G LP1Ra‐based therapies and DXA‐acquired musculoskeletal health outcomes: a focused meta‐analysis of placebo‐controlled trials

Abstract

Objective

The objective of this study was to evaluate the effect of glucagon‐like peptide‐1 receptor agonist (GLP1Ra)‐based therapies on change in dual‐energy x‐ray absorptiometry (DXA)‐acquired lean mass (LM) or bone mineral density (BMD).

Methods

PubMed and Web of Science were searched from database inception through January 29, 2024, for randomized, placebo‐controlled trials reporting on change in DXA‐acquired LM or BMD measures associated with 12+ weeks of GLP1Ra‐based treatment. Of 2618 articles, 9 trials met prespecified search criteria, with 7 reporting on change in total body LM and 2 reporting on change in BMD. For LM outcomes, a hierarchical Bayesian model was used to estimate treatment mean differences. BMD outcomes were described narratively.

Results

LM was reported in a total of 659 participants (GLP1Ra‐based therapies: n = 419; placebo: n = 240), with follow‐up times ranging from mean (SD) 12 to 72 (33.5) weeks. At baseline, participants were aged mean (SD) 41.7 (7.6) years, and 75% were female, with BMI values ranging from 30 to 43 kg/m². Compared with placebo, GLP1Ra‐based treatment was associated with significantly reduced total body weight (−6.9 kg; 95% credible interval [CI]: −10.7 to −3.0). GLP1Ra‐based treatment was also associated with significantly reduced LM (−1.9 kg; 95% CI: −3.5 to −0.2).

Conclusions

Approximately 30% of body weight lost with GLP1Ra‐based therapy is LM. More data are needed assessing BMD outcomes.

Study Importance.

What is already known?

• Weight loss is accompanied by loss of fat‐free mass, including muscle and bone.

• Glucagon‐like peptide‐1 receptor agonist (GLP1Ra)‐based therapies have garnered scientific, clinical, and public excitement due to their impressive weight‐loss efficacy.

What does this review add?

• In this focused meta‐analysis of 659 participants across seven randomized, placebo‐controlled trials reporting on change in dual‐energy x‐ray absorptiometry‐acquired total body lean mass (LM) associated with 12+ weeks of GLP1Ra‐based treatment, ~30% of total weight loss achieved with GLP1Ra‐based treatment was attributed to LM loss.

• Only two eligible randomized, placebo‐controlled trials reported on change in bone mineral density (BMD) outcomes with GLP1Ra‐based therapy.

How might these results change the direction of research or the focus of clinical practice?

• The field is ripe for studies identifying and evaluating interventions that maximize fat loss while minimizing fat‐free mass loss secondary to GLP1Ra‐based treatment.

• Additional studies are also needed assessing the effects of GLP1Ra‐based therapy on BMD, as well as the clinical relevance of musculoskeletal tissue loss.

INTRODUCTION

The epidemic of obesity continues to negatively impact health in the United States [1, 2]. Significant progress has been made in the pharmacological management of diseases concomitant to excess adiposity [3, 4]; however, obesity itself has historically been treated with lifestyle or surgical‐based therapies. Of late, the next generation of antiobesity medications have garnered scientific, clinical, and public excitement due to their impressive weight‐loss efficacy [5], with glucagon‐like peptide‐1 receptor agonist (GLP1Ra)‐based therapies being a particularly promising class of medications. Indeed, clinical trials have demonstrated substantial weight loss achievable with GLP1Ra‐based therapy, ranging from 12% to 15% over a 1‐year period for semaglutide [6] to weight loss rivaling what is seen with bariatric surgery for tirzepatide [7]. More recently, data showing a reduced risk of cardiovascular events have resulted in an additional Food and Drug Administration (FDA) indication for some GLP1Ra‐based therapies, highlighting the ability of these medications to reduce the risk of clinically relevant outcomes [8].

However, the impact of these medications on the quality of weight being lost is not as well understood. This may be of concern, particularly among older patients in whom the clinical consequences of muscle and bone loss (including incident fracture, disability, and even death) have been well described [9]. A 2019 review, which summarizes the effects of GLP1Ra‐based therapies on body composition (predominantly assessed via dual‐energy x‐ray absorptiometry [DXA]) reports that 20% to 50% of total lost weight is attributed to lean mass (LM), warranting attention and potential intervention [10]. In contrast, a more recent review evaluating the effects of obesity pharmacotherapy, including newer incretin‐memetics, on body composition concludes that a nonsignificant percentage of muscle mass is lost relative to fat mass (FM) [11]. Interestingly, preclinical studies have suggested that GLP1Ra‐based therapies may positively impact bone health by promoting bone formation and reducing bone turnover; however, currently available clinical data on effects of GLP1Ras on human bone metabolism are largely neutral (i.e., no consistent changes are observed in bone turnover biomarkers, bone mineral density [BMD], or fracture risk) [12]. Given the rapid and increasing uptake of GLP1Ra‐based therapies [13], coupled with the knowledge gap surrounding their impact on clinically relevant musculoskeletal outcomes, the purpose of this focused meta‐analysis was to identify and synthesize data from randomized controlled trials (RCTs) reporting on DXA‐acquired change in LM or BMD associated with GLP1Ra‐based treatment compared with placebo.

METHODS

Data sources and searches

This meta‐analysis followed the Preferred Items for Systematic Reviews and Meta‐Analysis (PRISMA) guidelines [14]. The study design protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO; Centre for Reviews and Dissemination [CRD] no. 42023487737) and was designed to identify the highest level of available evidence (i.e., published randomized, placebo‐controlled trials) to observe clinically meaningful weight loss associated with the intervention of interest (i.e., GLP1Ra‐based therapy), in the population of interest (i.e., human adults), measuring change in the outcomes of interest (i.e., DXA‐acquired LM or BMD). All searches and search strategy development were facilitated by a health science reference librarian (Colleen M. Foy). PubMed and the Web of Science Core Collection, including the Science Citation Index Expanded (1945 to present) and the Emerging Sources Citation Index (2005‐present), were searched on January 29, 2024, for literature published prior to the search date using the following terms: “glucagon‐like peptide 1 receptor,” “gastric‐inhibitory polypeptide,” “glucose‐dependent insulinotropic peptide,” “gastric‐inhibitory polypeptide,” “anti‐obesity agent,” “weight‐loss agent,” “incretin effect,” “hypoglycemic agent,” “semaglutide,” “weight loss,” “obesity,” “overweight,” “diabetes mellitus,” “noninsulin‐dependent diabetes mellitus,” “type 2 diabetes,” “body composition,” “dual‐energy x‐ray absorptiometry scan,” “x‐ray photodensitometry,” “adiposity,” “lean body mass,” and “body composition.” Search hedges for “adult,” “human,” and “randomized control trial” were included. The full search strategy and resources used in each database can be found in the online Supporting Information (Appendix A.1).

Study selection

Study selection criteria are fully described in the registered study protocol (PROSPERO; CRD no. 42023487737). Briefly, to limit our analyses to the most influential articles in the field, we restricted our search to peer‐reviewed publications. Identified articles were screened according to the following inclusion criteria: double‐blinded RCTs with GLP1Ra‐based treatments and placebo controls; intervention exposure spanning at least 12 weeks in duration; participants aged ≥18 years at baseline; and inclusion of DXA‐acquired LM or BMD outcomes, presented immediately before and after intervention. Exclusion criteria included open‐label, maintenance (i.e., reporting legacy treatment effects), or phase 2b or animal studies. The Covidence systematic review software platform was used as a citation screening and collaboration tool, as it provides comprehensive training and support resources as a part of its systematic review community. Duplicates and references considered as not reporting on RCTs were removed via the machine‐learning and automation features available in Covidence. Two “yes” votes were required per study in each review phase. Each phase of the review process was facilitated by reviewers with MDs in endocrinology (Tiffany M. Cortes and Fernando Reyes San Martin), reviewers with PhDs in exercise, nutrition, and preventive health (Kristen M. Beavers and Monica C. Serra), and a graduate student in health and exercise science (Lauren Dinkla). First, five reviewers (Kristen M. Beavers, Tiffany M. Cortes, Lauren Dinkla, Fernando Reyes San Martin, and Monica C. Serra) independently screened titles and abstracts of relevant citations. The same reviewers were then accountable for full‐text screening. Throughout both screening phases, inclusion and exclusion criteria were available on‐screen through the Covidence application. Reviewers not responsible for conflicting votes facilitated resolution based on inclusion and exclusion criteria.

Data extraction and quality assessment

Data extraction from relevant studies was facilitated by Microsoft Excel spreadsheet tools and two independent extractors (Kristen M. Beavers and Tiffany M. Cortes), with a third reviewer (Monica C. Serra) enlisted to resolve conflicts and provide consensus when required. The following data were extracted: first author name; publication year; PMID; study name; clinical trial registration number; population characteristics (e.g., age range, body mass index [BMI] range, diabetes status); study aims; intervention characteristics; comparator characteristics; and active and placebo treatment effects on prespecified outcomes, including LM or BMD, as well as total body weight and FM.

The revised Cochrane risk‐of‐bias tool for randomized trials was used to assess the quality of included studies according to the following five domains (D): D1, bias arising from the randomization process; D2, bias due to deviations from intended intervention; D3, bias due to missing outcome data; D4, bias in measurement of the outcome; and D5, bias in selection of the reported result. During the assessment phase, the Cochrane risk‐of‐bias 2.0 guidance document [15] was made available to assist assessors with quality judgments. Each study was classified as having “low,” “high,” or “some” concerns. Two reviewers (Lauren Dinkla and Fernando Reyes San Martin) assessed included studies independently. In the event of conflicting judgments, a third reviewer (Tiffany M. Cortes) facilitated resolution.

Data synthesis and analysis

Primary analyses focused on randomized mean treatment effects compared with placebo on total body weight, LM, and FM from the identified trials. Treatment‐specific means and standard deviations (SD) from each study were extracted when available and derived when necessary. The full code, including starting values and priors, and details regarding derivations, including handling of missing data, are included in Appendix A.4. Treatment effects for each outcome were estimated separately using a hierarchical Bayesian model assuming normally distributed study‐specific intercepts (change from baseline in control) and study‐specific treatment difference, with standard errors based on observed or derived standard deviations of change and sample sizes for each treatment arm. The model used the posterior predictive distribution for missing standard deviations of change to incorporate uncertainty in the resulting posterior estimates; furthermore, heterogeneity was quantified with the I ² statistic using model‐derived estimates of standard deviations when group change standard deviations were not available from the publications. A secondary model was fit with the addition of a main effect and hierarchical interaction term for “indication” (i.e., chronic weight or diabetes management), allowing treatment effects to be handled and presented separately across subgroups. Forest plots summarized the study‐specific data, study‐specific treatment effects and 95% credible intervals (CI), overall treatment means and 95% CI, and overall treatment effects and 95% CI for all three outcomes separately. A sensitivity analysis was performed to ensure the robustness of the results to the possible misspecification of one study's outlying variability estimate. Analyses were conducted using Markov chain Monte Carlo methods in R using JAGS 4.3.2 and the rjags package, and forest plots were constructed using the forestplot package [16]. Funnel plots were constructed using treatment effects and standard error estimates from the model using the package meta. The overall treatment effects were the means and 95% Bayesian CI of the hierarchically weighted treatment effect parameters. Statistical comparisons were deemed significant when the 95% CI excluded the null mean difference of zero; however, all analyses are considered exploratory due to the absence of confirmatory a priori power calculations.

RESULTS

Study selection and quality assessment

A total of 2618 articles were selected through the implementation of search strategies in PubMed and Web of Science. A total of 542 duplicates and 538 ineligible (i.e., non‐RCT) articles were detected and removed using Covidence automation tools, with an additional manual removal of 1 duplicate. A total of 1429 articles were excluded through a comprehensive evaluation of their title and abstract. During the full‐text review, 98 articles were further excluded primarily due to the lack of outcomes and incorrect study design (e.g., non‐RCT, phase 2b, duration <12 weeks). Ultimately, 10 articles (spanning 9 RCTs) were selected for inclusion (Figure 1), with 7 trials reporting on LM only [6, 7, 17, 18, 19, 20, 21] and 2 trials reporting both LM and BMD outcomes (and reported in 3 articles) [22, 23, 24]. Formal analysis of LM trials is presented later in this paper; however, because only two BMD trials were identified, these findings are summarized narratively.

FIGURE 1.

Flowchart of the study identification and selection process. [Color figure can be viewed at wileyonlinelibrary.com]

Formal quality and publication bias assessment was conducted on the LM articles only [6, 7, 17, 18, 19, 20, 21, 22, 24]. Quality judgments of the included articles were visually displayed in the Cochrane risk‐of‐bias tool (Figure 2). Briefly, five articles were judged as having a low risk of bias in every domain [6, 7, 18, 20, 22], and four articles were judged to have some concerns [17, 19, 21, 24], attributed to potential bias arising from the randomization process [17], deviations from intended intervention [19, 21], missing outcome data [17, 19], and measurement of the outcome [21, 24]. To assess potential publication bias, funnel plots were produced (Appendix A.2). All plots demonstrated a high degree of heterogeneity across articles (as reflected in the wide dispersion of points); however, given the bidirectionality of the standard errors, there does not appear to be strong indication of publication bias.

FIGURE 2.

Quality assessment findings using the Cochrane risk‐of‐bias tool. [Color figure can be viewed at wileyonlinelibrary.com]

Study characteristics

Table 1 contains pertinent study details for the nine eligible RCTs that had follow‐up whole‐body LM data collected by DXA scan [6, 7, 17, 18, 19, 20, 21, 22, 24]. Trials were published over a 6‐year period (2017–2023) and reported on LM outcomes in a total of 659 participants (GLP1Ra‐based therapies: n = 419, placebo: n = 240). An average of 73 participants per study (range: 35–160) was reported, with two articles presenting LM outcomes in only a subset of participants from the parent trial [6, 7]. At baseline, participants were aged 41.7 (7.6) years, and 75% were female, with baseline BMI values ranging from 30 to 43 kg/m². Target populations varied widely, with two trials specifically enrolling participants with type 1 diabetes mellitus [19, 20], two enrolling those with polycystic ovary syndrome [17, 21], one enrolling those with schizophrenia [22], and one enrolling those following gastric bypass resulting in poor weight loss [24]. Trial duration also varied widely, ranging from 12 to 72 (33.5) weeks. Five trials evaluated GLP1Ra‐based therapies used at dosages indicated for chronic weight management (i.e., liraglutide 3 mg and semaglutide 2.4 mg) or for the primary purpose of testing weight‐reducing effects (i.e., tirzepatide 5, 10, and 15 mg in people with overweight/obesity without type 2 diabetes) [6, 7, 18, 21, 24], whereas four evaluated GLP1Ra‐based therapies used at dosages indicated primarily for diabetes management [17, 19, 20, 22]. Specifically, six trials treated participants with liraglutide (doses ranging from 1.8–3.0 mg/day) [17, 18, 19, 20, 21, 24], one with semaglutide (2.4 mg/week) [6], one with tirzepatide (doses ranging from 5 to 15 mg/day) [7], and one with exenatide (2 mg/week) [22]. Five of these trials also included a lifestyle component in both the GLP1Ra and placebo groups [6, 7, 18, 21, 24], which typically included behavioral counseling to promote a dietary caloric deficit plus increased physical activity.

TABLE 1.

Characteristics of included RCTs reporting on DXA‐acquired LM and BMD as outcomes.

Author (publication date); study name	Target population	Duration (wk)	Sample size completing DXA (drug:placebo allocation)	Intervention description	Baseline characteristics (mean ± error b )	Intervention outcome
Ishøy (2017) [22]; Eriksson (2019) [23]; TAO study	Antipsychotic‐treated patients with schizophrenia spectrum disorder and BMI ≥ 30 kg/m2	12	40 (1:1)	2.0 mg/wk exenatide	Age a , 37 ± 11 y; Sex a (%F), 45; BMI a , 39.5 ± 3.5 kg/m2; Weight, 118.3 ± 16 kg; LM, 59.9 ± 11.7 kg; Lumbar spine BMD, 1.23 ± 0.19 g/cm2; Total hip BMD, 1.12 ± 0.12 g/cm2; Femoral neck BMD, 1.05 ± 0.13 g/cm2	Weight postintervention, 116.0 ± 16.9 kg; LM postintervention, 59.2 ± 12.5 kg; Lumbar spine BMD postintervention, 1.24 ± 0.19 g/cm2; Total hip BMD postintervention, 1.13 ± 0.11 g/cm2; Femoral neck spine BMD postintervention, 1.04 ± 0.13 g/cm2
				Placebo	Age a , 34 ± 11 y; Sex a (%F), 55; BMI a , 38.6 ± 6.3 kg/m2; Weight, 111.7 ± 18.0 kg; LM, 57.4 ± 7.5 kg; Lumbar spine BMD, 1.13 ± 0.16 g/cm2; Total hip BMD, 1.08 ± 0.11 g/cm2; Femoral neck BMD, 1.00 ± 0.10 g/cm2	Weight postintervention, 109.1 ± 19.0 kg; LM postintervention, 56.4 ± 8.2 kg; Lumbar spine BMD postintervention, 1.12 ± 0.16 g/cm2; Total hip BMD postintervention, 1.08 ± 0.11 g/cm2; Femoral neck spine BMD postintervention, 1.00 ± 0.10 g/cm2
Frøssing (2018) [17]; LIPT study	PCOS and BMI > 25 kg/m2 and/or presence of IR (fasting plasma C‐peptide > 0.6 nmol/L at screening)	26	65 (2:1)	1.8 mg/d liraglutide	Age a , 31 (range, 25–36) y; Sex a (%F), 100; BMI a , 33.3 ± 5.1 kg/m2; Weight, 94.2 ± 15.4 kg; LM, 55.8 ± 8.0 kg	Weight change, −5.2 ± 0.7 kg; LM change, −2.4 ± 0.4 kg
				Placebo	Age a , 26 (range, 25–32) y; Sex a (%F), 100; BMI a , 33.3 ± 4.6 kg/m2; Weight, 91.3 ± 13.6 kg; LM, 56 ± 7.2 kg	Weight change, 0.2 ± 0.9 kg; LM change, 0.1 ± 0.4 kg
Ghanim (2020) [19]	T1DM and BMI ≥ 25 kg/m2	26	64 (1:1)	1.8 mg/d liraglutide	Age a , 47 ± 2 y; Sex a (%F), 65; BMI a , 33.3 ± 1.2 kg/m2; Weight, (mean ± SEM) 94.2 ± 3.1 kg; LM, (mean ± SEM) 50.2 ± 2.1 kg	Weight postintervention, (mean ± SEM) 90.3 ± 3.0 kg; LM postintervention, (mean ± SEM) 51.8 ± 2.2 kg
				Placebo	Age a , 45 ± 3 y; Sex a (%F), 59; BMI a , 29.5 ± 1.3 kg/m2; Weight, (mean ± SEM) 83.5 ± 3.4 kg; LM, (mean ± SEM) 50.1 ± 2.2 kg	Weight postintervention, (mean ± SEM) 83.7 ± 3.5 kg; LM postintervention, (mean ± SEM) 50.1 ± 2.1 kg
Kadouh (2020) [18]	BMI > 30 kg/m2 (or ≥27 kg/m2 and 1 weight‐related complication)	16	35 (1:1)	3.0 mg/d liraglutide + lifestyle	Age a , 42 (IQR, 32, 51) y; Sex a (%F), 95; BMI a , 37.2 (IQR, 33.6, 41.0) kg/m2; Weight, 104.3 (IQR, 88.6, 110.7) kg; LM, 48.9 (IQR, 44.1, 53.8) kg	Weight change, −5.8 (IQR, −6.9, −4.45) kg; LM change, −1.3 (IQR, −1.8, −0.1) kg
				Placebo + lifestyle	Age a , 37 (IQR, 26, 51) y; Sex a (%F), 86; BMI a , 34.6 (IQR, 33.4, 38.9) kg/m2; Weight, 100.8 (IQR, 88.8, 109.9) kg; LM, 48.4 (IQR, 45.7, 52.1) kg	Weight change, −1.0 (IQR, −3.5, 2.5) kg; LM change, −0.7 (IQR, −1.1, 1.1) kg
Wilding (2021) [6]; STEP study	BMI 30–40 kg/m2 (or ≥27 kg/m2 and 1 weight‐related complication)	68	140 (2:1)	2.4 mg/wk semaglutide + lifestyle	Age a , 46 ± 13 y; Sex (%F), 76; BMI, 34.8 ± 3.6 kg/m2; Weight, 105.4 ± 22.1 kg; LM, 52.4 ± 11.6 kg	Weight change, −15.3 (ETD, −12.7; 95% CI: −13.7 to −11.7) kg; LM change, −5.3 (ETD, −3.43; 95% CI: −4.7 to −2.1) kg
				Placebo + lifestyle	Age a , 47 ± 12 y; Sex (%F), 76; BMI, 35.0 ± 3.6 kg/m2; Weight, 105.2 ± 21.5 kg; LM, 51.5 ± 10.8 kg	Weight change, −2.6 (ETD, −12.7; 95% CI: −13.7 to −11.7) kg; LM change, −1.8 (ETD, −3.4; 95% CI: −4.7 to −2.1) kg
Elkind‐Hirsch (2022) [21]	PCOS and BMI > 30 kg/m2	32	67 (2:1)	3 mg/d liraglutide + lifestyle	Age a , (mean ± SEM) 31 ± 1 y; Sex a (%F), 100; BMI a , (mean ± SEM) 41.6 ± 0.9 kg/m2; Weight, (mean ± SEM) 111.0 ± 2.8 kg; LM, (mean ± SEM) 55.0 ± 1.1 kg	Weight postintervention, (mean ± SEM) 104.7 ± 2.9 kg; LM postintervention, (mean ± SEM) 54.1 ± 1.1 kg
				Placebo + lifestyle	Age a , (mean ± SEM) 32.0 ± 1.1 y; Sex a (%F), 100; BMI a , (mean ± SEM) 43.9 ± 1.5 kg/m2; Weight, (mean ± SEM) 119.0 ± 4.7 kg; LM, (mean ± SEM) 58.5 ± 1.8 kg	Weight postintervention, (mean ± SEM) 117.9 ± 5.0 kg; LM postintervention, (mean ± SEM) 58.2 ± 1.9 kg
Jastreboff (2022) [7]; SURMOUNT‐1 study	BMI > 30 kg/m2 (or ≥27 kg/m2 and 1 weight‐related complication)	72	160 (1:1:1:1)	5 mg tirzepatide + lifestyle	Age a , 46 ± 13 y; Sex a (%F), 68; BMI a , 37.4 ± 6.6 kg/m2; Weight, 102.9 ± 20.7 kg; LM, data not presented	Weight change, −15.0% (95% CI: −15.9% to −14.2%); LM change, tirzepatide intervention groups combined, −10.9% (ETD, −8.3%; 95% CI: −10.6% to −6.1%)
				10 mg tirzepatide + lifestyle	Age a , 45 ± 12 y; Sex a (%F), 67; BMI a , 38.2 ± 7.0 kg/m2; Weight, 105.8 ± 23.3 kg; LM, data not presented	Weight change, −19.5% (95% CI: −20.4% to −18.5%)
				15 mg tirzepatide + lifestyle	Age a , 45 ± 12 y; Sex a (%F), 68; BMI a , 38.1 ± 6.7 kg/m2; Weight, 105.6 ± 22.9 kg; LM, data not presented	Weight change, −20.9% (95% CI: −21.8% to −19.9%)
				Placebo + lifestyle	Age a , 44 ± 13 y; Sex a (%F), 68; BMI a , 38.2 ± 6.9 kg/m2; Weight, 104.8 ± 21.4 kg; LM, data not presented	Weight change, −3.1% (95% CI: −4.3% to −1.9%); LM change, −2.6% (ETD, −8.3%; 95% CI: −10.6% to −6.1%)
Schmidt (2022) [20]; Lira Pump Trial	T1DM and BMI > 25 kg/m2	26	41 (1:1)	1.8 mg/d liraglutide	Age a , 54 (range, 37–61) y; Sex a (%F), 68; BMI a , 30.2 ± 2.0 kg/m2; Weight, 85.9 ± 10.0 kg; LM, 53.3 ± 9.5 kg	Weight change, −7.0 (95% CI: −8.5 to −5.5) kg; LM change, −2.5 (95% CI: −3.2 to −1.7) kg
				Placebo	Age a , 45 (range, 34–52) y; Sex a (%F), 68; BMI a , 29 ± 3 kg/m2; Weight, 88.6 ± 12.9 kg; LM, 56.4 ± 12.7 kg	Weight change, −0.3 (95% CI: −1.7 to 1.1) kg; LM change, 0.0 (95% CI: −0.7 to 0.7) kg
Mok (2023) [24]; BARI‐Optimize study	Patients with poor weight loss (<20% of body weight) and a suboptimal GLP‐1 response c 12 mo following primary gastric bypass or sleeve gastrectomy	24	47 (1:1)	3.0 mg/d liraglutide	Age a , 47 ± 11 y; Sex a (%F), 74; BMI a , 41.6 ± 6.9 kg/m2; Weight, 116.1 ± 23.6 kg; LM, 63.7 ± 11.0 kg; Total body BMD, 1.2 ± 0.1 g/cm2	Weight change, −9.5 ± 5.1 kg; LM change, −4.2 ± 3.0 kg; Total body BMD change, −0.01 ± 0.02 g/cm2
				Placebo	Age a , 48 ± 11 y; Sex a (%F), 74; BMI a , 44.6 ± 8.3 kg/m2; Weight, 123.5 ± 24.8 kg; LM, 67.1 ± 13.1 kg; Total body BMD, 1.2 ± 0.1 g/cm2	Weight change, −0.4 ± 3.9 kg; LM change, −1.2 ± 3.3 kg; Total body BMD change, 0.01 ± 0.04 g/cm2

Abbreviations: BMD, bone mineral density; DXA, dual‐energy x‐ray absorptiometry; ETD, estimated treatment difference; F, female; GLP‐1, glucagon‐like peptide‐1; IR, insulin resistance; LM, lean mass; PCOS, polycystic ovary syndrome; RCT, randomized controlled trial; T1DM, type 1 diabetes mellitus.

^a Data from parent study.

^b SD unless otherwise noted.

^c Twofold or less in circulating active GLP‐1 concentrations between 0 and 30 min following a 500‐kcal test meal.

Only two trials reporting BMD outcomes fit our inclusion criteria (Table 1) [23, 24], prohibiting formal data analysis; however, they are descriptively summarized here. Both trials were judged as having a low risk of bias in every domain. In the 3‐month trial conducted by Eriksson et al. [23], 45 patients living with schizophrenia and obesity were randomized to 2.0‐mg once‐weekly exenatide or placebo with total hip, femoral neck, and lumbar spine BMD measured at baseline and at 3 months. After intervention, similar and modest weight loss was observed in both groups (i.e., ~2.3 kg; 2% of baseline weight), with higher lumbar spine BMD reported after exenatide treatment (mean [SD] 1.24 [0.19] g/cm²) compared with placebo (1.12 [0.16] g/cm²; p = 0.02). In comparison, Mok et al. [24] evaluated the effect of 6 months of 3.0‐mg once‐daily liraglutide use on total body BMD among 70 adults who were 1 year out from an unsuccessful metabolic surgery (i.e., <20% body weight loss with a twofold or less increase in circulating active GLP‐1 30 min following a 500‐kcal meal). From baseline to 6 months, a greater reduction in body weight was observed in the 3.0‐mg liraglutide group compared with the placebo group (adjusted mean difference: −9.16 kg; 95% CI: −11.45 to −6.87); however, total body BMD was unchanged (adjusted mean difference: −0.00 g/cm²; 95% CI: −0.02 to 0.02).

Meta‐analysis of weight and DXA‐acquired lean and FM outcomes

Figure 3A–C presents forest plots for change in total body weight, total body LM, and total body FM for included RCTs. Treatment arm standard deviations use posterior expected values of the posterior predictive distributions when values could not be derived [6, 19, 21, 22]. Total body weight changed −8.1 kg (95% CI: −11.9 to −4.1) within the GLP1Ra group compared with −1.2 kg (95% CI: −2.3 to 0.0) in the placebo group. The GLP1Ra group experienced −2.5 kg (95% CI: −4.1 to −0.8) change in LM and −5.4 kg (95% CI: −8.4 to −2.3) change in FM, whereas the placebo group experienced changes of −0.6 kg (95% CI: −1.2 to 0.0) and −0.8 kg (95% CI: −2.0 to 0.3) for LM and FM, respectively. LM loss in the GLP1Ra group constituted 30.8% of total mass lost. Compared with placebo, GLP1Ra‐based treatment was associated with significantly reduced total body weight (treatment effect difference: −6.9 kg; 95% CI: −10.7 to −3.0), representing approximately −7.8% (95% CI: −11.5% to −4.0%) weight change in active treatment, compared with −1.1% (95% CI: −2.2% to −0.0%) change for placebo. Loss was derived mainly from FM (treatment effect difference: −4.6 kg; 95% CI: −7.6 to −1.6). GLP1Ra‐based treatment was also associated with significantly reduced LM (treatment effect difference: −1.9 kg; 95% CI: −3.5 to −0.2), on average, with study‐specific treatment effects ranging from gains of 1.2 kg [19] to losses of 4.5 kg [7]. Across body weight and composition outcomes, the greatest magnitude of weight loss was observed in the SURMOUNT [7] and STEP [6] trials, respectively. Heterogeneity was high (I ²  ≥ 0.84) for all outcomes.

FIGURE 3.

Meta‐analytic results for treatment effect on change in (A) total body weight, (B) lean mass, and (C) fat mass.

Appendix A.3 presents treatment effects on body mass and composition by medication indication. None of the posterior CI indicated significant differences in weight or composition by treatment indication. However, the magnitude of the treatment effects on total body mass was nearly two times greater in trials using GLP1Ra‐based therapies at dosages indicated primarily for chronic weight management [6, 7, 18, 21, 24] compared with trials using dosages indicated primarily for diabetes management [17, 19, 20, 22]: −10.2 kg (95% CI: −16.3 to −3.9) versus −5.5 kg (95% CI: −8.7 to −2.1), respectively. This twofold differential was mirrored in the magnitude of FM (−6.9 kg; 95% CI: −11.5 to −2.1 vs. −3.6 kg; 95% CI: −6.0 to −1.1) and LM (−3.3 kg; 95% CI: −6.2 to −0.4 vs. −1.1 kg; 95% CI: −2.6 to 0.5) losses. Appendix A.4.a presents sensitivity results assuming that the outlying standard deviation of Frøssing et al. [17] was a standard error. Only minor differences in point estimates resulted from this modified assumption; overall conclusions were not impacted.

DISCUSSION

The purpose of this meta‐analysis was to identify and synthesize across published randomized, placebo‐controlled trials reporting on change in DXA‐acquired LM and BMD associated with GLP1Ra‐based treatment. Herein, we report two main findings. First, total body LM loss comprises ~30% of total weight lost with GLP1Ra‐based therapy. Overall, this proportion is on the higher end of what might be expected with lifestyle‐based weight loss and aligns more closely with what is seen with bariatric surgery. With that said, the amount of LM lost during weight loss can be influenced by several factors that we could not fully account for in our meta‐analysis, and interpretation of LM loss must also weigh its impact on relevant health outcomes. Both considerations are discussed later in this paper. Second, there was a surprising lack of published data quantifying the effect of GLP1Ra‐based therapy on BMD outcomes.

The amount of LM lost during weight loss is known to vary widely depending on several factors, including the rate at which weight is lost and the degree of energy and nutrient deficiency, both of which can be influenced by the mode of weight loss, as well as whether exercise is incorporated into the weight‐loss prescription [25]. To illustrate, a 2017 publication by Weiss et al. [26] randomized 50 middle‐aged adults to achieve 6% to 8% weight loss over a 4‐month period via caloric restriction, exercise, or their combination. Despite achieving similar absolute weight loss (~7% of baseline weight), the percentage of weight lost as LM was greatest in the caloric restriction group (2% of baseline LM; 18% of total weight lost), less when exercise was added to caloric restriction (1% of baseline LM; 6% of total weight lost), and virtually nonexistent in the exercise‐only group (<1% of baseline LM; 2% of total weight lost). As with exercise, higher levels of dietary protein during weight loss also attenuate total LM losses, with meta‐analytic data showing that middle‐aged and older adults (i.e., aged 50+ years) who consume higher levels of dietary protein (>1.0 g/kg/day) during weight loss retain more LM comparison with those following normal protein diets (~21% vs. ≥30% of total weight lost as LM) [27]. Although over one‐half of included trials recommended some type of behavioral counseling, the lack of structured countermeasure interventions to mitigate LM loss (e.g., provision of protein supplements, supervised exercise sessions) may explain why the observed percentage of LM loss with GLP1Ra use is on the higher end of what is typically seen with lifestyle‐only obesity treatments; however, this hypothesis has yet to be tested.

In contrast to lifestyle‐based weight loss, bariatric surgery is typically associated with a greater rate and degree of total weight loss, both of which augment the proportion of weight loss‐associated LM loss; thus, the potential for LM loss is higher. A recent meta‐analysis examining body composition changes in ~300 individuals in the 6+ months following metabolic bariatric surgery found that 25% of weight lost comes from the fat‐free compartment [28]. Even then, it is worth noting that among the bariatric procedures there are differences, with biliopancreatic diversion with duodenal switch and Roux‐en‐Y gastric bypass resulting in a greater percentage of fat‐free mass loss than laparoscopic gastric banding [25], likely due to their malabsorptive nature. It is notable that the proportion of LM loss observed with GLP1Ra‐based therapy in this meta‐analysis aligns with bariatric surgery (~25%–30%); however, the amount of total achieved weight loss is significantly less. Although change in percentages, especially with small absolute numbers, can be deceptive, it does call attention to the possibility of additional mechanisms of LM loss, which accompany GLP1Ra‐based therapy, including inadequate nutrient (particularly dietary protein) intake, absorption, or utilization necessary to support LM maintenance [29], as well as effects on insulin signaling, which can affect protein synthesis rate within skeletal muscle [30]. Secondary analyses stratifying treatment effects by indication do not show statistically significantly increased LM loss when GLP1Ras are prescribed at doses indicated for chronic weight management. However, given the small number of studies and the magnitude of the treatment effects, further investigation is warranted as more data become available.

A certain degree of LM loss with weight loss is expected and is even considered adaptive; therefore, determining whether the loss is maladaptive requires evidence of negative impact(s) on health outcomes. Indeed, the same meta‐analysis reporting nearly one‐quarter LM loss secondary to bariatric surgery also reported preservation in muscle strength [28]. Likewise, many studies from our group [31, 32, 33] and others [34, 35] have shown improvements in muscle strength and function in older adults participating in lifestyle‐based weight‐loss interventions, despite experiencing significant weight and LM loss [36, 37]. Functional improvements following lifestyle or surgical weight loss are often attributed to fat loss‐associated improvements in muscle quality [38, 39]. However, clinical concern remains that there is likely a threshold at which LM loss becomes problematic, which may vary by individual [40]. Future trials in this area should continue to monitor treatment‐related change in body composition and assess clinically relevant outcome measures that may be impacted by LM loss [41], including, but not limited to, functional status, and particularly in high‐risk populations. This is especially important as the average weight‐loss effect continues to increase among emerging GLP1Ra‐based treatments [42, 43].

It was unexpected that only two studies meeting our inclusion criteria would be identified that reported on change in DXA‐acquired BMD [23, 24], with one showing no change in whole‐body BMD despite significant weight loss [24], and the other suggesting increased spine (but not total hip or femoral neck) BMD [23]. These limited findings align with preclinical data suggesting that GLP1Ra‐based therapies may positively impact bone health by promoting bone formation and reducing bone turnover [12, 44]. However, the relatively short study durations (i.e., 3–6 months; bone remodeling typically takes a minimum of 6 months) [45] and lack of sensitivity in whole‐body BMD to change could also explain these findings. Future longer‐term (i.e., 6+ months) studies specifically designed to address the effects of GLP1Ra‐based therapies on change in clinically relevant skeletal outcomes are needed to better understand the effect of these medications on bone health. Ideally, this would include incident fracture; however, given the difficulty in achieving adequate power for this outcome, total hip areal BMD [46] and state‐of‐the‐art measures of bone microstructure [47] are reasonable surrogates. Given the well‐described relationship between weight loss and fracture risk, particularly among older adults [48, 49, 50], monitoring clinically relevant bone endpoints in future GLP1Ra‐based therapy trials seems prudent.

A final consideration when assessing the impact of GLP1Ra‐based therapy on musculoskeletal health is that, as with lifestyle [51] and surgical interventions [52], weight regain following cessation of treatment is an expected occurrence [53]. At present, there is no limit as to how long these medications can be prescribed; however, outside factors such as supply chain issues [54] and cost [55] can influence their long‐term use. Indeed, recent data on the prevalence of GLP1Ra discontinuation suggest that one‐half of individuals living with obesity will stop taking these medications 1 year following initiation of treatment [54]. When barrier‐to‐access scenarios arise, clinicians often try to exchange a patient's current medication with a similarly effective one; however, this is not always possible due to availability or insurance coverage and can result in treatment lapses with concomitant weight regain or glycemic control deterioration [56, 57]. Because some long‐term data from lifestyle‐based weight‐loss interventions suggest that LM [58] and BMD [59, 60] may not be recovered with weight regain, better understanding of the effects of intermittent use of GLP1Ra‐based therapies on long‐term change in body composition represents another important area of further inquiry.

Strengths of this study include rigorous study selection criteria, which limited the meta‐analysis to only published randomized, placebo‐controlled trials; use of the Covidence systematic review software platform to reduce reviewer error and bias when screening and extracting data; limited evidence of risk of bias in included studies; and use of Bayesian hierarchical modeling approach, which better accommodates study‐specific heterogeneity and missing data than a traditional frequentist or fixed‐effects approach. Although selected due to clinical utility, focus on DXA‐acquired musculoskeletal outcome measures limits the scope and inferential ability of our work. In particular, we recognize that DXA‐acquired LM includes connective tissue, skin, organs, and water, in addition to the true outcome of interest (muscle mass). Many other ways to assess musculoskeletal health exist, and we implore future work to specifically include metrics of muscle quality and function to complement and extend DXA‐acquired LM. Moreover, the “GLP1Ra‐based therapy” intervention category collapsed across several medications with differing indications, dosing regimens, and effectiveness on weight‐loss outcomes. As GLP1Ra‐based therapies continue to evolve, and more studies become available, it would be of interest to parse out the effect of specific GLP1Ra medications on musculoskeletal health outcomes and determine whether effects vary by patient characteristics (e.g., baseline BMI status). Finally, we acknowledge search protocol limitations in this review. First, the exclusive use of PubMed and Web of Science databases may have omitted relevant studies. Second, although Covidence machine‐learning and automation features have been reviewed as having a very low and acceptable risk of missing duplicates and references considered as not reporting on RCTs [61], we acknowledge the limitation of relying on those features. Finally, as DXA‐related terms often appeared exclusively in supplementary materials of relevant studies, full‐text database searching for these terms imposed an inherent limitation and possible exclusion of relevant studies.

In conclusion, results from this meta‐analysis that focused on DXA‐acquired outcomes show ~30% of total weight lost achieved with GLP1Ra‐based treatment is attributed to LM. More data are needed to assess the effects of GLP1Ra‐based therapy on BMD, as well as the clinical relevance of musculoskeletal tissue loss. As with all medical decision‐making, it is important for individuals who are considering these weight‐loss methods to consult with health care professionals to determine the most appropriate approach based on their needs and circumstances. The field is ripe for studies identifying and evaluating interventions that maximize fat loss while minimizing fat‐free mass loss. Given their demonstrated effectiveness in lifestyle and surgical‐based weight‐loss trials, this may entail prescribing exercise and/or protein supplements alongside GLP1Ra‐based therapies or identifying alternate or adjuvant pharmacologic approaches capable of reducing FM while maintaining LM [62] and BMD [63].

AUTHOR CONTRIBUTIONS

Kristen M. Beavers: conceptualization, data curation, writing of original draft, review and editing, supervision; Jamy D. Ard: writing, review and editing; Daniel P. Beavers: methodology, software, formal analysis, data curation, writing of original draft, review and editing, visualization; Tiffany M. Cortes: data curation, writing of original draft, review and editing; Lauren Dinkla: data curation, writing of original draft, review and editing; Colleen M. Foy: methodology, software, validation, resources, writing of original draft, visualization; Fernando Reyes San Martin: data curation, writing of original draft, review and editing; and Monica C. Serra: data curation, writing of original draft, review and editing.

FUNDING INFORMATION

This work was supported by the National Institute on Aging (grant no. R01 AG059186: Kristen M. Beavers and Daniel P. Beavers; R01 AG074979: Kristen M. Beavers, Daniel P. Beavers, and Jamy D. Ard; P30 AG21332: Kristen M. Beavers; and P30AG044271: Tiffany M. Cortes), the National Institute of Arthritis and Musculoskeletal and Skin Diseases (grant no. U01 AR080969: Kristen M. Beavers, Jamy D. Ard, and Daniel P. Beavers), and the San Antonio Department of Veterans Affairs Geriatric Research Education and Clinical Center (Tiffany M. Cortes).

CONFLICT OF INTEREST STATEMENT

Jamy D. Ard declares grant/contracts from Nestlé Healthcare Nutrition; Eli Lilly and Company; Boehringer Ingelheim; Epitomee, Inc.; UnitedHealth Group R & D; KVK Tech; and WW International, Inc., as well as consulting fees from Nestlé Healthcare Nutrition; Eli Lilly and Company; OptumLabs R & D; Novo Nordisk A/S; Spoke Health, Inc.; Intuitive Health; Regeneron Pharmaceuticals; BrightSeed Bioactives; Level2; and WW International, Inc. The other authors declared no conflicts of interest.

Supporting information

Data S1. Supporting Information.

Beavers KM, Cortes TM, Foy CM, et al. G LP1Ra‐based therapies and DXA‐acquired musculoskeletal health outcomes: a focused meta‐analysis of placebo‐controlled trials. Obesity (Silver Spring). 2025;33(2):225‐237. doi: 10.1002/oby.24172