Bene t of 18F-FDG PET/CT in Treatment Naïve Nasopharyngeal Carcinoma

Shan-Shan Yang Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Yi-Shan Wu Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Wei-Chao Chen Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Jun Zhang Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Su-Ming Xiao Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Bao-Yu Zhang Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Zhi-Qiao Liu Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center En-Ni Chen Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Xu Zhang Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Pu-Yun OuYang Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center Fang-Yun Xie (  xiefy@sysucc.org.cn ) Sun Yat-Sen University Cancer Prevention and Treatment Center: Sun Yat-sen University Cancer Center https://orcid.org/0000-0003-4800-6025


Introduction
Nasopharyngeal carcinoma is a speci c head and neck cancer with unique geographical and ethnic distribution. Approximately 133354 new cases occurred in 2020 with the highest incidence in Southern China [1]. Radiotherapy is the main treatment modality for early-stage nasopharyngeal carcinoma, while concurrent chemoradiotherapy with or without induction chemotherapy is recommended for locoregionally advanced nasopharyngeal carcinoma.
The conventional work-up including head and neck magnetic resonance imaging (MRI), chest X-ray or computed tomography, abdominal sonography or computed tomography, and bone scan are recommended for tumor, node and metastasis (TNM) staging. As for patients with bilateral enlarged lymph nodes or palpable lymph nodes below the cricoid cartilage, 18 F-uoro-2-deoxy-D-glucose (FDG) positron emission tomography and computed tomography (PET/CT) is highly recommended, because of the high risk of occult distant metastasis [2,3]. In terms of N0-1 patients with Epstein-Barr virus (EBV)-DNA lower than 4000 copies/mL, however, prior study [3] insisted the low risk of distant metastasis and the comparable value of conventional work-up to PET/CT at initial staging, and nally did not recommend PET/CT considering the economic effectiveness. Also, a recent study con rmed that there was no survival bene t of adding PET/CT to conventional work-up for stage I-II nasopharyngeal carcinoma [4]. We suppose that precise detection of metastatic cervical lymph nodes and correct N-stage perhaps more affect the prognosis of these patients, instead of focusing on the value of detecting occult distant metastasis. Staged T3 nasopharyngeal carcinoma without distant metastasis is the most typical representation. For example, patients with T3N0M0 can achieve comparable overall survival by intensity modulated radiotherapy alone to patients with stage II disease [5], whereas the risk of distant metastasis is as high as 18% at 3 years after radical chemoradiotherapy for T3N2-3M0 nasopharyngeal carcinoma, and induction chemotherapy followed by concurrent chemoradiotherapy is strongly recommended for these patients [6][7][8]. Therefore, we attempted to nd if PET/CT can in uence the prognosis of nasopharyngeal carcinoma by accurate diagnosis of metastatic lymph nodes.
Currently, the optimal treatment mode for the subgroup of T3N1M0 remains the most controversial [9,10]. T3N1M0 patients belongs to locoregionally advanced disease, but clinical trials that justi ed the bene t of induction chemotherapy [8] did not include this sort of patients. A retrospective study found no survival bene t of the additional induction chemotherapy for these patients [9], whereas male patients staged with T3N1M0 and EBV DNA higher than 2000 copies/mL were the target population as suggested by another study [10]. Albeit EBV DNA showed prognostic value, it made little sense for the present in clinical practice due to the lack of uni ed test standard, robust serum level and accepted cutoff value. Considering the generality of PET/CT and MRI in different hospitals, we thus tried to nd the way to serve for individualized treatment by developing a radiologic score in a cohort of T3N1M0 nasopharyngeal carcinoma.

Methods
Patients and study design A total of 2759 patients who were pathologically diagnosed with nasopharyngeal carcinoma were obtained from May 2009 to May 2020 at Sun Yat-Sen University Cancer Center. 336 patients who underwent node ne-needle aspiration biopsy guided by ultrasonography or lymph node dissection, and examination of PET/CT and MRI before treatment constituted the Cohort A, to testify the accuracy of PET/CT in diagnosing metastatic lymph nodes. Cohort B   consisted of 1093 T3N1M0 nasopharyngeal carcinoma patients who received both PET/CT and head and neck MRI, while Cohort C contained 1377 T3N1M0 patients who underwent MRI alone. Cohort B and Cohort C were compared to nd the survival bene t of adding PET/CT to MRI. Specially, 838 patients in the cohort of B who received concurrent chemoradiotherapy with or without induction chemotherapy were identi ed as Cohort D to analyze the bene t of induction chemotherapy. The owchart was presented in Supplementary Fig. 1

Imaging analysis
The protocols of PET/CT and MRI were deposited in Supplementary Method. MRI images were read by two experienced radiologists, and PET/CT by two experienced nuclear physicians who were blinded to MRI results. Any differences were resolved by consensus. The metastatic lymph nodes were diagnosed according to the radiologic criteria [11]: (1) retropharyngeal lymph node with a minimal axial diameter of 5mm or greater, and cervical lymph node with a minimal axial diameter of 10mm or greater, respectively; (2) minimal axial diameter of 8mm for clusters of 3 or more lymph nodes; (3) lymph nodes with necrosis or extranodal extension. Same as previous study [12], radiologic extranodal extension was categorized into 4 grades: Grade 0, no extranodal extension; Grade 1, invasion to surrounding fat; Grade 2, coalescent nodes; Grade 3, in ltrating adjacent structures. As reported [13], the diagnostic criteria for nodal necrosis based on MRI included: (1) focal area of low signal intensity on T1-weighted images with or without enhanced edges; (2) focal area of high signal intensity on T2-weighted images. In PET/CT, lymph node was considered as positive when 18 F-FDG uptake increased signi cantly compared with background [14]. In the nal decision-making, PET/CT results were supplemented with MRI ndings (Fig. 1).

Treatment and follow-up
All patients received intensity-modulated radiotherapy. The prescribed doses were 66-72 Gy to gross tumor and lymph nodes. The treatment modality included concurrent chemoradiotherapy with or without induction chemotherapy, radiotherapy alone, and induction chemotherapy plus radiotherapy. After treatment, follow-up examinations were conducted at least every 3 months during the rst 2 years, and then every 6 to 12 months thereafter. Examinations including EBV DNA testing, nasopharyngoscopy, head and neck MRI, chest X-ray or computed tomography, and abdominal sonography or computed tomography were performed regularly. PET/CT was recommended if necessary, and recurrence or metastasis was con rmed by biopsy if possible.

Statistical analysis
Failure-free survival (FFS, time from diagnosis to failure or death) was de ned as the primary endpoint, and the second endpoints were distant metastasis-free survival (DMFS, time from diagnosis to distant metastasis or death), locoregional relapse-free survival (LRRFS, time from diagnosis to locoregional recurrence or death), and overall survival (OS, time from diagnosis to death).
Categorical variables were compared by Chi-square test. McNemar's paired-sample test or Chi-square test was used to compare PET/CT and MRI in terms of sensitivity, speci city, positive predictive value, and negative predictive value. And propensity score matching (PSM) method was applied to balance confounders between different groups at the ratio of 1:1. The cut-off values of continuous variables were determined by receiver operating characteristic curve (ROC) analysis. To con rm the bene t of PET/CT in N0-1 patients with EBV-DNA lower than 4000 copies/mL, the cut-off value of EBV-DNA was 4000 copies/mL in cohort B and cohort C [3], while the cut-off value was 2000 copies/mL in cohort D, in order to compared with the model reported in previous study [10].The survival rates were compared by Kaplan-Meier method. Univariate and multivariate Cox regression analysis were performed to determine independent factors. The model was evaluated by the concordance index (C-index). Statistical analysis was conducted by R 4.0.1 (http://www.r-project.org/) and SPSS 26.0. And a two-side P < 0.05 was considered to be statistically signi cant.
Among them, 96.7% (260/269) of positive and 75.9% (145/191) of negative lymph nodes were correctly detected by PET/CT, while only 88.5% (238/269) of positive and 70.7% (135/191) of negative lymph nodes were correctly diagnosed by MRI. PET/CT was signi cantly more sensitive than MRI for detecting cervical lymph node metastasis (p < 0.001). As for speci city, no signi cant difference was observed between two imaging methods (75.9% vs. 70.7%, p = 0.174). The negative predictive value, positive predictive value, and accuracy of PET/CT and MRI were 94.2% vs. 81.3%, 85.0% vs. 81.0%, and 88.0% vs.81.1%, respectively (Table 1). And the area under the curve (AUC) of PET/CT was higher than that of MRI (0.863 vs. 0.796, p < 0.05). Notably, 14.4% (66/460) of lymph nodes had discrepancies between two imaging tests. Among them, PET/CT showed true positive in 22 lymph nodes which was diagnosed as negative lymph nodes by mistake in MRI, and true negative in 27 lymph nodes misdiagnosed as positive lymph nodes by MRI. Nonetheless, 17 lymph nodes were wrongly diagnosed as positive lymph nodes by PET/CT according to histopathology.   Supplementary Fig. 3).
Prolonged survival rates of patients staged by PET/CT vs. MRI To nd if the advantage of PET/CT in diagnosis can contribute to the survival differences, the cohort C in which patients underwent MRI alone was compared with cohort B. As shown in Table 2, the baseline characteristics between PET/CT plus MRI and MRI alone were compared. However, the results showed there was an imbalance in age, lymph node location, and albumin between two groups (p = 0.021, p = 0.014, and p < 0.001, respectively). After PSM at the ratio of

Guiding individualized induction chemotherapy
In the Cohort B, 838 patients who received concurrent chemoradiotherapy with or without induction chemotherapy were selected to the Cohort D. However, there were signi cant differences in baseline characteristics between two treatment models (see Table 3).  Table 6) and multivariable analysis indicated that SUV-N higher than 9.35, in together with nodal necrosis and extranodal extension in ltrating adjacent structures, carried prognostic signi cance for FFS (p < 0.001, p = 0.002, and p = 0.002, respectively, Supplementary Table 7). Radiologic score was thus developed based on the number of the three factors. Patients with higher radiologic score had the lower FFS (p < 0.001, Supplementary Fig. 5). Thus, patients with one or more risk factors were classi ed into high-risk group (radiologic score > 0, n = 454), while patients with no risk factor were strati ed into low-risk group (radiologic score = 0, n = 244).
The survival curves showed patients in high-risk group had lower FFS, DMFS, LRRFS, and OS than those in lower-risk group (all p < 0.05, Supplementary  Fig. 6). Radiologic score model had higher C-index than the model with gender and EBV-DNA (0.   Fig. 3). But in the high-risk group strati ed by radiologic score, patients receiving induction chemotherapy plus concurrent chemoradiotherapy had higher 5-year FFS than those receiving concurrent chemoradiotherapy alone (82.2% vs. 71.5%; p = 0.00642, Fig. 3). After adjusting for covariates, multivariate analysis also con rmed that the addition of induction chemotherapy was an independent factor for FFS (HR: 0.53, 95%CI: 0.35-0.8, p = 0.0026; Supplementary Table 8 and Supplementary Table 9). By contrast, no survival difference was observed between the two treatment modes in the low-risk group (p = 0.074, Fig. 3). The same conclusion was also reached for DMFS and RRFS. And the detailed results of DMFS, LRRFS and OS were deposited in supplementary Tables and Figures.

Discussion
In this large cohort study, PET/CT was proved to be more accurate than MRI in diagnosing cervical lymph nodes con rmed by histopathology. Accordingly, as accurate N-staging more precisely uncovered the true prognosis of patients, PET/CT plus MRI identi ed T3N1M0 patients showed higher survival outcomes than MRI staged T3N1M0 patients, even if EBV-DNA was lower than 4000 copies/ml. And PET/CT based SUVmax of lymph nodes in together with nodal necrosis and extranodal extension involved with adjacent structures could build a radiologic score model and identify high-risk T3N1M0 patients who can bene t from the addition of induction chemotherapy.
In fact, this was not the rst report of advantage of PET/CT over MRI in diagnostic lymph nodes of nasopharyngeal carcinoma. But different from previous study [4,15], the included 460 lymph nodes were pathologically con rmed in our study, instead of clinical follow up. The nding of PET/CT superior to MRI was also consistent with the results of studies in head and neck cancer [16]. Although the sensitivity of PET/CT we found in nasopharyngeal carcinoma (96.7%) was a bit higher than that of head and neck cancer (90.0%), the speci city of PET/CT was only 75.9% herein, lower than that of head neck cancer (94.0%) [17] but still better than MRI (70.8%). Perhaps, newly deep learning algorithm might be a good assistant to further improve the diagnostic performance. Notably, 18.6% (118/633) of PET/CT diagnosed T3N1M0 were up-staged with T3N2-3M0 by MRI, while only 3.2% (20/633) of patients were down-staged with T3N0M0 by MRI, without any discrepancy in T3 staging. This also indicated the high potential of over-diagnosis by MRI. The undifferentiated survival rates ( Supplementary Fig. 2) across the misdiagnosed T3N0-3M0 patients by MRI but staged T3N1M0 by PET/CT testi ed again the higher possibility of getting closer to the true prognosis if staged by PET/CT, in terms of treatment outcomes. Certainly, MRI staged T3N1M0 patients were divided to T2N1M0 and T3N1M0 by PET/CT but no survival differences were observed between T2 and T3; as no gold standard to con rm T-stage of both examination equipment, we failed to draw a rm conclusion of PET/CT versus MRI in T-stage of nasopharyngeal carcinoma. Reviewing previous studies [15,18], MRI seemed to be more accurate than PET/CT in diagnosing the involvement of local structures. Therefore, the combination of PET/CT and MRI may be the best recommendation for diagnosing and staging of treatment naïve nasopharyngeal carcinoma.
To testify if the diagnostic advantage of PET/CT can transfer to bene t in treatment outcomes, we directly compared two cohorts of T3N1M0 patients staged by PET/CT plus MRI or MRI alone. The signi cant higher survival rates of patients with PET/CT plus MRI, regardless of the EBV DNA load, nally supported our suppose. Obviously, PET/CT as an examination test cannot directly alter the nal treatment outcomes by itself, but several prospective studies reported that adding PET/CT to conventional work-up could provide additional information and may change management approaches in 15.7%-33.8% of head and neck cancer patients [16,19,20]. Given the retrospective design of our study, the magnitude of actual changes in treatment modes before and after the application of PET/CT in these patients was not available. But on the other hand, we tried to investigate if PET/CT could predict the survival rate of patients and accordingly identify the high-risk patients to receive more intensive treatment. In a cohort of PET/CT and MRI staged T3N1M0 patients, induction chemotherapy showed no survival bene t for all the patients, as observed in prior studies [9]. Previous reported EBV DNA load and gender guided risk strati cation [10] did not work here, possibly because of its poor speci city and generalization. By contrast, SUVmax-N, nodal necrosis and extranodal extension involved with adjacent structures remained highly prognostic. In fact, these were not observed for the rst time [12,21,22], which indicated the potential of good generalization. The radiologic score model based on the three characteristics showed a signi cantly (p < 0.001) higher C-index (0.72) than the model based on EBV DNA and gender [10] (C-index = 0.56) in risk strati cation. Besides, the radiologic score model selected high-risk patients were just the target objectives who can bene t from the addition of induction chemotherapy. As shown in our study, the 5-year FFS rate of the high-risk T3N1M0 patients was similar to those more advanced patients included for clinical trials of induction chemotherapy [23]. Thus, it was not unreasonable that the high-risk T3N1M0 patients had signi cantly improved survival outcomes when receiving induction chemotherapy followed by concurrent chemoradiotherapy.
Based on 460 biopsied cervical lymph nodes, we rmly concluded the advantage of PET/CT in diagnosing lymph nodes. In a cohort of T3N1M0 patients, the interference of covariate factors including nodal size, nodal level, nodal laterality and T stage were totally excluded; we testi ed the survival bene t transferred from the accurate diagnosis by PET/CT and found the way to guide individualized induction chemotherapy for T3N1M0 patients by a radiologic score model based on PET/CT and MRI.

Declarations
Con ict of interest