Comprehensive analysis of prognostic value of lymph node staging classifications in patients with head and neck squamous cell carcinoma after cervical lymph node dissection

Abstract

Purpose:

To determine the optimal threshold of examined lymph node (ELN) number from cervical lymph node dissection for head and neck squamous cell carcinoma (HNSCC). Further to compare the prognostic value of multiple lymph node classification systems and to determine the most suitable scheme to predict survival.

Methods:

A total of 20991 HNSCC patients were included. Odds ratios (ORs) for negative-to-positive node stage migration and hazard ratios (HRs) for survival were fitted using the LOWESS smoother. Structural breakpoints were determined by the Chow test. The R square, C-index, likelihood ratio, and Akaike information criterion (AIC) were used to compare the prognostic abilities among AJCC N stage, number of positive lymph nodes (pN), positive lymph node ratio (LNR) and log odds of positive lymph nodes (LODDS) stages.

Results:

A minimal threshold ELN number of fifteen had the discriminatory capacities for both stage migration and survival. LODDS stages had the highest R square value (0.208), C-index (0.736) and likelihood ratio (2467) and the smallest AIC value (65874). LODDS stages also showed prognostic value in estimating patients with AJCC N0 stage. A novel staging system was proposed and showed good prognostic performance when stratified by different primary sites.

Conclusion:

Fifteen lymph nodes should be examined for HNSCC patients. LODDS stage allows better prognostic stratification, especially in N0 stage. The proposed staging system may serve as precise evaluation tools to estimate postoperative prognoses.

Introduction

Head and neck squamous cell carcinoma (HNSCC) is the sixth leading cancer by incidence and the eighth leading cause of cancer-associated mortality worldwide [1,2]. HNSCC often occurs in the oral cavity, oropharynx, larynx and hypopharynx. The American Joint Committee on Cancer (AJCC) TNM staging system for HNSCC is most widely used in clinical practice, and the N classification integrates information about lymph node (LN) status including number, size, laterality [3]. However, previous studies have questioned the prognostic and discriminative ability of the AJCC N classification within specific stages and for patients with bilateral metastatic LNs [4].

Radical surgical resection with cervical lymph node dissection remains the mainstay for obtaining precise information about metastatic LN for HNSCC. Cervical lymph node dissections are classified as either comprehensive or selective [5]. A comprehensive neck dissection is one that removes all lymph node groups that would be included in a classic radical neck dissection. Selective neck dissections have been developed based on the common pathways for spread of HNSCC to regional nodes [6]. However, there is no consensus about the extent of surgical lymphadenectomy with some literatures suggesting that a more extensive lymphadenectomy confers no survival advantage for survival but might result in greater surgical morbidity [7,8]. In addition, the number of examined LN (ELN) during lymphadenectomy can affect the accuracy of the N stage and the prediction about prognoses for HNSCC patients. Nevertheless, the number of positive LN (PLN), namely the pN classification is recognized as a critical factor which impacts the prognosis of HNSCC treated with surgical resection.

In addition the AJCC N stage and pN classification, other have shown greater lymph node ratio (LNR) in HNSCC portends inferior prognoses than those with lower LNR [9]. Moreover, log odds ratio (LODDS), which is the log of the ratio of PLN to the number of negative LN, has been suggested to being superior to the AJCC N stage and LNR classification for predicting the prognosis in this entity [10]. However, there is little clinical investigation comprehensively compare the prognostic ability of the aforementioned four LN classifications for HNSCC patients.

By analyzing a large cohort derived from the Surveillance, Epidemiology, and End Results (SEER) database, we aim to (1) determine the optimal ELN number threshold and (2) compare the prognostic value of LN classifications and determine the most suitable scheme to describe prognosis of HNSCC patients with a sufficient number of examined LNs.

Methods

Study population

A total of 20991 patients diagnosed with oral cavity, oropharynx, larynx and hypopharynx HNSCC who underwent surgical resection were identified from the SEER database and the inclusion and exclusion criteria are shown in Appendix Figure. Eligible patients were re-staged according to the AJCC 7th edition TNM staging system. We examined clinicopathological variables including the year of diagnosis, age, gender, race, marital status, primary site, histological grade, tumor size, ELN number and PLN number. The datasets were approved for research by the institutional review board of Fudan University Shanghai Cancer Center, which waived the requirement for the written informed consent of individual patients given the retrospective nature of the study.

According to the SEER program, survival time was defined as the time between diagnosis and the date of death, the last follow-up time, or the cutoff date of December 31, 2015 [11]. The primary endpoints were overall survival (OS), which was defined as the time between diagnosis and death from any cause, and cancer-specific survival (CSS), defined as the time from date of diagnosis to death caused by HNSCC. CSS with a competing risk of death from non-CSS causes was defined as the secondary endpoint.

Statistical analysis

Spearman coefficients were used to evaluate the correlation of LN classifications. The log-rank test and Kaplan-Meier survival curves were employed to compare survival between different subgroups. The cumulative incidence function (CIF) was used to describe the probability of death [12]. A two-sided p-value < 0.05 was considered statistically significant. Statistical analyses were conducted with R software version 3.5.2. The details of definitions of lymph node-based classifications, determination of the threshold of ELN for adequate N staging, proposal of novel TN_LODDS classification are shown in eMethod.

Results

Patient characteristics

A total of 20991 eligible HNSCC patients in the SEER database (1988–2016) undergoing cancer-directed surgery were included. The baseline characteristics are shown in eTable 1. The median follow up time was 43 months. The median number of ELN was 26 (IQR = 15, 40). The distribution of ELN number differed stratified by the primary sites; patients with tumor located in the hypopharynx and larynx had a larger number of ELN (median = 30) than those originated from oral cavity and oropharynx (median = 26 and 25, respectively).

Estimation and validation of the threshold of ELN number

To determine the ELN number that needed to be dissected to obtain the maximum differential for accurate lymph node staging and survival, ORs and HRs were extracted from the logistic and Cox regression analyses to construct LOWESS smoother fitting curves and to calculate the structural breakpoint. All the structural breakpoints were in agreement with one another, with fifteen ELN the cutoff point for both survival (Fig. 1A and B) and N staging (Fig. 1C).

Fig. 1.

LOWESS smoother fitting curves of stage migration and overall survival and determination of structural break points with use of the Chow test. The fitting bandwidth was 2/3. (A) and (B) Overall survival and cancer-specific survival were estimated by using the Cox proportional hazards regression model after adjusting for T stage, N stage, histological type, primary site, age, gender. (C) Stage migration was estimated by logistic regression after adjusting for T stage, N stage, histological type, primary site, age, gender in the all cohorts.

The chosen optimal cut-point was further validated. Survival analysis confirmed significant differences between the survival of patients with ≥15 ELNs and those with <15 ELNs (OS: HR, 0.870, 95% CI, 0.832–0.910, p = 0.012; CSS: HR, 0.841, 95% CI, 0.797–0.887, p < 0.001; Cancer-specific mortality: sHR, 0.865, 95% CI, 0.835–0.896, p < 0.001; Fig. 2A–C) after adjusting for other prognostic factors. The cut point was then validated in patients with negative-node and positive-node disease that used the aforementioned three end-points as measurements (Fig. 2D–I).

Fig. 2.

Stratification of overall survival (A, D, G), cancer-specific survival (B, E, H) and cancer-specific mortality (C, F, I) among HNSCC patients with overall (A, B, C), node-negative (C, D, E), and node-positive (G, H, I) diseases at the cut point of the number of harvested lymph nodes (15) on the basis of multivariable adjustments (other covariates were sex, age, T staging, primary site). ELN, examined lymph node.

Characteristics of lymph node classifications

Since our prior analysis had determined that fifteen ELN as the minimal threshold for HNSCC patients underwent surgical resection, 5194 patients with <15 ELN were excluded. For the remaining 15797 HNSCC patients in the final cohort, scatter plots were created to assess the relationship of LNR, LODDS and pN (eFigure 1). As shown in eFigure 1A and B, the LODDS was more closely related to LNR than pN (r = 0.881 vs. 0.767). eFigure 2 demonstrated the relationship between log hazard ratio and three LN classifications. As the value of each classification increased, the corresponding probability of death rose with nonlinear relation.

Survival analysis of LN classifications

HNSCC patients were stratified according to the AJCC N stage, pN, LNR and LODDS classifications. Fig. 3A–H showed the Kaplan-Meier survival curves for OS and CSS, which suggested that LODDS, LNR, pN classifications could provide effective stratification. However, the AJCC N classification plot showed an unclear pattern, in which there was an overlap for CSS between N1 and N3 stage (Figure 3A, 5-year OS, N1 vs. N3, 55.1% and 50.4%, p > 0.05), and N2c, rather than N3, had the worst 5-year CSS (27.4%). As shown in Fig. 3I–J, the CIF for cancer-specific mortality also suggested that other three LN classifications had superior prognostic abilities than the AJCC N classification. Patients diagnosed with N2b and N3 stage were not statistically separated as shown by the overlap of the cancer-specific mortality curves (5-year cumulative HNSCC death probability, 41.3% vs. 39.2%).

Fig. 3.

OS, CSS and CIF for cause-specific mortality stratified by AJCC N (A,E,I), pN (B,F,J), LNR (C,G,K) and LODDS (D,H,L) classifications for training cohort.

Moreover, the LODDS classification showed prognostic value in estimating patients with AJCC N0 stage. As shown in eFigure 3, there were significant differences in HNSCC N0 stage patients’ CSS and cumulative HNSCC death probability in LODDS groups (LODDS1 vs. LODDS2, 5-year CSS, 84.4% and 80.8%; 5-year cumulative HNSCC death probability, 14.8% vs. 18.2%).

The univariate Cox regression analysis was performed to identify prognostic factors of OS. As shown in eTable 2, clinicopathological variables including four N classifications, primary site, tumor size, T stage, race, PRCDA region, age, histological grade, postoperative radiotherapy (PORT), chemotherapy, marital status were significantly associated with prognosis. The results of multivariate analysis are provided in Table 1, which showed that primary site, tumor size, T stage, PORT, chemotherapy, age and marital status were prognostic factors both for OS and cancer-specific mortality and histological grade was identified as predictor for OS. Moreover, the LODDS classification model yielded the highest R² value (0.208), C-index (0.736) and likelihood ratio (2467) and the smallest AIC value (65874) among the four LN classifications, which suggested implied that the LODDS classification models were more robust than other three models.

Table 1.

Multivariate Cox analysis for overall survival and cancer-specific mortality stratified by different N classifications.

Characteristics	AJCC N stage		pN stage		LNR stage		LODDS stage
Empty Cell	HR (95%CI)	p value	HR (95%CI)	p value	HR (95%CI)	p value	HR (95%CI)	p value
Different N Classification
N0/pN0/LNR0/LODDS1	Ref		Ref		Ref		Ref
N1/pN1/LNR1/LODDS2	1.91 (1.74–2.10)	<0.001	1.67 (1.52–1.83)	<0.001	1.84 (1.71–1.98)	<0.001	1.35 (1.24–1.47)	<0.001
N2a/pN2/LNR2/LODDS3	1.42 (1.19–1.71)	<0.001	2.20 (2.00–2.42)	<0.001	2.34 (2.13–2.57)	<0.001	2.22 (1.93–2.55)	<0.001
N2b/pN3/LNR3/LODDS4	2.58 (2.35–2.83)	<0.001	3.61 (3.20–4.07)	<0.001	3.52 (3.18–3.90)	<0.001	3.27 (2.94–3.64)	<0.001
N2c	3.18 (2.81–3.60)	<0.001
N3	2.38 (1.88–3.01)	<0.001
Primary site
Hypopharynx	Ref		Ref		Ref		Ref
Larynx	0.96 (0.77–1.20)	0.551	0.97 (0.82–1.15)	0.743	0.97 (0.80–1.18)	0.768	0.98 (0.79–1.21)	0.715
Oral Cavity	1.13 (0.95–1.34)	0.209	1.15 (0.96–1.38)	0.209	1.14 (0.98–1.33)	0.132	1.14 (0.96–1.33)	0.287
Oropharynx	0.40 (0.32–0.50)	<0.001	0.41 (0.34–0.49)	<0.001	0.42 (0.35–0.50)	<0.001	0.38 (0.28–0.52)	<0.001
Tumor Size		<0.001		<0.001		<0.001		<0.001
≤ 2.6 cm	Ref		Ref		Ref		Ref
>2.6 cm	1.12 (1.03–1.22)		1.13 (1.03–1.24)		1.11 (1.03–1.20)		1.13 (1.02–1.25)
T stage 7th
T1	Ref		Ref		Ref		Ref
T2	1.53 (1.40–1.67)	<0.001	1.53 (1.41–1.66)	<0.001	1.53 (1.42–1.65)	<0.001	1.56 (1.42–1.71)	<0.001
T3	2.26 (2.02–2.53)	<0.001	2.25 (2.02–2.51)	<0.001	2.28 (2.05–2.54)	<0.001	2.32 (2.09–2.57)	<0.001
T4a	2.48 (2.25–2.73)	<0.001	2.51 (2.27–2.78)	<0.001	2.56 (2.31–2.84)	<0.001	2.61 (2.35–2.90)	<0.001
T4b	3.26 (2.31–4.60)	<0.001	3.14 (2.22–4.44)	<0.001	3.15 (2.22–4.47)	<0.001	3.28 (2.32–4.64)	<0.001
Race
White	Ref		Ref		Ref		Ref
Black	1.08 (0.99–1.18)	0.201	1.06 (0.96–1.17)	0.136	1.11 (0.96–1.28)	0.123	1.08 (0.97–1.20)	0.316
Others	0.86 (0.75–0.99)	0.024	0.86 (0.77–0.96)	0.042	0.86 (0.76–0.97)	0.036	0.85 (0.74–0.98)	0.021
PRCDA Region
East	Ref		Ref		Ref		Ref
Northern Plains	0.92 (0.81–1.04)	0.148	0.95 (0.88–1.03)	0.143	0.96 (0.90–1.02)	0.201	0.94 (0.86–1.03)	0.144
Pacific Coast	0.91 (0.83–0.99)	0.002	0.87 (0.79–0.96)	0.003	0.92 (0.86–0.98)	0.003	0.92 (0.85–0.99)	0.012
Southwest	1.01 (0.86–1.19)	0.776	1.01 (0.87–1.17)	0.884	1.01 (0.89–1.15)	0.883	1.03 (0.89–1.19)	0.666
Histological Grade
G1	Ref		Ref		Ref		Ref
G2	1.14 (1.01–1.29)	0.021	1.12 (1.02–1.23)	0.038	1.13 (0.98–1.30)	0.112	1.18 (1.04–1.34)	0.005
G3	1.18 (1.05–1.33)	0.008	1.22 (0.97–1.53)	0.088	1.19 (1.02–1.39)	0.034	1.24 (1.11–1.39)	0.001
G4	1.89 (1.33–2.69)	0.012	1.78 (1.52–2.08)	0.005	1.88 (1.49–2.37)	0.014	1.96 (1.46–2.63)	0.003
GX	1.04 (0.86–1.26)	0.672	0.99 (0.88–1.12)	0.729	0.97 (0.80–1.18)	0.785	1.03 (0.86–1.23)	0.626
PORT
No	Ref		Ref		Ref		Ref
Yes	0.77 (0.68–0.87)	<0.001	0.75 (0.67–0.84)	<0.001	0.76 (0.66–0.88)	<0.001	0.81 (0.72–0.91)	<0.001
Chemotherapy
No	Ref		Ref		Ref		Ref
Yes	0.88 (0.83–0.93)	0.003	0.84 (0.77–0.92)	<0.001	0.87 (0.79–0.96)	0.001	0.92 (0.87–0.97)	0.011
Age at diagnosis	1.03 (1.02–1.03)	<0.001	1.03 (1.02–1.03)	<0.001	1.02 (1.02–1.03)	<0.001	1.04 (1.01–1.07)	<0.001
Marital status
No	Ref		Ref		Ref		Ref
NR	0.76 (0.65–0.89)	0.001	0.78 (0.63–0.96)	0.001	0.76 (0.69–0.84)	0.001	0.78 (0.66–0.92)	0.003
Yes	0.72 (0.67–0.77)	<0.001	0.75 (0.67–0.84)	<0.001	0.75 (0.69–0.82)	<0.001	0.75 (0.72–0.78)	<0.001
Model performance
R square	0.195		0.197		0.204		0.208
AIC	66023.55		66001.57		65932.57		65874.61
C-index	0.727 (se = 0.004)		0.728 (se = 0.004)		0.733 (se = 0.004)		0.736 (se = 0.004)
Likelihood ratio	2311		2335		2409		2467
Empty Cell	AJCC N stage		pN stage		LNR stage		LODDS stage
Empty Cell	HRs (95%CI)	p value	HRs (95%CI)	p value	HRs (95%CI)	p value	HRs (95%CI)	p value
Different N Classification
N0/pN0/LNR0/LODDS1	Ref		Ref		Ref		Ref
N1/pN1/LNR1/LODDS2	1.82 (1.60–2.07)	<0.001	1.62 (1.45–1.81)	<0.001	1.75 (1.54–1.99)	<0.001	1.29 (1.14–1.46)	<0.001
N2a/pN2/LNR2/LODDS3	1.37 (1.15–1.63)	<0.001	2.15 (1.98–2.33)	<0.001	2.24 (2.12–2.37)	<0.001	2.10 (1.88–2.35)	<0.001
N2b/pN3/LNR3/LODDS4	2.45 (2.13–2.82)	<0.001	3.45 (3.21–3.71)	<0.001	3.21 (2.68–3.84)	<0.001	3.18 (2.92–3.46)	<0.001
N2c	3.04 (2.71–3.41)	<0.001
N3	2.23 (1.75–2.84)	<0.001
Primary site
Hypopharynx	Ref		Ref		Ref		Ref
Larynx	0.87 (0.62–1.23)	0.342	0.88 (0.63–1.23)	0.665	0.88 (0.74–1.05)	0.556	0.94 (0.77–1.15)	0.624
Oral Cavity	1.08 (0.96–1.22)	0.124	1.23 (1.04–1.45)	0.013	1.06 (0.85–1.32)	0.262	1.09 (0.87–1.37)	0.177
Oropharynx	0.34 (0.17–0.69)	<0.001	0.36 (0.21–0.61)	<0.001	0.43 (0.30–0.61)	<0.001	0.36 (0.27–0.48)	<0.001
Tumor Size		<0.001		<0.001		<0.001		<0.001
≤ 2.6 cm	Ref		Ref		Ref		Ref
>2.6 cm	1.11 (1.02–1.21)		1.12 (1.03–1.22)		1.12 (1.04–1.21)		1.13 (1.03–1.24)
T stage 7th
T1	Ref		Ref		Ref		Ref
T2	1.62 (1.38–1.90)	<0.001	1.53 (1.44–1.63)	<0.001	1.51 (1.37–1.66)	<0.001	1.54 (1.46–1.62)	<0.001
T3	2.18 (1.97–2.41)	<0.001	2.23 (2.08–2.39)	<0.001	2.15 (1.92–2.41)	<0.001	2.32 (2.05–2.63)	<0.001
T4a	2.60 (2.31–2.93)	<0.001	2.62 (2.33–2.95)	<0.001	2.45 (2.12–2.83)	<0.001	2.73 (2.36–3.16)	<0.001
T4b	3.20 (2.36–4.34)	<0.001	3.21 (2.30–4.48)	<0.001	3.21 (2.30–4.48)	<0.001	3.30 (2.36–4.61)	<0.001
Race
White	Ref		Ref		Ref		Ref
Black	1.07 (0.95–1.21)	0.151	1.07 (0.95–1.21)	0.179	1.09 (0.97–1.22)	0.215	1.09 (0.96–1.24)	0.146
Others	0.77 (0.60–0.99)	0.012	0.89 (0.54–1.48)	0.127	0.87 (0.78–0.97)	0.016	0.86 (0.74–0.99)	0.014
PRCDA Region
East	Ref		Ref		Ref		Ref
Northern Plains	0.90 (0.80–1.01)	0.216	0.91 (0.80–1.03)	0.289	0.91 (0.80–1.02)	0.212	0.92 (0.78–1.08)	0.132
Pacific Coast	0.86 (0.80–0.93)	0.001	0.87 (0.79–0.96)	0.002	0.88 (0.82–0.95)	0.001	0.88 (0.80–0.97)	0.010
Southwest	1.02 (0.90–1.16)	0.706	1.02 (0.90–1.16)	0.775	1.03 (0.74–1.43)	0.651	1.04 (0.85–1.27)	0.716
Histological Grade
G1	Ref		Ref		Ref		Ref
G2	1.16 (1.04–1.29)	0.016	1.13 (0.94–1.36)	0.214	1.14 (1.03–1.26)	0.032	1.18 (1.05–1.33)	0.003
G3	1.19 (1.06–1.34)	0.001	1.25 (1.04–1.50)	0.014	1.24 (1.03–1.49)	0.042	1.24 (1.11–1.39)	0.002
G4	1.88 (1.14–3.10)	0.002	1.80 (1.57–2.06)	0.014	1.85 (1.54–2.22)	0.002	0.96 (0.66–1.40)	0.666
GX	1.02 (0.85–1.22)	0.675	0.88 (0.67–1.15)	0.543	0.96 (0.85–1.09)	0.446	1.13 (0.81–1.58)	0.542
PORT
No	Ref		Ref		Ref		Ref
Yes	0.76 (0.65–0.89)	<0.001	0.72 (0.64–0.81)	<0.001	0.73 (0.61–0.88)	<0.001	0.82 (0.74–0.91)	<0.001
Chemotherapy
No	Ref		Ref		Ref		Ref
Yes	0.88 (0.82–0.95)	0.001	0.86 (0.77–0.96)	<0.001	0.86 (0.76–0.97)	0.001	0.85 (0.75–0.96)	0.001
Age	1.02 (1.01–1.03)	<0.001	1.03 (1.02–1.04)	<0.001	1.02 (1.02–1.03)	<0.001	1.02 (1.02–1.03)	<0.001
Marital status
No	Ref		Ref		Ref		Ref
NR	0.68 (0.57–0.81)	0.002	0.76 (0.63–0.91)	0.001	0.75 (0.64–0.88)	0.001	0.76 (0.65–0.89)	0.001
Yes	0.73 (0.66–0.81)	<0.001	0.72 (0.63–0.82)	<0.001	0.72 (0.64–0.81)	<0.001	0.73 (0.65–0.82)	<0.001
Model performance
C-index	0.721 (se = 0.004)		0.727 (se = 0.005)		0.732 (se = 0.004)		0.736 (se = 0.00

Proposal and validation of novel stages

To further demonstrate the prognostic value of LODDS classification in HNSCC patients underwent surgery, a novel TN_LODDS staging system was proposed (see Appendix Table). As shown in Fig. 4, the novel TN_LODDS staging reflected discernible differences in OS, CSS and cancer-specific mortality in distribution of HNSCC patients (p < 0.001 for all). Finally, when the survival analyses were repeated for HNSCC patients grouped by the four primary sites (oral cavity, oropharynx, hypopharynx and larynx), the discriminative ability of the novel TN_LODDS staging was verified (eFigure 4).

Fig. 4.

OS (A), CSS (B) and CIF of cancer-specific mortality (C) for HNSCC patients grouped by the novel TNLODDS classification. OS: overall survival; CSS: cancer-specific survival; CIF: cumulative incidence function.

Discussion

Cervical lymph node metastases are a vital prognostic indicator for HNSCC, and it was very important to accurately estimate the survival based on LN status. To our knowledge, by using a lager cohort from the SEER database and selecting three different end points, we compare the predictive ability of four common used LN classifications in the survival of HNSCC patients and found that LODDS classification had the best prognostic ability and discriminatory capacity. Furthermore, based on the N_LODDS classification, a refined TN_LODDS stage was proposed. Finally, our study was also the first on to determine the threshold ELN number to accurately staging HNSCC, which showed that at least fifteen LNs should be examined during surgical resection.

To obtain an accurate pN stage, cervical lymph node dissection should be performed. However, the ELN number during surgery could influence the PLN number, the more LNs examined, the smaller the risk of undetected PLN, leading to “stage migration” [13]. Moreover, previous published studies suggested that positive correlation between a greater ELN number and better prognosis [7,8]. Thus, in our study we sought to determine the adequate threshold of ELN number to minimize the risk of stage migration and allow a more accurated determination of prognosis. In our study, after considering the above two issues, an optimal cut-point (15 ELNs) was determined with good discrimintation ability of both survival and stage migration.

After excluding patients with non-adequate number of ELN, to guarantee the quality of surgery; our study further compared the prognostic ability of the four LN classifications, namely, the AJCC N stage, pN, LNR and LODDS classifications. Several possible underlying reasons for the best prognostic ability of LODDS exist. First of all, all of those with no LN metastasis belonged to the same stage measured by AJCC N stage (N0), pN stage (pN0) or LNR stage (LNR0) classification. In contrast, the LODDS classification could stratify these patients (eFigure 3). Secondly, according to our study, for patients with one PLN out of six ELN, the LN stage could move from pN1 stage (PLN = 1) to LNR2 stage (LNR = 0.167) and even to LODDS4 stage (LODDS = −0.564). Thus, the 5-year OS was estimated to be significantly different for these patients (pN1 stage, 61.5%; LNR2 stage, 49.4%; LODDS4 stage, 35.1%). It demonstrated that computation of the LODDS was able to change prognosis for individual patients in a meaningful way, given the distinct prognosis estimation by different classifications, which we observed.

The following limitations merits consideration. First, we were limited by the retrospective nature of this study and the inherent limitations in that study design. According to the AJCC 8th staging system, the LN classification has been revised, with extra-nodal extension (ENE) incorporated. Similar to other SEER-based studies, it is not possible to restage HNSCC patients according to the recent edition of AJCC staging system because of lack of information regarding ENE and HPV status [4]. However, recently published studies addressed several limitations of AJCC 8th N classification, which including the doubtful value of bilateral nodal disease and poor discrimination between N2 sub-stages [14]. Moreover, this staging system was suggested to be a crude category and failed to be a prognostic predictor for patients originated from the oropharynx [15,16]. In our study, the prognoses for AJCC N1 and N3 stage were comparable, which also supported the doubt of the prognostic ability of AJCC N classification. On the contrary, the proposed TN_LODDS staging system performed well with a discriminatory ability for the whole cohort but also showed good prognostic ability when stratified by different primary sites. Furthermore, the great sample size of this work provided adequate statistical power to estimate the prognostic ability of four LN classifications. Secondly, the information about location of metastatic lymph node was not included in the SEER database. Last but not the least, the details of surgical resection, especially the exact ELN number for specific nodal level was not provided by the SEER database, which makes it impossible for us to figure out the threshold ELN number of each LN level.

In conclusion, our study determined that fifteen ELNs as the threshold to obtain an adequate cervical lymph node dissection for HNSCC patients. Comprehensively comparison showed LODDS classification was more accurate in predicting survival than AJCC N stage, pN stage and LNR classification in HNSCC patients receiving surgery, especially for those with no PLN involved. The proposed TN_LODDS staging system based on LODDS classification could serve as an effective evaluation tool in predicting the postoperative prognosis of patients with HNSCC.

Abbreviations:

HNSCC

Head and neck squamous cell carcinoma

AJCC

the American Joint Committee on Cancer

LN

Lymph node

ELN

Examined lymph node

PLN

Positive lymph node

LNR

Lymph node ratio

LODDS

log odds ratio

SEER

the Surveillance, Epidemiology, and End Results

OS

Overall survival

CSS

Cancer-specific survival

OR

Odds ratio

HR

Hazard ratio

sHR

subdistribution hazard ratio

CI

Confidence interval

C-index

Harrell’s concordance index

Appendix

Table A1.

Proposed novel TN_LOODSM staging system for HNSCC patients.

Empty Cell	LODDS1	LODDS2	LODDS3	LODDS4
T1	stage I	stage I	stage II	stage II
T2	stage I	stage II	stage II	stage III
T3	stage II	stage III	stage III	stage IV
T4	stage II	stage III	stage IV	stage IV

Figure A1.

Flow chart illustrating how the inclusion and exclusion criteria were applied to achieve the eligible patients included in the analysis.

Data availability statement

The permission to access the SEER database was received from the National Cancer Institute (the private SEER ID 10425-Nov2018).