Skip to content
All resources

Reading the evidence

How to Read a PubMed Abstract: What It Says and What It Leaves Out

AR

By

· 11 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 · 11 sections
  1. What is on a PubMed record besides the abstract?
  2. How is a structured abstract organised?
  3. Who or what was actually studied?
  4. How many, for how long, compared with what?
  5. Which result is the primary one?
  6. Who paid, and who wrote it?
  7. Has it been retracted or flagged?
  8. How do MeSH terms and publication-type filters help?
  9. Why does this site check every PMID's first author?
  10. A checklist for any abstract
  11. Frequently asked questions

To read a PubMed abstract, work out five things before you look at the conclusion. Who or what was studied: species, cells or people. How many, for how long, and against what comparison. Which outcome was the prespecified primary one, and whether it was met. Who funded the work. And whether the record carries a retraction, correction or expression of concern. The conclusion is the authors' summary of their own work, and it is the part most likely to overstate it.

A printed journal abstract page under a magnifying glass beside reading glasses
Illustration: an abstract is a summary, and summaries leave things out.

What is on a PubMed record besides the abstract?

An abstract is one field on a record that carries more information than most readers use. Laid out from top to bottom:

FieldWhat it tells youWhat to check
Banner and titleWhat the paper claims to be aboutA "Retracted article" banner; whether the title names the species
Journal, date, DOIWhere and when it was publishedOnline date versus print date; follow the DOI to the full text
Authors and affiliationsWho did the workSponsor employees; one group behind a whole literature
Linked notices"Comment in", "Erratum in", "Retraction in", "Expression of concern in"Always open these
AbstractThe authors' summaryDesign, numbers per group, primary outcome
Registration numberA trial's ClinicalTrials.gov entry, if givenCompare the registered primary outcome with the reported one
Conflict of interest statementFunding and financial tiesPresent on many recent records, below the abstract
Publication types and MeSH termsHow PubMed has classified the paperUseful for filtering, but check them against the abstract
PMID and PMCIDThe record's identifier; a PMCID means a free full text existsUse the PMID to confirm a citation

The linked notices are easy to miss. Hróbjartsson and Gøtzsche's review of placebo trials carries an "Erratum in" link, and the ILLUMINATE trial of torcetrapib has eight linked comments. Neither undermines the paper, but they are where a paper's critics and corrections live.

How is a structured abstract organised?

Most clinical journals now ask for a structured abstract: background or objective, methods, results and conclusions, sometimes with headings for design, setting and participants. For randomised trials, the CONSORT extension for abstracts, published by Hopewell and colleagues, lists what an abstract should contain. That includes the design, the participants and setting, the interventions, the primary outcome, the number randomised and analysed in each group, and the result for each group with an effect size and its precision. It also asks for important harms, the trial registration and the funding source.

The checklist is useful to a reader as a list of what to look for and notice missing. A trial abstract that gives a P value but no group sizes, or describes "improvement" without saying which outcome was primary, has left out what you most need. Laboratory papers usually have unstructured abstracts, and there the methods are scattered through a single paragraph.

Who or what was actually studied?

This is the question titles most often obscure. Chang and colleagues' 2011 paper is titled "The promoting effect of pentadecapeptide BPC 157 on tendon healing involves tendon outgrowth, cell survival, and cell migration." The abstract shows what the experiments were: tendon explants and fibroblasts from rat Achilles tendon, in culture. Outgrowth, survival under oxidative stress, migration and signalling proteins were measured in dishes. Nothing was measured in a healing tendon, in an animal or in a person. That is a legitimate mechanistic study. The title's "tendon healing" is a description of the hypothesis, not of the experiment. The wider literature is in BPC-157 and connective tissue. BPC-157 is not approved by Health Canada, the FDA or the EMA.

Look for the words that fix the system: in vitro, cell line, explant, rat, mice, healthy volunteers, patients with a named condition. If the abstract does not say, assume it is not human. The reasons animal results so often fail to carry over are set out in how to read a peptide study.

How many, for how long, compared with what?

Numbers per group matter more than totals. TAZPOWER's randomised phase had 12 participants in a crossover design. In retatrutide's phase 1b trial, 72 people were enrolled and 29 stopped early. MMPOWER-3 randomised 218, split evenly between elamipretide and placebo. Each of those numbers changes how much weight the result can bear, and each is in the abstract. Duration matters for the same reason: a 12-week trial cannot say anything about year two.

Then the comparison. Placebo, an active drug, no treatment, a historical cohort, or nothing at all? An abstract that reports change from baseline with no comparison group is describing a single arm. Open-label vs placebo-controlled explains why that matters, and clinical trial phases explained covers what each trial size is built to answer.

Which result is the primary one?

A trial declares a primary outcome in advance; everything else is secondary or exploratory. When the primary outcome fails, abstracts often lead with something else. Boutron and colleagues examined 72 randomised trials whose primary outcomes were not statistically significant. They found spin, meaning reporting that steered the reader toward benefit, in the conclusions of 42 abstracts (58.3%). It appeared in the results sections of 27 abstracts (37.5%) and in 13 titles (18.0%).

Compare two abstracts from the same compound. MMPOWER-3's discussion says plainly that elamipretide did not improve the walk test or fatigue in mitochondrial myopathy. TAZPOWER's results section states that neither primary endpoint was met in the randomised part. Its conclusion then reports that elamipretide led to improvement in Barth syndrome symptoms, a finding drawn from the open-label extension. Both abstracts are accurate sentence by sentence. Only one puts the primary result where a skimming reader will see it.

The other habit to build is reading the numbers, not the adjectives. "Significantly" means a P value cleared a threshold, not that the effect was large. Relative vs absolute risk covers how to turn a ratio into something a reader can weigh.

Who paid, and who wrote it?

Many recent PubMed records show the conflict of interest statement below the abstract. In retatrutide's phase 1b trial, nine of the eleven authors are listed as employees and shareholders of Eli Lilly, the sponsor. In TAZPOWER, one author was an employee of Stealth BioTherapeutics, elamipretide's developer. Neither fact makes the papers wrong. Sponsors fund most drug trials, and their data are often the only data.

It is still a reasonable prior. Lundh and colleagues' Cochrane review found that industry-sponsored drug and device studies more often reported favourable efficacy results, with a risk ratio of 1.27. They more often reached favourable conclusions, at 1.34, and their conclusions agreed with their own results less often. The authors judged that this industry bias was not explained by standard risk-of-bias assessments.

Has it been retracted or flagged?

A retracted paper stays on PubMed, which is deliberate: the record of what was published, and withdrawn, is preserved. The signs are a "Retracted article" banner, a "Retraction in" link, and the publication type "Retracted Publication". The notice itself is a separate record with the publication type "Retraction Notice".

A current example from the GLP-1 literature: "Comparative efficacy of semaglutide versus liraglutide on weight loss and glycaemic control", by Ullah and colleagues in Endocrine Connections, was retracted in March 2026. The original record (PMID 41532579) carries the banner and links to the retraction notice (PMID 41932355). The PubMed record does not state a reason, which is common; the notice on the journal's site, reached through the DOI, is where the reason is given when there is one. A second example: a Retina article on semaglutide and optic neuropathy was published online in October 2025 and retracted in 2026, with a notice at PMID 42468029.

An expression of concern is a softer flag. Editors publish one when they have doubts about a paper, often pending an investigation, without yet retracting it. They appear under the publication type "Expression of Concern"; one dated October 2026 concerns a rat myocardial infarction study using a peptide scaffold (PMID 42810894). Treat any paper with such a notice as unreliable until the matter is resolved.

How do MeSH terms and publication-type filters help?

PubMed indexers tag records with Medical Subject Headings (MeSH), a controlled vocabulary, and with publication types. Searching on them is more precise than free text:

  • `randomized controlled trial[pt]`, `systematic review[pt]` and `meta-analysis[pt]` restrict a search by study design.
  • `retracted publication[pt]` finds withdrawn papers on a topic before you cite one.
  • `humans[mh]` and `animals[mh]` separate human and animal work, the single most useful filter for compound literatures.

Two cautions. New records are often not yet indexed, so a filter can silently drop the most recent papers. And indexing reflects how a paper is labelled, not what it is. The 168-week TAZPOWER report is an open-label extension, yet its record carries the publication type "Randomized Controlled Trial". The natural-history comparison study is tagged "Clinical Trial, Phase III" and "Observational Study" at once. Use filters to find papers, then read the abstract to classify them yourself.

Why does this site check every PMID's first author?

Because a PMID is the easiest thing on a page to get wrong while looking authoritative. When this site's resources section was first built from articles scraped from an earlier storefront, 17 of the first 38 citations resolved to the wrong paper. Most were real PMIDs adjacent to the correct one, the signature of a number guessed rather than looked up. Every reference on these pages is now checked automatically: the first author's surname in the reference must match the first author PubMed returns for that PMID.

You can do the same check by hand in a few seconds. Open the PMID and confirm that the title, first author, journal and year match the citation. Then do the part no script can: read the abstract and check that it supports the sentence it is attached to.

A checklist for any abstract

  • Is there a retraction, correction or expression of concern linked?
  • What system was studied: cells, animals, healthy people, patients?
  • How many per group, for how long, and against what comparison?
  • What was the primary outcome, and was it met?
  • Do the conclusions match the results, or lead with a secondary finding?
  • Who funded it, and are authors employed by the sponsor?
  • Is there a trial registration, and does the registered primary outcome match?

For judging reviews that pool many abstracts, see systematic review vs meta-analysis; for markers that stand in for outcomes, surrogate endpoints explained. Nothing here is medical, dosing, or therapeutic guidance.

Frequently asked questions

Is it enough to read the abstract? For deciding whether a paper is worth reading, often yes. For relying on its result, rarely. Abstracts omit randomisation methods, dropouts, secondary analyses, adverse-event tables and limitations. A PMCID on the record means a free full text is available in PubMed Central; otherwise follow the DOI to the publisher.

What does "Indexed for MEDLINE" mean? That the record has been fully processed by indexers and assigned MeSH terms. Newer records, and some journals, are in PubMed without that indexing. Such papers are still real publications, but they can be missed by MeSH-based searches and filters until indexing catches up.

How do I find out why a paper was retracted? Open the linked retraction notice, then follow its DOI to the journal, which usually publishes the reason: error, duplicated images, plagiarism, unverifiable data, or an author request. The PubMed notice itself often has no abstract. Retraction Watch maintains a searchable database of retractions and their stated reasons.

Why might a PubMed title not match the paper's original language? Foreign-language papers are indexed with an English translation of the title in square brackets, and sometimes an English abstract. Several Russian clinical studies of Semax appear this way. The full text remains in the original language, which limits how far a non-reader can check the methods.

Selected Research References

  1. Hopewell S, Clarke M, Moher D, et al.; CONSORT Group. CONSORT for reporting randomised trials in journal and conference abstracts. The Lancet. 2008;371(9609):281-283. pubmed.ncbi.nlm.nih.gov/18221781
  2. Boutron I, Dutton S, Ravaud P, Altman DG. Reporting and interpretation of randomized controlled trials with statistically nonsignificant results for primary outcomes. JAMA. 2010;303(20):2058-2064. pubmed.ncbi.nlm.nih.gov/20501928
  3. 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
  4. Chang CH, Tsai WC, Lin MS, Hsu YH, Pang JH. The promoting effect of pentadecapeptide BPC 157 on tendon healing involves tendon outgrowth, cell survival, and cell migration. Journal of Applied Physiology. 2011;110(3):774-780. pubmed.ncbi.nlm.nih.gov/21030672
  5. Reid Thompson W, Hornby B, Manuel R, et al. A phase 2/3 randomized clinical trial followed by an open-label extension to evaluate the effectiveness of elamipretide in Barth syndrome, a genetic disorder of mitochondrial cardiolipin metabolism. Genetics in Medicine. 2021;23(3):471-478. pubmed.ncbi.nlm.nih.gov/33077895
  6. Karaa A, Bertini E, Carelli V, et al. Efficacy and Safety of Elamipretide in Individuals With Primary Mitochondrial Myopathy: The MMPOWER-3 Randomized Clinical Trial. Neurology. 2023;101(3):e238-e252. pubmed.ncbi.nlm.nih.gov/37268435
  7. Urva S, Coskun T, Loh MT, et al. LY3437943, a novel triple GIP, GLP-1, and glucagon receptor agonist in people with type 2 diabetes: a phase 1b, multicentre, double-blind, placebo-controlled, randomised, multiple-ascending dose trial. The Lancet. 2022;400(10366):1869-1881. pubmed.ncbi.nlm.nih.gov/36354040
  8. Thompson WR, Manuel R, Abbruscato A, et al. Long-term efficacy and safety of elamipretide in patients with Barth syndrome: 168-week open-label extension results of TAZPOWER. Genetics in Medicine. 2024;26(7):101138. pubmed.ncbi.nlm.nih.gov/38602181
  9. Ullah H, Khan ZU, Wazir A, et al. RETRACTION: Comparative efficacy of semaglutide versus liraglutide on weight loss and glycaemic control. Endocrine Connections. 2026;15(3):e250723. pubmed.ncbi.nlm.nih.gov/41932355

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.