Osteoporosis International (2018) 29:2639–2644

https://doi.org/10.1007/s00198-018-4649-8

ORIGINAL ARTICLE

Glucagon-like peptide-1 receptor agonists and fracture risk: a network

meta-analysis of randomized clinical trials
Y. S. Zhang 1 & W. Y. Weng 1 & B. C. Xie 1 & Y. Meng 1 & Y. H. Hao 1 & Y. M. Liang 1 & Z. K. Zhou 1

Received: 18 April 2018 / Accepted: 19 July 2018 / Published online: 6 August 2018
# International Osteoporosis Foundation and National Osteoporosis Foundation 2018

Abstract
Summary Our network meta-analysis analyzed the effects of glucagon-like peptide-1 receptor agonists (GLP-1 RAs) on fracture
risk. By combining data from randomized controlled trials, we found that GLP-1 RAs were associated with a decreased bone
fracture risk, and exenatide is the best option agent with regard to the risk of fracture. This study is registered with PROSPERO
(CRD42018094433).
Introduction Data on the effects of GLP-1 RAs on fracture risk are conflicted. This study aimed to analyze the available evidence
on the effects of GLP-1 RAs on fracture risk in type 2 diabetes mellitus patients.
Methods Electronic databases were searched for relevant published articles, and unpublished studies presented at ClinicalTrials.gov
were searched for relevant clinical data. All analyses were performed with STATA 12.0 and R software (Version 3.4.4). We
estimated the risk ratio (RR) and 95% confidence interval (CI) by combining RRs for fracture effects of included trials.
Results There were 54 eligible random control trials (RCTs) with 49,602 participants, including 28,353 patients treated with
GLP-1 RAs. Relative to placebo, exenatide (RR, 0.17; 95% CI 0.03–0.67) was associated with lowest risk of fracture among
other GLP-1 RAs. Exenatide had the highest probability to be the safest option with regard to the risk of fracture (0.07 ‰),
followed by dulaglutide (1.04%), liraglutide (1.39%), albiglutide (5.61%), lixisenatide (8.07%), and semaglutide (18.72%). A
statistically significant inconsistency was observed in some comparisons.
Conclusion The Bayesian network meta-analysis suggests that GLP-1 RAs were associated with a decreased bone fracture risk
compared to users of placebo or other anti-hyperglycemic drugs in type 2 diabetes mellitus patients, and exenatide is the best
option agent with regard to the risk of fracture.

Keywords Fracture . Glucagon-like peptide-1 receptor agonists . Network meta-analysis . Type 2 diabetes mellitus

Introduction In recent years, glucagon-like peptide-1 receptor agonists

Diabetes is a major public health problem; type 2 diabetes
mellitus (T2DM) represents 90% of all cases of diabetes [1].
GLP-1 plays an important role in effectively regulating blood
glucose, by suppressing appetite and decelerating gastric,
emptying and inducing satiety [2].
Y. S. Zhang, W. Y. Weng and B. C. Xie contributed equally to this work.
Electronic supplementary material The online version of this article
(https://doi.org/10.1007/s00198-018-4649-8) contains supplementary
material, which is available to authorized users.
* Z. K. Zhou
zhikunzhou@126.com
1 decreasing calcium uptake in osteoblast cultures, while in-
Department of Pharmacy, Guangdong Medical University, No. 1,
Xincheng Dadao, Songshan Lake Science and Technology Industry
Park, Dongguan, 523808 China
(GLP-1 RAs), a new class of anti-diabetic agents based on
incretin therapy, can stimulate insulin secretion, improve in-
sulin resistance, and slow gastrointestinal motility [3].
Exenatide, one of the GLP-1 RAs, has been reported a bene-
ficial on bone [4].
Cumulative evidence indicates T2DM is associated with an
increased risk of bone fracture [5]. The factors leading to
fracture in patients with T2DM were reported: accumulation
of advanced glycation end products (AGEs) in diabetic bone
collagen contributes to reduced material properties and greater
susceptibility to fracture. And fracture risk was increased by
the higher levels of circulating AGEs. Other important factors
playing an important role are acute and chronic hyperglycemia
downregulating osteocalcin (OCN) expression and activity,
creasing PPAR-γ2 expression [6], but the consequences of
such T2DM complications are seriously dramatic, often
2640
requiring surgery and probably resulting in deformities, dis-
abilities, or even death [7, 8]. It is urgent and crucial for anti-
diabetes drugs to at least not increase the risk of bone fracture.
As a new and validated statistical methodology, network
meta-analysis was allowed to compare the safety of all treat-
ments, especially when these treatments have not been com-
pared head to head [9]. For instance, an indirect estimate of the
safety of exenatide to liraglutide can be achieved by obtaining
information from one trial comparing exenatide with placebo
and another trial comparing liraglutide with placebo.
Therefore, a Bayesian network meta-analysis was per-
formed to evaluate the risk of fracture among these
glucagon-like peptide-1 receptor agonists which was ap-
proved by the US FDA.
Methods
This network meta-analysis has been registered on the
PROSPERO website https://www.crd.york.ac.uk/
PROSPERO/ of which the registration number was
CRD42018094433. Detailed information can be found on
the website.
Search strategy and selection criteria
An extensive search of Medline, EMBASE, and Cochrane
Central Register of Controlled Trials was performed by two
of the investigators. Data were collected on all randomized
control trials in humans up to March 2018. To reach a con-
sensus, a divergence of views was resolved by joint review of
the literature. The search terms used were as follows: GLP-1,
glucagon-like peptide-1, semaglutide, exenatide, liraglutide,
lixisenatide, albiglutide, dulaglutide. The results of unpub-
lished data were identified through a search of the www.
clinicaltrials.gov/ website.
Study selection
We included studies if (1) randomized control trials in type 2
diabetes patients; (2) duration of at least 26 weeks; (3) the
intervention was with GLP-1 receptor agonists, including
semaglutide, exenatide, liraglutide, lixisenatide, albiglutide,
and dulaglutide; (4) comparators were GLP-1 receptor ago-
nists, other hypoglycemic drugs, or placebo; and (5) data on
bone fracture was available. We excluded studies with zero
events in both groups because these comparisons provide no
information on treatment effect.
Data extraction and quality assessment
The data, consisting of the first author’s name, registry num-
ber, sample size, types of GLP-1 RAs and comparators, length
Osteoporos Int (2018) 29:2639–2644
of follow-up, mean age, baseline level of HbA1c, number of
treatment groups, and reported outcomes, were extracted. Two
investigators (WY Weng and BC Xie) extracted data indepen-
dently, in duplicate. The Cochrane risk-of-bias tool was used
to assess the risk of bias for RCTs [10]. Any discrepancies
were resolved by a senior investigator (ZK Zhou).
Data analysis
To account for heterogeneity between studies [11, 12], we used
the random effects model for the Bayesian network meta-
analysis to evaluate the relative risk ratios (RR) of GLP-1
RAs and other anti-diabetes on bone fracture, allowing both
direct and indirect evidence to be taken into account simulta-
neously. RR with its 95% confidence intervals (CI) for bone
fracture was summarized. The analysis was done by using the
Markov chain Monte Carlo methods [13]. Four chains were
suitable, yielding 400,000 iterations (100,000 per chain) gen-
erating the posterior distributions of the model parameters.
Convergence was checked by using the Brooks-Gelman-
Rubin diagnostic. Heterogeneity was assessed by using
the I2 statistic [14]; the goodness of fit of the model was
assessed by calculating residual deviance [15]. And the
presence of inconsistency was evaluated by applying
Node-splitting method [16]; inconsistency was defined as
disagreement between direct and indirect evidence with a p
value > 0.05. In addition, rank probabilities were calculated
to evaluate the probability of each GLP-1 RAs being the
best in terms of the outcome.
We performed a meta-regression analysis to explore the
association between fracture risk and GLP-1 RAs concomi-
tantly treated with or without other anti-diabetic medications
[17]. We performed sensitivity analysis to determine the influ-
ence of trials. A comparison-adjusted funnel plot was created
to detect the presence of any publication bias [18]. All analy-
ses were performed with STATA 12.0 (STATA Corp, College
Station, TX, USA) and R software (Version 3.4.4; R
Foundation for Statistical Computing, Vienna, Austria). p val-
ue < 0.05 was considered as statistically significant [19].
Result
Study characteristics and quality
A total of 54 random control trials were included in the final
analysis, including 47 clinical trials from journals and 7 clin-
ical trials are available from www.clinicaltrials.gov/. Flow
chart of trials selection was shown in (Fig. 1). The
characteristics of the included studies and their quality
assessment are shown in Supplemental Table 1 and Table 2.
Individual specific and non-specific bone fractures were listed
in Supplemental Table 3. The options tested were as follows in
Osteoporos Int (2018) 29:2639–2644
Fig. 1 Trial flow diagram
the network: semaglutide, exenatide, liraglutide, lixisenatide,
albiglutide, dulaglutide, placebo, and Bother hypoglycemic
drugs^. Hypoglycemic drugs included insulin, metformin,
glimepiride, sitagliptin, and sulfonylurea. All these hypogly-
cemic agents were combined into one Bother hypoglycemic
drugs^ to make a connected network plot (Fig. 2).
Fig. 2 Network plot for the
Bayesian network meta-analyses
2641
Statistical analysis
The goodness of fit of the model was acceptable (ratio 1.15,
I2 = 14%) by calculating residual deviance.
The results of network meta-analysis are shown in (Table 1).
Relative to placebo, a significant decreased risk of fracture was
2642 Osteoporos Int (2018) 29:2639–2644

Table 1 Estimated relative treatment effects as risk ratios (RRs) and its corresponding 95% confidence intervals (CIs)

Treatment Semaglutide Exenatide Liraglutide Lixisenatide Albiglutide Dulaglutide Placebo

Semaglutide NA 2.76 (0.2, 33.31) 1.95 (0.08, 18.97) 1.18 (0.08, 20.52) 1.49 (0.08, 24.05) 1.5 (0.11, 17.7) 0.47 (0.04, 3.63)
Exenatide 0.36 (0.03, 4.99) NA 0.35 (0.07, 1.41) 0.44 (0.05, 4.24) 0.54 (0.07, 4.51) 0.54 (0.11, 2.74) 0.17 (0.03, 0.67)
Liraglutide 1.06 (0.11, 12.26) 2.87 (0.71, 13.52) NA 1.24 (0.22, 9.76) 1.55 (0.26, 10.89) 1.56 (0.42, 6.6) 0.49 (0.18, 1.13)
Lixisenatide 0.84 (0.05, 12.86) 2.3 (0.24, 18.9) 1.81 (0.1, 14.56) NA 1.24 (0.1, 14.41) 1.25 (0.13, 9.57) 0.39 (0.05, 1.79)
Albiglutide 0.67 (0.04, 11.9) 1.86 (0.22, 14.51) 0.64 (0.09, 3.79) 0.81 (0.07, 10.47) NA 1.01 (0.13, 7.48) 0.31 (0.04, 1.77)
Dulaglutide 0.67 (0.06, 8.86) 1.84 (0.36, 9.48) 0.64 (0.15, 2.4) 0.8 (0.18, 7.46) 0.99 (0.13, 7.8) NA 0.31 (0.06, 1.33)
Placebo 2.15 (0.28, 25.89) 5.96 (1.49, 31.36) 2.06 (0.89, 5.7) 2.57 (0.56, 19.72) 3.2 (0.57, 25.35) 3.21 (0.89, 15.49) NA

Comparisons should be read from left to right. The estimate is located at the intersection of the treatments in the column heads and the treatments in the
row heads. An RR value > 1 favors the column-defining treatment. An RR value < 1 favors the row-defining treatment; NA not applicable

seen for exenatide (RR, 0.17; 95% CI 0.03–0.67).
Whereas, semaglutide (RR 0.47; 95% CI 0.04–3.63),
liraglutide (RR 0.49; 95% CI 0.18–1.13), lixisenatide
(RR 0.39; 95% CI 0.05–1.79), albiglutide (RR 0.31; 95%
CI 0.04–1.77), or dulaglutide (RR 0.31; 95% CI 0.06–
1.33) were comparable; although the difference was not
significant.
When compared with exenatide, fracture risk was non-
significantly higher for semaglutide (RR 2.76; 95% CI 0.2–
33.31), liraglutide (RR 2.87; 95% CI 0.71–13.52), lixisenatide
(RR 2.3; 95% CI 0.24–18.9), albiglutide (RR 1.86; 95% CI
0.22–14.51), and dulaglutide (RR 1.84; 95% CI 0.36–9.48).
The pooled results favored exenatide (RR 0.35; 95% CI
0.07–1.41), albiglutide (RR 0.64; 95% CI 0.09–3.79), and
dulaglutide (RR 0.64; 95% CI 0.15–2.4) compared with
liraglutide. Relative to lixisenatide, the pooled results favored
exenatide (RR 0.44; 95% CI 0.05–4.24), albiglutide (RR 0.81;
95% CI 0.07–10.47), and dulaglutide (RR 0.8; 95% CI 0.18–
7.46).
The results of the sensitivity analysis were mostly similar to
the results of the main analysis (Supplemental Table 4).
Ranking probability
The probability ranking of GLP-1 RAs is shown in
Supplemental Table 5; exenatide had the highest probability
to be the safest option with regard to the risk of fracture
(0.07‰), followed by dulaglutide (1.04%), liraglutide
(1.39%), albiglutide (5.61%), lixisenatide (8.07%), and
semaglutide (18.72%).
Funnel plot and publication bias
The comparison-adjusted funnel plots of the network meta-
analysis were not suggestive of any publication bias (Fig. 3).

Inconsistence and heterogeneity check

A statistically significant inconsistency was observed in some

comparisons (Supplemental Table 6), including semaglutide
vs. placebo (p < 0.05) in the indirect comparison; and
exenatide vs. liraglutide (p < 0.05); albiglutide vs. liraglutide
(p < 0.05); lixisenatide vs. placebo (p < 0.05); and albiglutide
vs. placebo (p < 0.05) in the direct comparison. The global
heterogeneity parameter I2 values were 40% (Supplemental
Table 7).

Meta-regression and sensitivity analysis

No significant difference effect was noted in the meta-

regression analysis between fracture risk and GLP-1
RAs concomitantly treated with or without other anti-
diabetic medications (regression coefficient 0.56; CI
0.37–1.56). Fig. 3 Comparison-adjusted funnel plots of the network
Osteoporos Int (2018) 29:2639–2644
Discussion
Previous meta-analysis has not demonstrated adequately the
effects of GLP-1 RAs on fracture risk in patients with T2DM
due to its limitations. Therefore, we performed a Bayesian
network meta-analysis to provide a summary of current data.
Analysis of 54 RCTs indicated that GLP-1 RAs were asso-
ciated with a decreased bone fracture risk compared with other
anti-diabetic medications or placebo in patients with T2DM.
And we provided a probability of treatment rankings for ther-
apeutic strategies with respect to the fracture risk. The results
demonstrated that exenatide was associated with a lower risk
of fracture and had the highest probability to be the safest
option with regard to the risk of fracture. Unfortunately, there
is no pathophysiological interpretation regarding the de-
creased fracture risk among T2DM patients following treat-
ment with GLP-1 RAs; the results of non-GLP-1RAs and
fracture risk were irrelevant to clinical practice and may con-
fuse clinicians in using GLP-1 RAs clinically, because not all
other types of anti-diabetes treatment are considered having
the same effect on fracture risk. Therefore, the data about non-
GLP-1RAs and fracture risk should be neglected.
A well-established epidemiologic link was noted between
bone fracture and T2DM. Relative to the patients with diabe-
tes, those without diabetes were hardly to develop chronic
bone fracture [4]. Therefore, we need to control the risk factors
of fracture emphatically in the therapeutic strategies of T2DM.
Although one meta-analysis including 28 clinical trials in-
dicated that the use of GLP-1 RAs does not decrease the
fracture risk in patients with T2DM, another meta-analysis
including 16 RCTs demonstrated exenatide treatment was as-
sociated with an elevated risk of incident bone fractures; their
study was limited due to the number of trials [20, 21].
Moreover, the poor estimate of inter-comparisons between
GLP-1RA drugs could not be obtained in the study because
it failed to integrate all evidence into a single analysis. Their
results were not consistent with the result of our Bayesian
network meta-analysis.
In this study, we did integrate all evidence into a single
analysis with a Bayesian network meta-analysis to evaluate
the fracture risk of the GLP-1RAs; more statistical precision
was gained compared with previous meta-analysis by describ-
ing on indirect comparisons across trials [22]. In addition, our
network meta-analysis allowed a unified analysis of all avail-
able evidence to compare each GLP-1 RAs head to head,
which has rarely or never been done in a randomized control
trial.
Notably, there are 30 RCTs included patients who were
concomitantly treated with other anti-hyperglycemic drugs.
But the meta-regression analysis indicated that it was not ob-
viously influenced by the association between fracture risk
and GLP-1 RAs concomitantly treated with or without other
anti-hyperglycemic drugs.
2643
There are several limitations of this network meta-analysis
that should be considered. Firstly, although random effect
model was used to reduce the influence from the constraint
of common variance among trials, different drug doses might
be to contribute heterogeneity; we could not perform sub-
group analysis or meta-regression analysis to evaluate it due
to the extreme complexity of different doses of type of GLP-1
RAs. Secondly, a statistically significant inconsistency was
observed in some comparisons, including semaglutide vs. pla-
cebo; exenatide vs. liraglutide; albiglutide vs. liraglutide;
lixisenatide vs. placebo; and albiglutide vs. placebo; the in-
consistency may be caused by the extremely low incidence of
bone fractures [23]. Finally, we might have forgotten some
randomized control trials if these studies had not been regis-
tered in Medline, EMBASE, and Cochrane library, or were not
available from www.clinicaltrials.gov/.
Overall, evidence suggests that GLP-1 RA treatment aver-
aged over different studies to some extent prevents fractures
compared to users of placebo or other anti-hyperglycemic
drugs in type 2 diabetes mellitus patients, and that exenatide
seems to be the best option agent with regard to the risk of
fracture. Nevertheless, our results should be interpreted with
caution, since some limitations are present. Hence, large RCTs
with pre-specified, well-defined safety outcomes are warrant-
ed to assess the impact of GLP-1 ARs on fracture outcomes.
Funding information The present study was supported by grants from the
National Natural Science Foundation of China (grant no. 81774344).
Compliance with ethical standards
Conflicts of interest None.
References
1. Unnikrishnan R, Mohan V (2016) Diabetes in the tropics: preva-
lent, increasing and a major public health problem. Trans R Soc
Trop Med Hyg 110:263–264
2. Drucker DJ (2018) Mechanisms of action and therapeutic applica-
tion of glucagon-like peptide-1. Cell Metab 27:740–756
3. Gallwitz B (2015) GLP-1 receptor agonists in type 2 diabetes and
beyond - new insights 2015. Eur Endocrinol 11:21–25
4. Mabilleau G, Gobron B, Bouvard B, Chappard D (2018) Incretin-
based therapy for the treatment of bone fragility in diabetes
mellitus. Peptides 100:108–113
5. Goldshtein I, Nguyen AM, DePapp AE, Ish-Shalom S, Chandler
JM, Chodick G, Shalev V (2018) Epidemiology and correlates of
osteoporotic fractures among type 2 diabetic patients. Arch
Osteoporos 13:15
6. Epstein S, Defeudis G, Manfrini S, Napoli N, Pozzilli P (2016)
Diabetes and disordered bone metabolism (diabetic osteodystrophy):
time for recognition. Osteoporos Int 27:1931–1951
7. de Liefde II, van der Klift M, de Laet CE, van Daele PL, Hofman
A, Pols HA (2005) Bone mineral density and fracture risk in
type-2 diabetes mellitus: the Rotterdam Study. Osteoporos Int
16:1713–1720
2644
8. Schwartz AV, Vittinghoff E, Bauer DC, Hillier TA, Strotmeyer ES,
Ensrud KE, Donaldson MG, Cauley JA, Harris TB, Koster A,
Womack CR, Palermo L, Black DM (2011) Association of BMD
and FRAX score with risk of fracture in older adults with type 2
diabetes. JAMA 305:2184–2192
9. Faltinsen EG, Storebo OJ, Jakobsen JC, Boesen K, Lange T, Gluud
C (2018) Network meta-analysis: the highest level of medical evi-
dence? BMJ Evid Based Med 23:56–59
10. Higgins JP, Altman DG, Gotzsche PC, Juni P, Moher D, Oxman
AD, Savovic J, Schulz KF, Weeks L, Sterne JA (2011) The
Cochrane Collaboration’s tool for assessing risk of bias in
randomised trials. BMJ 343:d5928
11. Dias S, Sutton AJ, Ades AE, Welton NJ (2013) Evidence synthesis
for decision making 2: a generalized linear modeling framework for
pairwise and network meta-analysis of randomized controlled trials.
Med Decis Mak 33:607–617
12. Mills EJ, Thorlund K, Ioannidis JP (2013) Demystifying trial net-
works and network meta-analysis. BMJ 346:f2914
13. Mawdsley D, Bennetts M, Dias S, Boucher M, Welton NJ (2016)
Model-based network meta-analysis: a framework for evidence
synthesis of clinical trial data. CPT Pharmacometrics Syst
Pharmacol 5:393–401
14. Dias S, Welton NJ, Caldwell DM, Ades AE (2010) Checking con-
sistency in mixed treatment comparison meta-analysis. Stat Med
29:932–944
15. Cooper NJ, Sutton AJ, Lu G, Khunti K (2006) Mixed comparison
of stroke prevention treatments in individuals with nonrheumatic
atrial fibrillation. Arch Intern Med 166:1269–1275
Osteoporos Int (2018) 29:2639–2644
16. van Valkenhoef G, Dias S, Ades AE, Welton NJ (2016) Automated
generation of node-splitting models for assessment of inconsistency
in network meta-analysis. Res Synth Methods 7:80–93
17. Dias S, Sutton AJ, Welton NJ, Ades AE (2013) Evidence synthesis
for decision making 3: heterogeneity–subgroups, meta-regression,
bias, and bias-adjustment. Med Decis Mak 33:618–640
18. Chaimani A, Higgins JP, Mavridis D, Spyridonos P, Salanti G
(2013) Graphical tools for network meta-analysis in STATA.
PLoS One 8:e76654
19. Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie
D, Moher D, Becker BJ, Sipe TA, Thacker SB (2000) Meta-
analysis of observational studies in epidemiology: a proposal for
reporting. Meta-analysis Of Observational Studies in Epidemiology
(MOOSE) group. JAMA 283:2008–2012
20. Mabilleau G, Mieczkowska A, Chappard D (2014) Use of
glucagon-like peptide-1 receptor agonists and bone fractures: a
meta-analysis of randomized clinical trials. J Diabetes 6:260–266
21. Su B, Sheng H, Zhang M, Bu L, Yang P, Li L, Li F, Sheng C, Han Y,
Qu S, Wang J (2015) Risk of bone fractures associated with
glucagon-like peptide-1 receptor agonists' treatment: a meta-
analysis of randomized controlled trials. Endocrine 48:107–115
22. Zeng I, Lumley T (2018) Review of statistical learning methods in
integrated omics studies (an integrated information science). Bioinf
Biol Insights 12:1216874372
23. Gunhan BK, Friede T, Held L (2017) A design-by-treatment inter-
action model for network meta-analysis and meta-regression with
integrated nested Laplace approximations. Res Synth Methods 9:
179–194