Relationship between level of lymph node metastasis and survival in locally advanced head and neck squamous cell carcinoma

Abstract

Background

The current head and neck squamous cell carcinoma (HNSCC) staging system may not capture the full prognostic implications of regional lymph node involvement. We sought to investigate the impact of level of lymph node metastasis (LNM) on survival

Methods

The Surveillance, Epidemiology, and End Results (SEER) registry was queried for oral cavity (OC), oropharynx (OP), larynx (LAR), and hypopharynx (HP) HNSCC. Multivariate Cox proportional hazards model was used to evaluate whether level of LNM is an independent prognostic factor. Site-specific recursive-partitioning analysis (RPA) was performed to classify patients into different risk groups.

Results

Totally, 14,499 patients including OC (N=2,463), OP (N=8,567), LAR (N=2,332) and HP (N=1,137) were analyzed. Both AJCC N classification and level of LNM showed significant effects on overall survival (OS) in patients with OC, OP or LAR, but not in HP. In patients with N2 disease, AJCC subclassification (N2a, N2b, N2c) was significantly associated with OS of patients with OP and LAR, but not OC or HP, while level of LNM (primary, secondary, and tertiary) was significantly associated with OS in patients with OC, OP and LAR, but not HP. Using RPA, we designed a simple, primary site-specific prognostic tool integrating AJCC T classification, N classification, and level of LNM that can be easily utilized by health care providers in clinic.

Conclusions

Level of LNM is an independent prognostic factor for patients with locally advanced HNSCC and could add to the prognostic value of AJCC T and N classification in patients with locally advanced HNSCC.

Introduction

More than 50% of head and neck squamous cell carcinoma (HNSCC) patients present with regional lymph node involvement at the time of diagnosis.^{1, 2} Lymph node metastasis (LNM) is associated with poor survival,^2-5 and is one of the most important factors for determination of appropriate treatment.⁶

Previous studies have shown that patterns of lymph node drainage differ among HNSCC subsites.^7-9 While many studies have confirmed that extent of LNM is an important prognostic factor for locally advanced HNSCC, as reflected in the American Joint Committee on Cancer (AJCC) staging system,^{10, 11} little is known about the impact of level of LNM on HNSCC survival.

We hypothesized that level of LNM is an independent prognostic factor in patients with locally advanced HNSCC. To evaluate our hypothesis, we examined the impact of level of LNM on OS in patients with HNSCC of oral cavity (OC), oropharynx (OP), larynx (LAR), and hypopharynx (HP) using the Surveillance, Epidemiology, and End Results (SEER) registry.

Patients and Methods

The SEER registry was queried for HNSCC diagnosed from 2004-2009. HNSCC cases from four subsites (OC, OP, LAR and HP) were characterized as regards to age, race, T classification (AJCC), N classification (AJCC), level of LNM, and treatment (surgery and/or radiation therapy). Within our study period, there were 58,734 patients who had a pathologically confirmed diagnosis of HNSCC from these four subsites: OC, OP, LAR and HP. Cases were eligible for further detailed analysis if they fulfilled all of the following criteria: (1) lymph node positive diseases (≥N1); (2) squamous cell carcinoma histology (ICD-O-3 histology codes 8070-8076, 8078); (3) absence of LNM at levels VI, VII, facial, parotid, and subocciptal areas; (4) absence of distant metastases. Cases were excluded if there were missing or unknown data (codes ‘999’ or ‘988’) of any of the variables captured by SEER that would not allow level of LNM to be clearly defined.

Data checks were conducted to exclude patients with inconsistent information between N classification (as captured by SEER under variables AJCC_N_stage) and level of LNM (as captured by SEER under variables CS_SSF3, CS_SSF4, CS_SSF5 and CS_SSF6). The resultant final study cohort comprised 14,499 locally advanced HNSCC with AJCC 7^th edition stages III, IVA, and IVB, with full characterization of LNM confined to levels I-V.¹⁰

Definition of level of LNM for each subsite

Patterns of lymph node drainage differ among HNSCC subsites. Regional lymphadenopathy was classified as primary, secondary or tertiary LNM based on the frequencies of nodal involvement within each subsite of HNSCC. In brief, we analyzed previously published, independent data sets ^{8, 12, 13} and classified LNM as primary, secondary or tertiary according to frequencies of lymph node involvement (highest to lowest) in each subsite. We then confirmed that the pattern of nodal spread previously reported held true in our SEER dataset (Supplemental Table S1, online only), and used these pre-defined criteria for all subsequent analyses. The consistency of the SEER findings with the previously reported data increases confidence on the accuracy of the data collected in the SEER registry. As such, level 5 was considered a tertiary level of LNM for all subsites (OC, OP, LAR, HP). Level 4 was considered secondary level of LNM for all subsites. Levels 2 and 3 were considered primary levels of LNM for all subsites. Level 1 was considered primary level of LNM for OC, and secondary level of LNM for OP, LAR and HP. It should be noted that the terms primary, secondary and tertiary may be an oversimplification of the biology that determines LNM and might be best interpreted as preferential levels of cancer spread, as opposed to a (suboptimal) reflection of anatomic relationships.

Statistical analysis

The distribution of OS - defined as time from diagnosis to death or the time of last contact - was estimated by the Kaplan-Meier method.¹⁴ Log-rank test was performed to test the difference in survival between groups.¹⁵ Regression analyses of survival data based on the Cox proportional hazards model were conducted on OS.¹⁶ The proportional hazards assumption for the Cox proportional hazards model was evaluated. No major violations were detected. Regression diagnostics (e.g., generalized residuals, Martingale residuals, and Shoenfeld residuals) were examined to ensure that the models were appropriate. All the analyses only included complete cases on all of the variables. Wald test was used to estimate the 95% confidence interval of hazard ratios. Recursive-partitioning analysis (for censored survival data) was performed using S-Tree software (http://c2s2.yale.edu/software/stree/) for all four subsites separately using T classification, N classification and level of LNM. Patients were classified as having a low, (intermediate-low), intermediate, (intermediate-high), or high risk of death accordingly.¹⁷

Results

The distribution of eligible cases (N=14,499) according to primary tumor site was as follows: 2,463 OC, 8,567 OP, 2,332 LAR, and 1,137 HP. Median follow-up times for living patients were 1.75 years for OC, 2.17 years for OP, 1.92 years for LAR and 1.67 years for HP, respectively.

Survival impact of AJCC N classification versus level of LNM

The effect of AJCC N classification on survival is illustrated by the Kaplan-Meier curves in Figs 1A, 1C, 1E and 1G. A separation of the curves between N1, N2 and N3 subgroups was clearly evident for LAR subsite (Fig 1E). However, overlaps between the N2 and N3 curves were observed in OC (especially after 18 months of follow up – Fig 1A), and overlaps between the N1 and N2 curves were evident for OP (Fig 1C) and HP (Fig 1G). These results indicated that while there is some prognostic effect of the AJCC N classification, this variable alone suboptimally predicts OS in a large portion of HNSCC patients.

Figure 1. Kaplan-Meier survival plots and multivariate Cox proportional hazard model for patients with locally advanced HNSCC.

Survival plots show OS according to 7^th AJCC N classification (A, C, E, G) or level of LNM (B, D, F, H). Hazard ratios (HR), 95% confidence interval (95% CI) and p-values were computed using multivariate Cox model.

We then evaluated the effects of level of LNM on OS in each of the subsites (Figs 1B, 1D, 1F, and 1H). The results demonstrated a clear separation of the curves according to primary, secondary, and tertiary level of LNM status in all subsites except HP, raising the hypothesis that level of LNM may be a more accurate predictor of survival compared to AJCC N classification, and/or may add to the prognostic value of the already established AJCC N classification. To evaluate whether level of LNM and AJCC N classification is an independent prognostic factor in our patient population, we performed multivariate analysis including the aforementioned two variables in addition to other known/putative factors that could impact survival (e.g., age, gender, race, AJCC T classification, and treatment) in each of the four subsites (Table 1). The results demonstrated that both AJCC N classification and level of LNM were independent prognostic factors in OC (p=0.0008 and p=0.0008, respectively), LAR (p<0.0001 and p=0.0068, respectively) and OP (p=0.0357 and p<0.0001 respectively). Neither AJCC N classification nor level of LNM was an independent prognostic factor in HP in our multivariate model (p=0.3363, and p=0.3960, respectively). AJCC T classification was associated with survival in all four subsites. Because level of LNM may be more accurately captured in patients that undergo surgical resection as part of their treatment, we performed the same multivariate analysis including only the subset of surgically resected patients. Similar to the data in the general population, level of LNM was an independent prognostic factor for survival in OC (p=0.0012), LAR (p=0.0008), OP (p<0.0001), but not HP (p=0.1480) [Supplemental Table S2, online only].

Table 1. Multivariate Cox proportional hazard model of OS for patients with HNSCC.

	Demographic Parameters		All patients
			HR	95% CI		p
OC	Age	7 (80 -) vs. <30	3.726	1.830	7.589	<0.0001
		6 (70,80) vs. <30	2.487	1.224	5.051
		5 (60,70) vs. <30	1.637	0.807	3.322
		4 (50,60) vs. <30	1.393	0.687	2.824
		3 (40,50) vs. <30	1.428	0.696	2.928
		2 (30,40) vs. <30	0.930	0.397	2.178
	Race	Asian/Pacific Islander vs. Caucasian	0.911	0.711	1.167	0.0172
		Hispanic vs. Caucasian	0.912	0.715	1.163
		African American vs. Caucasian	1.299	1.084	1.557
	AJCC T classification	T4 vs. T1	2.398	2.007	2.864	<0.0001
		T3 vs. T1	2.282	1.865	2.791
		T2 vs. T1	1.556	1.306	1.855
	AJCC N classification	N3 vs. N1	1.165	0.818	1.658	0.0008
		N2 vs. N1	1.274	1.123	1.445
	Level of LNM	Tertiary vs. Primary	1.427	1.124	1.812	0.0008
		Secondary vs. Primary	1.294	1.072	1.563
	Radiotherapy	Yes vs. No	0.664	0.581	0.759
	Surgery	Yes vs. No	0.593	0.524	0.670
OP	Age	7 (80 -) vs. <40	4.148	2.684	6.412	<0.0001
		6 (70,80) vs. <40	2.861	1.888	4.334
		5 (60,70) vs. <40	1.506	0.999	2.272
		4 (50,60) vs. <40	1.176	0.780	1.771
		3 (40,50) vs. <40	0.986	0.644	1.510
	Race	Asian/Pacific Islander vs. Caucasian	1.035	0.754	1.422	<0.0001
		Hispanic vs. Caucasian	0.929	0.740	1.165
		African American vs. Caucasian	1.918	1.690	2.176
	AJCC T classification	T4 vs. T1	3.619	3.110	4.213	<0.0001
		T3 vs. T1	2.512	2.113	2.986
		T2 vs. T1	1.498	1.283	1.749
	AJCC N classification	N3 vs. N1	1.237	1.013	1.511	0.0357
		N2 vs. N1	0.965	0.869	1.070
	Level of LNM	Tertiary vs. Primary	1.228	1.053	1.433	<0.0001
		Secondary vs. Primary	1.249	1.127	1.384
	Radiotherapy	Yes vs. No	0.376	0.331	0.428
	Surgery	Yes vs. No	0.692	0.612	0.782
LAR	Age	7 (80 -) vs. <40	3.919	1.806	8.501	<0.0001
		6 (70,80) vs. <40	2.777	1.304	5.917
		5 (60,70) vs. <40	1.968	0.929	4.170
		4 (50,60) vs. <40	1.621	0.764	3.438
		3 (40,50) vs. <40	1.419	0.658	3.064
	Race	Asian/Pacific Islander vs. Caucasian	0.715	0.457	1.119	0.0106
		Hispanic vs. Caucasian	0.798	0.609	1.046
		African American vs. Caucasian	1.197	1.026	1.397
	AJCC T classification	T4 vs. T1	2.166	1.711	2.742	<0.0001
		T3 vs. T1	1.512	1.198	1.908
		T2 vs. T1	1.293	1.022	1.637
	AJCC N classification	N3 vs. N1	1.471	1.116	1.940	<0.0001
		N2 vs. N1	1.365	1.197	1.557
	Level of LNM	Tertiary vs. Primary	1.255	1.032	1.526	0.0068
		Secondary vs. Primary	1.203	1.054	1.374
	Radiotherapy	Yes vs. No	0.401	0.343	0.468
	Surgery	Yes vs. No	0.645	0.558	0.746
HP	Age	7 (80 -) vs. <50	2.516	1.715	3.691	<0.0001
		6 (70,80) vs. <50	1.405	1.010	1.955
		5 (60,70) vs. <50	1.114	0.809	1.534
		4 (50,60) vs. <50	0.923	0.668	1.277
	Race	Asian/Pacific Islander vs. Caucasian	0.735	0.483	1.117	0.0004
		Hispanic vs. Caucasian	0.594	0.378	0.932
		African American vs. Caucasian	1.362	1.099	1.687
	AJCC T classification	T4 vs. T1	3.186	2.119	4.790	<0.0001
		T3 vs. T1	2.352	1.528	3.619
		T2 vs. T1	1.811	1.204	2.724
	AJCC N classification	N3 vs. N1	1.323	0.911	1.922	0.3363
		N2 vs. N1	1.058	0.879	1.274
	Level of LNM	Tertiary vs. Primary	1.196	0.923	1.551	0.3960
		Secondary vs. Primary	1.052	0.866	1.279
	Radiotherapy	Yes vs. No	0.318	0.254	0.399
	Surgery	Yes vs. No	0.547	0.416	0.719

Survival impact of AJCC N subclassification versus level of LNM in patients with N2 disease

HNSCC with N2 lymph node involvement represents one of the most diverse groups of patients. In the AJCC staging system, N2 is divided into N2a, N2b and N2c subclasses based on number, size, and laterality of lymph nodes involved.¹⁰ To investigate if level of LNM provides additional prognostic information in this specific group of patients, we contrasted the effects on survival of the AJCC N2 subclassification and level of LNM within the N2 subgroup only.

As illustrated by the Kaplan-Meier curves in Fig 2, a separation of the curves between AJCC N2a, N2b and N2c subgroups was clearly evident for the OP subsite (Fig 2C). However, overlaps between the N2a and N2b curves were observed in OC (Fig 2A) and HP (Fig 2G), and N2b and N2c curves seem to overlap for LAR (Fig 2E). These results indicated that the AJCC N subclassification does not optimally predict OS in OC, LAR and HP patients with N2 disease. On the other hand, a separation of the curves between primary, secondary, and tertiary level of LNM status was observed in all subsites except HP (Figs 2B, 2D, 2F and 2H), suggesting level of LNM may be a more accurate predictor of survival compared to AJCC N subclassification in HNSCC patients with N2 disease (especially for LAR and OC).

Figure 2. Kaplan-Meier survival plots and multivariate Cox proportional hazard model for patients with locally advanced N2 HNSCC.

Survival plots show OS according to 7^th AJCC N classification (A, C, E, G) or level of LNM (B, D, F, H). Hazard ratios (HR), 95% confidence interval (95% CI) and p-values were computed using multivariate Cox model.

On multivariate analysis within the N2 subgroup (Table 2), AJCC N2 subclassification was associated with survival in OP (p<0.0001) and LAR (p=0.012), but not in OC (p=0.48), or HP (p=0.07). On the other hand, the association of level of LNM with OS was statistically significant in OP (p=0.0041), OC (p=0.0001), and LAR (p=0.02), but not in HP (p=0.24). These data again suggest that level of LNM may improve prognostication within the N2 subgroup of patients with HNSCC.

Table 2. Multivariate Cox proportional hazard model of OS for patients with N2 HNSCC.

	Demographic Parameters		N2 only Disease
			HR	95% CI		p
OC	Age	7 (80 -) vs. <30	3.202	1.390	7.375	<0.0001
		6 (70,80) vs. <30	2.216	0.966	5.085
		5 (60,70) vs. <30	1.559	0.683	3.557
		4 (50,60) vs. <30	1.198	0.526	2.732
		3 (40,50) vs. <30	1.178	0.507	2.740
		2 (30,40) vs. <30	0.721	0.254	2.042
	Race	Asian/Pacific Islander vs. Caucasian	0.941	0.676	1.310	0.025
		Hispanic vs. Caucasian	0.852	0.604	1.203
		African American vs. Caucasian	1.372	1.092	1.725
	AJCC T classification	T4 vs. T1	2.205	1.725	2.820	<0.0001
		T3 vs. T1	2.281	1.718	3.027
		T2 vs. T1	1.510	1.168	1.951
	AJCC N classification	N2c vs. N2a	1.055	0.774	1.439	0.4821
		N2b vs. N2a	0.941	0.706	1.254
	Level of LNM	Tertiary vs. Primary	1.544	1.166	2.043	0.0001
		Secondary vs. Primary	1.464	1.173	1.826
	Radiotherapy	Yes vs. No	0.555	0.461	0.668
	Surgery	Yes vs. No	0.618	0.522	0.731
OP	Age	7 (80 -) vs. <40	3.133	1.912	5.133	<0.0001
		6 (70,80) vs. <40	2.168	1.368	3.437
		5 (60,70) vs. <40	1.078	0.685	1.698
		4 (50,60) vs. <40	0.924	0.588	1.452
		3 (40,50) vs. <40	0.734	0.457	1.179
	Race	Asian/Pacific Islander vs. Caucasian	1.062	0.712	1.586	<0.0001
		Hispanic vs. Caucasian	0.856	0.642	1.141
		African American vs. Caucasian	1.842	1.566	2.166
	AJCC T classification	T4 vs. T1	3.536	2.881	4.340	<0.0001
		T3 vs. T1	2.460	1.955	3.095
		T2 vs. T1	1.472	1.195	1.813
	AJCC N classification	N2c vs. N2a	1.606	1.324	1.947	<0.0001
		N2b vs. N2a	1.235	1.032	1.479
	Level of LNM	Tertiary vs. Primary	1.116	0.927	1.343	0.0041
		Secondary vs. Primary	1.252	1.096	1.430
	Radiotherapy	Yes vs. No	0.416	0.350	0.495
	Surgery	Yes vs. No	0.750	0.638	0.881
LAR	Age	7 (80 -) vs. <40	7.563	1.823	31.385	<0.0001
		6 (70,80) vs. <40	6.061	1.491	24.634
		5 (60,70) vs. <40	4.199	1.038	16.990
		4 (50,60) vs. <40	3.529	0.872	14.284
		3 (40,50) vs. <40	3.369	0.822	13.815
	Race	Asian/Pacific Islander vs. Caucasian	0.837	0.485	1.444	0.0349
		Hispanic vs. Caucasian	0.704	0.476	1.043
		African American vs. Caucasian	1.223	1.008	1.484
	AJCC T classification	T4 vs. T1	2.361	1.692	3.295	<0.0001
		T3 vs. T1	1.617	1.160	2.252
		T2 vs. T1	1.310	0.932	1.840
	AJCC N classification	N2c vs. N2a	0.899	0.686	1.179	0.0120
		N2b vs. N2a	0.722	0.550	0.948
	Level of LNM	Tertiary vs. Primary	1.292	1.020	1.636	0.0203
		Secondary vs. Primary	1.241	1.040	1.481
	Radiotherapy	Yes vs. No	0.385	0.313	0.474
	Surgery	Yes vs. No	0.698	0.582	0.836
HP	Age	7 (80 -) vs. <50	2.134	1.263	3.604	<0.0001
		6 (70,80) vs. <50	1.385	0.900	2.131
		5 (60,70) vs. <50	0.911	0.603	1.377
		4 (50,60) vs. <50	0.709	0.462	1.090
	Race	Asian/Pacific Islander vs. Caucasian	0.826	0.501	1.362	0.0080
		Hispanic vs. Caucasian	0.527	0.284	0.979
		African American vs. Caucasian	1.393	1.045	1.857
	AJCC T classification	T4 vs. T1	3.026	1.774	5.160	0.0001
		T3 vs. T1	2.087	1.192	3.656
		T2 vs. T1	1.950	1.154	3.294
	AJCC N classification	N2c vs. N2a	1.109	0.776	1.584	0.0700
		N2b vs. N2a	0.820	0.594	1.133
	Level of LNM	Tertiary vs. Primary	1.315	0.938	1.843	0.2406
		Secondary vs. Primary	1.146	0.886	1.482
	Radiotherapy	Yes vs. No	0.319	0.233	0.437
	Surgery	Yes vs. No	0.565	0.396	0.806

Recursive-partitioning analysis

Site-specific recursive-partitioning analysis (RPA) was performed to identify prognostic factors with the most significance in a proportional-hazards model of OS. The analyses allowed us to classify patients into categories of low, (low-intermediate), intermediate, (intermediate-high), or high risk of death on the basis of AJCC T classification, AJCC N classification, and level of LNM. This approach provides a guide to integrate these three complex variables in a prognostic tool that can be easily utilized by health care providers.

For all four sites, T classification was the most important determinant of OS, followed by level of LNM for OC (within the T2 subgroup, Fig 3A), OP (within the T3 subgroup, Fig 3D) and HP (within the T2/3 subgroup, Fig 3G). Patients with OC were classified into three categories with respect to the risk of death: low risk, with a 3-year rate of OS of 55%; intermediate risk, with a 3-year rate of 48.2% (hazard ratio for the comparison with low risk, 1.41; 95% CI, 1.17 to 1.7, p=0.001); and high risk, with a 3-year rate of 28.3% (hazard ratio for the comparison with low risk, 2.43; 95% CI, 2.06 to 2.85, p<0.001) (Fig 3B). OP cases were classified into the following five categories: low risk, with a 3-year rate of OS of 84.8%; low-intermediate risk, with a 3-year rate of 76.4% (hazard ratio for the comparison with low risk, 1.51; 95% CI, 1.27 to 1.81, p<0.001); intermediate risk, with a 3-year rate of 67.3% (hazard ratio for the comparison with low risk, 2.11; 95% CI, 1.71 to 2.61, p<0.001); intermediate-high risk, with a 3-year rate of 58.4% (hazard ratio for the comparison with low risk, 2.82; 95% CI, 2.33 to 3.4, p<0.001); and high risk, with a 3-year rate of 47.2% (hazard ratio for the comparison with low risk, 4.1; 95% CI, 3.46 to 4.86, p<0.001) (Fig 3D). LAR cases were classified into three risk groups: low risk, with a 3-year rate of OS of 60.7%; intermediate risk, with a 3-year rate of 49.3% (hazard ratio for the comparison with low risk, 1.49; 95% CI, 1.14 to 1.95, p=0.004); and high risk, with a 3-year rate of 36.9% (hazard ratio for the comparison with low risk, 2.43; 95% CI, 1.85 to 3.19, p<0.001) (Fig 3F). For HP, the 3-year OS for low risk group was 59.3%; for intermediate risk 42.8% (hazard ratio for the comparison with low risk, 1.84; 95% CI, 1.23 to 2.75, p=0.003); and for high risk, 29.8% (hazard ratio for the comparison with low risk, 3.09; 95% CI, 2.06 to 4.61, p<0.001) (Fig 3H). Of note, AJCC N classification was not a determinant of OS for HP.

Figure 3. Site - specific risk classification for HNSCC and associated Kaplan–Meier survival.

A, C, E and G demonstrate the flow chart of resulting classifications. The number of cases in each risk group and its proportion of the total number of cases are shown under each risk group. LNM: P/S/T indicate level of lymph node metastasis to primary/secondary/tertiary groups respectively. B, D, F and H demonstrate the OS in the classified patients for OC, OP, LAR and HP respectively.

Discussion

In this SEER registry-based study, we demonstrated that level of LNM is an independent prognostic factor for survival in patients with locally advanced HNSCC (especially for patients with OC, OP and LAR cancers) and within the subgroup of patients with N2 disease, thus suggesting that level of LNM may add to the prognostic value of already established AJCC factors. Using RPA, we were able to design a primary site-specific prognostic tool integrating AJCC T classification, AJCC N classification, and level of LNM that can be easily utilized by health care providers.

The current AJCC N classification system is mainly based on number, size, and laterality of involved cervical lymph nodes. However, prognostication of HNSCC based on this system may not always capture the full implications of regional lymph node spread on OS. Indeed, in our analysis, survival curves for AJCC N1, N2 and/or N3 groups overlapped, to some extent, in patients with OC, OP, and HP primary sites. The HP cohort had the smallest number of patients (N=1,137) and the limited sample size could explain the lack of statistical significance of differences in survival according to the AJCC N classification. However, the cohort included 2,463 patients with OC and 8,567 patients with OP, which would provide a reasonable power to detect major differences in survival between the groups defined by AJCC N classification. As such, the data indicate that optimization of categorization of lymph node involvement is needed to improve prognostication.

The limitations of the current staging system in determining survival for patients with OP cancers has been documented by several groups,^{18, 19} largely due to the favorable prognostic effects of HPV-related cancers in this subsite. The incidence of HPV-related OP cancers has been increasing in the United States, and past HPV infections are now estimated to account for approximately 70% of all newly diagnosed OP cancer cases.^20-22 HPV-related cancers typically present with advanced nodal involvement and small primary tumors. Not surprisingly, recent series of survival analysis of OP cancers not stratified by HPV status have shown worse outcomes for N0 patients (and more frequently HPV-negative tumors), than patients with lymph node involvement (and more frequently HPV-positive tumors).^18,19 A more accurate way of determining the survival impact of lymph node involvement, if any, in OP (and largely HPV-positive) cancers is therefore clearly needed. As a result, a new prognostic model has been proposed for HPV-related oropharynx cancer, which takes into account TNM, as well as smoking status and age.²³ Unfortunately, the SEER database does not capture HPV status of HNSCC, or smoking history. Nonetheless, patients included in our analysis were diagnosed between 2004 and 2009, an era during which the frequency of HPV-related cancers was already high. Moreover, our analysis only focuses on node positive cancers. As such, it is very likely that our study population of OP cancers may largely represent HPV-related malignancies. Nonetheless, our findings in OP need to be confirmed in other cohorts with known HPV status.

Previous studies have evaluated the effects of level of LNM on outcomes. Our single-institution data (N=776 patients) demonstrated shorter survival in patients with OP cancers with lower level cervical lymph node involvement treated with intensity-modulated radiation therapy.²⁴ Li et al. demonstrated that the number of cervical levels with lymph node involvement was associated with a higher risk of distant metastases (N=391).²⁵ Based on these single institution experiences, we designed the current population-based study to investigate the impact level of LNM on survival, taking advantage of the more detailed collection of this information by the SEER database since 2004 at a large scale. SEER currently collects and publishes cancer incidence and survival data from population-based cancer registries covering approximately 28 percent of the US population. In addition, SEER provides information of patients from across the country in many different treatment settings, both community and academic practices, which makes our analysis more representative of the general US patient population. Because we included surgically and non-surgically treated patients, it is likely that the data obtained is broadly applicable for most clinical scenarios.

Our results showed level of LNM is an independent prognostic factor for patients with locally advanced HNSCC of OC, OP or LAR even after adjustment for AJCC N classification. We also demonstrated the limited prognostic value of the AJCC N2a, N2b, N2c subclassification in OC and LAR cancers. In contrast, level of LNM within N2 disease could predict survival for OP, OC and LAR subsites. The lack of association of nodal involvement in HP with survival (whether assessed by the AJCC N classification or level of LNM classification proposed herein) could possibly be explained by a different biology of this tumor, reflecting distinct molecular signatures²⁶ leading to aggressive clinical behavior less dependent on patterns of nodal spread.

The strengths of our analysis include: (1) the use of a large, US-representative sample, encompassing patients with multiple treatment modalities; (2) evaluation of the four major HNSCC primary sites individually, thus taking into account differences in natural history, treatment patterns and outcomes between these groups; (3) construction of an easy-to-use prediction tool that could integrate the complex variables of AJCC T classification, N classification, and level of LNM metastases. This tool was able to clearly separate risk groups within each of the four subsites and can be readily applied to patients in clinical practice. An example of how the use of such tool could improve prognostication is illustrated by the following scenario: a patient with a T2N2c LAR cancer would be classified as stage IVA by the AJCC 7^th edition. According to our tool, this patient could be classified within the intermediate or high risk groups, according to involvement of primary or secondary/tertiary levels of LNM, respectively. The 3-year survival for these two groups are estimated at 49% and 37%, respectively, representing clinically meaningful differences.

The limitations of our analysis include: (1) the relatively short median follow-up time for the cohort; (2) the lack of HPV data in SEER, particularly for OP cancers; (3) the lack of information on patterns of failure (i.e., locoregional recurrence, distant metastases, and development of second primary tumors); (4) the lack of details on treatment and pathology such as margins, perineural invasion, and extracapsular invasion. Recent studies have demonstrated that while HPV-related OP cancers have improved survival compared to non-HPV-related OP cancers, there are clearly subgroups with HPV-related HNSCC that are at increased risk of recurrence and death following definitive treatment. Smoking history, for example, has been consistently associated with worse prognosis in HPV-related OP cancers.^18-20 O'Sullivan et al. have demonstrated that the presence of T4 tumors and N3 disease were also associated with an increased risk of distant metastases and death in 505 patients with HPV-related OP cancers.²⁷ These data illustrate that refinement of prognostication using clinical and molecular criteria could assist with development of treatment strategies tailored to each patient's individual risk. Moreover, the ability to predict the risk of locoregional recurrence and distant recurrence using clinical, demographic or molecular criteria could lead to the study of treatment approaches to intensify local or systemic therapies in patients specifically at risk for locoregional or distant relapse, respectively. We speculate that, in the near future, novel molecular markers may be integrated to known prognostic factors to better assess each patient's individual risk of recurrence and death. As an example, preliminary data from our group demonstrate that genomic intra-tumor heterogeneity may be associated with higher risk of recurrence in early stage non-small cell lung cancer.²⁸ Intra-tumoral genetic heterogeneity in HNSCC has also been associated with worse survival in a small cohort of patients.²⁹ As these and other novel prognostic markers evolve, the development of prediction tools that can integrate detailed molecular, clinical and treatment outcomes data are likely to become part of the standard evaluation of patients in clinic.

Another limitation of our study is the lack of quality control of the data captured by SEER, with potential for misclassification of level of LNM, such as inaccurate assessment by radiologist (or false positive findings resulting from reactive nodes in non-surgically treated patients), incomplete nodal dissection, or inaccurate processing and/or labeling of surgical specimens by surgeons and/or pathologists. Nonetheless, a recent analysis demonstrated high concordance rates between lung cancer histology data captured by SEER with independent pathology review of the same data, thus providing further evidence for reasonable accuracy of one complex data aspect captured by SEER.³⁰ We caution, however, that results from any SEER analysis would be more robust (including the findings presented herein) if confirmed in independent, carefully curated data sets, and we strongly recommend validation of our findings by other groups before routine use in clinic. As an example of such framework, the SEER database was utilized as a validation tool for the revised TNM classification of non-small cell lung cancers originally proposed based on an independent international database and ultimately incorporated in the AJCC 7^th edition. This approach illustrates the significant potential for utilizing the SEER registry as a powerful tool to identify and/or validate novel prognostic factors.³¹

In conclusion, we demonstrated that level of LNM could add to the prognostic value of AJCC T and N classification in patients with locally advanced HNSCC. As we approach discussions regarding TNM classification of HNSCC for the AJCC 8^th edition, considerations should be given to refining the currently suboptimal N classification, both within HPV-related and HPV-unrelated HNSCC. Comprehensive data collection and analysis regarding novel clinical and molecular prognostic factors, as increasingly supported by SEER,³² will assist clinicians in improving risk assessment and developing user-friendly tools (such as the one presented herein) that can inform clinical decision making.