Induction Cisplatin and Gemcitabine Versus Cisplatin, Fluorouracil, and Docetaxel in Locoregionally Advanced Nasopharyngeal Carcinoma: a Network Meta-analysis


 Background and purposeCisplatin plus gemcitabine (GP), and cisplatin, fluorouracil plus docetaxel (TPF) has been approved in induction chemotherapy in locoregionally advanced nasopharyngeal carcinoma (LANPC). We compared the clinical outcomes and toxicities of these two regimens by network meta-analysis.Materials and methodsWe systematically searched studies comparing different regimens of IC plus CCRT versus CCRT alone for LANPC. Conventional and Bayesian network meta-analysis were conducted for statistical analysis.ResultsNine eligible studies with a total of 4,977 patients were involved. Compared with CCRT alone, IC+CCRT significantly improved PFS (HR=0.65 [95%CI, 0.54-0.77]) and OS (HR=0.68 [95%CI, 0.54-0.85]) in conventional meta-analysis. In Bayesian network meta-analysis, only TPF and GP improved clinical outcomes significantly. GP showed non-significant trends of superiority over TPF without increasing the incidence of ≥ grade 3 adverse events.Conclusion﻿Induction GP regimen showed non-significant trends of superiority in terms of clinical outcomes compared with TPF regimen in LANPC.


Introduction
Nasopharyngeal carcinoma (NPC) is a unique epithelial carcinoma arising from the mucosa of the nasopharynx with distinct geographical distribution. NPC is highly prevalent in South China. The estimated global new cases of NPC in 2020 is 133,354 (0.7%) [1]. In China, 60,600 new cases were estimated in 2015 with a ratio of about 2.5 between males and females [2]. The outcomes of NPC have been improved substantially with the advance in early disease screen, development of more sensitive diagnostic markers and imaging tools, and the use of optimized chemoradiotherapy.
NPC is sensitive to radiotherapy and chemotherapy and he prognosis of NPC has been greatly improved. The ve-year overall survival is over 90% for stage I-II, 83% for stage III, and 71% for stage non-metastatic IVa diseases [3].
Induction chemotherapy (IC) with platinum-based chemotherapy followed by concurrent chemoradiotherapy (CCRT) is the standard of care for locoregionally advanced (stage III-IVa) NPC [4]. The addition of IC to CCRT led to signi cant improved clinical outcomes in terms of progression-free survival, overall survival, locoregional control, and distant control, as compared with CCRT alone in locoregionally advanced NPC [5][6][7]. Currently, preferred regimens including GP (gemcitabine plus cisplatin, category 1 recommendation), TPF (docetaxel, cisplatin plus 5-uorouracil, category recommendation for EBV-positive NPC and category 2A for non-EBV-associated NPC) are recommended for IC in NPC. Other choice of IC could be TP (docetaxel plus cisplatin), PF (cisplatin plus 5-uorouracil), and PX (cisplatin plus capecitabine). The bene t of IC with TPF regimen was shown in a phase III randomized clinical trial (NCT01245959) that allocated 241 patients and 239 patients with locoregionally advanced NPC into the IC plus CCRT group, and the CCRT group, respectively. Compared with CCRT alone, IC plus CCRT led to signi cantly improved 5-year locoregional failure-free survival (83.8% versus 90.7%, p=0.044), distant failure-free survival (79.8% versus 88.0%, p=0.030), as well as overall survival (77.7% versus 85.6%, p=0.042) without adding to late toxicities [8,9]. Another phase III randomized clinical trial compared CCRT with (n=242) or without IC (n=238) using GP regimen (NCT01872962 Although both TPF and GP regimens showed promising effect in combination with CCRT, currently, no study has directly compared the bene t and safety of these regimens. Three-year recurrence-free survival (80% for TPF versus 85.3% for GP) and overall survival (92% for TPF versus 94.6% for GP) seemed comparable. Grade 3-4 adverse events such as neutropenia, leucopenia, diarrhea, mucositis were more comment in the IC phase for TPF, while vomiting was more comment for GP. In the CCRT phase, mucositis was more frequently seen for TPF, whereas risk of anemia and thrombocyteopenia was higher for GP [9,10]. In addition, GP was shown to be cost-effective compared with TPF [11].
Differed from conventional systematic reviews and pairwise meta-analysis that are only capable of comparing two "head-to-head" interventions at a time, network meta-analysis integrates information from all relevant studies, uses statistical methods such as frequentist or Bayesian approach, and thus provides new methods allowing direct and indirect comparison of effectiveness and safety among a network of interventions under the same scenario [12]. Although conclusions should be interpreted with caution due to inherent limitations of network meta-analysis such as potential biases, it could still give valuable information when result from direct comparative study is not available. [13] In this study, we performed a network meta-analysis to compare the effectiveness and safety pro les of multiple induction chemotherapy regimens in locoregionally advanced NPC with a focus on the comparison of TPF and GP regimens. We aimed to explore the optimal IC regimens and provide potential suggestions for clinical practice.

Literature search strategy
We searched published literature from electronic databases including PubMed, Embase and the Cochrane Library for studies that compared different regimens of IC plus CCRT or IC+CCRT versus CCRT in LANPC from Jan 1966 to Jan 2020. We used the following searching equation: "(((nasopharyngeal carcinoma) OR (nasopharyngeal neoplasms)) OR (nasopharyngeal cancer)) AND (((induction) OR (induction chemotherapy)) OR (neoadjuvant chemotherapy))". No limits were applied to language. Furthermore, published meeting proceedings or abstracts from conferences (2) studies with inferior quality or high risk of bias; (3) data was not available for analysis; (4) reported outcomes were not of current interest.

Data extraction and collection
Independently, two researchers (QW and SL) assessed the potential eligibility of the imported records by examining the titles and abstracts, and any divergence would be settled by two other authors (JL and YZ). A standardized data collection form was adopted to extract and gather publication information, experimental design and outcomes. All the digital index of outcomes were validated and managed by entering into the Review Manger (version 5.4, http:// www.cochrane.org). Above-mentioned procedures had gone through inspection.

Methodological quality assessment
Before constructing the Bayesian network meta-analysis, a methodological quality assessment was conducted on the included studies. For RCTs, we evaluated the quality from three aspects: 2 points for randomization, 2 points for blinding and 1 point for withdrawals and dropouts, using the Oxford Quality Score (also referred to as the Jadad Score) [14]. A cumulative score ≤ 2 was regarded as research of low quality, and ≥ 3 was considered as high quality research. Meanwhile, we assessed the quality of observational studies by using the Newcastle-Ottawa Scale (NOS), which also inspect three aspects of studies: 4 points for selection of objects, 2 points for comparability, and 3 points for measurement. A nal score ≤ 5 was regarded as research of low quality, and ≥ 6 was considered as high quality research. Ineligible studies would be excluded [15].

Outcomes
In order to capture all useful information, to reduce the risk of potential bias and to increase the credibility of the results, we set the overall survival rate (OS) at 3 years as the main endpoint, and the progression-free survival rate (PFS) at 3 years, treatment completion rate and ≥ grade 3 adverse events (AEs) as secondary endpoints, which were reported in most of the literature.

Statistical Analysis
We applied the following steps to conduct the statistical analysis.
Step 1: A traditional meta-analysis was constructed to compare OS and PFS of IC-CCRT vs CCRT. An overall HR < 1 meant favoring the experimental arm.
Step 2: Based on results of step one, we attempted to nd out which regimen of IC-CCRT was optimal. So a Bayesian network meta-analysis was created to explore comparable relationships among IC-CCRT regimens by using the package "gemtc" of the software R (MathSoft, version 4.0.3, https://www.r-project.org) [16,17].
Step 3: Regarding the heterogeneity among included studies, we chose random effects model and assessed the results of heterogeneity by I-square (I²) and forest plots.
Step 4: The Node-Splitting was use to calculate the inconsistency with the model of consistency. And if there were signi cant inconsistency, we would repeat the step with inconsistent model. Step 5: The nal results were visualized by forest plots and rank probability, and a HR < 1 or the highest bar located in far left meant favorable.

Study search and selection
After preliminary screening by reading article titles and abstracts, we included 17 articles for further assessment by reading full text and evaluating risk of bias. Finally, 9 of these 17 articles covering 4,977 patients were selected for subsequent statistical analysis, which contained 7 RCTs[10, 18-23] and 2 OBS [24,25] (Figure 1).

Study characteristics
Main characteristics of the 9 studies were summarized in Table 1. There were 7 RCTs, published from 2009 to 2019, involving 6 kinds of induction chemotherapies (namely, TP, docetaxel and cis-platinum; TPE, paclitaxel, epirubicin and cis-platinum; TGP, paclitaxel, gemcitabine and carboplatin; TPF, docetaxel, cis-platinum and uorouracil; PF, cis-platinum and uorouracil; GP, gemcitabine and cis-platinum). The two were OBS that assessed TPF+CCRT versus GP+CCRT, and TPF+CCRT versus PF+CCRT versus TP+CCRT, respectively. Additionally, the latter observational study (Peng et al. 2018) involved 3 different IC regimens (TPF, PF and TP), so we reused TPF to compared with PF and TP. The median follow-up time was from 3.4 years to 4.6 years. All studies had been assessed to have high methodological quality (the Jadad Score ≥ 3 or the NOS ≥ 6) ( Table 1).

Quality assessment
The analysis of both OS and PFS had acceptable heterogeneity (PFS: I²=0%; OS: I²=47%). The bias of publication was visualized by funnel plots. The scatters distributed evenly on both sides of the dotted lines in the two plots (Figure 2), indicating that the bias of the meta-analysis was acceptable.
We chose the random model and set the number of iterations as 50000 with other default options to construct the Bayesian network meta-analysis  Figure 6B). The rank probability map also indicated that GP might be superior to TPF ( Figure 6C). Therefore, current indirect evidence did not support signi cant difference between GP and TPF in terms of OS, although GP showed a slight tendency of superiority.

Treatment-related adverse events
The treatment-related toxicities might interrupt the treatment, mitigate the e cacy, or lower the quality of life of patients. We asked whether GP was superior to TPF in terms of treatment-related adverse events. We collected data of grade 3 and 4 adverse events, and found there was no difference in the incidence of severe adverse events (Figure 7). However the incidence of hematological toxicity, such as the most common leukopenia, GP was seem to be less than TPF (RR=0.55 [95%CI, 0.18-1.8]) (Figure 8).

Discussion
In this study, we applied network meta-analysis to compare the effectiveness and safety pro les among multiple induction chemotherapy regimens for locoregionally advanced NPC. We involved GP, PF, TP, TGP, TPF, TPE regimens and paid speci c attention on GP and TPF regimens. We found that only GP and TPF were associated with both improved progression-free survival and overall survival without marked increase in treatment-associated toxicities, while other regimens did not show statistically signi cant survival bene t. In addition, our study indicated that, although non statistically different, GP seemed superior to TPF in terms of PFS (HR=0.89 [95%CI, 0.48-1.7]) and OS (HR=0.8 [95%CI, 0.33-2.0]). These results suggest that GP and TPF are favorable regimens in IC for locoregionally advanced NPC, and comparative studies between these two regimens are needed.
Concurrent chemoradiotherapy was established as mainstay for locoregionally advanced NPC [26,27]. Numerous studies have since then explored the bene t of adding induction chemotherapy. Regimens such as PF [18,28], TP [21,29], PX [30], and others [31,32] have been investigated as IC protocols. While improved failure-free survival was reported in these phase II or phase III trials, whether there was a bene t in overall survival was debated [5,6,33,34]. Clear bene t in both failure-free survival and overall survival was obtained in two recent phase III randomized trials that evaluated the effectiveness and safety of IC with TPF [8,9] or GP [10,35] toxicities than TPF [37]. Another study showed similar results [25]. These data are in accordance with our ndings. Speci cally, we found that only induction TPF and GP improved both patients' progression-free survival and overall survival as compared with CCRT alone. However, other regimens seemed not resulted in clear improvement in patient survival. Although GP showed tendency to perform better than TPF, this effect was not statistically different. There was no difference in treatment-related toxicities between these groups.
Nevertheless, up to date, few prospective randomized trials that directly compare different IC regimens are reported. Given differed duration of hospitalization, requirement of central vena catheterization, treatment-related toxicities, patient compliance, potentially differed quality of life, and differed medical expense, there is a realistic demand in determining which IC regimen is an optimal one for locoregional advanced NPC. Currently, a number of trials are still ongoing, including TPC versus PF (NCT02940925), GP versus TPF (NCT02016417, NCT03723343), GP versus PF (NCT03840421), and others. These trials might give critical information on which IC regimen would be an optimal one.
Recently, immunotherapy and in particular immune checkpoint inhibitor treatment, has achieved evolutionally success in a number of solid cancers [38]. In recurrent and metastatic nasopharyngeal carcinoma, immune checkpoint inhibitor combined with conventional chemotherapy led to signi cant improvement in tumor response and patient survival [39][40][41]. It is expected that immunotherapy might also show advantage in LANPC.
Immunotherapy might increase the response rate of IC and might be active even as single treatment. Two single-armed phase II clinical trial that test the role of induction anti-PD-1 antibodies (Sintilimab, NCT03619824; Toripalimab, NCT04446663) with chemotherapy in LANPC is ongoing. These studies, in addition to the recent advance in induction chemotherapy in LANPC, might bring more novel choices for clinical practice.
There were some limitations in our study. With limited prospective studies in this area, our results were obtained from relatively small number of participants, which may lead to section bias and reduce its generalizability. Although we made indirect comparison among different IC regimens and between GP and TPF using network meta-analysis, these results still could not provide direct evidence on which would be the optimal IC regimen.
Therefore, while our study could bring several valuable clues for clinical practice, more prospective studies are needed.
In conclusion, our study indicated that both GP and TPF are favorable IC regimens in LANPC in terms of clinical bene t and safety pro les based on current evidence. Direct comparison between these two regimens in a prospective study might better help determine which one is optimal.