| Study type | Place in the hierarchy | What it can show | Planning use |
|---|---|---|---|
| Systematic review or meta-analysis of all relevant randomized controlled trials | Level I | Combines the results of many studies and gives more weight to higher-quality ones | Look for one first. Weigh it most when the trials inside it are sound. |
| Well-designed randomized controlled trial | Level II | Participants are randomly assigned to try the treatment or activity and are compared with a group that does not | The strongest single-study design. Still check size, follow-up and funding. |
| Well-designed case-control or cohort study | Level IV | Patterns in people who were observed rather than assigned | Use it to spot leads. Do not treat it as proof that a food or supplement causes an effect. |
| Opinion of authorities or expert committee reports | Level VII | Informed judgement without new trial data | Use it for context. It cannot replace trial or review evidence. |
Levels follow the evidence hierarchy in a University of South Alabama nursing evidence-based practice guide, and the descriptions of random assignment and reviews draw on McMaster Optimal Aging.
Knowing how to tell if a health study is reliable comes down to design, size, follow-up and funding: check how participants were assigned, who and how many were studied, for how long, and who paid. A single study rarely settles a claim.
The table is a ladder, and Understanding The Levels Of Evidence Pyramid: A Practical Guide For Clinicians And Students (2026) – Testpledge covers it in more depth. The sections below follow the order you would use on a real supplement or food claim: find the study, identify its design, test its quality, check who is behind it, and separate association from cause.
Finding the original study
A supplement or food claim is only as good as the study behind it, so the first job is to name that study. Product pages and social posts often cite none. McMaster Optimal Aging advises looking for proof that claims or recommendations are supported by study findings published in a credible academic journal. If a claim points to no study at all, treat it as unsupported until you locate one.
Trace the claim to a paper
- Write down what the claim actually says. Note the exact food or ingredient, the outcome promised and the group of people it supposedly applies to. You will compare these against the paper later.
- Use the details the article gives. Author names, journal and year are enough to find most papers. A vague reference such as “researchers found” is a warning sign in itself.
- Search PubMed by title. Paste the full title into PubMed. With no title, search the ingredient or food together with the health condition and add the words randomized or systematic review to surface stronger designs.
- Open the abstract, then the full text if you can. The abstract gives the design and headline result. The methods section shows how people were assigned and what was measured.
Read the paper, not the press release
Headlines compress. Compare the claim you wrote down with the paper’s population, intervention and outcome. A paper that tested one dose of one ingredient in one group does not automatically support a broad statement about the food or supplement in general.
Look at the conclusion and the limitations passage together. Authors often state the boundaries of their own findings, and those boundaries tend to disappear from retellings.
Check the trial registry
For a trial, search a registry such as ClinicalTrials.gov using the intervention and the condition. Compare the outcomes listed there with the outcomes the paper reports. If the paper emphasises results the registry never listed, treat the finding more cautiously and look for independent confirmation.
Registry entries also show how far a treatment has progressed. The 4 Phases Of Clinical Trials: How New Treatments Move From Lab To Clinic (What To Expect In 2026) – Testpledge explains why an early-stage trial answers a different question from a late-stage one.
Search for a review before leaning on one paper
Search the Cochrane Library for the ingredient or food. The Cochrane Database of Systematic Reviews is described as a highly regarded database of systematic reviews prepared by The Cochrane Collaboration. A review that already weighs many trials is a sturdier foundation than any single paper you happen to find.
Study designs and what each can show
Design decides how much weight a result can bear. Two papers can both report a benefit, yet one may be able to support a cause-and-effect conclusion while the other can only describe a pattern. Identify the design before you read the numbers.
Randomized controlled trials
A randomized controlled trial randomly assigns participants to try an activity or treatment and compares them with people who do not. McMaster Optimal Aging says the highest quality studies randomly select participants to try the treatment. For a full definition and the conditions under which you can trust one, see what is a randomized controlled trial.
An editorial in PubMed Central calls the randomised controlled trial the most reliable and frequent but rather clumsy study design. The same piece explains that a control group must carry out the same activities as the treated group without the hypothesised active ingredient. In a supplement trial, that means the comparison group follows the same routine without the ingredient under test. Reliability therefore depends on how well the trial was built, not on the label alone.
- Check the comparison. If the control group differs from the treated group in more than the ingredient, the result cannot be pinned on the ingredient.
- Check who knew what. Whether participants and researchers were aware of group assignment can colour the outcome, and Single-Blind Vs Double-Blind Studies: How To Tell Which Design Gives More Reliable Results In 2026 – Testpledge compares the arrangements.
- Allow for expectation. Belief in a treatment can change how people feel, and Placebo Effect Examples: 9 Real-World Cases That Reveal How Expectation Shapes Health (2026 Update) – Testpledge illustrates why a comparison group matters.
Case-control and cohort studies
Well-designed case-control and cohort studies sit at Level IV in the hierarchy, below randomized controlled trials. They watch or compare people rather than assigning them to a treatment. When a headline says a food is “linked to” an outcome, ask whether anyone was assigned to anything. If not, you are probably reading an observational finding.
Experiment Vs Observational Study: How To Choose The Right Method In 2026 – Testpledge sets out the broader choice. To separate the two common observational designs, read Case-Control Vs Cohort Studies: How To Choose The Right Observational Design In 2026 – Testpledge. A frequent variant is covered in What Is a Retrospective Cohort Study? A Practical Guide for Clinicians and Researchers (2026) – Testpledge.
Laboratory and animal research
Findings from cells, tissue or animals answer a narrower question than a study in people eating or taking the product. Supplement marketing sometimes leaps from a laboratory result straight to a recommendation. Ask whether the study took place in a living person, and if it did not, treat the result as background rather than a reason to buy. In Vivo vs In Vitro: How They Differ And When Each Matters (2026 Guide) – Testpledge and Ex Vivo Vs In Vitro: How They Differ And When To Use Each (Practical Guide 2026) – Testpledge explain the terms you will meet in the methods section.
Systematic reviews and Cochrane
A systematic review sits at the top of the hierarchy as Level I when it covers all relevant randomized controlled trials. McMaster Optimal Aging calls these the most trustworthy research findings that compare the results of many studies, giving more weight to higher-quality ones.
A University of California, Merced library guide notes that systematic reviews are created after reviewing and combining information from both published and unpublished studies, and that the included studies should collectively study the same outcome. Cochrane Summaries are described as drawing on 37,000 contributors and as free from commercial sponsorship and other conflicts of interest. For a supplement question, a Cochrane review on the same ingredient and outcome is usually the best single document to find.
Quality checks inside the paper
Once you know the design, test the execution. A well-chosen design can still produce a weak study if the sample is tiny, the follow-up is short or the numbers are read carelessly.
Who was studied and how many
McMaster Optimal Aging contrasts a small and a large result: if 7 of 10 people show a benefit, it is hard to interpret, whereas if 700 out of 1000 people show the same benefits, you can have more confidence. Small studies are also easy to over-read. The PubMed Central editorial warns that a chance finding in a small study exaggerates the benefit, and that the first observed effect often fails to replicate, a pattern called the winner’s curse.
Equally important is who the participants were. The editorial gives a plain rule: if we study healthy students, the answer may only apply to healthy students. A trial of a supplement in healthy young volunteers says little about older adults or people with a medical condition. Match the sample to the person the claim is aimed at before you take the result personally.
How long and how measured
Look for studies that use accurate, reliable ways to measure effects more than once and over a longer period. McMaster Optimal Aging gives follow-up of six months or a year as an example of a timescale that can show longer-term benefits and harms. A study that measures once, shortly after the start, cannot tell you whether an effect lasts or whether problems emerge later.
Rare harms and the numbers behind them
Absence of problems in a small run is not proof of safety. The PubMed Central editorial describes 20 procedures without problems, and notes that over 100 procedures with the same risk there would be a 95% chance that the number of deaths will be between 0 and 16. For supplements, the lesson is that short, small trials with no reported side effects cannot rule out uncommon harms.
When results are expressed as risk, the format changes the impression. Relative Risk vs Absolute Risk: How To Read Medical Statistics And Make Better Health Decisions (2026 Guide) – Testpledge shows how to read them, and what is number needed to treat offers a practical way to express benefit. A “significant” label is not the same as a large or useful effect, as What Does It Mean For Results To Be Statistically Significant? A Clear, Practical Guide (2026) – Testpledge explains.
Who is behind the claim
A paper can be well built and still be presented selectively. Look at who produced the claim and who funded the work. A Hofstra University library guide states that reliable sources are accurate, evidence-based, and created by people with relevant expertise. Use those three tests on the author of the paper and on the person relaying it.
- Match expertise to the topic. The same guide points out that an expert in one field, such as astrophysics, is not automatically an expert in another, such as nutrition. A confident influencer or a credentialed professional outside nutrition does not automatically carry authority on a food claim.
- Check the funding. McMaster Optimal Aging says it is wise to be skeptical of studies supported or promoted by for-profit businesses. Skeptical does not mean dismissive. It means looking for independent trials or a review that reaches the same conclusion.
- Separate the study from its promotion. A neutral trial may be quoted by a seller well beyond what it found. Compare the seller’s wording with the paper’s own conclusion.
Correlation, causation and other explanations
Correlation means two things occur together, such as a food eaten often and a health outcome. Causation means one produces the other. Observational data can show the first without establishing the second, because other factors may drive both. The PubMed Central editorial proposes a simple test for any claim: are there alternative explanations, have these been explored, how possible are they?
This is why a control group matters. By carrying out the same activities without the suspected active ingredient, it lets researchers see what changes because of the ingredient itself. When a paper has no such comparison, ask what else differs between people who use a product and people who do not. Causation vs Correlation: How To Tell What’s Really Driving Results (2026 Guide) – Testpledge develops the distinction, and what is a confounding variable shows how hidden factors distort a result and where to find them in a limitations section.
Deciding what to do with a claim
Match your response to the best evidence you found rather than to the most exciting headline.
- A sound systematic review exists. Give it the most weight, since it ranks highest in the hierarchy and draws on many trials. This suits readers deciding whether a long-promoted supplement deserves a place in their routine.
- Only a single trial exists. Treat the result as a promising lead, check the sample and follow-up, and wait for independent confirmation before changing habits.
- Only observational findings exist. Read them as patterns that suggest questions. They fit decisions with little downside, such as adding a food you already enjoy, better than decisions involving cost or risk.
- Only lab work, opinion or a seller’s page exists. Treat the claim as unsupported and do not pay extra for it.
If a product carries a possible harm, favour caution and wait for stronger evidence. A few remaining questions come up repeatedly, and the answers are below.
Frequently asked questions
Is a systematic review the same as a meta-analysis?
No. A systematic review gathers and assesses the studies on a question, while a meta-analysis also adds a statistical analysis that quantitatively combines them. The difference is set out in Systematic Review Vs Meta-Analysis: How They Differ And When To Use Each (2026 Guide) – Testpledge, drawing on a UC Merced library guide.
Is a systematic review always reliable?
No, because it is only as good as the studies inside it. A Cochrane review of acupuncture analysed 22 RCTs, yet its conclusions were limited by the low quality of evidence in those studies and by lack of blinding, according to the UC Merced guide.
Why do some trials need so many participants?
When the treatments being compared differ little in effect, a trial often needs hundreds of participants to detect the difference, according to the PubMed Central editorial. A large number can therefore reflect a hard question rather than padding.
Does an average describe a typical person in a study?
Not always. The choice of average changes what looks typical. The same editorial gives the example of UK household disposable income in the financial year ending 2022, where the median was about £32K and the average £40K, so check which measure a paper reports (PubMed Central).
