Basket trials can make rare cancer research more feasible by grouping patients according to shared genomic alterations rather than tumor location alone.
Evidence from BRAF studies and NCI-MATCH shows that the same molecular alteration can produce different treatment responses across cancer types.
Statistical borrowing and adaptive trial designs can strengthen small cohorts, but inappropriate pooling may create misleading conclusions.
Recruitment feasibility, biomarker testing, protocol availability, and multicenter coordination are central to the scientific validity of basket trials.
The future of rare cancer clinical research will likely integrate genomics, histology, and adaptive evidence generation rather than relying on any single classification system.
When Rarity Becomes a Research Problem
Rare cancers present a paradox for oncology research. Most individual cancer types meet the National Cancer Institute (NCI) definition of rare, meaning that they occur in fewer than 15 people per 100,000 each year, yet rare cancers collectively account for just over one-quarter of cancers in the United States. Their combined burden is substantial, but each disease may affect too few patients to support the conventional clinical development model used for more prevalent malignancies.1,2
Low incidence creates difficulties at nearly every stage of research. Investigators may have limited access to tumor specimens, validated disease models, natural history data, and clinicians with deep experience treating a particular cancer. Patients may be dispersed across regions or countries, requiring many clinical sites to identify even a modest study population. Some patients receive a diagnosis only after extensive evaluation, while others may never reach a center conducting relevant research. These obstacles can delay enrollment, weaken the evidence available to guide treatment, and reduce the commercial incentive to develop therapies for narrowly defined indications.2
Genomic classification adds another layer to this problem. A cancer that is already uncommon may contain several molecularly distinct subgroups, each representing only a fraction of the total patient population. A conventional trial restricted to one histology and one genomic alteration may therefore seek patients who are rare twice over: first because of the cancer type and again because of the biomarker required for eligibility.
Basket trials offer a way to reconsider how those patients are assembled for research. Instead of limiting enrollment to one organ or tissue of origin, a basket trial may evaluate a therapy across several cancer populations that share a molecular alteration or another defining characteristic. This approach can bring together patients who would otherwise be separated into multiple small, independently conducted studies. It does not remove the biological differences among cancers, but it can create a practical framework for investigating whether a shared genomic feature has therapeutic significance across them.
Reorganizing Trials Around Molecular Biology
Traditional oncology development has generally organized clinical trials around tumor location and histology. Patients with lung cancer enter lung cancer studies, patients with colorectal cancer enter colorectal cancer studies, and each development program builds evidence within a specified disease category. Molecular profiling has complicated that structure by revealing potentially actionable alterations that may appear across several histologies, sometimes at low frequencies within each one.
A basket trial evaluates one investigational therapy or therapeutic combination in multiple cancer populations defined by specified characteristics. Those characteristics may include histology, disease stage, previous treatment, genetic alterations, or other biomarkers. When genomic selection drives enrollment, the baskets commonly correspond to different tumor types that share the molecular feature being studied.3
Basket trials sit within the broader category of master protocols. A master protocol provides an overarching trial structure under which multiple drugs, multiple cancer types, or both can be evaluated. The shared infrastructure may include central governance, common screening procedures, standardized data collection, and coordinated rules for opening, expanding, modifying, or closing substudies. This structure can reduce the need to establish a separate protocol for every clinical question, although each basket must still produce evidence that can be interpreted on its own terms.4
Basket trials should not be treated as interchangeable with tissue-agnostic drug development. The former describes a clinical trial design. The latter describes a development approach in which a therapy targets a specific molecular alteration across cancers defined by different organs, tissues, or tumor types. A basket trial may contribute to a tissue-agnostic development program, but it can also reveal that activity is limited to selected histologies or that a biomarker is insufficient to predict response across cancers. The U.S. Food and Drug Administration’s (FDA’s) guidance on tissue-agnostic oncology development remains in draft form, but it recognizes that such programs raise scientific and regulatory questions that do not typically arise when development remains confined to one cancer type.5
The distinction matters because the architecture of a trial cannot establish biological equivalence. Grouping patients according to a shared alteration creates an opportunity to test whether that alteration predicts response across tumor types. It does not settle the question in advance.
Why Basket Trials Are Especially Relevant to Rare Cancers
The practical appeal of basket trials becomes clearest when a genomic alteration appears infrequently across several uncommon cancers. Conducting a separate trial for every combination of histology and biomarker may be infeasible. Each study would require its own activation process, investigator network, screening effort, monitoring plan, and analysis, even though the underlying therapeutic hypothesis might be similar.
A basket structure can consolidate those efforts. Patients with different cancers may undergo screening under a common protocol and enter the cohort corresponding to their tumor type or molecular profile. Investigators can evaluate whether early activity appears in one or more groups, discontinue cohorts with little evidence of benefit, and expand those that generate stronger signals. The usual basket trial is a single-arm study intended to estimate activity, often with overall response rate as the primary endpoint, although that pattern does not apply to every design.3
This flexibility is particularly useful when the prevalence of the target alteration is uncertain or unevenly distributed. A sponsor may know that a biomarker occurs across several cancers without knowing which histologies are most likely to respond or accrue enough patients for meaningful analysis. A basket trial can test several possibilities within one coordinated program and allow later development to focus on the populations that show the clearest evidence of activity.
The potential efficiency, however, comes from shared infrastructure and adaptive decision-making, not from treating all enrolled patients as one homogeneous population. A rare cancer basket may contain only a small number of participants, but those participants still represent a distinct clinical and biological context. Combining them indiscriminately with patients from other baskets can obscure meaningful differences and create misleading estimates of treatment effect.
For rare cancer research, the value of the design lies in its ability to accommodate fragmentation without pretending that fragmentation is irrelevant. It creates a pathway for studying small populations together while preserving the ability to learn where their outcomes diverge.
The BRAF Experience: A Shared Mutation Does Not Erase Tumor Context
An early basket study of vemurafenib in nonmelanoma cancers with BRAF V600 mutations demonstrated both the promise of molecularly organized trials and the limits of a shared-biomarker hypothesis. The phase II study included six prespecified cohorts defined by cancer type, along with an additional cohort that included other malignancies carrying the alteration. In total, 122 patients received treatment, including 27 patients with colorectal cancer who received vemurafenib in combination with cetuximab.6
The study tested a straightforward question: could a therapy already associated with activity against BRAF V600–mutated melanoma produce meaningful responses in other cancers carrying the same mutation? The results showed activity in some nonmelanoma cancers, but the degree of activity varied by histology. The presence of BRAF V600 did not produce a uniform response across all tumor types6.
That finding remains central to the interpretation of basket trials. A genomic alteration may help drive cancer growth in several tissues, but its functional importance can depend on the surrounding biology. Different tumors may activate alternative signaling pathways, rely on distinct compensatory mechanisms, or respond differently to inhibition of the same target. A biomarker that predicts sensitivity in one cancer may therefore be less informative in another.
The lesson is not that molecular classification fails or that histology should retain absolute priority. Molecular features can reveal therapeutic relationships that would remain invisible within organ-based development alone. The lesson is that genomic and histologic information answer different questions. The alteration may identify the mechanism a therapy is designed to affect, while tumor context helps determine whether targeting that mechanism will produce a clinically meaningful result.
Basket trials provide a disciplined way to examine that interaction. Separate cohorts allow investigators to test a common molecular hypothesis without collapsing the evidence across cancers. When responses differ, those differences can refine the biomarker strategy, support combination approaches, or redirect development toward the tumor types in which the target appears most consequential.
NCI-MATCH at Scale
The National Cancer Institute Molecular Analysis for Therapy Choice (NCI-MATCH) trial extended the basket concept across a broad range of molecular targets and refractory malignancies. The program used centralized molecular profiling to assign eligible patients to treatment arms based on actionable alterations rather than the site where their tumors originated.
Tumor biopsy specimens from 5,954 patients were analyzed using next-generation sequencing and selected immunohistochemistry. Molecular profiling succeeded in 93.0% of specimens, and 37.6% contained an alteration considered actionable within the trial. After molecular and clinical eligibility criteria were applied, 17.8% of screened patients were assigned to a treatment arm. The analysis estimated that 26.4% might have been assigned had all relevant subprotocols been available at the same time.7
The gap between identifying an alteration and assigning a patient illustrates an important constraint in genomically guided trials. A potentially actionable result does not automatically translate into enrollment. The appropriate treatment arm must be open, the patient must meet its clinical criteria, adequate trial capacity must be available, and the patient must remain well enough to participate. Molecular screening is therefore only one component of access.
The scale of NCI-MATCH also demonstrated the operational demands of broad precision oncology programs. A national screening network, standardized assays, centralized interpretation, numerous treatment arms, and coordinated eligibility rules were required to move patients from biopsy to assignment. The trial ultimately enrolled 1,201 patients across 38 treatment arms, creating a large body of evidence concerning both the feasibility and limitations of biomarker-matched development.8
For rare cancers, this infrastructure is particularly relevant. A single institution may encounter very few patients with a specific histology and alteration, but a distributed network can identify candidates across a much larger population. Central coordination can also promote consistency in testing and eligibility assessment. At the same time, the NCI-MATCH experience shows that network scale alone cannot eliminate the attrition that occurs between genomic testing and treatment. Trial design must account for subprotocol availability, patient condition, specimen adequacy, and the time required to complete each step.
What NCI-MATCH Revealed About Tumor-Agnostic Activity
NCI-MATCH did not produce a single answer about the value of molecular matching. Its treatment arms evaluated different targets, therapies, and biomarker hypotheses, and their outcomes varied accordingly. Across the arms included in a later review, 79 of 765 evaluable patients achieved a response, corresponding to an overall response rate of 10.3%. That aggregate figure provides context, but it cannot represent the performance of every individual arm.9
Some treatments demonstrated activity across multiple malignancies, supporting the view that selected molecular alterations can define clinically relevant populations beyond a single tissue of origin. Other arms produced limited or no broad response despite a plausible target–therapy match. These differences reinforce the importance of validating each molecular hypothesis rather than treating genomic selection itself as evidence that a therapy will work.8,9
The results also show why unsuccessful baskets can remain informative. A cohort that fails to meet a predefined activity threshold may indicate that the alteration is not a sufficient biomarker in that tumor type, that the therapy does not inhibit the target effectively enough, or that additional biological features influence sensitivity. Those possibilities may lead to refined eligibility criteria, better assays, alternative drugs, or combination strategies.
This kind of evidence is especially valuable in rare cancers, where conventional randomized studies may be difficult to complete and every enrolled patient contributes to a limited knowledge base. A negative signal can prevent further investment in an unproductive hypothesis. A response concentrated in one histology can direct resources toward a more focused trial. Activity across several baskets can justify broader development while identifying the tumor types that require additional confirmation.
The Statistical Challenge of Learning from Small Baskets
Basket trials must extract credible evidence from cohorts that may contain very few patients. If every basket is analyzed entirely independently, response estimates may be unstable, and true treatment activity may be difficult to distinguish from random variation. However, pooling all patients across tumor types can be equally problematic when the effect of treatment differs among baskets.
Statistical information borrowing offers a middle path. Methods can allow data from one basket to inform estimates in another when the cohorts appear sufficiently similar. This approach may strengthen inference for small tumor-specific groups by increasing the information available to estimate response. Its validity depends on whether the assumption of similarity is justified.10
Bayesian hierarchical models are among the approaches used for this purpose. Such models may treat tumor-specific response rates as exchangeable, meaning that they are assumed to arise from a common distribution rather than being completely unrelated. When that assumption reasonably reflects the biology and observed outcomes, borrowing can improve estimation. When treatment effects differ substantially among tumor types, however, borrowing can pull estimates toward an artificial average and distort the conclusions for individual baskets.10
Complete pooling creates an even clearer risk. If investigators assume that a biomarker confers the same response probability across all tumor types, a strong signal in one basket may influence the apparent result for another. Naive pooling under an incorrect homogeneity assumption can increase the false-positive rate, potentially suggesting activity in populations where the evidence is weak.10
The statistical plan must therefore address heterogeneity before outcomes are known. Investigators should define how similarity among baskets will be evaluated, how much information may be shared, how error rates will be controlled, and how cohort-level results will be reported. Simulation can help test how a proposed method behaves under different patterns of response, including scenarios in which all baskets perform similarly, only a subset benefits, or activity appears in a single tumor type.
For rare cancers, the temptation to borrow aggressively is understandable because recruitment may be exceptionally difficult. Scarcity of data, however, does not make populations biologically interchangeable. Statistical strength must come from a model that reflects plausible relationships among cohorts, not merely from combining every available observation.
Adaptation Without Overinterpretation
Adaptive basket trials can modify enrollment in response to accumulating evidence. A cohort showing little early activity may close before reaching its maximum sample size, while a promising basket may continue or expand. These decisions can concentrate patients and resources where the therapy appears most likely to provide benefit and reduce exposure in populations where the treatment is unlikely to succeed.3,10
Early stopping is not risk-free. A basket may enroll only a handful of patients before an interim analysis, and an initially weak result may reflect chance rather than true inactivity. A cohort that closes too quickly could eliminate a tumor type in which the therapy would have shown activity with additional enrollment. Conversely, allowing every cohort to continue despite consistently poor results can prolong an uninformative study and consume a limited pool of eligible patients.10
Prespecified decision rules are essential. The protocol should define the evidence required to stop, continue, or expand a basket and explain how repeated interim assessments affect statistical error. These rules should be evaluated in advance under a range of plausible recruitment and response scenarios. The goal is not to remove judgment from development but to ensure that major decisions do not depend on improvised interpretations of sparse data.
Expansion also requires caution. A strong response signal may justify enrolling additional patients and generating evidence that could support further development or, in some circumstances, a marketing application. The initial signal must still be interpreted within its cohort, with attention to sample size, response durability, prior treatment, and the reliability of the biomarker definition.3
Transparent reporting helps prevent the master protocol from masking these distinctions. Results should show how each basket performed, which adaptations occurred, how many patients contributed to each analysis, and whether information was borrowed across cohorts. A trial may begin with a shared molecular premise, but the evidence must remain visible at the level where clinically important differences emerge.
Trial Operations Are Part of the Scientific Design
A basket trial’s analytical framework can succeed only if the trial recruits the populations needed to test it. Biomarker prevalence may differ widely among tumor types, causing some cohorts to fill quickly while others remain open for extended periods. Results may therefore become available at different times, even when all baskets begin under the same protocol.10
Asynchronous recruitment can complicate both adaptation and information borrowing. Some statistical approaches assume that outcomes from several baskets will be available together, but a slowly accruing rare cancer cohort may not reach an interim analysis until other groups have completed enrollment. Investigators must decide whether to wait, update the model as data arrive, or analyze selected cohorts separately. Each choice can affect the operating characteristics of the trial.
Recruitment feasibility should be examined before baskets open. Estimates of biomarker prevalence, disease incidence, testing volume, referral patterns, and site access can help determine whether a cohort is likely to enroll. Even then, projections may prove inaccurate, particularly when a molecular alteration has not been studied extensively in a rare histology. Designs need enough flexibility to address those uncertainties without preserving nonviable cohorts indefinitely.10
The NCI-MATCH experience underscores the importance of the path from screening to treatment. Successful molecular profiling did not guarantee assignment, and the simultaneous availability of relevant subprotocols affected the proportion of patients who could enter a treatment arm. A trial may identify an appropriate molecular match but still lose the opportunity to enroll the patient if confirmation, eligibility review, or site activation takes too long.7
For rare cancers, geographic reach is equally important. Patients may receive care outside major academic centers, while the expertise and testing needed to identify them may be concentrated at a relatively small number of institutions. A multicenter or national network can expand access, but it also increases the need for consistent testing, rapid communication, coordinated specimen handling, and clear referral pathways.2
These are not separate administrative concerns layered onto an otherwise complete scientific design. They determine which patients enter the study, how representative each cohort becomes, whether planned analyses are feasible, and how quickly evidence can be generated. In a basket trial, operations shape the data on which biological conclusions depend.
Building the Next Generation of Rare Cancer Evidence
The future value of basket trials in rare cancer research will depend on more than the ability to identify patients who share a genomic alteration. The available evidence suggests that credible programs must align several elements: the biological rationale for the target, the definition and performance of the biomarker strategy, the boundaries of each cohort, the recruitment network, the statistical model, the adaptive rules, and the plan for interpreting or confirming any observed signal.
This is an analytical conclusion rather than a formal consensus statement. It follows from the recurring limitations visible across the evidence base. Rare cancer research begins with small and dispersed populations. Basket protocols can bring those populations into a coordinated study, but the vemurafenib experience shows that a shared mutation may have different consequences in different tumor types. NCI-MATCH demonstrates both the reach of large-scale genomic screening and the attrition between identifying an actionable alteration and assigning a patient to treatment. Statistical research shows that borrowing can strengthen inference, but only when assumptions about similarity are defensible. Operational analysis shows that uneven recruitment can undermine methods that appear efficient on paper.
Future designs will therefore need to preserve two forms of reasoning at once. They must be broad enough to detect activity that crosses conventional disease boundaries and sufficiently granular to recognize when histology, pathway context, or clinical features alter the effect of treatment. A molecularly defined program should be able to expand beyond one tumor type without requiring investigators to assume that every tumor carrying the alteration belongs to the same therapeutic population.
That balance may influence how evidence is generated in stages. An initial basket trial can screen for activity across several cancers. Cohorts with little evidence of benefit can close, while those with stronger signals can expand or proceed to more focused evaluation. Where responses appear consistent across histologies, the development program can investigate whether a broader tissue-agnostic strategy is justified. Where activity differs, subsequent trials can narrow the population, refine the biomarker, or evaluate combinations tailored to the biology of individual cancers.
The approach also changes the meaning of efficiency. A successful basket program is not simply one that enrolls fewer patients or reaches a decision quickly. It is one that uses a scarce patient population responsibly and produces conclusions that remain credible despite small cohorts. Closing an inactive basket early may be efficient. So may continuing a slow-accruing cohort when its biological rationale is strong and its evidence cannot be inferred safely from other tumor types.
Rare cancer research is unlikely to move from histology-based development to a system governed by genomics alone. A more plausible future combines molecular classification with tumor context, clinical characteristics, and adaptive evidence generation. Basket trials can support that integration by creating a common structure for testing related hypotheses while allowing the resulting evidence to separate where biology demands it.
For patients with rare cancers, that structure may open studies that would otherwise be impossible to conduct. For researchers, it can reveal which molecular relationships extend across malignancies and which remain dependent on tissue-specific biology. Its greatest contribution may be neither the abandonment of traditional cancer categories nor the universal adoption of tissue-agnostic development, but a more precise way to decide when those categories should be crossed.
References
1. “Definition of Rare Cancer.” National Cancer Institute Dictionary of Cancer Terms. Accessed 22 Jun. 2026.
3. FDA Modernizes Clinical Trials with Master Protocols.” U.S. Food and Drug Administration. 26 Feb. 2019. https://www.fda.gov/drugs/cder-small-business-industry-assistance-sbia/fda-modernizes-clinical-trials-master-protocols)
