Current drug development and trial designs in neuro-oncology: report from the first American Society of Clinical Oncology and Society for Neuro-Oncology Clinical Trials Conference

Abstract

Successful drug development for people with cancers of the CNS has been challenging. There are multiple barriers to successful drug development including biological factors, rarity of the disease, and ineffective use of clinical trials. Based upon a series of presentations at the First Central Nervous System Clinical Trials Conference hosted by the American Society of Clinical Oncology and the Society for Neuro-Oncology, we provide an overview on drug development and novel trial designs in neuro-oncology. This Review discusses the challenges of therapeutic development in neuro-oncology and proposes strategies to improve the drug discovery process by enriching the pipeline of promising therapies, optimising trial design, incorporating biomarkers, using external data, and maximising efficacy and reproducibility of clinical trials.

Introduction

Treatments of cancers of the CNS have not advanced rapidly in the last two decades.^1,2 Barriers to making therapeutic advances include biological hurdles such as the blood–brain barrier, tumour heterogeneity, and the immunologically privileged environment in the brain.^3–5 In addition to a limited pipeline of promising new therapies, there have been challenges in optimising clinical trial design.^4,6,7 The traditional drug development framework is expensive, slow, and difficult to execute; the inefficiency of the current system is also reflected in the low number of regulatory approvals for new types of treatments of CNS tumours in contrast to treatments of other tumour types.⁴ Furthermore, there is low patient participation and engagement, and inadequate dissemination of trial results.^8,9 Poor filtering of ineffective treatments also continues to be a concern, as many late-phase trials in the last two decades did not find effective treatments.⁸

To address the unmet needs in new drug development in neuro-oncology, two leading organisations, the American Society of Clinical Oncology (ASCO) and the Society for Neuro-Oncology (SNO) commenced a collaborative annual conference to help better define these barriers and facilitate discourse among stakeholders to outline a road map for improving translation of discoveries into effective therapies. Participants in the inaugural 2021 ASCO and SNO Clinical Trials conference included scientists, representatives from the US Food and Drug Administration (FDA) and pharmaceutical industry, patient advocacy groups, and clinical trialists from both within and outside the field of neuro-oncology. Innovations discussed at the conference included trial design, biomarker development, and leveraging of external data to improve and accelerate drug development for patients with CNS tumours. The discussion from the ASCO and SNO Clinical Trials Conference is used as a framework for this Review.

The gap between molecular advances and clinical advances

Molecular advances have occurred at a fast pace for CNS tumours but clinical advances have not followed suit. Without evidence of blood–brain barrier penetration, or target and pathway modulation, a therapy will be unlikely to be effective when tested on people with CNS tumours. Better pharmacodynamic testing, window of opportunity trials, and non-invasive molecular imaging might help address this challenge. Many therapies shown to be effective in preclinical models have not translated to improvements in patients; or, alternatively, the therapies currently being tested are only marginally effective. Improved models and enrichment of existing pipelines are desperately needed to successfully translate molecular advances to therapies that lead to improved outcomes. Innovative trial designs represent an important advancement in the field. Also, concerted and intentional efforts for stakeholders to align on motivation and direction with cooperative groups, professional societies, industry, regulatory agencies, and patient advocates provide an avenue to share lessons on previous drug trials that were unsuccessful and better identify opportunities to close the gap between molecular and clinical advances for CNS tumours.

Precision oncology clinical trial designs that integrate biomarker evaluation

In the era of precision medicine, drug development increasingly relies on identification of biomarkers that can predict the prognosis of patients or aid the selection of specific therapies. When a biomarker is used to predict the disease course of a patient (regardless of treatment received), it is referred to as a prognostic biomarker. In glioblastoma, the most well known prognostic biomarker is MGMT (O6-methylguanine–DNA methyltransferase) promoter, which has been validated as a strong prognostic biomarker;¹⁰ when present in the tumour (roughly in 30–40%), it is expected to confer a 50% improvement in median overall survival.

Some biomarkers are used to identify patients who are more or less likely to benefit from a particular therapy. These biomarkers are classified as predictive biomarkers.^11,12 For glioblastoma, MGMT promoter methylation is an established predictive biomarker. Patients with methylated MGMT are expected to derive greater benefit from temozolomide and radiotherapy than radiotherapy alone, compared with patients with unmethylated MGMT.¹⁰ In low grade gliomas, the chromosome 1p/19q codeletion status has also been verified in clinical trials as a predictive biomarker for guiding the use of procarbazine, lomustine, and vincristine for clinical benefit.^13–15 For the purpose of clinical decision making, predictive biomarkers are often dichotomised as positive or negative. The term biomarker-negative might refer to patients who are not good candidates for the targeted therapy, whereas biomarker-positive might refer to patients who are expected to benefit from receiving the alternative targeted therapy. Examples of predictive biomarkers include 1p/19q codeletion status,^14,16 MGMT promoter methylation status,¹⁰ BRAFV600E mutations,¹⁷ and TRK fusions¹⁸ in CNS tumours.

Clinical trials that evaluate targeted therapies often incorporate a predictive biomarker that might differentiate participants into a subgroup that benefits from targeted therapy (biomarker-positive) from the remaining patients for whom efficiency is not sufficiently strong for the therapy to be indicated (biomarker-negative).^19–22 Here, we review phase 3 clinical trial designs that incorporate a predictive biomarker into their design or analysis. We assume that at the time of the study, the assay for the biomarker has shown satisfactory performance and that a cutoff has been appropriately selected to yield a binary biomarker status (ie, biomarker-negative or biomarker-positive). This expectation is reasonable for a biomarker assay that will be deployed in a confirmatory phase 3 setting. We discuss the advantages and limitations of each design from both statistical and practical perspectives.

Enrichment design

When a targeted therapy might only benefit patients with a particular biomarker, trial eligibility might be restricted only to biomarker-positive patients. Eligible patients are randomly assigned to either the standard therapy or the targeted therapy. This trial design is referred to as an enrichment design (figure 1A). For example, the CODEL trial (NCT00887146) is a phase 3 enrichment trial to compare survival after radiotherapy alone versus radiotherapy with concurrent and adjuvant temozolomide in patients with newly diagnosed chromosome 1p/19q co-deleted anaplastic oligodendrogliomas. Chromosome 1p/19q codeletion serves as a screening biomarker to enrich the study cohort. If the biomarker is truly predictive, the enrichment design avoids treating patients who are unlikely to benefit from the targeted therapy. Since the biomarker is used for screening, it guarantees that biomarker status is known for all of the patients in the trial at study entry. Enrichment trials are efficient because the sample sizes are typically smaller than in trials without the biomarker screen. The sample size reduction results from the relatively large treatment effect expected in patients who are biomarker-positive.

Figure 1: Schematics of biomarker-based trial designs.

(A) Enrichment design, (B) all-comers design, and (C) biomarker-stratified design.

Enrichment designs have limitations, as they require a biomarker assay up front, which might be costly and inconvenient. Furthermore, if the prevalence of biomarker-positivity is low, many patients will need to be screened to identify biomarker-positive patients for enrolment, resulting in slow accrual. Finally, a positive enrichment trial provides no information about benefit among biomarker-negative patients. If the targeted therapy also benefits patients who are biomarker-negative, the trial result might limit the patient population indicated for the new treatment since the trial will have only tested the therapy in patients who are biomarker-positive.

All-comers design

In some situations, ascertainment of biomarker status before randomisation might not be feasible (eg, due to no available biomarker assay). An alternative approach would be to require specimen submission at enrolment with a plan to assess the predictive biomarker at trial completion. This method is referred to as the all-comers design. Eligible patients are randomly assigned to receive standard therapy or the experimental therapy. The trial is sized to detect the overall treatment effect regardless of biomarker status. Treatment effect is then estimated for biomarker-positive and biomarker-negative patients, separately (figure 1B). Evaluation of the predictive ability of a biomarker requires testing for the statistical interaction between treatment assignment and biomarker status.¹¹

In this setting, although the original trial is done prospectively, the predictive biomarker study is retrospective since each patient’s outcome is determined before analysis. To distinguish such biomarker studies (embedded within a treatment trial) from the non-experimental observational biomarker studies, Simon and colleagues²³ classified this design as prospective–retrospective and proposed conditions for conducting a rigorous prospective–retrospective study. Possibly due to the paucity of promising predictive biomarkers in neuro-oncology, most late-phase brain tumour trials have used the all-comers design.^10,24

A notable logistic advantage of the all-comers design is that biomarker assessment does not need to be done in real-time. However, since not all patients will have biomarker results available (eg, due to insufficient specimens or assay failures), concerns might arise regarding generalisability of results, particularly if patients who have biomarkers ascertained differ from the remaining trial patients in some important way. For example, in a study by Hegi and colleagues,¹⁰ which examined the predictive value of MGMT methylation status in the pivotal temozolomide efficacy trial, the proportion of patients who had debulking surgery was higher among patients who had MGMT methylation than among those who did not, casting doubt on the predictive ability of MGMT in the overall trial population.¹⁰ Retrospective subgroup analysis of the treatment effect often has poor statistical power, resulting in inconclusive evidence regarding treatment benefit within biomarker subgroups. Rigorous and realistic power estimation should be devised before embarking on a prospective–retrospective predictive biomarker study in an all-comer trial; existing methods proposed by several investigators are useful for this particular purpose.^25,26

Biomarker-stratified design

When strong evidence exists that targeted therapy is likely to benefit patients that are biomarker-positive but a potential treatment benefit cannot be confidently ruled out for patients that are biomarker-negative, a suitable design would be to use a biomarker-stratified design to randomly assign patients to the standard therapy and targeted therapy in the two biomarker subgroups, separately (figure 1C). Here, the biomarker serves as a randomisation stratification factor in the randomised controlled trial (RCT). The trial is powered to detect some prespecified treatment effects in the two biomarker subgroups. The biomarker-stratified design provides rigorous evidence of benefit across biomarker subgroups to inform treatment decisions. Like the enrichment design, biomarker status is required for all trial patients before randomisation. However, this design had the same limitation as the enrichment design in that the biomarker assay required for randomisation stratification might be costly and adds additional logistic burden to the trial. Also, if the biomarker prevalence (positive or negative) is low, the trial will need to screen many patients to fulfil accrual requirement for specific biomarker subgroups.

A biomarker-stratified design requires a large sample size because its structure is the same as two parallel RCTs in the two biomarker-defined subgroups. Since this design involves testing for treatment effect in both biomarker subgroups, the study-wide probability of recommending an ineffective therapy to any biomarker subgroup (type I error) might be increased. As such, some form of multiple testing adjustment is required, further increasing the sample size.²² Possibly due to this practical sample size constraint (and the paucity of predictive biomarkers), the biomarker stratified design has been infrequently used in brain tumour trials.

In summary, many anti-cancer agents might only benefit a subset of patients defined by a molecular biomarker. We review features of designs that incorporate a predictive biomarker into a phase 3 RCT (table 1). To contrast efficiency of these varying trial designs, we report a hypothetical example of a glioblastoma trial (appendix pp 1–2) to highlight tradeoffs for each design. We note that most appropriate phase 3 trial designs should be guided by the level of evidence in the biomarker’s ability to differentiate patients who benefit from the therapy. In general, there is no universal strategy; each design has its advantages and limitations. In choosing a phase 3 design, care should be taken to account for statistical and practical implications.

Table 1:

Features of various phase 3 trial designs incorporating a predictive biomarker

	Pros	Cons
Enrichment design	Simple and efficient—typically assumes a substantial treatment effect, resulting in a small sample size; biomarker status ascertained and known at study entry	Cost and inconvenience of real-time biomarker screening—some patients might not wish to undergo biomarker testing and wait for assay results; if biomarker prevalence is low, many patients will need to be screened, which will lead to slow accrual and a lengthy trial; positive trial results provide no information of treatment benefit in patients with biomarker-negative status; this might unnecessarily limit the patient population indicated for the new therapy
All-comers design	No need for real-time biomarker assay results at study entry; straightforward RCT—biomarker is not built into the treatment trial design	Not all trial patients will have biomarker results (eg, due to assay failures, insufficient tissue quantities); generalisability in question if patients who have biomarker status ascertained differ from the rest of the patients; poor statistical power in retrospective analysis of treatment effect in biomarker subgroups leading to inconclusive evidence regarding treatment benefit
Biomarker-Stratified design	Biomarker status ascertained and known at study entry; most rigorous and reliable evaluation of treatment benefit within two biomarker subgroups	Cost and inconvenience of real-time biomarker screening—some patients might not wish to undergo biomarker testing and wait for assay results; if prevalence of biomarker-positivity is low in either biomarker subgroup, many patients will need to be screened to identify enough patients; this might lead to slow accrual and a lengthy trial; requires large sample size as its structure resembles two parallel RCTs in biomarker-positive and biomarker-negative subgroups

RCT=Randomised controlled trial.

Basket, umbrella, and platform trials

To address inefficiencies in drug development for CNS tumours, an urgent need exists to increase the speed of testing therapies and decision making on whether a therapy should be advanced or not, which can in turn expedite the approval of new therapies. Such a change would shorten drug development, limit participant exposure to potentially ineffective drugs, foster precision medicine approaches, and lower drug development costs. One strategy that can address goals are master protocol trials. These are designs that test multiple therapies either individually or in combination in one or multiple diseases in parallel under a single overarching protocol and expedite assessment of outcomes. They include basket trials, umbrella trials, and platform trials, among others (table 2).^22,27,28

Table 2:

Basket, umbrella, and platform trials in neuro-oncology

	Trial	Experimental agents	Phase	Disease setting	Primary endpoint	Control group
Basket trials
NCT02034110	ROAR	Dabrafenib and trametinib	2	BRAFV600E-mutant low-grade and high-grade glioma	Objective response rate	No
NCT01524978	VE-BASKET	Vemurafenib	2	BRAFV600E mutant tumours (gliomas were included in cohort of other solid tumours)	Best overall response rate	No
NCT02637687	SCOUT	Larotrectinib	1/2	NTRK-mutant tumour in paediatric patients with advanced solid or CNS tumour	Phase 1: safety; phase 2: objective response rate	No
NCT02576431	NAVIGATE	Larotrectinib	2	NTRK-mutant tumour (primary CNS tumour cohort)	Best overall response rate	No
NCT04374305	ITUITT-NF2	Neratinib, brigatinib	2	Neuro-fibromatosis-2 associated CNS tumours	Radiographic response rate	No
NCT03173950	NCI-CONNECT	Nivolumab	2	Rare CNS tumour (prespecified rare diagnoses)	Objective response and progression-free survival	No
NCT04579380	SGNTUC-019	Tucatinib and trastuzumab	2	Previously treated solid tumour with HER2 alterations (including primary brain tumours)	Confirmed objective response rate	No
NCT02628067	KEYNOTE-158	Pembrolizumab	2	Microsatellite instability high or deficient DNA mismatch repair tumours	Objective response rate	No
NCT00918320	TOTEM2	Temozolomide and topotecan hydrochloride	2	Children with refractory or relapsed extracranial solid and CNS tumours	Objective response rate	No
Umbrella trials
NCT02523014	Alliance A071401	Vismodegib, FAK inhibitor GSK2256098, capivasertib, abemaciclib	2	Meningioma with mutation in SMO, NF2, NF2/CDK, or AKT matched to relevant inhibitor	6-month progression-free survival and response rate	No
NCT03994796	Alliance A071701	Abemaciclib, GDC0084, entrectinib, MRTX849	2	Brain metastases with known CDK, PI3K, or NTRK/ROS1 mutations	CNS response rate	No
NCT03158389	NCT Neuro Master Match N2M2 (NOA-20)	APG101, alectinib, idasanutlin, atezolizumab, vismodegib, palbociclib, temsirolimus	1/2	Newly diagnosed GBM, MGMT unmethylated	6-month progression-free survival	No
Adaptive platform trials
NCT03970447	GBM AGILE	Regorafenib; paxalisib; VAL-083; VT1021; troriluzole	2/3	Newly diagnosed GBM, MGMT unmethylated; newly diagnosed GBM, MGMT methylated; recurrent GBM	Overall survival	Yes, shared control arm
NCT02977780	GBM INSIGhT	Abemaciclib; CC-115; neratinib	2	Newly diagnosed GBM, MGMT unmethylated	Overall survival	Yes, shared control arm
ACTRN12620000048987*	MAGMA	Neoadjuvant chemotherapy or extended chemotherapy	2	Newly diagnosed GBM	Overall survival	Yes, shared control arm
NCT05009992	DMG-ACT	ONC201, paxalisib	2	Children and young adults with diffuse midline gliomas	6-month progression-free survival and overall survival at 7 months	No

GBM=Glioblastoma.

^*Platform trial with a plan for future transition to multiarm adaptive trial design.

Basket trials

Basket trials are master protocols designed to evaluate a single investigational drug or drug combinations in different disease populations defined by a common characteristic such as a genetic biomarker. The advantage of this design is that it allows access to targeted agents across tumour types, including CNS tumours, for which specific trials would be challenging to design. Examples of basket trials in neuro-oncology include the Rare Oncology Agnostic Research (ROAR) trial,¹⁷ testing dabrafenib and trametinib in patients with low-grade and high-grade gliomas, which reported a 33% objective response in patients with high-grade glioma and a 69% objective response among low grade gliomas. Other examples are the VE-BASKET²⁹ trial that tested the use of vemurafenib in non-melanoma tumours with a V600E mutation including gliomas, and NAVIGATE and SCOUT trials which allowed enrolment of NTRK fusion positive CNS tumours and showed a benefit of larotrectinib for CNS tumours.³⁰ These basket trials included different CNS tumours; designing individual trials for all these rare tumour types would have required more effort and resources, which could make their launch unrealistic. Basket trials have already led to accelerated regulatory approvals for indications with implications in neuro-oncology, including the ROAR trial, which led to FDA approval of dabrafenib and trametinib for tumours with BRAF mutations.^31,32

Although there are several important benefits of basket trials, a few limitations should be considered. Broadened patient eligibility criteria (eg, inclusion of paediatric and adult patients) can complicate the interpretation of the trial results, and there can be differences in responses across populations. Another limitation is the drug’s poor blood–brain barrier penetration, which would be expected to predominantly affect CNS tumours but not other tumour types. Studies on patients with CNS tumours also have unique features, such as neurocognitive testing, which might not be adequately addressed in basket trials due to the inclusion of broader cancer populations. Nevertheless, early phase trials might serve as a platform to collect preliminary data on neurological or neurocognitive endpoints, the findings of which might be further corroborated or validated in subsequent confirmatory RCTs.

Umbrella trials

Umbrella trials are master protocols designed to evaluate multiple drugs or drug combinations in treatment of a single disease. These trials usually include screening for the presence of one or more molecular targets, and possible patient assignment to different study groups. This design can prove very useful for the development of effective precision medicine approaches in neuro-oncology given that CNS tumours are frequently a constellation of many rare diseases. For example, work in meningiomas, the most common primary brain tumour, has shown that these tumours consist of genetic subtypes driven by genetic alterations including SMO mutations (5–10%), AKT pathway alterations (10–15%), and NF2 inactivation (15%).^33,34 To match the predominant molecular alteration in recurrent meningiomas with a targeted treatment, the Alliance for Clinical Trials in Oncology study A071401 was developed (NCT02523014). Each arm has a two-stage design that can further expedite the process via the rapid elimination of ineffective drugs at the end of the first stage. One of the study arms testing the use of the FAK inhibitor GSK2256098, which causes synthetic lethality in NF2-inactivated tumours, has already completed its initial phase. This arm met its primary endpoint of progression-free survival, both for grade 1 and grade 2/3 tumours, more than doubling the expected progression-free survival rates³⁵ as previously reported in the landmark response assessment in neuro-oncology (RANO)³⁶ systemic overview of medical therapies in meningiomas.

Another example is the Alliance biomarker-driven trial in brain metastases (A071701, NCT03994796), which is based on the discovery that brain metastases from solid tumours are frequently driven by different alterations as compared with the extracranial primary or metastatic disease.³⁷ A071701 has four treatment arms: abemaciclib, the PI3K inhibitor GDC0084, the KRAS G12C inhibitor MRTX849, and the NTRK/ROS inhibitor entrectinib (patients with lung cancer). In MGMT unmethylated glioblastoma, the NCT Neuro Master Match (N²M²) trial³⁸ also uses an umbrella trial approach to evaluate several targeted therapies on the basis of biomarker evaluation.

Platform trials

Platform trials have characteristics of both the basket and umbrella trials and they test multiple therapies in the context of a specific disease, but in an ongoing manner. Therapies enter or leave the platform on the basis of a predetermined decision algorithm. An example of such a trial in neuro-oncology is the GBM AGILE trial (NCT03970447),^39,40 a seamless phase 2/3 adaptive randomisation platform trial, which includes arms for newly diagnosed and recurrent glioblastoma. There is a single control arm for newly diagnosed and one for recurrent disease. The study was designed so that experimental arms can be added or graduated. Furthermore, the randomisation ratio is adapted on the basis of early outcome data. There is also a provision for biomarker-enriched arms, if such biomarkers are identified when targeted agents are introduced into the trial.

Another example of a platform trial is the Individualied Screening Trial of Innovative Glioblastoma Therapy (known as INSIGhT; NCT02977780) trial in glioblastoma. This phase 2 platform trial uses response-adaptive randomisation and deep genomic profiling in MGMT unmethylated glioblastoma to accelerate the identification of therapies for further testing. The study has a common control arm (temozolomide) and three experimental arms (abemaciclib, CC-115, and neratinib). During the study, the randomisation probability to CC-115 decreased on the basis of poor early progression-free survival results and the arm group closed due to toxicity concerns.⁴¹ Overall, this adaptive platform design facilitated more efficient and economical testing of multiple new agents for this indication.⁴²

An important consideration for platform trials is the selection of an appropriate control arm group. For early-phase trials, including some umbrella and basket trials, an appropriate control arm might not exist, be impractical, or be unethical. For example, in Alliance A071401, there is no universally agreed upon standard of care treatment for recurrent meningiomas after surgery and radiotherapy. Control arms are important for confirmatory platform trials. If a platform trial continues over a period of many years, the standard of care might change, and so the control would possibly need to be replaced. Additional issues include type I error considerations that needs to be controlled at the overall study-level, data sharing, transparency, and criteria for release of patient-level data.

Advantages of master protocols include the establishment of common screening platforms, which can result in shorter screening time and a lower number of patients screened who are deemed ineligible. Also, because of centralisation and shared governance, these studies tend to create experienced clinical trial networks with optimised coordination and oversight. These kinds of studies can increase the speed of drug development because common protocol elements do not need to be reinvented every time a group is added. These trials, especially the platform trials, mandate upfront longitudinal modelling and can allow flexibility of objectives. Decision rules and treatment effects of various experimental therapies are often evaluated by considering the outcome data across molecularly defined biomarker subgroups.²² For example, the GBM Agile trial was designed to include multiple stages of drug development from early phase to phase 3 evaluation preceding regulatory approval, thus facilitating seamless sharing of safety and efficacy data across different stages of disease and different drug development phases. This sharing of safety and efficacy data across the study would not be possible with any of the conventional designs. Furthermore, stakeholder and patient engagement can be increased given the multiple options for engagement. Challenges include complexity in design, longer preparation periods, logistical challenges such as coordinating data elements across different study sites, contracting with multiple industry partners, and complex statistical analysis and interpretation.

In summary, emergence of master protocols is an improvement in the drug development framework. Master protocols increase efficiency and shorten drug development timelines if multiple experimental treatments and multiple diseases are being tested. However, if a clinical development programme has a defined focus on a compound or combination of specific therapies, a master protocol might not be the most efficient approach. Optimal design choice and implementation require increased planning efforts, but this investment is justified by this type of programmes’ potential to expedite drug development.

External data for trial design and analysis

An important consideration in early phase trials is whether to use randomisation and include a control group. Single arm trials (SATs) have been commonly used in neuro-oncology but can be prone to unreliable prediction of treatment efficacy.⁴³ SATs in glioblastoma can inflate type I error rates and yield biased treatment effect estimates.^44,45 In contrast to SATs, RCTs have long represented the gold standard design for establishing treatment efficacy, particularly in late phase testing. Since randomisation accounts for known and unknown confounders, RCTs can provide more reliable treatment effect estimates than SATs.⁴⁶ Simulation studies in the early-phase glioblastoma trial context suggest that RCTs are superior to SAT designs without requiring large increases in sample size.⁴⁷ Although advantageous in some respects, RCTs are generally more expensive, require a longer time investment to complete, and can be more challenging for patient recruitment than are SATs.^48,49 Patients can be reluctant to enrol on a trial in which they might be randomly assigned to a standard-of-care therapy. This reluctance is particularly salient in neuro-oncology settings in which standard-of-care therapies result in poor outcomes. Patients might be more likely to withdraw from a trial after being randomly assigned to a control group in such disease settings, as was seen in the evaluation of nivolumab in recurrent glioblastoma in CheckMate 143;⁵⁰ enrolment was increased by 120 patients to compensate for patient withdrawals from the control group of the trial.

In seeking a compromise between the tradeoffs and risks of SAT and RCT designs, there has been an interest in trials using external data,^51,52 termed externally augmented clinical trial (EACT) designs. EACT designs have received increasing attention in neuro-oncology to improve trial efficiency.^53,54

Externally controlled single arm trials

A straightforward application of an EACT design would be to create an external control arm consisting of patient-level data with relevant covariates and endpoints for a representative cohort that might serve as a comparator for an externally controlled SAT (figure 2A). With this design, the external control arm can contextualise results and provide an estimate of treatment effect of a therapy after adjustment for relevant clinical covariates.⁵⁵ This design should be distinguished from historical control comparisons, which use crude benchmark values without adjustment. Statistical methods that can be used for these comparative analyses can be drawn from causal inference techniques (eg, propensity score matching).^56,57 Initial simulation-based research in newly diagnosed glioblastoma has shown that externally controlled SAT designs can reduce false positive rates relative to conventional SAT designs.⁵⁸

Figure 2: Schematic of various externally augmented clinical trial designs.

(A) Single arm trials with an external control arm allows for contextualisation of results. (B) Single-stage and (C) two-stage hybrid designs, incorporating an internal and external control. RCT=randomised controlled trial.

A completed prospective trial using an externally controlled SAT in neuro-oncology is the phase 2b trial in recurrent glioblastoma (NCT02858895), in which patients received convection-enhanced delivery of MDNA55, a fusion protein with IL-4 linked to Pseudomonoas aeruginosa exotoxin A. The overall survival was evaluated for patients that were enrolled compared with a prespecified, matched, external control.⁵⁹ The prespecified analyses used propensity score matching to adjust for pre-treatment profiles across the experimental and external control arms.⁶⁰ Another relevant example is the analysis of a phase 3 trial of autologous dendritic cells pulsed with autologous tumour lysate (DCVax-L), where overall survival results were difficult to interpret as more than 90% of patients crossed over and received the experimental therapy.^61,62 To bolster the survival analysis, contempor aneous external control data was used as a comparator arm to analyse patients treated with DCVax-L (NCT00045968).⁶³ Other examples of external control arms to contextualise SATs have been shown retrospectively in reanalyses of clinical trials in non-small cell lung cancer and early stage hormone receptor-positive breast cancer.^55,64,65

Hybrid EACT and RCT designs

Study designs that incorporate external data into an RCT design are termed hybrid EACT and RCT designs. In such designs, external data can bolster or augment an existing internal control arm of a clinical trial. One example is a trial in which external data directly supplements the internal control arm, so that the randomisation ratio of the experimental arm to control arm can be modified (eg, 3:1). This setup helps reduce the number of trial participants randomly assigned to the control group (figure 2B). Another possibility is a two-stage hybrid EACT and RCT design (figure 2C).⁵³ In this two-stage approach, there is conventional 1:1 randomisation in the first stage of the trial. During an early interim analysis, a comparison between the external control groups and study control group would evaluate the comparability of these cohorts. If the external and internal control groups are similar by prespecified statistical testing, the randomisation ratio could be altered to increase the probability of patients being randomly assigned to the experimental group. If the external and internal control groups were not similar, the study could proceed with 1:1 randomisation. The use of a two-stage hybrid EACT and RCT design is more conservative and flexible, but still ultimately leads to fewer patients enrolled to the control group relative to a conventional RCT.

An EACT and RCT hybrid design is planned for a registrational trial in the evaluation of MDNA55 in recurrent glioblastoma.⁶⁶ Since the trial is incorporating external data, the proposed trial will use 3:1 randomisation to the experimental arm relative to the control arm.⁶⁷ Such hybrid designs are also under active exploration in paediatric oncology.⁶⁸ Although EACT and RCT designs are garnering excitement, there still needs to be continued rigorous study to characterise possible risks with respect to trial operating characteristics with such designs.

Data quality and data sharing

EACT designs require external data that are high quality and appropriate for analytic purposes to minimise the risk of bias. Several factors are important in this consideration, including the source of data (previously completed clinical trial datasets versus real world evidence), patient population, data completeness, precision, contemporaneousness, consistency in measurements, and determination of endpoints.⁶⁹ Ideally, the data should be temporally and clinically representative of the trial population, with consistent and clear definitions and measurements of variables, and well delineated objective outcomes.⁵⁴

Difficulty in obtaining access to appropriate data is often a barrier to methodological research and implementation of EACT designs in prospective trials. Use of data from previously completed trial datasets is appealing, given that they are generally well annotated and often use central pathology and radiology review.⁷⁰ Clinical trial data, however, can be difficult to access and might not be contemporaneous, which could raise concerns for confounding factors that could affect analyses. Real world evidence represents an intriguing alternate source of data to use for EACT designs,^71,72 although data are scarce on whether non-trial patients are similar to patients enrolling onto neuro-oncology trials.^73,74

Future of EACT designs

In the era of precision medicine, there will possibly be trials targeting small biomarker-defined subgroups of patients in which RCTs might be increasingly challenging or impractical. EACT designs might serve as a solution to allow for efficient testing of novel therapeutics in these smaller populations. External data have the potential to inform trial design at various junctures of a clinical trial’s design and conduct including sample size selection, power calculations, and interim analyses.⁷⁵ Data science and biostatistical research will be important to continue to explore and identify the highest yield incorporation of external data into trial design and analysis to increase efficiencies without introduction of excessive bias. Moving forward, commitment from relevant stakeholders, including industry, regulatory bodies, cooperative groups, academics, and patients, will be necessary to reduce the barriers to data sharing, to focus on validation of novel trial designs, and to identify best practices.⁵³

Special considerations for CNS tumour trials

There are special considerations in developing therapies for CNS tumours. Crossing the blood–brain barrier has been a challenge in the development of therapies for brain tumours as it prevents most cancer drugs from reaching the tumour.⁷⁶ Opportunities for drug development include further exploration of nanodrug formulations that are being actively studied for their ability to effectively cross the blood–brain barrier.⁷⁷ Alternatively, small fragment T-cell engagers and nanobodies are also in development.⁷⁸ Another strategy of interest is the disruption of the blood–brain barrier with focused ultrasound to increase the ability of therapies to reach targets in the brain.^79,80

As attempts are made to harness the potential of precision medicine, access and use of next-generation sequencing has increased,⁷³ but most patients are not eligible for genome-targeted therapy.⁸¹ Lack of success in identifying therapeutic targets might be partly due to known intertumoral and intratumoral heterogeneity of CNS tumours. Functional precision medicine could represent a strategy to overcome this barrier with evaluation of treatment efficacies and vulnerabilities of ex-vivo tumour to better personalise treatment to a patient’s tumour.⁸² Tools to individualise therapy on the basis of genomic or functional precision medicine data raise unique operational challenges, especially if a trial aims to evaluate the strategy of predicting the benefit from treatment rather than one specific experimental therapy; clinical trial frameworks to address these unique challenges have been described⁸³ and could be incorporated into existing master protocols to increase the use of functional precision medicine approaches.

In terms of funding, there is a concern of decreased investment into CNS tumour-directed drug development given the financial risks associated with a disease setting that has produced few successes and many failures.⁸⁴ Although there was a positive trend in the number of trials between 2005 and 2015, particularly with more phase 2 trials,⁸⁵ the future of investment flow for CNS tumours is less clear. Moving forward, maximising the use of scarce resources will be important, and trials such as GBM INSIGhT and AGILE provide platforms to test multiple therapies by sharing a common trial infrastructure and thus reducing costs. Commitment from cooperative groups, the National Cancer Institute, and industry partners in addition to exploring novel financial sources might be necessary to ensure that constrained funding sources are maximised. Furthermore, continued emphasis on collaboration is important to align stakeholders and strategise priorities to maximise the chance of success.

Conclusion

There is a need for effective new treatments for CNS tumours. An essential aspect to achieve this goal is to have better clinical trial designs that allow more patients to enrol and expedite outcomes to discontinue development of ineffective therapies and advance effective therapies. We discussed achieving these goals by continuing collaborative efforts to use novel clinical trial designs that use and validate biomarkers, allow multiple drugs or diseases to be assessed simultaneously, or use external data to enrich trial design and expedite the evaluation of therapies.

Search strategy and selection criteria.

We searched the literature using PubMed for papers published in English with the search terms “clinical trial design”, “biomarkers”, “master protocol trials”, “adaptive platform trial”, and “external control arm” in combination with “neuro-oncology”, “glioblastoma”, and “glioma”, from Jan 1, 2020, to Jan 5, 2022. Articles were also identified through searches of the authors’ own files. The reference list for assessment was generated on the basis of relevance to the scope of this Review.