Optimal Therapies for Recurrent Glioblastoma: A Bayesian Network Meta-Analysis

Abstract

The optimal treatment of recurrent glioblastoma (GBM) remains controversial. Therefore, our study aimed to compare and rank active therapies in recurrent GBM. We performed a systematic review and a Bayesian network meta-analysis. We obtained a treatment hierarchy using the surface under the cumulative ranking curve and mean ranks. A cluster analysis was conducted to aggregate the separated results of three outcomes. The protocol was registered in PROSPERO (CRD42019146794). A total of 1,667 citations were identified, and 15 eligible articles with 17 treatments remained in the final network meta-analysis. Pairwise comparison showed no significant difference on the 6-month progression-free survival (6-m PFS) rate, objective response rate (ORR), and overall survival (OS). Among the reports, cediranib plus lomustine (CCNU) corresponded to the highest rates of grade 3-4 adverse events. Ranking and cluster analysis indicated that bevacizumab (BEV) plus CCNU and regorafenib had a higher efficacy on the ORR, 6-m PFS rate and OS, and that BEV monotherapy or BEV combined with active drug therapies was advantageous for the ORR and 6-m PFS rate. Additionally, tumor treatment fields (TTF) plus BEV showed a relatively higher SUCRA value in OS. According to ranking and cluster analysis, BEV plus CCNU and regorafenib are the primary recommendations for treatment. BEV monotherapy alone or combined with active drug therapies are recommended in patients with severe neurological symptoms. Advanced therapy, such as TTF and immunotherapy, remain to be investigated in future studies.

Keywords: Bayesian network meta-analysis; bevacizumab; combination therapy; recurrent glioblastoma; systematic review.

Figure 1

Study selection flowchart. A total of 1667 citations were identified by the primary research, and 15 eligible articles remained in the final network meta-analysis. The flowchart was made under the PRISMA guideline (13).

Figure 2

Network plot of all eligible comparisons involved. The size of every solid circle is proportional to the total sample size, and the width of the line is proportional to the number of clinical trials. Network plots of each outcome are shown in Supplementary Figures . BEV, bevacizumab monotherapy or combined with placebo; CCNU, lomustine monotherapy or plus placebo; TMZ, temozolomide; BEV_CCNU, bevacizumab plus lomustine; BV5_CCNU90, low-dose bevacizumab (5 mg/kg) plus low-dose lomustine (90 mg/m²); BEV_TMZ, bevacizumab plus temozolomide; BEV_ONA, bevacizumab plus onartuzumab; BEV_IRV, bevacizumab plus irinotecan; BEV_CAR, bevacizumab plus carboplatin; GAL, galunisertib; FOT, fotemustine; ENZ, enzastaurin; CED, cediranib; GAL_CCNU, galunisertib plus lomustine; CED_CCNU, cediranib plus lomustine; CED_GEF, cediranib + gefitinib; TTF_BEV, tumor treatment field plus bevacizumab; REG, regorafenib; RIN_BEV, rindopepimut plus bevacizumab.

Figure 3

Forest plot of the relative effects compared with BEV. All therapies were compared with BEV in 3 outcomes: (A) ORR, (B) 6-m PFS rate, and (C) OS. The size of every solid square is proportional to the total sample size. The abbreviations are defined in the legend of Figure 2 .

Figure 4

Heat-rank plot and cluster analysis of SUCRA. (A) Heat-rank plot of SUCRA. Each circle shows the SUCRA value for ORR, 6-m PFS, and OS from outside to inside. Interactions were labeled, and each sector was colored depending on the SUCRA value. The SUCRA value scale is shown, and “*” refers to missing data. (B) Cluster analysis of all 3 outcomes. All therapies involved were divided into three clusters and are shown in 3 circles with distinctive colors. Clusters are shown in a two-dimensional graph by PCA, and the attribution of the SUCRA value of three outcomes to the two dimensions is shown by vectors in different colors. SUCRA_ORR, SUCRA value for ORR; SUCRA_6m, SUCRA value for 6m PFS rate; SUCRA_OS, SUCRA value for OS.