Biomarker Endpoint

Biomarker Endpoint from Laboratory Measure to Trial Outcome and Economic Model

A biomarker endpoint turns a measured biological quantity, such as LDL cholesterol or blood pressure, into the outcome a trial is designed and analysed on. For health economists the question it raises is whether a treatment effect on that measure can be converted into survival, quality-adjusted life years and costs, because a cost-utility model needs final outcomes and the trial often reports only the marker. The answer depends on the regulatory standing the biomarker holds as a surrogate endpoint and on the validation evidence behind it.

What turns a biomarker into an endpoint

The FDA describes a biomarker as a defined characteristic objectively measured as an indicator of normal biological processes, pathological processes, or responses to an exposure or intervention. An endpoint adds the analysis, meaning the assay or image, the time point, the summary (change from baseline at week 12, proportion below a threshold) and the comparison between arms. Fixing these choices in the protocol stops a favourable time point or threshold being picked after the results are seen. Two trials reporting the same marker under different definitions are not estimating the same quantity, which matters when their results are pooled for a model.

Many biomarker endpoints are never meant to replace a patient outcome; early-phase trials use them to show that a drug reaches its target or to choose a dose. The economic stakes rise when the marker becomes the primary efficacy endpoint standing in for clinical benefit. The FDA defines a surrogate endpoint as a clinical trial endpoint used as a substitute for a direct measure of how a patient feels, functions, or survives, which does not measure that benefit itself but is expected to predict it. Used this way, the marker becomes the main source of the modelled treatment effect, so any error in the surrogate relationship flows straight into the incremental cost-effectiveness ratio.

Regulatory standing as candidate, reasonably likely or validated

The FDA grades a biomarker used as a surrogate by the strength of its clinical validation, and the level decides which approval route the endpoint can support. For an economic evaluator the level signals how much of the clinical benefit is observed and how much is predicted.

US law allows accelerated approval for a serious or life-threatening condition on an effect on a surrogate endpoint that is reasonably likely to predict clinical benefit (21 U.S.C. 356(c)). The approval may require postapproval studies to verify the predicted effect, and the FDA may withdraw approval through expedited procedures if such a study fails to verify the benefit. Section 507 of the Federal Food, Drug, and Cosmetic Act requires the FDA to publish a list of surrogate endpoints that were the basis of approval. The resulting Surrogate Endpoint Table lists, for example, serum LDL cholesterol for hypercholesterolaemia and blood pressure for hypertension as surrogates for traditional approval. The FDA decides acceptability case by case and leaves out of the table surrogates whose confirmatory trials after accelerated approval failed to show the expected benefit.

Validation evidence from a single trial to meta-analysis

The first formal criterion came from Prentice, who in 1989 proposed that a surrogate should yield a valid test of the null hypothesis of no treatment effect on the true endpoint. Operationally, the surrogate must capture the whole relationship between treatment and the true endpoint, so that the true endpoint rate is independent of treatment once the surrogate history is known. NICE DSU Technical Support Document 20 notes that meeting these criteria in a single trial does not guarantee a causal link between the treatment effects on the two outcomes, and quotes Fleming and DeMets, "A correlate does not a surrogate make".

Buyse and colleagues moved validation into a meta-analytic framework in 2000, separating an individual-level measure of agreement between the two endpoints from a trial-level measure of agreement between the treatment effects on them. The trial-level relationship is the one an economic model needs, because it supports predicting the effect on the true endpoint from an observed effect on the marker. TSD 20 recommends bivariate meta-analysis of treatment effects on the surrogate and final outcome, which allows for uncertainty in both effects, and observes that a weak surrogate relationship shows up as a wide prediction interval for the final-outcome effect. The equations and the trial-level R-squared are set out on the surrogate endpoint page and in multivariate meta-analysis.

What NICE expects when a model relies on a biomarker

The NICE manual (PMG36, section 4.6) states that for cost-utility analyses clinical end points reflecting how a patient feels, functions, or how long a patient lives are more informative than surrogate outcomes. When final end points cannot be used, the company must provide evidence supporting the outcome relationship and explain how the relationship is quantified for modelling. PMG36 adopts three levels of evidence from Ciani and colleagues (2017), listed below.

• Level 3. Biological plausibility of the relation between the surrogate end point and final outcomes.
• Level 2. A consistent association between surrogate and final outcomes, usually from epidemiological or observational studies.
• Level 1. The technology's effect on the surrogate corresponds to a commensurate effect on the final outcome in randomised controlled trials.

A surrogate counts as validated only when the relative effect on the surrogate is shown to predict the relative effect on the final outcome, preferably by a meta-analysis of RCTs reporting both outcomes using the TSD 20 methods. Biological plausibility should be shown for every surrogate, and committees may accept lower levels for some technologies for rare conditions and some diagnostics and devices. Validation is specific to the population and technology type. Extrapolating the relationship to another population or a different mechanism of action needs thorough justification, and the uncertainty in the surrogate relationship should be quantified and included in probabilistic sensitivity analysis, with further exploration in scenario analysis.

Carrying a biomarker effect into an economic model

A model that starts from a biomarker endpoint links three quantities, namely the treatment effect on the marker from the trial, a surrogate relationship that converts it into an effect on a final outcome, and the model structure (often a Markov model) that turns the final-outcome effect into life years, QALYs and costs. Each link carries its own uncertainty, and the second link is usually the weakest.

The choices that change the result are mostly about the second link. A relationship estimated from treatment effects across RCTs (level 1) predicts what happens when a treatment moves the marker. A relationship estimated from patient-level association in cohort data (level 2) predicts how people with different marker values fare, and using it assumes that a treatment-induced change carries the same prognosis as a natural difference between people. The modeller also decides how long the marker effect persists, whether the relationship holds for the new drug's mechanism and class, and whether to use the mean prediction or the full predictive distribution. These are sources of structural uncertainty, and scenarios that weaken the relationship, shorten persistence or use direct outcome data once mature show how much the decision depends on them.

Worked example of an illustrative lipid-lowering drug

All numbers in this example are illustrative and do not describe any real technology. A new drug lowers a cardiovascular marker by 1.2 units more than its comparator. A hypothetical trial-level meta-analysis gives a slope of $-0.20$ on the log relative risk of a major cardiovascular event per unit of marker reduction, with no intercept.

$$\ln(RR) = \beta \times \Delta S = -0.20 \times 1.2 = -0.24$$

where $RR$ is the predicted relative risk of the event over five years, $\beta$ is the slope of the surrogate relationship and $\Delta S$ is the treatment effect on the marker in units. The predicted relative risk is exp(-0.24) = 0.787.

With a five-year event risk of 10% on the comparator, the predicted risk on the new drug is 0.10 × 0.787 = 0.0787. Per 1,000 patients, 1,000 × (0.100 − 0.0787) = 21.3 events are avoided. Each event is assumed to cost 1.5 discounted QALYs and GBP 8,000, and the drug costs GBP 1,500 per patient over five years. Avoided events give 21.3 × 1.5 = 31.95 QALYs and save 21.3 × 8,000 = GBP 170,400, so the net cost is GBP 1,500,000 − 170,400 = GBP 1,329,600.

$$ICER = \frac{\Delta C}{\Delta E} = \frac{1{,}329{,}600}{31.95} = 41{,}615$$

where $ICER$ is the incremental cost-effectiveness ratio in GBP per QALY gained, $\Delta C$ is the incremental cost per 1,000 patients in GBP and $\Delta E$ is the incremental QALYs per 1,000 patients.

Suppose instead that only half of the marker effect passes through to events, as might happen if the drug acts partly off the pathway captured in the meta-analysis. The slope becomes -0.10, so ln(RR) = -0.12 and RR = exp(-0.12) = 0.887. Events avoided fall to 1,000 × (0.100 − 0.0887) = 11.3, QALYs to 11.3 × 1.5 = 16.95 and savings to 11.3 × 8,000 = GBP 90,400, giving a net cost of GBP 1,409,600 and an ICER of 1,409,600 / 16.95 = GBP 83,162 per QALY.

Halving the strength of the surrogate relationship roughly doubles the ICER, although the trial result on the marker is identical in both cases. A probabilistic analysis would draw the slope from its predictive distribution, including between-trial heterogeneity, so that the spread of ICERs reflects how well the marker is known to predict events.

Where biomarker endpoints have misled

The Cardiac Arrhythmia Suppression Trial is the standard warning. Ventricular premature beats after myocardial infarction were known to predict sudden death, and patients whose arrhythmia was suppressed on Holter recording were randomised to active drug or placebo. The preliminary report found higher total mortality with encainide or flecainide, 56 of 730 patients (7.7%) against 22 of 725 (3.0%) on placebo, a relative risk of 2.5. The marker moved in the intended direction while the final outcome moved the other way.

Common errors in economic work follow the same pattern. A strong prognostic association is read as evidence that changing the marker changes outcomes. A relationship validated for one drug class is applied to a first-in-class mechanism. Short trials miss late harms or loss of effect. The mean predicted effect is used without its prediction error, so probabilistic analysis understates decision uncertainty. When confirmatory trials or mature data arrive, models built on the marker should be re-run on the observed outcomes.

How biomarker endpoints relate to nearby concepts

The terms overlap and are often used loosely. The distinctions below follow the FDA and NICE usage above.

• A biomarker is the measured characteristic itself; the biomarker endpoint is its defined analysis in a trial.
• A surrogate endpoint substitutes for a patient outcome and need not be a biomarker; a biomarker endpoint is a surrogate only when used in that role.
• A final endpoint, such as overall survival, measures directly how a patient feels, functions or survives.
• Progression-free survival is often assessed as a surrogate for overall survival in oncology, but it is defined by progression events and death, not a single laboratory value.

Sources

• Buyse M, Molenberghs G, Burzykowski T, Renard D, Geys H. The validation of surrogate endpoints in meta-analyses of randomized experiments. Biostatistics. 2000;1(1):49-67. https://doi.org/10.1093/biostatistics/1.1.49
• Bujkiewicz S, Achana F, Papanikos T, Riley RD, Abrams KR. NICE DSU Technical Support Document 20: Multivariate meta-analysis of summary data for combining treatment effects on correlated outcomes and evaluating surrogate endpoints. Decision Support Unit, ScHARR, University of Sheffield. 22 October 2019. https://doi.org/10.15131/shef.data.33848854
• Cardiac Arrhythmia Suppression Trial (CAST) Investigators. Preliminary report: effect of encainide and flecainide on mortality in a randomized trial of arrhythmia suppression after myocardial infarction. New England Journal of Medicine. 1989;321(6):406-412. https://doi.org/10.1056/NEJM198908103210629
• Ciani O, Buyse M, Drummond M, Rasi G, Saad ED, Taylor RS. Time to review the role of surrogate end points in health policy: state of the art and the way forward. Value in Health. 2017;20(3):487-495. https://doi.org/10.1016/j.jval.2016.10.011
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 31 January 2022, last updated 13 October 2026. Section 4.6, accessed 5 October 2026. https://www.nice.org.uk/process/pmg36/chapter/economic-evaluation-2
• Prentice RL. Surrogate endpoints in clinical trials: definition and operational criteria. Statistics in Medicine. 1989;8(4):431-440. https://doi.org/10.1002/sim.4780080407
• United States Code. 21 U.S.C. 356(c), Accelerated approval of a drug for a serious or life-threatening disease or condition. Accessed 5 October 2026. https://www.law.cornell.edu/uscode/text/21/356
• US Food and Drug Administration. Surrogate endpoint resources for drug and biologic development. Content current as of 24 July 2018, accessed 5 October 2026. https://www.fda.gov/drugs/development-resources/surrogate-endpoint-resources-drug-and-biologic-development
• US Food and Drug Administration. Table of surrogate endpoints that were the basis of drug approval or licensure. Content current as of 29 April 2026, accessed 5 October 2026. https://www.fda.gov/drugs/development-resources/table-surrogate-endpoints-were-basis-drug-approval-or-licensure