Reading the evidence
Systematic Review vs Meta-Analysis: What Each Is and How to Judge One
AR
· 10 min read
A summary of published research, with citations. It states no amount, schedule or route, and makes no claim about what this compound does in a person. Supplied for laboratory research use only.
On this page · 8 sections
- What makes a review "systematic"?
- What does a meta-analysis add?
- What is heterogeneity, and what does I² mean?
- When two meta-analyses disagree: GLP-1 drugs and hair loss
- What does GRADE certainty mean?
- Why "garbage in, garbage out" applies
- How to judge a systematic review or meta-analysis in five minutes
- Frequently asked questions
A systematic review is a planned, reproducible search for every study that answers a defined question, followed by an appraisal of how trustworthy each one is. A meta-analysis is a statistical technique that pools the results of several studies into one estimate. Many systematic reviews include a meta-analysis, but not all: when the studies are too few or too different, the review describes them without pooling. A meta-analysis built on hand-picked studies rather than a systematic search is just a weighted average of a selection.

What makes a review "systematic"?
The difference from an ordinary narrative review is that every step is decided in advance and written down, so someone else could repeat it and get the same set of studies:
- A focused question, usually framed by population, intervention, comparison and outcome.
- A protocol, ideally registered before the work starts, often on the PROSPERO register, so that the questions and methods cannot be adjusted after the results are seen.
- A search of several databases, with the search strings reported, plus trial registries for unpublished work.
- Explicit inclusion criteria, applied by two people independently.
- A risk-of-bias assessment of every included study.
- Transparent reporting, conventionally to the PRISMA standard first published by Moher and colleagues and updated in 2020 (PMID 33782057). That includes the flow diagram showing how many records were found, screened, excluded and why.
A narrative review, by contrast, is an expert's account of a literature. It can be excellent, but nothing stops the author from citing the studies that fit and leaving out the ones that do not, and a reader cannot tell which happened.
What does a meta-analysis add?
A number. The meta-analysis takes each study's effect estimate, weights it by its precision (larger, less noisy studies count for more) and combines them into a pooled estimate with a confidence interval. The result is usually drawn as a forest plot: one line per study, a diamond for the pooled effect.
Two broad models are used. A fixed-effect model assumes every study is estimating the same underlying effect. A random-effects model assumes the true effect varies between studies and estimates its average. Most meta-analyses of real-world literatures use random effects, which gives wider intervals and is more honest about the variation.
| Narrative review | Systematic review | Meta-analysis | |
|---|---|---|---|
| Question | Broad, often a topic | Focused, prespecified | Focused, prespecified |
| Study selection | Author's choice | Reproducible search and criteria | Inherits it from the review, or not |
| Appraisal of bias | Optional, informal | Every study, with a stated tool | Should be carried into the analysis |
| Output | Prose | Prose and tables, sometimes a pooled estimate | A pooled estimate with a confidence interval |
| Example on this site | Many mechanism reviews | Poolton 2026 on BPC-157 (no pooling) | Cheng 2026 on GLP-1 drugs and hair loss |
What is heterogeneity, and what does I² mean?
Heterogeneity is variation between study results beyond what chance would produce. It is the first thing to look for in a meta-analysis, because a pooled average of very different results may describe no real population at all.
The usual summary is I², introduced by Higgins and colleagues. It estimates the percentage of the variation across studies that reflects real differences rather than chance. Low values suggest the studies agree; high values mean the pooled number is averaging over genuinely different effects, and the reader should ask why. Different populations, different drugs within a class, different follow-up and different definitions of the outcome are the usual reasons.
I² has limits. With only a handful of studies it is estimated imprecisely, so an I² of 0% across three studies is weak evidence of agreement. And statistical consistency says nothing about whether the studies were any good: three studies sharing the same bias will agree perfectly.
When two meta-analyses disagree: GLP-1 drugs and hair loss
Hair loss with GLP-1 medicines is a clean example of how the inputs determine the answer. The trial data are set out in GLP-1 hair loss; here the interest is the reviews.
Cheng and Chang pooled nine interventional studies, seven randomised and two not, covering 4,114 people on GLP-1 drugs. Against placebo, the risk ratio for hair loss was 3.25, with a 95% confidence interval of 1.44 to 7.36. Restricted to randomised trials in overweight or obesity it was 3.59.
Viquez Burboa and colleagues searched five databases and included 17 studies: cohort studies, pharmacovigilance analyses and case series, covering more than a million patient-exposures. Their pooled odds ratio for non-scarring alopecia was 1.40, drawn from three cohort studies with an I² of 0%.
Both are systematic reviews with meta-analyses, and they answer different questions. One pools adverse events recorded in trials against placebo; the other pools health records against other patients. They also use different effect measures, a risk ratio and an odds ratio, which are not interchangeable. Neither is wrong; quoting either as "the" risk would be. The second paper's conclusion also attributes the association mainly to micronutrient deficiency from weight loss. A pooled association cannot establish a mechanism, however large the sample.
A third review, by Gupta and colleagues, took 24 studies and did not pool them. It described patterns by drug and subtype instead. That is a legitimate choice when the studies measure different things.
What does GRADE certainty mean?
GRADE is the most widely used system for rating how much confidence a body of evidence deserves, set out by Guyatt and colleagues for the GRADE Working Group. It rates certainty as high, moderate, low or very low. Evidence from randomised trials starts high and evidence from observational studies starts low. Either can be downgraded for risk of bias, inconsistency, indirectness, imprecision or suspected publication bias.
The certainty rating is often more useful than the pooled number. Lundh and colleagues' Cochrane review is an example: industry-sponsored drug and device studies more often reported favourable efficacy results, with a risk ratio of 1.27, rated moderate-quality evidence. More often reaching favourable conclusions came in at 1.34, rated low-quality.
The BPC-157 systematic review by Poolton and colleagues shows what a systematic review looks like when there is almost nothing to pool. It was registered on PROSPERO and reported to PRISMA. It searched Embase, PubMed, the Cochrane Library and ClinicalTrials.gov, and found three human studies with 30 participants in all, none with a control group. The findings were synthesised narratively, without a meta-analysis. Overall certainty was rated very low for risk of bias, indirectness and imprecision. A well-conducted review of weak evidence is still a review of weak evidence. The animal literature behind that compound is in BPC-157 and connective tissue. BPC-157 is not approved by Health Canada, the FDA or the EMA.
Why "garbage in, garbage out" applies
Pooling does not remove bias; it averages it and narrows the confidence interval around it. A meta-analysis of small, unblinded studies can produce a precise estimate of the wrong number.
Wood and colleagues measured this across 146 meta-analyses containing 1,346 trials. For subjective outcomes, trials with inadequate or unclear allocation concealment exaggerated effects, with a ratio of odds ratios of 0.69. So did trials without blinding, at 0.75. For objective outcomes, the bias was small. Why blinding matters is the subject of open-label vs placebo-controlled.
A second problem is volume. Ioannidis counted 266,782 PubMed items tagged as systematic reviews and 58,611 as meta-analyses between 1986 and 2015. Annual output had grown by 2,728% and 2,635% respectively, against 153% for PubMed as a whole. He concluded that many are redundant, misleading or conflicted. A meta-analysis is a study design, not a seal of quality.
The third problem is specific to small fields. If most of the literature comes from one group, pooling twenty of its studies measures the consistency of one laboratory, not independent replication. How to read a peptide study discusses that pattern.
How to judge a systematic review or meta-analysis in five minutes
- Was a protocol registered before the search? Look for a PROSPERO number.
- Which databases were searched, up to what date, and were trial registries included?
- What study designs were allowed in, and were randomised and observational data pooled together?
- How was risk of bias assessed, and did the analysis change when high-risk studies were removed?
- What is I², and how many studies sit behind the pooled number?
- Is a GRADE certainty rating reported for the main outcome?
- Does the conclusion stay within what the pooled estimate can support?
For reading the individual papers a review is built from, see how to read a PubMed abstract; for turning a pooled ratio into absolute terms, relative vs absolute risk. Nothing here is medical, dosing, or therapeutic guidance.
Frequently asked questions
Is a meta-analysis always higher-quality evidence than a single trial? No. A large, well-run randomised trial can be more reliable than a meta-analysis of small, biased studies, and occasionally the two disagree. A meta-analysis is only as good as the studies it includes and the decisions made about which to include. Check the inputs and the certainty rating before the pooled number.
Can a systematic review have no meta-analysis? Yes, and it often should. When studies use different outcomes, populations or designs, pooling them produces an average that means little. Such reviews synthesise the findings in words and tables instead, sometimes following the SWiM reporting guideline, as the BPC-157 review by Poolton and colleagues did.
What does PROSPERO registration prove? That the review's question and methods were recorded before the results were known. It does not prove the review is good, but it lets a reader compare the published review against its plan and spot changes to outcomes or analyses made along the way.
Why do meta-analyses on the same question reach different answers? Usually because they included different studies: different search dates, different eligible designs, different outcome definitions. The GLP-1 hair-loss reviews above pooled trial adverse-event data in one case and health-record data in the other. Compare the inclusion criteria before comparing the numbers.
Selected Research References
- Moher D, Liberati A, Tetzlaff J, Altman DG; PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ. 2009;339:b2535. pubmed.ncbi.nlm.nih.gov/19622551
- Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327(7414):557-560. pubmed.ncbi.nlm.nih.gov/12958120
- Guyatt GH, Oxman AD, Vist GE, et al.; GRADE Working Group. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336(7650):924-926. pubmed.ncbi.nlm.nih.gov/18436948
- Wood L, Egger M, Gluud LL, et al. Empirical evidence of bias in treatment effect estimates in controlled trials with different interventions and outcomes: meta-epidemiological study. BMJ. 2008;336(7644):601-605. pubmed.ncbi.nlm.nih.gov/18316340
- Ioannidis JP. The Mass Production of Redundant, Misleading, and Conflicted Systematic Reviews and Meta-analyses. Milbank Quarterly. 2016;94(3):485-514. pubmed.ncbi.nlm.nih.gov/27620683
- Lundh A, Lexchin J, Mintzes B, Schroll JB, Bero L. Industry sponsorship and research outcome. Cochrane Database of Systematic Reviews. 2017;2:MR000033. pubmed.ncbi.nlm.nih.gov/28207928
- Cheng PL, Chang HC. Glucagon-like peptide-1 receptor agonists and hair loss: A systematic review and meta-analysis. Diabetes Research and Clinical Practice. 2026;237:113333. pubmed.ncbi.nlm.nih.gov/42155605
- Viquez Burboa GU, Flores Guillén MÁR, Duardo González VS. GLP-1 Receptor Agonists and Alopecia: A Systematic Review and Meta-Analysis of Incidence, Risk, Subtypes, and Mechanisms. Skin Appendage Disorders. 2026. pubmed.ncbi.nlm.nih.gov/42621629
- Gupta AK, Teasell EM, Economopoulos V, Mirmirani P. GLP-1 therapies and hair loss: A systematic review of current evidence and implications for counseling. Science Progress. 2026;109(2):368504261444578. pubmed.ncbi.nlm.nih.gov/41998799
- Poolton JT, Minett J, Kazmi A. Human evidence for BPC-157 in orthopaedic and regenerative medicine: a systematic review of efficacy and safety. Acta Orthopaedica Belgica. 2026;92(3). pubmed.ncbi.nlm.nih.gov/42885452
All products sold by NovaEvum are for research and laboratory use only. They are not intended for human or animal consumption, diagnosis, treatment, cure, or prevention of any disease.
