Development and validation of radiologic scores for guiding individualized induction chemotherapy in T3N1M0 nasopharyngeal carcinoma

Objectives We aimed to develop and validate radiologic scores from [18F]FDG PET/CT and MRI to guide individualized induction chemotherapy (IC) for patients with T3N1M0 nasopharyngeal carcinoma (NPC). Methods A total of 542 T3N1M0 patients who underwent pretreatment [18F]FDG PET/CT and MRI were enrolled in the training cohort. A total of 174 patients underwent biopsy of one or more cervical lymph nodes. Failure-free survival (FFS) was the primary endpoint. The radiologic score, which was calculated according to the number of risk factors from the multivariate model, was used for risk stratification. The survival difference of patients undergoing concurrent chemoradiotherapy (CCRT) with or without IC was then compared in risk-stratified subgroups. Another cohort from our prospective clinical trial (N = 353, NCT03003182) was applied for validation. Results The sensitivity of [18F]FDG PET/CT was better than that of MRI (97.7% vs. 87.1%, p < 0.001) for diagnosing histologically proven metastatic cervical lymph nodes. Radiologic lymph node characteristics were independent risk factors for FFS (all p < 0.05). High-risk patients (n = 329) stratified by radiologic score benefited from IC (5-year FFS: IC + CCRT 83.5% vs. CCRT 70.5%; p = 0.0044), while low-risk patients (n = 213) did not. These results were verified again in the validation cohort. Conclusions T3N1M0 patients were accurately staged by both [18F]FDG PET/CT and MRI. The radiologic score can correctly identify high-risk patients who can gain additional survival benefit from IC and it can be used to guide individualized treatment of T3N1M0 NPC. Key Points • [ 18 F]FDG PET/CT was more accurate than MRI in diagnosing histologically proven cervical lymph nodes. • Radiologic lymph node characteristics were reliable independent risk factors for FFS in T3N1M0 nasopharyngeal carcinoma patients. • High-risk patients identified by the radiologic score based on [ 18 F]FDG PET/CT and MRI could benefit from the addition of induction chemotherapy. Supplementary Information The online version contains supplementary material available at 10.1007/s00330-021-08460-1.


Introduction
In 2020, 133,354 new cases of nasopharyngeal carcinoma were reported, accounting for 0.7% of all cancers in the world, but over 70% of patients were from Asia, with an age standardized rate (world) of 3.0 per 100,000 in China [1,2]. Unfortunately, over 75% of patients are diagnosed with locoregionally advanced disease at presentation [3]. Despite advances in techniques, nearly 30% of patients experience treatment failure, especially distant metastasis [4]. Phase III randomized controlled trials have proven that induction chemotherapy added to concurrent chemoradiotherapy can significantly decrease the risk of distant metastasis and improve the survival of patients with locoregionally advanced nasopharyngeal carcinoma [5][6][7]. This treatment mode is thus the category 2A recommendation for these patients by National Comprehensive Cancer Network (NCCN) guidelines [8]. However, notably, these randomized trials did not enroll any patients staged with T3-4N0M0 or T3N1M0 at all. A retrospective study reported that patients with T3N0-1 do not benefit from induction chemotherapy [9], while male T3N1 patients with Epstein-Barr virus (EBV) DNA higher than 2000 copies/mL were the only target population for induction followed by concurrent chemoradiotherapy, as suggested by another study [10]. Therefore, the treatment modality of T3N1 nasopharyngeal carcinoma is still controversial. Although EBV DNA has been reported to have prognostic value, its extensive application is difficult in real-world practice due to the lack of recognized cutoff values and unified test standards. 2-Deoxy-2-[ 18 F] fluoro-d-glucose ([ 18 F]FDG) positron emission tomography/ computed tomography (PET/CT) and magnetic resonance imaging (MRI) have been widely applied for the diagnosis and staging of nasopharyngeal carcinoma [11]. Given the widespread use of [ 18 F]FDG PET/CT and MRI, the radiologic characteristics of the primary tumor and metastatic lymph nodes may prove useful for selecting individualized treatment.
The maximal standardized uptake value (SUVmax) of [ 18 F]FDG PET/CT, related to metabolic activity, has prognostic implications and is used for risk stratification [12,13]. The SUVmax of lymph nodes (SUVmax-N) and the lymph node-to-primary tumor SUVmax ratio are potential  [12,14]. However, the prognostic value of the SUVmax of the primary tumor (SUVmax-T) is in dispute [15]. Previous studies showed that ungraded radiologic extranodal extension determined by MRI had no prognostic significance in nasopharyngeal carcinoma [16,17]. After grading the radiologic extranodal extension, the sensitivity of diagnosing pathologic extranodal extension improved in head and neck cancer [18], and the consistency of determining radiologic extranodal extension also increased as extranodal extension grades increased. Recent studies demonstrated that high-grade radiologic extranodal extension with adjacent structure invasion significantly predicted a poor survival outcome [19][20][21][22]. However, the above studies did not eliminate the interference of confounding factors such as T stage, cervical lymph node level, laterality, and necrosis status. Thus, the reported prognostic value of radiologic characteristics in prior studies needs re-evaluation. Additionally, accurate diagnosis of T3N1M0 patients is another key point. [ 18 F]FDG PET/CT has advantages in detecting metastatic cervical lymph nodes and distant metastasis over MRI, but it is inferior in determining local tumor invasion and retropharyngeal nodal metastasis [23,24]. This conclusion was based on the judgment of metastatic lymph nodes by clinical follow-up instead of pathologic confirmation. Herein, we included patients who underwent both MRI and [ 18 F]FDG PET/CT examination before treatment. More importantly, one or more cervical lymph nodes of certain patients were histologically confirmed, so the performance of [ 18 F]FDG PET/CT and MRI in diagnosing the specific lymph nodes provided firm evidence for precisely identifying the subgroup of T3N1M0. Subsequently, we accurately developed and validated the radiologic score of the lymph node characteristics to identify high-risk patients who can gain an additional survival benefit from induction chemotherapy and finally suggested an individualized treatment mode for these patients. (3) receipt of concurrent chemoradiotherapy or induction followed by concurrent chemoradiotherapy; and (4) receipt of intensitymodulated radiotherapy. The patient flow chart is presented in Fig. 1. This study was approved by the Institutional Ethical Review Board (No. B2021-059-01), and informed consent was waived for the part of retrospective analysis. Patients in the validation cohort were derived from a prospective observational study, and informed consent regarding a second analysis of their data was obtained from all of the patients.

Image analysis
All patients received whole-body [ 18 F]FDG PET/CT and MRI examinations of the head and neck. The detailed MRI and [ 18 F]FDG PET/CT protocols are shown in the Supplementary Methods. All [ 18 F]FDG PET/CT images were evaluated by a researcher (SSY, 5 years of experience in treating nasopharyngeal carcinoma) with reference to the issued report and then checked by an expert nuclear medicine physician (X.Z., more than 20 years of experience). All MR images were assessed by a radiation oncologist (P.Y.O.Y.) with 10 years of experience and reviewed again by an expert radiation oncologist (F.Y.X.) with over 30 years of experience in treating nasopharyngeal carcinoma. Inconsistencies were discussed with a radiologist (Y.H.) who had interpreted head and neck MR images of over 500 patients per month for over 5 years.
The diagnostic criteria for the metastatic lymph nodes, radiologic extranodal extension, and nodal necrosis were the same as those in previous studies [19,26,27] and are detailed in the Supplementary Methods. SUVmax was defined as the highest decay-corrected activity concentration per injected dose per body weight. The treatment and follow-up are shown in the Supplementary Methods.

Statistical analysis
The primary endpoint was failure-free survival (FFS), which was defined as the time from diagnosis to failure (locoregional recurrence or distant metastasis) or death. The secondary endpoints were overall survival (OS, from diagnosis to death from any cause), regional relapse-free survival (RRFS, from diagnosis to regional recurrence or death), and distant metastasis-free survival (DMFS, from diagnosis to distant metastasis or death).
The sensitivity and specificity of [ 18 F]FDG PET/CT and MRI were compared using McNemar's paired-sample test, and confidence intervals for proportions were calculated according to the efficient-score method described by Robert Newcombe [28]. Time-dependent receiver operating characteristic (ROC) curve analysis was applied to determine the cutoff values of the continuous variables using the "survival ROC" package in R. The survival curves were compared by the log-rank test. Univariate and multivariate analyses were performed by Cox regression. Statistical analysis was conducted using SPSS 26.0 and R software (version 4.0.1, http:// www.r-proje ct. org/). A two-sided p < 0.05 was deemed statistically significant.

Patient characteristics
After screening, 542 and 353 eligible patients were enrolled in the training cohort and the validation cohort, respectively. In the training cohort, the median age was 44 years, ranging from 16 to 73 years. The cutoff value of SUVmax-N was 9.  Table 1.

[ 18 F]FDG PET/CT versus MRI
In the whole cohort, 174 patients underwent cervical lymph node fine-needle aspiration biopsy guided by ultrasonography. Among the 224 biopsied lymph nodes of 174 patients, 132 and 92 lymph nodes were pathologically confirmed positive and negative, respectively. [ 18   The analysis showed that patients with grade 3 radiologic extranodal extension had significantly lower FFS than those with grades 0, 1, and 2 radiologic extranodal extension (p < 0.001, p < 0.001, and p = 0.003). The survival curve is shown in Supplementary Fig. 3. As presented in Table 3, multivariate analysis demonstrated that SUVmax-N higher than 9.3, nodal necrosis, and grade 3 radiologic extranodal extension were independent factors of a poor prognosis for FFS (p = 0.035, p < 0.001, and p = 0.001, respectively). For graded radiologic extranodal extension, only grade 3, but not grade 0-2 radiologic extranodal extension, predicted an inferior FFS (hazard ratio [HR]: 2.703, 95% CI: 1.547-4.724, p = 0.001; Table 3).
Similarly, the above lymph node characteristics were also independent factors for DMFS, while grade 3 radiologic extranodal extension was the only significant independent factor for RRFS (Supplementary Table 1).

Radiologic score and risk stratification
Prognostic factors obtained from the multivariate analysis were used for risk stratification. One risk factor scored  Supplementary Fig. 4). Therefore, patients were stratified into a high-risk group (radiologic score > 0, n = 329) and a low-risk group (radiologic score = 0, n = 213) by their radiologic score. The baseline characteristics of participants in both risk groups are summarized in Supplementary

Benefit of induction chemotherapy
In the whole training cohort, FFS was not significantly different for patients with or without induction chemotherapy before concurrent chemoradiotherapy (HR: 0.72, 95% CI: 0.49-1.09, p = 0.12 by univariate analysis; Fig. 3a). However, in the high-risk group, patients who received induction followed by concurrent  Table 3, and Table 4.  , and 29 (8.2%) patients had grade 0, grade 1, grade 2, and grade 3 radiologic extranodal extension, respectively. Grade 3 radiologic extranodal extension, SUVmax-N (≥ 9.3), and nodal necrosis were confirmed as significant factors of a poor prognosis for FFS in the validation group (all p < 0.05; Supplementary Fig. 6). The prospective validation set was stratified into a highrisk group (n = 162) and a low-risk group (n = 191) according to the radiologic score identified in the training set. The 3-year FFS rate for patients in the high-risk group was lower than that for patients in the low-risk group (96.9% vs. 84.2%, p < 0.001; Supplementary Fig. 6e).
As shown in Fig. 3, there were no significant differences in FFS between the two treatment models in the whole validation cohort and the low-risk group (p = 0.1, p = 0.52). However, patients undergoing induction chemotherapy followed by concurrent chemoradiotherapy had a higher survival rate than those undergoing concurrent chemoradiotherapy alone in the high-risk group (3-year FFS: 92.2% vs. 80.2%; HR: 0.34, 95% CI: 0.13-0.93, p = 0.028; Fig. 3f).

Discussion
In this large cohort study, [ 18 F]FDG PET/CT was more accurate than MRI for detecting metastatic cervical lymph nodes, which provided firm evidence for precisely identifying T3N1M0 patients by both [ 18 F]FDG PET/CT and MRI. Radiologic lymph node characteristics, including SUVmax-N higher than 9.3, nodal necrosis, and grade 3 radiologic extranodal extension, were independent prognostic factors for nasopharyngeal carcinoma patients staged as T3N1M0. Accordingly, high-risk and low-risk groups could be stratified by these risk factors instead of sex and EBV DNA load. We demonstrated that only patients in the high-risk group could benefit from the addition of induction chemotherapy. These findings were verified again by the validation cohort from our clinical trial.
In this study, EBV DNA load was not confirmed for separating the subgroup of T3N1M0. As 206 and 96 patients in the training and validation cohorts had EBV DNA loads higher than 2000 copies/mL, an insufficient sample size may not justify this result. Similarly, female patients also did not show a superior survival rate. Consistent with a prior study [29], [ 18 F]FDG PET/CT did have better diagnostic accuracy than conventional imaging in nasopharyngeal carcinoma, as proven by the histological results of the lymph nodes. Therefore, the subgroup of T3N1M0 staged by both [ 18 F] FDG PET/CT and MRI in this study showed a relatively high survival rate, which was close to the reported rate of stage II patients [30]. As a result, extremely strong predictive markers are required for further separation of this subgroup.
[ 18 F]FDG PET/CT, as functional imaging, provides metabolic information and can guide prognostication [31]. Prior studies have reported that SUVmax-T, SUVmax-N, and SUV 75% of primary tumors are prognostic factors for nasopharyngeal carcinoma [32,33]. Thus, it was not absurd that SUVmax-N also acted as an independent prognostic factor in the subgroup of patients with T3N1M0. Pathologic extranodal extension has been introduced into the N classification for nonviral-related head and neck cancer in the 8th edition of the AJCC/UICC staging system. Due to the radiotherapy-based primary treatment, pathologic extranodal extension is not available for nasopharyngeal carcinoma. However, radiologic extranodal extension based on MRI or CT has good specificity and sensitivity (ranging from 70 to 90%) in predicting pathologic extranodal extension in head and neck cancer [34]. The specificity of radiologic extranodal extension infiltrating adjacent structures is nearly 100%, consistent with pathologic extranodal extension [18,35]. Therefore, radiologic extranodal extension based on MRI is an accepted surrogate of pathologic extranodal extension for nasopharyngeal carcinoma. Similar to previous studies [19][20][21], the most severe radiologic extranodal extension with the involvement of adjacent structures was correlated with poor survival outcomes in the subgroup of T3N1M0 nasopharyngeal carcinoma. In addition, nodal necrosis, a vital radiologic nodal feature, is a reliable sign for detecting nodal metastasis. MRI has similar sensitivity to CT in identifying nodal necrosis [27]. Previous studies indicated that nodal necrosis is a strong prognostic factor in nasopharyngeal carcinoma, as the survival rate of patients with nodal necrosis declined nearly 12% in comparison with that of patients without nodal necrosis [36]. Therefore, it was not unreasonable that nodal necrosis was a significant predictor of outcomes in the T3N1M0 subgroup.
In previous studies [19][20][21], the TNM stage of enrolled patients varied from stage I to stage IVa, which contained significant heterogeneity. The confounding factors, including T stage, nodal size, nodal level, nodal laterality, and treatment modes, could not be completely eliminated. In our study, the metastatic lymph nodes of T3N1M0 patients were located in the unilateral upper cervical region, which fully controlled the covariate factors of T stage and radiological lymph node characteristics, such as nodal level and nodal laterality. As mentioned above, after eliminating confounding factors, our study confirmed that the three radiological lymph node characteristics, namely SUVmax-N, extranodal extension, and nodal necrosis, were closely related to survival outcome, especially distant metastasis. Perhaps if lymph nodes have a high metabolic rate and tumors spread outside the nodal capsule, tumor cells can easily enter the blood circulation and finally develop metastasis in distant organs. Although intensity-modulated radiotherapy delivers radical doses to the primary tumor and metastatic lymph nodes and can achieve excellent locoregional control [4], the high risk of distant metastasis for patients with these sorts of nodal characteristics cannot be reduced. As we found in the present study, high-risk T3N1M0 patients had a similar 5-year FFS rate to patients who were enrolled in clinical trials of induction chemotherapy [37]. Given the confirmed benefit of induction chemotherapy in locoregionally advanced nasopharyngeal carcinoma from randomized controlled trials [5][6][7] and meta-analyses [38,39], it is highly reasonable that induction chemotherapy plus concurrent chemoradiotherapy can improve the survival rate of highrisk T3N1M0 patients.
There are several advantages of this study. First, all patients were restaged by [ 18 F]FDG PET/CT and MRI according to the 8th AJCC/UICC staging system. The subset of patients who underwent biopsy of certain cervical lymph nodes demonstrated the accuracy of [ 18 F]FDG PET/ CT in detecting positive lymph nodes, which supported the reliability of the T3N1M0 staging of the patients. Notably, all eligible patients were upper cervical lymph node positive and unilateral lymph node positive, which fully eliminated covariate factors, including T stage, nodal laterality, nodal level, and nodal size. In addition, the sample size was relatively large, and the results were verified by a validation cohort. Limitations of this study should also be noted. First, this was a single-center study, and WHO type III was the predominant pathology type. Second, the follow-up duration of the validation cohort was not long enough. Hence, subsequent follow-up is warranted.

Conclusion
In conclusion, T3N1M0 patients could be diagnosed more accurately by both [ 18 F]FDG PET/CT and MRI. The radiologic score of lymph node characteristics based on MRI and [ 18 F]FDG PET/CT could correctly identify high-risk patients who can obtain additional survival benefit from induction chemotherapy and it could be used to guide individualized treatment for nasopharyngeal carcinoma patients staged with T3N1M0 in clinical practice.