Accelerating discovery: Transformative clinical trial models in neuro-oncology Authors: Amy J Wisdom, Rifaquat Rahman Published: 2026 ABSTRACT AbstractProgress in neuro-oncology drug development has been slower than the pace of biological discovery, and trial design is increasingly recognized as a central contributor to this gap. Traditional trial paradigms, largely developed for systemic malignancies, face distinct challenges when applied to central nervous system (CNS) tumors - among them intratumoral heterogeneity, rapid clinical progression, limited tissue access, and the evolving molecular classification of gliomas. These biological and logistical realities have strained the assumptions underlying conventional phase II and III designs, and there is growing consensus that new approaches are needed. This review examines a framework of emerging trial models that better align clinical testing with the realities of CNS tumors. Master protocol designs enable simultaneous evaluation of multiple therapies within shared, molecularly informed infrastructures, supporting precision medicine approaches and improving efficiency. Bayesian adaptive frameworks allow trials to learn during conduct, reallocating patients toward promising therapies and incorporating emerging biological data in real time. Beyond efficacy assessment, trials can serve as active discovery platforms embedding longitudinal tissue sampling, window-of-opportunity designs, and multi-omic profiling to interrogate CNS drug penetrance, pharmacodynamic target engagement, and treatment-induced tumor evolution directly in human disease. Decentralized trial models and artificial intelligence offer additional tools to broaden access, improve enrollment, and reduce operational burden. Reimagining clinical trials as dynamic, biologically integrated, learning-based systems rather than static tests of individual agents is essential to accelerate therapeutic progress in neuro-oncology.Keywords: Neuro-oncology, Clinical trial design, Drug development Introduction: An Urgent Need to Transform Neuro-Oncology Trials Despite substantial advances in molecular understanding and drug development, outcomes for patients with primary central nervous system (CNS) tumors remain poor. Although both scientific and technological progress across neuro-oncology has dramatically accelerated biological discovery, the majority of investigational therapies have failed to demonstrate clinically meaningful benefit in late-phase trials, highlighting a persistent gap between discovery and therapeutic translation [1,2]. For example, glioblastoma (GBM), the most common primary malignant brain tumor, has not seen significant changes in standard-of-care therapy over the last two decades despite its devastating prognosis [3,4]. Traditional oncology trial paradigms rely on linear phase I through phase III development pipelines that were largely designed for systemic malignancies and perform best when patient eligibility is broad, response assessments are unambiguous, and tumor biology is relatively stable; these conditions often do not apply to primary CNS tumors [5]. High-grade gliomas and other CNS tumors exhibit marked intratumoral heterogeneity, can lead to rapid clinical deterioration, and have narrow windows for trial enrollment, complicating patient selection and trial completion. As a result, despite an increase in phase II trials in neuro-oncology, relatively few have translated into positive phase III results or changes in clinical practice [[6], [7], [8]]. In neuro-oncology, trial design itself has further limited progress (Fig. 1). An overreliance on single-arm phase II studies [2], many of which are underpowered [9], has contributed to the failure to generate practice-changing data [10]. Overly strict eligibility criteria, suboptimal endpoint selection, and ambiguous interpretation of patient responses have further undermined the translatability of many studies [11]. Potential barriers and corresponding design solutions across the clinical trial lifecycle. In response to these challenges, novel trial architectures have emerged that aim to better align clinical testing with the biological and logistical realities of CNS tumors [1,11,12]. While these strategies address important structural limitations, their success will ultimately depend on the development of viable, efficacious therapeutic candidates. In this review, we synthesize emerging clinical trial models in neuro-oncology and present a framework that emphasizes rigorously designed adaptive trials, biologically informed eligibility criteria, clinically meaningful endpoints, and deep integration of translational science. Accelerating discovery in neuro-oncology requires reimagining clinical trials as dynamic, learning-based systems capable of evolving alongside tumor biology and clinical care, rather than as static tests of individual therapeutic agents. Traditional Trial Paradigms Underperform in Neuro-Oncology In neuro-oncology, clinical trial participation remains limited; only approximately 8–11% of patients with newly diagnosed glioblastoma (GBM) enroll in clinical trials, constraining accrual and generalizability [2]. Multiple analyses have demonstrated that promising phase II signals in GBM frequently fail to translate into survival benefit when evaluated in larger, randomized phase III trials. In one representative analysis, only 1 of 8 completed phase III GBM trials reported a positive result, and only a minority of therapies demonstrating activity in phase II ultimately succeeded in phase III, highlighting fundamental inefficiencies in the therapeutic development pipeline [2]. Single arm trials vs. randomization A major contributor to this inefficiency is the continued reliance on single-arm phase II studies that use benchmark historical controls to infer therapeutic activity. These designs are often favored because they are faster, less costly, and more appealing to patients. However, historical comparisons are inherently vulnerable to bias, and randomized designs provide more reliable inference by reducing confounding and separating true therapeutic signal from variation related to patient selection, supportive care, and endpoint assessment [9,10,13]. Moreover, limited incorporation of longitudinal sampling and adaptive learning within conventional trials restricts the ability to account for tumor evolution during enrollment and treatment. The validity of historical benchmark values has been further eroded by evolving tumor classification systems. Successive updates to the World Health Organization (WHO) classification of CNS tumors increasingly define gliomas by molecular features rather than histologic appearance, fundamentally altering disease definitions and prognostic groupings [14]. As a result, benchmarks derived from older cohorts may no longer be biologically comparable to contemporary trial populations, diminishing their relevance as historical control comparators. Neuro-oncology trials face additional challenges that further limit the applicability of traditional paradigms. The restrictive permeability of the blood–brain barrier limits drug delivery and contributes to lack of efficacy for agents when they fail to reach therapeutic concentrations within the CNS. Despite this, early-phase trials often prioritize pharmacokinetic and pharmacodynamic assessments in peripheral blood or extracranial tissue, with limited direct evaluation of CNS drug exposure. Additionally, emerging strategies to improve intracranial delivery of systemic agents by disrupting the blood-brain barrier, including microbubble-enhanced focused ultrasound [15,16], convection-enhanced delivery [17], and nanoparticle approaches [18] hold substantial promise though require further validation. Furthermore, cerebrospinal fluid (CSF) sampling may be helpful though further validation will be necessary [19]. In addition, limited access to serial tumor tissue restricts biological validation and hinders mechanistic understanding of response and resistance. Endpoints Selecting appropriate endpoints in neuro-oncology trials remains a central challenge. Overall survival (OS) is widely regarded as the “gold standard” endpoint; however, it can be confounded by post-progression therapies and treatment crossover, and requires prolonged follow-up, which can sometimes limit feasibility in use in early-phase trials or indications with longer disease courses (e.g. IDH-mutant gliomas) [9]. Efforts to standardize imaging-based assessment, including the development of the Response Assessment in Neuro-Oncology (RANO) have improved consistency in response assessment, and RANO 2.0 criteria were recently updated [[20], [21], [22], [23]]. Progression-free survival (PFS) and PFS at 6 months (PFS-6) are often adopted as surrogate endpoints in glioma trials. While PFS can be informative for cytostatic agents, it may not consistently predict OS across therapeutic classes, especially in trials testing novel mechanisms of action or immunotherapies [24]. Notably, anti-angiogenic agents have shown the possible disconnect between PFS and OS [25,26]. Objective radiographic response rate (ORR), defined by reduction in tumor size, has also been proposed as an endpoint in neuro-oncology [27,28]. While RANO criteria have improved consistency in research settings, the implementation of radiographic assessment of CNS tumors into routine clinical practice remains challenging for numerous reasons. For example, therapies may cause radiographic changes that could confound imaging evaluation, as has been seen with treatment-related effects on vascular permeability rather than tumor burden [22]. Furthermore, inter-observer variability and technical differences in imaging acquisition may limit the generalizability and implementation of these criteria outside of academic institutions, and resource constraints in community centers that may lack dedicated neuroradiology expertise for precise volumetric assessments may hinder implementation of standard criteria into routine practice. The disconnect between validated research criteria and its implementation in real-world clinical practice underscores the need for approaches to response assessment that maintain scientific rigor without compromising feasibility [29]. A target ORR >25% has been suggested for recurrent glioblastoma trials [24], though prospective validation is needed. Minor response (>25%, <50% decrease in a target lesion) has not been fully validated but may serve as a means to illustrate a beneficial response to therapy, and it was used in the FDA approval of tovarafenib [30]. Along with consideration of response categories, defining durability of response remains an important need to be able to better characterize clinically meaningful therapeutic benefit. Further refinements to response assessment may come from broader adoption of contrast-enhanced T1 subtraction maps, rate of growth metrics, volumetrics and advanced imaging modalities may improve response assessment in the future [31]. Advanced imaging modalities, including perfusion MRI, diffusion imaging, spectroscopy, positron emission tomography (PET), and O-(2-[18 F]fluoroethyl)-l-tyrosine (FET) PET, are under active investigation as potential surrogate biomarkers that may better reflect tumor biology than contrast enhancement alone [[32], [33], [34], [35]]. However, these advanced modalities have generally not yet been broadly validated as reliable surrogates for survival benefit across therapeutic classes though ongoing efforts such as RANO-PET are advancing these efforts [36]. Finally, there is growing recognition that endpoints beyond radiographic disease status may better capture clinically meaningful benefits in patients with tumors of the CNS. Measures such as time to neurologic deterioration, steroid-sparing effects, neurocognitive function, and patient-reported quality of life provide critical context and may correlate more closely with patient experience than tumor size alone, but they remain underutilized as primary or co-primary endpoints in most trials [37]. Master Protocols and Novel Trial Architectures Master protocols represent a fundamental shift in clinical trial design by enabling the evaluation of multiple therapies, biomarkers, or strategies within a single overarching trial infrastructure. In contrast to traditional single-agent trials that test whether one drug is efficacious in a narrowly defined population, master protocols increase flexibility by allowing multiple hypotheses to be tested within a single clinical trial [[38], [39], [40]]. As targeted therapies increasingly benefit molecularly defined subsets of patients, master protocols are particularly well suited to support precision medicine approaches. An essential feature of successful neuro-oncology master protocols is the use of broad, biologically informed enrollment criteria. Master protocols can leverage shared trial infrastructure, which could incorporate centralized molecular screening, shared control arms, standardized endpoints, and unified statistical oversight. This shared architecture can reduce redundancy, improve efficiency, and facilitate continuous learning across trial conduct. Importantly, it also enhances patient enrollment by allowing consideration of multiple investigational options within a single protocol and potential for adaptive randomization. Collectively, basket, umbrella, and platform trials represent complementary master protocol designs that address distinct dimensions of molecular heterogeneity in neuro-oncology, each with unique design features and statistical implications. Basket trials enroll patients based on a shared molecular alteration rather than tumor histology, enabling evaluation of targeted therapies across diseases [[41], [42], [43]]. This approach is particularly attractive for patients with rare CNS tumors, where actionable alterations may occur infrequently within any single histology. Basket trials have enabled the study of rare genomic events such as BRAF V600E mutations and microsatellite instability across tumor types [[44], [45], [46], [47]]. The primary advantage of basket trials lies in their efficiency for studying rare alterations without requiring disease-specific trials that would be infeasible due to limited accrual. However, this design also carries important risks. Biological context matters, as target dependency and downstream signaling may differ substantially across tissues and disease states. In CNS tumors, differences in blood–brain barrier permeability, challenges in interpreting radiographic responses, and co-occurring genomic alterations can lead to heterogeneous responses within a single basket, complicating interpretation and regulatory decision making. Umbrella trials evaluate multiple targeted therapies within a single disease entity, stratifying patients based on molecular features. In gliomas, umbrella designs align naturally with the WHO molecular classification framework, enabling prospective assignment of therapies based on IDH status, MGMT promoter methylation, EGFR alterations, or other genomic features. For example, the NOA-20 trial evaluated multiple targeted compounds matched to molecular alterations in patients with newly diagnosed GBM [48,49]. However, umbrella trials also have limitations. Patients are not fully randomized across all interventions, molecular subgroups may be rare, and overlapping biomarkers can complicate patient assignment. Some patients may qualify for multiple arms or none at all, and small subgroup sizes can limit statistical power. Platform trials represent a flexible master protocol design, incorporating continuous enrollment, shared control arms, and the ability to add or drop treatment arms over time. These adaptive trials function as constant learning systems, allowing therapies to enter and exit the platform based on prespecified interim analyses and reducing the time that is required to evaluate each therapy [39,50]. Experimental arms demonstrating clinical benefit during the adaptive stage may transition to confirmatory evaluation, with biomarker-defined populations randomized at fixed ratios to a shared control arm. In neuro-oncology, platform trials such as GBM AGILE (NCT03970447) and INSIGhT (NCT02977780) represent ambitious attempts to implement adaptive platform trial architectures at scale. GBM AGILE integrates Bayesian adaptive randomization, shared control arms, and incorporates a pathway to registration while evaluating multiple agents in both newly diagnosed and recurrent GBM [51]. INSIGhT focuses on biomarker-driven testing in newly diagnosed MGMT-unmethylated GBM, emphasizing efficient screening and early signal detection [52]. However, one must note that, while demonstrating increased operational efficiency, these trials have not yet identified practice-changing therapeutics for glioma patients. These experiences highlight the gap between theoretical efficiency advantages and real-world clinical translation. For example, platform trials may offer substantial efficiency gains but require complex operational coordination, sustained funding, and close regulatory engagement. They also demand rigorous endpoint selection and statistical modeling to evaluate multiple scenarios to avoid biased results or propagation of false-positive signals within a continuously adaptive framework. The sophisticated data infrastructure required may exceed the resources available at many institutions, potentially limiting access and exacerbating disparities. Bayesian methods incorporate prior knowledge, are capable of integrating emerging evidence, and permit flexible decision-making during trial conduct [41,53]. Response-adaptive randomization reallocates patients toward arms with a higher probability of benefit while preserving the ability to learn from less effective treatment arms, increasing the likelihood that participants receive potentially beneficial therapies. Interim analyses enable early stopping for both futility and success, reducing patient exposure to ineffective therapies and accelerating development timelines, a particularly important consideration in neuro-oncology where patient numbers are limited and rapid disease progression is common. Bayesian designs also support biomarker-informed enrollment, allowing adaptive enrichment of molecular subgroups demonstrating early signals of efficacy and enabling trials to evolve alongside advances in biological understanding. However, while response-adaptive randomization is increasingly implemented in platform trials, its statistical validity and efficiency gains remain contested. Response-adaptive randomization introduces vulnerability to temporal trends, and shifts in patient populations, clinical practice, disease severity, or supportive care during enrollment can bias outcomes. The additional statistical and administrative complexity of these designs can also create challenges in trial conduct. Thus, trial design should carefully weigh the possible advantages of master protocol designs against uncertain real-world efficiency gains. When properly designed, Bayesian adaptive trials can achieve comparable or superior power with fewer patients than traditional fixed designs, particularly in settings with strong biological priors or large effect sizes. However, these approaches require rigorous upfront planning to avoid operational bias, endpoint instability, and overinterpretation of early signals [39]. By aligning enrollment criteria, statistical frameworks, and biological stratification with the realities of CNS tumors, these approaches enable more efficient, informative, and ethically sound evaluation of novel therapies. Trial designs incorporating external data Given the limited efficacy of current therapies for some CNS malignancies, particularly GBM, and the practical challenges facing randomized neuro-oncology trials including suboptimal standard of care therapies and increased drop-out rates in the control arm of a trial, there has been growing interest in alternatives to traditional control arms [54]. In this context, external control arms, sometimes referred to as synthetic controls, have emerged as a strategy to accelerate therapeutic development. By leveraging patient-level data from prior clinical trials or well-annotated real-world datasets, external controls can be used with contemporaneous, covariate-adjusted comparisons to support signal detection in early-phase studies, contextualize single-arm trials, and possibly reduce the number of patients assigned to standard therapies. The most straightforward incorporation of an external control is in an externally controlled single-arm trial, in which outcomes from the investigational cohort are compared with a matched or weighted external control dataset. This approach could be particularly appealing in rare CNS tumors and molecularly defined subgroups where achieving adequate sample size through randomization may be infeasible within a reasonable time frame. Hybrid designs incorporating a component of randomization represent a second strategy, in which external data are used to augment an internal control arm. The primary barrier to wider application of trial designs leveraging external control data remains data access. Importantly, appropriate fit-for-purpose data is critical for the use of an external control arm [55], and the Food and Drug Administration (FDA) has released guidance on considerations for design and conduct of externally controlled trials [56]. Given the need for high quality data, strategies to engage relevant stakeholders and promote data-sharing efforts of clinical trial datasets and real-world datasets is important to enable the use of such designs. Trials as discovery platforms to understand tumor biology and treatment-induced changes Adaptive trial architectures depend on real-time biomarker integration to inform interim decisions. Rather than generating mechanistic insights retrospectively, modern trials can embed biomarkers into prospective algorithms to trigger treatment modifications, patient enrichment strategies, and dose adjustments based on pre-specified decision rules. Clinical trials that prioritize mechanistic understanding require the routine collection of biospecimens before, during, and after treatment exposure. Tissue sampling, including pre-treatment tumor and post-treatment resection specimens, enables direct assessment of drug target engagement, pathway modulation, and evolutionary dynamics within the tumor microenvironment. Similarly, collection of blood and CSF provides accessible windows into systemic and CNS immune responses, circulating tumor DNA dynamics, and drug pharmacodynamics and pharmacokinetics [57]. Embedding these components into trial design ensures that biological insights inform not only whether an agent is active but how and why it exerts its effects. Window-of-opportunity trials represent formalized mechanisms to acquire such samples with the administration of investigational agents prior to planned resection, allowing measurement of drug CNS penetrance and target engagement directly in tumor tissue. These early biological signals can inform whether a therapy merits further development and provide unique biologic parameters that reduce the likelihood of advancing ineffective agents and thereby accelerating discovery. In master protocol designs, biomarker data collected during a trial can directly inform adaptive modifications without requiring protocol amendments. Incorporating correlative biological data into the design of clinical trials can create translational feedback loops that adapt enrollment criteria, combination therapy strategies, and sequencing decisions based on emerging mechanistic insights, such as the incorporation of on-treatment ctDNA, which can trigger specific clinical decision rules within adaptive randomization algorithms. Using contemporary methods, transcriptomic profiling may identify resistance pathways activated in responders versus non-responders within a trial, informing rational combinations that target resistance mechanisms. For neuro-oncology, CSF markers or immunogenomic data from tumor infiltrating lymphocytes can guide the inclusion of immunomodulatory agents or sequencing with other modalities and can enable real-time immune monitoring. Platform trials can integrate serial CSF cytokine assessment with predetermined stopping or modification rules, which may lead to different treatment decisions than standard 12-week imaging evaluations. Embedding translational endpoints within trials enables discovery to directly and dynamically influence clinical decision-making and subsequent protocol evolution. Importantly, there is increased awareness that tumor evolution under the selective pressures of surgery, radiation, and systemic therapy profoundly alters molecular and immunologic landscapes [5,58,59]. Antigenic drift, clonal selection, and microenvironmental remodeling can occur rapidly following treatment, undermining the predictive value of baseline genomics alone and underscoring the need to study post-treatment evolution directly [60,61]. Post-treatment tumor evolution often entails shifts in dominant clones, loss or gain of targetable alterations, and changes in immune infiltration and suppression [61,62]. These dynamic processes can render initial molecular classifications obsolete and contribute to treatment resistance. Consequently, baseline genomic biomarkers may not be useable for therapeutic selection if the target alterations are no longer present or biologically relevant at the time of tumor recurrence. Given the rapid evolution of tumor cell states and microenvironments under treatment pressure [63], eligibility criteria may increasingly need to reflect dynamic tumor state rather than diagnosis alone, particularly in the recurrent setting where prior therapies may reshape tumor biology. Incorporating serial biopsies or longitudinal sampling into trial design offers a strategy to capture this evolution in real time. Recent work in recurrent glioma underscores the extent to which treatment reshapes tumor genomics and the microenvironment [64], reinforcing the need for contemporaneous tissue profiling rather than reliance on archival specimens. Re-biopsy at recurrence, as well as planned on-treatment sampling or liquid biopsy approaches, can identify emerging resistance mechanisms, shifts in clonal architecture, and changes in immune composition that may directly inform adaptive therapeutic strategies. As a framework to implement liquid biopsies that allow for sampling and profiling of CSF emerges [65], validation studies with standardized approaches to monitor responses and identify predictive biomarkers remain a significant need. Emphasizing enrollment and comprehensive biologic profiling of patients who demonstrate durable responses may further uncover features associated with sustained benefit, generating hypotheses for next-generation combinations and biomarker refinement. Furthermore, the inherent heterogeneity within tumors, even among tumors with similar profiles of molecular alterations, can result in disparate responses to the same therapeutic agents. This has motivated integration of functional drug screening platforms into precision medicine trial designs. Patient avatar systems are a personalized medicine strategy in which a patient's tumor is implanted into a model organism, most commonly a mouse or a zebrafish, to test the efficacy of different therapies. These are composed of many complementary approaches, including orthotopic xenografts (PDX) [66], three-dimensional organoid cultures [67], zebrafish [68], ex vivo tissue platforms [69], or in silico screening [70]. This approach can evaluate drug sensitivity within a context that retains key features of the patient's native tumor including tumor-stroma interactions, genomic complexity, and (in some systems) endogenous immunity. In a randomized trial of patients with recurrent glioblastoma who failed standard of care therapy, OS and PFS were significantly improved when patient treatment was guided by an ex vivo assay [71]. Emerging therapies Numerous clinical trials testing immune-modulating approaches in primary CNS tumors are ongoing. These therapies may not demonstrate efficacy in a similar timeline to standard cytotoxic treatments. Thus, clinical trials must have specific consideration for these novel treatment modalities. For example, immune checkpoint inhibitors can be associated with pseudoprogression, and response assessment criteria must take this into account. Furthermore, symptoms caused by CNS- specific immune related adverse events such as aseptic meningitis or encephalitis may be attributed to progression rather than treatment, and thus require close monitoring by an experienced clinical team. Importantly, unlike treatment with cytotoxic chemotherapies, there is evidence that continuing checkpoint blockade through the adverse events (if tolerable) can be associated with improved outcomes. Chimeric antigen receptor (CAR) T cells and bispecific T cell engagers (BiTEs) direct T cell activity toward tumor cells based on the expression of specific tumor surface proteins [72]. However, heterogeneity in antigen expression at baseline and changes due to therapy must be characterized to understand which patients may derive the most long-term benefit. Trials incorporating serial biopsies as well as immune cell phenotyping over time will likely yield the most information about how heterogeneous primary tumors become resistant to these treatments. These therapies also require intensive CSF cytokine monitoring to characterize immune activation and neurotoxicity risk. Furthermore, adaptive dosing strategies may eventually be guided by biomarker kinetics to enhance both efficacy and safety of these treatments. Antibody-drug conjugates (ADCs) and radiopharmaceuticals target a therapeutic payload to tumor cells based on surface marker expression, and are both subject to the same requirements of tumor surface expression. ADCs, due to their size, may face specific challenges penetrating the blood–brain barrier, although some, such as HER2-targeted ADCs have demonstrated excellent CNS activity in the treatment of brain metastases [73]. Radioligand therapies also require unique considerations, particularly the calculation of cumulative dose in combination with external beam RT. Trials testing these treatments will require dose-adaptive designs to tailor therapy intensity to the predicted tumor and organ-specific absorbed doses. Additionally, emerging intra-arterial delivery approaches may lead to enhanced CNS tumor uptake, with additional post-procedural monitoring required. Pragmatic, Ethical, and Regulatory Considerations in Neuro-Oncology Trials CNS tumors pose unique logistical and ethical challenges, including rapid disease progression, limited access to specialized centers, and rare tumor types, rendering traditional centralized trial models increasingly inadequate. Decentralized trial strategies could offer an important opportunity to reduce geographic and institutional barriers that have historically limited participation in neuro-oncology trials. Patients with aggressive CNS tumors often experience significant physical and cognitive burdens that can make frequent travel to academic centers impractical. Incorporating remote consent, telemedicine-based monitoring, local laboratory and imaging assessments, and decentralized follow-up can meaningfully expand patients’ access without compromising data quality [74]. The FDA has issued guidance supporting decentralized and hybrid trial models, and several such trials are now actively underway, signaling growing regulatory acceptance and real-world implementation of these approaches [75,76]. From an equity perspective, decentralized trial designs may also help to address longstanding disparities in trial participation driven by geography, socioeconomic status, and access to specialized care. By lowering the logistical barriers to enrollment, these approaches have the potential to improve inclusion of historically underrepresented populations and to ensure that advances in neuro-oncology research benefit a broader patient community [77]. Furthermore, traditional phase III trials in neuro-oncology often employ highly restrictive eligibility criteria that optimize internal validity but result in trial populations that are poorly representative of routine clinical practice. Pragmatic trial designs can employ broader eligibility criteria, simplify data collection, and real-world outcome measurement to generate evidence that is directly applicable to clinical practice. Areas for expansion of inclusion criteria include elimination of irrelevant organ function requirements, flexibility in treatment protocols to reflect clinical variation, and emphasis on patient-centered outcomes. While these designs require careful planning, applying these features can enhance generalizability to ensure trial results meaningfully inform real-world practice. Ethical considerations are also further amplified in adaptive neuro-oncology trials. Response-adaptive randomization, while potentially increasing the probability that participants receive effective therapies, introduces new challenges for informed consent. Physicians must ensure that patients clearly understand that their treatment allocation may change over time, that assignment probabilities are dynamic rather than fixed, and that changes in treatment are not necessarily individualized to each patient. Clear communication and simplified consent materials are therefore essential to ensure that patients are adequately informed. Furthermore, adaptive randomization can conflict with the principle of equipoise. Once interim data suggest that one arm is superior, ethical justification for continued randomization is undermined, but adaptive trials specifically depend on such interim modifications to maintain efficacy. Ethics committees must prospectively establish both clear stopping rules for futility and superiority. Serial biospecimen collection may expose a patient to increased risk, particularly for CNS tumor tissue rebiopsies which may be necessary to characterize resistance mechanisms, antigen heterogeneity, or microenvironmental changes present real risks and should only be mandated when scientifically justified and when tissue is safely accessible. Adaptive trials may require re-biopsies at specific time points to enable treatment-switching decisions, potentially mandating tissue collection despite uncertainty about benefit to individual patients. Clinical trials typically have ethical obligations for post-trial care, and these are well-defined for fixed-duration studies. However, platform trials operate indefinitely, with new arms added and closed over extended periods of time. This structure creates novel questions about investigator obligations once trials conclude or specific arms are discontinued that have yet to be answered. Balancing speed with patient safety remains a central ethical concern. Adaptive designs enable early stopping for futility or success, reducing exposure to ineffective or harmful therapies. However, rapid iteration must not compromise careful toxicity monitoring, particularly when evaluating novel therapeutic combinations or sequencing strategies. Robust safety oversight, conservative early stopping boundaries, and predefined escalation and de-escalation rules are necessary to protect participants while preserving the efficiency advantages of adaptive trials. Artificial Intelligence in Clinical Trial Design The use of artificial intelligence (AI) in clinical trials remains in its early stages, yet it holds significant promise for improving efficiency, reducing cost, and enabling analyses that generate new insights. As most clinical trials fail to meet their initial enrollment timelines and targets, AI tools have the potential to streamline and accelerate patient recruitment by automating eligibility assessment, prescreening electronic health records, and identifying eligible candidates across health systems. This is particularly valuable in rare diseases such as CNS tumors, where eligible patients are uncommon and difficult to identify, and where natural language processing systems may offer a promising approach to automate eligibility prescreening and improve trial matching [78]. Beyond patient recruitment, AI can support data cleaning, harmonization across sites, remote patient monitoring, and complex analyses that generate new insights from existing datasets, all of which can be otherwise time-intensive and costly. In addition, virtual in silico trial simulations can be used to model anticipated outcomes and success probabilities as investigators define key design parameters, which allows for refinement eligibility criteria, sample size, and endpoint selection before trials launch to maximize their efficiency. Beyond recruitment, data analysis, and trial simulations, digital twins are another emerging option in AI-assisted trial design. Digital twins are dynamic computational models that integrate histopathologic, genomic, imaging, and clinical data to create virtual avatars of patients that can simulate disease progression and treatment response. In neuro-oncology, digital twins have predictive accuracy in glioblastoma [79], which enables simulation of patient-specific treatment responses to guide enrollment decisions and inform adaptive dosing strategies. Digital twins can also be used to generate synthetic control arms, though this approach is currently limited by the complexities of data integration and the robust validation requirements that would have to be demonstrated before incorporation into clinical trials. Real-time AI integration has the potential to transform clinical trials in terms of practical implementation of trial protocols. AI enables continuous biomarker monitoring throughout adaptive trial enrollment, treatment, and adaptation, which can provide real-time guidance about patient management; these approaches have started to be used in ongoing neuro-oncology trials. At trial entry, radiomics algorithms can analyze baseline MRI to noninvasively predict molecular biomarker status (such as IDH mutation and MGMT methylation status) with diagnostic accuracy exceeding 90% to enable immediate biomarker-driven stratification without delaying enrollment [80]. During CAR-T therapy, AI systems can continuously integrate CSF or serum biomarker data, which automatically monitors cytokine trajectories and has been shown to predict effect cell-associated neurotoxicity syndrome (ICANS) risk 3–5 days before clinical symptom emergence [81,82], enabling pre-emptive treatment or dose de-escalation to reduce the risk of severe complications. During radiotherapy, deep learning-based segmentation of both tumors and organs at risk can achieve >90% accuracy, significantly reducing the time needed for radiation contouring and planning. AI-guided adaptive radiation therapy can analyze daily imaging to account for tumor volume shrinkage, changes in edema, or anatomical shifts and re-plan the radiation target volume without treatment delay [83,84]. Integrating AI for biomarker and patient monitoring has the potential to support trial design based on individual patient features and real-time responses to therapy. Despite its considerable promise across medicine, including clinical trial design and execution, the integration of AI faces meaningful challenges. Many models are developed using homogeneous or single-center datasets, limiting generalizability and increasing the risk of embedded bias. Algorithmic bias, lack of transparency, and poor alignment with existing clinical workflows can undermine reliability and slow adoption. Rigorous external validation, use of diverse training datasets, and thoughtful integration into trial infrastructure will be essential to ensure that improvements in efficiency do not compromise equity or scientific rigor. As these challenges are addressed, AI-based tools have the potential to both enhance and accelerate clinical trial conduct in neuro-oncology. Conclusions Despite extraordinary advances in molecular profiling, imaging, and therapeutic development, neuro-oncology has seen limited translation of these biological discoveries into clinical treatment strategies that provide patients with durable benefits. As we outline, the failure of traditional trial paradigms in CNS tumors is not simply a function of aggressive biology, but is also due in part to structural mismatches between trial design and disease biology. Emerging clinical trial models including master protocols, platform trials, and Bayesian adaptive designs, offer a path forward by addressing many of these structural limitations. Accelerating progress in neuro-oncology also requires the continued emphasis on embedding biology directly into trial architecture. Trials will benefit by going beyond static baseline genomics, and window-of-opportunity studies and longitudinal profiling strategies are critical to assess CNS drug penetration, pharmacodynamic target engagement, and treatment-induced evolution. By aligning design, biology, and ethics, the next generation of neuro-oncology trials can more reliably identify therapies capable of delivering lasting benefits to patients with CNS tumors. Author contributions AW & RR contributed to conception, design, literature review, writing and review of the final manuscript. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Rifaquat Rahman reports a relationship with St. Lucia Consulting that includes: consulting or advisory. Rifaquat Rahman reports a relationship with Servier Pharmaceuticals LLC that includes: consulting or advisory and speaking and lecture fees. Rifaquat Rahman reports a relationship with Telix Pharmaceuticals Limited that includes: consulting or advisory. Rifaquat Rahman reports a relationship with National Brain Tumor Society that includes: consulting or advisory. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Footnotes This article is part of a special issue on published in Neurotherapeutics References