Induction Chemotherapy plus Concurrent Chemoradiotherapy in Endemic Nasopharyngeal Carcinoma: Individual Patient Data Pooled Analysis of Four Randomized Trials

This article is featured in Highlights of This Issue, p. 1771

Nasopharyngeal carcinoma (NPC) is a squamous cell carcinoma with a unique, unbalanced endemic distribution, with an especially high prevalence in east and southeast Asia (1). Nonkeratinizing [World Health Organization (WHO) types II and III] constitutes most cases of NPC in endemic areas (>95%), whereas keratinizing disease (type I) is more common in other regions of the world (1, 2). Unlike other head and neck cancers, radiotherapy is the primary treatment modality for nondisseminated NPC due to its anatomic location and radiosensitivity. The introduction of intensity-modulated radiotherapy (IMRT) substantially improved locoregional control (LRC), and distant metastasis is now the major pattern of treatment failure in locoregionally advanced NPC (3).

In recent decades, numerous trials have investigated the value of adding chemotherapy to radiotherapy in NPC. Concurrent chemoradiotherapy (CCRT) is now the standard treatment for locoregionally advanced disease (1, 4); the updated Meta-Analysis of Chemotherapy in Nasopharynx Carcinoma (MAC-NPC) demonstrated CCRT with or without adjuvant chemotherapy (AC) was related to a 5-year survival benefit of 5% to 12% (5). However, the efficacy of AC after CCRT remains uncertain, and our latest phase III trial observed no significant benefit for CCRT + AC compared with CCRT alone (6). Moreover, the toxic effects and the low rate of compliance to AC also need to be taken into account (4). Compared with AC, induction chemotherapy (IC) offers the advantages of better tolerability and early eradication of micrometastases (7); thus, sequential IC followed by CCRT may represent a promising strategy for NPC in the IMRT era.

Although four reported randomized controlled trials (RCT) have compared IC + CCRT versus CCRT alone in the endemic area, their conclusions remain controversial. A trial at the Prince of Wales Hospital (PWH, New South Wales, Australia), Hong Kong, compared induction docetaxel and cisplatin, followed by CCRT with CCRT alone in stage III–IVB NPC, and found IC significantly improved 3-year overall survival (OS) and had a nonsignificant effect on progression-free survival (PFS; ref. 8). The National Cancer Centre Singapore (NCCS) trial adopted gemcitabine, carboplatin, and paclitaxel (GCP) IC, and no significant differences in survival or distant control (DC) were observed between the IC + CCRT and CCRT alone arms (9). Recently, two multicenter phase III trials from Guangzhou have been reported: the GZ2008 trial found cisplatin and fluorouracil IC significantly improved disease-free survival, with a marginally significant effect on DC (10), whereas the GZ2011 trial found TPF IC improved failure-free survival, OS, and DC in locoregionally advanced NPC (11). Recent meta-analyses by the MAC-NPC Collaborative Group did not show IC + CCRT had obvious superiority over CCRT alone (5, 12). However, these studies included trials reported before 2013; thus, they could not confirm the value of IC + CCRT considering new RCTs reported in recent years.

Therefore, these controversial results highlight the need to determine whether the addition of IC to CCRT provides any additional benefit in NPC. Given the paucity of studies and potentially insufficient power to detect small improvements in the specific endpoints of each trial, the investigators of the four endemic trials launched an individual patient data (IPD) pooled analysis. We aimed to better determine the efficacy, compliance rates, and toxicity, as well as treatment interactions within specific patient subgroups, of adding IC to CCRT in locoregionally advanced NPC. To our knowledge, this is the first IPD analysis to provide a comprehensive overview of the precise role of additional IC in NPC.

We searched PubMed and EMBASE for RCTs in NPC using “nasopharyngeal neoplasms,” “nasopharyngeal tumors,” and “nasopharyngeal cancers” as keywords and “clinical trials” or “randomized controlled trial” as limits; the search was supplemented by searching the bibliographies of the retrieved articles. The final search date was June 24, 2017. To be eligible, trials had to be from endemic regions as indicated by Chua and colleagues (e.g., east and southeast Asia, etc.; ref. 1), have a randomized design, include patients with nonmetastatic NPC treated with definitive conventional fractionated radiotherapy, and had to compare IC followed by CCRT with CCRT alone.

The selection process is shown in Supplementary Fig. S1. Eventually, four RCTs were found to be eligible for this study: the PWH, NCCS, GZ2008, and GZ2011 trials (8–11). During the search process, we identified two other trials that assessed IC, followed by CCRT: one was conducted in a nonendemic area [the Hellenic Cooperative Oncology Group (HeCOG) trial; ref. 13)], and the other compared IC + CCRT with CCRT + AC (the NPC0501 trial; ref. 14). These two RCTs were included in supplementary analysis to provide a more comprehensive overview of the role of additional IC.

For the eligible RCTs, an updated IPD was established for patient and tumor characteristics, treatment details, date of randomization, failure and death, pattern of failure, cause of death, and adverse events during treatment. The studies were conducted in compliance with the Declaration of Helsinki and local regulatory requirements. Written informed consent was obtained from all patients; this study was approved by the ethics committee or Institutional Review Boards at each institution.

We examined the randomization process and intention-to-treat basis for each trial. Randomization was assessed by checking the methods used and balance between baseline characteristics. Patient follow-up was also compared between treatment groups in each trial (15). The analyses were sent to the investigators for review and validation, and all discrepancies were discussed by the investigators to reach consensus.

For the two RCTs included in the supplementary analysis, we reviewed the associated articles and previous IPD meta-analysis on NPC (5, 13, 14), then summarized the trial characteristics and extracted survival data (obtained directly or using the methods detailed by Parmar and colleagues; ref. 16). We also assessed the quality of these trials with respect to randomization and the intention-to-treat principle.

The primary endpoint was PFS, calculated from date of randomization to locoregional failure, distant failure or death from any cause, whichever occurred first. The secondary endpoints were OS, DC, LRC, and cancer and noncancer deaths. OS was calculated from the date of randomization to the date of death from any cause. DC and LRC were defined as the time from randomization to the occurrence of distant or locoregional failure, respectively. Patients with locoregional failure as a first event were censored for distant failure, and vice versa. If both distant failure and locoregional failure occurred at the same time, patients were considered to have a distant failure event only. Patients without distant and locoregional failure were censored at the date of death or last follow-up if still alive. Persistent primary/nodal disease was classified as locoregional failure. Surviving patients without any event for all endpoints were censored at the date of last follow-up. Deaths attributed to known causes other than NPC for patients with no reported progression were defined as noncancer deaths. All other deaths were defined as cancer deaths, including deaths from NPC, deaths from any cause in patients with previous progression and deaths from unknown causes. This definition has been adopted by the MAC-NPC Collaborative Group (5), as it prevents underestimation of deaths related to cancer and is less biased than other methods (17).

We performed analysis based on the intention-to-treat principle. Survival analyses were stratified by trial; the observed minus the expected number of deaths (O−E) and its variance was used to calculate individual and overall pooled HRs using a fixed effect model (16). Heterogeneity across trials was examined using the χ² test and I² statistic (18). Statistically significant heterogeneity was defined as a χ²P < 0.1 or an I² statistic >50%. If obvious heterogeneity existed, the DerSimonian and Laird random effects model was adopted (19).

Median follow-up was calculated via the reverse Kaplan–Meier method (20). The estimated Kaplan–Meier survival curves for the two treatment groups were plotted (21), and the absolute benefits at 3 and 5 years and 95% confidence intervals (CI) were calculated (17). The interactions between treatment effect and patient subgroups (i.e., sex, age, tumor category, nodal category, clinical stage, and radiotherapy technique) were estimated with an interaction test, by adopting a single Cox model stratified by trial and containing treatment effect, covariate (e.g., sex) effect, and treatment–covariate interaction (one-stage model method; ref. 22). An insignificant P value in the interaction test indicates the effect in the experimental arm versus control arm did not differ among that specific covariate group.

A supplementary analysis of PFS and OS between the experimental and control groups was performed, including both the eligible and supplementary RCTs. Moreover, to compare the potential differences in efficacy between different IC regimens, we performed a supplementary Bayesian network analysis including all six trials. The network analysis within a Bayesian framework using Markov chain Monte Carlo methods was built using the model proposed by Woods and colleagues (23); treatment effects were estimated by posterior means with corresponding 95% credible intervals (CrI; ref. 24). The fixed effects model was adopted as it resulted in lower deviance information criterion (DIC) statistics (DIC provides a measure of model fit that penalizes model complexity, with lower values suggesting a simpler model; 25). We did not assess the probability of treatment ranking as it could be produced even without clear statistical meaning, which is misleading; detailed methods have been described in our previous study (4). All tests were two-sided. Statistical analyses were performed using STATA version 12.0 (Stata Corporation) and WinBUGS 1.4.3 (MRC Biostatistics Unit).

Table 1 summarizes the key features of the PWH, NCCS, GZ2008, and GZ2011 trials. With respect to the intention-to-treat principle, all randomized patients in all four trials were analyzed; data were not available for only 8 patients in the NCCS trial, who were either found ineligible on retrospective review or withdrew before receiving treatment. Overall, a total of 1,193 patients were included in the current analysis, with 599 and 594 patients allocated to the IC + CCRT and CCRT arms, respectively. None of the four trials demonstrated unbalanced baseline characteristics between treatment arms. The median follow-up was 5.0 years; no major bias appeared between treatment arms in any trial as indicated by the reverse Kaplan–Meier curves. All trials recruited patients with WHO histologic type II or III NPC. The patient characteristics for the four eligible trials are summarized in Supplementary Table S1 and the two supplementary trials are described in Supplementary Table S2.

The disease status and patterns of failure are summarized in Supplementary Table S3. Overall, 208 of 1,193 (17%) patients died and 311 of 1,193 (26%) patients experienced disease progression. IC + CCRT improved PFS compared with CCRT alone (HR = 0.70; 95% CI, 0.56–0.86; P = 0.0009; from 64.6% to 73.9% at 5 years; Figs. 1A and 2A). IC + CCRT also improved OS, with an HR of death of 0.75 (95% CI; 0.57–0.99; P = 0.04) and 5.5% improvement at 5 years (Figs. 1B and 2B).

IC + CCRT was associated with a significantly lower rate of distant failure than CCRT alone (HR = 0.68; 95% CI, 0.51–0.90; P = 0.008; absolute reduction from 84.8% to 78.3% at 5 years; Figs. 1C and 2C). No significant difference in LRC was observed between IC + CCRT and CCRT alone, although IC + CCRT had a tendency to improve LRC (HR = 0.70; 95% CI, 0.48–1.01; P = 0.06; Figs. 1D and 2D). To confirm whether the results were sensitive to the choice of a fixed effects model as opposed to a random effects model, we also calculated pooled HRs and corresponding 95% CIs using the random effects model. The results were the same as those calculated with the fixed effects model. This is not surprising as the I² statistic (defined as variation in effect size attributable to heterogeneity) was equal to 0% for all endpoints (Fig. 1), which means no variation attributable to heterogeneity was detected, and effect size was not affected by the model used.

In total, 193 (16%) cancer deaths and 15 (1%) noncancer deaths occurred. IC + CCRT tended to reduce the risk of death, with an absolute difference of 4.8% (from 20.1% to 15.3%) at 5 years; there was no obvious difference in noncancer deaths between the two arms (Fig. 3).

Of the 599 patients randomized to the IC + CCRT arm, 578 (96%) received at least one cycle of IC, and 542 (90%) received IC as planned despite relatively lower compliance to the TPF regimen (88%; Supplementary Table S4). In total, 566 of 599 (94%) patients in the IC + CCRT arm and 564 of 594 (95%) in the CCRT arm initiated concurrent chemotherapy. Compliance to concurrent chemotherapy was significantly different between arms: more patients completed at least five/two cycles (if eight/three cycles were planned) of concurrent chemotherapy in the CCRT arm than the IC + CCRT arm (92% vs. 87%). No obvious differences in the numbers of patients starting and completing radiotherapy were observed between the IC + CCRT and CCRT alone arms.

Supplementary Table S5 summarizes the major grade 3–4 adverse events. During IC, the most common toxic effect was neutropenia (35%), followed by leukopenia (20%). Other toxicities were not common, with incidences lower than 10%. During CCRT, adverse events were similar between the two arms, although the IC + CCRT arm experienced significantly higher rates of grade 3 or 4 leukopenia (26% vs. 21%; P = 0.03) and neutropenia (15% vs. 9%; P = 0.003).

No significant heterogeneity was observed between trials for any endpoint. To determine whether the treatment effect of IC + CCRT versus CCRT alone differs among specific covariate groups, we performed subgroup analyses for PFS and OS in patients stratified by the following characteristics (covariates): sex (male, female), age (< 40, 40–59, ≥ 60), tumor category (T1–2, T3–4), nodal category (N0–1, N2–3), clinical stage (III, IV), and radiotherapy technique (2D-RT, IMRT). No interactions between these covariates and treatment were observed (all P > 0.1; Fig. 4), which means the benefit of additional IC did not differ among specific populations. However, considering the oldest subgroup had relatively few patients, large-scale studies are warranted to assess the benefit of IC in older patients.

When the two supplementary trials (13, 14) were added to the pooled analysis, the conclusions remained valid: compared with CCRT ± AC, IC + CCRT improved both PFS (HR = 0.72; 95% CI, 0.60–0.85; P = 0.0002) and OS (HR = 0.77; 95% CI, 0.62–0.97; P = 0.02) in locoregionally advanced NPC (Supplementary Fig. S2). Supplementary Bayesian network analysis was performed to help identify potential differences in the efficacy of different IC regimens. Supplementary Figure S3 shows the network established for PFS and OS. Figure 5 summarizes the results of multiple treatment comparisons; probably due to the lack of relevant trials, no significant differences were detected between the different IC regimens. Still, it should be noted that compared with no IC, only TPF regimen significantly improved both PFS (HR = 0. 70; 95% CrIs, 0.49–0.95) and OS (HR = 0.59; 95% CrI, 0.37–0.92), suggesting TPF IC may be more effective.

This IPD pooled analysis of patients from endemic regions shows the benefits associated with the use of IC followed by CCRT in locoregionally advanced NPC, including significant improvements in PFS and OS and a reduction in distant failure. The absence of interactions between patient characteristics and treatment effects indicates the benefit of additional IC does not differ among specific populations. The efficacy of IC + CCRT remained valid and was even strengthened after adding the supplementary trials, further supporting the conclusions.

Recently, the MAC-NPC Collaborative Group reported no significant differences between IC + CCRT and CCRT alone with respect to treatment outcomes in NPC in a network meta-analysis, although additional IC tended to improve DC (12). However, this study included trials reported before 2013 and did not include trials using new IC regimens (e.g., GCP, TPF). Considering the publication of new trials in recent years, we conducted this pooled analysis to confirm the role of IC followed by CCRT. We primarily focused on trials conducted in endemic regions. In the original trial reports, the PWH trial found additional IC significantly increased 3-year OS from 68% to 94%, but failed to detect a significant improvement in PFS (despite an obvious trend; ref. 8). The NCCS trial observed a small but insignificant increase in 3-year OS, disease-free survival, and distant metastasis-free survival (the absolute differences were 2%, 8%, and 4%, respectively; ref. 9). The multicenter GZ2008 trial observed an 8% improvement in 3-year disease-free survival and a marginally significant increase in distant metastasis-free survival (4% improvement), but no early OS benefit (10). The GZ2011 trial detected 8% and 6% improvements in 3-year failure-free survival and OS after IC (both P = 0.03), respectively (11). Combination of IPD from these four trials enabled a more precise and comprehensive estimation of the magnitude of the benefits provided by IC + CCRT compared with CCRT alone.

The results of this study demonstrate the superiority of additional IC over CCRT alone in terms of PFS and OS, with the survival benefit mainly associated with improved DC. Considering the widespread use of IMRT, it is not surprising that no significant improvement in LRC was observed. To further validate the conclusions, we included two additional trials (HeCOG and NPC0501 trials) in the supplementary analysis. The superior efficacy of IC + CCRT versus CCRT alone remained unchanged, and no heterogeneity was observed. Considering the varied regimens used for IC, these conclusions should be interpreted with caution; to provide more information on the efficacy of different IC regimens, we further conducted a supplementary Bayesian network analysis. Although no statistically significant differences in PFS or OS between different IC regimens were detected, the results may favor TPF over other regimens. The efficacy of adding docetaxel to the PF induction regimen has been demonstrated in locally advanced head and neck cancers via large-scale phase III trials and IPD pooled analysis (26–28). The unobvious superiority of one IC regimen over others may be result of a lack of trials directly comparing different IC regimens. Although TPF has the potential to be a better choice in IC for NPC, it is still of great significance to identify the optimal IC regimen. The gemcitabine-based IC regimen may also be effective, as it provides a significant advantage in advanced NPC (29). Several trials (NCT01872962 and NCT02512315) are being undertaken to assess other IC regimens, such as gemcitabine plus cisplatin, and these data should be publicly available in the near future. Moreover, further trials are required to directly compare different IC regimens.

Compliance to IC was satisfactory; about 90% of patients completed their planned cycles despite the relatively lower compliance to TPF (88%). During CCRT, relatively fewer patients in the IC + CCRT arm completed more than half of their planned cycles of concurrent cisplatin compared with the CCRT alone arm (87% vs. 92%), probably due to patient refusal and treatment toxicities (11). Overall, compliance to IC and CCRT was satisfactory compared with that of locally advanced head and neck cancers (about 80% of patients received IC as planned, with about 40% being able to receive concomitant chemotherapy, and about 70% starting planned radiotherapy), which could promote the clinical use of additional IC in NPC (28). The major grade 3 and 4 toxicities in the IC + CCRT arm were leukopenia (26%), mucositis (24%), neutropenia (15%), vomiting (12%), and nausea (11%), which were manageable and reversible; with the exception of leukopenia and neutropenia, adverse events were similar between arms. Moreover, no obvious differences in deaths due to toxicities were observed between groups, further reflecting the acceptable toxicity of additional IC.

The strengths of this pooled analysis are related to its size and its use of the IPD that allowed reanalysis of each trial (e.g., standardized multiple endpoints). Nevertheless, we should state the limitations of this analysis. First, these trials included different IC regimens. However, all regimens were platinum based, and no heterogeneity was observed for any endpoints. Second, we should be cautious when interpreting OS benefits. For some of the trials included, the failure to achieve a significant increase in OS may be explained by a high salvage rate, which dilutes the OS benefit, or relatively small sample sizes. With increased statistical power, our IPD analysis could help to detect a significant improvement in OS. Still, whether additional IC could provide a long-term OS benefit needs to be explored further when more trials with long-term follow-up results are available. However, the MAC-NPC Collaborative Group recently demonstrated that PFS and DC were valid surrogate endpoints for OS (30). Thus, considering the significant improvements in PFS and DC observed for IC + CCRT, additional IC may also provide a long-term survival benefit. Third, considering the uneven geographical distribution of NPC and the small number of RCTs available, we also included the HeCOG trial (from a nonendemic area) and NPC0501 trial (which also included adjuvant chemotherapy) in the supplementary analyses, which may cause potential bias. Still, the conclusions remained valid after including these supplementary trials, with no heterogeneity detected. The supplementary analyses help to provide a more comprehensive overview of the value of IC + CCRT in NPC.

The National Comprehensive Cancer Network (NCCN) currently recommends CCRT + AC for locoregionally advanced NPC (Category 2A); CCRT alone is also an option (Category 2B; ref. 31). IC + CCRT is recommended as a Category 3 option. According to the European Society for Medical Oncology (ESMO), platinum-based IC can be considered for locally advanced disease, but in no case should it negatively affect the administration of CCRT (32). This updated IPD pooled analysis provides important information to clarify the precise value of IC + CCRT in NPC. Our study highlights the importance of identifying high-risk patient groups that may benefit most from IC + CCRT; certain biomarkers such as plasma Epstein–Barr virus DNA load may be helpful for participant selection (NCT02135042). In the future, when data from other trials are available, the MAC-NPC Collaborative Group could provide a more comprehensive overview to help better understand the optimal treatment modality for NPC.

In conclusion, this IPD pooled analysis indicates the benefits associated with the addition of IC to CCRT in locoregionally advanced NPC; the precise value of IC in specific patient subgroups and the optimal regimens still need further assessment.

B.B.Y. Ma reports receiving commercial research grants from Novartis, reports receiving speakers bureau honoraria from MSD and Merck Serono, and is a consultant/advisory board member for Boehringer Ingelheim, Bristol-Myers Squibb, MSD, and Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y.-P. Chen, L.-L. Tang, B.B.Y. Ma, J. Ma

Development of methodology: Y.-P. Chen, J. Ma

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Q. Yang, S.-S. Poh, E.P. Hui, T. Tan, J. Wee, L. Chen, M. Tong, S.-L. Cheah, K.-W. Fong, K. Sommat, Y.L. Soong, Y. Sun, M.-H. Hong, S.-M. Cao, M.-Y. Chen, J. Ma

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y.-P. Chen, L.-L. Tang, Q. Yang, W.-S. Ong, T. Tan, W.-F. Li, S.-H. Tan, K. Sommat, Y. Guo, A.-H. Lin, Y. Sun, M.-Y. Chen, J. Ma

Writing, review, and/or revision of the manuscript: Y.-P. Chen, L.-L. Tang, Q. Yang, S.-S. Poh, E.P. Hui, A.T.C. Chan, W.-S. Ong, T. Tan, J. Wee, W.-F. Li, S.-H. Tan, S.-L. Cheah, K. Sommat, Y.L. Soong, Y. Guo, A.-H. Lin, Y. Sun, M.-H. Hong, S.-M. Cao, M.-Y. Chen, J. Ma

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.-S. Poh, Y. Sun, J. Ma

Study supervision: Y. Guo, A.-H. Lin, M.-H. Hong, J. Ma

This work was supported by grants from the National Science & Technology Pillar Program during the Twelfth Five-year Plan Period (2014BAI09B10), the Natural Science Foundation of Guang Dong Province (2017A030312003), the National Key R&D Program of China7 (2016YFC0902000), the National Natural Science Foundation of China (81572658), the Innovation Team Development Plan of the Ministry of Education (No. IR_17R110), and the Overseas Expertise Introduction Project for Discipline Innovation (111 Project, B14035). The authors thank the anonymous reviewers and editors for their insightful comments and great efforts to improve this manuscript. They also thank the Clinical Trials Centre, Sun Yat-sen University Cancer Centre, for assistance with data interpretation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.