GLP-1s and Bone: 10 Studies on a Risk Nobody Has Measured Properly

The best randomized bone data available, and it is more nuanced than the headline suggests. 195 adults carrying obesity, sorted after an eight-week low-calorie diet into four year-long arms — liraglutide 3.0 mg, supervised exercise, the two together, or neither — with density read at hip, spine and forearm. ★ Two results: the combination held bone density unchanged against placebo, and liraglutide alone lost more density than exercise alone at the hip and spine despite comparable weight loss. That second comparison is what complicates the drug-protects-bone story, and it points squarely at training as the countermeasure.

PMID 38916894 ↗DOI 10.1001/jamanetworkopen.2024.16775 ↗

The trial that started the protective hypothesis. 37 women with obesity who had already lost 12% on a low-calorie diet were randomized to liraglutide 1.2 mg daily or no drug for 52 weeks of weight maintenance. Bone mineral content fell significantly in the control group and not significantly on liraglutide — a four-fold difference — while the bone-formation marker P1NP rose 16% on the drug against a 2% fall in controls. ⚠ 37 people, measuring bone content rather than density, in a maintenance rather than active-loss phase. It generated a hypothesis; it did not settle one.

PMID 26043228 ↗DOI 10.1210/jc.2015-1176 ↗

The current randomized benchmark: 53 trials, more than 60,000 participants with type 2 diabetes, showing no significant fracture difference between GLP-1 drugs and their comparators — and no signal in any subgroup by drug, dose or duration. ⚠ The authors state the limits themselves. Follow-up is mostly under two years, and obesity populations without diabetes are essentially absent, which leaves the long-term question for older Wegovy and Zepbound users exactly where it started.

PMID 39603373 ↗DOI 10.1016/j.bone.2024.117338 ↗

The previous generation of the same analysis: 38 trials, roughly 39,000 participants, across six drugs in the class, with a pooled odds ratio of 1.05 and a confidence interval running 0.81 to 1.37 — flatly no effect, with little heterogeneity between trials. ⚠ It shares the structural flaw of every meta-analysis here: fractures were picked up as adverse-event reports rather than counted as a pre-specified outcome, which is a weaker way to find them.

PMID 30974033 ↗DOI 10.1002/dmrr.3168 ↗

The first attempt at the question, pooling 16 liraglutide and exenatide trials. The odds ratio came out at 0.75 — pointing toward fewer fractures — but the confidence interval ran from 0.28 to 2.02, which is another way of saying almost no fractures occurred and the estimate means little. ★ Its actual contribution was setting a default assumption of no harm that the two larger analyses above have since failed to overturn.

PMID 24164867 ↗DOI 10.1111/1753-0407.12102 ↗

The largest real-world comparison, and the one closest to a practical prescribing question. A nationwide South Korean cohort matched roughly 87,000 new SGLT2 inhibitor users against new users of incretin drugs, using an active-comparator new-user design that avoids the classic trap of comparing treated patients against untreated ones. Fracture rates came out similar. Neither class looks bone-safer than the other in type 2 diabetes.

PMID 37751205 ↗DOI 10.1001/jamanetworkopen.2023.35797 ↗

A review that does the useful thing of restricting itself to obesity rather than diabetes populations, pulling together the DEXA substudies from the STEP and SURMOUNT programs alongside the two randomized bone trials above. Its conclusion is the honest one: the density loss that shows up is what weight loss produces, a drug-specific protective effect remains plausible but unproven outside diabetes, and the only mitigation with evidence behind it is resistance training plus enough protein.

PMID 40920189 ↗DOI 10.1007/s00198-025-07664-1 ↗

A critical appraisal covering bone effects across every major weight drug — orlistat, phentermine-topiramate, naltrexone-bupropion, liraglutide, semaglutide and tirzepatide — which makes it the best single place to see how this class compares with the alternatives. For the incretins the verdict is: randomized fracture data reassuring but underpowered, density substudies consistently showing loss proportional to weight, and no long-term population data at all. ⛔ The authors name the gap directly, and it is the right one — millions of healthy adults are now expected to take these drugs for decades.

PMID 40555693 ↗DOI 10.1111/dom.16541 ↗

A disproportionality analysis of the FDA's adverse-event database, covering fracture reports for five drugs in the class through 2024. None showed a fracture signal against the database background. ⛔ Spontaneous reports have no denominator: nobody knows how many people took the drug and did not report anything, so this cannot produce a rate or establish cause. It earns its place as the largest such dataset that exists and because it agrees with the randomized evidence rather than contradicting it.

PMID 39556224 ↗DOI 10.1007/s00592-024-02415-w ↗

A prospective single-center study following 65 adults with type 2 diabetes for a year after starting semaglutide or dulaglutide, with density and trabecular bone score measured — the latter a texture measure that captures bone quality density alone misses. The treated group held or slightly improved both while a comparison group on standard care declined. ⚠ Not randomized, small, and single-center. It is consistent with the 2015 trial and it cannot confirm it.

PMID 38864922 ↗DOI 10.1007/s00223-024-01240-1 ↗