PHARMACOEPIDEMIOLOGY AND DRUG SAFETY 8] 076Ð080 "1999#

ORIGINAL REPORT

Low!Dose Corticosteroids and Avascular Necrosis of

the Hip and Knee
MARK BAUER MD0\ PAULETTE THABAULT RNC\ MS\ JD0\ DAN ESTOK MD0\ CINDY CHRISTIANSEN PhD1
AND RICHARD PLATT MD MS1
0
Harvard Pil`rim Health Care\ Watertown\ MA\ USA
1
Department of Ambulatory Care and Prevention\ Harvard Pil`rim Health Care and Harvard Medical School\
MA\ USA

SUMMARY
Background * Although high!dose corticosteroid use constitutes a well!known risk factor for avas!
cular necrosis "AVN#\ the risk of AVN with low!dose corticosteroids is not well established[
Objective * To assess AVN risk for high!dose and low!dose oral corticosteroid exposure compared
to baseline risk[
Methods * We employed a nested caseÐcontrol study design[ AVN of the hip and knee was determined
from coded diagnoses and full text record review[ Corticosteroid exposure was ascertained through
outpatient automated pharmacy dispensing _les[
Results * Thirty!one cases occurred during approximately 619\999 person years[ Eleven of 11 cases
that met all eligibility criteria received oral corticosteroids during the 2 years before diagnosis[ There
was a cumulative dose!related corticosteroid e}ect from no excess risk "relative risk  9\ 84) Con_dence
Interval  9Ð4# for doses 0Ð329 mg\ to a relative risk of 5 "84) CI 0Ð32# for doses of 339Ð0189 mg\
and an unde_ned relative risk "84) Lower CI 15# for doses greater than 0189 mg[
Conclusions * Systemic corticosteroids were strongly associated with AVN\ however\ the e}ect was
not clearly evident at 2!year cumulative doses ¾329 mg[ Copyright Þ 1999 John Wiley + Sons\ Ltd[

KEY WORDS * corticosteroids^ avascular necrosis

INTRODUCTION aseptic necrosis of the hip after a single dose of a

Practicing clinicians are often confronted with cho!
ices\ which must be made based on limited data[
The decision whether or not to prescribe a short
tapering course of corticosteroids for conditions
such as moderately severe contact dermatitis rep!
resents such a dilemma[ Scant evidence suggests
that this is a safe path^ however most primary care
clinicians have heard warning anecdotes from
orthopaedic colleagues about patients who develop
 Correspondence to] M[ Bauer MD\ Harvard Pilgrim Health
Care\ 374 Arsenal Street\ Watertown\ MA 91361!4983^ Tel] 506!
861!4129[ Fax] 506!861!4401[ E!mail] markðunspŁbauerÝ
vmed[org
Contract:grant sponsor] Harvard Pilgrim Health Care Foun!
dation[
corticosteroid[ The computerized medical record
and pharmacy data base of our HMO were utilized
to explore this issue[
Several studies\ including a case series0 and
cohort studies in high!risk populations\1\2 have
implicated systemic corticosteroids as an important
cause of avascular necrosis of large joints\ which
can cause disabling arthritis requiring joint replace!
ment[ Although corticosteroids are strongly sus!
pected to cause avascular necrosis\ the magnitude
of the risk has not been well de_ned and the associ!
ation with low!dose corticosteroids is controversial[
A recent review concluded {many of the studies to
date are anecdotal and poorly controlled|[3
Information regarding the risk of AVN after low!
dose corticosteroid exposure is important to guide
corticosteroid prescribing for conditions such as

Received 08 July 0888

Copyright Þ 1999 John Wiley + Sons\ Ltd[ Accepted 10 January 1999
077 M[ BAUER ET AL[
asthma exacerbations and severe contact derma!
titis[ We evaluated this relationship in an HMO
population for which both drug dispensing and
diagnosis information were available[ We sought to
"0# estimate the risk of AVN in the absence of
steroid exposure\ and "1# determine the relationship
between di}erent doses of steroids and new AVN[
METHODS
Population and data sources
The study base included adult members of the sta}!
model component of Harvard Pilgrim Health Care
representing an urban and suburban population of
metropolitan Boston\ Massachusetts[ The popu!
lation and data resources have been described in
detail[4 In many respects the HMO membership
resembles the population of the region\ with some
underrepresentation of the elderly[ The HMO
maintains automated pharmacy and medical record
systems\ the latter containing both coded diagnoses
and machine!retrievable full text[ Between 0880 and
0884\ the number of members at least 07 years old
receiving care at a centre with automated records
was approximately constant[ The adult mem!
bership at the midpoint of that interval\ 079\999\
was used to estimate the person time at risk for
AVN[ Approximately 89) of members had a drug
bene_t that provides 0 month|s prescription for ,4[
The study received approval from the Human
Studies Committee of Harvard Pilgrim Health
Care[
Case identi_cation
We screened for cases of avascular necrosis of the
hip and knee by searching ambulatory and hospital
records for coded diagnoses of AVN[ All records
with these codes were reviewed by two of us "M[B[\
P[T[#\ who con_rmed the diagnosis and its initial
date\ and recorded the involved joint\ demographic
information\ and concomitant diseases[ Coded rec!
ords were con_rmed as cases of AVN if there was
a supporting radiographic diagnosis[ Because all
radiographically con_rmed cases were referred to
orthopaedists\ we de_ned the index date as the _rst
visit to an orthopaedist for AVN[ Cases were
required to have been continuous members with
pharmacy bene_ts for 25 months prior to the index
date and to have used the pharmacy to obtain a
non!corticosteroid prescription in that period[
Copyright Þ 1999 John Wiley + Sons\ Ltd[
Selection of controls
For each case we identi_ed _ve controls with the
same membership and pharmacy bene_t require!
ments\ and matched on gender\ age "24 years# and
health centre[ Observation time for an individual
control began with the _rst pharmacy use within 2
years before the matched case|s incidence date and
continuing for 24 months[ For three cases we ident!
i_ed only four rather than _ve controls because of
these requirements[
Corticosteroid exposure
We used the automated pharmacy records to ident!
ify oral corticosteroid dispensing of cases from 29
days to 2 years before the diagnosis date[ Cortico!
steroid exposure in the 29!day period immediately
before the incident date was considered too brief a
time in which to develop AVN^ and a prior case
study suggested that most cases would develop
within 2 years of exposure[0 To assess dose
response\ three exposure ranges were de_ned] 0Ð
329 mg of prednisone "329 mg representing a typical
03!day tapered course#\ 339Ð0189 mg "a maximum
of three 03!day tapering courses in 2 years#\ and
×0189 mg[ No corticosteroids other than pre!
dnisone were prescribed in this outpatient setting[
No formal accounting was made of corticosteroid
exposures occurring during hospitalization\ or
more than 2 years before the index date\ since this
information was not uniformly available[
Statistical methods
Con_dence intervals for incidence rates were cal!
culated using exact Poisson methods[5 Cortico!
steroid!exposed person time was assumed to be
su.ciently low in the source population to allow us
to ignore it "as shown in Table 0\ exposed person!
time was ³9[4)# in computing the baseline risk of
avascular necrosis[ The association between
corticosteroid exposure and AVN was assessed by
computing odds ratios[ Because the odds ratio is
nearly identical to the risk ratio in this setting\ we
use the terms risk ratio and relative risk for ease of
interpretation[ Con_dence intervals for the odds
ratio were calculated using the mid!p method\
which adjusts the exact intervals because of the
discreteness of the data[6 This method is less con!
servative than the exact method and has been
shown to achieve nominal coverage in simulation
studies for common odds ratios[7 StatXact Version
Pharmacoepidemiolo`y and Dru` Safety\ 8] 076Ð080 "1999#
 LOW!DOSE CORTICOSTEROIDS 078
Table 0 * Systemic corticosteroid exposure and avascular necrosis

Total prednisone AVN cases Controls Relative risk

exposure "n  11# "n  096# "84) CI#
"for 24 months# "7 F\ 03 M# "17 F\ 58 M#

9 mg 00 87 Reference
¾329 mg 9 5 9 "9Ð4#
339Ð0189 mg 1 2 5 "0Ð32#
×0189 mg 8 9 Unde_ned "15Ðin_nity#

 Total mg dispensed in prior 2 years excluding 29 days prior to incidence date[

1[98 software was used for the analysis[ The attribu!
table risk percentage and population attributable
risk percentage were calculated by standard
methods[09 The results of matched and unmatched
analyses did not di}er appreciably\ and so the
unmatched results are reported[
RESULTS
We identi_ed 20 cases of AVN involving the hip
"14 cases# and knee "5# during approximately
619\999 person years\ yielding an overall incidence
of 3[2 cases per 099\999 person years "84) CI 1[8Ð
5[0#[
Twenty!two cases "60)# met all membership and
drug bene_t criteria for inclusion in this study[
Their median age was 35 years^ the range was 10Ð
62 years^ 7 "25)# were women[ Eleven "48)# cases
had received corticosteroid therapy between 29
days and 2 years before their index date[ Indications
for therapy in these 00 cases were] asthma "four
cases#\ and single cases of Wegener|s granu!
lomatosis\ chronic obstructive lung disease\ cer!
ebral oedema\ Crohn|s disease\ renal transplant\
gout and myasthenia gravis[ The estimated rate of
AVN in the absence of corticosteroid exposure was
0[4 per 099\999 person years "84) CI 9[7Ð1[6#[
AVN was strongly associated with corticosteroid
exposure "Table 0#\ with an overall relative risk for
any steroid exposure of 04[4 "4[4Ð3[5# yielding an
attributable risk percentage of 85)\ and a popu!
lation attributable risk percentage of 47)[ No
excess risk could be con_rmed for cumulative doses
of 0Ð329 mg prednisone "RR  9^ 84) CI  9Ð4#[
In contrast\ the RR estimate was indeterminately
elevated at doses ×0189 mg[ The estimated risk in
each exposure category is shown in Table 0[
DISCUSSION
Our study con_rms the relatively low risk of AVN
in a population not exposed to outpatient oral
corticosteroids[ We made several simplifying
assumptions in estimating the incidence of AVN in
the absence of steroid exposure[ We assumed that
turnover of the population did not impair rec!
ognition of prior steroid exposure among cases^ we
believe this was reasonable since 89) of the new
cases we observed had at least 1 years of prior
membership[ We also assumed that there was neg!
ligible corticosteroid exposure from sources other
than our pharmacies because of the low co!payment
for medications "for cases and controls# at the
health centre pharmacies and more importantly
because of the convenience of the patients _lling
their prescriptions at the same site where they were
prescribed[ Inpatient corticosteroid exposure\
which could not be reliably ascertained\ did not
enter into our classi_cation of corticosteroid
exposure[ The ambulatory record of one AVN case
classi_ed in this report as non!exposed noted such
inpatient corticosteroid therapy "Table 1#[ To the
extent that these assumptions are incorrect\ the inci!
dence rate of AVN among unexposed individuals
is overestimated[ The rate is underestimated if cases
were not identi_ed by our case search[ This is most
likely to have occurred for mild cases that were not
radiographically con_rmed and thus did not come
to the attention of one of the 06 orthopaedists who
cared for this population[ We have no reason to
believe there is serious over! or underestimation[
These data con_rm the strong reported associ!
ation between corticosteroids and AVN\ with
corticosteroids accounting for half of the AVN
observed in this population[ There are fewer poten!
tial biases in the assessment of steroid!associated
risk[ Nine patients were excluded from the assess!

Copyright Þ 1999 John Wiley + Sons\ Ltd[ Pharmacoepidemiolo`y and Dru` Safety\ 8] 076Ð080 "1999#
089 M[ BAUER ET AL[

Table 1 * Predisone dosage\ ethanol intake and associated diagnosis

Case Age Gender Cumulative Ethanol intake Associated Dx

prednisone dose "drinks:week# "includes trauma#
"mg#

0 30 M 9 19 None
1 23 M 9 03 Acute leukemia\ received corticosteroid 0889
2 35 M 9 Heavy None
3 33 M 9 Not documented None
4 55 M 9 Not documented None
5 52 F 9 Not documented None
6 41 F 9 Rare None
7 35 M 9 Not documented Trauma "spinal cord injury# 0876
8 62 M 9 19 None
09 61 M 9 Rare None
00 55 F 9 None None
01 32 M 599 Not documented Gout
02 26 M 549 Not documented Asthma
03 55 M 0249 Not documented COPD
04 55 F 2999 Not documented Wegener|s granulomatosis
05 17 M 2999 Not documented Post!trauma cerebral oedema
06 13 F 2139 Not documented Asthma
07 59 F 2559 None Asthma
08 10 M 7949 None Crohn|s disease
19 28 M 7199 None S:P kidney transplant
10 20 F 03\699 Not documented Myasthenia gravis
11 40 F 07\649 Not documented Asthma

ment of the relationship between AVN and cortico!
steroid exposure study because they failed to meet
criteria of 2 prior years pharmacy use[ Their mean
age was 34 years\ three of the nine had systemic
lupus erythematosus\ one had temporal arteritis\
one had asthma and three had no associated diag!
noses[ Since _ve of these nine excluded AVN cases
had diagnoses that are likely to have prompted
corticosteroid therapy\ their exclusion is not likely
to have reduced the magnitude of the observed
association\ especially for high doses[ The require!
ment for both cases and controls to have used the
pharmacy for a non!corticosteroid preparation may
have excluded relatively healthier controls who did
not use prescription drugs during the observation
period[ This might slightly understate the risk of
corticosteroid!associated AVN[ This study exam!
ined only the e}ect of cumulative exposure as
opposed to the rate of exposure to corticosteroid on
AVN risk[ Information on the relationship between
the rate of corticosteroid exposure and AVN is also
lacking in the medical literature[ Exposure to a
given amount of corticosteroid over a longer course
conceivably may incur a di}erent risk of AVN than
the same amount given over a shorter period[
We have found low!level corticosteroid exposure
to represent at most _ve times baseline and perhaps
no additional risk of AVN in the time period stud!
ied[ We are unable to exclude the possibility that
low!dose corticosteroid exposure may take longer
than 2 years to produce AVN^ however\ we think
our orthopaedists would have elicited this history\
since they commonly commented on prior cortico!
steroid exposure "and in one case noted high!dose
exposure 4 years previously#[ We believe the rec!
orded amounts of steroid exposure are plausible for
our population and especially for our cases\ e[g[
the Wegener|s patient received 2999 mg prednisone
which is not a low dose "particularly in an era when
Cytoxan is often used to spare prednisone#[ In any
event\ raising the doses of steroid received in the
high!dose group would not change our conclusions[
We cannot comment directly on whether cases we
coded as receiving no corticosteroids actually
received low!dose steroids in the hospital and none
in the ambulatory setting[ However\ in general\ low!
dose inpatient corticosteroid exposure with no out!
patient exposure is unusual and we know of no
reason why it should be more likely among the
cases[

Copyright Þ 1999 John Wiley + Sons\ Ltd[ Pharmacoepidemiolo`y and Dru` Safety\ 8] 076Ð080 "1999#
 LOW!DOSE CORTICOSTEROIDS
While this study provides estimates of the inci!
dence of AVN in a large population and the lack
of a strong e}ect of low!dose corticosteroids\ it
does not address the impact of other suspected risk
factors for AVN on our study results[ Four of 00
non!steroid exposed AVN cases were noted to be
heavy alcohol users^ this supports alcohol as a pre!
viously described risk factor for AVN[ It is quite
possible that the remaining seven "of 00# cases were
also heavy ethanol consumers^ similarly it is poss!
ible that alcohol consumption contributes to the
incidence of AVN in corticosteroid users[ There
were two high!dose steroid cases confounded by
other known risk factors "trauma and renal trans!
plant#^ a much larger study would be required to
de_ne their associated risks "and the risks of ethanol
consumption# separately[ The lack of complete
information about ethanol exposure or other risk
factors would be expected to obscure the observed
association between corticosteroids and avascular
necrosis[ However\ such missing information has
no bearing on the lack of an observed relationship
between low!dose corticosteroids and AVN\ and
the strong association between higher dose cortico!
steroids and AVN is clear despite whatever mis!
classi_cation of ethanol exposure exists[
The decision to use corticosteroids\ even in low
doses for short periods requires a balancing of the
bene_ts against the risks[ In addition to AVN\ these
risks include osteoporosis\ peptic ulcers\ diabetes
mellitus and cataracts[ These data\ which found no
clear evidence of excess AVN risk associated with
cumulative corticosteroid exposures up to 329 mg
of prednisone\ and which appear to exclude more
than a _ve!fold increase\ may help put this risk in
perspective and allow a more informed decision[
Further investigations to substantiate these _nd!
ings\ and to better characterize the risks of di}erent
dosages\ modes of administration\ and lengths of
therapy would aid in the safer prescribing of
corticosteroid therapy[
Copyright Þ 1999 John Wiley + Sons\ Ltd[
080
ACKNOWLEDGEMENTS
This work was supported in part by institutional
support from the Harvard Pilgrim Health Care
Foundation[ We thank Emily Cain\ Michelle Cor!
deiro and James Livingston for data collection and
analysis\ Dr Robert Miegel for orthopedic expertise
in study design\ and Dr Peter Choo for advice on
the presentation of results[
REFERENCES
0[ Fisher D\ Bickel W[ Corticosteroid induced alva!
scular necrosis[ Bone Joint Sur` 0860^ 42A] 748Ð762[
1[ Ono K\ Tohjima T\ Komazawa T[ Risk factors of
avascular necrosis of the femoral head in patients
with systemic lupus erythematosus under high!dose
corticosteroid therapy[ Clin Orthopaed Related Res
0880^ 166] 78Ð86[
2[ Bradbury G\ Benjamin J\ Thompson J\ Klees E\
Copeland J[ Avascular necrosis of bone after cardiac
transplantation[ J Bone Joint Sur` 0883^ 65"8#] 0274Ð
0277[
3[ Chang C\ Greenspan A\ Gershwin E[ Osteonecrosis]
current perspectives on pathogenesis and treatment[
Sem Arthrit Rheumat 0882^ 12"I#] 36Ð58[
4[ Platt R[ Studies of prescription drugs at Harvard
Community Health Plan[ In Pharmacoepidemiolo`y
"1nd edn#\ Strom B "ed#[ John Wiley and Sons] New
York\ 0883^ 167Ð176[
5[ Ulm J[ A simple method to calculate the con_dence
interval of a standardized mortality ratio[ Am J Epi!
dem 0889^ 020"1#] 262Ð263[
6[ Lancaster HO[ Signi_cance tests in discrete dis!
tributions[ J Am Stat Assoc 0850^ 45] 112Ð123[
7[ Mehta CR\ Walsh SJ[ Comparison of exact\ mid!
p and Mantel Haenzel con_dence intervals for the
common odds ratio across several 1×1 contingency
tables[ Am Statistic 0881^ 35] 035Ð049[
8[ StatXact[ A Statistical Packa`e for Exact Non!
parametric Inference[ Cytel Software Corporation]
Cambridge\ MA\ 0881[
09[ Hennekens CH\ Buring JE[ Epidemiolo`y in Medi!
cine\ Mayrent SL "ed#[ Little Brown and Company
"Boston:Toronto#] 0876[ 76Ð82[
Pharmacoepidemiolo`y and Dru` Safety\ 8] 076Ð080 "1999#