A peptide can have an impressive mechanism, a registered trial and an encouraging abstract without having an established clinical benefit. Reading the research means asking exactly what was tested, against what, and in whom. The most useful first sentence is a description of the experiment—not a verdict about the molecule.
This guide concerns interpretation of published research. It does not provide instructions for using a compound or turn a research product into a medicine.
Start with the question, not the phase number
Phase 1 generally investigates early human safety, tolerability and what happens to the compound in the body. Phase 2 explores effects in a relevant patient population and further characterises risk. Phase 3 expands the evidence needed to assess benefit and harm. These purposes overlap, and a phase label does not establish that a study succeeded. Nor does reaching a later phase automatically confer approval. FDA clinical-development definitions.
Write down the population, intervention, comparator and follow-up period. A trial in people with a particular disease does not directly answer a question about healthy adults. A result for one formulation or route does not automatically transfer to another. Look for random allocation and masking, but also check whether the comparison actually isolates the question being asked.
Find the endpoint and its time point
The primary endpoint is the prespecified measurement carrying the main research question. It is usually central to the sample-size calculation. “Improved function” is incomplete: which instrument measured it, how was improvement defined, and at what time? A registry can help locate those details before the results narrative starts directing attention elsewhere. ClinicalTrials.gov outcome definitions.
Secondary outcomes can be valuable, but a favourable secondary result does not erase an unsuccessful primary analysis. Be especially careful when a report highlights one favourable visit, subgroup or questionnaire among many. The question is whether the result was planned and how multiple comparisons were handled—not simply whether a p value crossed a threshold. FDA guidance on multiple endpoints.
A biomarker is a different kind of outcome
Changing a blood measurement can show that an intervention affected biology. It does not necessarily show that people feel better, function better or live longer. A surrogate endpoint stands in for a clinical outcome; confidence in that substitution depends on evidence linking the marker to the specific benefit. A familiar or plausible biomarker is not automatically a validated surrogate. FDA surrogate-endpoint resource.
For example, increased concentration of a metabolite and improved walking ability answer different questions. Both can be measured carefully, but they should not be described as interchangeable evidence of healthy ageing.
Ask what the estimand actually estimates
An estimand precisely defines the treatment effect the trial seeks to estimate. It brings together the population, interventions, outcome, summary measure and handling of events after treatment starts. Such events can include stopping treatment or taking an additional treatment. ICH E9(R1).
Consider two hypothetical questions: what happens after people are assigned to a strategy, including those who stop; and what might happen if everyone continued as intended? Those are different targets. A treatment-policy approach and a hypothetical approach can therefore produce different estimates without either being a simple arithmetic error. Read the stated assumptions. Stopping treatment is also different from losing all follow-up information; missing data require their own handling.
Translate the headline into absolute numbers
Imagine an invented study in which an event occurs in 10 of 100 people in the comparison group and 5 of 100 in the intervention group over the same period. The relative reduction is 50%, while the absolute difference is five percentage points. If the figures were 2 of 100 and 1 of 100, the same relative reduction would represent one percentage point. These are worked numerical examples, not peptide trial results.
Both views matter. Read the confidence interval around the between-group difference, not just changes within each group. A statistically uncertain estimate is not proof of equivalence, and a statistically significant estimate may still be too small to matter for the outcome. CONSORT recommends reporting relative and absolute effects for binary outcomes alongside precision. CONSORT 2025 explanation.
Read the missing information as carefully as the result
Use a short final check:
- How many participants were randomised, analysed and lost to follow-up?
- Were the protocol, statistical plan and registration available before results?
- Are harms and withdrawals reported for each group?
- Does the conclusion match the prespecified outcome and observed population?
- Who funded the work, and are interests and data-access arrangements disclosed?
These questions follow the reporting priorities in CONSORT 2025. Good reporting makes a study assessable; it is not itself proof of low bias or a useful clinical effect. A careful conclusion states the finding, its uncertainty and the boundary beyond which the experiment cannot speak.
Continue with What laboratory quality tests can and cannot show for a closer look at interpreting research evidence.
Sources and further reading
- FDA: drug development and clinical-study phases — Official regulatory definitions; Accessed 2026-10-31.
- ClinicalTrials.gov: protocol registration definitions — Official trial-registry definitions; Accessed 2026-10-31.
- ICH E9(R1): estimands and sensitivity analysis — Primary statistical guideline; 2020-02-17.
- FDA: surrogate endpoints and clinical benefit — Official scientific guidance; Accessed 2026-10-31.
- CONSORT 2025: explanation and elaboration — Original reporting guideline; 2025-04-14.
- CONSORT 2025: updated reporting statement — Original reporting guideline; 2025-04-14.
- FDA: multiple clinical-trial endpoints — Primary agency guidance; 2022-10.
