Identifying the best treatment choice for relapsing/refractory glioblastoma: a systematic review with multiple Bayesian network meta-analyses

Abstract

Background: Glioblastoma is a highly aggressive primary central nervous system tumor characterized by poor outcomes. In case of relapse or progression to adjuvant chemotherapy, there is no univocal preferred regimen for relapsing glioblastoma.

Methods: We conducted a systematic review and Bayesian trial-level network meta-analyses (NMA) to identify the regimens associated with the best outcomes. The primary endpoint was overall survival (OS). Secondary endpoints were progression-free survival (PFS) and overall response rates (ORR). We estimated separate treatment rankings based on the surface under the cumulative ranking curve values. Only phase II/III prospective comparative trials were included.

Results: Twenty-four studies (3733 patients and 27 different therapies) were ultimately included. Twenty-three different regimens were compared for OS, 21 for PFS, and 26 for ORR. When taking lomustine as a common comparator, only regorafenib was likely to be significantly superior in terms of OS (hazard ratio: 0.50, 95% credible interval: 0.33-0.75). Regorafenib was significantly superior to other 16 (69.6%) regimens, including NovoTTF-100A, bevacizumab monotherapy, and several bevacizumab-based combinations. Regarding PFS and ORR, no treatment was clearly superior to the others.

Conclusions: This NMA supports regorafenib as one of the best available options for relapsing/refractory glioblastoma. Lomustine, NovoTTF-100A, and bevacizumab emerge as other viable alternative regimens. However, evidence on regorafenib is controversial at best. Moreover, most studies were underpowered, with varying inclusion criteria and primary endpoints, and no longer adapted to the most recent glioblastoma classification. A paradigmatic change in clinical trials' design for relapsing/refractory glioblastoma and more effective treatments are urgently required.

Keywords: Bayesian; bevacizumab; glioblastoma; network meta-analysis; regorafenib.

Figure 1.

PRISMA flow-chart. Abbreviations: GB, glioblastoma; ORR, overall response rate; OS, overall survival; PFS, progression-free survival.

Figure 2.

Treatment networks of OS, PFS, and ORR. Direct comparisons are represented by the black lines connecting the treatments. Line width is proportional to the number of trials including every pair of treatments, whereas circle size is proportional to the total number of patients difference in the proportion of patients for each treatment in the network. (A) OS network; (B) PFS network; (C) ORR network. Abbreviations: BEVA, bevacizumab; CARBO, carboplatin; CED, cediranib; DASA, dasatinib; ENZA, enzastaurin; ERC1671, sitoiganap; ETO, etoposide; GALU, galunisertib; GEFI, gefitinib; IRINO, irinotecan; LD, low dose; LOM, lomustine; LOM90, lomustine at a dose of 90mg/m²; NIVO, nivolumab; NOVOTTF, NovoTTF-100A; ONAR, onartuzumab; ORR, overall response rate; OS, overall survival; PFS, progression-free survival; RADIO, radiotherapy; REGO, regorafenib; RINDO, rindopepimut; TEMO, temozolomide; TREB, trebananib; VB-111, ofranergene obadenovec; VOR, vorinostat.

Figure 3.

Forest plot of OS of all included regimens for relapsed/refractory glioblastoma compared to lomustine monotherapy and treatment ranking based on SUCRA. (A) A forest plot including the logHR of OS of each treatment vs lomustine monotherapy. Central dots represent posterior medians; thin lines represent 95%CrI, while thicker ones represent 80% CrI. Log scale was adopted to graphically represent the 95% CrI. The first column of values on the right reports the logHR with 95%CrI, the second column reports HR for OS with 95%CrI. Statistically significant results according to Bayesian posterior medians and 95%CrI are highlighted by asterisks. (B) Treatment ranking according to the OS endpoint, based on SUCRA values. Bar plot displaying SUCRA values for treatments analyzed in a Bayesian network meta-analysis. Each bar represents a treatment, with the height of the bar corresponding to its SUCRA value, which indicates the probability of a treatment being among the most effective options. Higher bars denote higher SUCRA values, suggesting greater effectiveness relative to other treatments. The y-axis lists the treatments, and the x-axis shows the SUCRA values as percentages. Abbreviations: BEVA, bevacizumab; CARBO, carboplatin; CED, cediranib; CrI, credible interval; DASA, dasatinib; ENZA, enzastaurin; ERC1671, sitoiganap; GALU, galunisertib; GEFI, gefitinib; HR, hazard ratio; LOM, lomustina; LOM90, lomustine at a dose of 90mg/m²; NIVO, nivolumab; NOVOTTF, NovoTTF-100A; ONAR, onartuzumab; OS, overall survival; RADIO, radiotherapy; REGO, regorafenib; RINDO, rindopepimut; SUCRA, surface under the cumulative ranking curve; TREB, trebananib; VB-111, ofranergene obadenovec; VOR, vorinostat. The ranking also includes regimens that are currently not approved or no longer in active development. *Identify regimens potentially available in clinical practice, although geographical restrictions exist due to heterogeneous approvals and reimbursement worldwide.

Figure 4.

Forest plots of PFS and ORR of all included regimens for relapsed/refractory glioblastoma compared to lomustine monotherapy. (A) A forest plot including the logHR of PFS of each treatment vs lomustine monotherapy. Central dots represent posterior medians; thin lines represent 95%CrI, while thicker ones represent 80% CrI. Log scale was adopted to graphically represent the 95% CrI. The first column of values on the right reports the logHR with 95%CrI, the second column reports HR for PFS with 95%CrI. Statistically significant results according to Bayesian posterior medians and 95%CrI are highlighted by asterisks. (B) A forest plot including the logOR for ORR of each treatment vs lomustine monotherapy. Central dots represent posterior medians; lines represent 95% CrI. Log scale was adopted to graphically represent the 95% CrI. The first column of values on the right reports the logOR with 95% CrI, and the second column reports OR with 95% CrI. Statistically significant results according to Bayesian posterior medians and 95% CrI are highlighted by asterisks. Abbreviations: BEVA, bevacizumab; CARBO, carboplatin; CED, cediranib; CrI, credible interval; DASA, dasatinib; ENZA, enzastaurin; ERC1671, sitoiganap; ETO, etoposide; GALU, galunisertib; GEFI, gefitinib; HR, hazard ratio; IRINO, irinotecan; LD, low dose; LOM, lomustina; LOM90, lomustine at a dose of 90mg/m²; NIVO, nivolumab; NOVOTTF, NovoTTF-100A; ONAR, onartuzumab; OR, odds ratio; ORR, overall response rate; PFS, progression-free survival; RADIO, radiotherapy. REGO, regorafenib; RINDO, rindopepimut; TEMO, temozolomide; TREB, trebananib; VB-111, ofranergene obadenovec; VOR, vorinostat.

Figure 5.

Global risk of bias analysis.