Research Literacy
Why Do Two Meta-Analyses on the Same Question Disagree?
The short answer
In 2024, two separate meta-analyses asked whether GLP-1 receptor agonists — the drug class behind semaglutide and liraglutide — change bone mineral density, and reached opposite answers. One, pooling 26 randomized trials and 2,268 participants, found a statistically significant increase in lumbar spine bone density (SMD 1.04, 95% CI 0.60-1.48) and femoral neck bone density (SMD 1.29, 95% CI 0.36-2.23). The other, restricting itself to 7 randomized trials with a true placebo or untreated control arm, found no significant change in bone density at the femoral neck, hip, or spine. Both were peer-reviewed, published the same year, and drew on overlapping trial literature. The difference wasn't fraud or incompetence on either side — it was how each review defined a 'control group,' what statistic it used to express the result, and how long the underlying trials ran.
Quick answers
- What is it?
- This uses two 2024 meta-analyses on GLP-1 drugs and bone density that reached opposite conclusions to show how methodology choices — not fraud or error — can make careful reviews of the same evidence disagree.
- Why are people interested in it?
- Readers who see two conflicting "meta-analysis says" headlines need to understand why that happens and what to check to judge which one to trust more.
- What has actually been studied?
- A 26-trial meta-analysis (2,268 participants) finding increased bone density with GLP-1 drugs, a stricter 7-trial meta-analysis finding no significant change, and a third 33-trial network meta-analysis (33,107 participants) that agreed with the first but flagged that trials under 52 weeks need cautious interpretation.
- How strong is the evidence?
- Both reviews were peer-reviewed and drew on overlapping trial literature, but they differed in what counted as a valid control group, what statistic they used to express results, and how long the underlying trials ran — mechanics the article walks through concretely.
- What don't we know yet?
- Whether GLP-1's apparent bone benefit reflects a true drug effect or comparison against drug classes already known to harm bone, what a review restricted to placebo-controlled trials running 52+ weeks would find, whether bone effects differ by individual drug, and whether the bone density signal actually changes fracture rates.
- Where can I read the original research?
- Sources at the bottom of this page: Li X, Li Y, Lei C, International Journal of Endocrinology, 2024; and Kim HJ, Choi SA, Gu MS, et al., Diabetes/Metabolism Research and Reviews, 2024.
The short answer: Systematic reviews and meta-analyses sit near the top of the evidence hierarchy precisely because they pool many trials instead of relying on one. But pooling doesn't guarantee agreement — two careful teams can build genuinely different reviews from an overlapping body of trials and land on opposite conclusions. A pair of 2024 meta-analyses on GLP-1 drugs and bone mineral density shows exactly how that happens, and exactly what to check when it does.
What the research shows
The case: two reviews, one drug class, opposite verdicts
GLP-1 receptor agonists — the class that includes semaglutide and liraglutide — are now used by millions of people, and researchers have asked whether they help or harm bone health as a side effect of the weight and metabolic changes they cause. In 2024, two teams answered that question using meta-analysis, in the same year, on largely the same underlying trial literature.
One review, published in the International Journal of Endocrinology, pooled 26 randomized controlled trials covering 2,268 participants with type 2 diabetes, searching the literature through May 2023. Compared against "other antidiabetic drugs or placebo," GLP-1 receptor agonists were associated with a significant increase in lumbar spine bone mineral density (standardized mean difference 1.04, 95% CI 0.60–1.48, p<0.00001) and femoral neck bone density (SMD 1.29, 95% CI 0.36–2.23, p=0.007), along with higher serum calcium, higher bone-formation markers, and lower levels of a bone-resorption marker.
The other review, published in Diabetes/Metabolism Research and Reviews, restricted itself to 7 randomized trials that met a stricter bar: an untreated control group or a true placebo group, at least 4 weeks of GLP-1 treatment, and formal risk-of-bias screening with the Cochrane RoB2 tool. Its search ran through June 2024 — more recent than the first review's. It found no significant change in bone mineral density at the femoral neck (mean difference 0.01 g/cm², 95% CI -0.01 to 0.04), total hip (-0.01 g/cm², -0.02 to 0.01), or lumbar spine (0 g/cm², -0.02 to 0.02) — while the same resorption marker that fell in the first review actually rose in this one (mean difference 0.04 μg/L, 95% CI 0.01–0.07).
Both papers passed peer review. Neither retracted or corrected the other. They simply disagree.
Mechanic one: what counted as a control group
The clearest, most checkable difference between the two reviews is what they each accepted as a valid comparison arm. The 26-trial review explicitly included trials where GLP-1 drugs were compared against other antidiabetic medications, not only against placebo. That matters because some other diabetes drug classes — thiazolidinediones, in particular, based on the broader network-analysis literature — are independently linked to worse bone outcomes. If part of a trial pool is comparing a GLP-1 drug against a medication that harms bone, GLP-1 can look "protective" simply by contrast, without doing anything to bone density on its own. The 7-trial review closed that door deliberately, keeping only trials with a placebo or untreated arm — a cleaner test of whether GLP-1 does something to bone, at the cost of a much smaller pool of eligible trials.
Mechanic two: how the result was expressed
The two reviews also reported their results using different statistics. The 26-trial review used standardized mean difference (SMD), a unit-less measure that expresses a result in terms of pooled variability across the included trials — useful for combining studies that measured bone density differently, but a statistic that can produce large-looking numbers (an SMD above 0.8 is conventionally called a "large" effect) when the underlying trials are small or heterogeneous. The 7-trial review reported raw mean difference in grams per square centimeter — a directly interpretable clinical unit, and one where the numbers involved (0.01, -0.01, 0 g/cm²) look close to zero in absolute terms even before considering statistical significance. Two different ways of describing "the same" measurement can make an identical underlying signal look dramatic in one framing and negligible in another.
Mechanic three: how long the trials actually ran
A third, independent 2025 network meta-analysis of 33 randomized trials and 33,107 participants — the largest pooled bone-outcomes analysis on this drug class to date — found the same direction as the first review: bone density increased at the lumbar spine, total hip, and femoral neck, and fracture risk fell by 20% (relative risk 0.80, 95% CI 0.65–0.94). But its authors added an explicit caution: "BMD findings from studies less than 52 weeks require cautious interpretation per clinical densitometry standards." Bone remodels slowly, and a scan taken too soon after treatment starts can reflect measurement noise rather than a real skeletal change. Since GLP-1 trials are typically designed around glycemic control or weight loss, not bone health, many of the trials feeding into either meta-analysis were not built to track bone over the timeline bone actually needs.
The broader pattern
Counted by number of independent reviews, the "bone density improves" conclusion has more company: alongside the 26-trial review, a 2025 analysis of 25 trials in Acta Diabetologica found the same direction of effect using raw mean differences (lumbar spine +0.07 g/cm², 95% CI 0.06–0.09, p<0.00001), and the 33-trial network meta-analysis agrees as well. But counting reviews isn't the same as weighing them — the 7-trial review is the only one that restricted itself to true placebo-controlled comparisons and applied a formal, current risk-of-bias tool to every included trial. That combination of stricter criteria and smaller trial pool is exactly the trade-off that produces disagreement: a review can be the outlier and the most methodologically conservative one at the same time.
What we don't know yet
Whether GLP-1's apparent bone benefit reflects a true drug effect, or reflects comparison against drug classes already known to harm bone. No retrieved review isolates GLP-1-versus-nothing cleanly enough, at scale, to rule this out.
What a review restricted to placebo-controlled trials running 52 or more weeks would find. That specific review does not yet exist in the retrieved literature — the trials needed to build it may simply not exist yet in sufficient numbers.
Whether bone effects differ meaningfully by individual drug — semaglutide, liraglutide, and tirzepatide are chemically distinct and dosed differently, but all four retrieved reviews analyze GLP-1 receptor agonists as a single pooled class.
Whether the bone density signal, in either direction, actually changes fracture rates. The fracture-specific data itself splits the same way: one review found a non-significant trend (relative risk 0.80, 95% CI 0.47–1.36, 25 trials) and another found a significant reduction using the same point estimate (relative risk 0.80, 95% CI 0.65–0.94, 33 trials) — the identical central estimate, reaching significance only once enough participants were pooled to narrow the interval.
Sources
- Li X, Li Y, Lei C. Effects of Glucagon-Like Peptide-1 Receptor Agonists on Bone Metabolism in Type 2 Diabetes Mellitus: A Systematic Review and Meta-Analysis. International Journal of Endocrinology, 2024;2024:1785321. DOI
- Kim HJ, Choi SA, Gu MS, et al. Effects of Glucagon-Like Peptide-1 Receptor Agonist on Bone Mineral Density and Bone Turnover Markers: A Meta-Analysis. Diabetes/Metabolism Research and Reviews, 2024;40(6):e3843. DOI
- Tan Y, Liu S, Tang Q. Effect of GLP-1 receptor agonists on bone mineral density, bone metabolism markers, and fracture risk in type 2 diabetes: a systematic review and meta-analysis. Acta Diabetologica, 2025;62(5):589-606. DOI
- Alalwani YJ, Alshahrani AS, Alsudays AI, et al. Differential effects of GLP-1 receptor agonists on diabetic osteopathy in type 2 diabetes: a patient-stratified network meta-analysis. BMC Musculoskeletal Disorders, 2025;26(1):742. DOI
Article records maintained via PubMed. Peptide Proof reports what studies found. It does not provide medical advice, dosing guidance, or treatment recommendations.
Sources
- Effects of Glucagon-Like Peptide-1 Receptor Agonists on Bone Metabolism in Type 2 Diabetes Mellitus: A Systematic Review and Meta-Analysis
- International Journal of Endocrinology, 2024 · n=2268 · DOI
- Effects of Glucagon-Like Peptide-1 Receptor Agonist on Bone Mineral Density and Bone Turnover Markers: A Meta-Analysis
- Diabetes/Metabolism Research and Reviews, 2024 · n=7 · DOI
- Effect of GLP-1 receptor agonists on bone mineral density, bone metabolism markers, and fracture risk in type 2 diabetes: a systematic review and meta-analysis
- Acta Diabetologica, 2025 · n=25 · DOI
- Differential effects of GLP-1 receptor agonists on diabetic osteopathy in type 2 diabetes: a patient-stratified network meta-analysis
- BMC Musculoskeletal Disorders, 2025 · n=33107 · DOI
Get the evidence, not the hype.
New research briefs, plainly written and fully sourced. No dosing advice, ever.