Executive Summary

Master protocols—encompassing basket, umbrella, and platform trials—represent a paradigm shift in oncology drug development. In these designs, a single overarching trial infrastructure is used to evaluate multiple treatments and/or patient subgroups, dramatically increasing efficiency compared to traditional one-drug–one-disease trials (^[1]) (^[2]). Basket trials test one or more targeted therapies across multiple tumor types sharing a common biomarker, whereas umbrella trials test multiple therapies within a single cancer type, stratified by molecular markers (^[3]) (^[2]). Platform (or multi-arm) trials are even broader, often allowing new arms to be added or dropped dynamically in one ongoing trial. These designs have grown rapidly: a 2019 landscape analysis identified 83 oncology master protocols (49 basket, 18 umbrella, 16 platform) (^[4]). FDA guidance in 2018 explicitly endorsed these approaches, reflecting regulatory support (^[5]).

The benefits are substantial. Master protocols can rapidly match targeted agents to patients and test combinations, accelerate accrual, and reduce the number of control patients by sharing comparison arms (^[6]) (^[7]). They have led to tissue-agnostic approvals: for example, pembrolizumab was FDA-approved in 2017 for microsatellite instability-high cancers across all histologies (^[8]), and larotrectinib was approved in 2018 for any NTRK-fusion solid tumor (^[9]). These milestones highlight how basket designs can uncover drug activity beyond traditional indications. Master protocols also afford flexibility: unsuccessful arms can be terminated early, and promising new therapies added mid-trial (^[10]) (^[11]). Moreover, adaptive statistical methods (Bayesian hierarchical models, response-adaptive randomization) can further enhance efficiency (^[12]) (^[13]).

However, these innovations introduce challenges. Statistically, the multiplicity of sub-studies risks inflated false-positive error rates if not carefully controlled (^[12]) (^[14]). Operationally, building and coordinating complex infrastructure across many institutions is demanding. Ethically, ensuring valid informed consent and sufficient enrollment in rare subgroups is a concern (^[15]) (^[16]). High screening burdens can limit match rates (e.g. early NCI-MATCH cohorts matched ~12% of patients (^[11])). Despite these hurdles, experience shows master protocols can succeed. For instance, the UK’s FOCUS4 colorectal trial screened 1434 patients to randomize 361 into biomarker-driven arms (^[17]). Industry and academia are increasingly adopting these trials: as of 2025 multiple pharmaceutical consortia and cancer centers run federated master studies.

Looking forward, master protocols will likely expand further, propelled by comprehensive genomic profiling and AI-driven patient selection. They may move beyond genomics to integrate multi-omic or imaging biomarkers. Global collaborative trials (e.g. GBM-AGILE in neuro-oncology) are emerging to tackle common cancers that lack effective treatments. Continued methodological advances (improved multiplicity control, seamless phase transitions) and regulatory harmonization will be crucial. In sum, basket and umbrella trials have proven their value in precision oncology (^[8]) (^[18]), and represent a key component of the future clinical trial landscape.

Introduction and Background

The challenge of modern oncology drug development is the explosion of tumor subtypes and targeted therapies. Traditional trials were organized by histology and stage, testing one treatment in one cancer population through sequential phase I–III steps (^[19]) (^[20]). This approach is becoming impractical. Advances in genomics mean tumors are now classified by myriad molecular alterations (^[1]) (^[19]). For example, lung cancer is subdivided by EGFR mutations, ALK fusions, and many rarer changes (^[21]) (^[22]). Enrolling enough patients for each mutation-specific study quickly exceeds feasible timelines. Furthermore, many new drugs (including immunotherapies) work only in biomarker-defined subgroups (^[23]) (^[24]). This has led to a “precision oncology” era in which patient selection is based on genetic drivers rather than tumor origin (^[1]) (^[20]).

Biomarker-based trial designs predated master protocols. Early examples include enrichment designs that only enroll mutation-positive patients, such as the NSABP N9831 trial of trastuzumab in HER2-positive breast cancer (^[25]), or the ToGA trial in HER2-positive gastric cancer (^[25]). In cases where the predictive value of a biomarker was unproven, marker-stratified designs randomized patients separately by marker status (e.g.INTEREST and MARVEL trials (^[26])). Extensions like the MaST and fallback designs allowed for sequential testing of subgroups (^[27]). These specialized designs proved the concept that targeting therapies by markers works. However, conducting hundreds of separate trials for each biomarker–tumor combination is prohibitively slow and expensive.

Master protocols emerged to unify and accelerate this process (^[1]) (^[28]). A master protocol is a single umbrella document under which multiple sub-studies are run concurrently (^[29]) (^[30]). All sub-studies share the same infrastructure (e.g. central lab testing, data systems, and consent forms) (^[29]) (^[30]). For oncology, the goal is to match therapies to patients more efficiently: patients are screened upfront with molecular profiling and then funneled into the appropriate sub-study. Patients may enter an arm for which they have an actionable alteration, or be placed on a waiting list for a new sub-study if none apply. The shared control (or natural history) can even let patients serve as “waiting controls” for one-arm sub-trials (^[31]) (^[28]). In effect, master protocols standardize procedures across many mini-trials (^[31]) (^[30]), reducing duplication of effort and speeding development.

The concept was popularized in the 2010s. Woodcock & LaVange (2017) outlined how master protocols could study “multiple therapies, multiple diseases, or both” under one protocol (^[32]). Early adopters include the I-SPY 2 breast cancer trial (launched 2010) and the Lung-MAP trial (launched 2014). Regulatory agencies took notice: in Sept 2018 the FDA released draft guidance endorsing master protocols, stating that basket and umbrella trials could efficiently advance oncology drugs (^[5]). A cascade of clinical trials and publications followed. By 2019, a systematic review had catalogued 83 such trials, mostly in oncology (^[4]). This primer will explore the basket vs umbrella designs in depth, drawing on multiple sources to explain definitions, practical experiences, and implications for cancer research.

Master Protocol Trial Types: Definitions and Key Concepts

Master protocol studies in oncology are generally categorized as basket, umbrella, or platform trials (^[33]) (^[34]). However, definitions can overlap, so clarity is essential. We adopt the common definitions:

• Basket Trials (Histology-Agnostic) – These test one (or a few) targeted therapy(ies) across multiple tumor types that share a common biomarker or risk factor (^[3]) (^[2]). Eligible patients come from various histologies but all have the same genetic alteration or protein target. For example, a basket trial might give a BRAF inhibitor to any patient with a BRAF V600E mutation, regardless of whether their cancer is melanoma, lung cancer, colon cancer, etc. The unifying eligibility criterion is the predictive biomarker (^[3]) (^[20]). Classically, these are single-arm (uncontrolled) studies with objective response as the primary endpoint, and each histology cohort is analyzed separately (^[35]) (^[20]). The recent approvals of drugs like larotrectinib (for NTRK fusions) and pembrolizumab (for MSI-H) are rooted in basket trial designs (^[8]) (^[9]).

• Umbrella Trials (Histology-Specific) – These run multiple targeted sub-studies within a single cancer type, stratifying patients by different biomarkers (^[3]) (^[20]). All patients have the same histology at entry (e.g. lung adenocarcinoma), but are assigned to distinct arms based on molecular features. For instance, lung cancer patients might be screened and those with EGFR mutations receive an EGFR inhibitor in one sub-study, those with ALK fusions get a different ALK inhibitor in another arm, and so on. Thus one disease is partitioned into multiple molecular subgroups, each with its own investigational therapy (^[3]) (^[20]). Umbrella trials often include randomization against a control within each subgroup and measure endpoints such as progression-free survival or overall survival (^[36]) (^[37]). Notable examples include the BATTLE-1 and Lung-MAP trials in lung cancer, ALCHEMIST in early-stage lung cancer, and FOCUS4 in colorectal cancer (^[38]) (^[39]).

• Platform Trials (Adaptive/Multi-Arm Multi-Stage) – These are broader designs that can be thought of as dynamic master protocols. A platform trial often has a common control arm and evaluates multiple investigational arms simultaneously, with pre-specified interim analyses allowing new treatments to be added or ineffective ones dropped (^[40]) (^[34]). Platform trials may or may not be strictly histology-specific. For example, the STAMPEDE trial in prostate cancer and the I-SPY 2 trial in breast cancer are platform trials that use adaptive randomization and seamless Phase II/III stages (^[34]) (^[13]). In immuno-oncology, even the RECOVERY trial for COVID-19 was a platform model (though non-oncology). In one sense, basket and umbrella trials can be seen as special cases within the more general platform framework (^[41]): if a platform trial tests a single drug (basket) or a single disease (umbrella), it effectively becomes that type of master protocol.

The distinctions are summarized in Table 1 below. Key differences include the scope of diseases enrolled, the number of interventions tested, and typical study phase and endpoints. Translating these definitions into practice can blur lines (for instance, NCI-MATCH has elements of both basket and umbrella designs (^[42]) (^[41])). Nevertheless, understanding the archetypes of basket vs. umbrella is crucial for researchers designing these trials.

Table 1. Features of Basket, Umbrella, and Platform Trials

Table 1. Comparison of master protocol trial types (Sources: (^[3]) (^[16]) (^[35]) (^[8]) (^[36]) (^[13]) (^[50]) (^[41])).

Basket Trials in Oncology

In basket trials, the central idea is that if a drug targets a specific molecular alteration, that alteration should predict sensitivity regardless of the tissue of origin. This concept was first popularized by Jain et al. and Redig & Janne (^[2]) (^[20]). By enrolling any patient whose tumor carries the biomarker of interest, baskets leverage rare mutations by pooling across many cancers. They are especially useful for “tumor-agnostic” development of drugs.

Design Principles. Typically, a basket trial has separate cohorts for each tumor type (or organ group). Example: the NCI-MATCH trial has dozens of cohorts, each for a different mutation–drug pair (^[55]). Many baskets are single-arm Phase II trials with objective response rate (ORR) as the endpoint (the “signal-finding” approach). Interim analyses may stop a cohort early for futility if responses are low, or expand enrollment for promising cohorts (^[49]) (^[11]). Simon two-stage designs are common per cohort (^[49]) (^[12]). Some basket trials incorporate Bayesian hierarchical models to allow information pooling across cohorts (^[12]). In such models, the assumed similarity of response rates among histologies can be encoded with a prior distribution (^[56]). Adaptive features are increasingly used: for example, I-SPY 2 (while technically in breast cancer) adapts by dropping ineffective arms based on accumulating data (^[13]).

Advantages of Baskets. Basket trials are efficient for rare subgroups. By testing a drug in multiple cancers simultaneously, they increase the chance of finding responsive patients without needing separate trials for each cancer. This was crucial in the approval of larotrectinib: NTRK fusions are individual exceedingly rare (<1% in each histology (^[57])), but the basket TRK inhibitor trial enrolled hundreds of patients across 20+ tumor types (^[44]), yielding a pooled response rate of ~75% (^[58]). The basket also accelerates enrollment speed; for instance, the Novartis “SIGNATURE” basket trial (Phase II, multiple cohorts) reported site start-up in 3.6 weeks on average vs. ~10.4 months for a traditional trial (^[6]). In [1]'s landscape, most basket trials were early-phase and small (median 205 patients) (^[46]) – highlighting their role in exploring signals. Furthermore, basket designs allowed tissue-agnostic FDA approvals: e.g., pembrolizumab for MSI-H/dMMR tumors in any histology (^[8]), and larotrectinib/entrectinib for NTRK fusions (^[9]). These are landmark clinical–regulatory successes enabled by basket data.

Challenges of Baskets. Despite advantages, baskets have pitfalls. Tumor biology is complex: a drug may be highly effective in one cancer type but not in another, even with the same mutation. For example, the BRAF V600E inhibitor vemurafenib works spectacularly in melanoma, but has modest effect in BRAF-mutant colorectal cancer (^[59]). If a basket trial pools across types and analyzes only the aggregate ORR, such tissue-specific differences might be missed. Indeed, a recent analysis of NCI-MATCH found that several “negative” subprotocols (no significant overall ORR) actually contained statistically meaningful responses in particular tumor types (^[60]). This suggests baskets can identify “hidden” activity, but also that interpreting overall results requires caution.

Statistically, basket trials must address multiplicity and heterogeneity. Running many independent cohorts inflates false-positive error risk (^[14]). If each cohort tests a null hypothesis (e.g. ORR ≤ baseline) independently, the chance of at least one spurious positive grows with number of cohorts. Methods like hierarchical modeling or multiplicity adjustment can mitigate this (^[12]) (^[13]), but require careful planning. In practice, many baskets deliberately avoid making formal statistical comparisons (focusing on estimation of ORR) to simplify analysis. Nevertheless, final interpretation (and potential regulatory decisions) must consider the multiple testing issue.

Operationally, basket trials may require screening large numbers of patients to fill cohorts. For rare mutations, this screen “funnel” can be substantial. In NCI-MATCH, over 6,000 patients were screened in under 2 years (^[61]), yet initial match rates were only ~12% (10 initial arms)–prompting the trial to add many more arms (^[11]). In practice, poor match rates stem from limited prevalence of targets and limited trial capacity. Large screening databases (e.g. national genomics networks) help identify eligible patients, but challenges remain. Ethically, enrolling patients in rare-biomarker arms also raises consent issues; participants may expect benefit based on the “precision” label, but actual benefit can be low for rare subtypes (^[15]) (^[60]). Trial designers must clearly communicate amid this complexity.

Case Studies: Basket Trials. Several high-profile basket trials illustrate these points. The NCI-MATCH trial (U.S. NCI, opened 2015) is perhaps the largest oncology basket. It screens adults with refractory solid tumors/lymphomas for a panel of gene alterations using a 143-gene assay (^[55]). Patients are then assigned to 35+ treatment subprotocols (each drug targets a specific alteration); each subprotocol enrolled ~35 patients (^[55]). In its first two years, MATCH accrued over 6,000 patients (^[61]). An interim analysis noted that only 12% of tumours matched an available arm (with 10 arms open) (^[11]), reflecting the rarity of many targets. The trial team thus expanded to 35 arms by 2019 (^[11]). Drilling down, a recent publication analysed tumor-specific responses in MATCH: it found several drug-tumor pairs (e.g. FGFR inhibitors in bladder cancer) with statistically significant activity that weren’t apparent from overall trial results (^[60]). This underscores a key insight: basket trials may reveal subpopulation signals even if aggregate metrics look negative.

Another example is the BRAF V600E basket trial (Hyman et al., 2015), which evaluated vemurafenib in non-melanoma cancers harboring BRAF V600E mutations (^[49]). The trial ran multiple parallel cohorts (e.g. colon, thyroid, brain) with response endpoints. It demonstrated REsponse rates in several histologies (including lung and glioma), enough to justify FDA’s later approval of vemurafenib for Erdheim–Chester disease (a rare BRAF-mutant blood cancer) based on a basket cohort (^[62]). Similarly, the series of AcSé (French) trials tested crizotinib in various rare tumors with ALK/MET/ROS1 alterations (^[63]), taking advantage of a basket structure to explore many disease types. These examples show how baskets function as “exploratory platforms” for targeted agents.

Summary: Strengths and Limitations of Basket Trials. In summary, basket trials revolutionize precision oncology by grouping patients by biology, not anatomy. They have yielded tissue-agnostic drug approvals and driven research into rare mutations (^[8]) (^[60]). Their efficiency in evaluating one drug across many cancers is significant. On the other hand, baskets demand rigorous statistical control and careful design to avoid misleading conclusions. Future baskets may increasingly use complex Bayesian approaches (^[12]) to borrow strength across cohorts, and may integrate multiple biomarkers (e.g. combining mutation + immune markers) to refine selection. Overall, baskets are ideally suited for signal-finding in genomic medicine (^[20]) (^[60]), but often need follow-up in confirmatory trials.

Umbrella Trials in Oncology

Umbrella trials take the reverse approach: one cancer type, multiple targeted therapies. Under a single protocol, patients with a given histology are molecularly profiled and assigned to the therapy arm matching their tumor’s biomarker (^[3]) (^[20]). This design mimics a “molecularly stratified” trial. Each biomarker-defined sub-study can be viewed as a mini-RCT or phase II, but all share infrastructure (screening, data systems, etc.).

Design Features. Umbrella trials typically start with a diagnostic screening step. For example, in an lung adenocarcinoma umbrella, all patients submit tissue and are tested for EGFR mutations, ALK fusions, KRAS mutations, PD-L1 expression, etc. Based on results, a patient is then sent to an arm: e.g., EGFR mutants go to an arm of an EGFR inhibitor, ALK fusions to an ALK inhibitor arm, PD-L1-high to an immunotherapy arm, etc (^[50]) (^[39]). Some trials include a “wild-type” or “non-match” arm for cases without actionable markers (e.g. Lung-MAP has a durvalumab arm for marker-negative patients (^[64])). The sub-studies are often randomized (drug vs standard of care or placebo) to confirm efficacy. Statistical endpoints are frequently clinical (PFS, OS) rather than just tumor shrinkage, since each arm is its own comparative trial. Complex umbrella platforms like MORPHEUS-type trials (e.g. NCT02332567) even use Bayesian adaptive frameworks to drop or add arms (^[65]).

Advantages of Umbrellas. Umbrella trials leverage common histology to ask multiple “Which targeted agent works best for which marker?” questions concurrently. This is efficient because patients are already engaged in one trial and routed to the appropriate treatment, rather than prescreened for various trials separately. For instance, the UK FOCUS4 trial in colorectal cancer screened newly-diagnosed metastatic patients for a panel of mutations (BRAF, PIK3CA, RAS, TP53) (^[17]). It then ran sub-trials: FOCUS4-B tested aspirin in PIK3CA-mutant CRC, FOCUS4-C tested adavosertib in TP53/RAS double-mutants, and others (^[17]). Rather than three separate studies, FOCUS4 housed them in one protocol. This multi-arm, biomarker-driven design speeds up evaluation of targeted combos in a coordinated way. It also spares patients from being randomized to irrelevant treatments: each patient only faces the therapy relevant to their tumor profile.

In practice, umbrella trials have proved nimble in adapting to new information. LUNG-MAP (squamous NSCLC) and ALCHEMIST (early lung adenocarcinoma) started before immunotherapy was standard, but were amended mid-stream. In LUNG-MAP, early substudies tested targeted drugs versus docetaxel (^[64]), but when new immunotherapies (PD-1/L1 inhibitors) were approved, the trial was restructured—docetaxel control was removed, and new immunotherapy sub-arms were added (^[64]). An immunotherapy arm for PD-L1–negative NSCLC was even created (^[64]). Similarly, ALCHEMIST (initially EGFR/ALK adjuvant arms) later added an immunotherapy arm when checkpoint inhibitors became standard (^[66]). These examples show how umbrella/platform trials can pivot, offering continuity in a rapidly changing field.

Challenges of Umbrellas. Umbrella trials also face hurdles. A key issue is patient accrual and turnover. Since patients must enroll at initial diagnosis or progression, screening failures or dropouts can be high. For example, in ALCHEMIST over 8,300 lung cancer patients had to be screened to find enough EGFR/ALK mutants (^[45]). This large funnel drives up time and cost. Dropout is also a problem: as Chen et al. note, many patients in ALCHEMIST declined further therapy after surgery and chemo, limiting accrual (^[67]). Placebo or observation arms may seem unappealing to patients (especially in adjuvant settings) (^[67]).

Statistically, umb relas can mitigate multiplicity by treating each sub-trial as a separate RCT. However, they must still plan for multiple comparisons across arms. Familywise error control is often less emphasized since each biomarker arm has its own question, but overall Type I control remains conceptually important. Some umbrella designs use adaptive randomization within arms or Bayesian stopping rules, which adds complexity (^[13]).

Case Studies: Umbrella Trials. Several landmark umbrella trials exist. BATTLE-1 (2006–2009) in advanced NSCLC was one of the first. It adaptively randomized patients to four targeted treatments (erlotinib, vandetanib, sorafenib, or erlotinib+bexarotene) based on tumor EGFR, KRAS, BRAF, VEGF, and RXR/CyclinD1 biomarkers (^[68]). Patients were reassigned probabilistically towards arms with better response. BATTLE-1 demonstrated the feasibility of adaptive randomization in a multi-arm trial (^[69]) (^[13]). Another seminal study is I-SPY 2 (ongoing), an adaptive platform trial in neoadjuvant breast cancer. Though called a platform trial, it is effectively an umbrella design for molecular signatures (HR/HER2 status, MammaPrint grade). I-SPY 2 uses Bayesian adaptive randomization to allocate patients among multiple investigational regimens (^[13]), with pathological complete response as the endpoint. Its success has encouraged similar designs in other cancer types.

In the UK, FOCUS4 is an ambitious example. It stratified 1434 screened colorectal cancer patients into multiple phases (A through D) based on biomarkers (^[17]). Three late-phase subtrials ran sequentially: FOCUS4-D (BRAF/PIK3CA/RAS wild-type vs AZD8931), FOCUS4-B (PIK3CA-mutant vs aspirin), and FOCUS4-C (TP53/RAS double-mutant vs adavosertib) (^[17]). All patients negative for the studied markers could enter a non-stratified arm (FOCUS4-N). By 2020, 361 patients (25%) had been randomized into these targeted sub-trials (^[17]). FOCUS4 also allowed adding new arms: ultimately 20 drug combinations were considered, though only 3 activated. Feedback from investigators highlighted the need for robust infrastructure, clear data workflows, and flexible designs† – echoing general lessons about master protocols (^[17]).

Summary: Strengths and Limitations of Umbrella Trials. Umbrella trials excel when a single cancer type has multiple rational targets. They allow head-to-head evaluation of personalized regimens under one organizational umbrella, and can accelerate multiple drug pathways at once. Indeed, ALCHEMIST and Lung-MAP have accelerated lung cancer drug approvals by combining forces of NCI, FDA, and industry (^[70]). On the downside, umbrellas share many challenges with baskets: complex logistics, potential for false leads in small sub-arms, and patient drop-out. Statistical correction for multiple biomarker-defined comparisons is required, and broad screening can be inefficient if many patients lack actionable mutations (as was seen when 75%+ of screened CRC patients in FOCUS4 belonged to the “no mutation” group). Nevertheless, when well-executed, umbrella trials provide a cooperative framework that benefits patients and developers by maximizing the information from each patient screened (^[71]) (^[17]).

Comparison of Basket vs. Umbrella Trials

It is instructive to directly compare basket and umbrella designs, as each has unique use-cases (Table 1 above and discussion below). In numeric terms, and based on systematic reviews, basket and umbrella trials have markedly different profiles (^[46]): in one analysis the majority of basket trials (47/49) were phase I/II and only 5% were randomized, whereas umbrella trials were more often intended as confirmatory phases and 44% (8/18) included randomization (^[46]). Basket trials tended to have smaller samples (median ~205 patients (^[46])) because each histology-specific cohort could be tiny, whereas umbrella trials ran larger (median ~346 participants) to power each arm (^[48]). Consequently, basket trial durations were shorter (median ~22 months) versus umbrella (~61 months) (^[48]). Umbrella trials also tested more interventions concurrently (median 5 different drugs) than baskets (usually 1 drug across cohorts) (^[48]).

Qualitatively, baskets and umbrellas differ in scientific focus. Basket trials ask “Is this molecular target’s role in disease biology consistent across cancers?” They are suited to phenotypically diverse patient sets. Umbrella trials ask “Which targeted therapies are effective in particular subtypes of this cancer?” They leverage a shared disease background but dissect it genetically. Baskets glean insight about genotype–phenotype relationships (e.g., “patients with ALK mutations in any tumor type”), whereas umbrellas glean insight about the landscape of effective drugs within a cancer.

Practically, this leads to different patient experiences. In a basket trial, a patient with a rare cancer may qualify by their biomarker, regardless of tumor site (^[35]) (^[20]). In an umbrella trial, a patient with common cancer qualifies by tissue type, then is stratified by marker. In either case, the chance of getting a targeted regimen (vs. standard care) is often higher than in traditional trials, which may improve enrollment appeal.

Both designs require molecular screening, but baskets often use more centralized molecular screening strategies because patients come from many histologies (e.g., a national program like NCI-MATCH (^[61])). Umbrellas may rely on pathology labs already screening, since all patients share one tumor type. On the other hand, because umbrella trials can offer randomization to potentially active treatments within a familiar disease setting, they may attract a broader investigator base (e.g., lung cancer specialists).

Table 2 (above) also listed examples of each design for clarity. Notably, some trials blur the lines. For example, NCI-MATCH is principally a basket, but it screens both solid tumors and lymphomas and could be considered “umbrella-like” due to its multi-arm nature (^[42]) (^[41]). FOCUS4 is officially an umbrella, but its adaptive, multi-arm protocol made it functionally a platform trial as well (^[17]) (^[41]). In the end, the choice between basket and umbrella (or platform) depends on the trial question and patient population.

Statistical and Methodological Considerations

Master protocols introduce unique statistical challenges beyond standard trials. Multiplicity and error control: running multiple sub-studies inflates the familywise error rate if each is tested independently. In early-phase exploratory trials this is often tolerated, but confirmatory master trials must address it. Approaches include pre-specifying hierarchical testing procedures, adjusting p-value thresholds, or using Bayesian methods. For instance, Bayesian hierarchical modelling can “borrow” strength across sub-studies, smoothing estimates and potentially improving power when effects are similar (^[12]). Thall and Berry’s methods allow sharing information across baskets and controlling error rates adaptively (^[12]). However, simulation studies (e.g. Freidlin & Korn) have shown hierarchical borrowing can sometimes incorrectly inflate type I error when subgroups truly differ (^[72]). Thus some designers prefer independent designs (e.g. one Simon two-stage per basket) with no borrowing, accepting the trade-off of larger sample sizes or missed subtle signals.

Adaptive designs: Many master protocols use adaptive rules. Common features include: interim futility stopping for a sub-arm (if it shows little activity after an initial cohort), or early success stopping (if a dramatic effect is seen). New arms can be seamlessly added via protocol amendments, as was done in FOCUS4 and Lung-MAP (^[64]) (^[17]). Bayesian response-adaptive randomization can be applied within umbrella/platform trials, assigning more patients to promising arms over time (used in I-SPY2 and BATTLE) (^[13]) (^[71]). Adaptive dose-finding or biomarker enrichment can also be embedded. These innovations complicate simulation and analysis but can greatly increase efficiency. Simon et al. and Chu & Yuan have proposed refined two-parameter Bayesian designs that adjust for multiple strata and heterogeneity (^[73]).

Patient accrual and timeline modeling: Given complex eligibility, models are needed to project recruitment. For example, clinical trial planners often combine epidemiological data on biomarker prevalence with screening capacity to estimate accrual. In masters, accrual is coupled: once a patient is screened negative for all arms, the protocol might follow them off-study, whereas basket/umbrella trials maximize the use of each screen. Availability of multi-site networks helps: NCI-MATCH leveraged 1,100 centers to get 6014 patients in 2 years (^[61]). Still, drop-out rates must be anticipated.

Data management and endpoint consistency: Having multiple sub-studies in one protocol allows standardized data collection (labs, imaging schedules, QoL metrics), which improves consistency. However, each arm may have slight differences (e.g. different imaging frequency or additional specific assessments). A unified statistical analysis plan (SAP) must account for varying endpoints. Some umbrellas combine endpoints (e.g. use PFS for most, OS for confirmatory arms). Platform trials must carefully define shared control usages to avoid erroneous comparisons between non-concurrent arms.

Overall, designing a master protocol demands a multidisciplinary team (as Burd et al. emphasize (^[71])). Experienced statisticians, clinicians, and operations experts must collaborate on common SAPs, adaptation rules, and oversight. While the initial design phase is lengthy, the payoff in trial efficiency can be enormous when done right (^[6]) (^[30]).

Regulatory and Ethical Perspectives

Regulators and ethicists have given much thought to master protocols. The FDA’s 2018 draft guidance on master protocols explicitly encourages basket and umbrella trials in oncology (^[5]), noting that they can expedite the evaluation of targeted therapies. The guidance provides recommendations on issues like pooling data, control arms, and reporting. In practice, several FDA approvals have hinged on basket/umbrella evidence (see table of pembrolizumab, larotrectinib, etc (^[8]) (^[9])). The EMA (European regulator) has similarly been open to innovative designs, though no formal EMA “master protocol guidance” exists publicly. Both agencies stress that the statistical integrity of each hypothesis test still must be maintained, and safety monitoring remains paramount.

On the ethics side, Strzebonska and Waligora have laid out concerns specific to basket/umbrella trials (^[15]). They highlight scientific validity: if subgroups are too small, the trial may lack power to find meaningful effects, potentially leading to false negatives. Conversely, conducting many small comparisons can increase false positives if no correction is made. This is an ethical issue because patients could be exposed to ineffective therapies based on unreliable endpoints. They also note issues in the risk–benefit balance: while these trials can generate knowledge on tumor diversity, individual patients may gain limited benefit if their assigned arm proves futile. For example, patients often must undergo waiting periods (weeks) for genotyping before treatment assignment (^[74]), which can delay care. The complexity of the design can also challenge informed consent. Patients may misunderstand terms like “precision” or “personalized” and believe the trial is guaranteed to target their cancer specifically. Ensuring that participants understand the adaptive nature of the trial, the possibility of no active treatment arm, and the uncertainties involved is crucial (^[75]).

To address these ethical challenges, best practice recommendations have been developed. The Blood Advances roadmap (ASCO) for master trials emphasizes several principles: careful justification of how many cohorts or arms to include, robust genomic testing protocols, and transparent consent processes (^[71]). Investigators should plan for “what-if” scenarios (e.g. if some arms grow faster than others) and establish governance (data monitoring committees) that oversee all sub-trials. Patient advocates have also stressed the need to communicate expectations: in many precision trials, the chance of receiving an experimental arm vs. standard care is higher, but outcomes are uncertain. Empowering informed decision-making is thus an ethical priority.

As Strzebonska & Waligora conclude, applying principles of validity, patient welfare, and consent is complex but achievable with diligence (^[15]) (^[75]). To date, master protocols have not run into major regulatory roadblocks in oncology—likely because of their success stories (like tissue-agnostic approvals) and the FDA’s encouragement. Nevertheless, investigators must remain vigilant about design pitfalls. For example, an umbrella trial must ensure each subgroup has a clear hypothesis and sufficient sample to justify randomization. A basket trial should predefine how to interpret results if some cohorts fail to recruit.

Case Studies and Key Examples

To illustrate how basket and umbrella trials work in practice, we review several prominent master protocols and their outcomes:

• NCI-MATCH (Master Protocol, Phase II) (^[61]) (^[11]): As discussed above, MATCH screens refractory solid tumor and lymphoma cases at 1,100 U.S. sites. It has (as of 2023) enrolled over 6,000 patients to ~35 arms (^[61]). Highlights include rapid accrual (6000 in <2 years) but an initial match rate of only 12% when only 10 arms were open (^[11]). The trial expanded to 39 arms by 2019 (^[11]). The impact of MATCH is both operational (showcasing national trials infrastructure for genomics) and scientific (e.g. confirming activity of larotrectinib and others in multi-histology cohorts). See Zhou et al. (2023) for detailed response patterns (^[60]).

• Lung-MAP (SWOG S1400, Phase II/III) (^[20]) (^[51]): An umbrella trial for previously treated squamous NSCLC, launched 2014. Patients are biomarker-screened and assigned to sub-studies targeting gene alterations (e.g. FGFR inhibitors, MET inhibitors). A non-match arm tests immunotherapy in marker-negative cases (^[64]). Lung-MAP has undergone significant modifications: originally docetaxel was control, but with the advent of PD-L1 inhibitors, the design was changed to remove chemotherapy control and add immunotherapy cohorts (^[64]). Lung-MAP exemplifies a public–private partnership (SWOG, NCI, pharma, NIH) and has multiple agents under study. Its adaptability has kept it current with standards of care.

• ALCHEMIST (NCI NCT02194738, Phase III) (^[50]) (^[45]): An umbrella trial in early-stage lung adenocarcinoma (non-squamous). Patients undergo EGFR and ALK testing after surgery/chemo; mutation-positive patients are randomized to adjuvant targeted therapy (erlotinib for EGFR, crizotinib for ALK) versus placebo (^[50]). A later amendment added a PD-1 inhibitor arm for patients negative for EGFR/ALK (^[66]). To identify enough EGFR/ALK mutants, ALCHEMIST needed to screen ~8,300 patients (^[45]), and accrual has indeed been large (screened >4,500 so far, ~100/month) with >4500 screened to date (^[66]). Challenges arose: many patients dropped out after aggressive upfront therapy (^[67]), and placebo arms were difficult to explain. Nevertheless, ALCHEMIST will generate OS evidence on using targeted adjuvant therapy, and it has demonstrated the feasibility of large-scale umbrella screening in the adjuvant setting.

• FOCUS4 (Cancer Res, UK, Phase II/III) (^[17]) (^[41]): A landmark prospective stratified umbrella trial in metastatic colorectal cancer. FOCUS4 enrolled 1434 first-line patients (Jan 2014–Oct 2020) and used molecular testing to allocate them into one of four sub-trials (B, C, D, or N) (^[17]). Each substudy investigated a targeted agent vs. control in a biomarker-defined subgroup (e.g. aspirin in PIK3CA-mutant [FOCUS4-B]). At least 20 candidate drug/biomarker pairs were considered, with 3 ultimately activated. 361 patients (25%) were randomized into sub-studies (^[17]). FOCUS4 used an adaptive multi-stage design (placeholder safety stage, then Ph2 proof-of-concept, etc.) to filter out ineffective arms early (^[17]). One of the key takeaways published by Brown et al. (2022) is that such complex trials require extensive upfront coordination, and that stakeholder feedback can generate useful guidelines for future master trials (^[17]) (^[71]).

• I-SPY 2 (Phase II breast cancer platform) (^[53]) (^[41]): Often cited as a model of adaptive master design. I-SPY 2 stratifies high-risk breast cancer patients by hormone receptor (HR), HER2, and 70-gene signature (MammaPrint) profiles. Multiple experimental drugs (e.g. neratinib, pembrolizumab, etc.) are tested concurrently against standard therapy, with Bayesian adaptive randomization favoring promising treatments in each subgroup (^[13]). Pathologic complete response (pCR) is the primary endpoint. Remarkably, I-SPY 2 has led to rapid “graduation” of regimens that went on to successful Phase III trials, demonstrating the strategy’s ability to triage leads quickly. While technically a platform trial, it functions like an umbrella for biomarker signatures (^[41]).

• BRAF V600 Basket (Phase II) (^[49]) (^[62]): Conducted by Hyman et al. (2015), this basket trial tested the BRAF inhibitor vemurafenib in patients with diverse non-melanoma cancers harboring BRAF V600E. It consisted of parallel cohorts for each tumor type (e.g. NSCLC, thyroid, glioma). Notably, it demonstrated substantial responses in some “rare” contexts, including rare histiocytoses (^[62]). One cohort (Erdheim–Chester disease) led directly to FDA’s full approval of vemurafenib for that indication (^[62]). This trial underscores how baskets can yield high-impact outcomes even in minuscule patient groups.

• MSK-IMPACT/NCI Genomic Tumor Board Platforms (Practice Models): Though not strictly trials, programs like MSK-IMPACT and the Pediatric MATCH embody the basket philosophy in clinic. Clinicians sequence patients with tumor-profiling panels; actionable results lead to treatment within dedicated subprotocols (even off-trial), or recommendations to clinical trial arms (^[76]). These have shown the feasibility of matching patients to therapies based on genetics at scale. The Pediatric MATCH trial (NCT03155620) is a pediatric basket trial mirroring NCI-MATCH, enrolling children with refractory solid tumors and assigning them by mutation (^[77]).

Each of these examples highlights different applications: basket trials for histology-agnostic questions (NCI-MATCH, Larotrectinib), and umbrella trials for exploring multiple agents in one cancer (ALCHEMIST, Lung-MAP, FOCUS4). Collectively, they provide case studies on how master protocols can streamline development. Many of them also illustrate iterative refinement (amendments for new biomarkers, adaptive arms, etc.) in response to emerging science.

Implications, Challenges, and Future Directions

The rise of basket and umbrella trials has profound implications for oncology research. These designs promise to accelerate drug development by answering many questions in parallel (^[6]) (^[30]). They encourage data sharing across arms and institutions, create ebbs in redundancy (one infrastructure instead of many), and keep patients at the center of precision care. Master protocols have catalyzed a “learning health system” mindset, where clinical research and routine care overlap via genomic screening networks.

However, broad adoption faces hurdles. Operationally, as Burd et al. emphasize, master trials require extensive coordination (^[71]). Setting up a single trial to cover multiple drugs involves multiple sponsors, different drug supply chains, and complex regulation. Tracking real-time enrollment in each sub-study is nontrivial. These trials also demand rigorous informatics (e.g. databases linking sequencing results to trial slots). Financially, it is challenging to fund a large, multi-arm trial; often public–private partnerships or multi-industry consortia are needed (as seen in Lung-MAP and ALCHEMIST (^[70])).

From a patient perspective, master protocols can improve access: one screening test can open doors to multiple trials. But equity is a concern: broad access to genomic testing is still uneven globally. In low-/middle-income settings, these designs are rare due to resource limits (^[78]). As precision medicine globalizes, building capacity for master protocols worldwide will be important.

Statistically, new methods will emerge. Current research aims to refine how to do multiplicity correction across many substudies. Bayesian decision criteria for arm continuation/discontinuation will likely become standard. Machine learning may help predict which combinations or cohorts to prioritize. At the same time, rigorous pre-specification of rules will be needed to convince regulators of validity.

Regulators will also face novel questions. For example: how to label a drug approved under a basket? The tissue-agnostic label was a new paradigm for FDA. Future approvals might be conditional on biomarkers (e.g. “drug X for patients with mutation Y, irrespective of tumor type”). Suppliers of diagnostics will be essential partners. On the ethics side, demonstrating patient benefit in each sub-cohort will remain under scrutiny. Long-term follow-up across sub-studies introduces data-management demands for regulators and trialists.

Looking ahead, master protocols are proliferating. Beyond oncology, similar frameworks are used in other fields (e.g. infectious diseases like COVID-19, neurodegeneration, etc.). In oncology, they may expand to incorporate multi-omic profiling (adding RNA, proteomics) and liquid biopsies for enrollment. Trials may evolve to platform designs that are truly perpetual, with infrastructure in place to launch new arms continuously (the “flagship” model). Training the next generation of trialists and statisticians in these complex methods is also critical (^[79]) (^[80]).

In the near term, we anticipate more master protocols targeting combinations (e.g. immunotherapy plus targeted agents) and understudied cancers (e.g. rare sarcomas, pediatric tumors). Pharmaceutical companies are increasingly willing to pool pipeline assets into joint master protocols to reduce competition and share costs. Overall, while basket and umbrella trials are still evolving tools, their trajectory is one of rapid growth and increasing impact. They bring us closer to the promise of precision medicine by matching the right drug to the right patient in the right trial – often concurrently.

Conclusion

Basket and umbrella (master protocol) trials represent a fundamental innovation in oncology clinical research. By re-architecting trials around molecular biology rather than organ of origin, they have enabled faster, more flexible testing of targeted therapies. From the first baskets in the 2010s to today’s large-scale adaptive platforms, this master-protocol approach has matured into a vital part of the oncology landscape. It has delivered high-profile successes (e.g., tissue-agnostic drug approvals (^[8]) (^[9])) and is reshaping how we think about trial efficiency.

While challenges remain—statistical, operational, ethical—the benefits are compelling. These include enrolling rare-patient subgroups, sharing resources, and accelerating discovery on multiple fronts simultaneously (^[6]) (^[30]). Expert consensus and case studies show that with careful planning, coordination, and innovative analytics, these designs can succeed on a large scale (^[71]) (^[17]).

As precision oncology continues to evolve, mast er protocols will likely expand their role. Advances in genomics, imaging, and data science will feed into richer eligibility criteria and endpoints. Regulatory frameworks are adapting in kind (^[5]) (^[8]). The future may see master protocols become the norm rather than the exception – a new standard for how to efficiently evaluate cancer therapies. Long-term, this could transform drug development timelines, lower costs, and ultimately improve patient outcomes by delivering effective, personalized treatments more quickly.

References: Detailed references for all data and claims are provided in the text above, e.g. (^[4]) (^[25]) (^[8]) (^[15]) (^[17]), etc., corresponding to peer-reviewed studies and regulatory documents. These include systematic reviews on master protocols (^[4]) (^[81]), overviews of trial designs (^[28]) (^[2]) (^[30]), and case reports of specific trials (^[35]) (^[50]) (^[61]), ensuring each statement above is supported by authoritative sources.