Biomarker Validation: Analytical vs Clinical Validation, Step by Step

Biomarker Validation: Analytical vs Clinical Validation, Step by Step

You have a promising biomarker and data behind it. The harder part is turning it into evidence that a regulator and a treating oncologist will trust. That is what biomarker validation is for, and where many programmes stall.

The cause is rarely weak science. More often the validation strategy is built too late, defined too loosely, or cannot survive the move from a discovery lab into a regulated, multi-site trial. This guide sets out the full validation ladder, how fit-for-purpose thinking sets the evidence bar, how assays move from research use to a companion diagnostic, and what immuno-oncology learned from PD-L1 and tumour mutational burden (TMB). It ends with a practical checklist.

Key takeaways

• Biomarker validation is three separate questions: does the assay measure the analyte reliably (analytical validation), does the result relate to the clinical outcome you claim (clinical validation), and does acting on it improve patient care (clinical utility). Passing one does not give you the next.
• The context of use (COU) statement is the foundation. Write it early and let it decide the depth of every validation step that follows.
• Validation is fit-for-purpose and staged: an exploratory phase 1 assay needs far less than one that selects patients in a pivotal trial.
• A predictive claim needs a control arm, pre-specified hypotheses and a biomarker-by-treatment interaction.
• The Blueprint PD-L1 and Friends of Cancer Research TMB projects exist because different tests gave different answers for the same patients.

What is biomarker validation?

Biomarker validation is the process of generating evidence that a biomarker, measured by a specific assay, is reliable and meaningful enough for a defined decision. It is never validated in the abstract, only for a particular use, population and test.

The standard reference for terminology is the BEST (Biomarkers, EndpointS, and other Tools) Resource, from the FDA-NIH Biomarker Working Group (first published 2016). It separates three ideas often blurred in practice:

• Analytical validation: establishing that the assay’s performance (accuracy, precision, sensitivity, specificity, range) is acceptable for its purpose.
• Clinical validation: establishing that the test result identifies, measures or predicts the clinical or biological state or outcome of interest in the intended population.
• Clinical utility: showing that using the test to guide decisions improves outcomes or the balance of benefit and harm.

Biomarker types: the BEST categories

Clinical biomarkers are classified by use, not by what they measure, and the same molecule can sit in more than one category with different evidence needs. The FDA’s summary of BEST lists seven categories.

The category decides the study design you will need: a prognostic claim can rest on single-arm or observational data, a predictive claim cannot.

Step 1: Write the context of use statement first

The context of use (COU) statement defines what your biomarker does, in which patients, at what stage and for what decision. The assay platform, depth of validation, substudy powering, pre-analytical controls and regulatory expectations all flow from it.

Step 2: Apply fit-for-purpose validation

Fit-for-purpose biomarker validation means the depth of validation matches the weight of the decision the data will support. The approach was set out in an influential 2006 paper by Lee and colleagues in Pharmaceutical Research, which grew out of an AAPS workshop and described validation as iterative: you validate to the level needed now, then extend it as the biomarker’s role grows.

FDA has since put the same principle into guidance. Its guidance Bioanalytical Method Validation for Biomarkers (issued January 2025, current version dated April 2026) points sponsors to ICH M10 as the starting point for biomarker assays used in regulatory decision-making, while acknowledging that some M10 criteria, written for drug concentration assays, may not apply to every biomarker platform. Sponsors are expected to justify their approach, which matters because biomarker assays measure endogenous analytes, often without a true reference standard.

Step 3: Analytical validation, parameter by parameter

Analytical validation proves the assay works, not that the biomarker works clinically. The parameters below apply, with platform-specific adjustments, to immunoassays, immunohistochemistry (IHC), flow cytometry and next-generation sequencing (NGS).

For diagnostics teams, the Clinical and Laboratory Standards Institute (CLSI) evaluation protocols are the common reference for designing these studies, including EP05 (Evaluation of Precision of Quantitative Measurement Procedures), EP06 (Evaluation of Linearity of Quantitative Measurement Procedures), EP07 (Interference Testing in Clinical Chemistry) and EP17 (Evaluation of Detection Capability). Check the current edition of each before writing protocols.

Pre-analytical variables are part of analytical validation

Collection, fixation, freezing, shipping and processing directly affect performance: cold ischaemia time, fixation time and block age for tissue; tube type, centrifugation, time to processing and shipping temperature for liquid biopsy. Define these limits during validation and write them into the lab manual.

Step 4: Clinical validation

Clinical validation shows that the test result is associated with the outcome you claim, in the intended population, using sensitivity, specificity and predictive values for selection claims, or hazard ratios and interaction tests for prognostic and predictive claims.

Prognostic or predictive?

A prognostic biomarker tells you about outcome regardless of treatment. A predictive biomarker tells you which patients benefit from a specific treatment. To validate a predictive claim you need a control arm, a pre-specified biomarker hypothesis and statistical evidence of a biomarker-by-treatment interaction. Without a control arm, a marker that looks predictive in a single-arm study may simply be prognostic: biomarker-positive patients do better on any therapy.

Any analysis not pre-specified in the protocol and statistical analysis plan will be treated as exploratory. Exploratory data can generate a hypothesis. They cannot validate a biomarker for labelling.

The cut-off is part of the claim

Choosing a threshold on the same data used to prove it works inflates performance. Derive it in a training set and confirm it in an independent one, and plan for samples near the threshold, where assay and reader variability become misclassification.

Build the substudy into trial design from the start

• Lock the assay, or at least the clinical trial assay version, before enrolment begins.
• Power the substudy for the biomarker endpoint, not only the overall trial endpoint.
• Account for sample attrition: missing, insufficient and failed samples erode biomarker-evaluable populations quickly.
• Establish central laboratory infrastructure and sample logistics early.

Design choices (enrichment, stratified, all-comers or adaptive) are covered in our guide to biomarker-driven clinical trials.

Tumour heterogeneity belongs in the same plan: pre-specify which sample type is the primary biomarker source, and use serial or liquid biopsy sampling where change over time matters.

Step 5: Clinical utility

Clinical utility is the rung most often skipped and the one payers care about most. A biomarker can be analytically sound and clinically valid yet have no utility if acting on it does not change outcomes. For a companion diagnostic, utility is largely shown through the drug trial itself. For complementary diagnostics, monitoring tests and prognostic tools, it usually has to be shown separately, through decision-impact studies, real-world evidence or health economic analysis. The distinction between the two diagnostic types is set out in our article on companion vs complementary diagnostics.

Step 6: Plan the assay transition from RUO to IVD

Biomarker development rarely uses one assay version from discovery to approval. Each transition, from research use only (RUO) to clinical trial assay to IVD or CDx, is a point where data can be lost if it is not planned.

Plan early for:

• Investigational device rules. FDA’s guidance on investigational IVDs used in clinical investigations of therapeutic products explains that when a test result decides a participant’s treatment, the use may be significant risk and need an investigational device exemption.
• Bridging studies. If the commercial CDx differs from the trial assay, you will need a concordance study on consented, banked trial samples.
• The LDT environment. FDA’s 2024 rule to regulate laboratory developed tests as medical devices was vacated by a federal court in Texas on 31 March 2025, and FDA formally rescinded it later in 2025. LDTs remain under CLIA. A drug label that depends on a test still needs an FDA-approved or cleared companion diagnostic. In the EU, the IVDR sets its own requirements; see our IVDR companion diagnostics guide.

The full regulatory route for a CDx is set out in our step-by-step guide to the FDA companion diagnostic approval process.

Step 7: Engage regulators early, through the right route

For a biomarker tied to one drug, the route is the drug programme itself: pre-IND and Type C meetings with CDER, plus a Pre-Submission (Q-Sub) to CDRH on the device validation plan before pivotal studies. A separate route exists for biomarkers meant to be used across many programmes.

The FDA Biomarker Qualification Program

Under section 507 of the Federal Food, Drug, and Cosmetic Act, added by the 21st Century Cures Act, the Biomarker Qualification Program uses a three-stage process: a Letter of Intent, a Qualification Plan and a Full Qualification Package. Once qualified, a biomarker can be used for its stated COU in any drug programme without the sponsor re-establishing its suitability.

In practice, qualification is slow and relatively few biomarkers have completed it. Most oncology predictive biomarkers reach the clinic through drug-diagnostic co-development instead, and qualification suits consortia developing safety, pharmacodynamic or enrichment biomarkers with broad use. Friends of Cancer Research has published a white paper on modernising the programme. Whichever route you take, arrive with a written COU, an analytical validation plan and a clear statement of the claim, not only data.

Step 8: Harmonise across sites and assays before the trial starts

Running a biomarker substudy across global sites is a different challenge from validating an assay in one lab. Immuno-oncology has two cautionary tales.

PD-L1 and the Blueprint project

Each early PD-1/PD-L1 inhibitor was developed with its own IHC assay, scoring algorithm and cut-off. Phase 1 of the Blueprint PD-L1 IHC Assay Comparison Project (Hirsch et al., Journal of Thoracic Oncology, 2017) stained 39 non-small cell lung cancer samples with the 22C3, 28-8, SP142 and SP263 assays. 22C3, 28-8 and SP263 stained tumour cells comparably, SP142 stained fewer, and immune cell staining varied more. In 37% of evaluable cases, a different PD-L1 classification would have resulted depending on the assay and scoring system used. Phase 2 confirmed this in routine clinical samples and added the 73-10 assay.

The lesson: assay, scoring algorithm and cut-off are one package. A 2019 analysis by Davis and Patel of 45 FDA approvals of checkpoint inhibitors found PD-L1 was predictive in only 28.9% of them, not predictive in 53.3% and not tested in 17.8%. More on the biology and the assays is in our article on PD-L1 as a predictive biomarker.

TMB and the Friends of Cancer Research harmonisation project

FDA approved pembrolizumab for TMB-high (at least 10 mutations per megabase) solid tumours in June 2020, with FoundationOne CDx as the companion diagnostic. But panel TMB values are not automatically comparable, because panel size, coverage and bioinformatics pipelines differ.

The Friends of Cancer Research TMB Harmonization Project addressed this in two phases. Phase 1 (Merino et al., Journal for ImmunoTherapy of Cancer, 2020) compared panel TMB with whole exome sequencing in silico using TCGA data and proposed reporting and validation recommendations. Phase 2, published in Annals of Oncology in 2021, used reference materials to align panel TMB across laboratories. Our TMB explainer covers what this means for clinical use.

For any multi-site programme: run a ring study before the first patient, certify readers for subjective scoring, and treat harmonisation as a trial design problem.

Biomarker validation checklist

Use this before IND submission, pivotal protocol lock or a CDx partnership agreement.

Where biomarker validation is debated in 2027

Validation strategy is a central thread in Stream 2: Biomarker & Companion Diagnostics at the immuno-oncology and biomarker conference in London, which runs on 3–4 March 2027 as the 2nd annual UK edition, with 400+ senior attendees expected. The published Stream 2 agenda includes a keynote on “Clinical Utility of Biomarkers” from Espen Walker, Head of Global Medical Diagnostics, AstraZeneca, a “Panel: Regulatory Harmonisation for Biomarker Qualification”, and a keynote on “Emerging Predictive & Prognostic Biomarkers in IO”. Stream 1 covers next-generation immuno-oncology in parallel, with keynotes, panels, case studies, pre-arranged 1:1 meetings and a Day 1 drinks reception. The 2026 London edition drew 300+ attendees from 100+ companies, 70% of them senior decision-makers. For the sponsor view of the series, see our biomarker conference page.

Register for ImmunoMark London 2027

If your company wants to meet this audience directly, whether you develop assays, run central laboratory or CRO services, or build companion diagnostics, see the London 2027 sponsorship opportunities.

FAQs

What is the difference between analytical and clinical validation?

Analytical validation shows that the assay measures the biomarker accurately and reliably: precision, accuracy, limit of detection, linearity, specificity, stability and reproducibility across sites. Clinical validation shows that the test result is associated with the clinical state or outcome you claim in the intended population. You need both.

What does fit-for-purpose biomarker validation mean?

It means the depth of validation matches the decision the data will support. An exploratory biomarker used for internal decisions needs only enough validation to trust the result, while an assay that selects patients for treatment needs full analytical and clinical validation.

What are the BEST biomarker categories?

The FDA-NIH BEST Resource defines seven categories: susceptibility/risk, diagnostic, monitoring, prognostic, predictive, pharmacodynamic/response and safety. The category matters because it determines the evidence needed; for example, a predictive claim requires a control arm and evidence of a biomarker-by-treatment interaction.

Do I need the FDA Biomarker Qualification Program to use a biomarker in a trial?

No. Most oncology biomarkers are accepted within a specific drug development programme through IND interactions and, for a companion diagnostic, a device submission. Qualification is a separate three-stage process (Letter of Intent, Qualification Plan, Full Qualification Package) for biomarkers intended to be used across many drug programmes for a stated context of use.

Why did PD-L1 and TMB need harmonisation projects?

Because different tests gave different answers for the same patients. The Blueprint project found that PD-L1 IHC assays and scoring systems could classify the same lung cancer samples differently, and the Friends of Cancer Research project found that TMB values vary across NGS panels and pipelines. Both show that assay, scoring method and cut-off must be validated together.