A better test result is not always a better health outcome

Name the outcome first

A study may measure a laboratory value, a scan, symptoms, hospital admissions or survival. These outcomes answer different questions. Before reading the result, write down exactly what was measured.

A biomarker is a measurement of a biological process or response. It may help identify disease, monitor safety or show biological activity without being a substitute for a patient benefit. A surrogate endpoint is used to stand in for a particular clinical outcome. That additional role needs evidence; it does not follow simply from the availability of a precise laboratory test.

Two groups, two measurements, two different stories

Imagine a randomized comparison of fictional Strategy A and Strategy B, with 100 people per group and complete follow-up. There is no real drug or disease behind the example. Both groups begin with a mean value of 100 units on an invented laboratory marker. After 12 weeks, A's mean is 70 and B's is 85. The decreases are 30 and 15 units; the week-12 means differ by 15 units.

Now read a different outcome: by 12 months, 12 people in A and eight in B have newly needed help with routine daily activities. Nobody needed that help at baseline. This illustrative functional outcome gives recorded proportions of 12% and 8%, a difference of four percentage points in the opposite direction from the favorable marker result. Each person is counted once.

Those invented counts do not prove A causes harm. We have not supplied an uncertainty analysis, a real outcome instrument or a full safety report. They demonstrate a narrower point: the marker numbers alone do not tell you the patient-outcome numbers. A headline claiming better daily functioning from the 12-week laboratory result would skip the evidence needed to connect the two.

A useful predictor is not automatically a useful substitute

People with worse marker values may also have worse future health. That can make a marker useful for prognosis. Surrogacy asks a different question: when an intervention changes the marker, does that reliably predict its effect on the clinical outcome?

A treatment might alter the measured signal without changing the main disease process, act on only one of several important pathways, or have other effects that offset a benefit. A biological explanation is valuable, but it needs to be tested against treatment effects on patient outcomes. Evidence across relevant interventions and trials can help assess that relationship; a strong association within one group is not enough by itself.

Why researchers use intermediate measures

Some outcomes take years to observe. A quicker measurement can make a study feasible and help explore a mechanism. Whether it reliably predicts a patient benefit must be established; a plausible explanation alone is not enough.

The value of an early marker result depends on the job it is doing. Showing that a strategy affects its intended biological target can justify further investigation without establishing that patients will feel better. A validated surrogate can support stronger conclusions in its accepted setting. Read an early-stage result as a step in learning, and look for what the next study is designed to resolve.

Candidate, reasonably likely, validated: keep the context attached

These labels describe different levels of support. A candidate surrogate is still being evaluated. A reasonably likely surrogate has a scientific rationale but insufficient evidence to count as validated. A validated surrogate has strong evidence that an intervention's effect on it predicts a specified clinical benefit in the relevant setting.

Do not detach a validation claim from its context. Which disease and stage? Which patient population? Which kind of intervention? Which clinical outcome? FDA's endpoint table explicitly warns that suitability in one development program should not be assumed in another. Inclusion in that table is not a universal certificate of validity, and scientific understanding can change. This guide uses U.S. FDA terminology; it does not summarize every regulator's approval rules.

Patient reports can be direct evidence of benefit

A person's report of pain, breathlessness or difficulty with daily activities can directly address an outcome that matters to that person. Patient-reported outcomes are one kind of clinical outcome assessment; clinician reports, observer reports and standardized performance tasks can contribute other perspectives.

The distinction is about the claim being made, not whether the measurement looks objective or uses a questionnaire. A suitable symptom measure can support a claim about symptoms. It does not automatically support a claim about survival. The instrument must still capture the intended concept reliably in the studied population, and readers should ask what the score difference means in daily life. The presence of a numerical scale alone settles none of those questions.

Earlier access leaves questions for confirmation

FDA's accelerated approval pathway can use a surrogate or an intermediate clinical endpoint considered reasonably likely to predict benefit for serious conditions with unmet needs. This is a reason to inspect the remaining evidence, rather than treating a surrogate-based approval as proof of every hoped-for outcome.

Confirmatory studies are needed to verify clinical benefit. FDA describes withdrawal or a change in the labeled indication as possible consequences if benefit is not verified or is insufficient relative to risk. For any particular product, the current approval, label and confirmatory evidence need their own up-to-date review; this methods guide does not assess individual approvals. A shorter route to an initial answer makes the quality and completion of later evidence especially important.

Keep the claim within the measurement

A change in a biomarker should be described as a change in that biomarker. Improvement in a symptom score should not quietly become a claim about longer survival. This distinction makes a result more useful, not less interesting.

For the invented study, say that A produced the lower observed mean marker at week 12, while the reported 12-month functional proportions were 12% and 8%. Do not translate a 30-unit decrease into a 30% health improvement. The units describe the marker, and the example provides no conversion from marker units to personal benefit.

Apply the same discipline when connecting studies. A marker-focused paper and a later patient-outcome paper may answer different questions. Check whether their populations, interventions and follow-up are comparable before treating the second as confirmation of the first. If a composite endpoint mixes different kinds of outcomes, read its components separately.

A useful reading checklist

Find the prespecified primary outcome, the follow-up period and the comparison. Then look at adverse effects and missing data. If several outcomes were tested, ask how the analysis accounted for multiple comparisons.

Make a three-line note: what was measured, what clinical benefit is being claimed, and what evidence connects them. Record missing information as unknown. An unreported patient outcome is not a zero-event result, and an absent safety table is not evidence of no harm.

For future studies, look for the uncertainty they can actually reduce: longer follow-up, direct measurement of daily function, better ascertainment of harms, or validation in a new treatment context. Faster assays and AI-based measurements may make research easier, but improved measurement technology still needs evidence connecting treatment-induced changes to the intended clinical outcome.

Keep the marker, patient outcome and missing information separate

All data are invented. The two fictional groups each contain 100 people, with complete marker and functional follow-up and no need for daily-activity help at baseline. The functional outcome is counted once per person. It is shown separately from the marker and has a different measurement time. The exercise does not establish efficacy, safety or a valid surrogate.