Anti-Hip Fracture Efficacy of Bisphosphonates:
A Bayesian Analysis of Clinical Trials
Nguyen D Nguyen, John A Eisman, and Tuan V Nguyen
ABSTRACT: In postmenopausal women, the efficacy of bisphosphonates on hip fracture risk is not clear. This
Bayesian meta-analysis quantitatively reviewed data from 12 randomized clinical trials with 18,667 patients
and found that bisphosphonate treatment was associated with a reduced risk for hip fracture by 42%.
Introduction: The efficacy of antiresorptive bisphosphonates therapy on reducing hip fracture is not clear,
because evidence from randomized clinical trials (RCTs) is inconclusive. This study was undertaken to quan-
titatively assess the effect of bisphosphonates on hip fracture using literature review and meta-analysis.
Materials and Methods: Bayesian methods of meta-analysis were applied to synthesize data from 12 RCTs
available between 1990 and 2004. The trials involved 18,667 postmenopausal women with low BMD or
osteoporosis who have been followed or treated for between 1 and 4 years. The medications used were
etidronate (two trials) alendronate (six trials), risedronate (three trials), and clodronate (one trial). The
primary endpoint was the incidence of hip fracture.
Results: When data from all 12 studies were pooled, treatment with bisphosphonates was associated with a
reduced risk for hip fracture by 42% (relative risk [RR], 0.58; 95% credible interval [CrI], 0.42–0.80). The
absolute rate reduction was 52 hip fractures per 10,000 women (95% CrI, 4–110) for a period of 3-year
treatment. The probability that bisphosphonates are better than placebo (in reducing hip fracture risk by at
least 30%) was 0.90.
Conclusions: In postmenopausal women with osteoporosis or low BMD, bisphosphonate treatment is asso-
ciated with reduced risk of hip fracture.
J Bone Miner Res 2006;21:340–349. Published online on September 6, 2005; doi: 10.1359/JBMR.050903
Key words: meta-analysis, hip fracture, Bayesian approach, bisphosphonates, osteoporosis, postmenopause,
etidronate, alendronate, risedronate, clodronate
tality risk, reduced quality of life, and incurs significant
health care cost.(1,2)Strategies for preventing or reducing
the burden of hip fracture in the general population have
included both pharmacologic and nonpharmacologic inter-
ventions. In women, hormone replacement therapy (HRT)
has been shown to reduce hip fracture risk by ∼34–38%.(3,4)
However, because of concerns about the possible deleteri-
ous effects of HRT on breast cancer risk and cardiovascular
outcomes, the treatment is not considered an optimal
choice. In recent years, bisphosphonates (i.e., alendronate,
IP FRACTURE IS the most serious consequence of osteo-
porosis, because it is associated with increased mor-
risedronate, and clodronate) have emerged as an alterna-
tive treatment of osteoporosis because these agents have
been shown in randomized clinical trials to be beneficial in
the reduction of vertebral fracture risk and increased
BMD,(5)while having few concerns about deleterious ef-
Because virtually all previous randomized clinical trials
(RCTs) were designed to test the efficacy of bisphospho-
nates on vertebral fracture risk, hip fracture risk was largely
considered a secondary outcome. As a result, the effect of
bisphosphonates on hip fracture risk is uncertain, because
results from different studies have been conflicting, ranging
from statistically nonsignificant to highly significant ef-
fects.(6–9)Nevertheless, a recent review of six clinical trials
found that alendronate treatment of postmenopausal
women with low BMD or established osteoporosis reduced
the risk hip fracture risk by 45%.(10)
In the presence of uncertain evidence, a systematic re-
Dr Eisman serves as a consultant and receives corporate ap-
pointments from Aventis, Eli Lilly and Company, Merck, Sharp &
Dohme Ltd., Novartis, NPS Pharmaceuticals, Organon, Roche,
and Servier. All other authors have no conflict of interest.
Bone and Mineral Research Program, Garvan Institute of Medical Research, St Vincent’s Hospital; Faculty of Medicine, University of
New South Wales, Sydney, New South Wales, Australia.
JOURNAL OF BONE AND MINERAL RESEARCH
Volume 21, Number 1, 2006
Published online on September 6, 2005; doi: 10.1359/JBMR.050903
© 2006 American Society for Bone and Mineral Research
view by compiling all available data and synthesizing them
into a coherent summary may provide a better and more
reliable conclusion about treatment efficacy than a single
trial does. This can be done by meta-analysis in which a
common effect size of all studies is estimated. This study
was undertaken to assess the efficacy of bisphosphonates
on hip fracture risk by using the meta-analysis approach.
MATERIALS AND METHODS
Search strategy and study inclusion
A systematic search the literature was carried out by us-
ing electronic resource including Pubmed, Ovid (from 1966
to March 2004), and the Cochrane Controlled Trials Reg-
istered from 1960 to March 2004. The keywords used for
searching included randomized placebo-and/or controlled
trial(s) or controlled-clinical trial, postmenopause, etidro-
nate, alendronate, risedronate, clodronate, pamidronate,
zoledronate, ibandronate, diphosphonates, bisphospho-
nates, fracture, nonvertebral/or hip fracture. Two reviewers
(NN and TN) identified eligible articles for which the ab-
stracts were recorded. If the abstract was consistent with
the inclusion criteria, the full article text was obtained.
The inclusion criteria were randomized placebo-control
trials comparing postmenopausal women receiving any bis-
phosphonates (e.g., etidronate, alendronate, risedronate, or
clodronate) to those not receiving bisphosphonates with the
duration of follow-up of at least 12 months and the inci-
dence of hip fracture recorded. Only original studies and
papers/abstracts written in English and published in peer-
reviewed journals were selected for analysis.
The full texts of all potentially relevant papers were ob-
tained, and two reviewers (NN and TN) independently
checked for data consistency. If more than one paper with
the same data were identified, only that which contained
the definitive data were included. For each study, relevant
data including details of study design, study duration, agent
used in the treatment, medication dose, inclusion and ex-
clusion criteria, and incidence of hip fractures were ex-
tracted. The primary outcome in this meta-analysis was in-
cidence of newly occurred hip fracture during the study
period (Table 1).
Data synthesis: In each study, the outcome data were
summarized by a two-by-two table format with four cells,
where a and b were the number of hip fractures in the
treatment and placebo group, respectively, and c and d
were the number of individuals without fractures in the
treatment and placebo group, respectively. The effect size
for each study was estimated by the relative risk (RRi),
which is defined as [a/(a + c)]/[b/(b + d)], where i indexes a
study (i = 1, 2, 3…12). The primary aim of the meta-analysis
was to make use of these RRito estimate an overall RR and
its uncertainty. This can be done by both a so-called fixed-
effects and random-effects model, which have widely been
described elsewhere.(5)Briefly, each logarithm of RRiwas
assumed to be normally distributed with a “true” but un-
known effect size ?iand a within-study variance ?i
collection of the logarithms of the RRiacross the different
studies is assumed to follow a normal distribution with
mean ? and variance ?2. Here, ? is the overall logarithmic
relative risk across studies and ?2is the between-study vari-
ance. The synthesis of data were performed both with the
classical random-effects(11)and Bayesian random-effects
models.(12,13)In the classical random-effects model, the pa-
rameters ?, ?2, and ?2are assumed to be fixed; however, in
Bayesian random-effects model, ?i
be random variables. In the Bayesian analysis, the prior
distribution for ? and ?2must be specified. In this analysis,
? was given a vague prior normal distribution of mean 0 and
variance of 10, whereas ?2was assumed to be uniformly
distributed with parameters (0, 10). These prior distribu-
tions were specified on the basis of the high homogeneity in
the population studies and the possible wide range of treat-
ment effects. Estimates of the Bayesian random-effects
parameters were obtained by the method of Markov
Chain Monte Carlo (MCMC) as implemented within the
WinBUGS program.(14)A similar Bayesian analysis based
on absolute difference, rather than logit of relative risk, was
also performed by using the MCMC technique.(13)In the
analysis of absolute risk difference, only 3-year period trials
Heterogeneity: The heterogeneity of effects across studies
was assessed by computing the Cochran’s Q statistic(15)and
the coefficient of inconsistency (I2) as described by Higgins
and Thompson.(16)I2is an estimate of the proportion of
total variation in study estimates that is caused by hetero-
geneity. To illustrate the heterogeneity between the treat-
ment effects in different trials, a funnel plot of sample size
against estimated treatment effect and funnel plot regres-
sion were provided as described by Macaskill et al.(17)Fi-
nally, recursive cumulative meta analysis(18)was also per-
formed to examine whether the magnitude of effects has
been changing markedly over time, as new studies have
been published on this topic. In this analysis, each calendar
year was considered as an informative step, in which evi-
dence was updated by studies published in the interim.
2and ?2are assumed to
Characteristics of studies
Between 1990 and 2004, there were 12 eligible RCTs
involving 18,667 individuals (range per study, 66–
5445),(6–9,19–26)among whom 263 hip fractures were ob-
served. Three RCTs related to etidronate treatment(27–29)
were excluded from the analysis because no hip fractures
were ascertained during the trial period. Furthermore, an-
other trial was excluded because of duplicated data.(30)
Among the 12 trials identified, 2 trials were on etidro-
nate,(23,24)6 were on alendronate,(6,7,19,20,25,26)3 were on
risedronate,(6–9,19–26)and 1 was on clodronate.(22)In the
etidronate trials, patients received intermittent oral etidro-
nate therapy: 400-mg daily dose for 2 weeks, followed by a
10- or 13-week period in which no drugs was given. In the
alendronate trials, the majority of patients initially received
EFFECTS OF BISPHOSPHONATES ON HIP FRACTURE RISK 341
Ross P, Miller PD, Licata AA, Chesnut CH III 1993 Four-year
study of intermittent cyclic etidronate treatment of postmeno-
pausal osteoporosis: Three years of blinded therapy followed
by one year of open therapy. Am J Med 95:557–567.
25. Greenspan SL, Schneider DL, McClung MR, Miller PD,
Schnitzer TJ, Bonin R, Smith ME, DeLucca P, Gormley GJ,
Melton ME 2002 Alendronate improves bone mineral density
in elderly women with osteoporosis residing in long-term care
facilities. A randomized, double-blind, placebo-controlled trial.
Ann Intern Med 136:742–746.
26. Bonnick SL, Rosen C, Mako B, DeLucca P, Byrnes C, Melton
ME 1998 Alendronate vs Calcium for Treatment of Osteopo-
rosis in Postmenopausal Women. Bone 23(Suppl 5):S476–
27. Miller PD, Watts NB, Licata AA, Harris ST, Genant HK,
Wasnich RD, Ross PD, Jackson RD, Hoseyni MS, Schoenfeld
SL, Valent DJ, Chesnut CH III 1997 Cyclical etidronate in the
treatment of postmenopausal osteoporosis: Efficacy and safety
after seven years of treatment. Am J Med 103:468–476.
28. Pouilles JM, Tremollieres F, Roux C, Sebert JL, Alexandre C,
Goldberg D, Treves R, Khalifa P, Duntze P, Horlait S, Delmas
P, Kuntz D 1997 Effects of cyclical etidronate therapy on bone
loss in early postmenopausal women who are not undergoing
hormonal replacement therapy. Osteoporos Int 7:213–218.
29. Wimalawansa SJ 1998 A four-year randomized controlled trial
of hormone replacement and bisphosphonate, alone or in com-
bination, in women with postmenopausal osteoporosis. Am J
30. Watts NB, Harris ST, Genant HK, Wasnich RD, Miller PD,
Jackson RD, Licata AA, Ross P, Woodson GC III, Yanover
MJ 1990 Intermittent cyclical etidronate treatment of post-
menopausal osteoporosis. N Engl J Med 323:73–79.
31. Tan SB, Dear KB, Bruzzi P, Machin D 2003 Strategy for ran-
domised clinical trials in rare cancers. BMJ 327:47–49.
32. Lilford RJ, Thornton JG, Braunholtz D 1995 Clinical trials and
rare diseases: A way out of a conundrum. BMJ 311:1621–1625.
33. Etzioni RD, Kadane JB 1995 Bayesian statistical methods in
public health and medicine. Annu Rev Public Health 16:23–41.
34. Harrell FE Jr, Shih YC 2001 Using full probability models to
compute probabilities of actual interest to decision makers. Int
J Technol Assess Health Care 17:17–26.
35. Lilford RJ 2003 Ethics of clinical trials from a bayesian and
decision analytic perspective: Whose equipoise is it anyway?
36. Goodman SN 1999 Toward evidence-based medical statistics.
1: The P value fallacy. Ann Intern Med 130:995–1004.
37. Goodman SN 1999 Toward evidence-based medical statistics.
2: The Bayes factor. Ann Intern Med 130:1005–1013.
38. Goodman SN 1999 Probability at the bedside: The knowing of
chances or the chances of knowing? Ann Intern Med 130:604–
39. McClung MR, Wasnich RD, Recker R, Cauley JA, Chesnut
CH III, Ensrud KE, Burdeska A, Mills T 2004 Oral daily
ibandronate prevents bone loss in early postmenopausal
women without osteoporosis. J Bone Miner Res 19:11–18.
40. Clark WF, Garg AX, Blake PG, Rock GA, Heidenheim AP,
Sackett DL 2003 Effect of awareness of a randomized con-
trolled trial on use of experimental therapy. JAMA 290:1351–
Address reprint requests to:
Tuan V Nguyen, PhD
Bone and Mineral Research Program
Garvan Institute of Medical Research
384 Victoria Road
Darlinghurst, Sydney, NSW 2010, Australia
Received in original form February 21, 2005; revised form August
2, 2005; accepted August 31, 2005.
EFFECTS OF BISPHOSPHONATES ON HIP FRACTURE RISK349