Biomarker-Driven Clinical Trials: Basket, Umbrella & Enrichment Designs

Biomarker-Driven Clinical Trials: Basket, Umbrella & Enrichment Designs

Most oncology programmes now carry a biomarker hypothesis from first-in-human onwards. The question is no longer whether to use a biomarker in the trial, but how: to select patients, to stratify them, to test one drug across many tumour types, or to test many drugs within one disease. Each choice shapes sample size, screening burden, the label and the companion diagnostic.

This guide for clinical development, translational and CRO teams defines the main designs for biomarker driven clinical trials, the trials that set the template, and the statistical, operational and regulatory issues that decide whether a biomarker strategy holds up at submission.

Key takeaways

• Enrichment designs enrol only biomarker-positive patients; stratified designs enrol all-comers and randomise within biomarker strata.
• Basket trials test one therapy across tumour types sharing an alteration; umbrella trials test several therapies within one tumour type; platform trials add and drop arms under a single master protocol.
• NCI-MATCH data contributed to the tumour-agnostic approval of dabrafenib plus trametinib for BRAF V600E solid tumours.
• Low prevalence drives screening volume and cost. NCI-MATCH matched about four in ten patients to a treatment arm, ranging from 6% in pancreatic cancer upwards.
• Assay choices made in phase 2 (central vs local testing, clinical trial assay vs final companion diagnostic) can force a bridging study later. Plan them early.
• Key FDA references: the 2019 enrichment guidance, the 2022 oncology master protocols guidance and the June 2026 revised master protocols draft.

What makes a clinical trial biomarker driven?

A trial is biomarker driven when a prespecified biomarker determines who is enrolled, how patients are allocated, or how the primary analysis is structured.

FDA’s guidance Enrichment Strategies for Clinical Trials to Support Determination of Effectiveness of Human Drugs and Biological Products (final, March 2019) defines enrichment as “the prospective use of any patient characteristic to select a study population in which detection of a drug effect (if one is in fact present) is more likely than it would be in an unselected population.” It separates variability-reducing, prognostic and predictive enrichment; most oncology biomarker work is predictive.

Our guide on how to validate an IO biomarker in a clinical trial covers the validation path the biomarker itself needs.

The five designs compared

Enrichment and stratified designs

The real question is what you must show about biomarker-negative patients. FDA’s 2019 guidance notes that “a critical question is almost always how much data will be needed in the off-target (nonenriched) population, particularly for predictive enrichment strategies.” If the biomarker is a continuous measure with an uncertain threshold, an enriched trial can lock you into a cut-off you later regret. A stratified design costs more but tells you where the effect actually starts.

Basket trials

A basket trial enrols patients by molecular alteration rather than by tumour site. The larotrectinib accelerated approval for NTRK gene fusion-positive solid tumours (26 November 2018) pooled three single-arm trials (LOXO-TRK-14001, SCOUT and NAVIGATE), with an objective response rate of 75% in 55 patients. Dabrafenib plus trametinib received tumour-agnostic accelerated approval for BRAF V600E solid tumours (excluding colorectal cancer) on 22 June 2022, supported by NCI-MATCH, ROAR and a paediatric study.

Umbrella trials

Umbrella trials screen one disease and route patients to sub-studies by biomarker. Lung-MAP (SWOG S1400) was the first biomarker-driven master protocol initiated with the US National Cancer Institute. From June 2014 to January 2019 it enrolled 1,864 patients with previously treated squamous NSCLC. In early 2019 a new screening protocol expanded Lung-MAP to all NSCLC histologies and added liquid biopsy.

Platform trials

I-SPY 2 in neoadjuvant breast cancer is the reference example. It classified tumours into eight subtypes by HER2, hormone receptor and 70-gene assay status, updated randomisation probabilities as pathological complete response data accrued, and “graduated” a regimen when its Bayesian predictive probability of success in a 300-patient phase 3 reached 85%. In 2016 the New England Journal of Medicine reported graduations for neratinib in HR-negative, HER2-positive disease and for veliparib plus carboplatin in triple-negative disease.

NCI-MATCH, ComboMATCH and KEYNOTE-158

NCI-MATCH’s successor, ComboMATCH, tests phase 2 combinations through a registration protocol (EAY191): sequencing at a designated laboratory, then algorithmic assignment weighing histology, variant, allele frequency and level of evidence.

Statistical and operational considerations

Prevalence and screen failure

Prevalence sets the screening multiple. A biomarker present in 5% of patients means roughly 20 screened per patient enrolled, before assay failure and insufficient tissue. NCI-MATCH’s match rate ranged from 6% in pancreatic cancer to over 25% in several other cancers. Lung-MAP lost patients at every step: 98.9% had tissue, 90.9% of those had biomarker results, and 83.9% of those were assigned. Budget screening as its own cost line.

Central vs local testing

Central testing gives one validated assay but adds shipping time and site friction; local testing is faster but heterogeneous. Accepting local results for eligibility and confirming centrally keeps speed and generates the concordance data a companion diagnostic submission needs. Liquid biopsy can cut screen failure for insufficient tissue; Lung-MAP added it in 2019. See our article on ctDNA and MRD in immunotherapy trials.

Assay concordance and bridging studies

If patients were enrolled with a clinical trial assay that differs from the final companion diagnostic, or with several local assays, you will usually need a bridging study: retesting stored samples on the final device to show the trial results apply to patients selected by it. That needs banked tissue from screened patients, including biomarker-negative ones, so build sample retention into the phase 2 protocol. FDA’s draft guidance Principles for Codevelopment of an In Vitro Companion Diagnostic Device with a Therapeutic Product (July 2016) covers clinical trial assays and device exemptions; see also our FDA companion diagnostic approval guide.

Multiplicity and small cohorts

Basket trials generate many small single-arm cohorts, and histology still matters: dabrafenib plus trametinib is not indicated in BRAF V600E colorectal cancer because of intrinsic resistance to BRAF inhibition. Bayesian models that borrow information across baskets help, but borrowing should be prespecified.

Why IO biomarker trials are different

Targeted therapy biomarkers are usually binary. Immuno-oncology biomarkers are mostly continuous, and the cut-off is a design decision.

PD-L1. Different checkpoint inhibitors were developed with different assays (22C3, 28-8, SP142, SP263) and scoring systems. Phase 1 of the Blueprint PD-L1 IHC Assay Comparison Project found that 22C3, 28-8 and SP263 stained tumour cells comparably while SP142 stained fewer, and that in 14 of 38 cases (37%) a different PD-L1 classification would result depending on the assay and scoring system used. Fix the assay, scoring algorithm and threshold before enrolment. See our piece on PD-L1 as a predictive biomarker.

TMB. The TMB-high approval for pembrolizumab (16 June 2020) came from a prospectively planned, retrospectively assessed analysis of ten KEYNOTE-158 cohorts, using FoundationOne CDx and a prespecified cut-off of at least 10 mutations per megabase. The response rate was 29% in the 102 TMB-high patients versus 6% in the 688 non-TMB-high patients. The threshold is still debated across tumour types and panels. Our TMB explainer covers harmonisation.

MSI-H/dMMR. The first tumour-agnostic approval of any kind was pembrolizumab for MSI-H or dMMR solid tumours on 23 May 2017, based on 149 patients across five single-arm trials including KEYNOTE-158. Dostarlimab followed for dMMR solid tumours in August 2021.

The lesson: use a stratified or biomarker-adaptive design early so the data show where benefit begins.

Regulatory guidance to know

For European programmes, see our IVDR companion diagnostics guide.

Discuss trial design with the people running these programmes

At the immuno-oncology and biomarker conference in London on 3–4 March 2027, Stream 1: Next-Gen Immuno-Oncology includes the session “Clinical Trial Design in Immunotherapy: Adaptive & Biomarker-Driven Models”, while Stream 2: Biomarkers & Companion Diagnostics runs “The Future of Biomarker-Driven Oncology” and “Panel: Regulatory Harmonisation for Biomarker Qualification”. Expect keynotes, panels, case studies, pre-arranged 1:1 meetings and a Day 1 drinks reception, with 400+ attendees targeted.

If you are US-based, the San Diego edition on 10–11 February 2027 includes “Artificial Intelligence for Optimising Clinical Trial Design and Patient Selection” in Stream 1.

Register for ImmunoMark London 2027

If your company wants to meet this audience directly, whether you run biomarker-driven trials as a CRO, supply central lab and NGS testing, or develop companion diagnostics, see the London 2027 sponsorship opportunities or our overview of oncology conference sponsorship.

FAQs

What is the difference between a basket trial and an umbrella trial?

A basket trial tests one therapy in patients who share a molecular alteration across several tumour types, such as NTRK fusions or BRAF V600E. An umbrella trial tests several therapies within a single tumour type, assigning patients to sub-studies according to their biomarker, as Lung-MAP did in non-small cell lung cancer. Both usually run under a master protocol.

What is an enrichment design in clinical trials?

FDA defines enrichment as the prospective use of any patient characteristic to select a study population in which a drug effect, if present, is more likely to be detected than in an unselected population. In oncology this usually means predictive enrichment: enrolling only biomarker-positive patients. The trade-off is limited data in biomarker-negative patients, which can restrict the label.

What is a platform trial in oncology?

A platform trial is a master protocol that evaluates multiple treatments over time, adding new arms and dropping or graduating others according to prespecified rules, often with a shared control arm. I-SPY 2 in neoadjuvant breast cancer is the best-known example, using Bayesian adaptive randomisation by biomarker signature.

Which drugs have tumour-agnostic approvals?

FDA tumour-agnostic approvals include pembrolizumab (MSI-H/dMMR, 2017; TMB-high, 2020), larotrectinib (2018) and entrectinib (2019) for NTRK fusions, dostarlimab (dMMR, 2021), dabrafenib plus trametinib (BRAF V600E, 2022), selpercatinib (RET fusions, 2022) and trastuzumab deruxtecan (HER2 IHC 3+, 2024).

When is a bridging study needed for a companion diagnostic?

A bridging study is usually needed when patients were enrolled using a clinical trial assay or local tests that differ from the final companion diagnostic. Stored samples are retested on the final device to show that the trial results apply to patients it selects. Retaining tissue from screened patients, including biomarker-negative ones, makes this possible.