B Arroll and F Goodyear-Smith

Corticosteroid injections for

painful shoulder:
a meta-analysis
Bruce Arroll and Felicity Goodyear-Smith

INTRODUCTION
ABSTRACT Shoulder pain is a common source of distress. In two
cross-sectional surveys based on patients registered
Background
There are no systematic reviews of corticosteroids for with general practices a prevalence of 11.7% and
shoulder pain that calculate the numbers needed to treat. 15%, respectively, was found.1,2
Aim Six previous reviews of the use of corticosteroid
We wished to determine the effectiveness in terms of
injections in shoulders have found conflicting
improvement of symptoms of intra-articular and
subacromial injections of corticosteroid for rotator cuff results.3-8 There are no systematic reviews of
tendonitis and frozen shoulder. corticosteroids for shoulder pain that calculate the
Design of study numbers needed to treat. A Cochrane review found
Systematic review and meta-analysis of randomised
that subacromial steroid injection was effective in
controlled trials.
Method improving range of abduction.3 A Health Technology
Data sources included the Cochrane Controlled Trials Assessment published in 1997 concluded that the
Register, Medline, EMBASE, hand searches and author evidence was less than compelling and reported a
contacts. The review methods required any randomised
number needed to treat of 33, with a confidence
controlled trial in which the effectiveness of subacromial
or intra-articular steroid injections versus placebo and interval (CI) including ‘no benefit’.4 A third review
versus non-steroidal anti-inflammatory medication, reported that the evidence was scarce and of poor
could be ascertained. The outcome was improvement quality. It did consider ‘improvement’ but did not
of symptoms. The data abstraction was done
independently, as was the validity assessment. The
pool the results.5 The fourth review reported that
data was pooled using Review Manager 4.1. local corticosteroid injections were effective in
Results rotator cuff tendonitis although it was critical of the
Seven studies were reviewed for corticosteroids versus quality of many of the studies.6 There was no pooling
placebo and three for corticosteroids versus non-
steroidal anti-inflammatory drugs (NSAIDs). The relative of results. The fifth review was conducted by the
risk for improvement for subacromial corticosteroid same authors as the Cochrane review.7 They
injection for rotator cuff tendonitis was 3.08 (95% reported that subacromial steroids were better than
confidence interval [CI] = 1.94 to 4.87). The number
placebo in improving the range of abduction.
needed to treat based on the pooled relative risk was 3.3
(95% CI = 1.8 to 7.7) patients to obtain one We consider improvement/remission a more
improvement. The relative risk for high dose (50 mg of important patient outcome than increases in range
prednisone or more) was 5.9 (95% CI = 2.8 to 12.6). The of motion or improvements on pain scales, as it
relative risk for improvement with steroids compared with
NSAIDs was 1.43 (95% CI = 0.95 to 2.16). The number
needed to treat for corticosteroids versus NSAIDs was B Arroll, PhD, FRNZCGP, associate professor of General
2.5 (95% CI = 1 to 9) for one significant study. The
Practice and Primary Health Care; F Goodyear-Smith, MGP,
relative risks for intra-articular steroid injection for rotator
FRNZCGP, senior lecturer in General Practice and Primary
cuff tendonitis were not statistically significant.
Health Care, Department of General Practice and Primary
Conclusion
Health Care, Faculty of Medical and Health Sciences,
Subacromial injections of corticosteroids are effective
for improvement for rotator cuff tendonitis up to a University of Auckland.
9-month period. They are also probably more effective
than NSAID medication. Higher doses may be better Address for correspondence
than lower doses for subacromial corticosteroid Associate Professor Bruce Arroll, Department of General
injection for rotator cuff tendonitis. Practice and Primary Health Care, Faculty of Medical and
Keywords Health Sciences, University of Auckland PB 92019,
adrenal cortex hormones; injections; meta-analysis; Auckland, New Zealand. E-mail: b.arroll@auckland.ac.nz
relative risk; review, systematic; shoulder impingement
syndrome; shoulder pain. Submitted: 7 February 2004; Editor’s response: 23 April
2004; final acceptance: 24 May 2004.
©British Journal of General Practice 2005, 55, 224–228.

224 British Journal of General Practice, March 2005

 Review Article

enables a number needed to treat to be calculated.

This view has some support in the literature.9 The
most recent review was a Cochrane review by some
of the same authors from the previous Cochrane
How this fits in
There are no systematic reviews of corticosteroids for shoulder pain that
review.3 They concluded that the effect of calculate the numbers needed to treat. We wished to determine the
subacromial steroids had a small benefit, but again effectiveness in terms of improvement of symptoms of intra-articular and
only considered continuous outcomes, not subacromial injections of corticosteroids for rotator cuff tendonitis and frozen
shoulder. A systematic review and meta-analysis of randomised controlled trials
complete remission. They also concluded that
indicates that subacromial injections of corticosteroids are effective for
subacromial steroids were no better than non-
improvement of rotator cuff tendonitis up to a 9-month period. They are also
steroidal anti-inflammatory drugs (NSAIDs).
probably more effective than non-steroidal anti-inflammatory drugs. Higher
Our objective was to systematically review the doses may be better than lower doses for subacromial corticosteroid injection
literature and statistically pool the results of for rotator cuff tendonitis.
improvement outcomes. The clinical question was
whether or not intra-articular and subacromial
injections of corticosteroid are effective compared
with placebo and NSAIDs in terms of improvement of fixed effects model was used throughout, as there
symptoms of rotator cuff tendonitis and frozen was no significant heterogeneity.13 The a priori
shoulder. We also wished to calculate a number sensitivity analysis included an analysis by high
needed to treat, as this has not been done before. quality studies versus low quality studies, high and
low dose, and an analysis of different medical
METHOD providers giving the injections. A high quality study
The Cochrane Controlled Trials Register, Medline was one with a Jadad score of three or more. The
1966–2004, and EMBASE 1980–2004 databases conduct of this review was undertaken according to
were searched using the MeSH terms ‘adrenal cortex the Quorom statement.14 CIs for the number needed
hormone’, ‘randomised controlled trial’, ‘shoulder to treat were calculated using the evidence-based
pain’ and ‘shoulder impingement syndrome’, and the medicine calculator on the University of Toronto
non-MeSH terms ‘rotator cuff tendonitis’, ‘frozen website.15
shoulder’ and ‘adhesive capsulitis’. Authors of
studies retrieved and included were contacted RESULTS
regarding any known unpublished work. The search for trials and reasons for exclusion are
The reference lists of retrieved papers were also presented in Figure 1, with the seven trials found to
searched for relevant papers. The selection criteria
required that the studies be randomised controlled Table 1. Quality scores of shoulder studies.a
trials in which the effectiveness of corticosteroids
could be assessed. This included studies of Jadad
corticosteroids versus placebo or NSAIDs, and Reference 1 2 3 4 5 6 7 8 9 10 11 score
studies of local anaesthetic and corticosteroids versus 16 + + ? + + + + + + + - 5
local anaesthetic. The participants needed to have a 17 + + ? + + + + + + + - 5
diagnosis of frozen shoulder or rotator cuff tendonitis 19 + + + + + + + + + + - 5
of any duration. The outcome needed to include 20 - + - + ? ? + - - + - 0
improvement, as this was considered the most 21 + + - + + ? + + - + - 2
significant patient-oriented outcome. Independent 23 + + + - + + - - - + - 1
assessment of included papers was undertaken and 22 + + - + - - + + + + - 2
any disagreements resolved by consensus. 18 + + + + + + - + - + - 3
A validity assessment was conducted using the a
Column numbers correspond to the following: 1 = study described as double blinded; 2 =
items from the Pedro scoring system (Table 1).10,11 subjects were randomly allocated to groups; 3 = allocation was concealed; 4 = groups
Scoring of studies was undertaken using the Jadad were similar at baseline; 5 = subjects were blinded; 6 = practitioners who administered the
system, which is a validated scoring system.12 Each intervention were blinded; 7 = assessors were blinded; 8 = measurements of key outcomes
were obtained from >85% of subjects; 9 = data were analysed by intention to treat; 10 =
of the included studies was assessed
statistical comparisons between groups were conducted; 11 = point measures and
independently for quality by the two authors and measures of variability were provided. + Indicates the criterion was clearly satisfied;
disagreement resolved through discussion. Data - indicates that it was not; ? indicates that it is unclear whether criterion was satisfied.
extraction was also done independently and Jadad score: 1. Was the study described as randomised? 2. Was the study described as
disagreement resolved through discussion. Data double blind? 3. Was there a description of withdrawals and dropouts? Give a score of 1
point for each ‘yes’ or 0 points for each ‘no’. Give 1 extra point if randomisation or blinding
were analysed using Review Manager 4.1 (Update
appropriate. Deduct 1 point if randomisation or blinding inappropriate. Score quality 0–5.
Software, Oxford). For improvement we calculated Poor quality <3.
the relative risk and the number needed to treat. A

British Journal of General Practice, March 2005 225

 B Arroll and F Goodyear-Smith

were 3.3 (95% CI = 1.8 to 7.7). There were no

important harms other than transient redness and
64 papers from CCTR,a discomfort. None of the studies reported tendon
Medline, EMBASE, hand
searches and authorsb rupture. Pooling our three high scoring (high quality)
studies resulted in a pooled relative risk = 5.9 (95%
CI = 2.8 to 12.6).16,17,19 Examination of study doses
10 duplicates
showed that the study with the highest dose
(100 mg of prednisone equivalent) had the highest
relative risk16 and the study with the lowest dose20
6 systematic reviews
48 randomised (26.66 mg of prednisone equivalent) had a non-
controlled trials significant result. Pooling of the three high dose
21 not randomised (also the high quality studies) had a relative risk of
trials, non-English or 5.9 (95% CI = 2.8 to 12.6). A sensitivity analysis was
not relevant to study conducted with the study by Blair et al17 removed as
question
the outcomes for the corticosteroid group were
6 systematic reviews assessed on average at 33 weeks while the placebo
27 randomised trials group was assessed at 28 weeks. The pooled
with usable information
relative risk was similar to that with Blair et al
19 conditions other included.
than rotator cuff There was only one study for intra-articular
syndrome or frozen
shoulder or no placebo corticosteroid injection and the effect was not
significant. 22 Pooling of the three trials of
6 systematic reviews corticosteroids versus NSAIDs (Table 3) found a
8 trials eligible for
assessment
pooled relative risk for improvement of 1.43 (95%
CI = 0.95 to 2.16).16,19,23 The numbers needed to treat
to obtain one remission by giving a corticosteroid
injection compared with an NSAID was 2.5 for the
one significant study. Pooling of the two high quality
studies of corticosteroids versus NSAID had a
5 randomised controlled
trials able to pool for relative risk = 1.9 (95% CI = 1.06 to 3.4).16,19 As no
meta-analysis non-English language studies were included it was
decided to do a sensitivity analysis with the one
study from a non-English speaking country (a
a
CCTR = Cochrane Controlled Trial Register. bAuthors = from contact with known authors. German study published in an English language
journal) being left out of the analysis.20 This did not
make a substantial difference, with the relative risk
Figure 1. Process of meet the ‘versus placebo or NSAID’ inclusion = 3.3 (95% CI 1.2 = 9.2). A funnel plot revealed a
inclusion of studies and criteria.16-22 An additional paper was found which met possible absence of small trials with small effects.
usable information.
the inclusion criteria for corticosteroids versus All of the clinicians giving the injections were
NSAIDs.23 rheumatologists, orthopaedic surgeons, internal
There were five studies that had data on medicine specialists or rehabilitation specialists.
improvement for subacromial injections versus There was no difference between these groups.
placebo for rotator cuff tendonitis (Table 2). The
terms that we interpreted as improvement DISCUSSION
(Supplementary Table 1) were ‘responder’, 16 Summary of main findings
‘decreased pain’,17 ‘remission’,19 ‘excellent result’20 Our results show a significant benefit for
and ‘complete remission’. 21 There were two subacromial corticosteroid injections versus
interventions in the Plafki et al study20 and hence in placebo for painful shoulder. This is the first review
Table 2 the placebo group is halved to avoid double to show a benefit for steroid injections in terms of
counting of that group. There was a significant the dichotomous variable improvement. The
improvement, with a relative risk of 3.08 (95% CI = numbers needed to treat range between 1.4 and 2.2
1.94 to 4.87). The numbers needed to treat for the patients, and, hence are clinically very significant.
statistically significant studies were between 1.4 The numbers needed to treat in similar ranges were
(95% CI = 1 to 2) and 2.2 (95% CI = 1 to 5). The obtained by applying the pooled relative risk to the
numbers needed to treat obtained from the pooled control event rates. It is also the first review to
relative risk using a control event rate of 14.3%24 suggest that higher doses of corticosteroids may

226 British Journal of General Practice, March 2005

 Review Article

give greater improvement. The interpretation of the

Table 2. Improvement for subacromial steroid for rotator cuff
subgroups of higher doses requires caution as there tendonitis.
was a range of doses and episode duration in the
three studies. It is probably not possible in a series Control Treatment Weight Relative risk (fixed)
of clinical trials to identify safety issues such as Reference n/total n/total % (95% CI)
tendon rupture. One reviewer claims that 16 14/20 0/20 2.67 29.00 (1.85 to 455.25)
corticosteroid injections into the rotator cuff have 17 16/19 8/21 40.63 2.21 (1.24 to 3.94)
not been shown to be deleterious but that it is 19 7/25 2/25 10.69 3.50 (0.80 to 15.23)
logical to limit the number of local corticosteroid 20a 11/16 0/5 3.95 8.12 (0.56 to 117.70)
injections.6 Our findings also suggest that, rather 20 b
8/16 0/5 3.95 6.00 (0.40 to 88.93)
than having no benefit when compared with 21 9/28 7/27 38.10 1.24 (0.54 to 2.86)
NSAIDs, there may in fact be a large benefit, with a
Total 65/124 17/103 100.00 3.08 (1.94 to 4.87)
number needed to treat of 2.5. However, this
Test for hetrogeneity χ2 = 9.14, degrees of freedom = 5 (P = 0.10). Test for overall effect
interpretation is based on two high quality studies,
Z = 4.79 (P<0.00001). aArm with dexamethasone. bArm with triamcinolone.
although only one was statistically significant. We
could find no steroids versus placebo studies for
adhesive capsulitis. Table 3. Improvement for subacromial steroid injection versus
NSAID for rotator cuff tendonitis.
Strengths and limitations of this study
Control Treatment Weight Relative risk (fixed)
A limitation of this review is possible publication bias,
Reference n/total n/total % (95% CI)
in that by missing unpublished or negative trials we
16 14/20 6/20 28.60 2.33 (1.13 to 4.83)
may have overestimated the beneficial effect of
19 7/25 5/25 23.80 1.40 (0.51 to 3.82)
subacromial corticosteroid injections. An analysis
leaving out the one study from a non-English 23 9/15 10/15 47.60 0.90 (0.52 to 1.55)
speaking country did not alter the findings. However, Total 30/60 21/60 100.00 1.43 (0.95 to 2.16)
we are confident that most research in this field was Test for hetrogeneity χ = 4.49, degrees of freedom = 2 (P = 0.11). Test for overall effect
2

identified by our comprehensive, systematic search Z = 1.70 (P = 0.09).

strategy including hand searching and author

contacts. All of the studies were conducted in of an effective (in the short-term) corticosteroid
outpatient settings and hence our findings are injection found no long-term difference between
generalisable to those settings. manipulation and physiotherapy, and that up to half
of the patients experienced recurrent complaints.26
Comparison with existing literature High doses of corticosteroids (50 mg equivalent of
Our findings differ from the other reviews in that we prednisone or greater) may be more effective than
report improvement. We feel that this is a more lower doses.
important patient-oriented outcome than increases
in range of movement and/or pain reduction as it Implications for future research and clinical
allows a number needed to treat to be calculated.25 practice
Only the two reviews by Green et al3,7 and the one by In summary, our findings suggest that subacromial
Buchbinder et al8 attempted to pool the results of injections of corticosteroids are probably effective in
the papers. They did not pool the papers by Blair et rotator cuff tendonitis. They are probably more
al,17 Plafki et al20 and Vecchio et al21 in part because effective than NSAID medication. There is insufficient
they did not have sufficient data to analyse for evidence to determine the effectiveness of intra-
continuous data. We feel their omission of these articular injections for rotator cuff tendonitis or for
papers is not warranted as they contain discrete frozen shoulder. Our finding that using improvement
data that are relevant to effectiveness and possibly as an outcome rather than pain or range of motion
more pertinent as they enable a number needed to was significant suggests that authors of other
treat to be calculated. musculoskeletal reviews may wish to consider a
The reviews by McQuay et al,4 Goupille and broader range of outcome measures. Further
Sibilia6 and van der Heijden et al5 commented on the research is needed to examine different doses and
poor quality of the literature and did not attempt to repeated injections. Outcomes need to include
pool their findings.4-6 dichotomous results so that numbers needed to treat
From our data the duration of benefit of can be calculated. The small numbers needed to
subacromial corticosteroid injections appears to be treat may make GPs more likely to use subacromial
from 3 to 38 weeks. The longer term benefit may steroids for rotator cuff syndrome as it is a relatively
not be so enduring, since a 2-year follow-up study easy procedure to perform.

British Journal of General Practice, March 2005 227

 B Arroll and F Goodyear-Smith

Supplementary information 5. Van der Heijden GJ, van der Windt DA, Kleijnen J, et al. Steroid
Additional information accompanies this article at injections for shoulder disorders: a systematic review of
http://www.rcgp.org.uk/journal/index.asp randomised clinical trials. Br J Gen Pract 1996; 46: 309–316.
Funding body 6. Goupille P, Sibilia J. Local corticosteroid injections in the
treatment of rotator cuff tendinitis (except for frozen shoulder and
This work was funded by the Accident Rehabilitation and
calcific tendinitis). Groupe Rhumatologique Francais de l’Epaule
Compensation Insurance Corporation (G.R.E.P.). Clin Exp Rheumatol 1996; 14: 561–566.
Competing interests 7. Green S, Buchbinder R, Glazier R, Forbes A. Systematic review of
None randomised controlled trials of interventions for painful shoulder:
selection criteria, outcome assessment, and efficacy. BMJ 1998;
REFERENCES 316: 354–360.
1. Badcock LJ, Lewis M, Hay EM, et al. Chronic shoulder pain in the 8. Buchbinder R, Green S, Youd JM. Corticosteroid injections for
community: a syndrome of disability or distress? Ann Rheum Dis shoulder pain. (Cochrane Review). In: Cochrane Collaboration.
2002; 61: 128–131. The Cochrane Library. Issue 1. Oxford: Update Software, 2003.
2. Palmer KT, Cooper C, Walker-Bone K, et al. Use of keyboards and 9. Guyatt G, Juniper EF, Walter SD, et al. Interpreting treatment
symptoms in the neck and arm: evidence from a national survey. effects in randomised trials. BMJ 1998; 316: 690–693.
Occup Med (Lond) 2001; 5: 392–395. 10. Herbert RD, Gabriel M. Effects of stretching before and after
3. Green S, Buchbinder R, Glazier R, Forbes A. Interventions for exercising on muscle soreness and risk of injury: systematic review.
shoulder pain: systematic review. Cochrane Musculoskeletal Group BMJ 2002; 325: 468.
Cochrane Database of Systematic Reviews 2002; 3. 11. Verhagen AP, de Vet HC, de Bie RA, et al. The Delphi list: a criteria
4. McQuay HJ, Moore RA, Eccleston C, et al. Systematic review of list for quality assessment of randomised clinical trials for
outpatient services for chronic pain control. Health Technol Assess conducting systematic reviews developed by Delphi consensus. J
1997; 1: i–iv, 1–135. Clin Epidemiol 1998; 51: 1235–1241.
12. Jadad AR, Moore RA, Carroll D, et al. Assessing the quality of
reports on randomised clinical trials. Is blinding necessary? Control
Clin Trials 1996; 17: 1-12.
13. Mulrow CD, Oxman AD. Cochrane collaboration handbook.
(Updated 1 March 1997.) In: Cochrane Collaboration. The
Cochrane Library. Issue 4. Oxford: Update software, 1997.
14. Moher D, Cook DJ, Eastwood S, et al. Improving the quality of
reports of meta-analyses of randomised controlled trials: the
QUOROM statement. Quality of Reporting of Meta-analyses.
Lancet 1999; 354: 1896–900.
15. Centre for Evidence Based Medicine. EBM calculator v1.2
http://www.cebm.utoronto.ca/palm/ebmcalc/ (accessed 24 Jan
2005.)
16. Adebajo AO, Nash P, Hazleman BL. A prospective double blind
dummy placebo controlled study comparing triamcinolone
hexacetonide injection with oral diclofenac 50 mg TDS in patients
with rotator cuff tendinitis. J Rheumatol 1990; 17: 1207–1210.
17. Blair B, Rokito AS, Cuomo F, et al. Efficacy of injections of
corticosteroids for subacromial impingement syndrome. J Bone
Joint Surg 1996; 78: 1685–1689.
18. De Jong BA, Dahmen R, Hogeweg JA, Marti RK. Intra-articular
triamcinolone acetonide injection in patients with capsulitis of the
shoulder: a comparative study of two dose regimens. Clin Rehabil
1998; 12(3): 211–215.
19. Petri M, Dobrow R, Neiman R, et al. Randomised, double-blind,
placebo-controlled study of the treatment of the painful shoulder.
Arthritis Rheum 1987; 30: 1040–1045.
20. Plafki C, Steffen R, Willburger RE, Wittenberg RH. Local anaesthetic
injection with and without corticosteroids for subacromial
impingement syndrome. Int Orthop 2000; 24(1): 40–42.
21. Vecchio PC, Hazleman BL, King RH. A double-blind trial
comparing subacromial methylprednisolone and lignocaine in
acute rotator cuff tendinitis. Br J Rheumatol 1993; 32(8): 743–745.
22. Berry H, Fernandes L, Bloom B, et al. Clinical study comparing
acupuncture, physiotherapy, injection and oral anti-inflammatory
therapy in shoulder-cuff lesions. Curr Med Res Opin 1980; 7:
121–126.
23. White RH, Paull DM, Fleming KW. Rotator cuff tendinitis:
comparison of subacromial injection of a long acting
corticosteroid versus oral indomethacin therapy. J Rheumatol 1986;
13: 608–613.
24. Picavet HS, Schouten JS. Musculoskeletal pain in the Netherlands:
prevalences, consequences and risk groups, the DMC(3)-study.
Pain 2003; 102: 167–178.
25. Liang MH, Lew RA, Stucki G, Fortin PR, Daltroy L. Measuring
clinically important changes with patient-oriented questionnaires.
Med Care 2002; 40(4 Suppl): II45–II51.
26. Winters JC, Sobel JS, Groenier KH, et al. Comparison of
physiotherapy, manipulation, and corticosteroid injection for
treating shoulder complaints in general practice randomised,
single blind study. BMJ 1997; 314: 1320–1325.

228 British Journal of General Practice, March 2005