2.1 Patient Demographics and Survival Outcomes

This comprehensive analysis included 3052 patients from the National Cancer Center cohort (nonendemic region, Beijing) and 3263 patients from the endemic regions, with 1996 patients from Guangzhou, 717 from Sichuan, and 550 from Hunan. Figure 1 provides the comprehensive overview of the process for selecting patients. The ratio of patients from endemic regions to those from nonendemic regions was 1.1:1. Cohort-specific and aggregated patient features are presented in Table S1. With a median age of 47 years, the cohort had a male-to-female ratio of 2.9:1. The majority of patients presented with locally advanced disease (48.2% stage III and 37.4% stage IV). The primary treatment modalities included intensity-modulated radiation therapy (IMRT) alone in 21.1% of cases, concurrent chemoradiotherapy (CCRT) in 44.9%, and induction chemotherapy (IC) followed by CCRT (IC+CCRT) in 25.2%. A detailed summary is presented in Table 1.

The median follow-up period for survivors was 76 months, of which 1548 patients (24.5%) experienced progression and 1006 patients (15.9%) died. The 5-year OS and PFS rates across all cohorts were 83.1% and 74.9%, respectively. The OS and PFS curves are presented in Figure S1. Univariate analysis revealed that sex, age, pathology, TNM staging, Epstein–Barr virus (EBV) DNA copies, lactate dehydrogenase (LDH), and treatment modality were significantly correlated with 5-year OS (Table 1). Figure 2A shows the annual hazard rate trajectory. Following initial treatment, 1014 (65.5%) cases of progression and 440 (43.7%) deaths occurred within 24 months. In concordance, smoothed hazard plots revealed that the peak annual hazard of progression was 10.3% within the initial 24 months and then decreased continuously. Furthermore, after 24 months, the annual hazard of death remained essentially constant at 4.0%.

The 5-year OS was considerably lower in the entire study cohort (83.1%) compared with that in the matched background population (94.7%). Additionally, the SMR from the initial treatment was 3.0 (Figure 2B). Of note, Kaplan–Meier analysis had an estimated 2-year PFS rate of 83.8%, which was most similar to the 5-year OS rate (83.1%) compared with PFS at 12 months (90.3%) and 36 months (79.9%). This suggests that the PFS24 milestone is an important metric for subsequent evaluations.

2.2 PFS24 and Subsequent OS

All 6315 patients completed adequate follow-up for the PFS24 assessment. Of these, 5301 (83.9%) achieved progression-free status after 24 months. In cases who achieved PFS24, the median OS was not reached (5-year sOS, 90.2%; Figure 2C). Minimal disparity was observed in the PFS24-associated actual death rate compared with the expected death rate (SMR 1.0, p = 0.478). Table 1 lists the OS and SMR at 5 years for the subgroups. Conversely, for patients experiencing progression within 24 months, the median sOS was limited to 25.0 months. The 5-year sOS rate post-progression was 21.9%, with an SMR of 23.8 (p < 0.001; Figure 2D).

2.3 Exploratory Subgroup Analysis Between PFS24 and Subsequent OS

Substantial variability of the clinical outcomes in the NPC subgroups prompted us to examine subgroup-specific outcomes contingent upon PFS24 (Table 2). Patients in most subgroups exhibited actual death rates similar to the expected death rates after reaching PFS24.

Of note, despite significant the divergence of OS among the subgroups with distinct baseline EBV DNA abundance (Table 1), these differences disappeared after achieving PFS24, with a 5-year sOS of 90.5% for the <2000 subgroup, 88.4% for the 2000–20,000 subgroup, and 90.4% for >20,000 subgroup (p = 0.600; Table 2). Similarly, comparable sOS values were observed between LDH-stratified groups after achieving PFS24 (90.4% for <240 U/L subgroup, and 88.2% for ≥240 U/L subgroup, p = 0.600).

Subsequent analyses examined outcomes related to clinical staging (Table 3). In early- to mid-stage disease, PFS24 was achieved in 98.9% of stage I, 92.4% of stage II, and 88.6% of stage III patients. For patients achieving PFS24, sOS at 5 years was 93.0%, 92.3%, and 94.3% for stage I, II, and III patients, respectively, which is similar to the general population matched by age and sex (92.2%, 94.3%, and 94.6%, Figure 2E; Table 3). The SMRs for stage I, II, and III patients achieving PFS24 were 0.5, 0.9, and 0.6, respectively. Conversely, for patients failing PFS24, the 5-year sOS was markedly lower at 50.0%, 31.6%, and 29.7% for stages I, II, and III, respectively (Figure 2F), with corresponding SMR values of 12.5, 15.3, and 18.0.

For advanced stage IVA disease, however, PFS24 was only achieved in 74.2% of the patients. Patients achieving PFS24 had a 5-year sOS rate of 83.3% with an SMR of 1.9, which indicates a significantly worse survival compared with the general population (95.1%; Figure 2G; Table 3). For patients failing to achieve PFS24, the 5-year sOS rate was considerably lower at 17.1% (general population, 95.5%; Figure 2H), with a high SMR of 29.1.

2.4 Loss of Residual Lifetime Analysis and Cumulative Mortality

Throughout the posttreatment phase, NPC patients experienced a reduction in residual lifetime by 1.09 months per year within the initial 10 years, compared to the expected survival for the background population. The detailed LoL outcomes by NPC subgroup are presented in Figure 3 and Table S2. For patients achieving PFS24, the residual lifetime was reduced by 0.01 months/year. In most patients achieving PFS24, the reduction in lifetime was less pronounced and statistically insignificant compared with the background population, even among those with an age ≥48 years, KPS score <80, EBV DNA ≥2000 copies/mL, and an LDH ≥240 U/L (Figure 3). For T4, N3, and stage IVA disease, however, LoL remained significantly high at 0.65, 0.38, and 0.51 months/year, respectively. For patients failing to achieve PFS24, the corresponding LoL estimates were 6.48 months/year (Figure 3).

The 5-year cumulative incidence of mortality in all patients was 17.3%, whereas a reduction to 6.0% was observed in patients who successfully achieved PFS24 (Figure S2A). Similarly, the cumulative incidence of progression at 5 years was 25.1% in all patients with NPC, which decreased to 10.8% in patients achieving PFS24 (Figure S2B). Multivariate analysis for individuals achieving PFS24 revealed that age >48 years, advanced T and N stage, high EBV DNA copies, and treatment without concurrent chemotherapy were related factors for late relapse (Table S3). Additionally, for individuals achieving PFS24, age >48 years, advanced T stage, and treatment without concurrent chemotherapy were strongly correlated with late mortality (Table S4).

2.5 Sensitivity Analysis Comparing PFS24 With Alternative Time Points

In the context of sensitivity analysis, we examined outcomes at the additional time points of PFS12 and PFS36. The squared linear correlation after adjustment for measurement error was strong, with an R² coefficient of 0.94 between the treatment effect of PFS24 and 5-year OS (Figure 4). The squared linear correlations of PFS12 and PFS36 with 5-year OS were suboptimal, yielding coefficients R² of 0.92 and 0.88, respectively.

For patients achieving PFS24, we observed no significant difference in the actual death rate at PFS24 compared with the expected death rate (SMR 1.0, p = 0.478); however, the actual death rates for patients achieving PFS12 and PFS36 were significantly different compared with those of the background population (SMR: 1.7 at 12 months and 0.7 at 36 months; Table 3). Furthermore, for patients failing to achieve the selected time point, there was minimal difference in 5-year sOS rates at 24 (22.0%) and 36 (23.3%) months, suggesting that there was minimal advantage in extending the timeframe beyond 24 months. Consistent outcomes were noted in patients with early-to-mid-stage disease (Table 3); however, for stage IVA patients, despite achieving PFS12, PFS24, or PFS36, the three- and 5-year sOS remained significantly lower than the expected survival rate in the background population, and the SMR was significantly higher (Table 3).

4.1 Study Design

In the present study, 6782 patients with non-metastatic NPC were retrospectively reviewed from the period spanning January 2007 to December 2020. The inclusion criteria encompassed patients treated with IMRT and those with complete baseline characteristics. To ensure analysis reliability, patients without progression but with follow-up periods of less than 24 months were excluded. Ultimately, the final cohort consisted of 6315 patients with non-metastatic NPC from four distinct regions of China. Clinical and demographic information were systematically collected. The study protocol was reviewed and approved by the Medical Research Ethics Committee of the National Cancer Center, Chinese Academy of Medical Sciences (Approval No. 23/353-4095). Given the retrospective nature of the study, which involved the use of anonymized data, the Committee waived the requirement for informed consent from individual participants. All procedures adhered to ethical standards as outlined in the Declaration of Helsinki, and the confidentiality and privacy of all patients were strictly maintained throughout the study.

4.2 Treatment and Follow-up

Each eligible patient received curative IMRT, with doses of 2.0–2.27 Gy per fraction, administered 5 days per week over a treatment duration of 6–7 weeks. The treatment targeted primary tumors and metastatic lymph nodes with doses exceeding 66 Gy, while regions prone to infiltration were given a minimum of 50 Gy. The chemotherapy regimen—including induction, concurrent, and adjuvant phases—was selected at the discretion of the treating clinician, with platinum-based agents being the most commonly used.

Systematic evaluations were conducted every 3 months during the initial 2 years, transitioning to semi-annual evaluations for the following 3–5 years, and subsequently transitioning to annual examinations.

4.3 Statistical Analysis

Categorical variables were described using frequencies and percentages, and continuous variables were summarized as medians with ranges. Continuous variables were classified using widely accepted cutoff points. Prior research has established that EBV DNA load cutoffs of 2000 and 20,000 copies/mL possess significant prognostic implications [8, 23]. Kaplan–Meier analysis was used to generate survival estimates. The annual hazard rates delineated the temporal trajectories of progression and morality, whereas the Epanechnikov kernel was used for smoothing. Statistical significance was defined as a two-sided p value of 0.05. Data analyses were conducted utilizing SPSS (version 26.0, IBM) and R software (version 4.2.0; http://www.r-project.org/).

OS represented the time interval from the initiation of treatment to death or last follow-up. PFS was defined as the time interval from the initiation of treatment to distant metastasis, local-regional recurrence, or death from any cause. PFS24 was defined as the status of survival without progression 24 months after initial treatment. The sOS is defined as survival from the 24-month point or progression within 24 months leading to mortality. A general population was derived from Chinese life tables and matched with the study cohort by age, gender, and calendar year using a conditional approach. LoL represented the reduction in life expectancy due to NPC, compared to the general population, such as cancer. The 10-year LoL was determined by calculating the area under the survival curves between NPC patients and the general population from treatment initiation to a 10-year follow-up period. The SMR was defined as the ratio of observed mortality to expected mortality. Expected survival was estimated using a conditional method, with the “survexp” function in R software.

For the sensitivity analysis, we investigated the relationship between the distributions of multipoint PFS rates and 5-year OS rates through linear regression analysis conducted in R software. This analysis was performed on subgroups defined by predefined, widely recognized prognostic factors. We assessed the correlation between PFS rates at additional landmark time points—namely at 12, 24, and 36 months—and the 5-year OS rates. The squared linear correlation coefficient (R²) was calculated to evaluate the association between PFS rates at various time points and 5-year OS, producing corresponding values for each subgroup. An R² value approaching 1 indicated a stronger correlation. Based on prior studies, R² values were categorized as excellent (R² > 0.9), very good (R² > 0.75), good (R² > 0.5), moderate (R² > 0.25), and poor otherwise.