The short answer
Confounding by indication occurs when the condition prompting treatment, rather than the treatment, causes the outcome observed. A 2026 Danish nationwide cohort demonstrates it directly: the same drugs showed an association in one indication and not in another.
Key facts
- Problem
- The reason for treatment causes the outcome
- Affects
- All observational drug studies
- Not fixed by
- Adjusting for measured confounders alone
- Worked example
- Hviid 2026 (PMID 41852577)
- The tell
- Association in one indication, not another
- Randomisation
- The only complete solution
What the problem actually is
People who receive a medicine differ from people who do not, and the most important way they differ is that they had a condition warranting it. If that condition also affects the outcome being studied, an association between drug and outcome appears even when the drug does nothing. The confounder is the indication itself.
Why it is harder to handle than ordinary confounding
Age or sex can be measured and adjusted for. The severity and duration of the condition that prompted prescribing often cannot be captured fully in registry data, and the residual is the part most strongly related to both treatment and outcome. Statistical adjustment reduces the problem without eliminating it.
Research material referenced
Retatrutide 10mg, third-party HPLC tested
The 2026 study that shows it clearly
Hviid and colleagues examined periconceptional GLP-1 receptor agonist exposure and obstetric outcomes in Danish health registries covering October 2009 to December 2023. Preterm birth risk was raised where the drugs were used for diabetes (liraglutide adjusted odds ratio 1.70, 95% confidence interval 1.17 to 2.48; semaglutide 1.84, 1.24 to 2.7), but not where they were used for weight management.
Why that pattern is the informative part
Same drugs, same registry, same outcome, same period. The only thing differing between the two groups is why the drug was prescribed. If the medicine were causing preterm birth it should do so regardless of indication. That it did not is strong evidence the association tracks the underlying diabetes, and the authors state that conclusion themselves.
Why this is a model for reading any observational drug study
Comparing across indications is a natural experiment built into the data. Where a study can show that an association holds in one indication and not another, it has tested its own confounding rather than merely adjusting for it. Most observational studies cannot do this, and it deserves attention when one can.
How coverage typically gets this wrong
By reporting the headline association and omitting the split. A summary saying these drugs are linked to preterm birth would be technically defensible and would invert the study's actual conclusion, which is that the medication is probably not the causal factor. The finding is in the comparison, not the association.
Frequently asked questions
What is confounding by indication?
When the condition that prompted treatment, rather than the treatment itself, causes the outcome observed, making a drug appear responsible when it is not.
Can it be adjusted away?
Not fully. Severity and duration of the underlying condition are often incompletely captured, and that residual is what matters most.
How did the 2026 study demonstrate it?
The same drugs showed raised preterm birth risk in the diabetes indication but not the weight-management one, so the association tracks the indication.
Continue researching
Related guides and reference material.
Primary sources & clinical trials
Published papers, trial registrations, regulator pages and company announcements. Links open the original.
- PubMedHviid KVR et al., Periconceptional GLP-1 receptor agonist exposure and obstetric outcomes: a Danish nationwide cohort study. Hum Reprod Open 2026 (PMID 41852577)pubmed.ncbi.nlm.nih.gov
- PubMedTang H et al., Cardiovascular and kidney outcomes of GLP-1 receptor agonists: a target trial emulation study. Diabetes Obes Metab 2025 (PMID 40874398)pubmed.ncbi.nlm.nih.gov
- PubMedPasternak B et al., GLP-1 receptor agonist use and risk of thyroid cancer: Scandinavian cohort study. BMJ 2024 (PMID 38683947)pubmed.ncbi.nlm.nih.gov
Written and published by UK Peptides, the trading name of AURYXRESEARCH LTD (company 17388727, registered in England and Wales).
Questions or a correction? Email hello@ukpeptides.org or call +44 7426 987848.
More GLP-1 & Incretin Science articles
- Why Age Changes the CalculationThe same weight reduction carries different consequences at 75 than at 45. Lean mass, bone and nutritional intake all sit differently.
- Where a Peptide Stops Being a DrugFDA treats 40 residues or fewer as a peptide - a drug. Longer is a protein, a biologic. That single number governs how generics reach the market.
- When the Intended Effect Goes Too FarA 2026 systematic review describes gastroparesis as an under-recognised complication. It is the drug's own mechanism crossing into pathology.
- The Same Effect, Past the Point of BenefitSome adverse effects are unrelated toxicities. Others are the drug working too well. The two need different responses, and they get conflated.
- What the SUSTAIN Programme Data ShowedVilsbøll and colleagues evaluated retinopathy data across the SUSTAIN trials. The title places the HbA1c reduction between the drug and the risk.