Biomarker strategy, explained.

Every measurement in a drug program needs a reason and a date: what a biomarker strategy does

One biomarker's path, from a promising paper to a decision in a trial

1. Someone finds a candidate markerA paper or an early study suggests that plasma p-tau217 tracks amyloid plaques in the brain.
2. The team names the decision it will support"Who goes on to a PET scan in study 201." This sentence, the context of use, sets everything after it.
3. The strategy ties marker, assay, evidence and date togetherWhich test, which laboratory, what validation, and when it is ready against first participant in. This is where a late assay is caught, or missed.
4. The laboratory validates and transfers the assayPerformance is established for this use, in the laboratory that will report results for real people.
5. The study runs on itFrom here, a wrong cut-off or a late test costs scans, screen failures, months and sometimes a bridging study, which links a new test to the old one, years later.

Step 3 is the subject of this guide. The program and dates in this example are fictional.

What newcomers get wrong about qualified markers, cleared tests and planning time

Context of use, fit for purpose, CLIA, companion diagnostic: what each means in a biomarker plan

Fourteen terms are defined in the one-page summary.

The BEST glossary's seven biomarker categories, and how they map to a program's four lanes

In shortFDA and the National Institutes of Health (NIH) publish one shared vocabulary, the BEST glossary, with seven categories of biomarker. FDA's draft qualification framework asks that a biomarker be classified by one of them. Teams then often plan in four working lanes: target engagement, pharmacodynamics, patient selection and safety. The categories are the regulator's words, and the lanes are a planning view.

BEST stands for Biomarkers, EndpointS, and other Tools. The FDA-NIH Biomarker Working Group created the resource on January 28, 2016. Listings we could read show its glossary chapter last updated on November 29, 2021, and other chapters carry other dates.8 It is a glossary, so it settles what words mean and sets no evidence bar.

The glossary defines a biomarker as "a defined characteristic that is measured as an indicator of normal biological processes, pathogenic processes", or of responses to an exposure or intervention.9 FDA adds the limit: "A biomarker is not an assessment of how an individual feels, functions, or survives."2 FDA's framework draft lists the seven categories and says a biomarker "should be classified" by one.10 We read the glossary's wording through the sources cited here, and we could not open its own page. Check it there before quoting it in a submission.

Writing a biomarker context of use: one category, one purpose, one stage, one population

In shortThe context of use is the sentence that says how a marker will be used. US law defines the term, FDA's draft framework gives it two components and four elements, and a good one sets how much evidence and validation the marker owes. Write it first, before the literature review.

Section 507 of the Federal Food, Drug, and Cosmetic Act, added by the 21st Century Cures Act on December 13, 2016, defines context of use as "the circumstances under which the drug development tool is to be used in drug development and regulatory review".11

FDA's framework draft of December 2018, still a draft, calls it "a concise description of a biomarker's specified use in drug development" with two components: "(1) the biomarker category and (2) the biomarker's proposed use".10 For qualification, it adds that each effort "should identify a single COU", meaning one context of use.10

Four context-of-use statements from one fictional Alzheimer's program

The FDA biomarker qualification program: three stages, few qualified, and why most programs skip it

In shortQualification makes a biomarker usable by anyone for a stated use, without re-arguing the evidence. It runs in three stages set by statute, and measured review times run about double FDA's targets. It suits consortia and shared problems. A single program can usually argue its marker in its own application.

The statute defines qualification as a determination that a tool and its context of use "can be relied upon to have a specific interpretation and application in drug development and regulatory review".11 Once qualified, a tool "may be used by any person in such context of use".11 FDA may rescind or modify a determination on new information, so a strategy that relies on one should record when it last checked.11

ICH E16, a guideline from the International Council for Harmonisation (ICH), describes qualification in the same terms, for genomic biomarkers. It was final at ICH in August 2010, and FDA issued it in August 2011.13

When a program needs qualification, and when its own application is enough

FDA's framework notes that biomarkers have long been accepted within specific development efforts, with qualification addressing broader use.10

In practice the question for one program is whether anyone else needs to rely on this marker for this use. If a consortium wants a shared enrichment tool, qualification is the route. If one sponsor wants one marker for one dose decision, its investigational new drug application (IND) and later application carry the argument, and the strategy's regulatory section should say why.

In the EU, the European Medicines Agency (EMA) offers qualification advice on plans, a formal qualification opinion from its human medicines committee on a specific context of use, and letters of support to encourage data sharing.16

Fit-for-purpose biomarker assay validation under FDA's April 2026 guidance

In shortFDA's current guidance is dated April 2026. It asks for a fit-for-purpose approach, full validation under ICH M10, the harmonized guideline on validating drug assays, when data support regulatory decisions such as approval or dosage instructions, and the sponsor's own judgment for internal decisions. It covers concentrations measured in sample materials, called matrices, such as blood or urine, so imaging reads and tissue scores need their own justification.

Older notes cite a January 2025 version, under three pages, which made no mention of context of use and drew complaints that it pointed to a standard that excludes biomarkers.17 The current text, final and dated April 2026, says "a fit-for-purpose approach should be used when determining the appropriate extent of method validation".12 One laboratory's commentary quotes the same words from the January version, so they may not be new.18 We found no published comparison of the two versions.

ICH M10, final in May 2022, remains the reference and "excludes biomarkers from consideration", yet FDA says its approach "should be the starting point", especially for chromatography and ligand-binding assays, two common laboratory methods.12,19 Where M10's criteria do not fit, FDA encourages sponsors to discuss their plans early and to justify the differences in their reports.12

From defining the purpose to routine use: how biomarker method validation runs in stages

FDA names the characteristics that define a biomarker method: "accuracy, precision, sensitivity, selectivity, parallelism, range, reproducibility, and stability".12 Parallelism asks whether a real sample, diluted, behaves like the calibrator, the reference material that sets the assay's scale. ICH M10 gives it special attention for substances the body also makes, and most biomarkers are of that kind.19 Cummings and colleagues judge a method's technical performance against its predefined purpose, since acceptance limits checked in isolation from the purpose do not amount to validation.20

FDA asks the same question from the method side: "How do sample collection, handling, and storage affect the reliability of the data?"12 So a strategy's assay section should hold the tube, the time to the centrifuge, the storage limit and how long samples will wait. A stability study that covers 9 months is little use to a study holding samples for 18.

CLIA assay validation: why a result that decides enrollment runs on a laboratory's calendar

In shortA laboratory that reports results about individual people falls under CLIA, and it must verify or establish its test's performance before reporting. A research assay validated elsewhere does not carry over by itself. That work has a start date, an end date and a queue, and those dates belong in the strategy.

CLIA's rules exempt "research laboratories that test human specimens but do not report patient specific results for the diagnosis, prevention or treatment of any disease or impairment of, or the assessment of the health of individual patients".22 A result that tells a site whether this person goes on to a scan is specific to that person, so sponsors commonly require a CLIA-certified laboratory for it. That is our reading and common practice. Ask your own counsel about a given design.

Rows one to three from 42 CFR 493.1253(b).23 Row four is our reading.

Companion diagnostic vs IVD: when a trial's selection test becomes a second regulated product

In shortAn in vitro diagnostic (IVD), a test on a sample taken from the body, becomes a companion diagnostic when a medicine cannot be used safely and effectively without it. FDA's guidance on this is final and dates from August 6, 2014. Its co-development guidance is still a July 2016 draft. Separately, any investigational test used in a trial needs a risk assessment under the device rules. Both questions belong on the strategy's first page.

Companion diagnostic definitions in the US and the EU

The consequence sits in one sentence. If FDA finds a test essential to a novel product, it "generally will not approve the therapeutic product or new therapeutic product indication if the IVD companion diagnostic device is not approved or cleared for that indication".25 FDA allows exceptions, for example where a serious condition lacks alternatives, and expects the test to follow.25 So a late companion test can hold the medicine.

Three device routes for a selection test used inside a trial

Before any marketing question, FDA expects the sponsor to assess the risk an investigational test presents to participants in that trial.29 The answer puts the test on one of three routes.

A blood pre-screen confirmed by PET invites the exemption argument, in our reading, though FDA's examples of exempt use in its draft are exploratory or retrospective analyses that do not direct treatment. A sponsor should document its position and can ask FDA, whose determination overrides its own.29 Exemption says nothing about whether the test will later need authorization with the medicine.29 FDA's general guidance on this is a December 2017 draft, and its streamlined route, final since October 2019, is for oncology trials.32,33

What FDA's companion diagnostic list shows in late 2026

FDA's list held 232 test-and-drug rows on the page dated September 28, 2026, plus nine group entries covering a class of oncology drugs, a route opened by guidance in April 2020.34,35 Rows count pairings, so one test can appear many times.

Clinical trial enrichment: the sample-size table, the screening funnel and what a blood pre-screen saves

In shortSelecting patients can shrink a trial many times over, on stated assumptions. FDA's table shows a 16-fold saving at one extreme and 2.6-fold at a milder one, and assumes a perfect test. Working the funnel through with real accuracy figures shows both the scans saved and the eligible people missed.

FDA's enrichment guidance, final since March 2019, sorts these strategies into three groups: decreasing variability, prognostic enrichment (people more likely to have the event) and predictive enrichment (people more likely to respond).42 Prognostic enrichment "would increase the absolute effect difference between groups but would not be expected to alter relative effect".42 Predictive enrichment claims more, since it says the drug works better in these people, and it depends on having a test that finds them.

A worked screening funnel: 1,000 candidates, a blood test, then PET

Take one published meta-analysis, a study that pools earlier studies, which reports 82% sensitivity and 86% specificity for plasma p-tau217 against amyloid on PET.43 Assume 40% of candidates are amyloid positive, a figure you should replace. Of 1,000 people, 400 are positive and 600 negative.

A published model points the same way. Schindler and colleagues compared four screening routes for an Alzheimer's trial. PET alone was the slowest and most expensive, adding a blood test at any point cut both time and cost, and more accurate blood tests added only small further savings.44 We read the abstract, which gives no figures, so take the model's numbers from the paper.

From a list with no decision to a test that arrives late: the eight common biomarker plan failures

In shortThe failures here are planning failures: a marker with no stated decision, a cut-off moved to another platform unchecked, a late laboratory, a device question nobody asked. Each of these can be found while drafting.

One marker, two cut-offs: what two published lung cancer trials show

Two trials of PD-1 antibodies as a first treatment for lung cancer were reported within a year of each other, and each used PD-L1, a protein read on tumor cells, to choose its primary group. KEYNOTE-024 enrolled patients with PD-L1 on at least 50% of tumor cells. CheckMate 026 read its primary result in patients at 5% or more.50,51 They used different drugs, and many things differ between them, so this is no controlled test of a cut-off.

Still, every biomarker team should know how far apart the published results fell. They are measured in progression-free survival, the time from randomization until the tumor progresses or the patient dies.52

A related lesson comes from the Blueprint project, which compared PD-L1 assays. Three of the four stained similar shares of tumor cells and one stained fewer, and agreement was much poorer on immune cells.47 Its authors treated interchange as a question to test, and patients could read positive on one assay and negative on another at the same cut-off.47

Enforcement: we found no FDA warning letter on an investigational selection test used without a risk assessment. FDA's guidance states the duty.

Patient selection biomarkers and approval odds: what the data show, and how much p-tau217 use has grown

In shortPrograms that used selection biomarkers reached approval about twice as often in the largest industry dataset, though that data cannot show the biomarker caused it. Registry records mentioning p-tau217 have jumped since 2024. We found no credible public figure for the size of this specific work.

Three things shape the trend. Blood tests reached the clinic: the Alzheimer's Association, in a statement dated August 24, 2026, called the Elecsys clearance the fourth FDA-cleared blood test for the disease, and the second in a week. FDA's record dates the decision August 19.56,46 FDA's table lists reduction in amyloid plaques as a surrogate endpoint, a stand-in measure, behind accelerated approval, which keeps amyloid measures central in this field.57 And FDA's April 2026 guidance now states the bar for an assay.12

FDA's guidance on developing drugs for early Alzheimer's disease remains a March 2024 draft.58 We found no credible public estimate of spending on biomarker strategy as a task. Market reports that size "biomarkers" count tests and instruments, and they tell you nothing about this planning work.

Clinical trial competitive intelligence for biomarkers: what registries must show, and how far to trust them

In shortUS law makes a rival's marker use largely public, often long before a paper. Registration is due within 21 days of the first participant for applicable trials. The narrow question to ask is which marker, at which cut-off, for which use, since when. Date every read, because registry records run late and sometimes disagree.

Under 42 CFR Part 11 the sponsor or other responsible party must register an applicable clinical trial within 21 calendar days after the first participant enrolls, and must update overall recruitment status within 30 calendar days of a change.59 Eligibility criteria and outcome measures sit in those records, and a marker used for entry often shows there. A cut-off may only surface in a poster or a paper. A record shows what a sponsor chose to register, so it cannot tell you a rival's strategy.

Trials registered late were a median of 111 days late.48 So a record tells you what was true when it was last updated, and a good strategy prints that date next to each row. Keep this reading to public sources: a rival's confidential data has no place in the plan.

How far AI can go in biomarker planning under FDA, EMA and EU rules, and which calls stay with scientists

In shortFDA's AI guidance is still a January 2025 draft and grades a model by its influence and the consequence of a wrong answer. Reading papers and drafting text may fall outside it, unless the output affects patient safety, quality or study reliability. Extraction helps when a person checks every value, and choosing a cut-off stays with scientists.

Three pieces of evidence set the limits. A 2025 preprint, not peer reviewed, tested a language model's data extraction with a person verifying each item across six systematic reviews, and reported 91.0% accuracy, 89.4% recall and 98.9% precision.68 That is useful with a checker, and it would be unsafe without one.

Discovery claims need more caution: a 2026 review of AI for transcriptomic biomarkers, read from gene activity, in COVID-19 warns about studies "that rely on internal validation", and most of them shared no code.69 And FDA has now qualified an AI tool, for reading liver biopsies in MASH trials, whose developer's announcement says it "is not for use in diagnostic procedures".6 A tool can be qualified, for a stated context of use, and that is all it is qualified for.

Four to eight weeks by hand: how a biomarker strategy moves from lead to scientist to board

In shortA translational medicine lead reads, pulls numbers, writes and chases, usually in a Word template. A biomarker scientist re-checks the values, section owners sign their parts, and a board reviews a static document. By one estimate, with no published figure behind it, most of the time goes on reading and drafting, and the assay-timing check often has no owner at all.

Who owns what on a biomarker strategy: lead, scientist, pharmacologist, diagnostics, operations, board

Who staffs biomarker planning: your own lead, a contract firm or software

From a late p-tau217 pre-screen to a costed option: the strategy map at work

In shortThis is the example from the top of the page, inside the product. Dr. Maya Chen, translational medicine lead, opens the program on its strategy map and sees one red card. The field lane shows two rivals already enrolling on that marker. She asks what would bring the test forward, gets two costed options, and proposes one. The diagnostics lead decides.

1

Nine biomarkers, four lanes, one red card

NWB-3140 is moving from Phase 1b to Phase 2. The map runs target engagement, pharmacodynamics, patient selection and safety as rows, and the phases as columns. Each card holds a marker, its assay, a state and the decision it supports. The header counts 9 biomarkers, 3 with an open gap, 1 not ready when needed, 3 rival trials compared.

In patient selection, under Phase 2a, plasma p-tau217 reads "Pre-screen not ready: CLIA transfer ends Jun 2027". Study 201's first participant is due in May 2027. The evidence is strong on this card, so the problem it names is the laboratory.

2

The field lane: two rivals already enroll on the same marker

Under the lanes, rival trials sit in the phase columns, read from registries, posters and papers, with the read date shown. HLV-210 pre-screens with plasma p-tau217 and confirms by PET, enrolling since March 2025. SLW-4 enriches with p-tau217 and Aβ42/40, since November 2024. BWP-77, a completed Phase 2, used PET for every candidate.

The call-out states what follows for this program: study 201 screens about 420 candidates by PET.

3

"See what would bring it forward"

Maya asks with one click, and each step runs in view. One agent, the trial design analyst, re-reads both rival registry records. Another, the assay and precedent reader, checks six CLIA laboratories for the same digital immunoassay platform, and finds Arden Clinical Laboratory with the platform in place and a validation slot from January 11.

4

Two options, with the estimates on screen

Option A: a second CLIA laboratory, ready March 2027, two months before first participant in, at $165,000. About 190 PET scans avoided, about $1.0 million saved and about 9 weeks of enrollment. Option B: keep the plan, with PET for every candidate and no transfer risk.

The estimates are printed: PET about $5,800 a scan, the pre-screen about $240 a test, 55% of candidates go on to PET. So 190 × $5,800 less 420 × $240 is about $1.0 million. The $165,000 transfer brings that down to about $836,000, the figure to take to a board.

5

She proposes; the diagnostics lead decides

Maya clicks "Propose option A". The card shows the pre-screen as proposed for study 201, the gap records the option, and the decision goes to Omar Haddad, the diagnostics lead, due November 14. Nothing is booked: an option stays a proposal until he decides.

Select any screen to open it larger. The screen gives a 14-week transfer, a January 11 slot and a March ready date. Fourteen weeks from January 11 runs to mid-April, so confirm a date like this with the laboratory. The guide shows them as the screens do.

The assay development plan, the gap list and a cited biomarker strategy ready for the board

In shortBehind the map sit the documents the board reads. The assay plan puts every activity against first participant in. The gap list gives five gaps an owner and a date. A scientist checks values beside their source passages. The strategy runs to eight sections with 112 cited claims in this sample, and the board chair decides with a reason.

1

Every assay against first participant in

The assay development plan lists each activity and its owner: the Kelbrook CLIA transfer, the proposed Arden transfer, two-platform bridging in 120 Phase 1b samples, re-validation of GFAP, a protein measured in blood, with stabilizer tubes, set-up of genotyping for the APOE gene, the central imaging charter, then the milestones for protocol final, the locked sample-collection plan and first participant in.

The critical path is stated at the foot: GFAP re-validation ends in mid-March, just before the sample-collection plan is locked.

2

Five gaps, each with an owner and a date

The gap and risk reviewer, an agent, reads the draft as a board member would. GFAP stability is High, owner Sam Patel, November 30. The p-tau217 pre-screen is High, Omar Haddad, November 14. Tau PET tracer at 9 of 40 sites is Medium, Kenji Mori, December 15. No two-platform bridging is Medium, Sam Patel, January 15. The cost of the intermediate zone is Low, Maya Chen, December 1.

The reviewer never closes a gap. A person assigns it or accepts the risk, and the device risk determination stays an open item inside gap two.

3

A number nobody can trace does not go in

Each extracted value sits beside the passage it came from. Plasma GFAP waits at 0.82 confidence: stability study SS-118 reports 84% recovery after 9 months at −80 °C, below its 85% criterion, and Phase 2 samples will wait up to 18 months. Values under the 0.90 threshold wait for a biomarker scientist, who confirms or corrects them.

4

One marker, all its evidence, with its context of use

The p-tau217 page shows its context of use, "Pre-screen to enrich enrollment for amyloid pathology before confirmatory PET", and readiness at the regulatory precedent rung. Sensitivity 91%, specificity 89% and an 18% intermediate zone come from a sample paper of 1,042 people. The paper and its journal are fictional. Real figures are in Part A.

5

Eight sections, every claim cited

Strategy v3 follows the house template: context of use, target engagement, pharmacodynamics, patient selection, safety, assay development plan, regulatory route, gap analysis. This sample holds 112 cited claims from 47 sources, with 0 uncited. Each section shows who checked or co-signed it, and each citation opens the source at its passage. It exports to Word, with the evidence table as a spreadsheet.

The safety section covers the trial plan only: imaging before dosing continues, and genotype before randomization.

6

The evidence table the scientist signs off

Nine biomarkers with role, readiness, key performance, validation, confidence and check. Two wait on Sam Patel. The rest show who checked them.

7

Co-signs in; the chair signs with a reason

Maya signed as author on October 6 at 4:42 pm. Sam Patel checked values on October 7 at 9:15 am. Lena Ortiz co-signed target engagement at 11:02 am and Omar Haddad patient selection at 1:30 pm. The pack closes October 14 for the October 21 board. Dana Okafor, the chair, approves, approves with conditions, or returns it for rework, and her electronic signature records the date, time and reason.

Ten programs at a glance, a companion test gap, rival designs compared and your evidence settings

In shortThe same discipline runs across a portfolio. A list shows each program's readiness and gaps. A second program shows how a companion test gap appears. One screen compares designs with rivals', and the dashboard tracks the work. The evidence rules are settings your team owns.

1

Every program, its next milestone and its board date

Ten sample programs in neuroscience, oncology, immunology and cardiometabolic disease, each with biomarkers by readiness, open gaps, the strategy's stage and its board. The sample's medicines appear here as a kind of medicine only.

2

A companion test gap in a second program

Veltrimab, a sample antibody–drug conjugate moving to Phase 3, selects patients with a tissue test. The trial assay is in use, and the Phase 3 card is red: "Companion diagnostic partner not contracted", Omar Haddad, November 30. The panel gives a lead time of about 24 months from contract to submission, an estimate in the sample, with Phase 3 starting in Q4 2027. Real target names, cut-offs and response figures are left off this screen.

3

Our design against the closest rival

How rivals use biomarkers across the portfolio, and a side-by-side table against HLV-210: pre-screen, intermediate zone, APOE ε4 homozygotes, GFAP and tau PET. The source shows the registry record's update date.

4

A dashboard of signed strategies, checked values and open gaps

Strategies signed, days from request to signed, sources read, values checked and open gaps, plus coverage by lane across programs and where evidence came from. These figures are sample data. The 6.5 days is a design target against an estimate of about 38 by hand, to be measured in your own use.

5

Your template, your evidence rules

The house template, citation style, independent sources per key claim, single-source flags, the scientist-check threshold, the recency window, preprint handling and the approvers for each section are settings your team owns. Preprints stay labeled as not peer reviewed.

Six agents draft the biomarker strategy; scientists, the diagnostics lead and the board decide

In shortFour agents read, one writes and one reviews. They can move a strategy no further than scientist review. A biomarker scientist checks values, section owners co-sign, the diagnostics lead decides routes and options, and the board chair signs. Controls support your compliance, and the obligations stay yours.

The six biomarker agents: what each one reads or writes, and the step it may not take

Inputs and outputs of a biomarker strategy: papers and registries in, a signed assay plan out

Records and systems a biomarker strategy rollout connects to

Maker-checker separation and the exportable trail behind each biomarker decision

Roles separate who extracts, who checks, who co-signs and who decides, so the maker of a value is never its only checker. Every agent step, value check, co-sign and decision goes on the activity trail, which can be exported for your quality team. Superseded strategies stay read-only. Botminds is SOC 2 Type II audited and manages information security to ISO 27001. GxP and 21 CFR Part 11 obligations remain yours, and these controls support your compliance with them.

The price of a late patient-selection assay, a wrong cut-off and a slow strategy

In shortStart from your own costs. Published figures frame the scale: what a day of a running trial costs, and how approval odds differ with selection biomarkers. No source publishes the hours in a biomarker strategy or how often a late assay delays a trial. You can measure both in a week from your own records.

Seven biomarker planning costs to find in a week, starting with your last strategy

Missed patients, tired scientists and quiet assumptions: what the totals leave out

One assay caught in time, and the lead's days back: the return on biomarker strategy software

In shortPut in your own numbers. The calculator values the lead's time released and, separately, what catching a late selection assay is worth. Defaults use published figures where they exist and the fictional example or illustrative values elsewhere. The hourly cost is an illustrative loaded cost, meaning pay plus overhead: replace it with your own.

The lead's time, scans, trial weeks and checked values: where biomarker planning automation saves

Strategies a year, the lead's days and one late assay on your own figures

Scoring a translational team's readiness: template, calendars, assays and sign-off

In shortPick the answer closest to your team today. Eight questions score 0 to 2, and the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

Twelve tests for biomarker strategy software vendors: page links, qualification claims, study calendars and sign-off

In shortAlmost every tool demonstrates well on a clean sample. These questions test what a demonstration hides: whether a value keeps its page, whether qualification claims have a record behind them, whether the tool knows your study calendar, and who may decide what. Ask us too.

Proving biomarker strategy software on your own work: 30 planted errors, one past program, one live strategy

In shortStart with a program you already know, because you know which assays ran late. Plant wrong values to measure the checks, then build one live strategy alongside your usual process. Decide on three numbers: the lead's days, planted errors caught, and real gaps found that your team had not listed.

The lead's days, planted errors caught and gaps your team had not listed

Biomarker strategy FAQ: BEST categories, context of use, qualification, CLIA and companion tests

In shortQuestions scientists and buyers search for, each answered on its own.