Novel Clinical Trial Designs in Neuro-Oncology

Abstract

Scientific and technologic advances have led to a boon of candidate therapeutics for patients with malignancies of the central nervous system. The path from drug development to clinical use has generally followed a regimented order of sequential clinical trial phases. The recent increase in novel therapies, however, has strained the regulatory process and unearthed limitations of the current system, including significant cost, prolonged development time, and difficulties in testing therapies for rarer tumors. Novel clinical trial designs have emerged to increase efficiencies in clinical trial conduct to better evaluate and bring impactful drugs to patients in a timely manner. In order to better capture meaningful benefits for brain tumor patients, new endpoints to complement or replace traditional endpoints are also an increasingly important consideration. This review will explore the current challenges in the current clinical trial landscape and discuss novel clinical trial concepts, including consideration of limitations and risks of novel trial designs, within the context of neuro-oncology.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13311-022-01284-x.

Introduction

The drug development process in neuro-oncology has been noted to be slow, expensive, and inefficient. Among central nervous system tumors, much of the focus on novel trial designs have centered on glioblastoma (GBM). GBM is the most common primary brain malignancy and is associated with a poor prognosis; standard of care has not substantially changed since the benefit of temozolomide (TMZ) was first demonstrated in 2005 [1]. In reviewing the GBM trial landscape, identified issues include long development times, suboptimal choice of clinical trial endpoints, and poor go/no-go decision making in early phase settings [2–4]. Novel clinical trial designs have attempted to tackle a number of these problems in trials for GBM as well as other brain tumors (Table 1) [5, 6].

Table 1.

Novel trial design elements and related examples

Trial design element	Challenge addressed	Trial example	Description
Common trial infrastructure	Inefficiency of enrollment	Larotrectinib for NTRK fusion (Basket Trial)	Similar target in different malignancies
		Alliance A071701: Genomically guided treatment in brain metastasis (Umbrella Trial)	Multiple therapies for a single molecular subgroup
Shared common control arm	Rapid investigation of multiple therapies	GBM AGILE (Adaptive Platform Trial)	Multiple experimental arms are compared to common control arm within clinical/biomarker patient subtypes
External control arm	Extrapolation of single-arm studies	NCT02858895: Evaluation of MDNA55	Outcomes of patients receiving MDNA55 were compared with an eligibility-matched synthetic control arm patients
Adaptation of trial arms using real-time reporting	Efficiency of identifying beneficial arms	GBM INSIGhT (Bayesian Adaptive Platform Trial)	Bayesian estimation of on-going results adapts probability of success, allowing treatment arms to be dropped and new arms to be added prior to completion of trial
Built in registrational trial component	Transition of phase 2 to 3 trials	GBM AGILE	Therapies successfully identified in early phase can be seamlessly transitioned to a registrational evaluation
Tumor-specific endpoints	Prolonged reporting of survival endpoints in diseases with favorable prognosis	Rate-of-growth in lower-grade gliomas (retrospective study)	Volumetric growth before and after treatment was associated with tumor biomarkers
Patient-reported outcomes as endpoints	Benefit of therapy beyond survival endpoints	NRG CC001	Hippocampal avoidance with meaningful neurocognitive benefit

The objective of a phase I study is to establish safety and identify the maximum tolerated dose of a therapy. A limitation of the traditional phase I study is the lack of emphasis on assessing response; in recent years, clinical trials have employed combined phase I/II designs to integrate the results of phase I study results into subsequent analyses [7]. Another limitation of traditional phase I studies is the rigidity of requiring a set number of patients at each dose level with extended periods of toxicity monitoring; some studies have utilized Bayesian statistics to adapt therapeutic dose based upon early toxicity results, allowing meaningful results with fewer patients [8].

Phase II clinical trials are intended to assess efficacy and determine which therapies warrant further investigation in a phase III setting. Single-arm phase II clinical trials are common in neuro-oncology, likely driven by shorter time to study completion, lower cost, and preference of physicians and patients to participate in nonrandomized trials [9]. Incentives related to pressure for academic productivity and competition to attract research funding and patient enrollment may also contribute to a propensity towards single-arm trial designs in spite of evidence in favor of the use of randomized designs [10, 11].

Despite an increase in published phase II trials in oncology, they have not necessarily translated to positive results in phase III trials or a change in practice [12–14]. Novel phase II/III study designs, such as master platform studies and Bayesian adaptive studies, use novel methods to increase efficiency of early phases and confirmatory trials without compromising scientific validity.

Early Phase Trial Design Considerations

Window-of-opportunity and perioperative trials are of increasing interest in neuro-oncology [15]. Window-of-opportunity trials are a type of perioperative study performed after a phase 1 study where the drug is administered between the window from tumor diagnosis and tumor resection, with a specific pharmacokinetic and pharmacodynamic focus related to drug detection within tumor tissue, engagement with molecular target, or association with tumor microenvironment [16, 17]. For example, numerous trials have demonstrated that checkpoint inhibition does not have the same therapeutic effect in GBM as in melanoma or non-small-cell lung cancer [18, 19]. In investigating further, a window-of-opportunity study found a scarcity of T cells within the tumor microenvironment and preponderance of CD68 + macrophages, suggesting a mechanism to potentially explain the lack of immune response with pembrolizumab alone in GBM [20]. In general, perioperative trial designs can document penetration of blood–brain barrier, drug concentrations in tumor, and free drug concentrations. The drug can then be subsequently resumed after surgery to evaluate PFS and OS [21, 22].

Causes for the low success rates of phase III trials for therapies that have shown promising results in phase 2 settings in GBM include overreliance on single-arm studies, the choice of primary outcomes, and limited data and incentives to better understand therapies that fail [2, 23, 24]. A number of studies have considered randomization with a control arm in early-phase studies to minimize confounding factors, though this may require a larger effect size as well as possibly larger enrollment [25]. A simulation-based study in GBM, holding sample sizes constant, however, supports the idea that randomized trials provide more accurate treatment effect estimates and better decisions on whether to continue drug development compared to single arms trials in the phase 2 setting [11]. Among approaches employed in the early phase of drug development, some studies have attempted to augment signal finding by utilizing functional imaging to better assess tumor biology effect [26], pretreatment growth trajectory with longitudinal imaging analysis [27], and monitoring presence of circulating tumor DNA in cerebrospinal fluid [28].

Master Protocols

Unlike traditional clinical trials that investigate a single drug in a single population, master protocols increase flexibility of trial design by investigating multiple hypotheses within a single trial infrastructure [29–31]. There are many master protocol trials in oncology given significant advances in molecular and genomic subtyping, particularly in the era of precision medicine with its emphasis on tailored therapies [32–34]. Due to significant coordination and resources required to test therapies, a common protocol and a trial network with shared infrastructure can be critical to gain efficiencies in trial conduct [29]. Centralization of trial infrastructure and shared governance also allows for standardization of biomarker analysis, including genomic, radiomic, and other advanced techniques, which ultimately drive enrollment into different intervention arms in master protocols.

Types of master protocol trials include basket trials, umbrella trials, and platform trials (Fig. 1). While there is potential for efficiency with these trial designs, there can be challenges due to complexities in designs with less familiar workflows and statistical analyses.

Fig. 1.

Examples of master protocol trial designs, including basket trial, umbrella trial, and platform trials

Basket Trials

As demonstrated by some of the first examples of basket trials (e.g., NCI-MATCH and NCI-MPACT), basket trials are particularly advantageous in its ability to include patients with a targetable mutation in rare patient populations [35–38]. Basket trials can also lead to regulatory approvals based upon specific mutations independent of disease type, as was seen with the approval of pembrolizumab for high microsatellite instability and high tumor mutational burden tumors [39, 40]. By allowing a broader entry population, basket trials allow for more efficient enrollment onto trials, though this may complicate results with heterogeneity in treatment efficacy across populations. An example of such a discordance is seen in the case of BRAFV600 inhibition, as there is more efficacy with targeted therapy in BRAFV600-mutant melanoma and hairy cell leukemia in comparison to BRAFV600-mutant colon cancer [41, 42].

The VE-BASKET study is an example of a basket trial in neuro-oncology that investigated the efficacy of vemurafenib, a BRAF-V600-mutant selective inhibitor, in patients with glioma [43]. The trial enrolled multiple histologies, including higher-grade gliomas as well as more favorable pleomorphic xanthoastrocytoma, and pilocytic astrocytoma, with the hypothesis that any tumor with a specific BRAF-V600 mutation should respond to the selective inhibitor irrespective of histological diagnosis [44]. The study found an overall objective response rate of 25% across all tumor types, with more responses seen in lower-grade gliomas [43]. A limitation of this study was the lack of central testing of glioma subtypes and BRAF mutation which could have been useful to estimate to estimate heterogeneity of responses.

Umbrella Trials

Umbrella trials investigate multiple therapies in a single cancer type to allow for testing of multiple subgroups within a heterogeneous disease. Umbrella trials, such as Alliance 071701, Alliance 071401, and Neuro Master Match (N2M2, NOA-20), allow biomarker-driven assessment of novel therapies in a single population, however, are limited in heterogeneous diseases, where conflicting biomarkers and disease factors may confound sensitivity to therapy [45–47]. Umbrella trials can more easily include a control arm since it studies one disease type, while basket trials can require distinct control groups specific of a disease or subpopulation [34].

The N2M2 trial, an open-label multicenter phase I/IIa trial, was designed by the German Cancer Society [47]. As patients with IDH-WT, MGMT-unmethylated GBM are thought to have less benefit from standard therapy with TMZ [48], N2M2 attempts to assign patients to an experimental therapy in combination with RT based on molecular signatures. Patients undergo timely molecular testing, including genome, exome, and transcriptome sequencing, within 4 weeks of enrollment, and match these patients to specific therapies that are associated with molecular signatures. Patients can be assigned to one of five targeted therapy subtrials based on matching molecular signatures; otherwise, patients without matching molecular signatures are randomized to biomarker-agnostic experimental arms or a control arm with TMZ. The phase I component of the trial utilizes Bayesian criterion for toxicity, while the phase II component utilizes 6-month PFS to assess efficacy.

A strength of N2M2 is the ability to efficiently assess molecular biomarkers for strategic enrollment, allowing for enrichment of trial population most likely to benefit from a specific therapy. Further, the trial allows for updating of drug targets and biomarkers, enabling modification and expansion based upon external data and feedback from study arms. There are a few limitations with this design. One limitation is the setting of overlapping biomarkers, which may limit accrual in arms with rarer biomarker expression. Another limitation is the lack of true randomization; while there is a control arm in biomarker-negative patients, endpoints such as PFS and OS are prone to variability in single-arm studies; this is a general issue for any trial of small biomarker subpopulations [49].

Platform Trials

Platform studies are a type of master protocol clinical trial that aims to study multiple targeted therapies in a single disease, with the ability for therapies to enter or leave the platform (Fig. 2) [29, 50]. Adaptive clinical trials utilize data in real time to inform trial adjustments that ultimately lead to decrease in time required for therapies to be assessed. Platform trials often incorporate response adaptive randomization with every patient update, which can adjust sample size and randomization of patients with particular biomarker subtypes [51, 52]. Response adaptive randomization allows experimental arms with less encouraging preliminary results to close early and reallocate those patients and resources to other experimental arms. Response adaptive adjustments can be integrated at pre-specified interim analyses, where information from interim analysis is used to inform trial decision-making, leading to trial optimization. Specific algorithms are prespecified to allow experimental arms to open or close over pre-planned interim analyses. The first examples of this come from other disease settings such as Lung-MAP in advanced squamous non-small-cell lung cancer. Lung-MAP was designed as a phase II/III platform study that closed experimental arms if they could not rule out at least 15% overall response rate [53].

Fig. 2.

Examples of clinical trial design, using a control arm, two experimental therapies (A, B) and a third experimental therapy (C) that is discovered after opening of trial. Traditional clinical trial can only investigate one experimental therapy (A) to a control arm at a time. Adaptive platform trials allow investigation of therapy A and B with a common control arm, and allow the addition of therapy C. Secondly, adaptive platform trials allow pre-specified interim analyses to close an ineffective therapy (B) early. Bayesian adaptive platform trials can allow biomarker-adjusted enrollment of each arm, allowing greater enrollment of patients with a specific “biomarker signature” as a trial progresses

The Individualized Screening Trial of Innovative Glioblastoma Therapy (INSIGhT) is an example of an ongoing Bayesian adaptive platform trial for GBM [54]. INSIGhT randomizes patients with IDH wild-type MGMT unmethylated GBM to three experimental arms or a control arm of standard-of-care radiation and temozolomide therapy. Genomic data is required to allow for biomarker grouping to inform adaptive randomization. INSIGhT utilizes Bayesian statistics to regularly assess therapeutic efficacy within biomarker grouping; it can then adjust enrollment of future patients to be targeted to experimental arms based on those biomarker groupings. The trial exemplifies some of the efficiencies allowed by adaptive platform trials: response-based adaptive randomization, a common shared control arm, and adding and dropping of arms. The ability to add experimental arms, with a corresponding increase in the control arm, allows for the trial machinery to accelerate evaluation of therapies relative to a standalone trial [55].

The Glioblastoma Adaptive Global Innovative Learning Environment (GBM AGILE) study is an example of an ongoing international phase 2/3 adaptive platform study with a registrational component [56, 57]. In GBM AGILE, patients are first randomized to one of many experimental arms versus a control arm using a central registration to allow rapid assessment of multiple therapies. As results of clinical data become available, experimental arms are weighed based on predicted success of therapies. This should allow a more targeted enrollment to experimental arms and higher chance of detecting meaningful efficacy. Experimental arms that show benefit in the initial adaptive stage can seamlessly “graduate” to a confirmatory stage, using a fixed randomization of patients with relevant biomarker signature in comparison with a control arm. Similar to INSIGhT, the AGILE trial allows for arms to be dropped and added during trial conduct, reducing potential downtimes. GBM AGILE is also able to consider future predictive biomarkers within their simulations and has built-in flexibility with adaptive randomization of currently unknown biomarker subgroupings.

Adaptive platform trials such as GBM AGILE and INSIGhT have built-in randomization with a shared control arm; in contrast to an umbrella trial such as N2M2, this may lead to rare biomarker-positive patients being randomized to a control arm rather than an experimental therapy.

Externally Augmented Clinical Trial Designs

Externally augmented clinical trials employ external data for decision making during the study or final analysis [55, 58]. External control arms consist of clinically annotated patient pre-treatment data and outcomes of previously completed studies [55]. An external control arm can be leveraged in early phase or late phase settings. In the early phase setting, an external control arm can supplement a single-arm study as a comparator arm. After adjustment of covariates, a treatment effect can be generated for the experimental therapy, an approach that may be superior to the conventional use of historical control benchmark estimates [58]. Another type of externally augmented clinical trial design is a hybrid randomized design, which integrates randomization to an internal control arm that is supplemented by an external control arm. In a hybrid design, an external control arm supplements an existing internal control arm to reduce the number of patients randomized to the control arm [59]. A hybrid randomized design with an external control arm can be considered for early phase or later phase trials, and a hybrid approach is planned to be used for the evaluation of MDNA55 in recurrent GBM in a registrational trial [60, 61]. The largest challenges of implementing externally augmented clinical trial designs are access to high-quality data and potential risks of bias (e.g., the effect of unmeasured confounders) with external data [62]. The evaluation of the use of externally augmented clinical trial designs remains an area of active investigation.

As investigators design trials in neuro-oncology, they must balance the advantages and disadvantages of different potential trial designs, and subsequent results must be interpreted within the respective limitations of each design. Master protocol trials represent an important step towards more efficiently identifying and evaluating therapies to advance the field.

Endpoints

Overall survival remains the gold standard endpoint for clinical trials, but this can be challenging in some clinical contexts. Given its ability to be ascertained sooner and prior to effects from subsequent salvage therapies, progression-free survival (PFS) is a very commonly used endpoint that has correlated with overall survival in GBM [63–66]. The use of non-survival endpoints must be used with caution, however, as several GBM trials have shown a prolongation of PFS that did not correlate with improved OS [67, 68]. As novel therapies may have unknown or complex relationships between PFS and OS, rigorous validation is warranted for appropriate application in clinical trials.

Response Assessment in Neuro-Oncology (RANO) criteria was established to improve on traditional imaging-based endpoints of tumor response in the context of advancements in imaging, specifically geared for clinical trials [69]. Historical response and progression criteria of high-grade gliomas were based upon two-dimensional measurements on imaging over time (Macdonald Criteria); this, however, failed to account for pseudoprogression, or non-enhancing recurrence on antiangiogenic agents [70]. While contrast enhancement has been used as a surrogate of brain tumor burden for decades, enhancement can be influenced by radiologic technique, seizure activity, treatment-related change, among other confounders. The RANO criteria for high-grade gliomas integrated changes to account for specific entities such as pseudoprogression, pseudoresponse, and T2/FLAIR non-enhancing components of disease, in addition to steroid use and clinical status [71]. As different CNS tumors can have unique response and assessments, RANO criteria have been established for many settings beyond high grade gliomas [72–74]. Modified RANO addressed limitations of the initial RANO criteria by allowing volumetric response evaluation, designating post-radiation scans as baseline for evaluation, and eliminating consideration of T2/FLAIR non-enhancing components of disease [75].

Given increasing testing of immunotherapy agents, immunotherapy RANO (iRANO) criteria have also been proposed to address specific considerations for immunotherapy-based trials in neuro-oncology [76]. Ongoing efforts include examination of the use of liquid biopsies, artificial intelligence, positron emission tomography, defining surgical resections and applications to leptomeningeal disease, meningiomas, pediatric neuro-oncology, and spinal tumors [77]. The RANO efforts have provided objective scales to provide standardization and address many challenges in neuro-oncology trials, though validation will be important for each of these initiatives moving forward.

Rate of Growth

To capture benefit of a therapy, there has been recent interest in the use of “rate of growth” of tumors before and after treatment, e.g., a volumetric change of a tumor over time per imaging scans. One multicenter retrospective study investigated growth trajectories in IDH-mutant lower-grade glioma patients who underwent standard treatment and had at least 3 MRI scans over a period of 6 months prior to treatment, as well as MRI scans after treatment [78]. They found that tumor growth rates can decrease after chemotherapy or radiation, in comparison to growth rates prior to treatment. A study in GBM patients has also shown possibility of stratifying patients for clinical outcomes with use of tumor growth kinetics [79]. Further studies will be needed to determine if rate of growth correlates with endpoints such as OS.

Rate of growth has been of interest in the evaluation of targeted therapies for IDH-mutant gliomas. For evaluation of ivosidenib, an IDH inhibitor, pretreatment growth trajectory was compared to posttreatment growth rate [27]; this showed benefit of therapy without meeting traditional response criteria. Likewise, vorasidenib is an IDH1/IDH2 inhibitor that has been shown to reduce growth rates in early phase testing [80]. INDIGO, an ongoing phase 3 RCT investigating vorasidenib, is one of the largest studies to date that will utilize tumor growth rate as a secondary endpoint [81].

Endpoints such as rate of growth have been proposed to measure therapeutic activity and to provide insight into clinical effect of a particular therapy that may not be captured by traditional endpoints [82]. Moving forward, there will be a need to standardize methodology for such endpoints in early phase trials, including definitions for minimum growth rate and clear requirements for assessment of pre-treatment growth and need for baseline scans that are closer to the date of starting a new experimental therapy [83]. Future directions could also include consideration of artificial intelligence for three-dimensional volumetric response of tumor [84], which would facilitate the use of endpoints such as rate of growth.

Neurocognitive/Patient-Reported Outcomes

In addition to radiographic changes, therapies can enable patients to achieve meaningful subjective responses, such as symptom relief or return of independent function. These clinical outcomes, which include neurocognitive function, symptom burden, quality of life, among other endpoints, represent a growing field of research that has been increasingly utilized as clinical trial endpoints. Furthermore, as studies have shown that clinicians fail to reliably detect symptoms and severity, patient-reported outcomes have been increasingly utilized in clinical trials [85].

A representative example of such clinical outcomes as endpoint is NRG CC001, a multi-institutional phase 3 RCT that randomized patients with brain metastases to whole brain radiation therapy (WBRT) vs. hippocampal avoidance WBRT (HA-WBRT) [86], with the hypothesis that lower RT dose to the hippocampus can mitigate neurocognitive effects of therapy. The primary endpoint was cognitive failure [87, 88], and the study met its primary endpoint of lower risk of cognitive failure with HA-WBRT.

Due to heterogeneity of symptoms that may impact patient experiences, several validated PRO measures (PROMs) have been utilized to capture meaningful outcomes. The M.D. Anderson Symptom Inventory-Brian Tumor Module (MDASI-BT) is one example of a validated PROMs in neuro-oncology that assesses 22 domains of symptom burden and disease interference in patients with brain tumors [89]. The Neurologic Assessment in Neuro-Oncology (NANO) scale was created to objectively assess neurologic function as a clinician-related outcome, consisting of 9 neurologic domains of direct observation and testing conducted during routine office visits [74]. Recent trials have employed MDASI-BT and PRO measures as valuable secondary endpoints. In RTOG 0825, a phase III randomized clinical trial of newly diagnosed GBM patients randomized to bevacizumab or placebo, worse baseline PROs were associated with worse OS and PFS independent of methylation status, RPA class, or neurologic factor [90]. This suggests that PROs can correlate with traditional endpoints and could represent an appropriate outcome to assess benefit of new therapies.

Conclusion

Several novel strategies are being investigated to increase efficiency and fidelity of clinical trials. The many unique challenges of therapeutic development in neuro-oncology have motivated investigators to identify endpoints that can measure meaningful benefits for patients, to maximize patient enrollment, to decrease therapeutic development timelines, and to leverage biostatistical methods to try to accelerate drug development. Strategies discussed here, including master protocol trials, response-based adaptive randomization, and platform trial infrastructure, address shortcomings of traditional clinical trial designs. Several large ongoing studies have adapted many of these novel strategies. While we wait for results from such ongoing trials, rapid accrual and efficient conduct of these studies suggests that many of these strategies may become more prevalent in the pursuit of novel therapeutics for central nervous system tumor patients.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

Lorenzo Trippa has been supported by the National Institutes of Health (NIH Grant 1R01LM013352-01A1).