GLP-1 receptor agonist use linked to more newly recorded ankylosing spondylitis diagnoses in diabetes cohorts
GLP-1 receptor agonists continue to carry a broad reputation for cardiometabolic benefit, but a large pharmacoepidemiology analysis has identified a comparative safety signal that deserves closer scrutiny. In adults with type 2 diabetes, initiation of a GLP-1 receptor agonist was associated with a higher rate of newly recorded ankylosing spondylitis diagnoses than initiation of a dipeptidyl peptidase-4 inhibitor. The absolute event rates were low, yet the relative difference was statistically persuasive and persisted across prespecified sensitivity analyses.
The finding does not establish causality. It does, however, place a rheumatologic diagnosis into the safety conversation around a drug class more commonly discussed for weight loss, glycemic control, and cardiovascular risk reduction.
A target trial emulation in a large U.S. network
Investigators used the TriNetX U.S. Collaborative Network to emulate a target trial among adults with type 2 diabetes who started either a GLP-1 receptor agonist or a dipeptidyl peptidase-4 inhibitor between 2018 and 2022. The design was new-user and active-comparator, a structure meant to reduce confounding by indication and other biases that often weaken observational comparisons. One-to-one propensity score matching produced a very large matched cohort: 208,970 patients in total, with 104,485 in each treatment arm.
A Cox model framework was used for the comparative analysis, and the Benjamini–Hochberg false discovery rate procedure covered six outcomes. After matching, ankylosing spondylitis emerged as the clearest divergence between the groups.
The signal centered on ankylosing spondylitis, not the broader arthritis outcomes
Newly recorded ankylosing spondylitis diagnoses occurred in 0.15% of patients who initiated a GLP-1 receptor agonist and in 0.10% of those who initiated a dipeptidyl peptidase-4 inhibitor. The hazard ratio was 1.49, with a 95% confidence interval of 1.16 to 1.91, and the P value was 0.001. In practical terms, the difference was small in absolute terms but notable in relative terms, especially because ankylosing spondylitis is an inflammatory disease that can present subtly before it is formally coded.
One sentence captures the clinical tension: a diabetes drug class associated with metabolic benefit was linked to a higher recorded rate of a chronic inflammatory rheumatologic diagnosis.
Polyosteoarthritis showed a nominal association, with rates of 1.80% versus 1.73% and a hazard ratio of 1.07. That association did not survive false discovery rate adjustment, with an adjusted P value of 0.093. Rheumatoid arthritis, other rheumatoid arthritis, gout, and Sjögren’s syndrome did not differ between treatment groups. The pattern argues against a broad arthritis effect and instead points to a more selective signal, or possibly a diagnosis-specific phenomenon related to clinical detection, coding practice, or immune biology.
Why a biologic mechanism is plausible but not proven
The observed association invites mechanistic speculation, but speculation must stay tethered to what is known. GLP-1 receptor agonists influence more than glucose handling; they affect weight, insulin sensitivity, gut physiology, and inflammatory signaling pathways. Those biological actions have made the class attractive in cardiometabolic disease, yet immune-modulating effects can cut in more than one direction depending on tissue context and disease susceptibility.
Ankylosing spondylitis itself arises from a complex interplay of immune activation, enthesitis, and genetic predisposition, with inflammation concentrated in the axial skeleton and adjacent structures. If a therapy alters systemic inflammatory tone, body composition, or gut-mediated immune signaling, it could theoretically change the timing at which early axial symptoms reach medical attention. Another possibility is less direct: patients who begin GLP-1 receptor agonists often have more intensive clinical follow-up, and frequent contact with the health system can increase the likelihood that musculoskeletal symptoms are evaluated and coded.
None of those explanations can be confirmed from the available data. The study is strongest as a comparative signal-detection exercise, not as proof of mechanism.
Sensitivity analyses held the same direction
Prespecified sensitivity analyses were directionally consistent with the main finding. That consistency matters because it reduces the chance that the association appeared only under one analytic setup or one narrow slice of the cohort. Even so, consistency across analyses does not remove the possibility of residual confounding, diagnostic misclassification, or differences in healthcare utilization between treatment groups.
The study design is still a strength. New-user, active-comparator analyses are better suited to real-world pharmacoepidemiology than simple exposed-versus-unexposed comparisons, particularly when therapies are prescribed to patients with different cardiometabolic profiles. Matching at this scale also helps balance measured covariates. Yet the outcome here is a newly recorded diagnosis, not adjudicated disease onset, and that distinction is central when interpreting a rare inflammatory condition in an administrative dataset.
What clinicians may take from the signal
For prescribers, the immediate consequence is not a change in standard GLP-1 receptor agonist use. The absolute risk remained low, and the analysis did not show a broad excess of other rheumatologic outcomes. But new inflammatory back pain, morning stiffness, or persistent axial symptoms in a patient starting a GLP-1 receptor agonist should not be dismissed simply because the drug class is usually framed around metabolic benefit.
The signal also matters for rheumatology and primary care documentation. Ankylosing spondylitis can be missed early, especially when symptoms are attributed to mechanical back pain, obesity, or diabetic musculoskeletal disease. A modest increase in recorded diagnoses may reflect earlier recognition rather than true excess incidence. Either way, the finding warrants careful validation in independent databases and, ideally, in studies that can distinguish genuine disease emergence from differential ascertainment.
A safety question that now needs independent confirmation
These data identify a potential comparative safety signal in a large real-world cohort, but they do not close the question. The central uncertainty is whether GLP-1 receptor agonist initiation is somehow linked to a true increase in ankylosing spondylitis risk, or whether the association is shaped by detection bias, coding behavior, or unmeasured clinical differences between treatment groups.
Until independent confirmation arrives, the result should be treated as hypothesis-generating and clinically alerting rather than practice changing. The study does, however, remind clinicians that drug safety surveillance does not end with cardiovascular endpoints or glycemic metrics. Inflammation-focused outcomes can surface in unexpected places, and rare events may become visible only when very large cohorts are examined with careful comparator design.