This retrospective investigation encompassed 569 individuals with newly diagnosed NPC who underwent complete treatment at our cancer center within the timeframe of January 2012 to December 2018. Rigorous participant selection criteria were adhered to, including: a) Biopsy-proven primary NPC; b) Availability of pertinent clinical and laboratory baseline data; c) Pre-treatment whole-body 18F-FDG PET/CT scans and head and neck MRI; d) Stage III‒IVA with lymph node metastasis; e) Radical treatment; To ensure study integrity, certain exclusion criteria were employed, involving: a) The presence of other malignancies; b) Other concomitant fatal disease; c) A history of malignancies treatment. The staging of all participants was meticulously carried out according to the 8th edition of the AJCC/TNM staging system. The study was approved by the Ethical Committee of Fujian Cancer Hospital (YKT2020-011-01).

Head and neck MRI examinations were performed using either a 1.5T MR system (Signa Excite 1.5T HD Twin Speed, GE Healthcare, Milwaukee, Wisconsin) or 3.0T MR (Achieva 3.0T, Philips Healthcare, Best, the Netherlands).16

For PET/CT imaging, the Gemini TF 64 PET/CT scanner from Philips (the Netherlands) was employed. The 18F-FDG radiochemical purity exceeded 95%, as ensured by the HM-10 cyclotron. The 18F-FDG SUV was determined by calculating the decay corrected tissue activity (nCi/mL), divided by the injected dose of FDG (nCi) and the patient’s body weight (g). We defined SUVmax-T and SUVmax-N as the maximum SUV of the primary tumor and the regional lymph nodes, respectively.

The rENE was defined based on the following criteria: a) Lymph node with only indistinct nodal margins or irregular enhanced nodal capsules; b) Infiltration into adjacent adipose tissue and surrounding structures; c) Matted node.15 Matted nodes were characterized as clusters of two or more nodes adjacent to each other without intervening fat planes.

The criteria for diagnosing Cervical Node Necrosis (CNN) at MRI were existing a focal area of high signal intensity on T2-weighted images or a focal area of low signal intensity on contrast-enhanced T1-weighted images, with or without a surrounding rim of enhancement. The tumor volume was calculated by the treatment planning software (Pinnacle, version 9.2, Philips Radiation Oncology System, Wisconsin), which could automatically reconstruct a 3D image by manually delineating each MR image.

Haematological biomarkers were collected from each patient before treatment including neutrophil count, platelets count, lymphocyte count, hemoglobin, Lactic Dehydrogenase (LDH), albumin and EB-DNA copies were collected. The Neutrophil-Lymphocyte Ratio (NLR) was determined by dividing the absolute neutrophil count by the absolute lymphocyte count. Subsequently, the Systemic Immune-Inflammation Index (SII) was derived by multiplying the absolute platelet count with the NLR. Finally, the Prognostic Nutrition Index (PNI) was computed by multiplying the absolute lymphocyte count by the sum of the serum albumin level +5.

Platinum-based chemotherapy was administered to a predominant majority of the patient cohort, encompassing 561 individuals (98.6% out of 569). Among the entire group, 551 cases (96.8%) underwent neoadjuvant chemotherapy, 419 cases (73.6%) received concurrent chemotherapy, and 132 cases (23.2%) received adjuvant chemotherapy. The most frequently employed regimen involved the combination of platinum with paclitaxel or gemcitabine. Intensity-Modulated Radiotherapy (IMRT) was applied for radiotherapy. The target volume and radiotherapy dose were implemented according to a previously described protocol.17 The radiotherapy prescription dose was GTV 70∼72.6Gy/31∼33F, CTV1 62∼62.7Gy/31∼33F, and CTV2 54.4∼56.2Gy/31∼33F.

Continuous variables were transformed into categorical variables. Cut-off values were calculated using Receiver Operating Characteristic (ROC) curve and were as follows: age (53years), SUVmax-T (7.86), SUVmax-N (10.28), PTV (42.75mL), NTV (34.88mL), ALB (32g/L), LDH (210 U/L), PLR (85.43), NLR (2.60), PNI (59.6), SII (633.9), EB-DNA (18,700).

Patients were followed by routine examination every 3-months for 2-years, every 6-months for years 3–5, then annually. The primary endpoint of this study was the assessment of Overall Survival (OS). This outcome measure was defined as the duration between the initial diagnosis date and either the date of death due to any cause or the date of the most recent follow-up.

Statistical analyses were performed using IBM SPSS Statistical software version 19.0 (IBM Corp., Chicago, IL, USA) and R version 3.4.0 (http://www. R-project.org/).

Patients were randomly divided into a training cohort and validation cohort at a ratio of 7:3 for construction and validation of the nomogram. Variables demonstrating a significant level of statistical significance at p< 0.1 in the univariate analyses were subsequently subjected to multivariable Cox regression analysis. Independent prognostic factors were identified if their impact remained statistically significant within the Cox model (p< 0.05).

Using these identified prognostic factors, we constructed a nomogram to effectively predict the 2-, and 3-year OS rates. The discriminative power of the nomogram was assessed through the Concordance index (C-index) and the area under the ROC curve analysis. A calibration curve was employed to juxtapose the observed probability against the predicted probability. Furthermore, the clinical validity of the nomogram was demonstrated using Decision Curve Analysis (DCA), which quantified net benefits across varying threshold probabilities. Subsequently, ROC curve was applied to stratify patients into two distinct risk groups. Kaplan-Meier survival analysis was then employed to generate survival curves for these different risk groups, with comparison facilitated by the log-rank test.

A total of 569 patients were enrolled in this study, with 398 individuals (69.9%) allocated to the training cohort and the remaining 171 (30.1%) to the validation cohort. Baseline characteristics were summarized in Table 1. Importantly, no statistically significant differences were observed in the baseline characteristics between the training and validation cohorts (p=0.114–0.970), except NLR (p=0.042). A total of 360 (63.2%) patients were presence of radiologic extranodal extension at initial diagnosis, including 261 (61.8%) in training cohort and 114 (66.7%) in validation cohort.

The median follow-up duration for the training cohort was 48 months (range=7 to 118 months), while the validation cohort had a median follow-up of 49 months (range=10 to 118 months). Over the course of follow-up, a total of 55 patients in the training cohort and 14 patients in the validation cohort had reached the endpoint of mortality. In addition, the 1-, 3-, and 5-year OS were 98.7%, 90.8%, 82.9%, and 98.8%, 95.3%, 91.8% for the training and validation cohorts, respectively. As expected, patients with rENE had significantly poor prognosis than other patients in both training and validation cohorts (p= 0.0012, p= 0.018, respectively) (Fig. 1).

Figure 1.

Kaplan-Meier survival curves for rENE of the training cohort (A) and validation cohort (B).

(0.2MB).

Univariate analysis indicated that rENE (p < 0.002), age (p< 0.012), Tumor stage (p= 0.014), Node stage (p< 0.008), SUVmax-T (p= 0.043), SUVmax-N (p= 0.011), PTV (p< 0.001), LDH (p< 0.001) and EBV DNA (p< 0.021) were associated with OS of NPC patients. In multivariate analysis for OS with Cox regression, the results showed that rENE (p< 0.002), age (p< 0.012), PTV (p< 0.001), Node stage (p< 0.008) and LDH (p< 0.001) remained independently prognostic. The detailed results of univariate and multivariate analyses are presented in Table 2.

Based on the above significant independent factors selected from the training cohort, a nomogram was constructed to predict OS for LA-NPC with LNM at 2-, and 3-years (Fig. 2), where larger points indicated a shorter OS. The C-index of the nomogram for OS was 0.75 (0.710–0.78). The AUCs of the nomogram (0.77 for 2-year; 0.75 for 3-year) were significantly better than the 8th edition of the TNM staging system (0.60 for 2-year; 0.65 for 3-year) in predicting 2- and 3-year OS (Fig. 3A and B). The calibration plots revealed an excellent prediction of 3-year OS and an acceptable prediction of 2-year OS (Fig. 4A and B). Ultimately, the DCA conducted within the training cohort demonstrated a consistently favorable net benefit associated with the nomogram among most threshold probabilities. This suggested a promising clinical utility for the nomogram model. (Fig. 5A and B).

Figure 2.

Nomogram model predicting 2 and 3year OS in LA-NPC patients with LNM.

(0.13MB).

Figure 3.

Receiver operating characteristic (ROC) curves for nomogram (A) and TNM staging system (B) for 2-year and 3-year overall survival rate in the training cohort.

(0.17MB).

Figure 4.

Calibration curves of the nomogram predicted and actual measured survival probabilities at 2- and 3-years of training cohort (A and B) and validation cohort (C and D).

(0.34MB).

Figure 5.

Decision curve analysis for 2-year and 3-year survival predictions of training cohort (A and B) and validation cohort (C and D).

(0.48MB).

The nomogram was validated in the validation cohort. The C-index of the nomogram in validation cohort for predicting OS was 0.76 (0.69–0.83). The calibration curve underscored the well-calibrated nature of the nomogram's predictions. Notably, the robust concordance observed between actual outcomes and nomogram predictions was particularly pronounced for the 3-year OS, as illustrated in Fig. 4 C and D. The DCA outcomes within validation cohort closely mirrored those observed in the training cohort (Fig. 5C and D).

The nomogram scores were calculated according to the total scores of each covariate in the model. The cutoff value of the nomogram scores for the stratification of OS was 162.4 by ROC curve. Then the cohorts were divided into high-risk and low-risk groups. The log-rank test disclosed a significant divergence in survival outcomes, a finding that held true within both the training cohort (p< 0.001) (Fig. 6A) and the validation cohort (Fig. 6B). Therefore, the risk stratifications could effectively discriminate OS for the proposed risk groups in LA-NPC with LNM.

Figure 6.

Kaplan-Meier survival curves for different risk groups of the training cohort (A) and validation cohort (B).

(0.2MB).