03 · What You Need to Know
Using Early Measurements Without Confusing Them With Final Benefits
What Is a Surrogate Outcome?
A surrogate outcome is a measurement used as a substitute for another outcome that is more directly relevant to the intended scientific, clinical, educational, or practical decision.
In clinical research, the term surrogate endpoint commonly refers to a biomarker or intermediate endpoint intended to substitute for a clinical endpoint and predict clinical benefit or harm.
For example, researchers may investigate whether a biomarker can stand in for a later clinical event.
In educational research, an immediate assessment may be proposed as a proxy for long-term competence or professional performance.
However, not every short-term educational outcome should automatically be called a validated surrogate. It may simply be a legitimate short-term outcome that addresses a narrower research question.
The defining issue is whether researchers intend to use the observed measure to make inferences about another outcome.
Distinguish Surrogate, Intermediate, and Final Outcomes
Final outcome
The result of principal substantive interest, such as sustained competence, functional improvement, or a clinically meaningful event.
Intermediate outcome
A result occurring before the final endpoint that may be important in its own right.
Surrogate outcome
An intermediate or substitute measure used to support an inference about a different, more consequential outcome.
Process or implementation measure
Information about intervention delivery, participation, adherence, or use that does not automatically represent a benefit.
A short-term outcome becomes a surrogate in the relevant sense when it is used to stand in for a different endpoint.
For example, immediate knowledge gain is a direct outcome for a question about immediate learning. It becomes a proposed surrogate when researchers use it to claim improvement in professional competence years later.
Why Do Researchers Use Surrogate Outcomes?
Final outcomes may require long observation periods, large samples, specialized assessments, or substantial financial resources.
Some occur infrequently, making adequately powered studies difficult.
Surrogate outcomes may occur earlier or be easier to measure, potentially reducing study duration and cost.
They can also support early investigation of promising interventions before larger or longer studies become feasible.
However, these practical advantages do not establish validity.
A convenient measurement may produce precise estimates while providing weak evidence about the final outcome.
Researchers should distinguish feasibility gains from the strength of the scientific inference.
Predicting an Outcome Is Not the Same as Predicting an Intervention Effect
This distinction is fundamental to surrogate endpoint evaluation.
Suppose students with higher immediate assessment scores generally demonstrate better professional performance several years later.
The association may establish that early scores have prognostic value.
However, it does not necessarily establish that an intervention increasing immediate scores will improve later professional performance.
The intervention might improve performance on the immediate test without changing durable knowledge or practical skills.
Similarly, a clinical biomarker may predict disease risk without reliably predicting the effect of a treatment on clinically important events.
Prentice's influential framework and subsequent methodological literature distinguish statistical associations from the stronger conditions required for valid surrogate-based inference.
Why a Strong Correlation May Be Insufficient
Researchers sometimes justify a surrogate by demonstrating a high correlation with the final outcome.
However, correlation between individuals' surrogate and final outcome values is not equivalent to correspondence between intervention effects on those outcomes.
For example, students with high examination scores may generally perform well in professional practice.
Nevertheless, an intervention that raises examination scores through test-specific preparation may not improve professional performance.
The correlation may reflect prior ability, motivation, or other common causes.
Researchers should therefore avoid treating individual-level predictive validity as sufficient evidence of trial-level surrogacy.
Individual-Level and Trial-Level Surrogacy Address Different Questions
Surrogate validation frameworks distinguish different forms of evidence.
Individual-level association concerns whether participants with different surrogate values also tend to have different final outcomes.
Trial-level association concerns whether interventions producing larger effects on the surrogate also tend to produce larger effects on the final outcome across studies or intervention comparisons.
Trial-level evidence is particularly relevant when researchers intend to predict an intervention's effect on the final endpoint.
However, trial-level association alone does not eliminate every concern about transportability, intervention mechanisms, or bias.
Buyse and colleagues have developed influential approaches to evaluating surrogate endpoints using evidence from multiple trials.
Such validation generally requires substantially more evidence than a single small study can provide.
Biological or Theoretical Plausibility Is Helpful but Not Sufficient
A surrogate may appear reasonable because it lies on a plausible causal pathway.
For example, an educational intervention may improve conceptual understanding, which may contribute to later professional competence.
However, professional competence may also depend on supervised practice, decision-making, workplace conditions, and other abilities.
An intervention could improve the proposed surrogate without improving these additional determinants.
Similarly, a treatment may affect a biomarker through one pathway while influencing the final clinical outcome through another.
Researchers should use causal theory to evaluate the proposed relationship while recognizing that theoretical plausibility does not establish empirical surrogacy.
The Surrogate Paradox
A particularly important concern is that an intervention may improve a proposed surrogate while having no beneficial effect, or even a harmful effect, on the final outcome.
This possibility is sometimes discussed as the surrogate paradox.
It can arise when an intervention affects the final outcome through pathways not adequately represented by the surrogate or when the assumed causal relationships are incomplete.
For example, a hypothetical educational intervention might improve immediate test scores through intensive practice while reducing opportunities for independent problem-solving.
The intervention could therefore improve the short-term measure without producing the intended long-term competence.
This example illustrates a causal possibility, not an established effect of any particular instructional approach.
Researchers should not assume that movement in a favorable direction on the surrogate guarantees movement in the same direction on the final outcome.
Surrogate Validity May Be Specific to an Intervention Class
A surrogate validated for one treatment mechanism may not perform equally well for another.
For example, two interventions might produce similar changes in a biomarker through different biological pathways.
The relationship between those changes and the final clinical outcome may differ.
Likewise, an immediate educational assessment might predict long-term performance under one instructional approach but not another that directly trains test-taking strategies.
Researchers should examine whether validation evidence applies to the intervention class, population, setting, and final outcome of interest.
Surrogate validity should not be treated as a permanent property of a measurement independent of its intended use.
Measurement Timing Matters for Both Outcomes
Surrogate outcomes are often selected because they occur earlier than final outcomes.
However, an early measure must be assessed at a time that corresponds to the process it is intended to represent.
For example, measuring knowledge immediately after instruction may capture short-term acquisition but not retention.
A delayed assessment may provide stronger evidence about persistence, although it may still not represent professional competence.
Researchers should consider the anticipated temporal relationship between intervention, surrogate, and final outcome.
They should also evaluate whether repeated surrogate measurements provide useful information about trajectories rather than relying on one arbitrary endpoint.
Measurement Quality Remains Essential
A surrogate must be measured with appropriate validity and precision for its intended use.
For example, a poorly designed knowledge test may fail to represent the learning construct even before questions about long-term prediction arise.
Measurement error can weaken observed associations and complicate surrogate validation.
Researchers should evaluate construct coverage, reliability or precision, responsiveness, and comparability across groups.
However, excellent measurement quality does not establish that the surrogate predicts treatment effects on the final outcome.
Measurement validity and surrogate validation are related but distinct requirements.
Can Surrogates Be Used as Primary Outcomes?
Yes, a study may designate a surrogate as its primary outcome when that choice is appropriate to its objectives and inferential limitations.
For example, an early-phase study may primarily investigate whether an intervention changes a biologically or theoretically relevant intermediate measure.
In that case, the study can legitimately make a claim about the surrogate itself.
However, the researchers should not automatically claim that the intervention improves the final outcome.
In some regulatory contexts, surrogate endpoints may support particular approval pathways under specified evidentiary requirements.
Those arrangements should not be generalized to all research settings or treated as evidence that every proposed surrogate is validated.
When Should Researchers Retain Final Outcome Measurement?
When feasible, collecting both surrogate and final outcomes can provide valuable evidence.
The surrogate may offer earlier information, while the final outcome allows researchers to evaluate whether the expected benefit actually occurs.
Long-term follow-up can also contribute to future surrogate validation.
However, continued measurement may require additional resources, participant retention procedures, and appropriate authorization.
Researchers should consider whether a follow-up extension, administrative record linkage, or subsequent study could provide final outcome evidence.
If the final outcome cannot be measured, the limitation should remain explicit.
Do Not Substitute a Convenient Proxy for an Unavailable Outcome Without Justification
Researchers may encounter an existing dataset containing only short-term indicators.
For example, a university may have immediate examination scores but no information about graduates' later professional performance.
The available data may support a study of immediate academic achievement.
They do not automatically support a study claiming improved professional competence.
Researchers should consider narrowing the research question to the outcome actually measured.
This is preferable to treating an unvalidated proxy as equivalent to the final outcome merely because it is the only available information.
Stakeholder Importance Should Influence Surrogate Selection
The final outcome is often important because it represents a benefit or harm that matters to patients, learners, practitioners, or other stakeholders.
Researchers should consider whether the surrogate captures a meaningful part of that benefit.
For example, students may value the ability to apply knowledge independently more than a small increase in an immediate test score.
A study using test performance as a surrogate for practical competence should explain the relationship and its limitations.
Stakeholder involvement can help identify when an intermediate indicator is insufficient for the intended decision.
See how stakeholder-important outcomes should influence outcome selection.
Use a Surrogate Outcome Evaluation Matrix
| Evaluation question |
Evidence to examine |
Potential concern |
| What final outcome is being substituted? |
Clearly defined final endpoint |
The target benefit is vague |
| Why is the surrogate proposed? |
Theoretical or biological rationale |
Convenience is the only justification |
| Does it predict final outcomes? |
Relevant longitudinal evidence |
Association may reflect confounding |
| Does it predict intervention effects? |
Appropriate trial-level or other causal validation evidence |
Individual correlation is mistaken for surrogacy |
| Is the evidence transportable? |
Population and intervention comparability |
Validation applies to another setting |
| Is measurement appropriate? |
Validity, precision, and timing evidence |
Surrogate itself is poorly measured |
| What claims are justified? |
Strength and limitations of validation |
Final benefit is overstated |
The matrix does not establish surrogate validity by assigning scores. It helps researchers identify the evidence required for the intended inference.
Watch Out
Do not assume that a surrogate is valid simply because it correlates strongly with the final outcome or lies on a plausible causal pathway. Predicting individual outcomes and predicting intervention effects are different tasks. A treatment may improve a surrogate without improving the outcome that matters most.
