Biomarker Endpoint

How a biomarker becomes an endpoint

A biomarker endpoint uses a measured biological characteristic as an outcome for evaluating disease, treatment response or safety. This page explains how a biomarker is defined and measured, when it can support conclusions about treatment effects, and why a change in a biological measure does not automatically establish a benefit that patients can feel or experience.

The biomarker is the biological characteristic, while the endpoint is the precisely defined variable analysed in a study. For example, blood pressure is a biomarker, but change from baseline in systolic blood pressure at 12 weeks is a biomarker endpoint.

Specifying exactly what the endpoint measures

A biomarker endpoint should identify the biological characteristic, specimen or imaging source, measurement method, time point and analytical rule. Without this precision, studies using the same biomarker name may not be estimating the same outcome.

A complete specification commonly includes:

• The biological characteristic being measured.
• The specimen, tissue, image or physiological source.
• The assay, device or assessment method.
• The unit and numerical scale.
• The baseline definition.
• The follow-up time or assessment schedule.
• Whether the endpoint is a measured value, change, percentage change, threshold or time-to-event result.
• The rules for repeated, missing or below-limit measurements.
• The analysis population and statistical contrast.

The endpoint should be prespecified before the results are known. Selecting a favourable time point, threshold or transformation after examining the data can produce a misleading treatment effect.

The roles biomarkers can play

Biomarkers can support different clinical and research purposes. A biomarker's classification depends on its intended use, so the same biological measure may serve more than one role in different contexts.

Common roles include:

• A diagnostic biomarker detects or confirms a disease or condition.
• A monitoring biomarker is measured repeatedly to assess disease status or response.
• A prognostic biomarker identifies the likelihood of a future clinical event regardless of treatment.
• A predictive biomarker identifies people more likely to experience a particular treatment effect.
• A pharmacodynamic or response biomarker shows a biological response after exposure to an intervention.
• A safety biomarker indicates the likelihood, presence or extent of toxicity.
• A susceptibility or risk biomarker identifies the likelihood of developing a disease or condition.

A prognostic association does not establish that changing the biomarker will improve outcomes. A predictive biomarker also requires evidence about variation in treatment effect rather than evidence that one group simply has a different prognosis.

Biomarker endpoints and patient-relevant outcomes

Patient-relevant outcomes describe how a person feels, functions or survives. Biomarker endpoints describe biological processes that may occur earlier or be easier to measure.

Biomarker endpoints can be valuable when they:

• Provide early evidence that an intervention reaches its biological target.
• Support dose selection.
• Detect toxicity before clinical harm becomes severe.
• Reduce the time or sample size required for a study.
• Help explain how an intervention produces its effects.
• Identify patients for targeted treatment.

These advantages do not make a biomarker equivalent to a patient-relevant outcome. The biological change may be too small, may not persist, may not lie on the causal pathway to benefit or may be offset by harmful effects elsewhere.

Biomarker endpoints and surrogate endpoints

A biomarker endpoint becomes a surrogate endpoint only when it is used as a substitute for a direct measure of how a patient feels, functions or survives. Not every biomarker endpoint is intended or validated to serve that role.

Three statements should be kept separate:

• The biomarker is associated with a clinical outcome.
• Treatment changes the biomarker.
• The treatment effect on the biomarker reliably predicts the treatment effect on the clinical outcome.

The first two statements do not prove the third. A biomarker may predict prognosis without capturing the pathway through which treatment affects the patient-relevant outcome.

Analytical validation

Analytical validation establishes whether the measurement method detects or quantifies the biomarker accurately and reliably. A biologically important marker cannot support a credible endpoint when its assay or imaging method performs poorly.

Analytical performance may include:

• Accuracy relative to an accepted reference method.
• Precision within and between runs.
• Reproducibility across laboratories, devices or readers.
• Analytical sensitivity and specificity.
• Limit of detection and limit of quantification.
• Linearity across the measurement range.
• Stability during collection, processing, storage and transport.
• Resistance to interference from other substances or conditions.
• Calibration and quality-control procedures.

Validation should match the intended specimen, population and setting. Performance established in one laboratory or platform should not be assumed for another method without supporting evidence.

Clinical validation

Clinical validation examines whether the biomarker is associated with the biological state or clinical outcome relevant to its intended use. The required evidence depends on whether the biomarker is diagnostic, prognostic, predictive, pharmacodynamic or related to safety.

Clinical validation should consider:

• The strength and consistency of the association.
• The population and clinical context.
• The comparator or reference standard.
• The timing of biomarker and outcome measurements.
• Confounding and selection bias.
• Whether thresholds were prespecified or data driven.
• Replication in independent data.
• Performance across relevant population groups.

A statistically strong association can still have limited clinical usefulness when values overlap substantially between people with different outcomes or when the association does not change decision making.

Establishing clinical utility

Clinical utility concerns whether using the biomarker to guide care improves outcomes or decisions. It goes beyond measuring the biomarker accurately or showing that it is associated with disease.

Utility may depend on whether:

• The result changes treatment or monitoring.
• The alternative action is effective.
• The benefit of changed care exceeds the harms of testing and treatment.
• The result is available at the time the decision must be made.
• The testing pathway is feasible and affordable.
• Patients and clinicians can interpret the result.

A biomarker can be analytically and clinically valid without improving care if no effective action follows from the result.

Measuring change from baseline

Many biomarker endpoints compare follow-up with a baseline measurement. The analytical method should account for baseline values and specify whether the endpoint is an absolute change, percentage change or another transformation.

Absolute change for participant (i) is:

$$ \Delta B_i = B_{i,t} - B_{i,0} $$

Percentage change is:

$$ %\Delta B_i = \frac{B_{i,t} - B_{i,0}}{B_{i,0}} \times 100% $$

where (B_{i,0}) is the baseline value and (B_{i,t}) is the value at follow-up time (t).

Percentage change can behave poorly when baseline values are close to zero or measured with substantial error. The choice of scale should reflect the biological meaning, statistical distribution and intended interpretation.

Threshold-based endpoints

A continuous biomarker may be converted into a responder endpoint by defining a threshold. Thresholds can aid clinical interpretation but discard information and can classify similar measurements differently when they fall on opposite sides of a cut-off.

A threshold should be supported by:

• Biological or clinical rationale.
• Evidence connecting the cut-off with risk or benefit.
• Analytical precision near the cut-off.
• Prespecification before outcome analysis.
• Validation in the intended population.

Data-driven selection of the threshold can exaggerate performance. Sensitivity analyses should examine how alternative plausible thresholds affect the conclusion.

Timing and repeated measurements

Biomarkers can change rapidly and may show transient, delayed or cyclical responses. The selected endpoint time should match the intervention mechanism and the clinical question.

Repeated measurements can describe response trajectories, peak change, duration or variability. The analysis should specify whether it uses a single time point, area under the curve, time-weighted average, slope or another summary.

Frequent measurement can improve understanding but may also introduce multiplicity and participant burden. Irregular timing can bias comparisons when measurement schedules differ between treatment groups or depend on patient condition.

Measurement error and biological variability

Observed biomarker values combine the underlying biological state with pre-analytical, analytical and within-person variation. Failure to separate these sources can make random fluctuation appear to be a treatment response.

Important sources include:

• Time of day, fasting status or recent activity.
• Specimen collection and handling.
• Storage duration and temperature.
• Assay batch and calibration.
• Device or reader differences.
• Acute illness or concomitant treatment.
• Natural within-person variation.

Standardised collection, blinded assessment, central analysis and quality control can reduce variation. Replicate measurements may be appropriate when a single value is insufficiently reliable.

Missing and censored measurements

Biomarker data may be missing because samples were not collected, could not be analysed or were affected by treatment discontinuation, illness or death. These causes have different implications and should not be combined without explanation.

Values below or above an assay's quantification range are not ordinary missing observations. Substituting a fixed value such as half the detection limit can bias results, particularly when censoring is common or differs between groups.

The analysis should report the extent and reasons for missingness, prespecify methods and test plausible alternative assumptions. Death or another clinical event that prevents measurement should be handled in a way consistent with the endpoint's interpretation.

Multiplicity and selective reporting

Studies may measure many biomarkers, time points and transformations. Testing numerous endpoints increases the probability of apparently favourable findings arising by chance.

The protocol should identify primary and secondary biomarker endpoints and specify any multiplicity adjustment or hierarchical testing strategy. Exploratory biomarker findings can generate hypotheses but should be labelled and validated independently.

Reporting only biomarkers that changed significantly can create a distorted account of biological effects. Complete reporting should include prespecified null and unfavourable results.

Using biomarker endpoints in trials

Biomarker endpoints can support early-phase development, dose finding, proof of mechanism, safety monitoring and confirmatory evaluation. Their role should match the development stage and the strength of evidence connecting the marker to patient outcomes.

Trial interpretation should examine:

• Whether assignment was randomised and allocation concealed.
• Whether biomarker assessment was blinded.
• Whether the assay was prespecified and validated.
• Whether groups differed in missing measurements.
• Whether the observed change was clinically meaningful.
• Whether other outcomes support or contradict the biomarker finding.
• Whether the result applies to the intended population.

A favourable biomarker endpoint can support biological activity without demonstrating net clinical benefit. Safety, symptoms, functioning and survival may still require direct evaluation.

Using biomarker evidence in health economic models

Economic models sometimes use biomarker changes to predict disease progression, treatment response, resource use or long-term health outcomes. This can be necessary when direct clinical outcomes are unavailable, but it introduces structural uncertainty.

The modelling pathway should identify:

1. Define the treatment effect on the biomarker endpoint.
2. Specify the relationship between the biomarker and later clinical outcomes.
3. Establish whether treatment changes that relationship.
4. Translate clinical outcomes into survival, quality of life and costs.
5. Represent uncertainty at every stage of the pathway.
6. Test alternative assumptions about persistence and surrogacy.

The model should not assume that treatment-induced biomarker change carries the same prognosis as naturally occurring differences between patients. That assumption requires specific evidence.

Valuing information from a biomarker endpoint

Earlier biomarker results may support faster decisions, smaller trials or targeted use, creating value through reduced research time and more efficient treatment. These benefits should be balanced against the risk of making a decision on an outcome that does not reliably predict patient benefit.

Economic consequences can include:

• Assay and infrastructure costs.
• Faster or smaller clinical studies.
• Earlier access to effective treatment.
• Treatment directed away from people unlikely to benefit.
• False-positive or false-negative classification.
• Costs and harms from acting on an invalid surrogate relationship.
• Additional confirmatory evidence requirements.

The value of biomarker information depends on whether it changes a decision and improves expected outcomes, not merely whether the measurement is statistically associated with disease.

A simplified example

Suppose a randomised trial evaluates a treatment intended to lower a biological marker associated with future cardiovascular risk. The mean marker falls by 18 units in the treatment group and by 6 units in the comparator group.

The difference in mean change is:

$$ Treatment\ effect = -18 - (-6) = -12\ \text{units} $$

The trial shows that the treatment changes the biomarker by an additional 12 units. It does not by itself show that cardiovascular events, survival or quality of life improve.

Using the result as a surrogate endpoint would require evidence that treatment effects on this marker reliably predict treatment effects on patient-relevant outcomes. The economic model should test the consequences if that relationship is weaker or shorter-lived than assumed.

Distinguishing biomarker endpoints from related concepts

Biomarker endpoints overlap with several outcome and test concepts but should not be treated as exact synonyms. The distinction depends on what is measured and how the result is used.

• A biomarker is the biological characteristic itself.
• A biomarker endpoint is the defined analysis of that characteristic in a study.
• A surrogate endpoint substitutes for a direct patient-relevant outcome and may or may not be biomarker based.
• A clinical endpoint directly reflects how a patient feels, functions or survives.
• A diagnostic test classifies whether a disease or condition is present.
• A predictive biomarker identifies variation in treatment effect rather than simply forecasting outcome.
• A composite endpoint combines several component outcomes and is not necessarily biomarker based.

Clear terminology prevents biological activity from being presented as clinical benefit without the required evidence.

Common misunderstandings

A biomarker endpoint can be objective and precisely measured while still being an uncertain basis for a treatment decision. Measurement quality and clinical meaning are separate questions.

Common misunderstandings include:

• Every biomarker endpoint is not a surrogate endpoint.
• Association with prognosis does not validate a biomarker as a treatment-effect surrogate.
• A statistically significant change is not automatically clinically important.
• Analytical validity does not establish clinical validity or clinical utility.
• Objective measurement does not remove selection, missing-data or analysis bias.
• A threshold discovered in one dataset may not perform similarly in another population.
• Change in a biomarker does not guarantee improvement in symptoms, functioning or survival.
• A validated biomarker for one intended use is not automatically valid for another.
• Precise laboratory results can still generate uncertain economic conclusions.

Interpreting a biomarker endpoint

Interpretation should begin with the endpoint definition, assay performance, timing and intended use. The conclusion should state whether the result demonstrates biological activity, predicts risk, guides treatment or substitutes for a patient-relevant outcome.

A useful biomarker endpoint provides reliable information for a defined decision without claiming more than its validation supports. When the endpoint is used to predict long-term health or economic value, the relationship to patient outcomes and the uncertainty around that relationship should remain explicit.