Nasopharyngeal carcinoma (NPC) is one of the most common head and neck cancers in China and Southeast-Asia. The last decade has seen significant improvements on the loco-regional control rate of NPC owing to the advances in treatment modality and techniques. However, distant metastasis (DM) remains common and has become the major cause of mortality for NPC [
The current N classification system does have limitations. Firstly, based on two-categorical nodal laterality, level, and size, the N classification may miss the importance of quantitative lymph node (LN) burden; for instance, patients with extensive metastatic LNs could be staged the same as those with single LN, yet they empirically fare much poorer prognosis. Secondly, the use of multiple parameters may bring more confounders and increase the interobserver inconsistency of N classification. Meanwhile, the use of two-category variables may cause vital loss of information. In addition, the current N classification system was derived from source datasets with OS as endpoint, which was confounded by deaths from local recurrence (T) and thus was unable to distinguish the specific pattern of failure related to nodal metastasis; distant metastasis-free survival (DMFS) would be a more reasonable endpoint to distinguish the actual effect of N on prognostication.
The number of metastatic LNs is a promising novel predictor of survival with demonstrated superiority to the 8th edition AJCC N classification in a variety of squamous cell head and neck cancers [
A total of 354 consecutive nonmetastatic NPC patients treated at our center between September 2010 and March 2011 were included in this study. Each patient underwent a pretreatment workup of complete physical examination, laboratory tests, endoscopy, MR imaging of head and neck, positron emission, and computer tomography (PET/CT) or a combination of chest computed tomography (CT), abdominal sonography/CT, and bone scintigraphy to exclude distant metastases. Those with previous history of cancers or incomplete MR images were excluded.
Medical records and MR images were retrospectively reviewed for this study under approval of the Institutional Review Board. Patients were then restaged using the 8th edition of AJCC staging system.
All MR images were acquired on a standard 1.5-Tesla scanner using a head and neck coil. Based on the T1-weighted fast spin-echo (FSE) images in axial and sagittal planes, T2-weighted FSE images in the axial plane, as well as postcontrast T1-weighted images with fat saturation in axial and coronal planes, radiological features were reassessed by an experienced radiologist and a radiation oncologist independently. At their disagreements, discussion was necessary to make a consensus.
Diagnostic criteria for retropharyngeal lymph node (RPN) metastasis include (1) lateral RPN with minimal axial diameter (MID)≥5mm; (2) grouping RPNs; (3) RPNs of any size with central necrosis; (4) any visible RPNs in the medial group. Criteria for metastatic cervical lymphadenopathy include (1) MID≥10mm for individual LNs; (2) borderline MID of 8-10mm for three or more contiguous LNs; (3) nodes of any size with central necrosis or extracapsular extension (ECE) [
The distribution of lymph nodes was mapped following the updated 2013 consensus guideline of node level delineation for head and neck tumors [
All patients received definitive intensity modulated radiation therapy with simultaneous integrated boost technique (SIB-IMRT), with a prescribed dose of 66-70.4 Gy in 30-32 fractions to primary tumor, 66 Gy to metastatic cervical nodes, 60 Gy to high-risk subclinical and nodal regions, and 54 Gy to elective low-risk nodal regions. All target volumes were delineated according to the definition of International Commission on Radiation Units and Measurements Reports 50, 62, 71, and 83. Residual disease was treated with dose boost using external beam IMRT, or brachytherapy to local residue and electron beam irradiation to palpable nodes. Details of our institutional radiation protocol have been previously reported [
Most patients with locoregionally advanced NPC (stages III–IVB in 7th edition of AJCC system) and part of stage II cases with bulky nodes were given cisplatin-based concurrent chemotherapy with/without neoadjuvant/adjuvant chemotherapy, while early stage patients (T1-2N0) received radiation only. Neoadjuvant/adjuvant chemotherapy regimens included 2-3 cycles of alternative docetaxel/cisplatin/fluorouracil (TPF), docetaxel/cisplatin (TP), cisplatin/fluorouracil (PF), and gemcitabine/cisplatin (GP). Concurrent cisplatin was administered weekly or every 3 weeks.
Follow-up frequency of patients was every 3 months for the first two years and then every 6 months thereafter. MRI of head and neck was performed every 3-6 months. Chest CT, abdominal sonography/CT were done at least annually. Bone scintigraphy or PET/CT was recommended at the discretion of doctors when there are patient-reported new symptoms. Close follow-up tests were suggested at the existence of suspected radiologic findings. Follow-up duration was measured from the date of histological diagnosis, and endpoints of interest included DMFS-time to distant metastasis, OS-time to death of any cause, and disease-free survival (DFS)-time to recurrence or death of any cause.
Suspected distant recurrence was based on (1) appearance of new isolated or multiple lesions in remote regions, including distant lymph nodes, lung, bone, liver, or others, as detected by PET/CT or conventional work-up of surveillance; (2) a progressive change in size/number of lesions within a period of close follow-up; (3) typical radiological characteristics identified under the consensus of at least two radiologists; (4) no evidence of a second primary tumor. Verification of distant metastasis was based on image-guided biopsy or surgical histology when indicated.
Baseline nodal characteristics in patients with different LRN were compared with Chi-square test. Actuarial survival rates were estimated by the Kaplan-Meier method and compared with the log-rank test. Maximally selected rank statistics, as described by Lausen et al. [
Demographic and clinical characteristics of the included 354 patients are listed in Table
Patient characteristics.
Characteristics | No. of patients (%) |
| |
Age, median (range), y | 49 (12-81) |
| |
Gender | |
Male | 266 (75.1) |
Female | 88 (24.9) |
Histological type | |
WHO I | 1 (0.3) |
WHO II | 353 (99.7) |
T classification (AJCC | |
T1 | 119 (33.6)/119 (33.6) |
T2 | 52 (14.7)/64 (18.1) |
T3 | 47 (13.3)/88 (24.9) |
T4 | 136 (38.4)/83 (23.4) |
N classification (AJCC | |
N0 | 34 (9.6)/34 (9.6) |
N1 | 149 (42.1)/143 (40.4) |
N2 | 114 (32.2)/87 (24.6) |
N3 | 57 (16.1)/90 (25.4) |
Clinical stage (AJCC | |
I-II | 100 (28.3)/102 (28.8) |
III-IV | 254 (71.7)/259 (71.2) |
Treatment modality | |
RT alone | 54 (15.2) |
IC+RT | 37 (10.4) |
CCRT | 49 (13.8) |
IC+CCRT | 138 (38.9) |
CCRT+AC | 14 (3.9) |
IC+RT+AC | 62 (17.5) |
Cumulative cisplatin dose | |
≤ 300mg/m2 | 301 (84.8) |
> 300mg/m2 | 53 (14.9) |
Abbreviations: AJCC: American Joint Committee on Cancer; RT: radiation therapy; IC: induction chemotherapy; CCRT: concurrent chemoradiotherapy; AC: adjuvant chemotherapy.
With a median follow-up duration of 63 months, the actuarial 5-year OS, DMFS, and DFS were 84.4%, 85.0%, and 73.7%, respectively. 5-year DMFS was 96.4% (N0), 92.5% (N1), 86.0% (N2), 45.2% (N3a), and 57.8% (N3b), respectively according to the 7th edition of N classification and 96.4% (N0), 92.3% (N1), 81.5% (N2), and 70.3% (N3) to the 8th edition. 5-year DFS was 90.8%, 83.9%, 69.1%, 45.2%, and 45.9% according to the 7th edition and 90.8%, 83.3%, 65.4%, and 59.5% to the 8th edition. 5-year OS was 94.1%, 89.9%, 84.5%, 61.5%, and 67.6% according to the 7th edition and 94.1%, 89.6%, 80.6%, and 77.3% to the 8th edition.
In this study, two cut points for LRN were identified for risk stratification using maximally selected rank statistics. Patients with 0-1 LRN were found with the lowest risk of DMFS, and those with more than 7 LRN had the highest risk. Therefore, LRN was stratified into three categories: LRN 0-1, LRN 2-6, and LRN
The frequency of LN involvement in each level was Ib 5.1%, IIa 58.5%, IIb 76.8%, III 56.7%, IVa 19.6%, IVb 8.6%, Va 30.5%, Vb 11.5%, Vc 2.6%, VIIa 76.6%, VIIb 2.3%, and VIII 4.4%. No metastasis to level Ia, VI, IX, or X was found. The median count of LRN was 4 (range 0-20). As a continuous variable, LRN was statistically correlated with other nodal characteristics, including N classification, LN laterality, level, size, level, ECE, and necrosis (P<0.001). The correlation between categorical LRN and other LN features is shown in Table
Correlation between LRN and other nodal characteristics.
Variable | LRN | | ||
---|---|---|---|---|
0-1 | 2~6 | ≥7 | ||
N classification (AJCC | <0.001 | |||
N0-1 | 78 (100%) | 99 (55.0%) | 0 | |
N2 | 0 | 55 (30.6%) | 32 (33.3%) | |
N3 | 0 | 26 (14.4%) | 64 (66.7%) | |
SCF involvement | <0.001 | |||
SCF (-) | 78 (100%) | 176 (97.8%) | 65 (67.7%) | |
SCF (+) | 0 | 4 (2.2%) | 31 (32.3%) | |
Lower neck involvement | ||||
Lower neck (-) | 78 (100%) | 162 (90.0%) | 35(36.5%) | |
Lower neck (+) | 0 | 18 (10.0%) | 61 (63.5%) | |
Laterality | <0.001 | |||
Nil | 65 (83.3%) | 2 (1.1%) | 0 | |
Unilateral | 13 (16.7) | 112 (62.2%) | 7 (7.3%) | |
Bilateral | 0 | 66 (36.7%) | 89 (92.7%) | |
MLD | <0.001 | |||
≤ 6cm | 78 (100%) | 168 (93.3%) | 77 (80.2%) | |
> 6cm | 0 | 12 (6.7%) | 19 (19.8%) | |
ECE | <0.001 | |||
Negative | 75 (96.2%) | 85 (47.2%) | 17 (17.7%) | |
Positive | 3 (3.8%) | 95 (52.8%) | 79 (82.3%) | |
Nodal necrosis | <0.001 | |||
Negative | 74 (94.9%) | 104 (57.8%) | 33(34.4%) | |
Positive | 4 (5.1%) | 76 (42.2%) | 63 (65.6%) |
Abbreviations: LRN: number of involved lymph node regions; AJCC: American Joint Committee on Cancer; MLD: maximal lymph node diameter; SCF: supraclavicular fossa; ECE: extracapsular extension.
DMFS was chosen as primary endpoint in our analysis based on the following considerations: (1) distant metastasis has become the major pattern of failure for NPC nowadays; (2) the commonly used endpoints, OS and DFS, could easily be complicated by local failure (T factor), making it difficult to distinguish the actual effect of N factor on survival; (3) LN metastasis has been well known with its impact on distant dissemination. According to univariable analysis in Kaplan-Meier method, LRN, whether as a continuous or categorical variable, was strongly predictive for worsening DMFS (P<0.001). When using the three-categorization stratification, the estimated 5-year DMFS for LRN 0-1, 2-6, and ≥7 was 97.0%, 86.7%, and 69.7%, respectively; 5-year DFS was 88.7%, 76.9%, and 55.2%, respectively; 5-year OS was 97.1%, 84.9%, and 74.2%, respectively (Figure
Univariable and multivariable analyses for distant metastasis-free survival.
Variable | Univariable | Multivariable | |||
---|---|---|---|---|---|
No. (%) | Hazard Ratio (95% CI) | | Hazard Ratio (95% CI) | | |
Age | 0.118 | ||||
≤ 49 | 177 (50.0) | 1 (reference) | |||
> 49 | 177 (50.0) | 1.02 (0.99-1.04) | |||
Gender | 0.819 | ||||
Male | 266 (75.1) | 1 (reference) | |||
Female | 88 (24.9) | 1.07 (0.58-1.98) | |||
T classification (AJCC | 0.019∗ | ||||
T1-2 | 183 (51.7) | 1 (reference) | |||
T3-4 | 171 (48.3) | 2.73 (1.34-5.60) | |||
N classification (AJCC | <0.001∗ | ||||
N0-1 | 177 (50.0) | 1 (reference) | |||
N2 | 87 (24.6) | 2.76 (1.46-5.22) | 0.009 | ||
N3 | 90 (25.4) | 4.93 (2.78-8.75) | <0.001 | ||
Laterality | 0.008 | ||||
Nil/Unilateral | 199 (56.2) | 1 (reference) | |||
Bilateral | 155 (43.8) | 2.11(1.22-3.66) | |||
Lower neck involvement | 0.001 | ||||
No | 275 (77.7) | 1 (reference) | |||
Yes | 79 (22.3) | 2.49 (1.45-4.36) | |||
SCF involvement | 0.001 | ||||
No | 319 (90.1) | 1 (reference) | |||
Yes | 35 (9.9) | 3.72 (2.02-6.86) | |||
Necrosis | |||||
No | 211 (59.6) | 1 (reference) | |||
Yes | 143 (40.4) | 3.55 (2.00-6.32) | |||
ECE | 0.008 | 0.080 | |||
No | 177 (50.0) | 1 (reference) | 1 (reference) | ||
Yes | 177 (50.0) | 3.73 (1.96-7.09) | 1.98 (0.92-4.26) | ||
MLD | <0.001 | <0.001 | |||
≤ 6cm | 323 (91.2) | 1 (reference) | 1 (reference) | ||
> 6cm | 31 (8.8) | 6.20 (3.44-11.14) | 4.11(2.23-7.56) | ||
LRN | <0.001∗ | <0.001∗ | |||
0~1 | 78 (22.0) | 1 (reference) | 1 (reference) | ||
2~6 | 180 (50.8) | 5.34 (1.58-17.95) | 0.034 | 4.59 (1.36-15.49) | 0.039 |
≥7 | 96 (27.1) | 13.78 (4.13-45.93) | <0.001 | 9.78 (2.88-33.25) | 0.002 |
Cumulative cisplatin dose | 0.040 | 0.028 | |||
< 300mg/m2 | 301 (84.8) | 1 (reference) | 1 (reference) | ||
≥ 300mg/m2 | 53 (14.9) | 0.34 (0.12-0.95) | 0.32 (0.12-0.88) |
Abbreviations: AJCC: American Joint Committee on Cancer; MLD: maximal lymph node diameter; SCF: supraclavicular fossa; ECE: extracapsular extension; LRN: number of involved lymph node regions.
∗Overall P value for multiple categorical variables.
Kaplan-Meier estimate with the three-categorical LRN on (a) distant metastasis-free survival; (b) disease-free survival; (c) overall survival. LRN: number of metastatic lymph node regions.
By including these significant factors in a stepwise backward selection procedure, multivariable Cox regression model found LRN, MLD> 6cm, and cumulative cisplatin dose as independent predictors for DMFS (P<0.05), while ECE retained a marginal significance (P=0.08) (Table
As pretreatment nodal features, LRN and MLD were chosen into recursive partitioning analysis for clustering of DMFS risk. ECE dropped out of the model relative to other covariables. The conditional inference tree was plotted as in Figure
Proposed N classification derived from recursive partitioning analysis in patients with nasopharyngeal carcinoma. MLD: maximal lymph node diameter; LRN: number of metastatic lymph node regions; DMFS: distant metastasis-free survival.
Compared with the 7th and 8th edition of AJCC N classification, the new N classification showed improved discrimination capability of DMFS (c-index 0.74, 0.69 in 8th, 0.72 in 7th), OS (c-index 0.71, 0.66, 0.68), and DFS (c-index 0.70, 0.67, 0.69), while information loss of the model was reduced (AIC for DMFS 2234, 2242, and 2235; AIC for OS 2151, 2157, and 2154; AIC for DFS 2261, 2267, and 2262). Kaplan-Meier estimates of survival by the three systems were shown in Figure
Kaplan-Meier estimate with the 7th edition (left), 8th edition (middle), and the proposed N classification (right) on (a–c) distant metastasis-free survival, (d–f) overall survival, and (g–i) disease-free survival.
Over the past decades, a cumulating body of data has highlighted the prognostic importance of quantitative LN burden for malignant tumors. In gastric cancer [
However, such effect has never been investigated in NPC, primarily due to the difficulty of LN quantification in a nonsurgical setting with no histological evaluation. Moreover, as ECE is not rare in NPC [
To our knowledge, this is the first study to demonstrate the prognostic value of quantitative LN regions in NPC patients. We identified two cut-off points of LRN to generate a three-category stratification. It turned out that LRN was significantly correlated with other nodal features, indicating that anatomical spread of LNs is usually accompanied with ECE, necrosis, and enlargement of LNs. Survival analysis suggested that incremental LRN strongly correlated with increased DM risk. After adjusting for therapeutic factors and potential confounders, LRN remained a predominant predictor for DMFS. The underlying mechanism might be that higher LRN reflects increased tumor burden, which is highly associated with distant metastasis and overall survival [
Our data also confirmed the prognostic value of LN size in NPC, which remained controversial in previous reports. According to Lee et al. [
A related finding of the present study is that when accounting for LRN, classic variables in AJCC N classification including LN laterality, SCF level, and lower neck level were no longer significant prognostic factors for DMFS. Correlation analysis showed strong collinearity between all these variables and LRN, suggesting that they might be surrogates for quantitative LN burden. This was in concordance with previous evidence in other head and neck cancers, where features like LN contralaterality were eclipsed by LN number in prognostic value [
In NPC, removal of lower neck and SCF from the prognostic model was unexpected but could be properly explained by the latest advances in LN biology. Emphasis on the importance of SCF over long time was based on its proximity to the thoracic duct, which possibly mediates systemic dissemination via lymph-venous conjunction [
We proposed a novel N classification schema using recursive partitioning analysis algorithm. By retaining LRN and MLD for risk stratification, the new system was simplified, showing an improved predictive power of survival over the AJCC (7th and 8th edition) staging systems. The potential advantages of the new schema include the following: (1) it is based on LRN, an independent factor that drives outcomes, rather than surrogates; (2) the three-categorization criteria of LRN better partitioned risk than the classic two-category criteria did, with reduced information loss; (3) 3-dimensional measurement of MLD was more reasonable for outlining the prognostic value of LN size. Our system appeared also superior to the historical reports on 8th staging, with a higher c-index in both DMFS and DFS [
In our patient population, the 8th edition of AJCC N classification did not show superiority to the 7th edition in prognostic power, especially in distinguishing OS of N2 and N3 (P>0.05). This was quite similar to Yang’s reports, which failed in separating N1 with N0, and N3 with N2 by using the 8th edition [
The present study had some limitations. Being conducted in an institutional population, it might require external validation with larger cohorts in future. Besides, the conclusions are confined by the retrospective nature of this study. In addition, use of PET/CT was limited in this study due to the problem of reimbursement in China. More data of PET/CT guided LN evaluation should be incorporated in future, according to the recommendation of National Cancer Comprehensive Network (NCCN) guideline. Moreover, well-known prognostic factors such as tumor volume and Epstein-Barr virus DNA were not included in this study; it is yet unknown if incorporating these factors will alter the conclusions of this study. Further efforts of incorporating these predictors into the current scheme will be worthwhile.
In summary, our study demonstrated that LRN is an independent predictor of DMFS in patients with NPC, predominantly outweighing other classic factors such as LN laterality and level in prognostic value. By combining LRN and MLD, the novel N classification confers significant improvement over the present staging systems in prognostication. Future data for validation of this schema will be warranted.
The original Excel data used to support the findings of this study have not been made available because it is temporarily not allowed according to the institutional regulations.
The authors have no conflicts of interest to declare.
Great appreciations are owed to Dr. XM Ou and Dr. YQ Yang for the assistance in data collection and analysis; to Dr. CY Shen, Dr. TT Xu, and Dr. JH Ding for providing guidance of radiological review; and to Dr. CS Hu for the technical support with study design and progress. Also we sincerely thank all the patients and their families for kindly cooperating in our follow-up.