TECHNOLOGY EXPLAINER
Biomarkers and Clinical Endpoints
— The technique of standing in for "what we really want to know" with "what can be measured early"
Whether a new drug is truly useful is decided, in the end, by how patients feel, how they function and whether they survive. But finding that out takes a long time. So we substitute something that can be measured early, such as a blood value or the shrinkage of a tumor on an image. This article sorts out how far such a substitution is allowed, and where it breaks down.
- Biomarkers and clinical endpoints in brief
- The BEST definitions — types of biomarker
- "What we measure" versus "what we want to know" — three tiers of surrogate endpoint
- Endpoints in oncology — OS, PFS and response rate
- Our calculation: the smaller the effect, the more sharply the required number of events grows
- When a surrogate endpoint fails — the lesson of CAST
- A materials engineer's view (1): a surrogate endpoint is like an accelerated test
- Accelerated approval and surrogate endpoints
- Companion diagnostics — approving a drug and a test as a set
- A materials engineer's view (2): a measurement is a system of materials
- Open issues, and what has not been settled
- Glossary / References / Claim-to-source audit
Sourced = stated in published material (link given)
Our calculation = a figure this article derived from assumptions stated in the text
Not yet confirmed = a plan or target whose outcome has not been confirmed
Interpretations, such as how the material is organized or what it means for materials and measurement design, are marked separately as Commentary.
This article explains how clinical development is "measured" and how the regulatory system works. It does not evaluate the effect or safety of any individual drug or test and is not medical advice. It does not rank one drug or test against another. The trials and products that appear in the text are treated only as examples to explain how endpoints are thought about. Please consult a healthcare professional about any decision on treatment or testing.
1. Biomarkers and clinical endpoints in brief
- What a biomarker is: the glossary compiled by the joint FDA-NIH working group (BEST) defines it as "a defined characteristic that is measured as an indicator of normal biological processes, pathogenic processes, or responses to an exposure or intervention, including therapeutic interventions"Sourced. It includes molecular, histologic, radiographic and physiologic characteristics
- What it is not: the same definition says plainly that a biomarker is "not a measure of how an individual feels, functions, or survives"Sourced. That is measured by a clinical outcome
- Why it matters: clinical outcomes take a long time to confirm. So a biomarker expected to predict clinical benefit is sometimes used as the measure in a trial (a surrogate endpoint). This is where a trade between speed and certainty arises
A surrogate endpoint is not "the real answer"; it is "a measurement that predicts the real answer." BEST defines a surrogate endpoint as one that does not measure the clinical benefit of primary interest in and of itselfSourced. That is why the US system sorts surrogate endpoints into tiers according to how far they have been validated, and why, when a drug is approved on the strength of one that is not sufficiently validated, it asks for a post-approval trial to confirm the real benefitSourced.
2. The BEST definitions — types of biomarker
BEST (Biomarkers, EndpointS, and other Tools) is a glossary that the FDA-NIH working group first published on 28 January 2016 and last revised on 16 January 2025Sourced. It serves as a "common dictionary" so that the meaning of terms does not drift between researchers, companies and regulators (our commentary). The current glossary lists the following eight types of biomarkerSourced.
| Type | BEST definition (summary) |
|---|---|
| Susceptibility / risk | Indicates the potential for developing a disease in an individual who does not currently have clinically apparent disease |
| Diagnostic | Detects or confirms the presence of a disease, or identifies an individual with a subtype of the disease |
| Monitoring | Measured repeatedly to assess the status of a disease, or exposure to a medical product or environmental agent (or its effect) |
| Prognostic | Identifies the likelihood of a clinical event, recurrence or progression in patients who have the disease |
| Predictive | Identifies individuals who are more likely to experience a favorable (or unfavorable) effect from exposure to a medical product, compared with similar individuals without the biomarker |
| Response | Shows that a biological response has occurred in an individual exposed to a medical product. Includes pharmacodynamic biomarkers and biomarkers used as surrogate endpoints |
| Safety | Measured before or after exposure to indicate the likelihood, presence or extent of toxicity as an adverse effect |
| Multi-component | Two or more individual biomarkers combined in a defined way and interpreted together |
All Sourced (BEST glossary [Ref. 1]). The wording of the definitions is this article's summary.
For the pharmacodynamic biomarker, BEST says it indicates the biological activity of a medical product without necessarily informing about efficacy or disease outcome, and lists its uses as proof of concept, help in dose selection and measuring responseSourced. Not "evidence that the drug works," but "evidence that the drug has reached the body and is doing something": this distinction underlies what follows (our commentary).
3. "What we measure" versus "what we want to know" — three tiers of surrogate endpoint
BEST defines an endpoint as "a precisely defined variable intended to reflect an outcome of interest that is statistically analyzed to address a particular research question"Sourced. It then defines a surrogate endpoint as follows.
"An endpoint that is used in clinical trials as a substitute for a direct measure of how a patient feels, functions, or survives. A surrogate endpoint does not measure the clinical benefit of primary interest in and of itself, but rather is expected to predict that clinical benefit or harm based on epidemiologic, therapeutic, pathophysiologic, or other scientific evidence."Sourced
BEST goes on to say that, from the US regulatory standpoint, surrogate endpoints fall into three tiers according to the degree of clinical validationSourced.
| Tier | BEST definition (summary) | How it is used in approval |
|---|---|---|
| Validated | A clear mechanistic rationale and strong clinical data show that an effect on the measure predicts a specific clinical benefit | In a defined context, can support approval without additional trials directly showing clinical benefit |
| Reasonably likely | A strong mechanistic and/or epidemiologic rationale exists, but there are not enough clinical data to show that it is validated | Can be used for accelerated approval. A post-approval confirmatory trial has been required |
| Candidate | Its ability to predict clinical benefit is still being evaluated | — |
All Sourced (BEST glossary [Ref. 1]). BEST says that in the US regulatory context a validated surrogate endpoint in most cases refers to a biomarker.
US law (section 507(e)(9) of the FD&C Act) likewise defines a surrogate endpoint as a marker, such as a laboratory measurement, radiographic image, physical sign or other measure, that is not itself a direct measurement of clinical benefit, and distinguishes two uses: one known to predict clinical benefit can support traditional approval, and one reasonably likely to predict it can support accelerated approvalSourced. Under this provision the FDA publishes a table of surrogate endpoints that were the basis of drug approval or licensure and says it is updated every 6 monthsSourced. The table lists, for example, blood pressure in hypertension, serum HbA1c in type 2 diabetes and serum LDL cholesterol in hypercholesterolemia as bases for traditional approvalSourced.
4. Endpoints in oncology — OS, PFS and response rate
Surrogate endpoints are discussed most often in cancer. In its December 2018 guidance the FDA sorts out the endpoints used to approve cancer drugsSourced.
| Endpoint | Definition (as described in the guidance) | Advantages (guidance table) | Disadvantages (guidance table) |
|---|---|---|---|
| Overall survival (OS) | The time from randomization until death from any cause. Measured in the intention-to-treat (ITT) population | Easily and precisely measured. Based on objective, quantitative assessment | Can be affected by crossover from the control to the treatment group and by subsequent therapy. Needs long follow-up. Includes non-cancer deaths |
| Progression-free survival (PFS) | The time from randomization until objective tumor progression or death, whichever comes first | Can be assessed earlier and with fewer patients than a survival trial. Stable disease is included in the assessment | Assessment can be biased in open-label trials. Definitions differ between trials. Needs frequent imaging. May not correlate with survival |
| Objective response rate (ORR) | The proportion of patients whose tumor shrinkage of a pre-specified amount was sustained for a minimum period. The FDA generally defines it as the sum of partial and complete responses | Assessed early and with few patients. The effect on the tumor can be attributed to the drug (not to natural history) | Definitions differ between trials. Needs frequent imaging. May not correlate with survival |
All Sourced (FDA, "Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics," December 2018 [Ref. 2]). The wording is this article's summary.
The guidance describes OS as the most reliable cancer endpoint and, where a trial able to assess survival adequately can be run, usually the preferred endpointSourced. On the other hand, it lists long follow-up in large trials as a difficultySourced.
For PFS, it says that for a given number of patients, the size of the effect on PFS can be larger than the effect on overall survival, and at the same time that there are usually not enough data to robustly assess the correlation between effects on overall survival and on PFSSourced. ORR is described as the surrogate endpoint most commonly used to support accelerated approvalSourced.
The guidance says that ORR and PFS can be a surrogate endpoint supporting accelerated approval, a surrogate endpoint supporting traditional approval, and in some cases a direct measure of clinical benefit itself, depending on the disease, the context of use, the size of the effect, available therapy and the balance of risks and benefitsSourced. The name of an endpoint alone does not settle what it is evidence of (our commentary).
5. Our calculation: the smaller the effect, the more sharply the required number of events grows
Why are OS trials large and long? In comparing survival times, what is needed is not the number of patients as such but the number of observed "events" (deaths or progressions). Schoenfeld (1983) derived a formula for the number of observations needed to test whether two survival distributions are equal under the proportional-hazards modelSourced.
We estimate it with an approximation commonly used for a two-group comparison with 1:1 allocation, D = 4 × (zα/2 + zβ)² ÷ (ln HR)²Our calculation.
- Assumption: two-sided significance level of 5% (z = 1.960) and 80% power (z = 0.842). Then 4 × (1.960 + 0.842)² ≈ 31.4
- Assumption: the hazard ratio (HR) is constant (proportional hazards). If survival times follow an exponential distribution, the ratio of medians is 1 ÷ HR
Calculation
- HR 0.50 (equivalent to the median doubling): 31.4 ÷ (ln 0.50)² = 31.4 ÷ 0.480 ≈ 66 events
- HR 2/3 (≈ 0.67; equivalent to a median going from 6 to 9 months): 31.4 ÷ 0.164 ≈ 191 events
- HR 0.75: 31.4 ÷ 0.0828 ≈ 380 events
- HR 0.80: 31.4 ÷ 0.0498 ≈ 631 events
Assumptions and limits: this is a textbook approximation; actual trial designs take account of interim analyses, dropout, allocation ratios, non-proportional hazards and more. The HR values are hypothetical numbers chosen for this article and are not the results of any drug or trial. "Median 6 to 9 months" is the correspondence if an exponential distribution is assumed; actual survival curves need not behave that way.
What we can read from this is that to show a small difference in OS, one has to wait until many death events have occurred. And, as the guidance points out, OS is affected by subsequent therapySourced. Behind the demand for surrogate endpoints lie these two problems, the waiting time and the contamination (our commentary).
6. When a surrogate endpoint fails — the lesson of CAST
A classic example of how risky a surrogate endpoint can be is the Cardiac Arrhythmia Suppression Trial (CAST). According to the abstract of the 1991 New England Journal of Medicine paper, the trial was designed to test the hypothesis that suppressing ventricular ectopy after a myocardial infarction reduces the incidence of sudden deathSourced.
Patients in whom ventricular ectopy could be suppressed with a drug were randomly assigned to active drug or placebo. The use of two of the drugs (encainide and flecainide) was discontinued because of excess mortalitySourced.
Of 1,498 patients, 857 were assigned to encainide or its placebo (432 active drug, 425 placebo) and 641 to flecainide or its placebo (323 active drug, 318 placebo). After a mean follow-up of 10 months, 89 patients had died: deaths from arrhythmia numbered 43 on active drug and 16 on placebo, nonarrhythmic cardiac deaths 17 and 5, and noncardiac deaths 3 and 5Sourced.
- Active-drug groups: 432 + 323 = 755 patients. Deaths 43 + 17 + 3 = 63. 63 ÷ 755 ≈ 8.3%Our calculation
- Placebo groups: 425 + 318 = 743 patients. Deaths 16 + 5 + 5 = 26. 26 ÷ 743 ≈ 3.5%Our calculation
- Ratio: 8.3% ÷ 3.5% ≈ about 2.4-foldOur calculation
Assumptions and limits: these are crude proportions from simply adding up the abstract's numbers for the two drugs, not an analysis that accounts for differences in follow-up and so on. The result concerns the population of that trial (patients after a myocardial infarction with ventricular ectopy) and does not evaluate the current uses or safety of these drugs.
7. A materials engineer's view (1): a surrogate endpoint is like an accelerated test
In materials reliability evaluation, to see degradation that would take 10 years in real use, we use accelerated tests such as high temperature and high humidity. That a material that passes an accelerated test will also last in real use rests on the premise that the degradation mechanism is the same (our commentary).
A surrogate endpoint has much the same structure. BEST gives, as the grounds on which a surrogate endpoint is expected to predict clinical benefit, epidemiologic, therapeutic, pathophysiologic or other scientific evidenceSourced. That is, it presumes a causal pathway in which "if the measure moves, the patient's outcome moves too."
What happened in CAST can be read as this: the measured value (ectopic beats) moved as intended, but the drug affected the outcome by another route (our commentary). In accelerated-test terms, it corresponds to a situation in which a different degradation mode, one that does not appear under the accelerated conditions, dominated in real use.
- The correlation is guaranteed only "within that mechanism of action": even the same measure can lose its predictive power for a new drug that acts differently. That the FDA's oncology guidance judges the standing of ORR and PFS by "the disease, the context of use, the size of the effect" and the like corresponds to this pointSourced
- The FDA's table also has a "mechanism of action" column: the FDA's table of surrogate endpoints (adult, non-cancer) has a "Drug Mechanism of Action" column, in which blood pressure is shown as "Mechanism agnostic" (spanning many mechanisms of action), HbA1c as "Glucose-lowering" and LDL cholesterol as "Lipid-lowering"Sourced. The same page says that acceptability depends on the disease, the patient population, the mechanism of action of the therapy and the availability of existing treatments, and that a surrogate endpoint appropriate for one development program should not be assumed to be appropriate for a program in a different clinical settingSourced. It also says that a surrogate endpoint used for accelerated approval, for which the confirmatory trial then failed to show the expected benefit, is no longer accepted and is not included in the tableSourced
- Whether it is validated depends on "how much data there is": the three BEST tiers (candidate, reasonably likely, validated) correspond to the process by which an acceleration factor goes from "an estimate" to "a value confirmed by measurement" (our commentary)
In the world of materials, too, whether an existing accelerated test can be applied as it is to a new material system is always debated. "Can an existing surrogate endpoint be used as it is for a drug of a new modality?" is the same kind of question (our commentary).
8. Accelerated approval and surrogate endpoints
The US accelerated approval pathway uses a reasonably likely surrogate endpoint to approve a drug for a serious disease sooner. The FDA's December 2024 draft guidance sums up the history of the system as followsSourced.
| Year | Event (as described in the guidance) |
|---|---|
| 1992 | The FDA issued the accelerated approval regulations (21 CFR 314 subpart H and 601 subpart E) |
| 1997 | Congress put the system into law through the FDA Modernization Act (FDAMA), adding section 506 to the FD&C Act |
| 2012 | The FDA Safety and Innovation Act (FDASIA) amended it so that the seriousness, rarity and prevalence of the condition and the availability of alternative treatments are to be considered |
| 2023 | Amended by the Consolidated Appropriations Act, 2023 (section 3210). At approval, conditions for the confirmatory trial (milestones such as enrollment targets, the protocol and target completion dates) are set. The FDA may require that the confirmatory trial be "underway" before approval. Progress reports are submitted about every 180 days. A new expedited withdrawal procedure is created |
All Sourced (FDA draft guidance, "Expedited Program for Serious Conditions — Accelerated Approval of Drugs and Biologics," December 2024 [Ref. 4]). The 1992 regulations are also described in the oncology guidance [Ref. 2]. A draft guidance is not the final versionNot yet confirmed.
The guidance says that where a confirmatory trial fails to verify benefit, or is not conducted with due diligence, among other cases, the FDA may withdraw approval through an expedited procedureSourced. So how many have in fact been "confirmed," and how many "withdrawn"? The FDA publishes a list of accelerated approvals based on surrogate endpoints.
The list is titled "CDER Drug and Biologic Accelerated Approvals Based on a Surrogate Endpoint" and contains 350 entries in totalSourced. Counting the "Conversion-Withdrawal Status" of each row, we get the followingOur calculation.
- Converted (to traditional approval): 210 = 60.0%
- Not Yet Converted: 87 = 24.9%
- Withdrawn: 53 = 15.1%
- Withdrawn, of those settled (converted + withdrawn = 263): 53 ÷ 263 ≈ 20.2%
Assumptions and limits: the counts simply tally the per-indication rows, so several indications of the same drug are counted separately. The reasons for withdrawal (failure of the confirmatory trial, a company decision and so on) are not distinguished. What will happen to the 87 "not yet converted" is not settledNot yet confirmed.
9. Companion diagnostics — approving a drug and a test as a set
Among biomarkers, a predictive biomarker is used to choose "the people this drug is likely to work for." A test approved as a product for that purpose is a companion diagnostic.
FDA: a companion diagnostic is a medical device, often an in vitro diagnostic (IVD) device, that provides information that is essential for the safe and effective use of a corresponding drug or biological productSourced. Its uses include identifying patients who are most likely to benefit, identifying patients at increased risk of serious side effects and monitoring response to treatmentSourced. The FDA also writes that if a diagnostic test is inaccurate, treatment decisions based on the test may not be optimalSourced.
Japan: an MHLW notification of 1 July 2013 describes companion diagnostics and similar products as those used to improve the efficacy or safety of a specific drug, and as in vitro diagnostics or medical devices essential for the use of that drug (excluding those intended simply to diagnose disease)Sourced. It also says that, for a drug whose companion diagnostic has not been approved, in principle the application for the companion diagnostic should be filed at the same time as the drug's applicationSourced.
In the United States, the decision date for the premarket approval (PMA P980018) of HercepTest, an immunohistochemistry test for HER2 overexpression, is 25 September 1998, and the initial approval date of the corresponding anti-HER2 antibody, Herceptin (BLA 103792), is also 25 September 1998Sourced. The approval document says the test is used as an aid in assessing patients for whom Herceptin treatment is being consideredSourced. So the drug and the test were approved on the same day (this article's cross-check of two published dates). The FDA issued its final companion diagnostic guidance, "In Vitro Companion Diagnostic Devices," on 6 August 2014Sourced.
Table 1 of the FDA's "List of FDA-Authorized Companion Diagnostic Devices (In Vitro and Imaging Tools)" (for tests tied to a specific therapeutic product) had 226 rows when we checked (page updated 23 July 2026)Sourced. One row is a combination of test × corresponding drug × indication. Classifying the sample type from each row's "Indication - Sample Type" column gives the followingOur calculation.
- Tissue: 153 rows = 67.7%
- Plasma: 42 rows = 18.6%
- Whole blood, bone marrow, serum, saliva and others: 23 rows = 10.2%
- Tissue or plasma: 6 rows = 2.7% / no sample listed: 2 rows = 0.9%
A simple count of the biomarker column gives, in descending order, EGFR 30 rows, HER2 29, BRAF 19, KRAS 16 and PD-L1 15Our calculation.
Assumptions and limits: variations in notation in the list (spaces, upper and lower case) were merged by us when classifying. Because the same test appears in separate rows for different drugs and indications, this is not a count of test products. The list is updated from time to time.
10. A materials engineer's view (2): a measurement is a system of materials
BEST defines a test, tool or instrument as an assessment system made of three necessary components: (1) the materials for the measurement, (2) the assay (in the case of a biomarker) and (3) the method or criteria for interpreting the measurementSourced. Note that materials come first in the definition (our commentary).
Validation is also split in two. Analytical validation confirms performance such as sensitivity, specificity, accuracy and precision by a defined technical procedure that may include the procedures for sample collection, handling and storage, and BEST takes care to stress that it is not a validation of the usefulness of the itemSourced. Usefulness is the job of clinical validation: confirming that the test acceptably identifies, measures or predicts the concept of interestSourced.
The FDA's April 2026 guidance "Bioanalytical Method Validation for Biomarkers" lists, among the questions to be settled, besides the accuracy, precision and quantification range of the measurement, how sample collection, handling and storage affect the reliability of the data obtained from the method, and gives as concrete examples whether samples must be frozen in transit, the storage temperature and duration, and the effect of bench-top handlingSourced. For data used for decisions such as approval, it calls for full validation in line with ICH M10Sourced.
From here, the points of contact for materials and components come into view (our commentary).
- Materials that "keep" the sample: blood collection tubes, preservation solutions, freezing containers. As Fig. 6 shows, the main samples for companion diagnostics are tissue and plasma. In a test that measures the small amounts of nucleic acid in plasma, whether the sample stays unchanged between collection and analysis decides the result
- Materials that "read" the sample: antibodies used in immunostaining, the enzymes and reagents that amplify nucleic acids, and the reference materials that serve as the basis of measurement. The FDA list includes a rabbit monoclonal primary antibody used for HER2 staining as a companion diagnostic in itselfSourced
- Sources of variation: scatter in a measurement can bury the difference we want to see in a clinical trial (the HR of Chapter 5). Holding down lot-to-lot variation in materials also helps protect the power of a clinical trial
In materials terms, analytical validation is close to "instrument calibration and gauge repeatability and reproducibility (gauge R&R)," and clinical validation to "verifying that the measured property predicts product life." The former can be secured with materials and procedures, but the latter can be shown only with clinical data (our commentary).
11. Open issues, and what has not been settled
(1) The "validation" of a surrogate endpoint is re-asked whenever the type of drug changes
The FDA's oncology guidance says the handling of ORR and PFS is judged by the disease, the context and the size of the effectSourced, so there is no uniform rule. How far existing surrogate endpoints apply to drugs with a new mechanism of action is something for which we could not find, as of our research (September 2026), a primary source giving a general answerNot yet confirmed.
(2) The new accelerated approval rules are at the stage where their operation is still being settled
The FDA guidance following the 2023 legal amendment was still a draft as of December 2024Sourced. How the criteria for judging whether a confirmatory trial is "underway," and similar points, will finally be applied is Not yet confirmed.
(3) The relation to Japan's expedited and conditional approval
This article has focused on the US system. The comparison of expedited programs in Japan, the US and Europe is covered in our explainer on regulatory submissions and real-world data.
- A biomarker is an "indicator," not a measure of "how one feels, functions or survives" (the BEST definition)Sourced
- Surrogate endpoints divide by degree of validation into "candidate, reasonably likely and validated"Sourced
- OS is the most reliable but needs long follow-up. Going from an HR of 0.75 to 0.80 raises the required number of events about 1.7-foldOur calculation
- There is an example in which the surrogate improved but the outcome worsened (CAST)Sourced
- Of the FDA's 350 accelerated approvals, 60.0% converted to traditional approval and 15.1% were withdrawn (we tallied the list as of 30 June 2026)Our calculation
- For companion diagnostics, about two thirds of samples are tissue and about one fifth plasma (we tallied the FDA list)Our calculation
12. Glossary
- Biomarker
- A defined characteristic measured as an indicator of a biological process or a response to an intervention. Not a measure of how one feels, functions or survives.
- BEST
- The glossary of biomarker and endpoint terms compiled by the FDA-NIH working group.
- Predictive biomarker
- A biomarker that identifies people more likely to have a favorable (or unfavorable) effect from a particular drug.
- Prognostic biomarker
- A biomarker that indicates the likelihood of recurrence, progression and so on in people who have a disease.
- Pharmacodynamic biomarker
- A response biomarker that shows the biological activity of a drug. Not necessarily a conclusion on efficacy.
- Endpoint
- A precisely defined variable analyzed statistically in a clinical trial.
- Surrogate endpoint
- An endpoint used in place of directly measuring clinical benefit. Expected to predict that benefit.
- Clinical outcome
- An outcome expressing how a patient feels, functions and survives.
- Overall survival (OS)
- The time from randomization until death from any cause.
- Progression-free survival (PFS)
- The time from randomization until tumor progression or death, whichever comes first.
- Objective response rate (ORR)
- The proportion of patients whose tumor shrinkage of a set amount was sustained for a set period. The sum of partial and complete responses.
- Hazard ratio (HR)
- The ratio of the rates at which events occur in two groups. The smaller it is below 1, the fewer events in the treatment group.
- Accelerated approval
- A US pathway for early approval based on, for example, a reasonably likely surrogate endpoint. A confirmatory trial is a condition.
- Confirmatory trial
- A trial run after accelerated approval to verify actual clinical benefit.
- Companion diagnostic
- A test that provides information essential for the safe and effective use of a specific drug. Most are in vitro diagnostics.
- Analytical validation
- Confirmation of the technical performance of a test (sensitivity, specificity, accuracy, precision and so on). Not a validation of usefulness.
- Clinical validation
- Confirmation that a test acceptably identifies, measures or predicts the concept of interest.
- Fit-for-purpose
- The idea that the degree of validation is sufficient to support the proposed use.
13. References
- FDA-NIH Biomarker Working Group "BEST (Biomarkers, EndpointS, and other Tools) Resource — Glossary." Published 28 January 2016, revised 16 January 2025 (NCBI Bookshelf) — ncbi.nlm.nih.gov
- FDA "Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics — Guidance for Industry." December 2018 (PDF) — fda.gov
- FDA "Table of Surrogate Endpoints That Were the Basis of Drug Approval or Licensure" (page updated 29 April 2026) — fda.gov
- FDA "Expedited Program for Serious Conditions — Accelerated Approval of Drugs and Biologics — Guidance for Industry (Draft)." December 2024 (PDF) — fda.gov
- FDA (CDER) "NDA and BLA Accelerated Approvals — CDER Drug and Biologic Accelerated Approvals Based on a Surrogate Endpoint." As of June 30, 2026 (PDF) — fda.gov
- Echt, D.S. et al. (CAST) "Mortality and morbidity in patients receiving encainide, flecainide, or placebo. The Cardiac Arrhythmia Suppression Trial." New England Journal of Medicine 324(12), 781–788 (1991). doi:10.1056/NEJM199103213241201 — pubmed.ncbi.nlm.nih.gov
- Schoenfeld, D.A. "Sample-size formula for the proportional-hazards regression model." Biometrics 39(2), 499–503 (1983). doi:10.2307/2531021 — pubmed.ncbi.nlm.nih.gov
- FDA "Companion Diagnostics" (page updated 20 June 2023) — fda.gov
- FDA "List of FDA-Authorized Companion Diagnostic Devices (In Vitro and Imaging Tools)" (page updated 23 July 2026) — fda.gov
- FDA "Premarket Approval (PMA) — DAKO HERCEPTEST, P980018" — accessdata.fda.gov
- FDA "Drugs@FDA — BLA 103792 (HERCEPTIN)" — accessdata.fda.gov
- Ministry of Health, Labour and Welfare (MHLW), Director of the Evaluation and Licensing Division, Pharmaceutical and Food Safety Bureau "Points to consider in applications for approval of companion diagnostics and related drugs," Notification Yakushoku-shinsa-hatsu 0701 No. 10, 1 July 2013 (PDF, in Japanese) — pmda.go.jp
- FDA "Bioanalytical Method Validation for Biomarkers — Guidance for Industry." April 2026 (PDF) — fda.gov
14. Claim-to-source audit
| Claim in the text | Basis | Category |
|---|---|---|
| The definition of a biomarker (a defined characteristic measured as an indicator of normal biological processes, pathogenic processes or responses to an intervention; includes molecular, histologic, radiographic and physiologic characteristics; not a measure of how one feels, functions or survives). The eight types (susceptibility/risk, diagnostic, monitoring, prognostic, predictive, multi-component, response, safety) and their definitions. The definition and uses of the pharmacodynamic biomarker (proof of concept, dose selection, measuring response). The definitions of endpoint, clinical outcome and surrogate endpoint; that a surrogate endpoint does not itself measure clinical benefit and is expected to predict it on epidemiologic, therapeutic, pathophysiologic or other evidence. The three tiers (validated, reasonably likely, candidate), that a confirmatory trial has been required for accelerated approval, and that a validated surrogate endpoint in most cases refers to a biomarker. The definition of a companion diagnostic. That a test, tool or instrument consists of three components (materials for measurement, the assay, and the method or criteria for interpretation). The definitions of analytical validation (sensitivity, specificity, accuracy, precision; may include sample collection, handling and storage procedures; not a validation of usefulness) and clinical validation. The definition of fit-for-purpose. Published 28 January 2016, last revised 16 January 2025 | BEST glossary (NCBI Bookshelf). Reference 1 https://www.ncbi.nlm.nih.gov/books/NBK338448/ | Sourced |
| The definitions of OS, PFS and ORR, and the advantages and disadvantages given in Table 2. That OS is the most reliable cancer endpoint and usually preferred where survival can be assessed adequately, that long follow-up in large trials is a difficulty, and the effect of subsequent therapy. That the effect on PFS can be larger than on OS, and that data are usually insufficient to robustly assess the correlation. That ORR is the most commonly used surrogate endpoint for accelerated approval. That ORR and PFS can be surrogate endpoints for accelerated or traditional approval, or direct measures of clinical benefit, depending on context and other factors. That the accelerated approval regulations were issued in 1992. Issued December 2018 | FDA guidance on cancer endpoints. Reference 2 https://www.fda.gov/media/71195/download | Sourced |
| That section 3011 of the 21st Century Cures Act created section 507 of the FD&C Act, and the definition of a surrogate endpoint in section 507(e)(9) (two categories: traditional and accelerated approval). That the table is updated every 6 months. That blood pressure in hypertension, serum HbA1c in type 2 diabetes and serum LDL cholesterol in hypercholesterolemia are listed as Traditional in the "Type of Approval Appropriate for" column, with "Drug Mechanism of Action" entries of Mechanism agnostic, Glucose-lowering and Lipid-lowering respectively. That acceptability depends on disease, patient population, mechanism of action and existing treatments, and that a surrogate appropriate in one clinical setting should not be assumed appropriate in another. That surrogate endpoints for which a confirmatory trial did not show benefit are not included in the table | FDA page on the table of surrogate endpoints. Reference 3 https://www.fda.gov/drugs/development-resources/table-surrogate-endpoints-were-basis-drug-approval-or-licensure | Sourced |
| The history of accelerated approval (the 1992 regulations, section 506 added by the 1997 FDAMA, the amendment by the 2012 FDASIA, the amendment by section 3210 of the Consolidated Appropriations Act, 2023). Setting conditions for the confirmatory trial (milestones including enrollment targets, protocol and target completion dates), the authority to require that the confirmatory trial be underway before approval, progress reports about every 180 days, and the new expedited withdrawal procedure and its requirements. That the guidance is a draft of December 2024 | FDA draft guidance on accelerated approval. Reference 4 https://www.fda.gov/media/184120/download | Sourced |
| That CDER's list of accelerated approvals based on a surrogate endpoint has 350 entries in total as of 30 June 2026, and that each row gives a "Conversion-Withdrawal Status" | FDA (CDER) list of accelerated approvals. Reference 5 https://www.fda.gov/media/151146/download | Sourced |
| That CAST was designed to test the hypothesis that suppressing ventricular ectopy after myocardial infarction reduces sudden death; that patients in whom ectopy could be suppressed were randomized; that encainide and flecainide were discontinued because of excess mortality; the group sizes (857 = 432 + 425, 641 = 323 + 318, 1,498 in total), the mean follow-up of 10 months and the deaths by cause (arrhythmia 43 vs 16, nonarrhythmic cardiac 17 vs 5, noncardiac 3 vs 5) | Echt et al. (1991), abstract. Reference 6 https://pubmed.ncbi.nlm.nih.gov/1900101/ | Sourced |
| That Schoenfeld (1983) derived a formula for the number of observations needed to test the equality of two survival distributions under the proportional-hazards model | Schoenfeld (1983), abstract. Reference 7 https://pubmed.ncbi.nlm.nih.gov/6354290/ | Sourced |
| The FDA definition of a companion diagnostic and its three uses; that if a diagnostic test is inaccurate, treatment decisions based on it may not be optimal; that the final guidance "In Vitro Companion Diagnostic Devices" was issued on 6 August 2014 | FDA page on companion diagnostics. Reference 8 https://www.fda.gov/medical-devices/in-vitro-diagnostics/companion-diagnostics | Sourced |
| That Table 1 has 226 rows, the "Indication - Sample Type" and biomarker entries of each row, that the PATHWAY anti-Her2/neu (4B5) Rabbit Monoclonal Primary Antibody is listed as a companion diagnostic, and that the page was updated on 23 July 2026 | FDA list of companion diagnostics. Reference 9 https://www.fda.gov/medical-devices/in-vitro-diagnostics/list-fda-authorized-companion-diagnostic-devices-in-vitro-and-imaging-tools | Sourced |
| That the decision date of the HercepTest (PMA P980018) is 25 September 1998, and that it is a semi-quantitative immunohistochemical assay that determines HER2 overexpression, used as an aid in assessing patients for whom Herceptin treatment is being considered | FDA PMA database. Reference 10 https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P980018 | Sourced |
| That the initial approval (ORIG-1) of Herceptin (BLA 103792) was on 25 September 1998 | Drugs@FDA. Reference 11 https://www.accessdata.fda.gov/scripts/cder/daf/index.cfm?event=overview.process&ApplNo=103792 | Sourced |
| The scope of companion diagnostics in Japan (used to improve the efficacy or safety of a specific drug, and in vitro diagnostics or medical devices essential for the use of that drug), and that the application for the companion diagnostic should in principle be filed at the same time as the drug's application. The notification number and date | MHLW notification (Yakushoku-shinsa-hatsu 0701 No. 10). Reference 12 https://www.pmda.go.jp/files/000213148.pdf | Sourced |
| Taking a fit-for-purpose approach to the validation of biomarker assays, full validation in line with ICH M10 when used for regulatory decisions such as approval, and the questions to be settled (accuracy and precision, quantification range, effects of sample collection, handling and storage, whether freezing in transit is needed, storage temperature and duration, and the effect of bench-top handling). Issued April 2026 | FDA guidance on bioanalytical method validation for biomarkers. Reference 13 https://www.fda.gov/media/185276/download | Sourced |
| The numbers of events required (66 at HR 0.50, 191 at 2/3 (≈ 0.67), 380 at 0.75 and 631 at 0.80) and the about 1.7-fold rise. The CAST death proportions (8.3% on active drug, 3.5% on placebo, about 2.4-fold). The breakdown of the 350 accelerated approvals (210 converted, 60.0%; 87 not yet converted, 24.9%; 53 withdrawn, 15.1%; 20.2% withdrawn of those settled). The breakdown of the 226 companion-diagnostic rows by sample (tissue 153 rows, 67.7%; plasma 42 rows, 18.6%; blood and others 23 rows, 10.2%; tissue or plasma 6 rows, 2.7%; none listed 2 rows, 0.9%) and the tally of the biomarker column (EGFR 30, HER2 29, BRAF 19, KRAS 16, PD-L1 15). The cross-check that HercepTest and Herceptin were approved on the same day | Our calculation. Two-sided 5%, 80% power, 1:1 allocation, proportional hazards and exponential distribution are assumptions set in this article, and the HRs are hypothetical. The CAST figures are crude proportions from simply pooling the numbers in the abstract. The list tallies are simple row counts, with variations in notation merged by us. None evaluates the efficacy or safety of any individual drug or test | Our calculation |
| What will happen to the 87 "not yet converted" accelerated approvals. The finalization and operation of the accelerated approval draft guidance | As of the research for this article (September 2026) the outcome is not settled. The draft guidance is not the final version. Reference 4 https://www.fda.gov/media/184120/download | Not yet confirmed |
| A general answer on how far existing surrogate endpoints apply to drugs with a new mechanism of action | As of the research for this article (September 2026), no primary source giving a general answer could be found (commentary) | Not yet confirmed |
| The effect, clinical results and relative merits of individual drugs and tests | Not covered: this article explains endpoints and the regulatory system, and makes no assessment of treatment effect or safety. CAST is quoted, from the paper's abstract only, as a historical example to explain the idea of a surrogate endpoint (commentary) | Commentary |
| The arrangement of biomarkers by setting. Reading a surrogate endpoint as analogous to an accelerated test of materials (that the correlation is guaranteed only within the mechanism, and the three tiers as a move from an estimated to a measured acceleration factor). Reading CAST as a situation in which a different degradation mode dominated in real use. Seeing measurement as a system of materials, and sorting it into materials that keep the sample, materials that read it and sources of variation. Likening analytical and clinical validation to gauge R&R and to verification of life prediction. The framing of OS as "waiting time and contamination" | Organization and commentary by this article based on published content; not views expressed by the FDA, NIH, MHLW or the authors of the papers | Commentary |
| Figures combine ImageGen-generated explanatory images with vector drawings, distinguished from real objects and measured images | Note by this article | Commentary |
Last updated: 26 September 2026 / Sources are limited to primary sources (the FDA-NIH working group glossary, FDA guidance, published lists and approval databases, an MHLW notification, and peer-reviewed papers). Because the article includes readings of endpoint design and of measurement, these are separated as "Commentary" from sourced facts. How far existing surrogate endpoints apply to drugs with a new mechanism of action, the final operation of the accelerated approval draft guidance, and the fate of accelerated approvals not yet converted are marked "Not yet confirmed" because there is no confirmed primary source. This article explains endpoints and the regulatory system; it does not evaluate the effect or safety of any individual drug or test, and it is not medical advice. All figures are explanatory concept diagrams. Figures combine ImageGen-generated explanatory images with vector drawings. Generated images are not photographs, microscope images or measured data of real objects. The original diagrams with numbers and sources are either shown as they are or kept in a form that expands below the new image.