A pooled number can look more decisive than a table of separate studies. It is useful only when the combination has a defensible meaning. Peptide papers may measure different molecules, experimental systems and outcomes, so the decision to pool must begin with the scientific question rather than the availability of numbers.
Write the sentence the pooled estimate would complete
Try the sentence: this estimate describes the effect of this defined comparison on this outcome in these settings. If the sentence requires several unrelated meanings of effect, the proposed synthesis needs to be separated or reframed.
Cochrane advises judging whether studies are sufficiently similar for their combination to be meaningful. Differences between populations or methods are not an automatic ban on pooling, but the intended question must justify the combination.Cochrane Handbook — Chapter 10: Analysing data and undertaking meta-analyses (opens in a new tab)
An original hypothetical collection contains receptor-binding affinity, a cellular reporter response and an animal functional score. They may all concern Peptide S, yet they do not estimate the same property simply because each provides a numerical result.
The shared compound name is a starting point for organising the literature. It is not, on its own, a statistical rationale for averaging every reported effect.
Compare the quantities before comparing the estimates
| Dimension | Compatibility question |
|---|---|
| Material | Are the entities within the defined question? |
| Comparison | Do the contrasts answer the same question? |
| Outcome | Do the measurements represent a common construct? |
| Timing | Are these comparable stages of response? |
| Units | Can values be converted on a justified basis? |
A change of units can be harmless when it is a known conversion of the same quantity. Converting nanomolar to micromolar does not change what was measured. Converting an unrelated outcome to a unitless score does not create the same kind of equivalence.
Likewise, standardising effects can help compare suitable measures of a common construct, but it does not prove that the construct is common. That justification comes from the outcome definitions and scientific context.
Keep unresolved material identity visible. A full-length peptide and a fragment might belong in a deliberately broad question, but they should not be presented as the same intervention through an unstated assumption.
Do not ask a model to repair the research question
A random-effects model allows related studies to have different underlying effects. Cochrane stresses that this does not remove heterogeneity or make it unimportant; the reasons for variation and the interpretation of the average still require attention.Cochrane Handbook — Chapter 10: Analysing data and undertaking meta-analyses (opens in a new tab)
For an original arithmetic illustration, values of +10 and −10 average to zero. That calculation is correct, but a zero average hides the opposite directions. This is not a meta-analysis: no study weights, uncertainty estimates or statistical model have been supplied.
The example shows why an average should be accompanied by the pattern it summarises. If opposing results occur in different models or time windows, those distinctions may be central to the explanation.
Use a structured synthesis when one number is unhelpful
Group studies by a scientifically relevant distinction and present their estimates, uncertainties and limitations within those groups. Explain why some results were not pooled, using criteria that can be checked rather than a vague statement that studies differed.
Avoid replacing an unsuitable average with a count of statistically significant results. Such a count discards effect sizes and differences in precision, and it can make small inconclusive studies appear to contradict larger informative ones.
A useful synthesis can identify which comparisons have been repeated, which outcomes remain isolated and where material definitions are unclear. These findings can be more informative than a single combined estimate.
If pooling is defensible, state its target clearly and show the constituent evidence. Readers should be able to understand both the average and the limits of applying it to a particular peptide, model or outcome.
Sources and further detail
- Cochrane Handbook — Chapter 10: Analysing data and undertaking meta-analyses (opens in a new tab)
Sections on meaningful combination, heterogeneity and random-effects interpretation read. The Peptide S collection and +10/−10 arithmetic are original illustrations; no real studies were pooled.
Sources checked 19 September 2026. Worked examples are illustrative unless a supplied report is explicitly identified. This article has not undergone independent scientific peer review.