The Role of Pretreatment ¹⁸F-FDG PET/CT for Early Prediction of Neoadjuvant Chemotherapy Response in Patients with Locoregionally Advanced Nasopharyngeal Carcinoma

Introduction

Nasopharyngeal carcinoma (NPC) is an extremely common head and neck cancer affecting 10 to 25 cases per 100,000 in Southern China.¹ Unfortunately, more than 70% of patients with NPC are categorized with locoregionally advanced disease at initial diagnosis.^2,3 Despite the use of intensity-modulated radiotherapy and chemotherapy, outcomes for patients with advanced disease still remain unsatisfactory, with 30–50% of patients suffering from disease relapse post treatment.³ It is challenging for clinicians to manage locoregionally advanced NPC (LANPC). The standard treatment for LANPC is concurrent chemoradiotherapy (CCRT).⁴ Neoadjuvant chemotherapy (NAC), given before CCRT, has been used increasingly to control LANPC. Several randomized trials^5,6 showed the superiority of additional NAC over CCRT alone in terms of overall survival (OS) and disease-free survival (DFS). However, current meta-analyses by the MAC-NPC Collaborative Group failed to identify the advantages of NAC followed by CCRT compared to CCRT alone.⁷ In contrast, patients treated with NAC plus CCRT had more grade 3–4 adverse events than those treated with CCRT alone.⁵ Hence, early assessment of NAC response is of utmost importance for NPC due to its contribution in minimizing ineffectual and toxic chemotherapies.

To assess NAC response, conventional imaging methods (ultrasound, SPECT/CT [single-photon emission computed tomography-computed tomography] and MRI [magnetic resonance imaging]) have been utilized.^8–10 Yet, these approaches have offered limited accuracy. Du et al⁹ proposed that prediction of chemotherapy responses in NPC patients can be elicited from the uptake ratio of ^99mTc-MIBI in early phases. However, calculating ^99mTc-MIBI by utilizing planar images can be subject to personal error given lower resolution, contributing to inconsistency in results across various studies.¹¹ Segara et al¹² offered a scoring metric to evaluate the pathologic response to NAC, demonstrating that breast MRI provides increased accuracy over other imaging modalities. Nevertheless, MRI has been reported with varied accuracy and diminished specificity for the breast.¹³ A promising field of radiomics has emerged recently, and several studies suggest that radiomics may be a beneficial instrument in individualized NAC treatment of LANPC.^14,15 However, the clinical applicability of radiomics analysis, which demands specialized workstations and trained professionals, has been largely limited by the complexity of its application at present. Hence, it is of utmost importance to identify novel factors to guide NAC treatment.

¹⁸F-Fluorodeoxyglucose (FDG) positron emission tomography (PET)/CT is frequently utilized for clinical diagnosis and determination of tumor staging for NPC. Prior to this, the utilization of PET/CT was not limited to the diagnosis and detection of metastatic lesions in NPC.¹⁶ The maximum standardized uptake value (SUVmax) of PET/CT is an index that reflects tumor metabolism. Several studies^17,18 have suggested that higher SUVmax from the primary tumor site (SUVmax-PT) was associated with poorer prognosis in NPC patients. Apart from SUVmax-PT, total lesion glycolysis (TLG) from the primary tumor site (TLG-PT) may be a more accurate prognosis marker, which quantifies the metabolic activity and volumetric burden of tumors.¹⁹ A recent meta-analysis of ¹⁸F-FDG PET/CT reported by Li et al²⁰ showed that both SUVmax-PT and TLG-PT were significantly associated with NPC patients’ treatment and outcomes. However, there are few studies to evaluate baseline ¹⁸F-FDG PET/CT and its utilization in predicting the NAC response among patients with NPC. Notably, studies have confirmed ¹⁸F-FDG PET/CT to be useful in early prediction of NAC response to breast and bladder cancer.^21,22 We therefore hypothesize that ¹⁸F-FDG PET/CT may be useful in predicting NAC response in NPC.

Given this background, we conducted this study to determine the accuracy of baseline ¹⁸F-FDG PET/CT in assessing the response to NAC among patients with LANPC. Moreover, a combination of metabolic parameters and clinical variables was included to create a nomogram model to more accurately predict NAC response.

Materials and Methods

Participant Inclusion

Between June 2017 and July 2020, newly diagnosed NPC patients who received treatment at our institution were retrospectively analyzed. The eligibility criteria were as follows: (1) non-metastatic, histologically demonstrated NPC; (2) stage III or IVA NPC (8^th AJCC staging system); (3) received no less than two cycles of NAC; (4) whole-body ¹⁸F-FDG PET/CT performed by our institution prior to NAC; (5) completed MRI prior to NAC and after the second cycle of NAC; (6) available medical records, including age, gender, TNM stage, NAC regimens, serum Epstein‑Barr virus (EBV) DNA load, and lactate dehydrogenase (LDH). Patients with additional active cancers, pregnant or lactating, or with improperly medicated cardiac disease were excluded. Figure 1 flowchart represents patient selection. In total, 121 patients who met the eligibility criteria were retrospectively analyzed in the current study. The median age of all the participants was 50 years old (range: 19–74 years), with 94 (77.7%) males and 27 (22.3%) females. Detailed information on the treatment is available in Supplement 1. The Ethics Committee of The Fifth Affiliated Hospital, Sun Yat-sen University has approved this retrospective study, and it was run in accordance with the 1964 Declaration of Helsinki and all its amendments.

Figure 1 Flowchart demonstrating patient selection in final analysis along with exclusion criteria. Abbreviations: NPC, nasopharyngeal carcinoma; RT, radiotherapy; CCRT, concurrent chemoradiotherapy; AC, adjuvant chemotherapy; NAC, neoadjuvant chemotherapy.

¹⁸F-FDG PET/CT Data Acquisition

Prior to the examination, all patients were mandated to fast for a minimum of 6 hours prior to examination. ¹⁸F-FDG was manufactured in line with standards set by our hospital, utilizing a 112-ring digital light guide PET/CT scanner (uMI780, United Imaging, China). The radiochemical purity was maintained at 98%, with the final products being diluted with saline and sterilized via a 0.22-μm Millipore filter through a sterile syringe. Doses of intravenously injected ¹⁸F-FDG were calculated utilizing each patient’s weight (3.7 MBq [0.1 mCi/kg]). The bounds for the images were set to CT scan from the head to upper thighs. PET scans were performed promptly post CT scan with 1.5 minutes of scanning time per image, utilizing 3-D mode for all images.

Two experienced nuclear medicine physicians (YW and JJW) examined all PET/CT images. For quantitative analysis, the SUVmax-PT was normalized to body weight and automatically calculated for the primary tumor. TLG-PT was calculated using SUVmean × MTV (metabolic tumor volume, whcih recorded at the absolute SUV threshold of 2.5 and the relative SUVmax-PT threshold of 70%). In the current study, we did not include SUVmax and TLG for the neck lymph nodes in the analysis, given that neck lymph node inflammation and the complex biology of lymph nodes may inadvertently influence the results of the research.²³

Magnetic Resonance Imaging (MRI) Scan

The MRI study was done using a 3.0-T system (Verio; Siemens Healthcare, Erlangen, Germany). A combined head-and-neck coil was applied to inspect the area from the sternal end of the clavicle to the suprasellar cistern. Complete data on MRI protocol is described in the Supplement 2. All patients completed two MRI studies. The MRI scans were performed prior to treatment and repeated two weeks after the second NAC administration.

Tumor Response Evaluation

Tumor response was evaluated by two radiologists (JBG and GJW) specializing in NPC MRI based on Response Evaluation Criteria for Solid Tumor (RECIST) 1.1 Criteria,²⁴ with disagreements resolved by consensus. Specifically, the NAC response was determined by combining the results of the evaluation of nasopharynx lesion and of metastatic lymph nodes in this study. The variation of the maximum diameter of nasopharynx lesion and the short axis of neck lymph nodes was calculated. The nasopharynx lesion with the longest diameter was recorded and only neck lymph nodes with the short axis of at least 15 mm were considered, and no more than five target lesions were evaluated. In addition, all the diameters and short axis measurements were done on the transverse plane. The overall biological responses after NAC were grouped as complete response (CR), partial response (PR), stable disease (SD) and progressive disease (PD). Participants who experienced CR/PR were categorized as “responders”; while participants who suffered from SD/PD were categorized as “non-responders” in this study.

Statistical Analysis

Receiver-operating characteristic (ROC) curves were utilized on metabolic parameters to assess an optimal cut-off value for NAC response. The AUC ([area under the ROC]) was computed to understand the sensitivity and specificity of all metabolic parameters in calculating NAC response. Assessed covariates include SUVmax-PT (low vs high), TLG-PT (low vs high), gender, age (<50 vs ≥50 years), T stage (T1–2 vs T3–4), N stage (N0–1 vs N2–3), serum EBV DNA (<4000 vs ≥4000 copy/mL), LDH (<245 vs ≥245 U/L), and NAC regimen (GP [cisplatin 80 mg/m² D1 and gemcitabine 1 mg/m² D1, D8] vs TP [cisplatin 75 mg/m² D1 and docetaxel 75 mg/m² D1 or paclitaxel 150–180 mg/m² D1]/PF [cisplatin 80–100 mg/m² D1 and 5–fluorouracil 800–1,000mg/m², by 120-h continuous intravenous infusion] vs TPF [cisplatin 60 mg/m² D1, docetaxel 60 mg/m² D1 or paclitaxel 135 mg/m² D1, and 5-fluorouracil 500–800 mg/m², by 120-h continuous intravenous infusion]). The cutoff value was set at 4000 copies/mL and was selected to categorize patients into high- and low pre-EBV DNA groups as described in prior NPC studies.²⁵ Normal LDH ranged from 109 to 245 U/L; thus, a cutoff value of 245 U/L was selected to categorize patients into high- and low LDH.

In this study, we first conducted a univariate analysis for each independent variable with the NAC response, then selected the independent variables with a P value <0.1 and included them into the final multivariate analysis. To determine which patients are most likely to benefit from NAC, recursive partitioning analysis (RPA) based on metabolic parameters was applied. A random number generator was used to sort patients into a training cohort containing 81 patients and an internal validation cohort of 40 patients. Nomograms were created using the rms package within R, and gender was included into the nomogram given the importance of gender in clinical decision-making.³ The concordance index (C-index) was utilized to estimate bias-correction or overfitting within the model. P < 0.05 were categorized as significant, and all statistical assessments were two-sided. All statistical analyses were performed in R version 3.3.2 (http://www.r-project.org/).

Results

Patient Characteristics

The clinical characteristics of all 121 enrolled participants are categorized in Table 1. Of the 121 patients, 54 (44.6%) received GP regimen of NAC, 48 (39.7%) received TP/PF regimen, and 19 (15.7%) received TPF regimen. Disease in stage III accounted for 88 (72.7%) cases, and stage IVA was identified for 33 (27.3%) cases. The SUVmax-PT ranged from 2.7 to 25.7, with a median value of 10.6. The TLG-PT ranged from 0.6 to 205.4, with a median of 54.5. After two cycles of NAC, 96 participants were categorized as “responders” (7 having CR and 89 having PR) and 25 participants as “non-responders” (22 having SD and 3 having PD) by the RECIST 1.1 Criteria.

Table 1 Clinical Characteristics of 121 Patients with LANPC

Correlation Between Tumor Response and Metabolic Parameters

An ROC curve was plotted to determine an absolute cut-off value for SUVmax-PT and TLG-PT in tumor response to NAC. Optimal cut-offs for SUVmax-PT and TLG-PT were 11.8 (AUC value, 0.744; specificity 69.8%, sensitivity 72.0%; p < 0.001) and 38.5 (AUC value, 0.720; specificity 63.5%, sensitivity 76.0%, p < 0.001), respectively. Among the entire group, 47 patients (38.8%) showed high SUVmax-PT, and 54 (44.6%) showed high TLG-PT. Univariate analysis showed that SUVmax-PT (HR, 5.94; 95% CI, 2.24–15.76; p < 0.001) and TLG-PT (HR, 5.52; 95% CI, 2.02–15.12; p < 0.001) were significantly associated with NAC response (Table 2). In multivariate analysis, non-responders were significantly associated with high SUVmax-PT (HR, 3.49; 95% CI, 1.17–10.36; p = 0.024), TLG-PT (HR, 4.45; 95% CI, 1.44–13.78; p = 0.010) and LDH (HR, 1.83; 95% CI, 1.15–5.34; p = 0.048), but not with chemotherapy (HR, 1.30; 95% CI, 0.53–3.21; p= 0.830) (Table 3).

Table 2 Univariate Analysis of Prognostic Factors for NAC Response

Table 3 Multivariate Analysis of Prognostic Factors for NAC Response

Benefit Stratification for NAC According to Metabolic Parameters

Since both SUVmax-PT and TLG-PT were shown to be independently related to NAC response in univariate and multivariate analyses, these above two PET parameters were integrated into an RPA model for the prediction of chemotherapy response. The RPA algorithm separated the 121 patients with NPC into 4 classes (Figure 2). Because the NAC responses of classes 2 and 3 were not significantly different (class 2 vs 3, p = 0.767), these two groups were combined. Thus, the final RPA model categorized patients into the following three prognostic groups: high response (low SUVmax-PT with low TLG-PT; n = 53), intermediate response (low SUVmax-PT with high TLG-PT, and high SUVmax-PT with low TLG-PT; n = 35), and low response (high SUVmax with high TLG-PT; n = 33), with corresponding NAC response rates of 94%, 80%, and 55%, respectively.

Figure 2 Prognostic grouping by recursive partitioning analysis in patients with LANPC for predicting NAC response. Abbreviations: LANPC, locoregionally advanced nasopharyngeal carcinoma; NAC, neoadjuvant chemotherapy.

Establishment of Metabolic and Clinical Nomogram

For the prediction of NAC responses, a nomogram was developed utilizing results from multivariate analysis (Figure 3A). Since SUVmax-PT and TLG-PT were predictors of NAC response after multivariate analysis, this elicited inclusion in the nomogram. The established nomogram yielded a high C-index of 0.787 (95% CI, 0.533–1.000; Figure 3B) in the training cohort and 0.825 (95% CI, 0.570–1.000; Figure 3C) in the validation cohort. Individual PET parameters were utilized to assess the validity of the nomogram by comparing ROC curves. The AUC values of SUVmax-PT, TLG-PT, and the nomogram were 0.744, 0.720, and 0.787, respectively (p < 0.001; Supplement 3). The results revealed that the nomogram predicted NAC response, use of PET parameters in conjunction with clinical risk variables, was superior to an individual PET parameter.

Figure 3 Nomogram (A) to predict NAC response in LANPC. Calibration curves for the nomogram to calculate NAC response with the training cohort (B) and with an internal validation cohort (C). Abbreviations: NAC, neoadjuvant chemotherapy; LANPC, locoregionally advanced nasopharyngeal carcinoma.

Discussion

¹⁸F-FDG PET/CT was identified as a prognostic instrument for NPC based on the high sensitivity and noninvasive assessment methods for identifying tumor metabolism.²⁶ However, there were a limited number of reports supporting the use of ¹⁸F-FDG PET/CT for the identification of NPC tumor response to NAC. This case series using ¹⁸F-FDG PET/CT to predict NAC response in LANPC is reported to add to the literature on non-invasive and highly sensitive assessment methods. Preliminary outcomes found SUVmax-PT and TLG-PT to be highly effective in identifying the subpopulation sensitive to NAC. Furthermore, we also built a nomogram based on ¹⁸F-FDG PET/CT bounds in conjunction with other variables to precisely predict individual response to NAC, which would be useful to guide clinical decisions.

Tumour response to NAC is a significant prognostic factor in NPC.^27–29 Generally, favourable NAC response was significantly associated with improved disease control, locoregional control, and overall survival for NPC. In contrast, unfavorable NAC response was correlated with poor clinical outcomes. Hence, evaluating NAC response may assist in prognostication and refining of treatments for high-risk NPC patients. Clinical studies on the efficacy of ¹⁸F-FDG PET/CT in predicting NAC response have been presented for lung cancer, breast cancer, and head and neck cancer.^30–32 For example, Vallius et al assessed the benefit of ¹⁸F-FDG PET/CT to identify NAC in non-responsive patients after three or four cycles;³³ and a decrease in FDG uptake after one cycle of NAC was associated with improved tumor response to ovarian cancer. However, patients with chemo-resistant NPC have been shown to have a delayed need for radiotherapy, reduced toxic exposure as well as reduced cost with the use of NAC. Thus, early screening to determine chemo-resistant NPC would allow clinicians to personalize treatment protocols.

For this study, all patients completed baseline ¹⁸F-FDG PET/CT and received two cycles of standard NAC regimen. Our results indicated that patients who responded favorably to NAC had significantly lower SUVmax-PT and TLG-PT than non-responders. The potential mechanisms underlying this observation may be explained as follows. It is well recognized that intratumoral heterogeneity contributes to drug resistance and treatment failure.^34,35 Prior studies have confirmed that the intratumoral heterogeneity could be identified based on variations in glucose metabolism,^36,37 with relevant molecular and cellular characteristics including, fibrosis, necrosis, and hypoxia. Consistent with our study,³⁸ Wang et al indicated an uptake ratio of ^99mTc-MIBI at the primary tumor site, which may predict a reaction to NAC in NPC. Still, the calculation of ^99mTc-MIBI using planar imaging is vulnerable to human error due to limited resolution, contributing to discrepancies in the findings among studies.¹² Compared to ^99mTc-MIBI, PET/CT is highly advantageous for its production of images with intrinsically greater contrast allowing for increased specificity and sensitivity in lesion detection.²⁶ Given that NPC is often enclosed by normal tissues that will take up a substantial amount of ^99mTc-MIBI, adoption of ¹⁸F-FDG PET/CT improves measurement accuracy of tumor uptake ratios and minimizes the impact of standard uptake, thus maintaining stability within the results.²⁶

To understand which patients may benefit from NAC, researchers often use multivariate analysis to elucidate relevant prognostic factors.^31–33 However, these prognostic factors are often combined with independent factors, creating confusion for clinicians when several prognostic factors coexist. To avoid the interaction of prognostic factors with risk of tumor response, we analyzed the role of metabolic parameters by hierarchical analysis, which is unique compared to traditional statistical analysis methods used to date.³⁰ RPA is one of the most successful models for stratifying patients into subgroups with homogeneous performance and has been implemented to create prognostic algorithms for numerous malignancies.^39,40 We found that these objective risk factors (SUVmax-PT and TLG-PT) were confirmed to influence NAC response significantly in certain subgroups in the RPA model. As a result, the RPA model then categorized patients into different response-risk groups according to metabolic parameters. In further stratified analysis, we found that the NAC response is more apparent in intermediate- and high-response groups. By contrast, nearly 50% of patients in the low-response group responded poorly to NAC. As a result, clinicians can screen and identify patients with LANPC who will benefit most from NAC while providing them with sound advice on NAC. Moreover, some low-response patients can avoid adverse effects from NAC and experience reduced overall treatment costs.

Our nomogram utilizing ¹⁸F-FDG PET/CT boundaries provided an interesting result. While methods traditionally use prognostic variables such as TNM stage, sex, and blood parameters, our nomogram provides biological heterogeneity for tumors. Thus, this prognostic nomogram may offer a simple and precise method of identification for improved patient recovery from NAC in LANPC. One recent study established an MRI-based radiomics nomogram containing 120 patients with NPC, for which it provided a high AUC value of 0.822, to predict responses to NAC. However, the establishment of MRI radiomics models to predict NPC should consider several important points. Primarily, MRI-based radiomic studies face technological challenges. Both image acquisition, which distorts tissue properties, and scanner limitations, which decrease the reproducibility of tumor features, affect the quality of MRI-based studies.⁴¹ Secondly, MRI varies in scanner properties, affecting the reproducibility of images and, in turn, the tumor features identified within them.⁴² Extreme variability in scans and systems limits generalizability and merit for these systems to be used in clinical decisions. In contrast to radiomic features from MRI, metabolic parameters indicated by ¹⁸F-FDG PET/CT are easier to obtain, and our nomogram may be more clinically practical.

Another biological mechanism associated with NPC is plasma EBV DNA.⁴³ Prior studies indicate that plasma EBV DNA may be utilized for clinical and molecular monitoring in NPC patients.⁴³ However, the prognostic benefits of plasma EBV DNA in predicting tumor reaction to NAC have not been sufficiently studied. Recently, Liu et al²⁹ assessed the prognostic benefit of EBV DNA values on tumor reaction in NPC and identified detectable EBV DNA following NAC to be associated with poor tumor response. However, baseline EBV DNA levels failed to predict tumor response to NAC in the current study. One reason underlying this observation may be differences in the time points of EBV DNA examination between the Liu et al study and this study. Currently, NAC response is calculated most often by the change in tumor size, yet the tumor size in non-responders shrank insignificantly or increased.²⁴ Usually, patients with a poor response to NAC are more likely to have a larger residual tumor lesions, and EBV DNA load was confirmed to be a reliable indicator of tumor burden in patients with NPC. Therefore, it is reasonable to conclude that a detectable EBV DNA after NAC treatment is associated with poor tumor response.

The current study has some limitations. Primarily, all data was obtained exclusively from a single center with few patients. Consequently, future multi-center studies with a higher number of patients may be warranted. Second, confined to the follow-up duration, the prognostic value of glucose metabolic parameters on survival outcomes was lacking for the current study. Although prior analysis of NPC indicated that the reaction to primary chemotherapy is related to outcome,^28,29 prolonged follow-up times are essential to assess long-term outcomes. Finally, the efficacy of a variety of NAC regimens may have confounded the main findings of this research. Yet, to date, no evidence has been found to suggest an ideal NAC regimen, and all regimens used for this study are in line with the National Comprehensive Cancer Network Guidelines.

In summary, our findings suggest that metabolic parameters (SUVmax-PT and TLG-PT) measured using ¹⁸F-FDG PET/CT were significantly associated with NAC response in LANPC. Moreover, the nomogram based on metabolic parameters with other variables had good prognostic accuracy in predicting NAC response. We therefore cautiously suggest that ¹⁸F-FDG PET/CT would be a useful non-invasive method for early indication of NAC efficacy in guiding clinical decisions. Future studies with larger sample sizes are still needed to validate these findings.