Begin with the exact claim
Evidence does not have one strength in the abstract. It is strong or weak in relation to a specific question.
A cell experiment may directly answer whether a compound activates a receptor in that test system. The same experiment is indirect evidence for a claim about a meaningful health outcome in people.
Different study types answer different questions
Mechanistic work asks how an effect might occur. In vitro research studies cells, molecules, or tissues outside a living organism. In vivo animal research examines effects in a whole nonhuman organism. Clinical research studies people.
Each step can add useful information. FDA explicitly notes that preclinical research can answer basic questions about toxicity and biological activity, but it is not a substitute for studying how a drug interacts with the human body.
Human evidence still needs appraisal
Observational studies examine what happens without randomly assigning the exposure. They can reveal associations and may be valuable when randomization is impossible or when studying uncommon or longer-term events.
Randomized trials assign participants to groups by chance. When designed and conducted well, randomization helps reduce confounding—other differences between groups that could explain the outcome. Randomized does not mean flawless; missing data, selective reporting, small samples, short duration, and other problems can still weaken confidence.
Look beyond a single study
Confidence grows when multiple relevant studies point in the same direction, use sound methods, measure outcomes that matter, and produce estimates precise enough to be useful.
A result can be statistically significant and still be small, uncertain in practice, or limited to a narrow population. Relay therefore keeps population, comparison, outcome, duration, and limitations attached to the finding.
Relay diagram
Closer to a human outcome—not an automatic quality score
What this can—and cannot—tell us
What it can tell us
- How directly a study addresses a specific claim.
- Which biases and limitations may change confidence.
- Whether findings are consistent, precise, and applicable to the population of interest.
What it cannot establish alone
- That every randomized trial is automatically trustworthy.
- That an animal or cell result will produce the same outcome in people.
- That one positive study settles every question about benefits, harms, or long-term effects.
Go deeperOptional · about 1 minute
Why Relay avoids a single evidence score
Evidence appraisal involves several dimensions: risk of bias, directness, consistency, precision, and the possibility of missing or selectively reported evidence. Compressing these into one unexplained number can conceal the reason confidence is limited.
Relay instead shows the evidence type and the limitation that matters. A user can see whether uncertainty comes from indirect animal evidence, a small sample, conflicting results, a short study, or another specific problem.
Why observational evidence still matters
Cochrane notes that some intervention questions cannot or are unlikely to be answered by randomized trials. Nonrandomized evidence may be necessary, especially for population-level interventions or some harms.
The tradeoff is greater concern about confounding and selection bias. Useful does not mean equivalent; the study must be interpreted for the question it can reasonably answer.
The takeaway
If you only remember one thing from this guide:
The strongest evidence is not the fanciest study label—it is evidence that directly and reliably answers the specific question.
Sources and support5 sources
- Step 2: Preclinical Research U.S. Food and Drug Administration
Describes in vitro and in vivo preclinical research and the questions it addresses before human testing.
- Step 3: Clinical Research U.S. Food and Drug Administration
States that preclinical research is not a substitute for studies in people and outlines core clinical-trial design questions.
Explains confounding, randomization, and the need to assess bias in randomized-trial results.
Explains both the uses of nonrandomized studies and their greater potential for confounding and bias.
- Methods Guide for Effectiveness and Comparative Effectiveness Reviews Agency for Healthcare Research and Quality
Supports multidimensional appraisal using domains such as risk of bias, directness, consistency, and precision.