Glioblastoma (GBM) is a primary central nervous system (CNS) tumor associated with poor prognosis and significant invasiveness (1). The prognosis of GBM remains poor despite first-line therapy, and the median overall survival (OS) is 12-15 months (2), while the 5-year survival does not exceed 5% (3). Tumor recurrence is the leading cause of death, and the rate of recurrence is rather high after initial treatment (4). Most patients (90%) experience recurrence in situ, and no significant relationship has been observed with the choice of therapy. Therefore, GBM recurrence has become a clinically significant issue.

Currently, participating in ongoing clinical trials prior to standard chemotherapy is strongly recommended for patients with recurrent GBM according to the National Comprehensive Cancer Network (NCCN) guideline. Other preferred treatments include bevacizumab (BEV), temozolomide (TMZ), lomustine (CCNU) or carmustine, PCV and regorafenib. The combination therapy of carmustine or CCNU + BEV and TMZ + BEV is also recommended when BEV monotherapy loses efficiency. BEV is considered to be ineffective in OS prolongation according to the European Association for Neuro-Oncology (EANO) guideline (5). Several clinical trials are currently in progress for improving the prognosis of recurrent GBM. The main exploratory therapies are chemotherapy, targeted therapy, vaccines, and tumor treatment fields (TTF). However, the recommendation categories of the second-line treatments in the NCCN guidelines are 2A, 2B, and 3, and further evidence is urgently needed to determine the feasibility of the relevant therapies.

Network meta-analysis (NMA) is a statistical approach that provides both direct and indirect evidence, which anchors the arms of the treatments in different clinical trials (6). NMA can summarize the results of several different randomized controlled trials (RCTs) by forming an evidence chain and combining studies for analysis in a network. It is also able to rank the treatment approaches and present the statistical results in a graphical format (7). NMA has already been used successfully in several fields of medicine (8).

This NMA study established a clinically meaningful hierarchy of efficacy and safety of different therapies for adult patients with recurrent GBM from published RCTs. Our study is the first to systematically integrate and compare the results of clinical trials on treating adult patients with recurrent GBM and aims to provide potential guidance for therapies for recurrent GBM.

A total of 1,667 citations were identified by the primary research, and 15 eligible articles remained in the final network meta-analysis ( Figures 1 , S1 , S5 , S9 ). The baseline characteristics of the involved studies are shown in Table 1 . A total of 2253 patients were randomly assigned to 18 treatment arms, which included 17 active drug therapies, of which the majority are based on BEV or CCNU, and TTF plus BEV ( Figure 2 ). Four phase III (15, 16, 25, 26) and 11 phase II RCTs (14, 17–24, 27, 28) were included with a median age of involved patients ranging from 52 to 61 years. Assessment of model fitting and risk of bias are shown in the supplementary documents. The median heterogeneity variances were estimated as 0.16 (95%CrI, 0–0.35) for ORR, 0.14 (95% CrI, 0–0.33) for 6m-PFS rate and 0.11 (95%CrI, 0–0.27) for OS, which shown a low-to-moderate heterogeneity ( Table S2 ). Six of 17 included RCTs shown a high risk of bias, which were all referred to non-blinded design, based on Cochrane Collaboration Handbook ( Table S4 ). The qualitative evaluation of evidence was done followed GRADE guideline, and we evaluate most evidences as low-to moderate ( Tables S5 – 7 ).

To date, a number of clinical trials investigating multiple monotherapies or combination therapies of chemotherapy, targeted therapy and immunotherapy for recurrent GBM have been published. Recently, the EANO guidelines pointed out that there are no well-defined standards of care for recurrent GBMs (5). A relatively large number of treatments make it difficult for clinicians to make a proper selection in clinical practice. Nevertheless, no direct comparison of existing regimens is available, and it is difficult to draw conclusions.

The purpose of our network meta-analysis was to synthesize all the available evidence from current research, compare the efficacy and safety of recurrent GBM therapies, and contribute to identifying the more effective therapies among those currently available. To our knowledge, this study represents the most comprehensive overview of the efficiency and safety data available for the treatment of recurrent GBM, and its main findings can be summarized in the following paragraphs.

There are no statistically significant differences among the included therapies in the three efficiency indicators of OS, 6-m PFS, and ORR. This result may be due to the fact that most clinical trials in recurrent GBM did not yield positive results. It is not so realistic to find a rigidly efficient cure for recurrent GBM in the short term, thus analyzing the published results for a better combination is a more practical option. However, considering that it is difficult for a specific therapy to appear in the short term, and clinicians need to choose among the current therapies, we have ranked the current therapies and determined the relatively effective ones.

The analysis results suggested that BEV (10 mg/kg) and CCNU (90 mg/m²) combination therapy and regorafenib are relatively superior for improving OS, 6-m PFS, and ORR, and the adverse effects profile of these two therapies is acceptable. Therefore, BEV and CCNU combination therapy and regorafenib monotherapy are recommended in clinical practice considering its relatively high efficiency. The efficacy of regorafenib has also been emphasized in the latest NCCN guideline, where regorafenib is listed as a preferred regimen.

Our results recommend BEV and CCNU combination therapy after comprehensive analysis, but the phase III EORTC 26101 trial in 2017 reported that although PFS is prolonged in the combination group, the effect of combination therapy did not exceed the effect of CCNU monotherapy in OS, and was only listed in other recommended regimens in NCCN guideline that can be considered after BEV monotherapy failure (26). However, combining the two trials statistically, we found BEV and CCNU had a relatively prior efficacy (17, 26), and this should be interpreted imprecisely in the clinical practice due to tremendous heterogeneity of patients with recurrent GBM.

Recently, molecular pathological status shown as a biomarker of treatment and prognosis for patients with GBM. In our research, five trials reported IDH mutation status (17, 19, 20, 23) (IDH mutant patients accounted for 3–10%) and five reported MGMT promoter methylation (patients with MGMT promoter methylation accounted for 37–68%). In terms of a single study conclusion, the phase II BELOB trial showed that both PFS and OS of IDH mutant patients after BEV and CCNU treatment were significantly higher than that of IDH wild-type patients (17). Considering that the proportion of IDH mutant patients was relatively low, their influence on the result analysis was relatively small. At the same time, the studies showed that patients with MGMT promoter methylation received BEV and CCNU combination therapy (17) and onartuzumab plus BEV treatment (23), PFS and OS were significantly higher than those without MGMT promoter methylation, while OS in patients receiving CCNU and BEV or OS-6 of patients treated with fotemustine or BEV was also higher in methylation-positive patients than those without MGMT promoter methylation (20, 26). Therefore, our research suggests possible future clinical research directions, as multiarmed head-to-head RCTs with subgroup analysis, which was descripted in GBM AGILE trial (30).

The EGFRvIII-targeted therapies have become a specific research field in GBM. In previous studies, the peptide vaccine rindopepimut showed benefits in prolonging OS and symptom control in patients with recurrent GBM with EGFRvIII mutation (28). A majority of EGFRvIII-mutant GBMs maintain the mutation at recurrence. However, a subset of patients may experience EGFRvIII expression change at recurrence, thus reassessing the EGFRvIII status for patients with recurrent GBM is recommended before EGFRvIII-targeted therapies (31).

The included trial only reported the efficiency of TTF and BEV combination therapy on OS. The effect of this therapy on prolonging OS is only second to BEV and CCNU combination therapy. Therefore, this treatment is recommended if the family can financially afford the treatment and the patient has good compliance and can receive better care. We suggest that future clinical trials should focus on the impact on PFS and ORR to determine the efficiency of TTF and BEV combination therapy in the treatment of recurrent GBM. We look forward to further reports on the combination therapies of TTF and other drugs.

The combination therapies including BEV were extremely effective in prolonging PFS and ORR, and this result was related to the effect of BEV in controlling symptoms. Thus far, studies have confirmed the role of BEV in alleviating necrosis and brain edema after radiotherapy (32), and it has been proven to be effective in prolonging PFS for patients with recurrent GBM. Therefore, BEV has been recommended as preferred regimens in the latest version of NCCN guidelines, while the BEV-based combination therapies listed as other recommended regimens. This conclusion suggests that combination therapies, including BEV, are of great significance, especially for patients with tumor-related symptoms such as cerebral edema.

In this study, we performed indirect comparisons and comprehensive rankings of therapies that have not been directly compared in clinical trials, which may support clinicians in the efficient health care of recurrent GBM patients and help clinical teams and patients make joint medical decisions, thereby increasing the possibility of prolonging survival and improving treatment effectiveness. Our study also statistically summarized the incidence of adverse events of various therapies, which can assist clinicians in increasing the surveillance, prevention and management of drug-related toxic effects. We also performed the assessment of RCTs heterogeneity, evidence certainty and risk of bias. We found a low-to-moderate heterogeneity between included RCTs, so we can combine the direct and indirect comparisons between those treatments. However, we found a low-to-moderate qualitative level of each evidence chain, because we downgraded all pairwise comparisons by one level for almost half of trials included were sponsored commercially and with a relatively small sample size, and several single-arm clinical trials were not included. Also, a few clinical trials were not included for a various interaction in control group. And this is one of the limitations of our study and of vital importance when we discuss the optimal therapy for recurrent GBM.

Due to the relatively low incidence of GBM and high degree of malignancy and low mean OS in patients, there are indeed several trials adopted single-arm treatment or non-randomized trial design. Our search was unable to include ongoing studies and trials with additional unpublished data, also, single-arm clinical trials, therefore, a considerable amount of meaningful studies were not included. Secondly, the definitions of 6-m PFS and OS were not exactly same among the included studies. To be specific, five studies calculated efficacy from randomization and two studies defined efficacy from start of treatment, while eight articles did not clarify the specific initial node for efficacy evaluation. The inconsistency among researches may bring certain bias to the statistical results. In addition, the included published results lag behind the current therapy option. Besides TTF, new therapies including neoadjuvant checkpoint inhibitor and laser interstitial thermotherapy has not been included. And these advanced therapies remain to be explored in future studies.

Because the network meta-analysis requires the included studies to form an evidence chain, the regimen not involved in other studies cannot be included. In our analysis, there were nine studies that met the inclusion criteria, but there was no corresponding comparable regimen in other trials. Among them, seven compared the efficacy and safety between two therapies, and the other two explored the efficacy of different doses of cilengitide (33) and ABT-888 (34). In the studies comparing different therapies, most of them did not achieve the statistical different results of the primary endpoints (including TMZ and dose-dense TMZ versus PCV regimen (35), carboplatin and RMP-7 combination therapy versus carboplatin (36), TTF vs physician’s choice chemotherapy (37), afatinib monotherapy and afatinib plus TMZ versus TMZ (38), different dose of single-agent CT-322 and irinotecan combination therapy (39), and cintredekin besudotox versus carmustine (40). TMZ and dose-dense TMZ regimens showed no significant survival benefit than PCV. PD-1 inhibitor pembrolizumab improved both OS and PFS compared with only adjuvant therapy (41). The requirement of the evidence chain also results in some high-quality studies not being included for comparison, which to some extent affects the comprehensiveness of the results.

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.

Study design and manuscript writing: YueW, WC, YuW, and WM. Clinical trials searching: YueW, WC, PL, and LL. Model constructing and statistical analysis: YueW, WC, and BZ. Manuscript formatting and revising: YueW and WC. All authors contributed to the article and approved the submitted version.

Beijing Municipal Science & Technology Commission [7202150]; Beijing Municipal Science & Technology Commission [19JCZDJC64200(Z)]; Chinese Academy of Medical Sciences [2016-I2M-2-001]; Peking Union Medical College Hospital [2019ZLH101].

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.