The shift towards precision oncology has fundamentally changed how clinical trials are designed. Where trials once enrolled broad populations defined by tumour histology, contemporary early-phase studies frequently restrict enrolment to patients who meet a molecular eligibility criterion — a gene alteration, protein expression level, or immune signature — predicted to identify those most likely to benefit. Biomarker-driven patient selection is now the norm in early-phase oncology, and an increasingly important consideration in Phase III design.
The distinction between predictive and prognostic biomarkers is central to trial design and is frequently conflated in the literature.
Confusing these two categories leads to important errors. A trial that restricts enrolment to high-risk patients based on a prognostic marker, then attributes improved outcomes to the treatment, may produce spuriously positive results. The appropriate statistical test for a predictive biomarker is a treatment-by-biomarker interaction — not a comparison of outcomes within the biomarker-positive subgroup.
Two related trial architectures have emerged to efficiently test biomarker-driven hypotheses across multiple oncology settings:
Basket trials enrol patients with a shared molecular alteration regardless of tumour histology. The premise is that a cancer defined by, for example, BRAF V600E mutation may behave similarly — and respond similarly to BRAF-directed therapy — whether it arises in melanoma, colorectal cancer, or thyroid cancer. NCI-MATCH and the vemurafenib basket trial in BRAF-mutant non-melanoma cancers established this design. The key analytical challenge is handling heterogeneous response rates across histological baskets — a Bayesian hierarchical model that borrows information across baskets is now commonly used.
Umbrella trials work in the opposite direction: a single histological tumour type is molecularly profiled, and patients are assigned to different treatment arms based on their tumour's genetic profile. LUNG-MAP and I-SPY 2 are established examples. The infrastructure is shared but the treatment hypotheses are distinct across arms.
Key consideration for site investigators: Both basket and umbrella designs require prospective tumour molecular profiling as a screening step — often using next-generation sequencing. Sites must have access to a CLIA-certified sequencing laboratory and be prepared for the lag time between biopsy and molecular results before eligibility can be confirmed.
When a biomarker is integral to the treatment hypothesis — where the drug is only expected to work in biomarker-positive patients — it must be co-developed alongside the therapy. FDA's companion diagnostic (CDx) framework applies when a biomarker test is required to identify the patient population for whom the drug is approved. The CDx must be analytically validated, clinically validated for the specific drug-biomarker pair, and approved at the same time as the drug.
Early-phase trials should be designed with co-development in mind from the outset. This means selecting an analytically robust assay early, including biomarker-negative patients in Phase II if the goal is to demonstrate a differential treatment effect, and pre-specifying the biomarker threshold used for patient selection — not retrospectively optimising it.
FDA's Biomarker Qualification Programme (BQP) provides a regulatory pathway to formally qualify a biomarker for a specific context of use — independent of any individual drug application. A qualified biomarker can then be used by any sponsor in trial design without re-validating it from scratch. The qualification dossier requires a substantial evidence package, and the process is lengthy, but qualified biomarkers carry significant regulatory weight.
For sponsors who do not seek formal qualification, FDA's 2018 guidance on clinical trial enrichment strategies provides practical direction on selecting and validating enrichment biomarkers, including the use of an adaptive enrichment design — where the trial begins in a broader population and may narrow to a biomarker-defined subgroup at an interim analysis if differential efficacy is observed.
Sites participating in biomarker-selected trials face operational demands that differ significantly from conventional histology-defined trials. Investigators should expect higher screen failure rates — often 50–80% in molecularly selected studies — with corresponding implications for enrolment timelines. Tissue acquisition for molecular testing, including sufficient archival material or a fresh mandatory biopsy, must be factored into the site qualification assessment. Liquid biopsy-based eligibility criteria, while less invasive, introduce their own analytical variability that sites must understand and document.
KCLEAGENICS MEDICAL's precision oncology programme includes biomarker-stratified designs across our active portfolio. For clinicians interested in co-investigator participation or patient referral to our biomarker-selected studies, please contact our clinical research team for current eligibility criteria and site capability requirements.
Adaptive Platform Trial Design
For CliniciansPre-specification in Clinical Trials
For CliniciansMetabolic Risk and Haematological Malignancy
Subscribe to our monthly Clinical Intelligence Briefing
