GLP 1 ankylosing spondylitis signal appears in a records study of 208970 adults with diabetes

A records study published on 20 October in Scientific Reports reports a possible GLP 1 ankylosing spondylitis signal in adults with type 2 diabetes. People who started a GLP 1 receptor agonist had slightly more new diagnoses of the inflammatory spine disease than matched people who started a DPP 4 inhibitor. The absolute difference was tiny, and the authors call it a signal that needs independent confirmation.

How the comparison was built

The team, from Shin Kong Wu Ho Su Memorial Hospital in Taipei, used the TriNetX US Collaborative Network, a large pool of electronic health records from American health systems. They designed the analysis as an emulation of a target trial, which means they wrote down the trial they would have liked to run and then copied its rules with existing data. Adults with type 2 diabetes entered the study when they first started either a GLP 1 receptor agonist or a DPP 4 inhibitor between 2018 and 2022. Comparing two groups of new users of different diabetes drugs, rather than users against non users, is meant to reduce bias, since both groups had a reason to begin a new treatment.

Each GLP 1 user was then paired with a DPP 4 user who looked similar on recorded characteristics, a method called propensity score matching. After matching, the study held 208,970 people, 104,485 in each group. The researchers followed the groups for six outcomes, ankylosing spondylitis, rheumatoid arthritis, other rheumatoid arthritis, polyosteoarthritis, gout and Sjögren's syndrome, using Cox models to compare how often each new diagnosis appeared over time. Because testing six outcomes raises the chance of a false alarm, they applied the Benjamini and Hochberg false discovery rate correction. The paper is an accepted manuscript posted early, so the final edited version may still change in wording or detail.

The ankylosing spondylitis numbers

Newly recorded ankylosing spondylitis appeared in 0.15% of people who started a GLP 1 drug and in 0.10% of those who started a DPP 4 inhibitor. The hazard ratio was 1.49, with a 95% confidence interval from 1.16 to 1.91 and a P value of 0.001. In relative terms that is roughly half again as many new diagnoses. In absolute terms it is about five extra recorded cases for every 10,000 people, a gap small enough that most clinicians would never notice it in practice. The authors report that the direction of the result held across their prespecified sensitivity analyses, which repeat the comparison under different assumptions.

In absolute terms it is about five extra recorded cases for every 10,000 people, a gap small enough that most clinicians would never notice it in practice.

The other five outcomes told a quieter story. Polyosteoarthritis was recorded in 1.80% of GLP 1 starters and 1.73% of DPP 4 starters, a hazard ratio of 1.07 with a P value of 0.031 before correction. After the false discovery rate adjustment the P value rose to 0.093, so the authors treat it as nominal rather than significant. Nominal here means the result crossed the usual threshold only before accounting for the number of outcomes tested, a common source of chance findings in large records studies. Rheumatoid arthritis, other rheumatoid arthritis, gout and Sjögren's syndrome showed no difference between the groups. The table below sets out the main figures as given in the abstract.

Why a GLP 1 ankylosing spondylitis link is hard to read

Several features of the data limit what the result can mean. The outcome is a newly recorded diagnosis code, not a confirmed case reviewed by a rheumatologist. Ankylosing spondylitis often goes undiagnosed for years, and anything that brings people into more contact with doctors, such as the visits and scans that may accompany a new weight loss medicine, could raise the chance that existing back pain finally gets a label. Matching only balances what the records capture. Body weight, smoking, family history and earlier back symptoms may be recorded unevenly, and any imbalance could shift a rare outcome. Records studies of this kind can point to a question but cannot establish that a drug causes a disease.

There is also no clear biological story yet. GLP 1 receptors appear on some immune cells, and laboratory studies have generally linked these drugs with less inflammation rather than more. Our report on gut glucagon like peptides described mouse work in which the hormones protected the intestine from immune attack. The new paper does not test any mechanism, and it offers no laboratory or imaging data that might explain one. Large records studies of these drugs keep producing long lists of associations, some protective and some harmful, as a veterans records analysis of 175 outcomes showed. Each such signal needs a second dataset and ideally trial data before it changes anything.

What would confirm or dismiss the signal

The authors describe the finding as a potential comparative safety signal that warrants independent confirmation, and that framing fits the evidence. A useful next step would be the same analysis in a separate population, such as a national registry outside the United States, with diagnoses checked against imaging or specialist notes. Another would be to look for ankylosing spondylitis among adverse events in the large randomised trials of GLP 1 drugs, where the comparison groups were assigned by chance. If the signal appeared again with a similar size, researchers could then ask whether it differs between individual drugs, doses or durations, which this paper did not report in its summary.

For readers taking these medicines, the paper gives no reason to stop or change treatment on its own. The absolute risk in both groups was well below one in five hundred over the study period, and the comparison was against another diabetes drug, not against no treatment. People with persistent inflammatory back pain, especially pain that improves with movement and worsens with rest, should raise it with a clinician whatever drug they take. The study also had funding from the authors' own hospital, which the paper says had no role in its design. The full text of the Scientific Reports paper lists its methods and sensitivity analyses.

Conclusion

The GLP 1 ankylosing spondylitis finding is a small relative increase in recorded diagnoses among people who started GLP 1 drugs instead of DPP 4 inhibitors, about five per 10,000. It comes from health records, rests on diagnosis codes and was not tested against a mechanism. Until another dataset repeats it, the GLP 1 ankylosing spondylitis link remains a signal to check, not a known side effect.

Source Lu YC, Li TH, Liu YF, Tsai MH. GLP 1 receptor agonist initiation and newly recorded ankylosing spondylitis diagnoses in type 2 diabetes: a target trial emulation. Scientific Reports (2026). DOI 10.1038/s41598-026-75731-6

Amino Weekly reports research. It is not medical advice and never gives doses.