Propranolol and the risk of hospitalized myopathy: translating chemical genomics findings into population-level hypotheses
Propranolol and the risk of hospitalized myopathy:Translating chemical genomics findings intopopulation-level hypothesesSoko Setoguchi, MD, DrPH, a ,d John M. Higgins, MD, b,d,e Helen Mogun, MS, a Vamsi K. Mootha, MD, c andJerry Avorn, MD a Boston, MA
Background A recent large-scale, chemical screening study raised the hypothesis that propranolol may increase the riskof myopathy. We tested this hypothesis in a large population to assess whether (1) propranolol use is associated with anincreased risk of myopathy and (2) the concurrent use of propranolol with a statin may further increase risk of myopathy. Methods New users of propranolol and other β-blockers (BBs) aged ≥65 were identified using data from Medicare anddrug benefit programs in 2 states (1994-2005). The primary end point studied was hospitalization for myopathy orrhabdomyolysis. We used stratified Cox proportional hazards regression to estimate the multivariate-adjusted effect ofpropranolol compared to other BBs and controlled for demographic variables, risk factors for myopathy, other comorbidities,and health service use measures. We also assessed whether co-use of propranolol and statin further increases the risk, byincluding an interaction term for use of propranolol and statins. Results We identified 9,304 initiators of propranolol and 130,070 initiators of other BBs and found 30 cases ofhospitalized myopathy in 15,477 person-years (PYs) of propranolol use and 523 in 343,132 PYs of other BB use. Comparingpropranolol with other BB users, the adjusted hazard ratio was 1.45 (95% CI 1.00-2.11) for myopathy and 1.48 (95% CI0.82-2.67) for rhabdomyolysis. We could not detect interaction between propranolol and statins due to limited power. Similarresults were observed when propranolol users were compared to other antihypertensive drug users. Conclusions Propranolol may be associated with a 45% increased risk of hospitalized myopathy in the elderly. Ourstudy illustrates how results from in vitro chemical screens can be translated into hypotheses about drug toxicity at thepopulation level. (Am Heart J 2010;159:428-33.)
We recently performed a large-scale, chemical genomic
metoprolol or atenolol, gave rise to a very similar
screen of nearly 2,500 drugs in cultured mouse muscle
signature of toxicity. Moreover, the study revealed that
and discovered a molecular and physiologic signature of
combination treatment of these cells with a statin and
statin toxicity.The signature of toxicity reported in this
propranolol gave rise to an additive toxicity in a dose-
cell-based study is consistent with previous reports
dependent manner. A subsequent study demonstrated
suggesting that statins may cause myopathy via a
increased cellular toxicity for propranolol as compared to
mitochondrial Surprisingly, we found that
other β-blockers (BBs) in a different cell type.
treatment of muscle cells with propranolol, but not
These cell-based studies raise the possibility that
propranolol use in humans might be associated with
From the aDivision of Pharmacoepidemiology and Pharmacoeconomics, Department of
increased risk of in vivo mitochondrial toxicity and
Medicine, Boston, MA, bDepartment of Pathology, Brigham and Women's Hospital and
possibly clinically significant myopathy. In the current
Harvard Medical School, Boston, MA, and cCenter for Human Genetic Research,
article, we conducted a cohort study using large popula-
Massachusetts General Hospital, Broad Institute of MIT and Harvard, and Department of
tion-based health care use databases to assess whether (1)
Systems Biology, Harvard Medical School, Boston, MA. dCofirst authors.
propranolol may be associated with an increased risk of
eCurrent address: Center for Systems Biology and Department of Pathology, Massachusetts
myopathy and (2) the concurrent use of propranolol with a
General Hospital, and Department of Systems Biology, Harvard Medical School.
statin may further increase the risk of myopathy.
Submitted June 9, 2009; accepted December 2, 2009. Reprint requests: Soko Setoguchi, MD, DrPH, Division of Pharmacoepidemiology andPharmacoeconomics, Brigham and Women's Hospital, 1620 Tremont St, Suite 3030,Boston, MA 02130.
We conducted a cohort study pooling health care use
2010, Mosby, Inc. All rights reserved.
databases from 2 states: (1) Medicare beneficiaries enrolled in
Table I. Characteristics of cohort patients with age ≥65 (Medicare and Pharmacy Assistance Program in PA and NJ combined; 1995-2005)
Covariates were assessed during 1 year before initiation of study drugs. Values represent percentage for binary variables and median (interquartile range) for continuous variables. ACEI, Angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker; BB, beta-blocker; CCB, calcium-channel blocker; DMARD, disease-modifying antirheumatic drug;
SSRI, selective serotonin reuptake inhibitor; THI, thiazide diuretic.
the Pharmaceutical Assistance Contract for the Elderly (PACE) in
threshold but less than approximately $35,000, thus, including
Pennsylvania from January 1, 1994, to December 31, 2005, and
primarily lower middle-class elderly. The linked Medicare/state
(2) Medicare beneficiaries enrolled in the Pharmaceutical
drug benefit program data provide basic demographic and
Assistance to the Aged and Disabled (PAAD) or in Medicaid in
coded diagnostic and procedural information as well as
the state of New Jersey from January 1, 1994, to December 31,
complete pharmacy dispensing information with high accura-
2005. Both drug benefit programs in Pennsylvania and New
cy.The Institutional Review Board of the Brigham and
Jersey provided comprehensive pharmacy coverage with a small
Women's Hospital and Massachusetts General Hospital ap-
or no copayment. Patients were eligible for coverage by PACE or
proved this study, and data use agreements were established.
PAAD if their income is above the Medicaid annual income
All potentially traceable personal identifiers were removed from
the data before analyses to protect patients' privacy. The authors
rhabdomyolysis would be a subset of hospitalized myopathy
had full access to the data and take full responsibility for its
cases because ICD-9 codes used to defined rhabdomyolysis in
integrity. All authors have read and agreed to the article
the algorithm by Andrade and the new code for rhambdomyo-
lysis (728.88) were part of ICD-9 codes used to define
In the databases, we identified a cohort of subjects aged ≥65
years who were newly started on propranolol or another β-blocker. New use of propranolol was defined as having filled a
prescription for propranolol during the study period and not
Potential confounders were measured during the 12 months
having used the drug during the 12 months before the index use.
before the exposure to propranolol or other BB, using diagnosis
Patients who used other BBs during the 12 months before the
and procedure codes and/or prescription information in the
index use of propranolol were considered as new users of
data, including demographic variables; risk factors for myopathy
propranolol. The same definition was applied to define new
including renal impairment, hypothyroidism, hyperthyroidism,
users of other BBs. This “new user” design is preferable because
liver disease, other comorbidities; and use of other medications
including prevalent users can underestimate the true effect of an
(see for the list of comorbidities and medications).
exposure by missing events that might have occurred soon afterthe first exposures, as well as by focusing on patients who were
less susceptible to a given risk.All patients were required tohave at least one filled prescription and use of at least one
Cox proportional hazards regression was used to estimate the
clinical service during each of 2 consecutive 6-month periods
unadjusted, age-sex–adjusted, and multivariate-adjusted effect of
before the index use of any BB, to ensure ongoing eligibility and
propranolol versus other BB on the occurrence of hospitalization
to assess prior comorbid conditions. The earliest observation
for myopathy or rhabdomyolysis. Patients from the 2 states were
included in the analyses was January 1, 1996.
combined and analyzed in stratified Cox proportional hazardsregression, allowing different baseline incidence of the out-comes between the 2 regions. The model was also stratified by
calendar year to adjust for any trend or variation in the exposure
The exposure of interest was use of propranolol. We chose
and outcomes. We also adjusted potential confounders using
other BB users as a comparison group because the comparison
propensity score methoThe propensity scores were
between active users of similar medications can help protect
estimated as a probability of receiving propranolol compared
against confounding by indication and other selection biases
to another BB, given all potential covariates that predicted the
related to use of preventive medications.Cohort follow-up
use of propranolol. To estimate propensity scores, we also
started at the first prescription of propranolol or other BB during
included several covariates not included in the final multivariate
the study period. We did not allow patients to cross over
models: use of other medications and health service. Dose
between categories and instead censored them as soon as they
response was assessed by replacing indicator variables with
stopped taking the exposure medication of interest. We assessed
propranolol categories versus the comparison drug.
dose response by determining the daily dose of propranolol
To test the hypothesis that concomitant use of statin and
based on the closest dispensing to the outcome or censoring
propranolol may be associated with a further increase in the risk
event, we then categorized the dose into low (≤40 mg) and high
of myopathy, we assessed the most recent use of statins in
(N40 mg), given a median dose of 40 mg/d.
patients taking propranolol and other comparison drugs beforethe previously specified outcomes or censoring events. We then
included an interaction term between propranolol and statin usein the fully adjusted Cox model to determine whether there was
Subjects were censored at the earliest of (1) the last use of
a synergistic effect for statins and propranolol.
propranolol or other BBs, (2) death, or (3) end of the study
We conducted sensitivity analyses using other comparison
period. The last use of propranolol or other BB was defined as
groups within the same population. We identified new users of 3
the last date of prescription plus the number of days supplied,
other classes of antihypertensive medications: angiotensin
plus a 14-day grace period to account for the time lag between
blockers (ABs), calcium-channel blockers (CCBs), and thiazide
filling a prescription and the actual intake of the medication. The
diuretics (THI). We repeated the same analyses comparing
primary end point studied was the first incidence of severe
propranolol users to AB users, CCB users, THI users, and all
myopathy after the initiation of the study drugs, defined as
comparison drug users (AB + CCB + THI + other BB users).
hospitalization in an acute care facility with myopathy-relatedcodes including International Classification of Diseases, NinthEdition (ICD-9) code for rhabdomyolysis (ICD-9 of 710.4,
728.8X, 728.9, 729.1, 791.3, 359.4. 359.8, 359.9) as the primary
or secondary diagnosis listed in the discharge summary. The
We identified 9,304 initiators of propranolol and
secondary outcome was the first incidence of rhabdomyolysis. A
130,070 initiators of other BBs. Among the 130,070
specific ICD-9 code for rhabdomyolysis (728.88) became
other BB users, the most frequent BB used was metoprolol
available only after October 2003. We therefore defined
(57%) followed by atenolol (27%). We also identified new
rhabdomyolysis using a previously developed algorithm byAndrade et up to October 2003. After October 2003, we
users of AB (n = 110,328), CCB (n = 70,976), or THI (n =
defined rhabdomyolysis as hospitalization in an acute care
81,411) for sensitivity analyses. presents the
facility with rhabdomyolysis (ICD-9 of 728.88) as the primary or
characteristics of the study population aged ≥65 and older
secondary diagnosis in the discharge summary. The cases of
measured during the 12-month period before exposure to
Table II. Number of cases, person-years, and incidence rate of
likely that the misclassification bias brought the estimate
toward the null. After October 2003, the specific ICD-9code for rhabdomyolysis became available. Analysis of
the subset of the data with patients at risk for developingrhabdomyolysis after October 2003 found that the HR for
rhabdomyolysis comparing propranolol to other BB users
was 1.96 (95% CI 0.97-3.97) and the HR comparing
propranolol to all other comparison drug users was 2.09
The HR of hospitalized myopathy for high-dose
propranolol (HR 1.68, 95% CI 1.01-2.77) was somewhat
higher than that for low-dose propranolol (HR 1.26, 95%
CI 0.74-2.15), suggesting a possible dose response.
Propensity score analyses yielded similar results to the
multivariate analyses, with the propensity score-adjustedHR having narrower CIs ). These results were
P-Y, Total person-years; IR, incidence rate (per 10000).
consistent when propranolol users were compared tousers of AB, CCB, and THI separately.
the study drugs. Age was similar across the groups, butpropranolol users generally had fewer comorbidities
Concurrent use of statins and lack of synergistic effect
compared to other BB users or all other users of
Concurrent use of any statin was assessed at the time of
comparison drugs. Depression, hyperthyroidism, liver
initiating propranolol, the time of the last prescription
disease, and migraine were slightly more common in
before the outcome, or at a censoring event. The co-use
propranolol users. Propranolol users were more likely to
of statins was relatively infrequent, for example, the use
use antipsychotics, selective serotonin reuptake inhibitor,
of statin at the time of last dispensing was 17% (n = 1,557)
and nonselective serotonin reuptake inhibitor antidepres-
for propranolol users, 30% (n = 39,171) for other BB
sants than users of comparison drugs and were less likely
users, 25% (n = 19,600) for angiotensin-converting
than other BB users to have a history of antiplatelet
enzyme inhibitor/angiotensin receptor blocker users,
21% (n = 15,251) for CCB users and 23% (n = 27,657)for THI users. We did not find any evidence of a
synergistic effect between the use of propranolol and
statins in causing myopathy. Among 30 myopathy
We identified 30 cases of hospitalized myopathy in
hospitalizations for the propranolol users, only 6 were
15,477 person-years of propranolol use and 523 cases in
exposed to statin at the same time. We therefore did not
343,132 person-years of other BB use (and 12
pursue further analyses assessing additive interactions.
admissions for rhabdomyolysis in propranolol users and
Concurrent use with any statin was also assessed at the
227 in other BB users. Compared to other BB users or all
time of initiating the study drug, but we also did not find
other comparison drug users (other BB, AB, CCB, and THI
any significant interaction between propranolol and
combined), the incidence of hospitalization for myopathy
and rhabdomyolysis was elevated in propranolol users(crude rate ratio of 1.3:1.7 for hospitalized myopathy and1.2:1.6 for rhabdomyolysis).
Using very large population-based databases of typical
Association between propranolol and myopathy
elderly patients, we found that propranolol might be
associated with a 45% increase in the risk of severe
After adjusting for potential confounders in the Cox
myopathy. We also found a statistically nonsignificant
proportional hazards models, we continued to find a
48% increase in the risk of rhabdomyolysis in propranolol
significantly increased risk of hospitalized myopathy in
users. These results were consistent using multiple
propranolol users compared to other BB users or all other
comparison groups. These results are compatible with
comparison drug users ). For rhabdomyolysis,
the hypothesis raised by an integrated high-throughput
we found a similar degree of increase in the risk, but the
chemical biology and gene expression study.Although
95% CIs were wider due to a smaller numbers of events
there have been a few case reports associating propran-
(hazard ratio [HR] for rhabdomyolysis comparing pro-
pranolol users to other BB users was 1.48, 95% CI 0.82-
to our knowledge, this is the first study to
2.67). Because the definition of rhabdomyolysis by
suggest that propranolol may be associated with hospi-
Andrade et alhad positive predictive value of 75%, it is
talization for myopathy at the population level.
Table III. Cox analyses (propranolol vs other BB)
Cox analyses (propranolol vs all other comparison drug users)Unadjusted (crude)⁎
⁎ Cox proportional hazards model stratified by calendar year of exposure and state with study time as a time-scale.
† Cox proportional hazards model stratified by calendar year of exposure and state with study time as a time-scale and age, sex, and race in the model.
‡ Cox proportional hazard model stratified by calendar year of exposure and state with study time as a time-scale and age, sex, and adjusted for demographic information (age, race,gender), comorbidities (history of acute coronary syndrome, other coronary artery disease, cerebrovascular disease, peripheral vascular disease, hypertension, chronic kidneydisease, chronic airway disease, diabetes, cancer, depression/anxiety, hypothyroidism, hyperthyroidism, liver disease, anemia, depression, inflammatory myositis), and healthservice use measures (prior nursing home, number of prior hospitalization, number of physician's visits, and number of medications).
One of the promises of modern biomedical research is to
sufficiently specific. However, such misclassification is
inform best practices for patient management with the
likely to be nondifferential and therefore may have led to
insights emerging from high-throughput chemical and
underestimation of the true risk. Second, our population-
genomic studies that are now possible. Many previous
based database does not have precise clinical information
attempts to extrapolate isolated molecular studies or isolated
on all risk factors for myopathy such as body mass index or
genomic or proteomic analyses to human populations have
history of muscle injuries, including creatine kinase
failed because these limited experimental systems do not
elevations. Although we adjusted for liver dysfunction in
always reflect the true complex dynamics of the organism
our study, the condition is likely to be undercoded in the
We note that the experimental findings motivating this
claims data and therefore likely to lead to residual
current study are based on an integrated analysis of multiple
confounding. However, by selecting users of classes of
experimental data sources including studies of cell viability,
drugs that have similar indications as a comparison group,
gene expression, and cell physiologySome compounds
we may have been able to minimize the potential
may show a false-positive correlation based on the analysis
confounding. Finally, propranolol can be used for
of any single source of data, but a correlation based on
treatment of hyperthyroidism, which is also associated
the integration of several different experimental datasets as
with myopathy. To address this potential confounding,
in this analysis is much more likely to yield a robust
we assessed diagnoses for hyperthyroidism and adjusted
prediction. The precise molecular mechanism of the
for the condition in the analyses. We also conducted
toxicity of propranolol in muscle is still unclear and requires
analyses excluding patients who had diagnosis for
hyperthyroidism, which yielded similar hazard estimates
The in vitro study by Wagner et noted at least an
(HR was 1.55 with 95% CI of 0.86-2.81 for rhabdomyolysis
additive, and possibly synergistic, effect of propranolol and
and 1.45 with 95% CI of 0.98-2.13 for hospitalized
statins in causing muscle toxicity. In the present popula-
myopathy comparing propranolol to other BB initiators).
tion-level study, we were not able to detect a synergistic
However, residual confounding by misclassification of
effect because small numbers of dually exposed patients
hyperthyroidism cannot be ruled out. Nonetheless, the
limited the power of these data to elucidate this
degree of residual confounding is expected to be small
relationship. Alternatively, an additive effect of the
due to the low prevalence of the condition and relatively
combination of statin and propranolol on mitochondrial
small imbalance of the condition in our population.
toxicity may not necessarily translate into a synergistic
Our data indicate that propranolol use may pose a 45%
effect of these drugs at the population level. Limited by the
greater risk of severe hospitalized myopathy compared to
small number of cases in the the propranolol users, we
other BBs or other antihypertensive medications. These
were unable to pursue further analyses testing interactions.
findings need to be confirmed in other populations. More
The present study has a few limitations. First, we used
generally, this study illustrates the potential value of
ICD-9 diagnosis codes or a previously validated algorithm
translating findings from chemical and genomic screen-
to define hospitalization for myopathy or rhabdomyolysis.
ing studies into testable hypotheses about drug efficacy
These codes and validated algorithm may not have been
4. Stergachis AS. Record linkage studies for postmarketing drug
surveillance: data quality and validity considerations. Drug Intell Clin
We thank Shawn Murphy and Henry Chueh and the
Partners Health Care Research Patient Data Registry
5. McKenzie DA, Semradek J, McFarland BH, et al. The validity of
group for facilitating use of their database.
Medicaid pharmacy claims for estimating drug use among elderly
Dr Setoguchi had full access to all of the data in the
nursing home residents: the Oregon experience. J Clin Epidemiol
study and takes responsibility for the integrity of the data
and the accuracy of the data analysis.
6. Ray WA. Evaluating medication effects outside of clinical trials: new-
user designs. Am J Epidemiol 2003;158:915-20.
7. Glynn RJ, Schneeweiss S, Wang PS, et al. Selective prescribing led to
overestimation of the benefits of lipid-lowering drugs. J Clin Epidemiol
Financial support: Divisional fund; American Diabetes
Association/Smith Family Foundation.
8. Andrade SE, Graham DJ, Staffa JA, et al. Health plan administrative
databases can efficiently identify serious myopathy and rhabdo-myolysis. J Clin Epidemiol 2005;58:171-4.
9. Rubin DB. Estimating causal effects from large data sets using
1. Wagner BK, Kitami T, Gilbert TJ, et al. Large-scale chemical dissection
propensity scores. Ann Intern Med 1997;127(8 Pt 2):757-63.
of mitochondrial function. Nat Biotechnol 2008;26:343-51.
10. Blessing W, Walsh JC. Myotonia precipitated by propranolol
2. Dirks AJ, Jones KM. Statin-induced apoptosis and skeletal myopathy.
Am J Physiol Cell Physiol 2006;291:C1208-12.
11. Forfar JC, Brown GJ, Cull RE. Proximal myopathy during beta-
3. Cheong HI, Johnson J, Cormier M, et al. In vitro cytotoxicity of eight
beta-blockers in human corneal epithelial and retinal pigment
12. Sobie EA, Guatimosim S, Song LS, et al. The challenge of molecular
epithelial cell lines: comparison with epidermal keratinocytes and
medicine: complexity versus Occam's razor. J Clin Invest 2003;111:
dermal fibroblasts. Toxicol In Vitro 2008;22:1070-6.
The format of this leaflet was determined by the Ministry of Health and its content was 1. NAME OF THE MEDICINAL PRODUCT COLESTID 2. QUALITATIVE AND QUANTITATIVE COMPOSITION One dose (1 packet or 1 level teaspoon) of COLESTID Granules contains 5 grams of colestipol hydrochloride For excipients, see section 6.1 3. PHARMACEUTICAL FORM Light yellow, tasteless and odorless gra
Gastric ulcers: avoidable or just a fact of life? The problem The prevalence of gastric ulceration in Thoroughbreds in race training varies from 70 to 94%, and most sport horses are similarly affected. The horse’s stomach contains two different types of lining – the squamous mucosa on the top half and the glandular mucosa on the bottom. Ulcers can happen in either location, but