Main

On a global scale, nasopharyngeal carcinoma (NPC) causes ∼65 000 deaths annually, although its incidence varies widely by region (Ferlay et al, 2010). While rare in North America and Europe, the incidence of NPC exceeds 20 cases per 100 000 people in endemic regions, such as Southern China, Southeast Asia and the Middle East/North Africa (Chang and Adami, 2006). Different factors are thought to contribute to the pathogenesis of cases in endemic and non-endemic regions. Epstein−Barr virus (EBV) infection, environmental factors and genetic predisposition are proposed to be the main contributing factors in endemic regions, while the classic risk factors for other head and neck tumors such as smoking, alcohol and human papilloma virus (HPV) infection are thought to account more for cases in non-endemic areas (Vaughan et al, 1996; Chua et al, 2016).

Approximately 5−11% of patients present with de novo metastatic disease, while a further 15−30% of patients who were treated for locally advanced NPC will develop local recurrence and/or distant metastatic disease (Lee et al, 2015). Most recurrent cases are not amenable to salvage therapy with surgery and/or radiotherapy with or without concurrent chemotherapy. Hence, treatment options for the majority of patients with recurrent or metastatic NPC (RM-NPC) are largely limited to palliative systemic therapies (Lee et al, 2015).

NPC is a chemosensitive disease with some studies reporting response rates of over 80% with platinum-based chemotherapy regimens in recurrent or metastatic settings (Leong et al, 2008; Chen et al, 2013). Durable responses and prolonged survival have been observed in a subset of patients (de Graeff et al, 1987; Siu et al, 1998; Taamma et al, 1999). Therefore, despite the lack of direct comparison to supportive care, systemic chemotherapy is presently the mainstay of treatment for patients with RM-NPC. Due to the relatively low disease incidence, clinical trials investigating systemic therapies for RM-NPC typically enroll a heterogeneous patient population in a single treatment arm, which makes modelling of historical patient outcomes difficult (Lee et al, 2015).

This systematic review qualitatively examined studies of systemic therapies in RM-NPC to determine the relative efficacies of available drug regimens, and provide estimates of aggregated progression-free survival (PFS) and overall survival (OS); In particular, we seek to compare platinum to non-platinum regimens and investigate aggregate survival outcomes in the second line and beyond treatment setting. Trends over time in overall response rates (ORR), PFS and OS were also investigated. The Downs and Blacks (DB) checklist was chosen to assess the methodological quality of identified clinical trials as it allows assessment of both randomised and non-randomised studies (Downs and Black, 1998). Analysis of the association between clinical trial outcomes (ORR, PFS and OS) and characteristics of the study, including region where it was conducted, number of patients involved, study DB score and the year the trial was performed.

Materials and methods

Search strategy

A systematic search of MEDLINE (from 1950), EMBASE (from 1980), and the Cochrane Library was conducted to March 2015. Clinical trials investigating cytotoxic and molecularly targeted agents in adult patients with RM-NPC were included. The search was limited to human trials and studies published in the English language. Dose escalation or phase I clinical trials, immune therapies, concomitant and/or sequential use of radiotherapy or surgery, curative-intent systemic therapies, trials of patients with non-NPC histology where patient outcomes were not reported separately, conference proceedings and abstracts were excluded. The same search inclusion and exclusion criteria were applied to studies in models A and B; however, only studies with analysable Kaplan−Meier curves were used in model B analysis.

All search results were initially screened for relevance on the basis of article titles and abstracts by two authors (AP and AH). Full text articles were then retrieved for shortlisted studies. Additional trials were identified through manual searches of reference lists from the included articles. Reviewers were not blinded to study authors or outcomes. The decision to include a study for review was made by consensus between two authors (AP and AH). Disagreements would be resolved by a third author, but there were no unresolved differences.

Data extraction

Data were extracted by two teams of two reviewers (AP with AH and SO with TC). The following data were extracted for each trial: (1) study characteristics including clinical phase, region conducted and number of patients involved; (2) reported outcomes including ORR, PFS and OS; (3) grade 3 and 4 toxicities affecting at least 25% of study population; (4) patient characteristics including age, gender, prior treatment in the metastatic setting; and (5) details of the systemic therapy regimen under investigation.

Data analysis

All relevant studies were assessed for quality using the DB checklist (Downs and Black, 1998), which provided an overall score indicating reporting quality, internal validity (bias and confounding), power and external validity (Downs and Black, 1998). Aggregated data analysis and Student’s t-tests were performed for all identified studies (model A). Two-sided P-values were calculated with a cutoff of 0·05 for significance. For studies that published Kaplan−Meier (KM) curves with sufficient resolution and details regarding censoring, survival data were extracted from the KM curves using electronic software, DigitizeIt, Germany, and individual patient-level data were recreated (Guyot et al, 2012). Survival analysis with marginal proportional hazard model methods was used to account for the clustering of patients within trials (model B) (Lin, 1994).

Response rate analyses were performed for all identified studies in model A. To account for clustering of data within studies, a random intercept model was used in model B (Raudenbush and Bryk, 2002).

Fisher’s Exact test was used to examine the difference in rates of occurrence of grades 3 and 4 haematological toxicities of the treatment regimens. Spearman correlation was used to analyse the relationship between study outcomes (ORR, PFS and OS) and the region where it was conducted, number of patients involved, study DB score and the year the trial was conducted.

All analyses were conducted in SAS (version 9·4, Cary, NC, USA). P-values of <0·05 were considered statistically significant.

Role of the funding source

There is no specific funding source for this study.

Results

Systematic search

Systematic search results are summarised in Figure 1. A total of 56 articles were included in the final analysis of this review (summarised in Table 1).

Figure 1
figure 1

PRISMA flow diagram.

Full size image
Table 1 Study summary
Full size table

Assessment of study quality

All 56 studies were included in model A analysis and 26 trials had analysable KM curves for inclusion in model B. The mean DB score for all studies was 15.3 (range 8–25). The 26 studies included in model B analysis had significantly higher mean DB scores than the other 30 studies (17.3 (95% CI 15.5–19.0) vs 13.7 (95% CI 12.2–15.1), P=0.002).

Survival analyses

For RM-NPC patients undergoing systemic therapy (chemotherapy and molecularly targeted agents) in the first-line setting, the estimated median OS by model A was 15.7 months (95% CI, 12.3–19.1), while model B estimated this at 19.3 months (95% CI, 17.6–21.1). For patients undergoing second line or higher therapies (2nd+), the estimated median OS by model A was 11.5 months (95% CI 10.1–12.9), while model B estimated this at 12.5 months (95% CI 11.9–13.4). Aggregate KM curves for OS from model B analysis are presented in the Supplementary Appendix Figures S1 and S2. PFS estimates for patients undergoing first-line chemotherapy by model A was 7.6 months (95% CI, 6.2–9.0), while model B estimated this at 8.0 months (95% CI, 7.6–8.8). For patients undergoing therapy in the 2nd+ setting, the estimated PFS by model A was 5.4 months (95% CI, 3.8–7.0), which was closely approximated by the model B analysis at 5.2 months (95% CI, 4.7–5.6). Aggregate KM curves for PFS from model B analysis are presented in the Supplementary Appendix Figures S3 and S4.

Combination chemotherapy regimens

The use of combination therapy in the first-line setting produced a statistically significant PFS improvement over single agent using both statistical models. Estimated PFS by model A was 8.4 months (95% CI, 6.9–9.8) for patients treated with combination therapy regimens in the first line setting, in comparison to 3.5 months (95% CI, 1.1–5.9, P=0.007) for monotherapy. Hazard ratio is estimated at 0.48 (95% CI 0.41–0.56, P=<0.0001) using model B analysis, in favor of combination therapy regimens. The OS outcomes were different between the two models as outlined in Table 2a, 2b. Model A showed a statistically significant improvement with use of combination therapy, but this was not corroborated in model B analysis.

Table 2a Survival analyses for use of combination vs single agent regimens in the first-line settings
Full size table
Table 2b Response rate analyses for use of combination vs single agent regimens in the first-line settings
Full size table

Studies using combination chemotherapy regimens were more likely to report the occurrence of grades 3 and 4 haematological toxicities affecting at least 25% of all the patients enrolled in the study (56.7 vs 9.1%, P=0.011). Fifteen studies were excluded from this analysis due to missing toxicity reporting.

Platinum-containing regimens

Model A analysis supported the use of combination platinum-containing regimens in the first line setting with an estimated PFS of 8.3 months (95% CI, 7.0–9.6), in comparison to 3.5 months for non-platinum-containing therapies (95% CI, 1.1–5·9, P=0.007). Hazard ratio for the two treatment strategies was estimated at 0.48 by model B analysis (95% CI, 0.41–0.56, P=<0.0001), in favor of the platinum-containing regimens. The two statistical models again differed in their OS analyses, where model A showed statistically significant improvement with the use of first-line platinum-containing chemotherapy over non-platinum regimens, while the difference was not significant for model B. Statistical results and study distributions are summarised in Table 3a, b.

Table 3a Survival analyses for use of platinum-based vs non-platinum based regimens in the first line settings
Full size table
Table 3b Response rate analyses for use of platinum-based vs non-platinum based regimens in the first line settings
Full size table

Studies using platinum-containing regimens are more likely to report the occurrence of grades 3 and 4 haematological toxicities affecting at least 25% of all the patients enrolled in the study (55.2 vs 12.5%, P=0.010). Eleven studies were excluded from this analysis due to missing toxicity reporting.

Response rate analysis

Model A analysis of mean and weighted ORR supports the use of combination chemotherapy in the first-line setting. The improvement in ORR with the use of platinum-based regimens for treatment-naive patients; however, was not found to be statistically significant when comparing to non-platinum based regimens (Tables 2a, b and 3a, b). ORR estimations by model B further supported the results from model A.

Use of targeted agents

Four clinical trials investigating three molecularly targeted agents (MTAs) were identified in this systematic review. MTAs investigated include pazopanib, gefitinib and cetuximab (Chan et al, 2005; Chua et al, 2008; Ma et al, 2008; Lim et al, 2011). One of the studies investigated cetuximab in combination with carboplatin (Chan et al, 2005), while the other studies investigated the pazopanib and gefitinib in three separate studies as monotherapy (Chua et al, 2008; Ma et al, 2008; Lim et al, 2011). In three of the studies that investigated MTAs in the 2nd+ settings, the median PFS was lower than that estimated from all studies using both models A and B analyses. With the exception of one study investigating gefitinib, the median OS for MTAs in the 2nd+ setting was also lower than that estimated from all studies using both models A and B analyses (Chua et al, 2008; Ma et al, 2008; Lim et al, 2011). Currently, no MTA has been approved for treatment of RM-NPC (Lee et al, 2015).

Analysis of outcome generalisability

Wilcoxon signed-rank test of PFS comparing studies conducted in Asian vs non-Asian populations showed statistically significant differences in PFS, with higher values in non-Asian populations (P=0.02). However, there was no statistically significant difference in ORR or OS outcomes (P=0.54 and P=0.53, respectively).

Spearman correlation analyses did not demonstrate any significant relationship between ORR or PFS and the year the study was published, number of patients enrolled, or DB score. However, there is moderate correlation between OS and the year of publication (Spearman correlation coefficient=0·43). See Table 4 for Spearman correlation coefficients.

Table 4 Spearman correlation coefficients
Full size table

Discussion

The first randomised phase III clinical trial in RM-NPC compared two platinum-based combination chemotherapy regimens in the first-line setting (Zhang et al, 2016). The reported PFS with gemcitabine and cisplatin (GP) was seven months compared with 5.6 months for 5-fluorouracil and cisplatin (FP). PFS outcome from the superior GP regimen is comparable to the aggregate PFS presented in our systematic review (7.6 months by model A, and 8 months by model B analyses). Compared to the aggregate OS however (15.7 months by model A, and 19.3 months by model B analyses), the reported median OS from the phase III clinical trial is higher at 29.1 months and 20.9 months for the GP and FP patient cohort, respectively (Zhang et al, 2016). Factors that may account for this difference in OS outcomes include differences in the characteristics of the patient population under study, better supportive or palliative care, improved management of chemotherapy-related side effects, as well as opportunities for patients to receive subsequent treatment on clinical trials. In addition, the earlier identification of progression with improved medical imaging techniques may further introduce lead time bias for OS.

However, this phase III trial does not address whether combination treatment is more effective than single agent therapy, or if platinum-containing regimens are superior to non-platinum-containing schedules. Our systematic review addressed this gap in knowledge. Our analyses support the use of platinum combination regimens in the first line setting with demonstrated superior PFS and OS outcomes of combination regimens over monotherapy; and platinum treatments over non-platinum regimens. First line platinum combinations for treatment of patients with RM-NPC is also recommended by expert consensus opinions such as the National Comprehensive Cancer Network (NCCN) guideline (Network, NCC, 2016).

Beyond the first line setting there is no standard of care regimen; however, if patients are suitable for further treatment they either receive another chemotherapy option such as docetaxel or participate in a clinical trial (Network, NCC, 2016). This review now provides important benchmarks (OS 11.5−12.5 months; PFS 5.2−5.4 months) for patients receiving second line treatment, which can inform future clinical trials of systemic therapies in RM-NPC. This information will be especially useful for studies adopting a single-arm design because there is now a robust historical control for comparison.

To our knowledge this is the first review of its kind in RM-NPC. Furthermore, the results can be interpreted as reliable given that the two different models typically produced similar estimates. The reason for the discrepancy between the two models for OS estimates for both platinum vs non-platinum regimens and combination vs monotherapy treatments, is probably attributable to the fewer studies included in model B because there are less publications reporting Kaplan−Meier survival curves in an extractable form and the low number of trials investigating non-platinum-containing chemotherapy regimens and single-agent chemotherapy in the treatment-naive patient population. In addition, the significantly higher PFS rates in studies conducted in non-Asian populations compared with Asian based studies is likely due to non-Asian studies typically investigating combination treatments. Notwithstanding, the broad range of included studies would permit the application of this information to most, if not all, populations of RM-NPC patients.

Our study has several limitations. The literature review was conducted to March 2015, hence results from the phase III study comparing two platinum-based combination chemotherapy regimens in the first-line setting (Zhang et al, 2016) was not included in the analysis. Original or reconstructed individual patient data from all trials was not used to calculate the OS and PFS estimates. Thus the analysis is impacted by the quality of reporting of the trial information. Trials that did not have KM curves of sufficient image quality could not be used, and thus reduced the number of studies included which as result increased the variance around the estimates. The heterogeneity of included patients in the reconstruction of survival data using model B analysis may limit the robustness of the results. Differences in DB scores further highlight the heterogeneity of clinical trials design and reporting. Furthermore, given that NPC is endemic in Asia, the exclusion of studies not published in the English language may be another limitation. Hence the results of this review must be interpreted within that context. It is important to note that the analyses performed in this systematic review were not designed to identify the optimal chemotherapy regimen, which remains to be defined. We have been particularly careful not to report the superiority of one specific regimen.

Interestingly our analysis demonstrated no significant changes in ORR or PFS over time, although there is moderate correlation between OS and the year of study publication. However, this result needs to be interpreted with caution due to the potential for biases given the majority of studies are single-armed with low patient numbers, and supportive care practices may have changed over time.

Ongoing challenges facing the interpretation of clinical trials investigating systemic therapies in RM-NPC include diversity in the patient demographics enrolled in the trials, lack of standardised treatment regimens, and relatively small patient populations studied (Zhang et al, 2016). A number of randomised clinical trials are presently ongoing (Butte et al, 2007; Lutzky et al, 2014; Ng et al, 2016), the results of which are eagerly awaited to resolve treatment dilemmas with generalisable result to various patient populations. It is hoped that novel agents such as immunotherapies may provide meaningful PFS and OS improvements in future.