Research Literacy
What Is Publication Bias, and How Would You Spot It?
The short answer
Publication bias is the tendency for studies with positive, significant results to get published — and published faster — more often than studies that find nothing, which skews the visible literature toward exaggerated effects. Researchers test for it using funnel plots and Egger's test, a statistical check for lopsided asymmetry in how study results cluster. In a 2025 meta-analysis of weight regain after stopping anti-obesity drugs (36 studies), Egger's test flagged significant bias for one drug, liraglutide (p=0.013), but not for semaglutide, exenatide, or orlistat. A separate 2026 meta-analysis of lean mass loss during incretin therapy (20 trials, 15,782 people) ran the same test and found no evidence of bias (p=0.23). Both outcomes are normal — the test doesn't always find something, and that's part of what makes it useful.
Quick answers
- What is it?
- Publication bias is the tendency for studies with positive, statistically significant results to get published faster and more often than studies that find nothing, skewing the visible literature toward exaggerated effects.
- Why are people interested in it?
- Readers relying on published research to judge a treatment need to know whether the studies they can see are a fair sample of everything that was actually tried.
- What has actually been studied?
- How researchers detect this using funnel plots and Egger's statistical test, illustrated with a 2025 meta-analysis of weight regain after stopping anti-obesity drugs (36 studies) and a 2026 meta-analysis of lean mass loss during incretin therapy (20 trials, 15,782 people).
- How strong is the evidence?
- A well-illustrated methodological concept: Egger's test found significant asymmetry for liraglutide specifically in one meta-analysis but not for three other drugs in the same paper, and came back clean in a separate lean-mass meta-analysis — both normal outcomes of a working check.
- What don't we know yet?
- Whether Egger's test is reliable when a meta-analysis pools fewer than roughly ten studies, what specific studies are actually missing when asymmetry is found, why liraglutide's evidence base showed asymmetry and the other three drugs' didn't, and how registered-but-unreported trials factor into this.
- Where can I read the original research?
- Sources at the bottom of this page: Kolli RT, Aoutla S, Jyothi N, et al., Cureus, 2025; and Eisa N, Barood O, Diabetes, Obesity and Metabolism, 2026.
The short answer: Publication bias is the tendency for studies with positive, statistically significant findings to get published — and published faster, in more prominent journals — than studies that find no effect. Because the visible literature is a filtered sample of everything that was actually tried, that filtering can make a treatment look more consistently effective than it is. Researchers test for this using funnel plots and Egger's test, a statistical check for asymmetry in how study results cluster around the pooled estimate. In a 2025 meta-analysis of weight regain after stopping anti-obesity drugs, pooling 36 studies, Egger's test found significant asymmetry for liraglutide specifically (p=0.013) but not for semaglutide, exenatide, or orlistat. In a separate 2026 meta-analysis of lean mass loss during incretin therapy, covering 20 randomized trials and 15,782 participants, the same kind of test came back clean (p=0.23). Both results are normal outcomes of a working check — the point of running Egger's test isn't to always find bias, it's to actually look.
What the research shows
Why the published record isn't a random sample
A meta-analysis pools every study its authors could find on a question and combines the results into a single estimate. That estimate is only as trustworthy as the studies feeding into it — and studies don't reach publication at random. A trial that finds a drug works is more attractive to journals, more likely to be written up quickly, and more likely to be accepted on the first submission. A trial that finds nothing is more likely to sit in a file drawer or land, slower, in a smaller journal. None of this requires dishonesty — it's a structural tilt built into incentives that reward positive findings, and it means the studies a reader can actually find are not a neutral sample of every study ever run.
The funnel plot: a visual check for a lopsided literature
The standard tool for detecting this is the funnel plot: a scatter plot with each study's effect size on one axis and a measure of its precision — typically standard error or sample size — on the other. Large, precise studies should cluster tightly near the pooled effect estimate; small, imprecise studies should scatter more widely around it, in both directions, producing a shape that looks roughly like an inverted funnel when no bias is present. If small studies showing an unfavorable or null result are systematically missing from the literature — because they were never published, or took years longer to appear — the funnel comes out lopsided: small studies cluster on the "favorable" side only, with a gap where the unfavorable small studies should be.
Egger's test: putting a number on the asymmetry
A funnel plot is a visual judgment call, which is exactly why researchers pair it with Egger's test — a regression-based statistical test that checks whether the funnel's asymmetry is more pronounced than chance alone would produce. A significant Egger's test result (conventionally p<0.10, sometimes p<0.05) is read as evidence of small-study effects, of which publication bias is one plausible explanation among several. This is a meaningful distinction: Egger's test detects an asymmetric pattern. It does not, by itself, prove that specific unpublished studies exist, or that publication bias specifically (rather than genuine clinical heterogeneity between small and large trials) is the cause.
A real result, and a real non-result, side by side
Two recent meta-analyses in the GLP-1 literature show what this looks like when researchers actually run the check and report what they find, rather than assuming an answer either way. A 2025 meta-analysis in Cureus examined weight regain after stopping four anti-obesity drugs — semaglutide, liraglutide, exenatide, and orlistat — pooling 36 studies. The authors ran Egger's and Begg's tests separately for each drug's pooled estimate. For liraglutide, Egger's test came back significant (p=0.013), flagging funnel plot asymmetry; for the other three drugs, "no major bias was found." The same paper, in other words, produced two different verdicts on the identical statistical question depending on which drug's evidence base it was applied to — exactly the granularity a reader should expect, rather than a single blanket judgment about "GLP-1 research" as a category.
A second example, a 2026 meta-analysis in Diabetes, Obesity and Metabolism, pooled 20 randomized controlled trials — 15,782 participants — comparing lean mass loss during incretin-based drug therapy against lifestyle interventions. Its Egger's test came back non-significant (p=0.23): no detected evidence of publication bias in that particular pooled comparison. A third example, a 2023 network meta-analysis of GLP-1-based therapies in EClinicalMedicine covering 40 trials, likewise reported "risk of bias and publication bias were generally low throughout studies" based on visual funnel plot inspection and Egger's test, while flagging high heterogeneity separately as a distinct issue from bias.
What this means for a reader, practically
The presence of a stated bias assessment is itself informative — a meta-analysis that reports running Egger's test, whatever the result, is telling you the authors checked. A meta-analysis that never mentions funnel plots, Egger's test, or any comparable check is one where that question was either not asked or not reported, and a reader is left to wonder which. When a check was run and came back significant, the honest next step isn't to discard the pooled estimate outright — it's to treat it as a flag: read which specific comparison triggered it, how many studies fed into that estimate, and whether the authors offered any explanation beyond the statistical signal itself.
What we don't know yet
Whether Egger's test is reliable with few studies. The test performs poorly and can give misleading results when a meta-analysis pools fewer than roughly ten studies — a limitation affecting some subgroup analyses even within otherwise large meta-analyses, and one not always flagged when results are reported.
What's actually missing when asymmetry is found. None of the three papers discussed here identify specific unpublished studies. A significant Egger's test result describes a pattern in the visible data; it doesn't reconstruct what isn't there.
Why liraglutide's evidence base showed asymmetry and the other three drugs' didn't. The Cureus meta-analysis reports the finding without explaining the mechanism — genuine publication bias, a smaller and older evidence base, or some other structural difference between the drugs' trial histories.
How registered-but-unreported trials factor into this. Peptide Proof's companion article on reading ClinicalTrials.gov describes trials that reached Completed or Terminated status without ever posting results. Whether any specific unreported trial would have shifted a given funnel plot isn't something a published meta-analysis can determine from the outside.
Sources
- Kolli RT, Aoutla S, Jyothi N, et al. Rebound or Retention: A Meta-Analysis of Weight Regain After the Discontinuation of Glucagon-Like Peptide-1 (GLP-1) Receptor Agonists and Other Anti-obesity Drugs. Cureus, 2025;17(10):e94926. DOI
- Eisa N, Barood O. Lean Mass Changes With Incretin Therapy Versus Lifestyle Intervention: A Systematic Review and Meta-Analysis of Randomised Controlled Trials. Diabetes, Obesity and Metabolism, 2026;28(6):4818-4827. DOI
- Caruso I, Di Gioia L, Di Molfetta S, et al. Glucometabolic Outcomes of GLP-1 Receptor Agonist-Based Therapies in Patients With Type 2 Diabetes: A Systematic Review and Network Meta-Analysis. EClinicalMedicine, 2023;64:102181. DOI
Article records maintained via PubMed. Peptide Proof reports what studies found. It does not provide medical advice, dosing guidance, or treatment recommendations.
Sources
- Rebound or Retention: A Meta-Analysis of Weight Regain After the Discontinuation of Glucagon-Like Peptide-1 (GLP-1) Receptor Agonists and Other Anti-obesity Drugs
- Cureus, 2025 · DOI
- Lean Mass Changes With Incretin Therapy Versus Lifestyle Intervention: A Systematic Review and Meta-Analysis of Randomised Controlled Trials
- Diabetes, Obesity and Metabolism, 2026 · n=15782 · DOI
- Glucometabolic Outcomes of GLP-1 Receptor Agonist-Based Therapies in Patients With Type 2 Diabetes: A Systematic Review and Network Meta-Analysis
- EClinicalMedicine, 2023 · DOI
Get the evidence, not the hype.
New research briefs, plainly written and fully sourced. No dosing advice, ever.