to 1.04; P < .001; Table 2).
Table 1.
Univariable and Multivariable Analyses of Overall Survival in Oral Cavity Cancer
Fig 1.
Kaplan-Meier estimates of overall survival in oral cavity cancer, stratified by number of positive metastatic lymph nodes in (A) all patients, (B) patients with N2b disease, and (C) patients with N2c disease. LN+, lymph node–positive;
Fig 2.
Adjusted hazard ratio (HR) with increasing number of positive metastatic lymph nodes (LNs) and LNs examined in oral cavity cancer. Blue dashed lines represent estimated 95% CIs of the predicted HRs. (A) Gold solid line represents smoothed restricted cubic spline plot of the natural logarithm of predicted adjusted HR versus the number of positive metastatic LNs, with a reference value of 0. Gray vertical line represents the estimated change point at four positive LNs. (B). Gold solid line represents smoothed restricted cubic spline plot of the natural logarithm of predicted adjusted HR versus the number of LNs examined, with a reference value of 10. Gray vertical line represents the estimated change point at 35 examined LNs.
Table 2.
Summary of Hazard Ratios for No. of Positive Metastatic LNs and No. of LNs Examined in Oral Cavity Cancer, Stratified by Change Point
Number of LNs Examined
An increasing number of LNs examined was associated with improved OS in multivariable analyses (P < .001). As with the number of metastatic LNs, number of LNs examined exhibited a nonlinear relationship with mortality. A multivariable model with a three-knot restricted cubic spline function showed that the risk of death decreased continuously with each additional node harvested (with a baseline of 10 LNs examined) up to a change point of 35 LNs (HR, 0.98; 95% CI, 0.98 to 0.99; P < .001; Fig 2B). However, no significant improvement in survival was appreciated beyond 35 LNs (HR, 1.00; 95% CI, 0.99 to 1.00; P = .126; Table 2). Because stage I to II patients (T1-2N0) are often treated with surgery alone, they were separately compared with stage III to IV patients (T1-2N1-3/T3-4N0-3), who are often treated with surgery and adjuvant therapy. Subset analysis found similar change points for number of LNs dissected and magnitude of benefit on survival (Appendix Fig A2).
Metastatic LN Features
After adjustment for covariables, including positive metastatic LNs and number of total nodes examined, both ENE (HR, 1.41; 95% CI, 1.20 to 1.65; P < .001) and lower neck (level 4-5) involvement (HR, 1.16; 95% CI, 1.06 to 1.27; P < .001) were independently associated with mortality risk. However, LN size and contralateral LN involvement (N2c disease) had no significant impact on survival (Table 1).
Proposed Nodal Staging System
RPA using nodal covariables independently associated with survival generated a novel schema comprising metastatic nodal number and ENE (Fig 3). Kaplan-Meier estimates of the schema and AJCC (8th edition) system are illustrated in Figure 4. Lower neck (level 4-5) involvement dropped out of the model relative to other covariables. Patients with one positive LN who were ENE positive and patients with two positive LNs clustered separately in the RPA analysis, but were grouped together because of similar survival rates (Appendix Table A2, online only). The most advanced nodal category (N3b) showed HRs of 6.54 (95% CI, 5.43 to 7.89) and 3.68 (95% CI, 3.25 to 4.16) for the proposed system and AJCC (8th edition) system, respectively (Appendix Table A3, online only). The optimism-corrected c-index for the proposed system showed improvement in predictive ability (0.706; 95% CI, 0.694 to 0.718) over the AJCC (8th edition) system (0.703; 95% CI, 0.691 to 0.715).
Fig 3.
Novel proposed nodal staging system developed by recursive partitioning analysis in patients with oral cavity cancer with determinable American Joint Committee on Cancer (8th edition) stage. Bonferroni-adjusted P values are given in the inner nodes, and Kaplan-Meier estimates for 3-year overall survival (OS) are displayed in the terminal nodes. Given similar OS rates, one LN+/ENE+ and two LN+ categories were merged to N2 status. ENE–, extranodal extension–negative; ENE+, extranodal extension–positive; LN+, lymph node–positive; OS, overall survival.
Fig 4.
Kaplan-Meier estimates for (A) proposed and (B) AJCC (8th edition) N classification systems in oral cavity cancer. AJCC, American Joint Committee on Cancer; ENE–, extranodal extension–negative; ENE+, extranodal extension–positive; LN+, lymph node–positive.
DISCUSSION
In this study, we demonstrated that the absolute number of metastatic LNs is a critical predictor of oral cavity cancer mortality, surpassing other nodal covariables, including size, contralaterality, ENE, and lower neck involvement. Using a continuous multivariable regression model, we found that successive positive nodes increased the risk of death without plateau. Each positive LN conferred an added 34% increased risk of death through four positive nodes, whereas each successive positive node beyond this increased relative mortality by 3% (Table 2). In addition, we found that the number of positive LNs significantly affected prognosis among N2b and N2c cohorts (Fig 1B and 1C), suggesting that such conventional subgroups themselves comprise a wide spectrum of outcomes.
These results build on previous studies assessing metastatic LN number on head and neck cancer outcome.3,22 However, our study design contains several meaningful differences, including adjustment for covariables that are both nodal (eg, LN size, ENE, lower neck involvement) and non-nodal (eg, adjuvant chemoradiation, margin status). In contrast to prior reports, we excluded oropharyngeal malignancies because of their fundamentally different relationship between nodal burden and prognosis,23,24 now reflected in a separate AJCC nodal staging system for human papillomavirus–positive oropharyngeal cancer.2 Our analysis focused on the HNSCC sites (eg, tongue, buccal mucosa, hard palate), for which surgery, and specifically neck dissection, is the predominant treatment modality and depicts a granular representation of each metastatic node’s added impact on survival.
A related finding is that when accounting for the number of cancerous LNs, classic prognostic factors used by AJCC staging (ie, LN size and contralaterality) were no longer independent predictors of survival, suggesting that they may be surrogates for overall nodal burden. This result supports a prior multicenter study demonstrating similar outcomes in patients with N2b and N2c disease, when accounting for the fact that patients with N2c disease tend to have more metastatic nodes than do patients with N2b disease.25 Conversely, in our analysis, ENE and lower neck involvement were independently associated with worse survival. Both elements have been linked to distant metastasis in head and neck cancer26-29 and are incorporated into HNSCC and nasopharyngeal cancer staging systems, respectively.
We proposed a novel nodal staging schema (Appendix Table A2) using an agnostic recursive partitioning analysis algorithm, illustrating the importance of pure metastatic LN number in prognosis and management. ENE was retained in this schema, but only for patients with a single positive LN (Fig 3). The predictive power of the proposed system was improved over that of the AJCC (8th edition) staging system, although the absolute difference in c-indices was relatively modest. This may be because conventional nodal factors in the AJCC (8th edition) system (ie, LN size and contralaterality) serve as proxies for absolute metastatic LN number.
There are several advantages of the proposed schema beyond mildly improved prediction of survival. First, it is based on factors that independently drive outcomes, rather than surrogates. It also represents a concise stratification, relying almost entirely on a single variable. All the N categories in the proposed system identify patient groups with distinct, nonoverlapping prognoses (Fig 4).The proposed system furthermore partitions risk over a greater spectrum: patients classified as N3b (≥ 8 LN+) in the proposed system have 6.5 times the risk of death as patients classified as N0 (Appendix Table A3), with 3-year OS of 14.5%. In comparison, the HR and 3-year OS for AJCC (8th edition) N3b patients is 3.7 and 35.3%, respectively. Given the poor outcomes in patients with a high metastatic LN burden (≥ 8 positive LNs), these patients may derive greater benefit from intensification of adjuvant therapy such as concomitant chemoradiation. They may also be excellent candidates for novel therapeutic regimens, such as the addition of immunologic checkpoint inhibitors to standard chemoradiation. Collectively, the proposed system encapsulates a parsimonious model that exhibits greater discrimination at the high end of patient risk.
A final key finding is that the number of LNs examined (benign or malignant) is associated with improved survival and that this effect is much less pronounced than the impact of number of cancerous LNs. Specifically, the relative risk of death was reduced by 2% for each additional LN examined, up to 35 LNs, with no significant improvement in survival beyond this number (Fig 2B). A growing body of literature supports the number of LNs examined as an important physician-modifiable determinant of outcome in head and neck cancer,4-6,30,31 with most investigators choosing 18 nodes examined as a threshold. There are several factors that likely contribute to these different cut points, including our focus on oral cavity cancer and the requirement of at least 10 LN examined, which excludes biopsies and minor neck procedures. Our results suggest that although examining 18 LNs is associated with decreased mortality risk, survival continues to improve with more extensive neck dissections that yield nearly twice this number.
Although the exact reason why an increasing number of LNs examined improves survival is unclear, plausible hypotheses exist. First, a more thorough neck dissection may be therapeutic, increasing the probability of eliminating micrometastatic nodal deposits. Second, higher nodal yield may be a measure of surgeon acumen. Finally, the number of LNs reported in part depends on the diligence of the pathology department and may be an indirect gauge of institutional expertise.32,33 Together with work from other solid malignancy sites (eg, colorectal, breast, gastric, bladder),34-37 our data support the notion that surgeons and pathologists should strive for thorough compartmental dissection and pathologic evaluation. Neck dissection may deserve a larger role for diagnostic, therapeutic, and staging purposes.38,39
Several caveats to this analysis require mention, including its retrospective nature and absence of disease-specific survival metrics. Although the data are broad in scope, there is a lack of information on certain prognostic features, including smoking status, alcohol consumption, and perineural invasion. Factors such as chemotherapy regimen, bilateral versus ipsilateral neck dissection, and radiation quality are also not assessable in the NCDB. Our results should be validated in independent data sets to determine whether the survival detriment is due to locoregional or distant relapse, which would have implications for when to use adjuvant therapy. Finally, it is unclear whether our results can be translated to clinically staged patients, given that clinical and radiographic identification of positive LN number is often less exact than pathologic LN assessment.40,41 Nevertheless, our results represent the most compelling evidence to date of the importance of metastatic LN number in oral cavity squamous cell carcinoma.
In summary, we established that metastatic nodal burden is a central predictor of mortality in patients with oral cavity cancer, with each additional metastatic LN conferring escalated risk of mortality. Classic factors such as LN size and contralateral nodal metastasis lack independent prognostic value when accounting for number of metastatic nodes. Our data suggest that deeper integration of quantitative nodal burden could better calibrate the wide spectrum of risk that staging systems presently capture. Such adjustments would be a promising means to more effectively articulate patient prognosis, tailor clinical trial design, and ultimately advance clinical decision making.
Appendix
Fig A1.
CONSORT diagram. LN, lymph node.
Fig A2.
Adjusted hazard ratio (HR) with increasing number of lymph nodes (LNs) examined in (A) stage I to II (T1-2N0) compared with (B) stage III to IV (T1-2N1-3/T3-4N0-3) oral cavity cancer. Gold solid lines represent smoothed restricted cubic spline plots of the natural logarithm of predicted adjusted HR versus the number of LNs examined, with a reference value of 10. Gray vertical lines represent the estimated change point of (A) 31 LNs examined and (B) 37 LNs examined. Three knots for the number of LNs examined were placed at (A) 13, 25, and 49 and (B) 14, 30, and 60, each corresponding to 10th, 50th, and 90th percentiles, respectively. Blue dashed lines represent estimated 95% CIs of the predicted HRs.
Table A1.
Baseline Patient Demographics Stratified by Nodal Status
Table A2.
Comparison of New Proposed Nodal Staging System With AJCC (8th edition) System in Patients With Oral Cavity Cancer With Determinable AJCC (8th edition) Stage
Table A3.
Multivariable Analyses With Proposed N Classification System and AJCC (8th edition) N Classification System in Oral Cavity Cancer
AUTHOR CONTRIBUTIONS
Conception and design: Allen S. Ho, Zachary S. Zumsteg
Collection and assembly of data: Allen S. Ho, Zachary S. Zumsteg
Data analysis and interpretation: All authors
Manuscript writing: All authors
Final approval of manuscript: All authors
Accountable for all aspects of the work: All authors
AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST
Metastatic Lymph Node Burden and Survival in Oral Cavity Cancer
The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc.
Allen S. Ho
No relationship to disclose
Sungjin Kim
No relationship to disclose
Mourad Tighiouart
No relationship to disclose
Cynthia Gudino
No relationship to disclose
Alain Mita
Speakers' Bureau: Genentech
Kevin S. Scher
No relationship to disclose
Anna Laury
No relationship to disclose
Ravi Prasad
No relationship to disclose
Stephen L. Shiao
No relationship to disclose
Jennifer E. Van Eyk
No relationship to disclose
Zachary S. Zumsteg
Leadership: Scripps Proton Therapy Center
Consulting or Advisory Role: EMD Serono
REFERENCES
- 1.Cerezo L, Millán I, Torre A, et al. : Prognostic factors for survival and tumor control in cervical lymph node metastases from head and neck cancer. A multivariate study of 492 cases. Cancer 69:1224-1234, 1992 [DOI] [PubMed] [Google Scholar]
- 2. Amin MB, Edge S, Greene F, et al (eds): AJCC Cancer Staging Manual (ed 8). New York, NY, Springer, 2017. [Google Scholar]
- 3.Roberts TJ, Colevas AD, Hara W, et al. : Number of positive nodes is superior to the lymph node ratio and American Joint Committee on Cancer N staging for the prognosis of surgically treated head and neck squamous cell carcinomas. Cancer 122:1388-1397, 2016 [DOI] [PubMed] [Google Scholar]
- 4. doi: 10.1002/cncr.30204. Divi V, Harris J, Harari PM, et al: Establishing quality indicators for neck dissection: Correlating the number of lymph nodes with oncologic outcomes (NRG Oncology RTOG 9501 and RTOG 0234). Cancer 10.1002/cncr.30204 [epub ahead of print on July 15, 2016] [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. doi: 10.1200/JCO.2016.67.3863. Divi V, Chen MM, Nussenbaum B, et al: Lymph node count from neck dissection predicts mortality in head and neck cancer. J Clin Oncol 34:3892-3897, 2016. [DOI] [PubMed] [Google Scholar]
- 6.Kuo P, Mehra S, Sosa JA, et al. : Proposing prognostic thresholds for lymph node yield in clinically lymph node-negative and lymph node-positive cancers of the oral cavity. Cancer 122:3624-3631, 2016 [DOI] [PubMed] [Google Scholar]
- 7.Little RJ: A test of missing completely at random for multivariate data with missing values. J Am Stat Assoc 83:1198-1202, 1988 [Google Scholar]
- 8.van Buuren S, Groothuis-Oudshoorn K: mice: Multivariate imputation by chained equations in R. J Stat Softw 15:1-67, 2011 [Google Scholar]
- 9.van Buuren S: Multiple imputation of discrete and continuous data by fully conditional specification. Stat Methods Med Res 16:219-242, 2007 [DOI] [PubMed] [Google Scholar]
- 10. Rubin DB: Multiple imputation for nonresponse in surveys. New York, NY, John Wiley & Sons, 1987. [Google Scholar]
- 11.Schemper M, Smith TL: A note on quantifying follow-up in studies of failure time. Control Clin Trials 17:343-346, 1996 [DOI] [PubMed] [Google Scholar]
- 12. Kalbfleisch JD, Prentice RL: The statistical analysis of failure time data. New York, NY, John Wiley & Sons, 1980. [Google Scholar]
- 13.Cox DR: Regression models and life tables. J R Stat Soc [Ser A] B34:187-220, 1972 [Google Scholar]
- 14. Yamashita T, Yamashita K, Kamimura R: A stepwise AIC method for variable selection in linear regression. Commun Stat Theory Methods 36:2395-2403, 2007. [Google Scholar]
- 15. Grambsch PM, Therneau TM: Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 81:515-526, 1994. [Google Scholar]
- 16. Harrell F: Regression modeling strategies: With applications to linear models, logistic regression, and survival analysis (ed 2). New York, NY, Springer, 2001. [Google Scholar]
- 17.Toms JD, Lesperance ML: Piecewise regression: A tool for identifying ecological thresholds. Ecology 84:2034-2041, 2003 [Google Scholar]
- 18. Hothorn T, Hornik K, Zeileis A: Unbiased recursive partitioning: A conditional inference framework. J Comput Graph Stat 15:651-674, 2006. [Google Scholar]
- 19.Segal MR: Regression trees for censored data. Biometrics 44:35-47, 1998 [Google Scholar]
- 20.Harrell FE, Jr, Lee KL, Mark DB: Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 15:361-387, 1996 [DOI] [PubMed] [Google Scholar]
- 21. The R Foundation: The R Project for statistical computing. http://www.R-project.org/
- 22.Sinha P, Kallogjeri D, Gay H, et al. : High metastatic node number, not extracapsular spread or N-classification is a node-related prognosticator in transorally-resected, neck-dissected p16-positive oropharynx cancer. Oral Oncol 51:514-520, 2015 [DOI] [PubMed] [Google Scholar]
- 23.Huang SH, Xu W, Waldron J, et al. : Refining American Joint Committee on Cancer/Union for International Cancer Control TNM stage and prognostic groups for human papillomavirus-related oropharyngeal carcinomas. J Clin Oncol 33:836-845, 2015 [DOI] [PubMed] [Google Scholar]
- 24.O’Sullivan B, Huang SH, Su J, et al. : Development and validation of a staging system for HPV-related oropharyngeal cancer by the International Collaboration on Oropharyngeal cancer Network for Staging (ICON-S): A multicentre cohort study. Lancet Oncol 17:440-451, 2016 [DOI] [PubMed] [Google Scholar]
- 25.Ebrahimi A, Gil Z, Amit M, et al. : The prognosis of N2b and N2c lymph node disease in oral squamous cell carcinoma is determined by the number of metastatic lymph nodes rather than laterality: Evidence to support a revision of the American Joint Committee on Cancer staging system. Cancer 120:1968-1974, 2014 [DOI] [PubMed] [Google Scholar]
- 26.Riaz N, Setton J, Tam M, et al. : Patients with low lying lymph nodes are at high risk for distant metastasis in oropharyngeal cancer. Oral Oncol 50:863-868, 2014 [DOI] [PubMed] [Google Scholar]
- 27.Ellis ER, Mendenhall WM, Rao PV, et al. : Does node location affect the incidence of distant metastases in head and neck squamous cell carcinoma? Int J Radiat Oncol Biol Phys 17:293-297, 1989 [DOI] [PubMed] [Google Scholar]
- 28.Myers JN, Greenberg JS, Mo V, et al. : Extracapsular spread. A significant predictor of treatment failure in patients with squamous cell carcinoma of the tongue. Cancer 92:3030-3036, 2001 [DOI] [PubMed] [Google Scholar]
- 29.Greenberg JS, Fowler R, Gomez J, et al. : Extent of extracapsular spread: A critical prognosticator in oral tongue cancer. Cancer 97:1464-1470, 2003 [DOI] [PubMed] [Google Scholar]
- 30.Ebrahimi A, Clark JR, Amit M, et al. : Minimum nodal yield in oral squamous cell carcinoma: Defining the standard of care in a multicenter international pooled validation study. Ann Surg Oncol 21:3049-3055, 2014 [DOI] [PubMed] [Google Scholar]
- 31.Ebrahimi A, Zhang WJ, Gao K, et al. : Nodal yield and survival in oral squamous cancer: Defining the standard of care. Cancer 117:2917-2925, 2011 [DOI] [PubMed] [Google Scholar]
- 32. Cormier B, Sauthier P, Lussier C, et al: Determinants of lymph node count in endometrial cancer surgical staging. Int J Gynecol Cancer 22:1361-1366, 2012. [DOI] [PubMed]
- 33.Meijer RP, Nunnink CJ, Wassenaar AE, et al. : Standard lymph node dissection for bladder cancer: Significant variability in the number of reported lymph nodes. J Urol 187:446-450, 2012 [DOI] [PubMed] [Google Scholar]
- 34.Le Voyer TE, Sigurdson ER, Hanlon AL, et al. : Colon cancer survival is associated with increasing number of lymph nodes analyzed: A secondary survey of intergroup trial INT-0089. J Clin Oncol 21:2912-2919, 2003 [DOI] [PubMed] [Google Scholar]
- 35.Weir L, Speers C, D’yachkova Y, et al. : Prognostic significance of the number of axillary lymph nodes removed in patients with node-negative breast cancer. J Clin Oncol 20:1793-1799, 2002 [DOI] [PubMed] [Google Scholar]
- 36.Volpe CM, Driscoll DL, Douglass HO, Jr: Outcome of patients with proximal gastric cancer depends on extent of resection and number of resected lymph nodes. Ann Surg Oncol 7:139-144, 2000 [DOI] [PubMed] [Google Scholar]
- 37.Konety BR, Joslyn SA, O’Donnell MA: Extent of pelvic lymphadenectomy and its impact on outcome in patients diagnosed with bladder cancer: Analysis of data from the Surveillance, Epidemiology and End Results Program data base. J Urol 169:946-950, 2003 [DOI] [PubMed] [Google Scholar]
- 38.Seethala RR: Current state of neck dissection in the United States. Head Neck Pathol 3:238-245, 2009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Ferlito A, Rinaldo A, Devaney KO, et al: Detection of lymph node micrometastases in patients with squamous carcinoma of the head and neck. Eur Arch Otorhinolaryngol 265:1147-1153, 2008. [DOI] [PubMed]
- 40.Koch WM, Ridge JA, Forastiere A, et al. : Comparison of clinical and pathological staging in head and neck squamous cell carcinoma: Results from Intergroup Study ECOG 4393/RTOG 9614. Arch Otolaryngol Head Neck Surg 135:851-858, 2009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Greenberg JS, El Naggar AK, Mo V, et al. : Disparity in pathologic and clinical lymph node staging in oral tongue carcinoma. Implication for therapeutic decision making. Cancer 98:508-515, 2003 [DOI] [PubMed] [Google Scholar]