GLP-1 Receptor Agonists and Musculoskeletal Outcomes: A Systematic Literature Review and Meta-Analysis

Abstract

Introduction

Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) are increasingly used for the treatment of type 2 diabetes and obesity, but their effects on musculoskeletal health remain completely misunderstood.

Objective

This systematic review/meta-analysis aims to synthesise clinical data on the effects of GLP-1 RAs on key relevant bone, muscle, and joint outcomes.

Methods

MEDLINE, Cochrane Central Register of Controlled Trials (CENTRAL) (both via Ovid^® platform) and Embase were searched from inception to March 2025 to identify relevant randomised controlled trials (RCTs) or real-world evidence (RWE) studies to be included. This bibliographic search was completed manually. A random-effect model meta-analysis was performed for any outcome reported in at least 2 studies. Subgroup analyses were performed on the type of GLP-1 RAs, type of comparator used and study design. Sensitivity analyses (i.e., leave-out sensitivity analyses and analyses restricted to the most adjusted effect estimate) were performed to test the robustness of the data. The strength of evidence was assessed using GRADE. This work has been performed in adherence with PRISMA statement. (PROSPERO Record ID: CRD420251024082).

Results

From 1148 potentially relevant references, 60 articles (46 RCTs, 13 RWE studies and 1 pharmacovigilance study, comprising 1,250,717 individuals) met our inclusion criteria. Different GLP-1 RAs were represented across the panel of studies, i.e., semaglutide, liraglutide, exenatide, dulaglutide, tirzepatide (dual agonist gastric inhibitory polypeptide [GIP]/GLP-1) and others. No effect on bone outcomes (i.e., bone mineral density [all sites] and fractures [all sites]) were observed when the meta-analytical models included the most adjusted effect size. Regarding muscle outcomes, a significant decrease of lean body mass/fat-free mass was consistently observed with GLP-1 RAs in the global model (k = 28, standardised mean difference [SMD] 0.52, 95% confidence interval [CI] −0.8; −0.23, I² 88%, p-value for heterogeneity <0.0001), which remained robust in all sensitivity analyses. Subgroup analyses showed that the effect was mainly driven by liraglutide and semaglutide, with a decrease in lean body mass/fat-free mass observed when GLP-1 RAs were compared with placebo. No publication bias was found. Regarding joint outcome, models revealed no significant change in The Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain, physical function and stiffness.

Conclusions

This meta-analysis is the first to investigate the effects of GLP-1 RAs on a large panel of musculoskeletal health outcomes. While no significant effects were observed on bone- or joint-related outcomes, GLP-1 RAs were associated with reductions in lean body mass/fat-free mass, although the certainty of evidence was low and these changes appeared largely related to weight loss. Whether these changes translate into clinically meaningful impairments in muscle function or physical performance remains uncertain. Further studies in this field, including those looking at muscle function, strength or performance and using multivariate models considering confounding are needed to better reinforce the models and final findings.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s40265-026-02365-3.

Key Points

GLP-1 receptor agonists (GLP-1 RAs) are increasingly used for the treatment of type 2 diabetes and obesity, but their effects on musculoskeletal health remain completely misunderstood.

In this systematic review and meta-analysis of 60 studies including more than 1.2 million individuals, GLP-1 RAs showed no adverse effects on bone- or joint-related outcomes but were associated with reductions in lean body mass/fat-free mass, apparently in line with body weight reduction.

The clinical significance of the observed reduction in lean mass remains uncertain, highlighting the need for future studies evaluating muscle strength, physical performance, and other clinically relevant musculoskeletal outcomes.

Introduction

Glucagon-like peptide-1 (GLP-1) receptor agonists (GLP-1 RAs) are a class of medications primarily used to treat type 2 diabetes mellitus (T2DM) [1], that have also been harnessed for obesity treatment [2, 3] due to the effect of GLP-1 to induce satiety and slow gastric emptying [4].

Beyond their established anti-diabetic and anti-obesity effects, recent pre-clinical [5–9] and robust evidence from clinical trials and synthesis of evidence suggest that these agents may also improve relevant outcomes in body composition and musculoskeletal health, making them a promising option for addressing unmet needs in conditions such as sarcopenia, osteoporosis, and osteoarthritis [10]. However, the mechanisms underlying these potential musculoskeletal effects remain incompletely understood and, in some cases, controversial. In particular, uncertainty persists regarding the expression and physiological relevance of GLP-1 receptors in several musculoskeletal tissues, including bone and skeletal muscle.

Preclinical studies have suggested that GLP-1 signalling may influence bone metabolism through multiple direct or indirect pathways, including modulation of osteoblast activity, osteoclastogenesis, inflammatory pathways, and metabolic homeostasis. In 2019, Zhang et al. [11] found that GLP-1 RAs could support bone metabolism in ApoE(−/−) mice by promoting osteoblast proliferation and differentiation while reducing the accumulation of advanced glycation end products (AGEs). Clinical data remain inconclusive for the effect of GLP-1 RAs on bone mineral density, but some recent observational studies and meta-analyses have suggested a potential reduction in fracture risk with GLP-1 RAs use [12–14]. Regarding muscle health, experimental studies have reported favourable effects of GLP-1 RAs on pathways involved in muscle metabolism, mitochondrial function, and oxidative stress, although whether these effects result from direct skeletal muscle signalling remain to be debated. Furthermore, initial clinical observations suggest that GLP-1 RAs treatment in older individuals may contribute to improved muscle mass and reduced fat accumulation, thereby supporting its potential role in sarcopenia prevention [4, 15]. However, recent evidence suggests that GLP-1 RA-induced weight loss may also be accompanied by reductions in lean body mass, the functional significance of which remains uncertain. Finally, evidence reports that the well-known weight-reduction effect of the GLP-1 RAs may reduce symptoms of osteoarthritis of the knee, including pain [16].

Taken together, the data remain controversial with strong mechanistic evidence from preclinical studies contrasting with heterogeneous findings in humans. It may be crucial to provide updated and robust evidence about the safety and potential beneficial effects of GLP-1 RAs on musculoskeletal health, both to improve understanding of potential risks and because effective pharmacological options remain relatively limited in some musculoskeletal diseases, such as osteoarthritis [17], and are still unavailable for others, such as sarcopenia [18]. Clarifying their effects on these key relevant outcomes may open a promising way to ensure safety, advancing drug development, and potentially revisiting existing drug indications, thereby supporting progress in expanding therapeutic options beyond the current state of the art.

To address these gaps, the WHO Collaborating Centre for Epidemiology of Musculoskeletal Health and Aging, Liège, Belgium, was requested by the World Health Organization (Doctor Amuthavalli Thiyagarajan), as part of its Term of Reference, to conduct a systematic review and meta-analysis to synthesise clinical data on the effects of GLP-1 RAs on key relevant bone, muscle, and joint outcomes.

Methods

The proposed systematic review was conducted and reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) [19]. The completed PRISMA checklist is available in Online Resource 1. A protocol has been developed and published in the PROSPERO registry (RecordID: CRD420251024082).

A project has been created on Open Science Framework (OSF) (https:// https://osf.io/zbhwc/), a platform for sharing scientific research. All materials and resources associated with this work are accessible through this open-access repository.

Research Question

The research question can be summarised using the PICOS format: P (Population): adults of any age or sex with obesity and/or type 2 diabetes mellitus; I (Intervention); any GLP-1 RAs (tirzepatide accepted as a dual agonist gastric inhibitory polypeptide [GIP]/GLP-1), at any dose and for any treatment duration; C (Comparator): placebo, no treatment, or active comparators (other antidiabetic therapies); O (Outcome): musculoskeletal outcomes, including but not limited to bone mineral density, fracture incidence, muscle mass, muscle function, and joint-related outcomes; S (Study design): randomised controlled trials (RCTs) and real world evidence (RWE) observational studies.

Literature Search

MEDLINE, Cochrane Central Register of Controlled Trials (CENTRAL) (both via Ovid platform) and Embase were searched from inception to March 2025 to identify potentially relevant studies for inclusion in this systematic review and meta-analysis. The search strategies used for each of the bibliographic database are available in Online Resource 2.

Additionally, a manual search for the bibliographies of the relevant manuscripts that had been identified was performed to complete the initial bibliographic database search. Forward reference searching of the included studies was also conducted using Web of Science to identify other research that had cited any article of interest. Experts in the field were consulted to provide any missing references.

The search results from both electronic sources and manual searches were imported into Covidence^® software for data management. Covidence^® is a web-based collaboration software platform that streamlines the production of systematic and other literature reviews (https://support.covidence.org/help/how-can-i-cite-covidence).

Study Selection

All identified articles were screened for their eligibility by two independent reviewers, in pairs, (Y.M., C.B. and D.S.R.), first based on their titles and abstracts and second, based on their full texts. Any conflicts were resolved by a third reviewer (C.B.). Articles that met the following criteria were included in this systematic review and meta-analysis: (i) original published studies, interventional (RCTs) or observational longitudinal studies (RWE studies); (ii) using placebo, no treatment, or any other active treatment as comparator; (iii) including data from adult humans of any gender and any age; (iv) using GLP-1 RAs as intervention, at any dose and any length of treatment; (v) reporting the effect of treatment on musculoskeletal parameters including: bone mineral density (BMD), fracture incidence, joint pain in two sites (i.e., knee and hip) assessed using validated questionnaires (e.g., The Western Ontario and McMaster Universities Osteoarthritis Index [WOMAC]), joint function assessed using validated questionnaires (e.g., WOMAC), handgrip muscle strength, skeletal muscle mass (i.e., muscle mass, lean mass, fat-free mass, appendicular lean mass, skeletal muscle index) assessed using either Dual-Energy X-ray absorptiometry (DXA), bioelectrical impedance analysis (BIA), CT-scan, or magnetic resonance imaging (MRI), or muscle performance (using validated instruments such as the Short Physical Performance Battery [SPPB] test, gait speed, chair stand test, etc.); (vi) published in English [20, 21]. No limit in terms of date of publication was applied. Studies performed on inflammatory rheumatic diseases were excluded. We excluded studies that were restricted to populations with specific comorbid conditions unrelated to the conventional indications for GLP-1 RAs (i.e., type 2 diabetes and obesity), to ensure generalisability to the target clinical population. Conference abstracts were not considered eligible.

Data Extraction

Data were extracted by one reviewer (Y.M.) according to a standardised data extraction form that had been pretested on a sample of four studies. A second reviewer checked data extraction (C.B. and D.S.R.). The following information was collected: study characteristics (author, year of publication, journal, DOI), population characteristics (description, mean age, gender ratio, comorbidities), treatment characteristics (groups, type of treatment, dose, length of follow-up), outcomes of interest, conflict of interest and funding. Any disagreements were resolved through consensus between reviewers. Authors of individual papers were contacted in case of missing information.

Risk of Bias and Quality Appraisal

The Cochrane Risk of Bias 2.0 tool was used to evaluate the risk of bias of RCTs [22], and the Newcastle–Ottawa Scale (https://ohri.ca/en/who-we-are/core-facilities-and-platforms/ottawa-methods-centre/newcastle-ottawa-scale) was applied to assess the quality of observational studies. Quality assessment was performed in duplicate (Y.M. and S.VH.).

Strategy for Efficacy Data Synthesis

When at least three comparable studies were available for pooling in a meta-analysis (i.e., three studies reporting a similar outcome), random-effects models were applied. For binary outcomes (e.g., fracture incidence), relative risks (RRs) with 95% confidence intervals (CIs) were calculated. For continuous outcomes (e.g., percentage change in BMD, joint pain, muscle mass, etc.), either mean differences (MDs) or standardised mean differences (SMDs), with their corresponding 95% CIs, were estimated. Separate meta-analyses were performed for each outcome type. Separate meta-analyses were performed combining RCTs and RWE together but also separately for each study design.

For binary outcomes, when only odds ratios (ORs) were reported, RRs were recalculated using crude values. When adjusted RRs were reported, the most adjusted estimate available was used. Intent-to-treat data were used whenever available. For both binary and continuous outcomes, if studies reported multiple follow-up periods, the longest available follow-up was considered.

Results were examined for heterogeneity using Cochran’s Q statistic and the I² statistic. Heterogeneity was further explored by subgroup analyses (e.g., type of GLP-1 RAs, type of comparator), whenever sufficient data were available. Publication bias was assessed using funnel plots and Egger’s regression asymmetry test when possible (≥10 studies included in the model). In the presence of significant publication bias, the Trim-and-Fill method was applied to evaluate the impact of potentially missing studies on the pooled effect size.

As some studies sometimes included multi-arms, we performed sensitivity analyses to include all possible scenarios in the forest plot without duplicating any arm (e.g., avoiding duplicating the same control group for, e.g., two intervention groups). A multitude of scenarios were therefore run to assess the model's robustness. Sensitivity analyses were also performed restricting the analyses to studies including the following GLP-1 RAs: liraglutide, semaglutide and tirzepatide (dual agonist GIP/GLP-1). Additional one-way sensitivity analyses were also performed to evaluate the robustness of the findings by removing one study at a time (i.e., leave-one-out (LOO) analysis).

When data were not available in the right format or were incomplete, authors of individual studies were contacted to obtain missing values. If the missing data could not be obtained from the authors, different strategies to obtain the missing information were used: (1) application of the methods described in Section 7.7.3 of the Cochrane Handbook for Systematic Review to obtain missing standard deviations (SDs) from standard errors (SEs), from p-values or 95% CIs; (2) when no information was provided to obtain the missing SDs, they were extracted from figures or from another study with a similar sample size, (3) when only median and interquartile ranges were available, the formula proposed by Hozo et al [23] to convert them into mean and SDs was used.

For all analyses, a 2-sided p-value of 0.05 or less was considered statistically significant. Analyses were performed using R software and appropriate statistical packages. When a meta-analysis could not be performed, a narrative synthesis of the results was provided.

Strength of Evidence

The GRADE (Grading of Recommendations, Assessments, Development, and Evaluation) assessment was used to evaluate the level of evidence of the investigated outcome [24]. Starting with a high level of evidence when evidence emerged mainly from RCTs and starting with a moderate level of evidence when evidence emerged mainly from observational studies, the association was downgraded if the meta-analyses met one of the following criteria: (1) a high risk of bias in >75% of the included studies; (2) significant heterogeneity (I > 50%) that could not be explained, proving inconsistency; (3) factors limiting the generalisability of the results, thus indicating indirectness; (4) imprecise estimates with wide 95% CI, leading to potential alterations in the recommendations if the actual effect lies within the 95% CI; (5) significant publication bias.

Results

Studies Characteristics

A total of 1431 references were identified through the search strategies applied on bibliographic databases in March 2025. After removing duplicates, 1131 references were assessed for eligibility based on their title/abstract. Among those references, 98 were assessed based on their full text and 43 studies met the inclusion criteria. The list of excluded studies in the full-text review as well as reason for exclusion is available on an Open Science Framework deposit (https://osf.io/zbhwc/). Manual search identified 17 additional references, making a total of 60 studies that were further included in this systematic review and meta-analysis (Fig. 1).

Fig. 1.

Preferred reporting items for systematic reviews and meta-analyses (PRISMA 2020) flowchart of study selection. GLP-1 glucagon-like peptide-1

Among the 60 studies included, 46 were RCTs, 13 were RWE studies and 1 was a pharmacovigilance study. Main study characteristics are summarised in Table 1 for RCTs Table 2 for RWE and pharmacovigilance studies.

Table 1.

Randomised controlled trial (RCT) characteristics (n = 46, newest first)

First author, year	Population description (health condition, mean age, sex)	Arms description (treatment description + dose in each group)	Sample size per groupa	Treatment duration (weeks)	Outcomes
Bliddal, 2024 [16]	Type of population: Male or female adults with obesity (BMI ≥30.0 kg/m2) and clinical diagnosis of knee OA (American College of Rheumatology (ACR) criteria) Age: 56 (10) y Sex: 81.6% women	G1: Semaglutide 2.4 mg G2: Placebo	G1: 246 G2: 122	68	WOMAC Pain score WOMAC Stiffness score WOMAC Total score SF-36 physical-function score
Golubic, 2024 [25]	Type of population: Overweight and obese adults with T2D, aged ≥ 30 and ≤ 75 y, BMI >28 and ≤ 40 kg/m2, HbA1c) ≤8.0%, with glucose control managed with metformin, with or without a dipeptidyl peptidase IV (DPPIV) inhibitor, sodium–glucose co-transporter-2 inhibitor (SGLT2i), sulfonylurea, or glitinide, where no significant dose change (increase or decrease >50%) has occurred in the 3 months before screening Age: G1: 59.5 (8.4) and G2: 62.2 (7.2) y Sex: 10.7% women	G1: Cotadutide, a fixed-dose escalation of 100 μg daily for the first 4 days, 200 μg daily for the following 4 days and 300 μg daily for the remaining 34 days G2: Placebo	G1: 12 G2: 7	6	Lean body mass (kg)
Hansen, 2024 [26]	Type of population: Post-menopausal men and women (menopause > 5 y) aged 40–85 Age: 63.1 (5.5) y Sex: 86% women	G1: Semaglutide 1.0 mg. Dosages were escalated from an initial dose of 0.25 mg to 0.5 mg after 4 weeks, and from 0.5 mg to 1.0 mg after another four weeks G2: Placebo	G1: 32 G2: 32	52	Lumbar spine BMD (g/cm2) Total hip BMD (g/cm2) Femoral neck BMD (g/cm2)
Kushner, 2024 [27]	Type of population: Patients with overweight or obesity Age: 61.6 (8.9) y Sex: 28% women	G1: Semaglutide is injected once weekly (on the same day each week) into the abdomen, thigh, or upper arm. The treatment starts at 0.24 mg and increases every 4 weeks up to 2.4 mg G2: Placebo	G1: 8803 G2: 8801	104	Hip fracture Lumbar vertebral fracture
Kondo, 2024 [28]	Type of population: Patients with T2DM, age 20–85 y, metformin with or without basal insulin treatment, stable diabetes treatment over 8 weeks, and baseline HbA1c ≤9.5% Age: G1: 65.8 (9.3); G2: 65.5 (10.4) y Sex: 48% women	G1: Dulaglutide 0.75 mg/wk G2: Trelagliptin 100 mg/wk	G1: 25 G2: 25	24	Skeletal muscle mass (kg)
Akyay, 2023 [29]	Type of population: Postmenopausal women with T2DM and HbA1c of 7% to 10% despite using metformin (BMI >30 kg/m2) Age: 53 y Sex: 100% women	G1: Exenatide (started at 5 mcg 2 × 1/day, and after 1 month, revised to 10 mcg 2 × 1/day) G2: Insulin glargine (started at a dose of 0.2 IU/kg at 10.00 PM, and titration was made according to the average 3-day morning fasting blood glucose value)	G1: 15 G2: 15	28	BMD (g/cm2): • Lumbar L1–L4 • Femur neck • Femur total
Corbin, 2023 [30]	Type of population: Healthy volunteers aged 18–50 y  with overweight and obesity, stable weight and BMI 28–40 kg/m2. Obesity associated with mild concomitant diseases were allowed (e.g., mild hypertension) as long as not considered clinically relevant Age: 36.0 (7.7) y Sex: 32.1% women	G1: SAR425899, a dual GLP-1R/GCGR. The SAR425899 doses were administered sc: 0.06 mg from Days 1 to 4, 0.12 mg from Days 5 to 8, 0.16 mg from Days 9 to 12, and 0.2 mg from Days 13 to 19 G2: Placebo	G1: 11 G2: 17	19 days	Fat free mass (kg)
Heise, 2023 [31]	Type of population: T2DM on metformin: patients treated with diet and exercise and stable dose(s) of metformin, with or without 1 additional stable dose of oral antihyperglycaemia medication other than metformin, with HbA1c) ≥7% and ≤ 9.5% if on metformin only; or ≥6.5% and ≤9.0% if on metformin in combination with oral antihyperglycaemia medications other than metformin, and BMI ≥25 and ≤45 kg/m2 with stable weight Age: G1: 63.7 (5.9); G2: 61.1 (7.1); G3: 60.4 (7.6) y Sex: 27%  women	G1: Semaglutide 1 mg G2: Placebo G3: Tirzepatide 15 mg	G1: 43 G2: 24 G3: 41	28	Fat free mass (kg)
Mok, 2023 [32]	Type of population: Patients with poor weight loss (<20% WL) and a suboptimal nutrient-stimulated GLP-1 response at least 12 months following primary sleeve gastrectomy (SG) or Roux-en-Y gastric bypass Age: 47.6 (10.7) y Sex: 74% women	G1: Liraglutide 3.0 mg + Lifestyle intervention. Week 1: 0.6 mg once daily; Week 2: 1.2 mg od; Week 3: 1.8 mg od; Week 4: 2.4 mg od; Week 5–24: 3.0 mg od G2: Placebo + Lifestyle intervention	G1: 23 G2: 24	24	Lean body mass (kg)
Silver, 2023 [33]	Type of population: Obese (BMI ≥30 kg/m2) and prediabetic adults, aged 18 to 65 y, with a HbA1c 5.7–6.4% Age: 50.3 (10.8) y Sex: 68% women	G1: Liraglutide 1.8 mg/d and an oral placebo. Liraglutide dosing started with 0.6 mg/d for study Week 1, escalated to 1.2 mg/d for study Week 2, and was then maintained at 1.8 mg/d for 14 weeks (study Weeks 3–16); G2: Sitagliptin 100 mg/d and a Flexpen sc placebo. The sitagliptin group received an oral dose of 100 mg/d; G3: Caloric restriction alone (−390 kcal/d)	G1: 44 G2: 22 G3: 22	14	Total lean mass (g)
Bartholdy, 2022 [34]	Type of population: Overweight or obese adults diagnosed with knee osteoarthritis (OA). Clinical diagnosis of knee OA (American College of Rheumatology [ACR] criteria) with early to moderate radiographic changes (Kellgren-Lawrence (KL) grades 1, 2, or 3), Age ≥18 y and <75 y, BMI ≥ 27 kg/m2, Stable body weight during the previous 3 months (<5 kg self-reported weight change), motivated for weight loss Age: G1: 58.8 (11.3); G2: 58.6 (9.6) y Sex: G1: 65.1% women; G2: 63.8% women	G1: Liraglutide 3 mg starting with 0.6 mg/d increasing biweekly by 0.6 mg/day until 3 mg/d was reached G2: Placebo	G1: 66 G2: 69	52	KOOS Function (0–100)
Jastreboff, 2022 [35]	Type of population: Overweight or obese adults with a BMI ≥30 kg/m2 or a BMI ≥27 and at least one weight-related complication (e.g., hypertension, dyslipidaemia, obstructive sleep apnoea, or cardiovascular disease) Age: 42.2 y Sex: 73% women	G1: Tirzepatide 5 mg G2: Tirzepatide 10 mg G3: Tirzepatide 15 mg G4: Placebo	Pooled tirzepatide (G1: G2: G3): 124 G4: 36	72	SF-36 Physical function score Lean body mass (kg)
Johansen, 2022 [36]	Type of population: Adults with T1DM + insulin therapy, T1DM according to WHO criteria with duration of ≥1 year, BMI >22.0 kg/m2, and HbA1c >7.5% and <10.0% Age: G1: 50.1 (14.2); G2: 50.4 (14.0) y Sex: 28% women	G1: Exenatide 10 µg, tid, sc injection prior to main meals G2: Placebo	G1: 52 G2: 53	26	Femoral neck BMD Total neck BMD Total Lumbar spine BMD
Schmidt, 2022 [37]	Type of population: Adults aged 18–70 y with T1DM and insulin pump use for ≥1 year, overweight or obese with BMI >25 kg/m2, and HbA1c 7.5% Age: G1: 54 (14); G2: 43 (12) y Sex: 68% women	G1: Liraglutide 1.8 mg. The treatment was initiated with an injection of 0.6 mg liraglutide, increased to 1.2 mg liraglutide after one week, and increased to 1.8 mg after another week G2: Placebo	G1: 22 G2: 22	26	Total lean mass (kg)
Shoemaker, 2022 [38]	Type of population: Patients aged 10–26 y with hypothalamic obesity (HO) with suprasellar tumours. Diagnosis of HO with age- and sex-adjusted BMI ≥95% or BMI ≥30 kg/m2 if over 18 y. History of craniopharyngioma or another tumour located in the hypothalamic area Age: 16 y Sex: 61% women	G1: Exenatide extended release 2 mg sc once weekly G2: Placebo	G1: 22 G2: 15	36 weeks (2 weeks placebo run-in)	Lean mass (kg)
Cai, 2021 [39]	Type of population: Adults ≥40 years with T2DM of at least 6 months duration, with 6.5% ≤ HbA1c ≤ 10%, and the glycaemic control was stable for at least 90 days (unadjusted hypoglycaemic treatment) Age (based on the Excel): eG1: 62.95 (7.41); G2: 57.42 (7.89); G3: 64.36 (9.27); G4: 62.00 (4.99) y OR (based on the comment by YM) Mean (± SEM): eG1: 62.95 (1.70); dG2: 57.42 (1.81); iG3: 64.36 (2.93); G4: 62.00 (1.21) Sex: 44.6% women	G1: Exenatide (2 mg/wk) G2: Dulaglutide (1.5 mg/wk) G3: Insulin glargine (6 unit/day) G4: Placebo (once weekly)	G1: 19 G2: 19 G3: 10 G4: 17	52	BMD-L1 (g/cm2) BMD-L2 (g/cm2) BMD-L3 (g/cm2) BMD-L4 (g/cm2) BMD-L1-L4 (g/cm2) BMD-femoral neck (g/cm2) BMD-total hip (g/cm2)
Gibbons, 2021 [40]	Type of population: Adults aged 18–75 y with T2DM for at least 90 days, HbA1c 6.0–9.0%, BMI 20–38 kg/m2, stable body weight (<3 kg change during 90 days prior to screening) and who were treated with diet and exercise and/or stable dose of metformin for more than 30 days Age: 58.2 (37–73) y Sex: 13% women	G1: Semaglutide. The starting dose of od oral semaglutide was 3 mg (Weeks 0–4), escalating to 7 mg (Weeks 4–8) then 14 mg (Weeks 8–12) G2: Placebo	G1: 15 G2: 15 (cross-over trial)	12	Whole body lean mass (kg)
Gudbergsen, 2021 [41]	Type of population: Overweight/obesity with clinical diagnosis of knee OA (ACR criteria) with early to moderate radiographic changes (Kellgren-Lawrence (KL) grades 1, 2, or 3), age ≥18 and <75 y, BMI ≥ 27 kg/m2, stable body weight during the previous 3 months Age: 59.2 (10.3) y Sex: 65% women	G1: Liraglutide 3 mg starting with 0.6 mg/d and followed by incremental biweekly dose escalation steps of 0.6 mg/d to liraglutide 3 mg/d G2: Placebo	G1: 80 G2: 76	52	WOMAC Pain score WOMAC Stiffness Score WOMAC Function WOMAC Total score
Lundgren, 2021 [42]	Type of population: Adults with obesity (BMI 32–43), no diabetes complication Age: 42.84 (11.87) y Sex: 64% women	G1: Liraglutide 3 mg The starting dose was 0.6 mg/d with weekly increments of 0.6 mg until reaching a dose of 3.0 mg/d G2: Placebo G3: Exercise G4: Liraglutide 3 mg + exercise	G1: 49 G2: 46 G3: 48 G4: 49	52	Lean mass (kg)
Neeland, 2021 [43]	Type of population: Overweight or obese adults without T2DM, with BMI ≥30 kg/m2 or ≥ 27 kg/m2 with metabolic syndrome, defined as at least three of the following: waist circumference >102 cm in men and 88 cm in women, triglycerides >150 mg/dL or on treatment for hypertriglyceridaemia, HDL cholesterol <40 mg/dL in men and <50 mg/dL in women, BP >130/85 mmHg or on treatment for hypertension, fasting glucose >100 mg/dL Age: 50.2 (9.4) y Sex: 92% women	G1: Liraglutide 3.0 mg/d. Escalate the liraglutide (active) dose to 3.0 mg/d over a 4-week period following a 4-weekly dose of 0.6 mg/d and weekly dose escalation steps of 0.6 mg/d through sc injection G2: Placebo	G1: 73 G2: 55	40	Total body lean mass
Wilding, 2021 [44]	Type of population: Overweight or obese and without diabetes, BMI ≥30.0 kg/m2 or ≥27.0 kg/m2 with the presence of at least one of the following weight-related comorbidities (treated or untreated): hypertension, dyslipidaemia, obstructive sleep apnoea, or cardiovascular disease Age: 46 y; subsample total lean body max, 51 (12) y Sex: 74.1% women; subsample total lean body mass, 75.7% women	G1: Semaglutide 2.4 mg + Lifestyle intervention Semaglutide, administered with a prefilled pen injector, was initiated at a dose of 0.25 mg once weekly for the first 4 weeks, with the dose increased every 4 weeks to reach the maintenance dose of 2.4 mg weekly by Week 16 (lower maintenance doses were permitted if participants had unacceptable side effects with the 2.4-mg dose) G2: Placebo + lifestyle intervention	G1: 1296 G2: 650 Subsample lean body mass G1: 95 G2: 45	68	Total lean body mass (kg) SF-36 Physical functioning score
Ghanim, 2020 [45]	Type of population: Overweight and obese T1DM patients, with HbA1c of ≤10.0%, between the ages of 18–75 y, with BMI ≥25 kg/m2 Age: G1: 47 (2) and G2: 45 (3) y Sex: 47.6% women	G1: Liraglutide 0.6 mg/d for 1 week followed by 1.2 mg/d for 1 more week and then at 1.8 mg/d for rest of the study G2: Placebo	G1: 37 G2: 27	26	Total lean mass (kg) Total BMD (g/cm2)
Kadouh, 2020 [46]	Type of population: Healthy overweight adults (BMI ≥27 kg/m2) and adults with obesity (BMI >30 kg/m2), aged 18–65 y Age: 39.9 (11.6) y Sex: 87.5% women	G1: Liraglutide. The FDA-recommended dose escalation was used, starting at 0.6 mg sc od, increased weekly by 0.6 mg until maintenance dose of 3.0 mg daily is reached G2: Placebo	G1: 19 G2: 21	16	Lean body mass (kg)
McCrimmon, 2020 [47]	Type of population: Adults with T2DM on stable daily dose of metformin, with HbA1c of 7.0–10.5% Age: 58.2 (10.0) y Sex: not reported	G1: Once-weekly semaglutide 1.0 mg sc injections and od canagliflozin placebo oral tablets G2: od canagliflozin 300 mg oral tablets and once-weekly sc semaglutide placebo	G1: 88 G2: 90	52	Total lean mass (kg or %) Visceral lean mass (kg or%)
Nakaguchi, 2020	Type of population: Patients with T2DM on insulin, HbA1c ≥7.0%, ≤9.5% Age: G1: 67.2 (9.0); G2: 66.3 (9.5) y Sex: 31% women	G1: Liraglutide 0.9 mg/d for 24 weeks Starting at 0.3 mg od with weekly increments of 0.3 mg, reaching a final daily dose of 0.9 mg by the end of the second week G2: Empagliflozin 10 mg/d for 24 weeks Provided that from 12 weeks onward, if FPG is ≥180 mg/dL or HbA1c ≥8.5%, the dose of empagliflozin is increased up to 25 mg/d	G1: 32 G2: 32	24	Lean tissue mass (kg)
van Eyk, 2020 [48]	Type of population: Overweight and obese patients with T2DM, treated with metformin, and with a HbA1c ≥7.0 and ≤ 10.0%. Concomitant treatment with sulfonylurea derivatives and insulin was optional, but the dosage of all glucose-lowering medication needed to be stable for at least 3 months prior to participation Age: G1: 59.9 (6.2); G2: 59.2 (6.8) y Sex: 41% women	G1: Liraglutide. The starting dose of the study medication was 0.6 mg/d, which was titrated in two weeks to a maximum dose of 1.8 mg/d, if tolerated. Dose: s.c. 0.6 mg (0.1 mL) od. After 1 week, the dose was increased to 1.2 mg (0.2 mL) od G2: Placebo	G1: 22 G2: 25	26	Lean body mass (kg)
Wadden, 2020 [49]	Type of population: Patients with obesity (BMI above or equal to 30 kg/m2) Age: G1: 45.4 (11.6); G2: 49.0 (11.2) y Sex: 83% women	G1: Liraglutide 3.0 mg Administered sc (under the skin) od for 56 wk. Dose gradually increased to 3.0 mg + IBT G2: Placebo + IBT	G1: 142 G2: 140	56	6-min walk test (m)
Feng, 2019 [50]	Type of population: Patients with T2D aged 18–70 y, no hypoglycaemic drug use during the preceding 3 months, HbA1c levels of 7.0–14%, BMI of 20–38 kg/m2, with non-alcoholic fatty liver disease and weight fluctuations of <10% within the past 3 months Age: 47.2 (1.2) y Sex: 27.96% women	G1: Liraglutide. The sc dose of liraglutide was 0.6 mg qd during the first week, 1.2 mg qd during the second week and 1.8 mg qd from the third week to the end of the study G2: Metformin The oral dose of metformin was 250 mg tid during the first week, 500 mg tid during the second week and 1000 mg bid from the third week to the end of the study G3: Gliclazide. The initial oral dose of gliclazide was 30 mg before breakfast; this was gradually increased a maximum of 120 mg/d in order to reach the target for a fasting capillary plasma glucose concentration of <7.0 mmol/L	G1: 29 G2: 29 G3: 27	24	Total lean tissue mass (kg)
Gerstein, 2019 [51]	Type of population: Patients with T2DM with HbA1c equal to or less than 9.5% (≤81 mmol/mol), Anti-hyperglycaemic drug naive or treated with up to 2 oral hyperglycaemic drugs with or without a glucagon-like peptide-1 analogue or basal insulin, or basal insulin alone Age: 66.2 (6.5) y Sex: 46% women	G1: Dulaglutide, 1.5 mg administered once weekly, sc G2: Placebo	G1: 4949 G2: 4952	5.4 years (IQR: 5.1–5.9)	Fractures
Hygum, 2019 [52]	Type of population: Men and women aged ≥ 30 y diagnosed with T2DM, not diagnosed with osteoporosis, HbA1c > 48 mmol/mol Age: G1: 62 (59–65); G2: 64 (61–67) y Sex: 50% women	G1: Liraglutide 1.8 mg/d. A starting dose of 0.6 mg/d of the investigational drug with a dose escalation of 0.6 mg everyone/two weeks depending on tolerability up to 1.8 mg G2: Placebo	G1: 30 G2: 30	26	Hip total areal bone mineral density (aBMD) Hip neck aBMD Lumbar spine total BMD
Wagner, 2019 [53]	Type of population: T2DM patients with a HbA1c 7–10% with oral agents and/or intermediate-/long-acting insulin Age: 52.8 (11.7) y Sex: 62.2% women	G1: Liraglutide 1.2–1.8 mg/d was given, as a daily, self-administered sc injection using a pre-filled injection pen containing 3 mL (18 mg). The target dose will be 1.8 mg/d, but 1.2 mg/d will be accepted if the higher dose is not tolerated Participants will administer 0.6 mg/d during the first week, increasing to 1.2 mg/d during the second week and to 1.8 mg/d during the third week G2: Placebo	G1: 12 G2: 12	24	6-min walk test
Dubé, 2018 [54]	Type of population: Adults with T1DM, with diabetes duration ≥5 y, nonsmoker, and with BMI ≥25 kg/m2 Age: 35.8 (1.7) y Sex: 53% women	G1: Liraglutide + insulin. Participants injected study drug (liraglutide/placebo) 0.6 mg/d for 7 to 10 days (upon tolerance), escalated to 1.2 mg/d for 7 to 10 days (upon tolerance) and 1.8 mg/d for the rest of the 24 weeks of treatment G2: Placebo + insulin	G1: 15 G2: 15 (cross-over trial)	24	Fat free mass (kg)
Ludvik, 2018	Type of population: Patient with T2DM on SGLT-2i therapy, HbA1c concentration ≥7.0% and ≤9.5%, a BMI of ≤45 kg/m2 Age: G1: 56.17 (9.26); G2: 58.55 (9.14); G3: 57.10 (9.59) y Sex: 50% women	G1: Dulaglutide 1.5 mg, once weekly G2: Dulaglutide 0.75 mg, once weekly G3: Placebo	G1: 142 G2: 142 G3: 140	24	Fractures
Yin, 2018 [55]	Type of population: Overweight and obese patients with T2DM receiving a stable metformin dose of ≥1.5 g/d for > 8 weeks; aged 18 to 70 y; HbA1c ≥7.0 and ≤10.0%; BMI ≥24 kg/m2 (based on the China Obesity Task Force criteria) Age: G1: 47.6 (2.5); G2: 48.3 (2.3) y Sex: 35% women	G1: Exenatide. The initial dosage was 5 μg bid for 4 weeks followed by 10 μg bid for the remainder of the trial + metformin treatment G2: Insulin glargine. The starting dose was 8 IU od, followed by a titrated dosage of ≥2 IU every 3 days based on fasting blood glucose (FBG) levels until the peripheral blood glucose level reached 6.1 mmol/L, following which a maintenance dose with fixed glargine was administered for the remaining 12 weeks + metformin treatment	G1: 19 G2: 18	16	Total lean mass (kg)
Blundell, 2017 [14]	Type of population: adults with obesity (BMI 30–45 kg/m2), non-diabetic, HbA1c <6.5, and stable body weight (< 3 kg change during the 3 months prior to screening) Age: 42 (21 to 70) y Sex: 33.3% women	G1: Semaglutide 1.0 mg. The starting dose was 0.25 mg (4 weeks), escalating to 0.5 mg (4 weeks) and then 1.0 mg (4 weeks). Subjects received a fifth dose (administered at the clinic) of 1.0 mg at the last visit of each treatment period and assessments were conducted G2: Placebo	G1: 30 G2: 28 (cross-over trial)	12 + 12 weeks (cross-over trial)	Body lean mass (kg)
Bouchi, 2017 [56]	Type of population: Adult patients with T2DM on insulin treatment, BMI ≥25 kg/m2, with HbA1c levels between 7 to 10% Age: 59 (13) y Sex: 53% women	G1: Liraglutide 0.9 mg + insulin Liraglutide was administered from 0.3 mg/day and the dosage was increased to 0.6 mg after one week and 0.9 mg after a further week. The dose was maintained until study completion G2: Insulin glargine	G1: 8 G2: 9	24	Skeletal muscle index (SMI)
Jorsal, 2017 [57]	Type of population: Stable chronic heart failure patients with and without diabetes Age: G1: 65 (9.2); G2: 65 (10.7) y Sex: 11% women	G1: Liraglutide 1.8 mg od Study medication was introduced at a dose of 0.6 mg/d, which was increased to 1.2 mg/d after 1 week and to 1.8 mg/d thereafter. A dose increase could be postponed depending on the patient’s tolerance to the trial product or reduced at any time during the trial if required G2: Placebo	G1: 122 G2: 119	24	6-min walk test (m)
Mensberg, 2017 [58]	Type of population: Overweight, dysregulated patients with T2DM, treated with diet and/or metformin; HbA1c between 7% and 11%, BMI >25 kg/m2, and sedentary lifestyle (self-reported physical activity <150 min/wk) Age: G1: 56.5 (9); G2: 55.6 (12) y Sex: 30% women	G1: Liraglutide + exercise. Od, sc injections of 0.1 mL of study medication (0.6-mg liraglutide) in the evening for 1 week, 0.2 mL (1.2 mg-liraglutide) the following week and thereafter 0.3 mL (1.8 mg liraglutide) for the remaining study period G2: Placebo + exercise	G1: 17 G2: 16	16	Lean body mass (kg)
Margulies, 2016 [59]	Type of population: Patients with advanced heart failure and reduced ejection fraction Age: 60.3 (11.9) y Sex: 81.6% women	G1: Liraglutide 1.8 mg The protocol involved up-titration of study drug dosage as tolerated every 14 days from 0.6 mg/d to 1.2 mg/d to 1.8 mg/d during the first 30 days of the trial G2: Placebo	G1: 154 G2: 146	180 days	6-min walk test (m)
Gilbert, 2015 [60]	Type of population: Patients with T2DM aged 18 to 80 y, BMI 45 kg/m2 who were either drug-naïve, treated with diet and exercise, or had failed to achieve control with a single oral agent at 50% of maximum approved dose, with HbA1c value of 7% to 11% if treated with diet and exercise or 7% to 10% if treated with oral anti-diabetic monotherapy Age: G1: 54.0 (8.5); G2: 55.3 (10.8), G3: 53.5 (13.0) y Sex: 54.1% women	G1: Liraglutide 1.8 mg/d G2: Liraglutide 1.2 mg/d G3: Glimepiride 8 mg/d	G1: 14 G2: 9 G3: 6	104	BMD (g/cm2)
Iepsen, 2015 [61]	Type of population: Obese, glucose-tolerant, nondiabetic women aged 18–65 y, with BMI 30–40 kg/m2) Age: 46 (2) y Sex: 100% women	G1: Liraglutide 1.2 mg/d, sc injection G2: Placebo	G1: 18 G2: 19	8 week-low calorie diet programmes AND 52 weeks of treatment	Total BMD
Astrup, 2011 [62]	Type of population: Obese individuals with BMI ≥30.0 or ≤40.0 kg/m2, with stable body weight (< 5% self-reported change within the last 3 months) postmenopausal with T2DM and HbA1c of 7% to 10% despite using metformin (BMI>30 kg/m2) Age: G1: 47.2 (9.7); G2: 45.5 (10.9); G3: 45.0 (11.1); G4: 45.9 (10.7); G5: 45.9 (10.3); G6: 45.9 (9.1) y Sex: 73% women (full database)	G1: Liraglutide 1.2 mg starting at 0.6 mg/d and increasing weekly (dose escalation) G2: Liraglutide 1.8 mg G3: Liraglutide 2.4 mg G4: Liraglutide 3.0 mg G5: Placebo G6: Orlistat, the open-label comparator group (n = 95) was randomised to receive orlistat capsules (3 x 120 mg) with each main meal for the full 2-year period	G1: 15 G2: 13 G3: 15 G4: 15 G5: 14 G6: 12	20	Lean tissue (kg)
Bunck, 2011 [63]	Type of population: Metformin-treated patients with T2DM, with HbA1c 6.5–9.5%, BMI 25–40 kg/m2 Age: 59 (8) y Sex: 34.8% women	G1: Exenatide G2: Insulin glargine	G1: 36 G2: 33	44	Total BMD (g/cm2)
Bunck, 2010 [64]	Type of population: Metformin-treated patients with T2DM, with HbA1c 6.5–9.5%, and BMI 25–40 kg/m2 Age: 59 (8) y Sex: 35% women	G1: Exenatide G2: Insulin glargine	G1: 36 G2: 33	52	Total lean mass (kg)
Jendle, 2009 [65]	LEAD-2 Type of population: Obese metformin-treated patients with T2DM Age: G1: 58 (10) G2: 59 (8) G3: 58 (9) G4: 56 (10) G5: 56 (9) Sex: 35% women	G1: Liraglutide 0.6 mg G2: Liraglutide 1.2 mg G3: Liraglutide 1.8 mg G4: Placebo G5: Glimepiride 4 mg	G1: 30 G2: 27 G3: 29 G4: 12 G5: 32	26	Total lean body tissue mass (kg)
	LEAD-3 Type of population: Obese patients with T2DM Age: G1: 55 (11) G2: 54 (9) G3: 54 (13) Sex: 54% women	G1: Liraglutide 1.2 mg G2: Liraglutide 1.8 mg G3: Glimepiride 8 mg	G1: 9 G2: 14 G3: 6	52	Total lean body tissue mass (kg)
Harder, 2004 [66]	Type of population: Obese adults with T2DM, HbA1c for diet-treated subjects of 7–12%, both inclusive; HbA1c for sulfonylurea-treated subjects ≤10%, BMI ≥ 27 kg/m2 Age: 60.0 (9.5) y Sex: 66% women	G1: Liraglutide 0.6 mg G2: Placebo	G1: 21 G2: 12	8	Lean mass (%)

Conflict of interest and funding information for each study, where available, can be viewed on the OSF platform: https://osf.io/zbhwc

bid twice daily, BMI body mass index, BP blood pressure, Hb1Ac glycated haemoglobin, IBT intensive behavioural therapy, OA osteoarthritis, od once daily, qd four times daily, sc subcutaneous, T2DM type 2 diabetes mellitus, tid three times daily

Table 2.

Real-word evidence (RWE) and pharmacovigilance studies' characteristics (n = 14, newest first)

First author, year	Population description (health condition, mean age, sex)	Arms description (treatment description + dose in each group)	Sample size per groupa	Treatment duration (weeks)	Outcomes
Song, 2024 [67]	Type of population: Adults aged 19–65 y with obesity (BMI ≥25.0 kg/m2) or overweight (BMI ≥23.0–24.9 kg/m2), and treated with phentermine, phentermine/topiramate, liraglutide, naltrexone/bupropion, lorcaserin, or orlistat for at least 6 months Age: 39.3 y Sex: 71.2% women	G1: Liraglutide G2: Phentermine G3: Phentermine/topiramate G4: Naltrexone/bupropion G5: Lorcaserin G6: Orlistat	G1: 62 G2: 39 G3: 51 G4: 17 G5: 24 G6: 12	26	Skeletal muscle mass (kg)
Xiao, 2024 [68]	Type of population: Patients with T2DM Age: 61 to 64 y Sex: % women not reported	G1: GLP-1 RA (exenatide, liraglutide, albiglutide, lixisenatide, dulaglutide, semaglutide) G2: Non GLP-1 RA	G1: 98625 G2: 391482	991 weeks (Jan 2004 to Dec 2022)	Fracture events
Agcakaya, 2023 [69]	Type of population: Adults diagnosed with T2DM, using metformin + SGLT2-i or metformin + exenatide Age: 56.99 y Sex: 79% women	G1: Metformin + SGLT2i (dapagliflozin 10 mg or empagliflozin 10 mg od) G2: Metformin + GLP-1 RA (exenatide 10 mcg od)	G1: 50 G2: 50	24	Total body fat-free mass percentage (%) Total body muscle percentage (%)
Ko, 2023 [70]	Type of population: Postmenopausal women aged ≥45 y with T2DM with ≥1 prescription of SGLT2 inhibitors or DPP4 inhibitors Age: Cohort study before weighting by propensity score fine stratification, G1: 62.52 (10) and G2: 61 (6.1) y; Cohort study after weighting by propensity score fine stratification, G1: 62.5 (10.2) and G2: 61.4 (9.8) y Sex: 100% women	G1: GLP-1 RA (albiglutide, dulaglutide, exenatide, or lixisenatide) G2: SGLT2i (dapagliflozin, empagliflozin, ipragliflozin, or ertugliflozin)	Cohort study before weighing by propensity score fine stratification, G1: 8181 and G2: 113622 Cohort study after weighting by propensity score fine stratification, G1: 8177 and G2: 111835	Cohort study before weighing by propensity score fine stratification, NR Cohort study after weighting by propensity score fine stratification, G1: 0.82 (0.85) y and G2: 1.43 (1.33) y	Overall fractures Vertebral fractures Hip fractures
Osaka, 2023 [71]	Type of population: Older patients aged ≥70 years with T2DM who received glargine/lixisenatide or basal insulin Age: G1: 76.7 (5.2); G2: 76.8 (5.4) y Sex: 40% women	G1: Glargine/lixisenatide G2: Basal insulin	G1: 10 G2: 10	Nine in-hospital days from baseline to follow-up of the two BIA assessments Recruitment period: May 2019–Feb 2021	Skeletal muscle mass (kg)
Patil, 2023 [72]	Type of population: Veteran patients with T2DM who were prescribed a SGLT-2i (empagliflozin, canagliflozin, or dapagliflozin) or GLP-1RA (liraglutide, dulaglutide, lixisenatide, or semaglutide) Age: Unmatched cohort study, G1: 64.73 (10.36) and G2: 65.27 (9.87) years. Matched cohort study, G1: 64.73 (10.44) and G2: 64.87 (9.98) Sex: Unmatched cohort study, G1: 6.96% and G2: 3.94% women; matched cohort study, G1: 4.23% and G2: 4.22% women	G1: GLP-1 RA [3.415 (37.46%) dulaglutide; 19.429 (54.26%) liraglutide; 2.952 (8.24%) exenatide; 14 (0.04%) lixisenatide; 0 (0%) semaglutide] G2: SGLT2i [5.510 (99.16%) empagliflozin; 79 (0.22%) dapagliflozin; 209 (0.58%) canagliflozin; 12 (0.33%) ertugliflozin]	Unmatched cohort study, G1: 37149 and G2: 75370; matched cohort study, G1: 35347 and G2: 35347	G1: median follow-up time, 1.54 y G2: median follow-up time, 1.01 y (analysis of all outcomes)	All clinical fractures Hip fractures
Zhu, 2023 [73]	Type of population: Patients aged >45 y with T2DM and KOA Age: G1: 60.7 (8.7); G2: 61.2 (8.6) y Sex: 72.99% women	G1: GLP-1 RA (liraglutide, oligopeptide, semaglutide, losenatide, risenatide, exenatide) G2: Non GLP-1 RA	G1: 233 G2: 1574	G1: 7.7 (1.5) y, 402 weeks G2: 7.8 (1.6) y, 407 weeks	WOMAC Total score WOMAC Pain Sub-score WOMAC Stiffness sub-score WOMAC Function sub-score
Al-Mashhadi, 2022 [74]	Type of population: All individuals who initiated a combination of metformin and GLP- 1 RA or metformin and DPP-4i treatment between Jan 1st 2007 and Dec 31st 2018 in Denmark. As subjects were included when either treatment combination was initiated, any previous use of metformin, GLP-1 RA or DPP-4is alone or in combination with any other glucose-lowering drug was allowed Age: cohort study, G1: 56.6 (12.0) and G2: 63.6 (12.4) y; matched cohort study, G1: 57.5 (11.3) and G2: 57.9 (11.0) y Sex: cohort study, G1: 43.1% and G2: 40.3% women; matched cohort study, G1: 42.3% and G2: 41.3% women	G1: GLP-1 RA G2: DPP4i	G1: cohort study, 16723; matched cohort study, 16133 G2: cohort study, 26093; matched cohort study, 16133	G1: cohort study, 637 days (222–1403); matched cohort study, 642 days (223–1414) G2: cohort study, 519 days (196–1133); matched cohort study, 529 days (207–1131)	Mayor Osteoporotic Fractures (MOF) Any Fractures Hip Fractures Vertebral Fractures
Al-Mashhadi, 2022 [75]	Type of population: Adults with diabetes treated with metformin in combination with either SGLT2 inhibitors or GLP-1 receptor agonists and no concurrent treatment with any other glucose-lowering drugs between Jan 1st 2012 and Dec 31st 2018 Age: cohort study, G1: 60.01 (11.4) and G2: 57.4 (12.1) y; matched cohort study, G1: 61.1 (11.3) and G2: 58.5 (12.0) y Sex: cohort study, G1: 35.8% and G2: 42.4% women; matched cohort study, G1: 38.5% and G2: 40.0% women	G1: SGLT2i G2: GLP-1 RA	Cohort study: G1: 13775 and G2: 13768 participants Matched cohort study: G1: 9190 and G2: 9190 participants	Cohort study: G1: 334 (139–662) days and G2: 497 (185–1077) days Matched cohort study: G1: 355 (126–779.8) and G2: 372 (136.2–766)	Mayor Osteoporotic Fractures (MOF) Any Fractures Hip fractures Vertebral Fractures
Keskin, 2022 [76]	Type of population: Adults with T2DM with BMI ≥30 kg/m2 and HbA1c 6–10% Age: 49.70 (7.93) y Sex: 67% women	G1: Liraglutide 3 mg + Metformin G2: Metformin	G1: 138 G2: 138	12	Muscle mass (kg)
Patorno, 2021 [77]	Type of population: Older patients (aged ≥66 y) with T2DM with cardiovascular disease with an SGLT2i, i.e., canagliflozin, dapagliflozin, or empagliflozin, or a GLP-1RA, i.e., albiglutide, dulaglutide, exenatide, or liraglutide Age: 71 y Sex: G1: 56.46% and G2: 50,64%; propensity score matching, G1: 54% and G2: 54.40% women	G1: GLP-1 RA (liraglutide 58.7%; exenatide 23.5%; dulaglutide 14.8%; albiglutide 3.0%) G2: SGLT2i (canagliflozin 76.9%; dapagliflozin 13.1%; empagliflozin 11.1%)	G1: 64417 and G2: 72900; propensity score matching, G1: 45047 and G2: 45047	NR	Bone fractures (humerus, wrist, hip, pelvis)
Zhuo, 2021 [78]	Type of population: Patients aged ≥66 y with T2DM who were newly prescribed an SGLT-2i, a DPP-4i, or a GLP-1RA Age: G1: 71.46 (4.84); G2: 74.69 (6.71); G3: 71.94 (5.17) y. Propensity score matching: G1: 71.67 (4.97); G2: 71.64 (5.13); G3: 71.60 (4.96) Sex: 55% women; propensity score matching, 53% women	G1: GLP-1 RA (albiglutide, dulaglutide, exenatide, liraglutide, lixisenatide, semaglutide) G2: DPP4i (alogliptin, linagliptin, saxagliptin, sitagliptin) G3: SGLT2i (canagliflozin, dapagliflozin, empagliflozin)	G1: 66016 G2: 338463 G3: 62454 Propensity score matching: G1: 45889 G2: 45889 G3: 45889	NR Propensity score matching: G1: 250 (249) days, 35.71 weeks G2: 295 (278) days, 42.14 weeks G3: 268 (262) days, 38.29 weeks	Fractures
Ueda, 2018 [79]	Type of population: All patients in Denmark and Sweden, aged ≥35 y, who filled their first prescription for either an SGLT2 inhibitor or a GLP-1 RA during the study period Age: 60–62 y Sex: 40% women	G1: GLP-1 RA G2: SGLT2i	G1: 27278; G2: 21008 Propensity score matching: G1: 17213; G2: 17213	182 (Jul 2013 to Dec 2016)	MOF
Driessen, 2015 [80]	Type of population: Adult patients with T2DM with at least one prescription for a non-insulin anti-diabetic drug Age: G1: 53.5 (10.5), G2: 61.0 (15.1) y Sex: 47% women	G1: GLP-1 RA users G2: Never GLP-1 users	G1: 8364 G2: 208462	G1: 4.61 (1.12) G2: 3.46 (2.67)	MOF Hip fracture Vertebral fracture

Conflict of interest and funding information for each study, where available, can be viewed on the OSF platform: https://osf.io/zbhwc

bid twice daily, BMI body mass index, BP blood pressure, DPP4i dipeptidyl peptidase-4 inhibitor, GLP-1 RA glucagon-like peptide-1 receptor agonist, Hb1Ac glycated haemoglobin, IBT intensive behavioural therapy, OA osteoarthritis, od once daily, qd four times daily, sc subcutaneous, SGLT2i sodium-glucose cotransporter-2 inhibitor, T2DM type 2 diabetes mellitus, tid three times daily

Studies included a total of 1,250,717 individuals. Sample sizes ranged from 15 [40, 54] to 17,604 [35] in the RCTs and from 20 [71] to 466,933 [78] in the RWE studies. The highest number of participants was observed in the pharmacovigilance study with 490,107 participants. Of the 46 RCTs included, participants exhibited heterogeneous characteristics in terms of body weight and metabolic profile. The study populations comprised individuals who were overweight, obese, or of normal weight. Some participants with type 2 or type 1 diabetes were treated with metformin or insulin. Other trials included specific clinical populations, such as patients who had undergone metabolic surgery, individuals receiving haemodialysis, those with hypothalamic obesity associated with a suprasellar tumour, patients with advanced heart failure and reduced ejection fraction, and individuals with knee osteoarthritis (KOA). Among the 13 RWE and 1 pharmacovigilance studies included, participants were predominantly diagnosed or newly diagnosed with T2DM. Consequently, GLP-1 RAs therapy was either initiated or already established. One study specifically reported a population with KOA. The length of treatment ranged from less than 3 weeks (19 days [30]) or 6 weeks [25] to 104 weeks [60] in RCTs and from 9 in-hospital days [71] to 991 weeks [68] in RWE studies.

Different GLP-1 RAs were represented across the panel of studies, including semaglutide, liraglutide, exenatide, Dulaglutide and lixisenatide tirzepatide (dual agonist GIP/GLP-1) but also others (i.e., SAR425899 [bamadutide], albiglutide, loxenatide, risenatide, oligopeptide). Glucagon-like peptide-1 RAs were compared to placebo, sodium-glucose cotransporter 2 inhibitors (SGLT2-i), DPP-4 inhibitors (DPP-4i), other classical oral antidiabetics (i.e., metformin, insulin and combination, gliclazide, etc.) or other anti-obesity medications (i.e., orlistat, phentermine, Lorcaserin, etc.).

A variety of different outcomes were reported across the panel of studies, including bone-related outcomes (i.e., bone mineral density [BMD]: lumbar spine BMD (5 studies), femoral neck BMD (5 studies), total hip BMD (4 studies), and total BMD (5 studies) and incidence of fractures: hip fracture (6 studies), vertebral fracture (5 studies), any fracture (11 studies), and major osteoporotic fracture (MOF) (4 studies), muscle-related outcomes (i.e., lean body mass [24 studies], muscle mass [2 studies], skeletal muscle mass [3 studies], skeletal muscle mass index [1 study], fat-free mass [4 studies]), and joint related outcomes (i.e., KOOS function [1 study], WOMAC physical function score [3 studies], WOMAC pain score [3 studies], WOMAC stiffness score [3 studies], WOMAC total score [3 studies], and the 6-minute walk test [4 studies]).

Study Quality

Risk of bias was assessed using the Cochrane Risk of Bias 2.0 tool for RCTs and the Newcastle-Ottawa Scale for RWE [22]. The complete results of the study quality per study are available in Online Resource 3, 4 and 5. For analyses conducted according to the intention-to-treat approach, 74% of studies were considered to be at low risk of bias, 22% raised some concerns, and 4% were considered at high risk of bias. The domains most frequently associated with concerns were the randomisation process, deviations from intended interventions, and selection of the reported result. A high risk of bias was primarily driven by issues related to outcome measurement (Online Resource 3). For per-protocol analyses, 50% of studies were considered at low risk of bias, 18.8% raised some concerns, and 31.3% were at high risk of bias (five studies). In studies that raised some concerns, the domains most commonly affected were the randomisation process, missing outcome data, and selection of the reported result. High risk of bias in per-protocol analyses was mainly associated with deviations from intended interventions, missing outcome data and outcome measurement (Online Resource 4). In addition, three randomised crossover trials were evaluated separately using the Cochrane Risk of Bias 2.0 tool for crossover trials. All three trials were judged to raise some concerns, mainly related to the randomisation process and measurement of the outcome (Online Resource 5).

Regarding RWE studies assessed using the Newcastle–Ottawa Scale, 1 study received 6 stars, two received 7 stars, 2 studies received 8 stars, and 8 studies received 9 stars (Online Resource 6).

Effect of GLP-1 RAs on Bone-Related Outcomes

The complete results, including the global model, the sensitivity analyses, the subgroup analyses, and the publication bias assessment, are available on OSF platform: https://osf.io/zbhwc/.

Bone Mineral Density (BMD)

Overall, 10 unique RCTs assessed BMD, reporting lumbar spine (k = 5), femoral neck (k = 5), total hip (k = 4), and total body BMD (k = 5), with overlap across skeletal sites among these 10 RCTs. Studies ranged from 24 to 52 weeks and included between 35 and 108 randomised individuals with T1DM or T2DM (n = 7), obese individuals (n = 2) or individuals with increased risk of fractures (n = 1).

The random-effect model meta-analyses showed no effect of GLP-1 RAs on lumbar spine BMD (k = 5 RCTs, n = 245 individuals, MD 0.01, 95% CI −0.03; 0.04, I² 16.7%, p-value for heterogeneity 0.31), femoral neck BMD (k = 5 RCTs, n = 245 individuals, MD 0.03, 95% CI −0.05; 0.12, I² 79.5%, p-value for heterogeneity <0.001), total hip BMD (k = 4 RCTs, n = 265 individuals, MD 0.04, 95% CI −0.06; 0.14, I² 84.2%, p-value for heterogeneity <0.001) and total BMD (k = 5 RCTs, n = 254 individuals, MD 0.00, 95% CI −0.01; 0.01, I² 64.7%, p-value for heterogeneity 0.02) compared to controls. No difference was observed when analyses were stratified by the type of GLP-1 RAs or by the nature of controls. Results remained stable across all sensitivity analyses, including different scenarios for multi-arm studies (i.e., Cai et al, R4-arm trial [39]) and LOO sensitivity analyses. Publication bias was not assessed, as fewer than 10 studies were included in each meta-analysis.

Fractures Incidence

Twelve studies (8 RWE cohort studies and 4 RCTs), most T2DM individuals, reported fractures as an outcome and included between 29 and 490,107 individuals. The length of treatment ranged from 250 days to 4.6 years.

Hip fractures were investigated in 5 studies (i.e. 4 RWE studies and 1 RCT). When combining both RCTs and RWE studies, random-effect model meta-analysis showed no effect of GLP-1 RAs on hip fractures (k = 5 studies, n = 489,658 individuals, OR 0.86, 95% CI 0.20–3.72, I² 90.1%, p-value for heterogeneity <0.001). Restricting the evidence to RWE studies only confirmed the absence of association (k = 4 RWE studies, n = 472,054 individuals, OR 0.61, 95% CI 0.11–3.34, I² 92%, p-value for heterogeneity <0.001).

Major osteoporotic fractures were investigated in 3 RWE studies. The random-effect model meta-analysis showed no effect of GLP-1 RAs on MOF (k = 3, n = 315,778 individuals, OR 0.68, 95% CI 0.25–1.84, I² 97.4%, p-value for heterogeneity <0.001).

Any fractures were investigated in 10 studies (i.e.,7 RWE studies and 3 RCTs). The random-effect model meta-analysis showed no effect of GLP-1 RAs on any fracture (k = 10, n = 972,713 individuals, OR 0.93, 95% CI 0.71–1.22, I² 98.4%, p-value for heterogeneity <0.001) compared to controls (Fig. 2). No publication bias was reported (Egger test p-value = 0.12). Results remained stable when evidence was restricted to RWE studies only (k = 7, n = 961,643 individuals, OR 0.91, 95% CI 0.65–1.25, I² 98.9%, p-value for heterogeneity <0.001) or RCTs only (k = 3, n = 11,070 individuals, OR=1.05, 95% CI 0.89–1.25, I² 0%, p-value for heterogeneity 0.99). No difference was observed when analyses were stratified by the type of GLP-1 RAs or by the nature of controls. Results remained stable in adjusted and non-adjusted analyses and across all sensitivity analyses, including different scenarios for multi-arm studies (i.e., Al-Mashhadi et al, R3-arm study [74]) and LOO sensitivity analyses.

Fig. 2.

Effect of GLP-1 RAs on hip fractures (A), vertebral fractures (B), any fractures (C) and MOF (D). CI confidence interval, GLP-1 RAs glucagon-like peptide-1 receptor antagonists, MOF major osteoporotic fracture, OR odds ratio, RCT randomised controlled trial, RWE real-world evidence

Vertebral fractures were investigated in 4 studies (i.e., 3 RWE studies and 1 RCT). When combining both RCTs and RWE studies, the random-effect model meta-analyses showed a significant reduction of the incidence of vertebral fractures with GLP-1 RAs (k = 4, n = 418,964 individuals, OR 0.66, 95% CI 0.54–0.80, I² 7.1%, p-value for heterogeneity 0.36) compared to controls (Fig. 2). Restricting the evidence to RWE studies only confirmed the association (k = 3, n = 401,360 individuals, OR 0.66, 95% CI 0.54–0.81, I² 37.7%, p-value for heterogeneity 0.20). However, this reduction was no longer observed when the meta-analysis included the most adjusted effect sizes (i.e., hazard ratio [HR]) reported by authors (k = 3, pooled HR 0.95, 95% CI 0.36–2.51, I² 47.6%, p-value for heterogeneity 0.15). Results remained stable across all sensitivity analyses, including different scenarios for multi-arm studies (i.e., Al-Mashhadi et al, R3-arm study [74]) and LOO sensitivity analyses. Full results available on OSF platform (https://osf.io/zbhwc/).

Effect of GLP-1 RAs on Muscle-Related Outcomes

Lean Mass/Fat-Free Mass

Twenty-eight studies (27 RCTs and 1 RWE studies) reported lean body mass or fat-free mass as an outcome. Most of these studies included populations with overweight/obesity, T2DM, T2DM with overweight/obesity or T1DM with overweight/obesity. Sample size varied between 21 and 195 individuals and the length of follow-up varied between 19 days and 72 weeks. Most of the studies reported the absolute change in lean body mass/fat-free mass.

The random effect model meta-analyses showed a moderate and significant reduction of lean body mass/fat-free mass with GLP-1 RAs compared to controls (k = 28, n = 1765 individuals, SMD −0.52, 95% CI −0.80; −0.23, I² 88%, p-value for heterogeneity <0.001) (Fig. 3). A significant difference was observed when stratified by GLP-1 RA types, with liraglutide and semaglutide being the main drivers of reduction (p-value for subgroup difference <0.001). When stratifying by comparators, a large and significant reduction of lean body mass/fat free mass is observed when the use of GLP-1 RAs was compared to placebo (k = 20, n = 1188 individuals, SMD −0.70 95% CI −1.07; −0.33). When restricting the analysis to studies reporting specifically “fat-free mass” as an outcome, no significant reduction of fat-free mass is observed (k = 4, n = 242 individuals, SMD −0.08, 95% CI −0.44; 0.27, I² 39.7%, p-value for heterogeneity 0.18). Results remained stable across all sensitivity analyses, including different scenarios for multi-arm studies (i.e., Astrup et al, R3-arm study [62] and Wilding et al, R3-arm study [44]) and LOO sensitivity analyses. No publication bias was found (Egger test p-value = 0.98) (Fig. 4).

Fig. 3.

Effect of GLP-1 RAs on lean body mass/ fat-free mass, stratified by GLP-1 RAs. CI confidence interval, GLP-1 RAs glucagon-like peptide-1 receptor antagonists, SD standard deviation, SMD standardised mean difference

Fig. 4.

Funnel plot effect of GLP-1 RAs on lean body mass/fat-free mass. GLP-1 RAs glucagon-like peptide-1 receptor antagonists

Skeletal Muscle Mass

Five studies (2 RCTs and 3 RWE studies) reported skeletal muscle mass as an outcome. All studies measured this parameter using a BIA. Sample size ranged from 19 to 205 individuals with diabetes (n = 4) or obesity (n = 1). The length of treatment ranged from 19 days to 2.5 years.

In the general random-effect model comprising the 5 studies, a significant improvement in skeletal muscle mass with GLP-1 RAs (k = 5, n = 435 individuals, SMD 0.46, 95% CI 0.03; 0.89, I² 58.4%, p-value for heterogeneity 0.05) compared to controls was observed. When stratified by study design, this increase of skeletal muscle mass was observed only for RWE studies (k = 3, n = 370 individuals, SMD = 0.57, 95% CI 0.35; 0.89) and not in RCTs (k = 2, n = 65 individuals, SMD 0.33, 95% −0.86; 1.52). Results were not robust according to the LOO analysis, as no effect of GLP-1 RAs was observed when either the study of Bouchi et al. [56], Keskin et al. [76], Osaka et al. [71] or Song et al. [67] were removed from the model. No publication bias was reported (Egger test p-value = 0.94).

Effect of GLP-1 RAs on Joint-Related Outcomes

Three studies reported WOMAC physical function, WOMAC pain, WOMAC stiffness and WOMAC total as outcomes (2 RCTs and 1 RWE study), including between 156 to 1807 individuals with knee osteoarthritis and T2DM or obesity. The length of treatment ranged between 52 and 68 weeks in RCTs. Length of follow-up was 7.7 years in the RWE studies.

Physical Function

The random-effect model meta-analyses revealed no significant improvement of physical function with GLP-1 RAs compared to controls (k = 3, n = 2331 individuals, MD −4.62, 95% CI −13.7; 4.47, I² 89.8%, p-value for heterogeneity <0.001). Four additional RCTs (including from 24 to 300 individuals with liraglutide versus placebo during a range of 180 days to 56 weeks) also reported the 6-min walking test as an outcome. The model including either WOMAC physical function or the 6-min walking test did not show any improvement of physical function in individuals treated with GLP-1 RAs (k = 7, n = 3147 individuals, SMD −0.03, 95% CI −0.23; 0.18, I² 77.2%, p-value<0.001) (Fig. 5). No intergroup difference was observed when analyses were stratified by the type of GLP-1 RAs, by the nature of controls or studies design (i.e., RCTs vs RWE studies). Results remained stable in the LOO sensitivity analysis.

Fig. 5.

Effect of GLP-1 RAs on WOMAC function (+6 minutes walking test), WOMAC pain (B) and WOMAC stiffness (C). CI confidence interval, GLP-1 RAs glucagon-like peptide-1 receptor antagonists, MD mean difference, SD standard deviation, SMD standardised mean difference

WOMAC Pain, WOMAC Stiffness and WOMAC Total

The random-effect model meta-analyses showed no significant improvement of WOMAC pain with GLP-1 RAs compared to controls (k = 3, n = 2370 individuals, MD −5.31, 95% CI −13.37; 2.75, I² 90.5%, p-value for heterogeneity <0.001), no significant improvement of WOMAC stiffness with GLP-1 RAs compared to controls (k = 3, n = 2331 individuals, MD = −3.42, 95% CI −15.11; 8.28, I² 93.1%, p-value for heterogeneity <0.001) and no significant improvement of WOMAC total with GLP-1 RAs compared to controls (k = 3, n = 2331 individuals, MD = −4.70, 95% CI −13.82; 4.43, I² 90.5%, p-value for heterogeneity <0.001) (Fig. 5). An important heterogeneity in the results was noticed as one study (Bliddal et al. [16]) demonstrated significant improvement of WOMAC pain/stiffness/total with the use of GLP-1 RAs whereas the 2 other studies (Gudbergsen et al. [41] and Zhu et al. [73]) demonstrated either a significant decrease of WOMAC scores with the use of GLP-1 RAs or no effect of GLP-1 RAs on these outcomes.

Strength of Evidence

The certainty of evidence was assessed for all outcomes. Evidence was rated as high for lumbar spine BMD and any fractures (evidence from RCTs only) and moderate for femoral neck BMD, total hip BMD, total body BMD, vertebral fractures, and WOMAC function. The certainty of evidence was rated as low or very low for all other outcomes (Online Resource 7).

Discussion

To our knowledge, this systematic review and meta-analysis provide the most comprehensive synthesis to date of the effects of GLP-1 RAs on a wide panel of musculoskeletal outcomes, encompassing bone, muscle, and joint-related outcomes. By integrating evidence from RCTs and RWE studies across heterogeneous populations, our findings contribute to additional insights into the musculoskeletal safety profile of this rapidly expanding therapeutic class.

The integration of RCTs and RWE was deliberately chosen to provide a more pragmatic assessment of musculoskeletal outcomes, combining the high internal validity of clinical trials with the greater external validity and longer-term perspective offered by observational studies. This approach is particularly relevant for musculoskeletal outcomes, as fractures are relatively uncommon events in clinical trials, osteoarthritis progression is typically slow, and changes in body composition, muscle health, and physical function may require prolonged follow-up to become clinically apparent. Importantly, the consistency of findings across study designs, as demonstrated in a large majority of subgroup analyses, further reinforces the robustness of the overall conclusions.

Bone-Related Outcomes: Absence of Deleterious Effects Despite Promising Mechanistic Data

Our analyses showed no significant effect of GLP-1 RAs on BMD at any skeletal site (with a high to moderate level of evidence), nor on the overall incidence of fractures (with a moderate to low level of evidence), when the most adjusted effect estimates were considered. Pooled analyses demonstrated effect estimates close to null for lumbar spine BMD (MD 0.01 g/cm²), femoral neck BMD (MD 0.03 g/cm²), total hip BMD (MD 0.04 g/cm²), and whole-body BMD (MD 0.00 g/cm²). Similarly, fracture risk estimates remained neutral, with point estimates centred around unity for hip fractures (OR 0.86), MOF (OR 0.68), and any fracture (OR 0.93). Although a reduced risk of vertebral fractures was observed in crude analyses (OR 0.66), this association was no longer observed when the most adjusted models were considered (HR 0.95). These findings were consistent across study designs, comparator types, and GLP-1 RA molecules, and the general conclusion remained unchanged even after extensive sensitivity analyses.

These results contrast with current preclinical evidence suggesting a potential osteoprotective role of GLP-1 signalling. Indeed, mechanistic models have demonstrated that GLP-1 receptor activation may promote osteoblast differentiation, inhibit osteoclast activity, and reduce the accumulation of advanced glycation end products, thereby improving bone quality [11, 81]. Moreover, several meta-analyses focusing on biochemical markers of bone turnover have reported favourable effects of GLP-1 RAs on markers of bone formation and resorption [82–84]. The mechanisms of action through which GLP-1 RAs impact musculoskeletal health and more specifically bone remodelling are complex. It is well known that weight loss (caloric restriction, surgical by-pass, unintentional bone loss, etc.) might decrease the beneficial effect of weight-bearing on bone formation, inducing a relatively mild but intransient decrease in biochemical markers of bone formation (e.g., Procollagen Type 1 N-terminal propeptide [P1NP]) [85]. This is particularly true when muscle mass is significantly reduced, impairing the beneficial bone-muscle cross-talking [86]. Insulin has been repeatedly shown to have a beneficial effect on bone formation, directly stimulating proliferation and differentiation of osteoblast but also improving the tri-dimensional structure (microarchitecture) of the skeleton – hence, improving its resistance, and acting synergistically with insulin growth factor-1, which is also a potent stimulator of bone formation. However, the impact of GLP-1 RAs on circulating insulin levels may vary according to basal treatment (with or without insulin therapy) and depending on the degree of achieved weight loss. Thus, the effects of GLP-1 RAs more specifically on bone formation are most presumably multi-factorial and complex, which may explain some discrepant results seen in different studies [87, 88].

However, our findings align with previous meta-analyses with clinical results indicating that these biological effects do not translate into measurable improvements in BMD or clinically meaningful fracture reduction in clinical setting [89, 90]. It is important to acknowledge that none of the studies included in our systematic review and meta-analysis had bone outcomes as their primary endpoint; hence, being inappropriately designed (inclusion criteria) and drastically underpowered to demonstrate fracture reduction. Another potential explanation lies in the relatively short duration of most RCTs in the field, which could be insufficient to capture structural BMD and other bone-related changes or fracture outcomes. Additionally, the weight loss induced by GLP-1 RAs, often substantial, could counteract potential anabolic effects on bone through reduced mechanical loading and changes in hormonal balance, particularly in older or metabolically vulnerable individuals [91]. Clinical trials including patients with established osteoporosis or at very high risk of fracture risk and including a fracture endpoint could provide discrepant results.

Taken together with prior scientific literature, our findings support the musculoskeletal safety of GLP-1 RAs with respect to bone health, but do not provide evidence for a clinically relevant osteoprotective effect in the studied populations.

Muscle Outcomes: Consistent Reduction in Lean Mass Accompanying Weight Loss

In contrast to bone outcomes, our meta-analysis demonstrated a consistent and statistically significant reduction in lean body mass/fat-free mass associated with GLP-1 RAs (with a low level of evidence), with a moderate pooled effect size (SMD −0.52). This effect was robust across all sensitivity analyses and appeared mainly driven by studies involving liraglutide- and semaglutide-based interventions, particularly when compared with placebo. This finding should not be interpreted as evidence of a molecule-specific deleterious effect on muscle tissue, but rather as reflecting the fact that these compounds were among the most frequently studied GLP-1 RAs and were associated with substantial weight loss [92, 93]. The substantial heterogeneity observed in the pooled analysis of lean body mass (I² = 88%) warrants careful interpretation. Interestingly, heterogeneity was markedly reduced after stratification according to the type of GLP-1 receptor agonist, suggesting that differences between compounds and treatment regimens may explain an important proportion of the observed variability. While exenatide studies showed virtually no heterogeneity and liraglutide studies only moderate heterogeneity, substantial heterogeneity persisted among semaglutide studies. In this subgroup, two studies reported considerably larger reductions in lean mass than the remaining studies. In particular, the study by Wilding et al, which evaluated the higher obesity dose of semaglutide (2.4 mg/week) over a prolonged treatment period, may reflect the greater magnitude of weight loss achieved with intensive weight-management regimens.

Our findings are in line with recent systematic reviews and meta-analyses showing that GLP-1 RA-induced weight loss was accompanied by a reduction in lean mass, accounting for approximately 15%–25% of total weight loss [4]. Importantly, this proportional loss of lean mass appears comparable to that observed with other weight-loss interventions, including caloric restriction and bariatric surgery, and was substantially smaller than the concomitant reduction in fat mass.

Mechanistically, GLP-1 signalling has been shown in preclinical models to exert neutral or even beneficial effects on myogenesis and mitochondrial function [94]. In addition to mitochondrial adaptations, GLP-1 signalling appears to influence muscle vascularisation. Infusion of GLP-1 acutely increases skeletal muscle microvascular blood volume and perfusion in humans, which expands the endothelial surface area available for nutrient and oxygen delivery to myocytes. Enhanced microvascular recruitment and vasodilation could facilitate improved substrate exchange and metabolic support for muscle tissue [95]. Nevertheless, in clinical setting, the main driver of lean mass reduction seems likely be the negative energy balance induced by appetite suppression and reduced caloric intake [94]. The absence of systematic resistance exercise or protein supplementation in most included RTCs could further exacerbate lean mass loss.

When restricting analysis to studies reporting specifically fat-free mass, no significant reduction was observed, which carried several clinical implications. Since BIA-derived measures are highly sensitive to changes in hydration and extracellular water content, both of which can be affected by GLP-1 RAs through appetite suppression and early weight loss, the observed reductions in lean body mass may mainly reflect shifts in body water rather than actual loss of contractile tissue. The stable preservation of fat-free mass, a more reliable indicator of structural tissues, suggests that GLP-1 RAs may not cause clinically significant tissue wasting, a finding that only more accurate imaging techniques, such as MRI, could confirm.

Nevertheless, while the observed reduction in lean body mass could raise concerns, particularly in older adults or individuals at risk of sarcopenia, it is critical to interpret these findings in context. Few studies assessed muscle strength or physical performance, which are now recognised as more clinically relevant indicators of sarcopenia than muscle mass alone [96]. Thus, whether the observed lean mass loss translates into functional impairment remains largely unknown.

Joint-Related Outcomes: Evidence Beyond Weight-Mediated Effects

Our analyses did not identify a significant improvement in joint-related outcomes, including WOMAC pain, stiffness, or physical function. These findings contrast with recent high-quality evidence from the STEP 9 RCT, which demonstrated clinically meaningful reductions in knee osteoarthritis pain and improvements in physical function with semaglutide in individuals with obesity and moderate-to-severe KOA [16]. This apparent discrepancy could be explained by differences in study design, population selection, and outcome of interest between studies included in the meta-analytical model. Most studies included in our meta-analysis were not designed to evaluate osteoarthritis as a primary outcome and often involved heterogeneous populations without radiographic or symptom confirmation of joint disease. Moreover, joint outcomes were frequently secondary or exploratory endpoints, limiting statistical power. Collectively, these elements highlight the need for future robust and specifically designed studies investigating osteoarthritis-related outcomes under GLP-1 RA therapies.

Importantly, the benefits observed in STEP 9 [16] seem likely to be mediated mainly by weight loss rather than direct disease-modifying effects on joint tissues. This interpretation is supported by evidence demonstrating a strong dose-response relationship between weight reduction and improvements in osteoarthritis symptoms [97].

Intra-articular liraglutide is currently the centre of a large investigational programme based on promising pre-clinical data suggesting a beneficial effect on chondrocytes [98–100] and synovitis [99], hence offering a potential benefit on symptoms and structure modification in KOA.

Recent publications also reported that in patients without osteoarthritis at inclusion, a prescription of GLP-1 RAs increases the five-year incidence of progression to osteoarthritis, at the hip and at the knee, in obese diabetic and non-obese diabetic patients. However, these results were not confirmed in obese non-diabetic patients and no impact on total hip arthroplasty (THA) or total knee arthroplasty (TKA) was observed in this cohort [101]. However, another study, conducted in obese patients, who were propensity score matched for diabetes, reported an increase in the onset of hip and knee osteoarthritis, as well as in conversion to TKA (without significant effect on THA), but the latter was only observed in patients who were not osteoarthritic at inclusion [102]. Conversely, in patients with prevalent osteoarthritis, the rates of conversion to THA or TKA were significantly reduced. Altogether, these results suggest that in patients with no prevalent OA, GLP-1 RA might increase onset, severity and conversion to arthroplasty of the hip or of the knee whereas in patients with prevalent osteoarthritis, a protective effect might be observed. The clinical significance of these findings together with a careful interpretation of the interaction with changes in weight or body composition are needed before clinical relevance can be firmly established.

Clinical and Research Implications

From a clinical perspective, our findings provide confirmation regarding the overall musculoskeletal safety of GLP-1 RAs, particularly with respect to bone health and fracture risk. However, the consistent reduction in lean mass underscores the importance of implementing preventive strategies, such as resistance exercise, adequate protein intake, and individualised risk assessment, when prescribing GLP-1 RAs, especially in older adults or those with pre-existing frailty. Usually, muscle quality and strength are not evaluated but it remains important to monitor physical performance during GLP-1 RA therapy, especially in older adults or individuals experiencing rapid weight loss. Since joint diseases and sarcopenia could be frequently associated, whether the effect of GLP-1 RA therapy on physical function is explained by improvements in muscle function or by clinical improvement in joint symptoms should be determined by using appropriate tests and questionnaires.

From a research perspective, several gaps warrant careful awareness. Future RCTs could, for instance, systematically incorporate functional muscle outcomes, as mentioned, but also longer follow-up durations, stratification by age, sex, baseline body composition, and comorbidity burden but also use more precise imaging technologies such as MRI. Also, recent evidence suggests potential sex-specific responses to GLP-1 signalling, possibly mediated by interactions with oestrogen pathways [103], which currently remain insufficiently explored in musculoskeletal research. Moreover, since GLP-1 RAs are now indicated, in some parts of the world, for the treatment of obesity/overweight, without T2DM, with a different dosage for each indication, it might be of interest to investigate whether the effects on bone/joint/muscle are influenced by the GLP-1 RA doses administered.

Compared with previous systematic reviews and meta-analyses, the present work provides several important additions to the existing literature. Previous evidence syntheses have generally focused on a single musculoskeletal domain, such as bone health, fracture risk, or body composition, and often relied exclusively on RCTs. In contrast, our review simultaneously evaluated bone-, muscle-, and joint-related outcomes and integrated evidence from randomised trials, real-world observational studies, and pharmacovigilance data. This broader approach allowed a more comprehensive assessment of the musculoskeletal safety profile of GLP-1 RAs across a large spectrum of clinically relevant outcomes. Importantly, such an integrated perspective may be particularly relevant in the context of ageing, where bone fragility, loss of muscle mass and function, and joint disorders frequently coexist and interact to influence disability, mobility limitation, falls, fractures, and overall quality of life. By evaluating these domains together rather than in isolation, our findings provide clinicians with a more holistic overview of the potential musculoskeletal consequences of GLP-1 RA therapies, thereby supporting more informed therapeutic decision-making in populations at risk of age-related musculoskeletal decline.

Strengths and Limitations

From a methodological point of view, the strengths of this work include its comprehensive scope, rigorous methodology, and the integration of evidence from good quality RCTs, real-world observational studies, and pharmacovigilance data, allowing a broad and complementary assessment of the musculoskeletal effects of GLP-1 RAs. The application of random-effects models, the systematic use of the most adjusted effect estimates when available, and the extensive series of sensitivity and subgroup analyses further strengthen the robustness and internal consistency of the findings. In addition, the inclusion of multiple GLP-1 RA molecules and a wide range of comparators enhances the external validity and clinical relevance of our findings.

Nevertheless, several limitations must be acknowledged. First, substantial heterogeneity across study populations, outcome definitions, and assessment methods may have influenced the magnitude and precision of the pooled effect estimates, despite the consistency of the overall direction of effects. Second, the duration of follow-up in most RCTs was relatively short, particularly for bone- and joint-related outcomes, limiting the ability to capture long-term structural changes and clinically relevant events such as fractures or disease progression. Moreover, the restricted number of studies included in the meta-analytical model prevented us from running subgroup analyses on the duration of follow-up. Third, references and data on muscle function, physical performance, and muscle quality were sparse and inconsistently reported, which did not allow us a comprehensive evaluation of the functional implications of the observed changes in body composition. Similarly, evidence on joint-related outcomes remained limited and heterogeneous, with most studies not specifically designed to assess osteoarthritis-related endpoints. Fourth, it was not possible to restrict the evidence to T2DM or obesity separately, as many included studies with mixed populations and did not report condition-specific results. Fifth, no subgroup analyses according to study quality were conducted. However, this choice was driven by the already substantial analytical burden of work and, most importantly, by the fact that the majority of the included studies were of high methodological quality. Moreover, the robustness of the findings was further supported by leave-one-out sensitivity analyses, which implicitly accounted for the potential influence of studies with lower methodological quality. Finally, although pharmacovigilance and RWE studies provided valuable information on rare and long-term outcomes, residual confounding and exposure misclassification cannot be excluded in these designs as underscored by our sensitivity analyses. As an additional comment, more than a limitation, because most pooled analyses relied on intention-to-treat estimates whenever available, the observed effects should be interpreted as treatment-strategy effects at the group level rather than pure on-treatment pharmacological effects among fully adherent individuals.

A final comment about limitations – although meta-analytical approaches provide a useful structured overview of increasingly complex literature, all pooled estimates should nevertheless be interpreted cautiously in light of the substantial clinical and methodological heterogeneity across GLP-1 RA compounds, study populations, and outcome definitions.

Conclusion

This meta-analysis is the first to investigate the effects of GLP-1 RAs on a broad spectrum of musculoskeletal outcomes, including bone-, muscle-, and joint-related endpoints. While no adverse effects on bone and joint outcomes were found, potential effects on muscle outcomes (i.e., decreases in lean body mass/fat-free mass) could be suggested. However, the clinical significance of this finding remains uncertain, particularly given the limited evidence available on muscle strength, physical performance, and other functional outcomes.

Taken together, the current evidence supports the overall musculoskeletal safety of GLP-1 RAs but does not allow definitive conclusions regarding their impact on osteoporosis, osteoarthritis, or sarcopenia. Indeed, most available studies were not specifically designed to investigate these conditions and rarely considered musculoskeletal outcomes as primary endpoints. These findings therefore support the need for integrated and multidisciplinary strategies aimed at preserving musculoskeletal health in individuals treated with GLP-1 RAs, particularly those at risk of frailty and age-related musculoskeletal decline. Future studies should be conducted in populations selected on the basis of osteoporosis, osteoarthritis, or sarcopenia and incorporate clinically relevant musculoskeletal outcomes, such as fractures, pain, joint function, mobility disability, physical performance, and the need for joint replacement. Until such dedicated evidence becomes available, any global assessment of the effects of GLP-1 RAs on osteoporosis, osteoarthritis, or sarcopenia should be interpreted with caution.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

The contribution of JYR to this manuscript has been funded by the Distinguished Scientist Fellowship Program (DSFP) of the King Saud University, Riyadh, Kingdom of Saudi Arabia.

Declarations

Conflict of Interest

D.S-R. reports personal speaker fees from Nutricia outside the submitted work. J-Y.R. is President of the European Society for Clinical and Economic Aspects of Osteoporosis, Osteoarthritis and Musculoskeletal Diseases (ESCEO), a not-for-profit organisation receiving Unrestricted Educational Grands from corporate partners. J-Y.R. is an Editorial Board member of Drugs. J-Y.R. was not involved in the selection of peer reviewers for the manuscript nor any of the subsequent editorial decisions. Other authors did not report any conflicts of interest in relation to this work.

Disclaimer

The authors alone are responsible for the views expressed in this article, and they do not necessarily represent the views, decisions, or policies of the institutions with which they are affiliated.

Ethics Approval

Not applicable.

Consent to Participate

Not applicable.

Consent for Publication

Not applicable.

Code Availability

All materials related to this work are freely available on the Open Science Framework deposit – https://osf.io/zbhwc/

Availability of Data and Material

All materials related to this work are freely available on the Open Science Framework deposit – https://osf.io/zbhwc/

Authors’ Contribution

J-Y.R. and J.A.T. are at the initiative of the research question. C.B. and Y.M. are the main investigators of the present research. C.B. drafted the protocol, which was reviewed by A.S. and J-Y.R. C.B. developed the search strategies and identified potential references for inclusion in the project. Y.M. C.B. and D.S.R. screened the references and selected the relevant ones. Conflicts were resolved by C.B. and Y.M. Y.M. performed the manual literature search. Y.M., C.B. and D.S.R., S.v.H. extracted the data, and performed the risk of bias assessment. Meta-analyses models were run by Y.M and C.B. The results were first interpreted by C.B., J-Y.R., A.S. and then by the rest of the experts, including A.M., R.R., and G.D. The first draft of the manuscript was written by C.B. All authors reviewed and approved the final manuscript.