Validation of the 9th AJCC Staging System for NPC in Non-High-Incidence Areas in China

Nasopharyngeal Carcinoma (NPC) is a malignant tumor with distinct geographical distribution characteristics, with over 80% of global cases concentrated in East Asia and Southeast Asia. These two regions-particularly southern China (including Guangdong, Guangxi, Hong Kong, and Taiwan), Malaysia, Vietnam, and the Philippines-are high-incidence areas for NPC and prioritize NPC as a key malignant tumor for prevention and control. In other parts of the world, such as Europe and North America, the incidence of NPC is relatively low. In non-endemic regions of China (e.g., northern and eastern China), significant gaps in knowledge remain regarding NPC's epidemiological characteristics, etiological mechanisms, and clinical management strategies. Epidemiological studies have shown that differences in Epstein-Barr virus (EBV) seropositivity rates, dietary patterns, and environmental exposures between populations in non-endemic and endemic regions may lead to significant variations in tumor biological behavior and treatment responses. More critically, over 70% of patients are diagnosed at the locally advanced NPC (LA-NPC) stage initially. Even with the current standard treatment regimen-induction chemotherapy (IC) combined with concurrent chemoradiotherapy (CCRT)-20%-30% of patients still face recurrence or metastasis, highlighting the urgency of optimizing treatment strategies. Notably, previous large-scale clinical studies have been based exclusively on populations in endemic regions, while data from non-endemic regions have long been overlooked. The unique clinicopathological features and potential differential treatment needs of NPC in non-endemic regions urgently require clarification through targeted research. Accurate AJCC/UICC staging serves as the cornerstone for formulating individualized treatment strategies and assessing prognosis. The newly released 9th edition AJCC/UICC Staging System (hereafter referred to as TNM-9) in 2024 has achieved key breakthroughs, addressing the limitations of TNM-8. First, TNM-9 has optimized the N staging. Extranodal Extension (ENE) refers to the infiltration of tumor cells beyond the lymph node capsule into surrounding tissues. In recent years, the prognostic value of radiological Extranodal Extension (rENE) in NPC has received widespread attention. rENE is typically evaluated via magnetic resonance imaging (MRI) and classified into the following grades: G1-rENE: MRI shows tumor invasion of surrounding adipose tissue without involvement of adjacent structures (e.g., muscles, neurovascular structures, skin). G2-rENE: Multiple adjacent lymph nodes fuse to form a mass, with loss of normal anatomical spaces. G3-rENE: Tumor significantly invades adjacent structures beyond perinodal adipose tissue (e.g., muscles, neurovascular structures, skin, or salivary glands). This grade indicates that the tumor has penetrated the lymph node capsule and directly invaded critical surrounding tissues, correlating with a higher risk of distant metastasis and poorer prognosis. Ai QY and King AD conducted a retrospective analysis of MRI data from 546 NPC patients, evaluating ENE, lymph node size, location, and necrosis. They found that advanced ENE (invading muscles or skin) significantly increased the risk of distant metastasis (HR=4.742, p\<0.001), regional recurrence rate (p=0.014), and mortality risk (HR=2.672, p\<0.001). Its prognostic significance was comparable to that of stage N3, leading the authors to recommend its inclusion in the N3 classification. Canadian scholar Chin et al. also validated the prognostic role of ENE in NPC, emphasizing its general applicability in Western populations. However, controversy remains regarding whether ENE in retropharyngeal lymph nodes (RLNs) should also be classified as N3. Jiang et al. analyzed 4485 patients with non-metastatic NPC and found that patients with stage N1-2 and advanced RLN ENE had a significantly better 5-year overall survival (OS) than those with stage N3 (HR=0.60, 95% CI: 0.38-0.93). Additionally, in multivariate analysis, advanced RLN ENE was not an independent prognostic factor (HR=1.08, p=0.21). In summary, domestic and international studies consistently confirm that \*\*advanced ENE\*\* (definite invasion of adjacent muscles, skin, and/or neurovascular structures) is an independent poor prognostic factor for all endpoints (HR=1.67; 95% CI=1.26-2.19). Based on this evidence, the 9th edition N staging incorporates radiologically confirmed advanced lymph node invasion into the definition of N3 for the first time. The new N3 criteria are explicitly defined as unilateral or bilateral cervical lymph node metastasis meeting any of the following conditions: 1. Maximum diameter \> 6 cm; 2. Lymph nodes extending inferiorly beyond the caudal margin of the cricoid cartilage; 3. Presence of advanced ENE (i.e., invasion of adjacent muscles, skin, or neurovascular bundles). TNM-9 has also made important adjustments to the overall staging system, reorganizing the overall stages and refining subgroups: * The scope of Stage I has been expanded to include more patients. * Significant changes have been made for patients with locally advanced disease: most patients classified as Stage III in TNM-8 are downstaged to Stage II in TNM-9, while most patients with Stage IVA in TNM-8 are reclassified as Stage III. * Finally, TNM-9 refines the M staging by dividing Stage IV into IVA (M1a, ≤3 metastatic lesions) and IVB (M1b, \>3 metastatic lesions). Studies have shown that compared with TNM-8, TNM-9 exhibits significantly improved prognostic discriminatory ability, with obvious advantages in statistical validation-outperforming TNM-8 in terms of risk differentiation consistency, C-index, and Brier score. Meanwhile, Guo et al. proposed incorporating plasma EBV DNA into the TNM staging system based on 979 patients and constructing new subgroups via recursive partitioning analysis (RPA), which demonstrated better prognostic discriminatory ability than TNM-8. This study identified EBV DNA as an independent prognostic factor for progression-free survival (PFS) (HR=1.214, p=0.001), OS (HR=1.288, p\<0.001), and distant metastasis-free survival (DMFS) (HR=1.386, p\<0.001), and recommended 2000 copies/mL as the cutoff value for risk stratification. With the advancement of molecular diagnostic technologies, future staging systems may need to integrate molecular biomarkers. The 9th edition AJCC/UICC Staging System was revised based on large-sample studies in high-incidence East Asian regions, significantly optimizing the definition of LA-NPC. However, the prognostic stratification efficacy of this system highly depends on the epidemiological characteristics of endemic regions (e.g., EBV positivity rate \> 98%). Significant differences exist in dietary habits, lifestyles, and socioeconomic conditions between southern and eastern China, potentially leading to variations in the pathogenesis of NPC across different risk regions in China. For example, previous epidemiological studies have shown that the EBV positivity rate in endemic regions is slightly higher than in non-endemic regions. The uniqueness of NPC in non-endemic regions of China may profoundly affect the applicability of TNM-9 and the selection of treatment strategies. First, differences in treatment tolerance cannot be ignored; in addition, variations in environmental factors, accessibility to medical resources, and socioeconomic status may result in differences in treatment completion rates and toxicity profiles between LA-NPC patients in non-endemic and endemic regions. Therefore, the applicability of TNM-9 in populations from non-endemic regions still needs to be validated. In non-endemic regions of NPC (e.g., Jiangsu, Zhejiang, and Anhui provinces), the survival outcomes and follow-up performance of locally advanced patients under TNM-9 and TNM-8 may differ from those in endemic regions. This study, leveraging multicenter long-term follow-up data, focuses on exploring the applicability of the TNM-9 staging system in non-endemic regions, aiming to fill gaps in research on the generalizability of TNM-9 and further clarify its stability and potential heterogeneity in populations from non-endemic regions. Furthermore, this study addresses the new changes in treatment strategies brought about by the updated staging system. One key change in TNM-9 is the restructuring of the LA-NPC patient population. The revision of N3 has expanded the scope of LA-NPC, making risk stratification of locally advanced disease even more critical. This restructuring, based on precise prognostic stratification, breaks away from the traditional TNM staging-based treatment decision-making model. It emphasizes the need to more clearly distinguish between low-risk and high-risk groups among LA-NPC patients and further explore optimized treatment intensity strategies corresponding to different risk levels. According to the CSCO Clinical Practice Guidelines for Head and Neck Tumors (2025), the treatment of LA-NPC is based on CCRT, and the combined application of IC, adjuvant chemotherapy, and immunotherapy should be explored. Current clinical studies indicate that CCRT is the standard treatment for LA-NPC, with cisplatin being the most commonly used drug. Sequential CCRT after IC is another treatment modality for LA-NPC; previous studies have shown that IC helps improve local control rates but does not significantly improve OS. Sequential adjuvant chemotherapy after CCRT is an alternative treatment modality for LA-NPC, but previous studies have reported suboptimal completion rates due to radiotherapy-related toxicity. The optimal adjuvant chemotherapy regimen, treatment cycles, and beneficiary populations remain to be determined, and the relationship between adjuvant therapy after CCRT and IC in terms of overall treatment efficacy also requires further research. In recent years, various immune checkpoint inhibitors have been incorporated into clinical trials based on CCRT, including the full-course, neoadjuvant, adjuvant, and standalone adjuvant phases. Some studies have shown that these inhibitors may improve 2-3 year event-free survival (EFS) or PFS, but the optimal combination phase, neoadjuvant combination approach, duration of adjuvant therapy, and their impact on OS remain unclear. The following section highlights selected ongoing immunotherapy studies. Immunotherapy has become a transformative approach in cancer treatment, revolutionizing therapeutic strategies for various malignancies. NPC possesses a unique tumor microenvironment, characterized by abundant lymphocyte infiltration and a high PD-L1 expression rate (83%-92%), which provides a theoretical basis for immunotherapy in NPC. Preliminary results from the DIPPER study show that adjuvant PD-1 blockade with camrelizumab significantly improves EFS with manageable toxicity, highlighting its potential in the treatment of LA-NPC. Additionally, the completed phase III CONTINUUM trial provides pivotal evidence for the integration of immunotherapy into the management of high-risk LA-NPC, marking a milestone in this field. This study enrolled 425 patients with high-risk LA-NPC (30% Stage III, 70% Stage IVA), who were randomly assigned to receive either standard IC+CCRT or IC + sintilimab + CCRT + sintilimab maintenance therapy. After a median follow-up of 50.6 months, the sintilimab group demonstrated significantly superior EFS (the primary endpoint) compared to the standard treatment group. A subgroup analysis further revealed that patients with Stage N2/N3 derived particularly significant benefits, which is highly consistent with the high-risk nature of N3 (with advanced ENE) in TNM-9. The CONTINUUM trial confirmed that adding PD-1 inhibitors to standard IC+CCRT confers significant benefits in EFS, DMFS, and local recurrence-free survival (LRFS) for patients with high-risk LA-NPC, providing level III evidence for an intensified treatment model in this high-risk population. However, the increased toxicity associated with this regimen (especially immune-related adverse events) and its applicability in non-high-risk populations require careful evaluation. Based on the above background, this study focuses on NPC populations in non-endemic regions of China, leveraging long-term follow-up data and the 9th edition AJCC/UICC Staging System, with the following objectives: 1. Validate the prognostic evaluation efficacy of the 9th edition AJCC/UICC Staging System in non-endemic regions; analyze differences in patient survival (OS, DFS, DMFS, PFS) between the 9th and 8th editions of the AJCC/UICC Staging System; and assess the consistency of risk stratification (C-index, AUC) of the 9th edition AJCC/UICC Staging System in populations from non-endemic regions. 2. Explore subgroups of locally advanced patients under the 9th edition AJCC/UICC Staging System; evaluate the feasibility and potential clinical value of adjusting treatment intensity strategies in low-risk and high-risk populations. 3. Explore subgroups of patients with N3 and lymph node metastasis (ENE) in non-endemic regions under the 9th edition AJCC/UICC Staging System; assess the accuracy of survival prediction and risk identification for these subgroups; and further analyze survival benefits associated with different treatment strategies. The implementation of this study will provide detailed long-term follow-up data for NPC in non-endemic regions of China, filling research gaps in this field. Additionally, it may provide pivotal evidence for advancing the treatment decision-making framework, promoting clinical practices such as precise intensified treatment for high-risk patients and rational de-escalated treatment for intermediate-risk patients, thereby helping to bridge regional research gaps and improve patients' quality of life. To achieve the above objectives, the specific workflow is as follows: 1. Patient Enrollment and Data Preparation: Patients with NPC were screened according to the aforementioned criteria at three centers: Jiangsu Cancer Hospital, Zhejiang Cancer Hospital, and Anhui Cancer Hospital. Original data (including name, hospital number, gender, age, treatment modality, and follow-up information) were uniformly named, cleaned, and coded. Baseline characteristic tables were constructed based on staging and data from internal/external hospitals. 2. Staging Re-evaluation and Variable Organization: Multiple experts completed re-evaluation of staging according to the 9th and 8th editions of the AJCC/UICC Staging System. All variables-including study variables (e.g., age, T stage, N stage, targeted therapy) and confounding factors (e.g., treatment modality, gross tumor volume \[GTV\], residual lesions)-were organized into an analyzable format. 3. Survival Analysis and Model Validation: Based on a minimum of 36 months of follow-up data, with OS, PFS, LRFS, and DMFS as outcomes, univariate screening was performed using Kaplan-Meier curves and log-rank tests. Subsequently, the prognostic performance of the TNM-9 staging system (C-index, time-dependent ROC, Brier score) was constructed and compared in Cox multivariate models. Finally, a risk score was developed based on independent predictors, and its stability was validated using Bootstrap resampling. 4. Exploration of Treatment Strategies for Risk Subgroups: Based on the aforementioned extensive survival analyses, low-risk and high-risk subgroups of TNM-9-defined locally advanced patients were identified to explore the potential benefits of guiding targeted/immunotherapy for different risk groups. This study is a retrospective cohort study, and all data were obtained from existing electronic medical record systems and radiotherapy/follow-up databases of three tertiary cancer hospitals (Jiangsu Cancer Hospital, Zhejiang Cancer Hospital, and Anhui Cancer Hospital). Researchers extracted relevant data from patients' original medical records (including initial consultation information, chemoradiotherapy regimens, follow-up data, and imaging reports) and entered them into standardized electronic data collection forms (Excel spreadsheets) in a timely, complete, accurate, and clear manner. Case Report Forms (CRFs) were also completed to ensure that all data could be traced back to the original records. Although no electronic data capture (EDC) system was used, all electronic data collection forms were designed with unified variable field formats and restrictions on invalid inputs. A codebook (instruction document for data entry) was developed for researchers at each center to reference. Each study participant was identified by a unique research ID to ensure data anonymization and traceability. Data verification was jointly conducted by the principal investigator and designated data managers at each center. The verification process included the following components: 1. Eligibility Criteria Review: Individual verification of compliance with predefined inclusion/exclusion criteria to prevent incorrect enrollment or exclusion. 2. Completeness Check: Confirmation that no core variables (e.g., staging, outcomes, survival time, treatment modality) were missing. If missing data were identified, original records were rechecked or the data were deemed ineligible for analysis. 3. Logical Consistency Check: Verification of logical relationships (e.g., follow-up time cannot be earlier than the initial treatment date; date of death cannot be earlier than the end date of radiotherapy). 4. Outlier Review: Flagging of values outside the normal range or with statistical anomalies, followed by verification with the respective center to confirm whether the values were entry errors. 5. Data Lock Requirements: After all data were verified as correct, the data manager marked the dataset as the "final version," which was uniformly locked before proceeding to the statistical analysis phase. No modifications to locked data were allowed unless a written application and explanation of the modification reason were submitted, with a traceable record of changes. After completing data entry and verification as required, the CRFs of this study were archived and stored in numerical order, with a searchable catalog for reference. All electronic data files of the study were stored in categories, with multiple backups saved on different disks or storage media to prevent data loss or damage. All original records and study data will be retained for at least 10 years, in strict compliance with data confidentiality regulations and relevant medical information management requirements. This study does not involve the transfer of data to countries or regions outside China, nor does it involve the collection, storage, or export of human genetic resources. It strictly adheres to the Regulations on the Management of Human Genetic Resources of the People's Republic of China\* and other relevant laws and regulations. Study data are used exclusively for the purpose of this research, not for commercial use, and will not be disclosed to third parties, ensuring data privacy and compliance. This study is a non-interventional, retrospective cohort study, and all participants were derived from real-world clinical data. Therefore, no prospective sample size calculation was performed. It is estimated that approximately 1500-1700 patients with locally advanced NPC diagnosed between 2011 and 2023 at the three centers will be enrolled. The sample size is sufficient to provide strong statistical power for analyzing the primary endpoint (OS), secondary endpoints (PFS, LRFS, DMFS), and multivariate analyses. 1. Hypothesis Testing Principles Although no preset sample size was required, the following hypothesis testing principles were followed in statistical inference: * Null Hypothesis (H₀): The 9th edition TNM staging system cannot effectively distinguish between different survival risk groups among patients with locally advanced NPC in non-endemic regions (i.e., no significant survival differences between stages). * Alternative Hypothesis (H₁): The 9th edition TNM staging system can effectively distinguish between different survival risk groups among patients with locally advanced NPC in non-endemic regions (i.e., significantly different survival prognoses corresponding to different stages). All statistical inferences were set with a type I error rate (α) of 0.05 and a type II error rate (β) of 0.20 (i.e., 80% statistical power), using two-tailed tests. To control systematic errors and improve the reliability of statistical conclusions, the acceptable type I error rate (α) was set at 0.05 and the type II error rate (β) at 0.20, corresponding to 80% statistical power. All hypothesis tests used two-tailed tests unless otherwise specified. 2. Statistical Methods Appropriate statistical tests were selected based on the types of predictor variables and outcome variables. The outcome variables in this study are all time-dependent (OS, PFS, LRFS, DMFS), and the independent variables include categorical variables (e.g., T/N stage) and continuous variables (e.g., age). Therefore, Kaplan-Meier curves combined with log-rank tests were used to compare survival differences between groups, and Cox proportional hazards models were further used for multivariate regression analysis-these are standard methods matching the variable types. For intergroup comparisons of baseline variables (non-outcome variables), methods such as chi-square tests, independent samples t-tests, and Wilcoxon rank-sum tests were used, all of which comply with statistical selection criteria based on variable types and distribution characteristics. Data analysis was primarily performed using R software (Version 4.5.0) for survival analysis, visualization, and model performance evaluation; partial data preprocessing and summary statistics were completed using SPSS Statistics (Version 30.0; IBM Corp.). * Quantitative Data (e.g., age, EBV-DNA level, GTV): Described using mean ± standard deviation or median (interquartile range), with statistical indicators selected based on normality and variance homogeneity. * Categorical Data (e.g., gender, T/N stage, treatment modality, use of targeted/immunotherapy): Described using frequencies and percentages. For intergroup comparisons: * For quantitative data: Independent samples t-tests were used if normality and variance homogeneity were satisfied; otherwise, Wilcoxon rank-sum tests (Mann-Whitney U tests) were used. * For categorical data: Chi-square tests were used; if the chi-square test assumptions were not met (e.g., excessively small expected frequencies), Fisher's exact tests were used. * For ordinal data (e.g., staging grades): Wilcoxon rank-sum tests were used. Except for the primary evaluation indicators, all statistical tests used two-tailed tests unless otherwise specified, with a two-tailed P\<0.05 considered statistically significant. 3. Survival Outcome Analysis Survival outcome variables included OS, LRFS, DMFS, and PFS. These variables are all time-dependent outcome variables. Kaplan-Meier methods were used to estimate survival rates and plot survival curves, and log-rank tests were used to compare intergroup differences. Given the presence of multiple confounding factors in observational studies, multivariate analysis methods are ideal for adjustment. Common multivariate analysis methods include generalized linear models (e.g., logistic regression models, Cox models, Poisson models), linear mixed-effects models (LMM), and generalized linear mixed-effects models (GLMM). The selection of models depends primarily on the type of outcome variable. This study included multiple clinical variables that may affect survival outcomes; to account for confounding effects, univariate and multivariate analyses were performed using Cox proportional hazards regression models. Variables with P \< 0.10 in univariate Cox models were first screened and then included in multivariate models to identify factors independently affecting OS (e.g., age, T stage, N stage, receipt of targeted therapy). The analysis results are reported as hazard ratios (HR), 95% confidence intervals (CI), and P values, with visualization via forest plots. Although some predictor variables are multicategorical (e.g., T stage, N stage, treatment modality)-with T/N stages being ordinal and treatment modality being nominal-Cox models can directly handle such variables without the need for additional ordinal or nominal regression models. This study does not involve binary or multicategorical outcome variables, nor does it use methods such as logistic regression, Poisson regression, LMM, or GLMM. This study further employed multi-level statistical methods to achieve subgroup classification of locally advanced NPC. First, based on the multivariate Cox regression model combined with LASSO (Least Absolute Shrinkage and Selection Operator) and Bootstrap resampling techniques, three different risk stratification models (Models A/B/C) were established. The stability of the models was evaluated using variable selection frequency, while discriminatory validity was verified through Kaplan-Meier survival curves, Log-rank tests, and C-index. Results showed that age, HighRisk grouping, and targeted therapy were the most stable and critical prognostic factors. All three models effectively distinguished patients with different risk levels, among which Models B and C exhibited greater potential for refinement.