Among nearly a million people with type 2 diabetes, the ones started on an SGLT2 inhibitor were diagnosed with osteoarthritis slightly less often over the next five years than the ones started on a GLP-1 drug. The gap was small, about 8 percent fewer diagnoses (hazard ratio 0.921), but it ran opposite to the joint-protection story that has been building around GLP-1 medicines.
Osteoarthritis is the wear-and-tear joint disease, the slow cartilage loss that drives most knee and hip replacements. Both drug classes in this study are taken for diabetes and both cause weight loss, which takes mechanical load off the joints. GLP-1 receptor agonists ↗, the class that includes semaglutide ↗ (sold as Ozempic and Wegovy) and tirzepatide ↗ (Mounjaro and Zepbound), also blunt inflammation. So do SGLT2 inhibitors, the sugar-dumping pills like empagliflozin and dapagliflozin. On paper both should be easy on the skeleton. The study asked which is easier.
What the study compared
Orthopaedic surgeons at Chi Mei Medical Centre in Tainan, Taiwan, ran the comparison through TriNetX, a network that pools deidentified electronic health records from many hospitals. Writing in PLoS One ↗ on July 20, they selected adults with type 2 diabetes who were newly prescribed either an SGLT2 inhibitor or a GLP-1 receptor agonist between 2017 and 2019, then used propensity matching, a method that pairs people with similar age, sex, weight, and lab values so the two groups look alike on paper. That left 452,445 patients in each arm, followed for five years.
The SGLT2 group came out marginally ahead. Overall osteoarthritis was about 8 percent less common (hazard ratio 0.921). The clearest signal was in the knee, where osteoarthritis ran roughly 17 percent lower (hazard ratio 0.831). SGLT2 users also received fewer major joint injections, the cortisone and lubricant shots that hold surgery off (hazard ratio 0.915).
The catch the ratios hide
Read where the effect lands. The SGLT2 edge showed up in osteoarthritis diagnoses and in joint injections, the softer endpoints. It did not show up where it would matter most. Rates of total knee replacement and total hip replacement, the operations that mark joint disease bad enough to cut for, were no different between the two groups. The advantage lives in the diagnosis codes and the cortisone shots, not in the operating room.
That distinction is the whole story. A diagnosis of osteoarthritis is partly a function of how often a patient sees a doctor and gets imaged, which differs between drug classes for reasons that have nothing to do with cartilage. A joint replacement is a hard event that a coding artifact cannot manufacture. When the soft endpoints move and the hard one does not, the honest read is that the difference may be as much about detection as about disease.
A comparator problem
The finding also complicates a picture this section covered in June, when a separate TriNetX analysis found GLP-1 users had fewer fractures and less osteoarthritis than people on other weight-loss drugs ↗. Put the two side by side and they are not contradictory, they are relative. GLP-1 drugs look joint-friendly next to older obesity medicines and slightly less so next to SGLT2 inhibitors. The comparator sets the verdict, and neither study randomized anyone.
Both share the same ceiling. This is a retrospective look at care that already happened, not a trial that assigned patients to a drug. Which class a doctor reaches for is not random, and the paper offers no mechanism for why a sugar-dumping pill would spare a knee that a gut-hormone drug does not. It reports an association and says, in effect, worth a closer look.
For the GLP-1 receptor ↗ drugs peptidemodel tracks, the read is narrow. The class keeps picking up incidental benefits in large diabetes cohorts, on the heart, the kidney, and the skeleton. This is the first big head-to-head where a non-peptide class edged it on a joint outcome, and even then only on the diagnoses, not the replacements.