Randomised Trial vs Observational Study vs Meta-Analysis: What's the Difference?

Start hereEvidence pyramids put meta-analyses at the top. Yet Cochrane's own handbook warns that a meta-analysis can mislead seriously. So how do you tell which kind of study you are looking at, and what it can show?

A randomised trial lets chance decide who gets which treatment, then compares the groups. An observational study watches what people already do, without assigning anything. A meta-analysis is not a new study at all: it combines the results of studies that already exist. Each one answers a different kind of question, and each has its own blind spots.

011. At a glance

	Randomised trial	Observational study	Meta-analysis
In one line	Treatments are selected by a random process	The investigator does not assign participants to specific interventions	The statistical combination of results from two or more separate studies
Main strength	Chance balances known and unknown factors between groups, on average	Better suited to spotting rare or late side effects	Better precision than small studies alone
Main weak spot	Cannot answer every important question about a treatment	Confounding, because who gets what is often shaped by other factors	Can mislead seriously if biases in the studies are ignored
Reporting checklist	CONSORT 2025: a 30-item checklist and a flow diagram	STROBE, for cohort, case-control and cross-sectional studies	PRISMA 2020, for systematic reviews

Read across a row and you can see the trade-offs.

022. A simple picture

Say you want to know whether a new study app raises test scores. This is an invented example, so there are no results here, only the logic.

• Randomised trial. You let chance put some students on the app and some not, then compare scores. Chance balances known and unknown factors between the groups, on average.
• Observational study. You compare students who already chose the app with students who did not. Our reading: the students who chose it might also study more, and that alone could explain better scores. This is confounding. Cochrane’s handbook says confounding occurs when there are common causes of group assignment and outcome.
• Meta-analysis. You gather ten such studies and combine their results into one estimate. Most meta-analysis methods are variations on a weighted average of the studies’ results.

033. What each one is, and where it came from

Randomised trial

In a randomised trial, the treatments to be given are selected by a random process, such as a random-numbers table. The trial compares at least one test treatment with one control treatment, enrolled and followed at the same time.

Why chance? If randomisation works, it stops known or unknown factors from steering who ends up in which group. This means that, on average, each group has the same prognosis before the treatment starts. The handbook adds that chance imbalances between groups are not a source of systematic bias.

Controlled comparisons of treatments go back centuries. In 1747, James Lind, a Scottish naval surgeon, compared six proposed treatments for scurvy in a prospective controlled trial. He tested them in 12 patients on board ship.

For a long time, many trials split patients by alternation instead. Each new patient went to the next group in strict alternating order. The UK Medical Research Council’s trial of streptomycin, published in 1948, used a method of random allocation designed by the statistician Austin Bradford Hill.

Was it the first? Not quite. A 2025 history notes sporadic earlier use of random allocation, and says the 1948 trial is generally regarded as marking the emergence of the modern clinical trial. The James Lind Library calls its report a methodological landmark, because it described the trial in detail, especially the steps taken to prevent foreknowledge of treatment assignments.

Observational study

In an observational study, people may receive treatments or other interventions, but the investigator does not assign them. The STROBE reporting guideline covers the three main designs: cohort, case-control and cross-sectional studies.

• Cohort study. The investigators follow people over time, then see who develops which outcome.
• Case-control study. Investigators compare exposures between people with an outcome (cases) and people without it (controls).
• Cross-sectional study. Investigators assess everyone in a sample at the same point in time.

These designs fill gaps that trials leave. The STROBE authors write that randomised trials cannot answer all important questions about a given intervention. Observational studies are more suitable to detect rare or late adverse effects, and more likely to give an indication of what is achieved in daily medical practice.

The trade-off is a likely higher risk of bias. Cochrane’s handbook says potential biases are likely to be greater for non-randomised studies than for randomised trials when judging the effects of interventions. It also says these studies vary in their ability to estimate a causal effect.

Systematic review and meta-analysis

These two often travel together, but they are not the same thing.

A systematic review is a careful search-and-sort. It attempts to collate all the evidence that fits pre-specified eligibility criteria, to answer a specific research question. It uses explicit, systematic methods chosen with a view to minimising bias.

A meta-analysis is the maths that may come next. Cochrane defines it as the statistical combination of results from two or more separate studies.

Not every systematic review includes one. An important step is deciding whether it is appropriate to combine the numerical results of all, or some, of the studies. Cochrane’s handbook has a chapter on synthesis methods for reviews without meta-analysis.

044. Where they overlap

• A meta-analysis can pool either kind of study. Cochrane says review authors may be justified in including non-randomised studies when a question cannot be answered by randomised trials. It adds that results should always be interpreted with caution when they are included.
• The pooled answer inherits the studies’ quality. A 2016 paper argued that a meta-analysis of well-conducted trials at low risk of bias cannot be equated with one of observational studies at higher risk of bias.
• All three have reporting checklists. CONSORT, STROBE and PRISMA each list what a paper should report, so a reader can judge it. CONSORT’s site says the guidelines give readers the information they need to critically appraise the research.

055. Why the evidence pyramid is a simplification

You may have seen research drawn as a pyramid. Its versions show weaker designs such as case series at the bottom, and case-control and cohort studies in the middle. Randomised trials come next, with systematic reviews and meta-analyses at the very top.

The authors of that 2016 paper, Murad and colleagues, start by giving the pyramid its due. They call the pyramid intuitive and likely correct in many instances. But they also argue that “Study design alone appears to be insufficient on its own as a surrogate for risk of bias”.

They point to a rating system called GRADE. In the early 2000s the GRADE Working Group built a framework that judged certainty on numerous factors, not solely on study design. So they redrew the pyramid. They proposed to “remove systematic reviews from the top of the pyramid and use them as a lens through which other types of studies should be seen”.

Others urge care too. Oxford’s Centre for Evidence-Based Medicine, which still publishes its own levels of evidence, writes that hierarchies of evidence have been somewhat inflexibly used, and criticised, for some decades. When reviews include non-randomised studies, Cochrane’s handbook recommends focusing on specific features of a study’s design rather than on labels such as case-control versus cohort.

Our reading: the pyramid is a fair place to start, and a poor place to stop.

066. Common mix-ups

A meta-analysis is always the strongest evidence. It depends on what went in. Cochrane’s handbook says meta-analyses have the potential to mislead seriously, particularly if study designs, biases within studies, differences between studies and reporting biases are not carefully considered.

Systematic review and meta-analysis mean the same thing. One is the search and appraisal; the other is the statistics. Murad and colleagues describe the systematic review as “the process of selecting the studies” and the meta-analysis as “the statistical aggregation that produces a single effect size”.

Observational studies are worthless. They are better suited to some questions, such as rare or late side effects. Their weak spot is judging cause and effect, where bias is likely to be greater than in randomised trials, and some designs estimate a causal effect better than others.

Randomised means the groups are identical. Randomisation makes them alike on average, not identical. Cochrane’s handbook says imbalances between groups that are consistent with chance should not be read as a sign of bias.

077. Reading a headline about a study

Our longer guide, how to read a study without a science degree, goes through nine questions. Here are five that fit this page.

1. Did the researchers assign the groups, or watch people? Assignment by chance is what balances unknown factors, on average.
2. If they watched, what else differs between the groups? Our guides to correlation vs causation and numbers that mislead show how a link can run through a third thing.
3. If it is a review, were the rules for choosing studies set in advance? Rules set in advance are part of what makes a review systematic.
4. If it is a meta-analysis, what kinds of studies went in? A pool of well-run trials and a pool of higher-risk observational studies are not equal.
5. Has it been repeated? Our reading: one result is rarely the last word. Our look at famous psychology findings shows how some held up when tested again and some did not.

Where to go next

• The Cochrane Handbook: the official guide to preparing Cochrane systematic reviews on the effects of healthcare interventions. It can be read online; chapters 1 and 10 explain reviews and meta-analysis.
• CONSORT and SPIRIT: a standard way to report trial protocols and findings for randomised trials, with checklists.
• STROBE: the reporting checklists for observational studies. Its documents are open-access and available to download.
• PRISMA: a guideline designed to improve the reporting of systematic reviews.
• The James Lind Library: it illustrates the development of fair tests of treatments through history.
• MeSH, the National Library of Medicine’s index terms: short official definitions of study types. Its subject headings appear in MEDLINE/PubMed.

08Quick answers

09Use this

10Quick quiz

How we checked this · Last checked October 2026

• We read chapters 1, 8, 10, 12 and 24 of the Cochrane Handbook (version 6.5) on Cochrane's website, and the National Library of Medicine's MeSH definitions of randomised trials and observational studies.
• We read the STROBE statement and its explanation paper, and the CONSORT and PRISMA websites. Reporting checklists are revised from time to time, so we will check the versions again.
• For the history we read the James Lind Library's records and a 2025 history of randomised trials. We do not call the 1948 trial the first randomised trial, because that history notes earlier, sporadic use of random allocation.
• Oxford's Centre for Evidence-Based Medicine page was read in a browser. Its levels-of-evidence documents are download-only, and we did not use them.

How we check every article →

11Sources