LAUSR

Given the significant advances using both targeted therapies and immunotherapies in other solid tumors, the therapeutic advances in glioblastoma (GBM) are particularly disappointing. One might think that an important factor is a lack of understanding of the underlying biology of GBM. However, GBM is actually one of the solid tumors with the most comprehensive genomic and molecular profiling available, and was one of the first tumors profiled by The Cancer Genome Atlas (TCGA) Network.1 A more detailed understanding of molecular subtypes of GBM and other infiltrating gliomas has recently led to a revision of the WHO diagnostic and classification criteria for these tumors.2,3 These efforts highlight the observation that while the molecular alterations in individual genes are heterogeneous in GBM, a number of the alterations affect a core set of pathways that are dysregulated in a high percentage of tumors.1,4 Thus, attempting to target these consensus pathways in GBM with appropriately potent drugs appears to be a reasonable approach. However, determining therapeutic effect and the relationship to specific biomarkers in such a situation is much more challenging than for a high-frequency consensus alteration such as BRAF mutations in melanoma. Given these challenges, it is possible that we need to look to different study designs that allow for testing of multiple hypotheses in multiple tumor subgroups simultaneously.5 Traditional phase II trials are designed to test a single hypothesis in a specific patient population with a predefined statistical power. While appropriate for this task, these designs are not efficient for testing multiple hypotheses across a heterogeneous tumor population. To address some of these limitations, a number of proponents have advocated using a “platform” approach, often with Bayesian or adaptive statistical designs.6,7 These approaches offer several potential advantages compared with 2-arm randomized studies. First, the platform trial includes multiple experimental arms with a single shared and contemporaneous control, which increases efficiency compared with running multiple synchronous or asynchronous 2-arm studies, each requiring its own control arm. Second, adaptive designs allow for changes in the trial design in response to incremental information and data that are acquired during the performance of the trial, including: preferential randomization of patients to arms performing the best, early closing of arms due to poor performance, and replacement of arms with new arms (which can also incorporate information gained during the study). The utility of both the platform and Bayesian approaches is exemplified by several current trials in various cancers, including the NCI Match and I-SPY2 trials.8,9 In the manuscript by Tanguturi and Trippa et al in the current issue, the authors take an initial step toward the design and implementation of a multi-arm biomarker-based adaptive trial for newly diagnosed GBM.10 The authors focus on biomarker subtypes that were prioritized by the Targeted Therapies Working Group of the Brain Malignancy Steering Committee, including: epidermal growth factor receptor (EGFR) amplification and/or mutation, phosphatidylinositol-3 kinase (PI3K) pathway activation, cell cycle targets, and the p53 axis.11 The authors utilize a large single institution patient cohort and validation using the dataset of TCGA to investigate the individual biomarker prevalence, biomarker overlap, and relationships to clinical covariates and endpoints in clinical trial simulations. Tumors were assigned to biomarker positive or negative for each group (EGFR, PI3K, p53, and cyclin-dependent kinase [CDK]) based on available molecular profiling data. Multiple clinical variables and outcomes including progression-free survival (PFS), survival post-progression (SPP), and overall survival (OS) were included in the univariate and multivariate analyses. The 4 primary biomarker groups were relatively balanced in incidence, although the p53(+) subclass occurred slightly less frequently. EGFR(+) and p53(−) subtypes were significantly more likely to be isocitrate dehydrogenase wild type. In terms of overlap of biomarker subtypes, EGFR(+) and CDK(+) subtypes were most frequently associated with inclusion in other biomarker positive groups. None of the 4 individual biomarkers demonstrated independent differences in PFS or OS after accounting for the statuses of O6-DNA methylguanine-methyltransferase (MGMT) and isocitrate dehydrogenase. There were also no strong associations between individual biomarker subtypes and differences in MGMT promoter methylation. The PI3K(+) subgroup was significantly older than other subgroups and demonstrated shorter PFS on univariate analyses, but this PFS difference was not significant after accounting for other 880 Neuro-Oncology