Our current network meta-analysis compared four treatment arms (CRT, CRT + adjuvant, induction + CRT, and RT alone) in advanced NPC and found that induction + CRT was better than CRT in reducing distant metastasis and locoregional recurrence, while CRT + adjuvant was superior to CRT in improving OS. Notably, induction + CRT did not significantly differ from CRT + adjuvant. RT alone was always worse than the other three treatments. Further, sensitivity analysis after excluding the study by Kwong et al38 also confirmed these findings.
As intensity-modulated radiotherapy has become the predominant treatment for patients with NPC, locoregional control of advanced disease has improved greatly and distant metastasis has become the main failure pattern.41,42 Therefore, combined strategies of chemotherapy with radiotherapy have been intensively investigated. A possible and feasible strategy is additional cycles of chemotherapy, such as adjuvant chemotherapy and induction chemotherapy, to CRT. In fact, both CRT + adjuvant and induction + CRT have been recommended as the standard cure for patients with advanced NPC by the National Comprehensive Cancer Network (NCCN) guidelines. However, this produces another question: which one is better? CRT + adjuvant or induction + CRT? As far as we know, there are no head-to-head clinical trials comparing CRT + adjuvant with induction + CRT being reported yet. Given this concern, we performed this network meta-analysis to provide preliminary answer for this question. Our findings indicated that CRT + adjuvant and induction + CRT were equally effective.
A recent individual patient data pooled analysis of four randomized clinical trials comparing induction + CRT with CRT revealed that induction + CRT was superior over CRT in reducing distant metastasis and improving OS.17 Similar to these findings, our study also demonstrated that induction + CRT was better than CRT both in direct and network meta-analysis. However, when comparing induction + CRT with CRT + adjuvant, no significant difference was found in both original and sensitivity analysis for all endpoints, which was consistent with the findings in an individual patient data network meta-analysis.15 Obviously, the incorporation of new evidence11,12,14 about induction + CRT into the network meta-analysis still failed to demonstrate the superiority of induction + CRT over CRT + adjuvant. One possible explanation is that previous value regarding CRT + adjuvant is too strong since RT alone was the control arm,4–7,39 therefore increasing the weight of CRT + adjuvant in the network comparison. Given these findings and concerns, head-to-head randomized clinical trials comparing induction + CRT with CRT + adjuvant are urgently warranted. Possibly, only results from such trials could be conclusive.
To validate the results of original analysis, we performed sensitivity analysis after excluding the study by Kwong et al38 because a standard concurrent chemotherapy regimen of cisplatin was not used in that trial. Generally, the results of sensitivity remained valid, indicating that our meta-analysis was reliable. Of note, CRT + adjuvant was better than CRT for OS in original analysis, while they were comparable in sensitivity analysis. The reason being only one trial comparing CRT + adjuvant with CRT existed8 after excluding the study by Kwong et al.38 Actually, no significant difference between CRT + adjuvant and CRT was found in the original text.8 Undoubtedly, a lack of such trials would affect the final results of network meta-analysis. Therefore, future meta-analysis should include more trials.
It should be pointed out that we could not only rely on the P-score to select treatment although we employed it to rank treatment. A treatment ranking probability would still be generated without definitive statistical significance even if differences of effect size between treatments were small and nonsignificant. Thus, we should not overinterpret the P-score in network meta-analysis.
We should also address the limitations of this study. Since we have no access to individual patient data, potentially reporting bias could be produced. To minimize such bias, we set strict criteria for study inclusion to reduce heterogeneity between studies. Another important issue should be the application of different radiotherapy technique (intensity-modulated radiotherapy vs conventional radiotherapy) across different trials which may affect the results. Moreover, we did not evaluate the endpoint of PFS because the definition of PFS varied greatly across trials. Of course, these limitations should be addressed in future studies.