Surrogate Endpoints vs. Health Outcomes: What Did the Study Actually Improve?

An improved number is an outcome, but which outcome?

A study can convincingly show that an intervention changes a laboratory value. That is a finding about the laboratory value. Whether it also establishes fewer symptoms, better functioning, or longer survival depends on the relationship between the measurement and those outcomes.

This is the problem surrogate endpoints are intended to address. Some direct health outcomes take years to observe or require very large studies. A measurement available sooner can be useful if it reliably predicts the benefit of interest. FDA’s resources describe both the role of surrogates and the importance of the evidence supporting them.

Biomarker and surrogate are not synonyms

A biomarker is a measured characteristic that can provide information about biological processes or responses. Its use might be diagnostic, prognostic, predictive, or something else. A surrogate endpoint has the more specific job of standing in for a clinical outcome in an evaluation.

The FDA-NIH BEST resource defines a validated surrogate in terms of evidence that an intervention’s effect on the surrogate reliably predicts clinical effectiveness in a defined context. The phrase “defined context” matters. A relationship established for one disease, intervention type, or population may not transfer unchanged to another.

For example, a measurement could help identify people at higher risk without being a reliable target for improving outcomes. Predicting a problem and demonstrating that changing a predictor prevents the problem are different scientific tasks.

A hypothetical shortcut that fails

Imagine a fictional marker, M, that tends to be higher among people who later experience an adverse outcome. Researchers develop an intervention that lowers M. It is tempting to infer that the intervention must reduce the outcome too.

Several explanations could break that inference. M might be a consequence of another process rather than a cause. The intervention might lower M without changing the process that matters. Or it might improve one pathway while causing harm through another. The marker could move in the desired direction while the overall health outcome remains unchanged or worsens.

This example does not imply that surrogates are generally useless. It shows why a correlation between a marker and an outcome is not enough to validate the marker as a substitute for measuring the intervention’s benefit.

Why the same surrogate can work differently in another setting

Suppose evidence supports a marker for one class of interventions. A new intervention changes the marker through a different mechanism. The earlier relationship may remain useful, but it cannot simply be assumed. The intervention’s other effects and the population’s circumstances can alter the connection between marker change and benefit.

Time also matters. A short-term marker response might not persist, while a direct outcome may depend on sustained change. A trial that measures only the early response leaves a different uncertainty from one that follows the marker and the clinical outcome over a relevant period.

What direct outcomes add

Outcomes such as symptoms, functioning, hospitalization, and survival are closer to the consequences people often care about. They still need clear definitions. A symptom score depends on the instrument and timing. A hospitalization outcome can depend on access and local practice. Survival can be measured from different starting points.

“Patient-important” therefore does not mean free of measurement problems. It means the outcome has a more direct relationship to the experience or consequence being evaluated. Strong interpretation still asks how it was measured, which people were included, and what uncertainty remains.

Composite endpoints create a related reading problem

A composite endpoint combines several outcomes into one measure. It is not necessarily a surrogate, but it can create a similar gap between a headline and the result underneath it. A composite may improve mainly because a more frequent, less serious component changed, while a rarer, more serious component did not clearly differ.

In a fictional composite of an extra clinic visit, hospitalization, or death, a reduction in extra visits could drive the overall result. Saying only that “major outcomes improved” would obscure which component changed. The component results and their uncertainty help explain the practical meaning.

Reading the conclusion at the right level

A study showing improvement in a marker can be described as evidence of that improvement. If the marker is a well-supported surrogate for the relevant context, a stronger interpretation may be justified. If the connection remains uncertain, the conclusion can identify the promising signal while leaving the clinical question open.

That phrasing preserves useful evidence without dismissing early research or treating a convenient measurement as a complete health outcome. The key is matching the conclusion to what was actually measured and how well that measurement supports the benefit being claimed.