Evidence Levels: From a Single Study to a Systematic Review

"A study shows" can mean almost anything, from a single experiment in mice to a review of dozens of large human trials. Knowing the main levels of evidence helps you place a claim on that spectrum quickly, and knowing their limits stops you from over-trusting the label at the top.

This article walks through each level of evidence, explains how researchers rate the certainty of evidence rather than relying on study type alone, looks at what happens when different levels disagree, and shows how to put this to use. For six quick questions to ask of any health claim, see How to Tell If a Health Claim Is Backed by Science.

The levels of evidence, one by one

Each type of study answers a different kind of question, and each has characteristic limits.

Lab and animal studies are an essential early step, but many results do not reach people. A 2024 umbrella review of 122 systematic reviews covering 367 therapies first tested in animals found that about 50% reached any human study, 40% reached a randomized trial, and 5% gained regulatory approval [1].

Systematic reviews and meta-analyses are usually placed at the top of evidence pyramids. Some methodologists argue they are better seen as tools for appraising and applying the studies beneath them, rather than as a separate, higher type of evidence [2]. Either way, their value depends on the quality of what goes into them.

Why study type is a starting point, not a verdict

Researchers who write clinical guidelines rate the certainty of a whole body of evidence, not just the study design. The most widely used system, GRADE, was developed because guidelines were inconsistent in how they rated evidence, and it has been increasingly adopted by organizations worldwide [3].

GRADE uses four categories: high, moderate, low, and very low certainty. "Randomized trials begin as high-quality evidence, observational studies as low quality" [4]. But that starting point can move. Certainty can be rated down for risk of bias, imprecision, inconsistent results, indirectness (for example, studies in a different population than the one you care about), and publication bias, and several factors can raise confidence in an estimate [4].

This explains why a single large, careful observational study can sometimes carry more weight than a small, poorly run trial, and why a guideline may describe the evidence for a treatment as "low certainty" even when randomized trials exist. When you see a certainty rating in a guideline or review, it is usually more informative than the study type alone.

When the levels disagree

Observational studies and randomized trials often agree on average, but they can diverge on specific questions. A 2024 Cochrane methodology review compared effect estimates from 2,869 randomized trials and 3,924 observational studies addressing the same health questions. It found "no difference or a very small difference" between the two on average, and 23 of 34 reviews reported estimates that were on average in agreement [5]. The authors rated the certainty of this finding as low, and noted that differences in the populations, interventions, comparisons, and outcomes studied can explain disagreements [5].

So observational evidence is not worthless, and agreement on average does not guarantee agreement on every question. A well-known example in which observational studies and a large trial pointed in opposite directions, involving beta carotene, is described in our guide to judging health claims. When an important decision depends on cause and effect, randomized evidence is still the most direct test.

Not every systematic review is reliable

The top of the pyramid has quality problems of its own. A 2016 analysis found that publication of systematic reviews and meta-analyses rose sharply, by more than 2,600% between 1991 and 2014, compared with 153% for all PubMed-indexed items [6]. The author concluded: "Possibly, the large majority of produced systematic reviews and meta-analyses are unnecessary, misleading, and/or conflicted" [6].

That is one researcher's assessment, and it is deliberately provocative, but it is a useful reminder to check a review before relying on it. Reporting guidelines such as PRISMA 2020 set out what a review should disclose: why it was done, what the authors did, and what they found [7]. Signs of a more reliable review include a clear search strategy, an assessment of the quality of included studies, and a rating of how certain the overall evidence is.

Putting the levels to use

When you come across a claim, three questions place it quickly:

• What level is the evidence? One lab or animal study, an observational association, a randomized trial, or a review of many trials?
• How certain is it? If a review or guideline gives a certainty rating, use it; "low" or "very low" means the answer could easily change.
• Does it apply to the claim? Check whether the studies used the same population, intervention, dose, and outcome that the claim describes.

For probiotics, the last question matters especially, because evidence belongs to specific strains; see Are All Probiotics the Same? Why Strains Matter and Probiotics, Explained: What They Are and What Research Shows.

The bottom line

Levels of evidence rank study designs by how well they support cause-and-effect conclusions, from expert opinion and lab studies up through observational studies, randomized trials, and systematic reviews. The ranking is a starting point: certainty also depends on bias, precision, consistency, and how directly the studies apply, and even systematic reviews vary widely in quality. When you meet a claim, ask what level of evidence supports it, how certain that evidence is, and whether it actually applies to the claim being made.