A Bayesian network meta-analysis of the primary definitive therapies for locoregionally advanced nasopharyngeal carcinoma: IC+CCRT, CCRT+AC, and CCRT alone

Background The major aim of this Bayesian network analysis was to determine the optimal treatment strategy for locoregionally advanced nasopharyngeal carcinoma (LANPC). Method We systematically searched databases and extracted data from randomized clinical trials involving LANPC patients randomly assigned to receive induction chemotherapy followed by concurrent chemoradiotherapy (IC+CCRT), CCRT followed by adjuvant chemotherapy (CCRT+AC), or CCRT. Results In the network analysis, IC+CCRT was significantly better than CCRT alone for 5-year FFS (odds ratio [OR]: 1.63, 95% credible interval [CrI] 1.16–2.29), DMFS (OR: 1.56, 95% CrI 1.08–2.22), and LFRS (OR: 1.62, 95% CrI 1.02–2.59), but not OS (OR: 1.35, 95% CrI 0.92–2.00). Rank probabilities showed that IC+CCRT was ranked the best followed by CCRT+AC and CCRT for all 5-year outcomes. Although compared to IC+CCRT and CCRT, CCRT+AC did not significantly improve survival but had the highest 5-year survival rates. Conclusions IC+CCRT could be recommended as a front-preferred primary definitive therapy for patients with LANPC.

Since 2013, the category of evidence supporting CCRT for LANPC has been cut to 2B [6]. At the same time, CCRT followed by adjuvant chemotherapy (CCRT+AC) was given a 2A recommendation [6]. However, the cited studies do not fully support the use of CCRT+AC over CCRT, as the two trials demonstrated the superiority of CCRT+AC compared to RT but not CCRT [1,7]. Additionally, another two prospective clinical trials failed to show that the addition of AC to CCRT improved survival outcomes [8][9][10].
According to the 2020 NCCN guideline, IC+CCRT, CCRT+AC, and CCRT alone are all the primary definitive therapies for LANPC. Nevertheless, the cited studies recommending the application of CCRT+AC have not been updated. Since the absence of a randomized trial directly comparing IC+CCRT to CCRT+AC and to explore the optimal therapeutic strategy for LANPC, we conducted this Bayesian network analysis to comprehensively compare the efficacies of IC+CCRT, CCRT+AC, and CCRT.

Methods
This study was conducted based on the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) extension statement for network meta-analysis [16].

Search strategy
A systematic search of articles was conducted on PubMed, EMBASE, Web of Science, and Cochrane Library for clinical trials comparing at least any two of the three types of treatments on Nov 1, 2020. All the identified trials and relevant reviews were identified through reference lists to ensure completeness. The search terms in PubMed, EMBASE, and Web of Science online databases were "induction OR neoadjuvant OR adjuvant" and "concurrent OR concomitant" and "cisplatin" and "chemotherapy OR chemoradiotherapy OR radiotherapy" and "nasopharyngeal" and "carcinoma OR cancer OR tumor" and "study OR trial", while the search terms used in searching Cochrane Library included: "induction OR neoadjuvant OR adjuvant" and "concurrent OR concomitant" and "cisplatin" and "chemotherapy OR chemoradiotherapy OR radiotherapy" and "nasopharyngeal carcinoma OR nasopharyngeal cancer OR nasopharyngeal tumor".

Selection criteria
Eligible trials were requested to satisfy the following "PICO" inclusion criteria: (P) patients were newly diagnosed with LANPC; (I+C) LANPC patients randomly received at least two of the three treatment strategies, including IC+CCRT, CCRT+AC, and CCRT; (O) full-text articles and data of survival rates were available. Additional criteria comprised: (1) CCRT was cisplatin-based conventional concomitant chemoradiotherapy; (2) trials should be officially registered prospective phase II-IV clinical studies; (3) target therapy was prohibited during the whole care process; (4) published language was English. Any discrepancies were resolved by discussion.

Outcomes
The primary endpoints were the rates of 5-year overall survival (OS) and failure-free survival (FFS). The secondary endpoints were 5-year rates of distant metastasis-free survival (DMFS) and locoregional recurrence-free survival (LRFS).

Data extraction and risk of bias assessment
Bi-Cheng Wang and Zhan-Jie Zhang independently extracted the survival rates of all outcomes and the data of basic study characteristics and treatment modalities. All raw data were directly collected from the eligible clinical trials without further modification or adjustment. Review Manager 5.3 software (Cochrane Collaboration's Information Management System) was used to assess the risk of bias.

Data synthesis and statistical analysis
Bayesian network meta-analyses were built using the Bayesian Markov Chain Monte-Carlo (MCMC) sampling in WinBUGS or GeMTC. The random-effects model was proposed. The effect size was estimated with odds ratios (ORs) and 95% credible intervals (CrIs) using the available raw data extracted from the trials. ORs were computed on averages of the 100,000 iterations (four chains with a thinning interval of 10) after a training phase of 50,000 burn-ins. Node-splitting approach was performed to evaluate if there was inconsistency in the closedloop. Pairwise meta-analyses were applied to evaluate incoherence between direct and indirect evidence. The probability of each treatment being the best among the three therapies was estimated by using the distribution of the ranking probabilities.
Pooled OS, FFS, DMFS, and LRFS rates and 95% CIs were analyzed by using STATA 14.0 software. The heterogeneity between rates was tested by I 2 statistic percentages. Whether the heterogeneity was high (I 2 > 50%) or low (I 2 � 50%), random effects were performed to reduce the bias.
Although all eligible trials were well-randomized studies, the methodological quality of this analysis was at moderate risk for reporting bias owing to the open-label design (S2 Fig).

Discussion
Our study fully collected the prospective registered and randomized clinical trials and comprehensively analyzed the efficacies of IC+CCRT, CCRT+AC, and CCRT on LANPC. Based on our Bayesian network analysis, IC+CCRT ranked the best for all 5-year survival outcomes. In addition, CCRT+AC showed the highest response rates in the pooled analyses.

IC+CCRT vs CCRT
Multiple phase III clinical trials have confirmed the critical role of IC+CCRT in treating patients with LANPC. Compared with CCRT alone, IC+CCRT had better HRs in terms of OS, FFS, DMFS, and LRFS [13].
In Zhang's study [21] published in 2019, gemcitabine plus cisplatin as IC combined with CCRT increased the 3-year OS rate to 94.6% versus 90.3% in CCRT group. IC could be well tolerated as 96.7% of the patients in IC+CCRT group completed three cycles of IC. In the newest NCCN guideline [15], gemcitabine and cisplatin regimen was recommended as a category 1 treatment for LANPC patients given IC+CCRT. In addition to this regimen, taxane (including docetaxel, paclitaxel, and nab-paclitaxel) plus cisplatin strategy is used in our hospital. A propensity-score matching analysis indicated that there were no significant differences in clinical outcomes and safety profiles between docetaxel plus cisplatin and gemcitabine plus cisplatin [26]. Therefore, we suggest taxane or gemcitabine plus cisplatin as a front IC option for LANPC patients.

CCRT+AC vs CCRT
CCRT+AC failed to show the superiority compared to IC+CCRT or CCRT alone in the Bayesian analysis. However, the survival rates in patients received CCRT+AC were numerically highest. Since most data of CCRT+AC were contributed by Chen's studies [9,10], that is the reason why CCRT+AC did not have higher ORs than the other two therapies. Although there were no significant differences between the groups in Chen's clinical trial, the survival rates in CCRT+AC group were numerically higher than those in CCRT group. Kim and colleagues retrospectively detected the benefit of addition AC to CCRT and found that CCRT+AC showed higher 3-year OS (86% vs 80%, p = 0.894) and FFS (75% vs 66%, p = 0.018) rates against CCRT [28]. However, another retrospective study determined that patients with LANPC might not receive significant survival benefits from adding AC to CCRT (OS HR 0.77, 95% CI 0.37-1.57; FFS HR 1.26, 95% CI 0.69-2.28) [29]. Combined with our Bayesian network analysis, to date, there is still a lack of evidence to confirm the superiority of CCRT+AC compared to CCRT.
In 2021, Hui et al. detected the prediction function of plasma Epstein barr virus (EBV) deoxyribonucleic acid (DNA) and found that the patients with detectable post-radiotherapy plasma EBV DNA who experienced subsequent plasma EBV DNA clearance might be the potential target population of AC [30]. In further explorations, the ongoing phase II and III study, NRG-HN001, attempt to risk-stratify AC using post-radiotherapy plasma biomarker EBV DNA levels. The future results of NRG-HN001 might help us determine the suitable population and whether omitting AC will result in non-inferior survivals compared to patients treated with CCRT+AC.

IC+CCRT vs CCRT+AC
Large-scale phase III clinical trials that directly compare IC+CCRT with CCRT+AC are limited. NPC-0501 was the only prospective randomized study we identified [25]. Compared with cisplatin plus 5-fluorouracil AC arm (75%), cisplatin plus 5-fluorouracil IC arm had a similar 3-year FFS rate (79%), while cisplatin plus capecitabine arm showed a significantly higher 3-year FFS rate (81%). Therefore, different chemotherapeutic regimens might have impacts on responses. Moreover, NPC-0501 indicated that accelerated fractionation radiotherapy did not achieve more benefits but incurred higher toxicities against conventional fractionation radiotherapy. According to the retrospective results reported by Setakornnukul and colleagues, IC+CCRT was not superior to CCRT+AC (adjusted OS 3-year: 84% vs 81%; 5-year: 74% vs 70%) [31].
Based on the results of published studies and our analysis, IC+CCRT and CCRT+AC display comparable effects on LANPC patients. However, we would recommend IC+CCRT based on the improved probability of being the preferred regimen in all 5-year Bayesian models.

Limitations
(1) all enrolled clinical trials were open-label studies, which might increase the bias of evaluation; (2) only two trials provided the data for AC, leading to weak evidence for supporting the accurate efficacy of CCRT+AC on LANPC patients; (3) weekly and triweekly cisplatin regimens were conducted during CCRT, however, there is still a debate on the response rates of between the two treatment modalities (week NPC); (4) Analyses of individual patient data are lacking.

Conclusions
Among the three standard therapeutic strategies, IC+CCRT can be recommended as the preferred choice for previously untreated LANPC patients. Even if current studies suggest CCRT +AC is not inferior to IC+CCRT, evidence comparing CCRT+AC to the other two strategies remains limited.