Nonsteroidal Anti-inflammatory and Corticosteroid Injections for Shoulder Impingement Syndrome: A Systematic Review and Meta-analysis

Abstract

Context:

To determine optimal treatment strategies for shoulder impingement syndrome (SIS).

Objective:

To compare subacromial nonsteroidal anti-inflammatory injections (SNIs) and subacromial corticosteroid injections (SCIs) on pain relief and functional improvement in individuals with SIS. Second, to perform a cost analysis of the 2 injections.

Data Sources:

MEDLINE, SPORTDiscus, CINAHL, Embase, Web of Science, and SCOPUS databases were searched for randomized controlled trials using several keywords.

Study Selection:

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were utilized, and 10 studies comparing changes in pain or function in humans with SIS receiving SNIs or SCIs were included. Quality and risk of bias were assessed using the Consolidated Standards of Reporting Trials (CONSORT) 2010 scale and the Cochrane Collaboration tool.

Study Design:

Systematic review and meta-analysis.

Level of Evidence:

Level 1.

Data Extraction:

Baseline and follow-up scores of the visual analog, Constant-Murley, and University of California Los Angeles shoulder scales were extracted to calculate effect sizes (ESs), represented as Cohen d. Metaregression and publication bias analyses were performed. Procedural and medication costs were extracted from Medicare guidelines.

Results:

A total of 7 high and 3 good quality studies were included, with a mean score of 21.1. Only 1 study had a high risk of bias. The meta-analyses produced pooled ESs of 0.05 (P = 0.83), 0.12 (P = 0.71), and 0.07 (P = 0.79) for each scale, respectively, with CIs crossing 0. Procedural costs were equal between groups, whereas ketorolac was the least costly medication ($0.47). There was no significant difference in side effects between the 2 injections.

Conclusion:

SNIs are as effective as SCIs for short-term pain relief and improving function in patients with subacromial impingement syndrome. In addition, they are less expensive and cause no major difference in complications, providing a viable, cost-effective alternative for injection therapy in patients with SIS.

Shoulder pain is the third most common musculoskeletal complaint seen by primary care physicians and orthopaedists. ³⁹ With an estimated annual prevalence of 14.7 new cases per 1000 patients, ¹⁹ it trails only back pain and knee pain. ⁶⁰ Shoulder impingement syndrome (SIS) accounts for 44% to 65% of these visits, making it the most common shoulder disorder evaluated. ³⁷ Thus, advancements in the treatment of SIS could be quite impactful to musculoskeletal healthcare.

SIS occurs due to entrapment of the rotator cuff tendons within the anatomic spaces of the shoulder. There are 2 types: external impingement, which occurs in the subacromial space, and is the primary focus of our study, and internal impingement, which involves the glenohumeral joint. External, or subacromial impingement, is a result of the entrapment and inflammation of the supraspinatus tendon within a narrowed subacromial space. Intrinsic causes of this narrowing include tendon or bursa swelling, as seen in tears or bursitis, bony growth formation, as seen in arthritis, or a curved acromion, an anatomic variant. Extrinsic factors include muscle imbalance and scapular instability. Regardless of the underlying cause, patients with SIS suffer from pain and or decreased range of motion. It is a clinical diagnosis based on history and physical examination. ³³

Several guidelines including those from the American College of Physicians, the American Academy of Family Physicians, and the American Academy of Orthopedic Surgeons recommend treating SIS conservatively at first, with activity modification, topical or oral nonsteroidal antiinflammatory drugs (NSAIDs) and physical therapy (PT). If this approach fails, or if PT is intolerable, a subacromial corticosteroid injection (SCI) is recommended to reduce pain and inflammation, and subsequently allow for PT. Consultation with an orthopaedic surgeon is recommended should this approach not meet the patient’s expectations.^4,23,50

NSAIDs have been a mainstay in the treatment of acute and chronic musculoskeletal injuries since the early 1950s. ¹² Many studies have shown that oral NSAIDs and SCIs significantly improve pain in patients with shoulder pain, including those with SIS.^2,8,35,48,65 However, both oral and topical NSAIDs, as well as SCIs, have side effects that limit their use. Oral NSAIDs can cause gastrointestinal damage, cardiovascular complications, and impair renal function,^26,44 while topical NSAIDs pose a risk of local skin reactions. ³² In addition, while topical NSAIDs can penetrate to the synovium, tissue concentration varies significantly; thus, their analgesic and anti-inflammatory effects may be inconsistent. ⁵² SCIs can result in both local and systemic effects such as bone weakening, tendon adhesions, cartilage necrosis and hyperglycemia that render them unsafe for frequent or prolonged use.^53,58 An alternative to oral or topical NSAIDs and SCIs are subacromial nonsteroidal anti-inflammatory injections (SNIs). Limited data exist on their side effects, but they pose no negative effects on articular cartilage or long-term joint function. ⁵¹ In addition, SNIs have been shown to cause less gastrointestinal and renal effects than oral NSAIDs in patients with knee arthritis. ²²

Several studies have been conducted over the past decade comparing pain relief of SNIs and SCIs in patients with SIS,^{1,3,17,25,30,31,38,54,59,66} many of which report similar pain reduction between the 2 approaches. Our primary purpose was to examine the pain relief, functional improvement, and cost of these 2 injections by way of a systematic review and metaanalysis. We build on previous reviews^20,57,62 by limiting to randomized control trials (RCTs) that only utilized injection therapy, adding 7 studies since the last systematic review, ⁵⁷ and conducting a cost analysis to provide a more comprehensive evaluation of this important question.

Methods

Protocols and Registration

Study selection, eligibility criteria, and data extraction calculations were conducted based on a prespecified protocol before writing and registered on PROSPERO (CRD42021264119).^10,11 For the selection of articles to review, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement and the PRISMA flowchart were utilized.^40,46

Eligibility Criteria

Inclusion criteria included RCTs that compared changes in pain or functionality in humans with symptomatic SIS after an SNI or SCI. Studies included patients of any age, gender, or pain severity. Exclusion criteria included studies evaluating individuals with shoulder pain or a history of shoulder pain due to other pathologies besides SIS, or patients using alternative therapies such as oral or topical analgesics and PT. Animal studies or non-RCTs were also excluded.

Information Sources

A search strategy using the keywords pain, shoulder, impingement syndrome, nonsteroidal antiinflammatory drug, corticosteroid, and injection was developed. A search for RCTs was then conducted in MEDLINE (PubMed), SPORTDiscus (EBSCO), CINAHL (EBSCO), Embase (Elsevier, Embase.com), Web of Science, and SCOPUS from each database’s inception date through October 2021. RCTs were identified using the Cochrane Handbook for Systematic Reviews of Interventions (version 6.2, 2021). ²⁷ Boolean terms are available in Appendix 1 (available in the online version of this article). The search was performed between August and October 2021.

Study Selection

Two reviewers (RZ and KP) independently performed an eligibility assessment on each trial in the database search by initially eliminating studies that did not investigate SIS. Titles and abstracts were screened, followed by a full-text article if the trial was deemed eligible by at least 1 reviewer. Discordance was resolved by a third reviewer (TMB).

Data Collection

Utilizing Microsoft Excel, data extraction included the following: study author, study year, the type and dose of medications used, number of patients, mean age, dropout rate, major inclusion and exclusion modality, duration of intervention, adverse effects, and outcome scores. The outcome scores were either mean difference (MD) with SD, mean baseline and follow-up values with SDs, or the median and interquartile range values. Follow-up values were from 4 or 6 weeks postinjection.

Assessment of Methodological Quality and Risk of Bias

Studies were independently assessed by 2 reviewers (R.Z. and K.P.) for methodological quality and risk of bias using the Consolidated Standards of Reporting Trials (CONSORT) 2010 scale. ⁴⁷ Scoring discrepancies within 2 points were averaged together, whereas discrepancies greater than 2 points were resolved by a third independent reviewer (TMB). A score greater than 20 was considered high quality, a score between 16 and 20 was considered good quality, a score between 11 and 15 was deemed fair quality, and a score of 10 or less was considered poor quality.

Similarly, 2 reviewers (RZ and KP) independently assessed the risk of bias using the Cochrane Collaboration tool for assessing the risk of bias in randomized trials and supplementary material from the Cochrane handbook. ²⁸ Any disagreement was resolved by a third independent reviewer (TMB).

Effect Sizes and Meta-analysis

For each study, scores for pain by the visual analog scale (VAS), and functionality by the Constant-Murley scale (CMS) and/or the University of California Los Angeles (UCLA) shoulder scale were used to determine the treatment effect. MD was extracted if possible, however, for studies reporting only the mean baseline and follow-up scores, MD was calculated by subtracting the 2. Meanwhile, for studies reporting median values and interquartile range, the baseline and follow-up means and SDs were calculated using the following equations: ⁶³ $\mathrm{Mean}=\frac{\mathrm{Median}+\mathrm{Q}1+\mathrm{Q}3}{3}$ and $\mathrm{SD}=\frac{\mathrm{Q}3-\mathrm{Q}1}{1.35}$ , and MDs were subsequently calculated as previously. However, for these calculated MDs, the corresponding SDs had to be imputed. A systematic protocol, derived from a mathematical model between baseline SD, follow-up SD, and difference SD, was determined before the commencement of data extraction to reduce the risk of researcher bias. Of note, it was assumed that the data sampled at baseline and follow-up were normally distributed, and that the values imputed originated from a normal distribution. In addition, it was stated that the log of baseline SD, follow-up SD and difference SD are related by a normal trivariate distribution. Then, as previously documented,^34,43 the following equation was applied: ${\mathrm{SD}}_{\mathrm{difference}}^2={\mathrm{SD}}_{\mathrm{baseline}}^2+{\mathrm{SD}}_{\mathrm{endpoint}}^2-\left[\left(2{\mathrm{SD}}_{\mathrm{baseline}}\right)\left({\mathrm{SD}}_{\mathrm{follow}-\mathrm{up}}\right)\right].$ A Bayesian approach was selected and therefore defined the posterior assumption for ρ to range from 0 to 1. A sensitivity analysis for ρ using the values 0, 0.25, 0.5, and 0.75 was used for each scale and the same ρ was used across studies using the same scale. Each study was analyzed using a 2-tailed t test. The MD was divided by the pooled SD to produce the standardized MD, or Cohen d, to represent the ES, which we characterized by the standard interpretation (0.2 = small, 0.5 = moderate, and 0.8 = large). ¹⁸ Using the Statistical Package for the Social Sciences (SPSS), an inverse variance approach was used to assign weights to our studies to account for precision, and all studies were included to produce a single cumulative ES with 95% CIs, depicted as a forest plot.

Heterogeneity

Although the patients in each study were randomly allocated, inherent differences between the patients in each study may influence the treatment effect. Thus, the I² method was employed to test for heterogeneity, and the random effect model was used if I² > 0.5. Of note, the average heterogeneity for a meta-analysis using less than 20 studies has been reported as 0.26 to 0.87. ²⁹

Metaregression Analysis

Metaregressions, represented as bubble plots, were conducted to account for heterogeneity or determine underlying causes of study bias. Two regressions were performed for each scale, with the first covariable being percentage female, and the second being age (years). A R² = 0 means no relationship, R² < 0.3 means very weak, R² = 0.3 to 0.5 means weak, R² = 0.5 to 0.7 means moderate, and R² > 0.7 means strong. ⁴¹

Publication Bias

Multiple methods of imputation were used to account for each study’s data; however, there is still the possibility of publication bias among smaller studies. Thus, we provide a publication bias analysis presented as funnel plots to evaluate for this possibility. For ease of interpretation, the inverse of SE was assigned the vertical axis and the ES was assigned the horizonal axis.

Cost Analysis

Costs in US dollars were derived using the most recent Medicare reimbursement guidelines for procedures and medications, dated April 1, 2022 through June 30, 2022.^14,15 The national average cost for both a nonfacility (ie, outpatient office) large bursa injection (ie, subacromial injection) with and without ultrasound guidance were extracted. The payment limit amounts for each medication were extracted.

Results

Study Selection

A total of 323 potentially relevant studies was identified. After removal of 36 duplicate studies, the remaining 287 were screened by title and 276 were removed due to nonrelevance to SIS. Abstracts were reviewed for the remaining 11 studies, and 1 was further excluded secondary to study design. Therefore, 10 studies met the inclusion criteria for subsequent analysis per the PRISMA guidelines (Figure 1).

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 flow diagram for new systematic reviews, which included searches of databases.^40,46 For more information, visit: http://www.prisma-statement.org/.

Assessment of Methodologic Quality and Risk of Bias

Nine studies were double blinded and 1 was not blinded (Aksakal et al ³ ). Three studies used unclear allocation concealment methods (Çift, ¹⁷ Abolhasani et al, ¹ Goyal et al ²⁵ ). One study (Aksakal et al ³ ) had a high risk of bias due to a lack of allocation concealment, blinding of the participants and blinding in the outcome assessment (Figure 2).

Figure 2.

Cochrane Collaboration’s tool for assessing risk of bias, where (-) or green = no risk, (?) or yellow = unclear risk, and (+) or red = high risk.

A total of 3 studies were of good quality, and 7 were of high quality. The mean score was 21.1, with a low grade of 16.5 and a high grade of 25 (Table 1).

Table 1.

Study characteristics (type and dosage of medication used, total patients, total patients per group, mean age, dropout rate, major inclusion and exclusion modality, duration of intervention, adverse effects and CONSORT score)

Study (Author, Year, Reference)	Type and Dosage of NSAID (N) and Corticosteroid (C)	Total Patients (M:F) a	Total Patients per Group (M:F) a	Mean Age (years) ± SD	Dropout Rate (n)	Major Inclusion Modality	Major Exclusion Modality	Duration of Intervention (weeks)	Adverse Effects	CONSORT Score (x/25)
Karthikeyan et al (2010) 30	N: Tenoxicam 20 mg C: Methylprednisolone 40 mg	56 (31:25)	N: 30 (15:15) C: 26 (16:10)	N: 58.0 ± 9.8 C: 60.0 ± 13.0 Total: 58.9 ± 11.3	N: 1 C: 1 Total: 2	Physical examination	x-ray	6	0	22
Min et al (2013) 38	N: Ketorolac 60 mg C: Triamcinolone 40 mg	32 (25:7)	N: 17 (13:4) C: 15 (12:3)	N: 39.6 ± 9.4 C: 39.1 ± 10.5 Total: 39.4 ± 9.8	N: 7 C: 9 Total: 16	Physical examination	x-ray ± MRI	4	0	25
Çift (2015) 17	N: Tenoxicam 30 mg C: Methylprednisolone 40 mg	40 (18:22)	N: 20 (10:10) C: 20 (8:12)	N: 45.3 ± 8.8 C: 46.5 ± 11.0 Total: 45.9 ± 9.9	0	Physical examination or MRI	x-ray ± MRI	6	N: 2 C: 0	16.5
Aksakal et al (2017) 3	N: Lornoxicam 8 mg C: Betamethasone 9 mg	70 (26:44)	N: 35 (12:23) C: 35 (14:21)	N: 53.0 ± 5.5 C: 53.0 ± 5.3 Total: 53.0 ± 5.4	0	Physical examination and MRI	x-ray	2, 4, and 6	0	20.5
Taheri et al (2017) 59	N: Ketorolac 60 mg C: Methylprednisolone 40 mg	40 (17:23)	N: 20 (8:12) C: 20 (9:11)	N: 49.8 ± 4.8 C: 47.5 ± 6.9 Total: 48.7 ± 6.0	0	Physical examination or MRI	x-ray ± MRI	4 and 12	0	23
Yu et al (2018) 66	N: Lornoxicam 8 mg C: Triamcinolone 40 mg	41 (21:20)	N: 19 (7:12) C: 22 (14:8)	N: 37.2 ± 10.1 C: 44.7 ± 9.5 Total: 41.2 ± 10.4	N: 6 C: 5 Total: 11	Physical examination or MRI	x-ray ± MRI	6	N/A	19
Abolhasani et al (2019) 1	N: Ketorolac 60 mg C: Betamethasone 8 mg	70 (31:39)	N: 35 (16:19) C: 35 (15:20)	N: N/A C: N/A Total: 37.4 ± 9.5	0	Physical examination	x-ray	4 and 6	N/A	22
Goyal et al (2020) 25	N: Ketorolac 60 mg C: Methylprednisolone 40 mg	67 (24:43)	N: 34 (14:20) C: 33 (10:23)	N: 51.6 ± 13.2 C: 52.7 ± 11.8 Total: 52.1 ± 12.4	N: 1 C: 2 Total: 3	Physical examination	x-ray	4 and 12	0	21.5
Kim et al (2021) 31	N: Ketorolac 30 mg C: Triamcinolone 40 mg	60 (41:19)	N: 30 (19:11) C: 30 (22:8)	N: 66.6 ± 6.0 C: 68.8 ± 6.0 Total: 67.7 ± 6.1	0	Physical examination	x-ray ± MRI	3, 6, and 12	N: 0 C: 2	22.5
Siddique et al (2021) 54	N: Ketorolac 60 mg C: Methylprednisolone 40 mg	218 (117:101)	N: 109 (54:55) C: 109 (63:46)	N: 39.1 ± 9.9 C: 38.1 ± 8.6 Total: 38.6 ± 9.3	0	Physical examination	x-ray	6	0	19
Total		694 (351:343)	N: 349 (168:181) C: 345 (183:162)	Total: 46.1 ± 13.3	N: 15 C: 17 Total: 32				N: 2 C: 2 Total: 4	21.1

CONSORT, Consolidated Standards of Reporting Trials; M:F, number of men and women; MRI, magnetic resonance imaging; NSAID, nonsteroidal anti-inflammatory drug.

^aAfter accounting for dropouts.

Study Characteristics

The total number of participants across the 10 RCTs was 694 (351 men and 343 women). The number of individuals who received an SNI was 349 (168 men and 181 women), whereas the number of individuals who received an SCI was 345 (183 men and 162 women). The mean age for the 10 studies was 46.1 ± 13.3 years (Table 1).

All studies utilized physical examination as the primary tool for diagnosing SIS. Specifically, each study considered a positive Neer test and/or Hawkins sign as being indicative of SIS. Every study employed imaging to aid in participant inclusion and exclusion. Four studies included participants with magnetic resonance imaging (MRI) confirmed SIS, while 5 used MRI to exclude other pathologies such as rotator cuff tears. Each study utilized x-ray to exclude other diagnoses such as fractures or arthritis (Table 1).

Duration of intervention varied among studies, from 2 to 12 weeks. Five studies measured follow-up results at more than 1 time frame. Seven studies reported outcome at 6 weeks, 5 at 4 weeks, 3 at 12 weeks, and 1 at 2 weeks (Table 1).

Dropout rates were reported for all 10 studies. Six had no loss of participants. Of the remaining 4 studies, there were no statistical differences between both groups. Dropout was either due to loss of follow-up (n = 19), an alternative diagnosis being discovered (n = 10), expiration of consent (n = 1), or participants not being followed up in time (n = 2). A total of 32 participants dropped out across the 4 studies, with 15 belonging to the SNI group and 17 to the SCI group (Table 1).

Eight of the 10 RCTs reported postinjection adverse effects with 6 reporting no cases. Of the remaining 2 studies, 1 (Çift) ¹⁷ had 2 patients in the SNI group who experienced temporary hypotension that was not quantified, whereas the other (Kim et al ³¹ ) had 2 patients in the SCI group who suffered from facial flushing for 1 week. Each patients’ symptoms resolved without additional treatment (Table 1).

Formulation and Dosing

There were differences in the dose and type of medications utilized. Six of the 10 studies had ketorolac as their NSAID, with 5 using 60 mg and 1 using 30 mg. Lornoxicam 8 mg was the NSAID for 2 studies. Two studies employed the NSAID tenoxicam, with dosages of 20 and 30 mg, respectively. Five of the studies used 40 mg of methylprednisolone as the corticosteroid. Triamcinolone was the corticosteroid for 4 studies, with 3 using 40 mg and 1 using 80 mg. Two studies utilized the corticosteroid betamethasone, with a dosage of 7 mg and 6 mg, respectively (Table 1). Nine studies had a single SNI and SCI group, whereas 1 had a single NSAID group but 2 separate, different corticosteroid groups.

Patient-Reported Outcomes

Eight studies evaluated pain with the VAS. Five demonstrated small ESs, 1 produced a moderate ES, and 2 resulted in large ESs. Small ESs favoring SNIs were seen in Kim et al ³¹ and Min et al, ³⁸ which compared ketorolac with triamcinolone. Two (Taheri et al, ⁵⁹ Goyal et al ²⁵ ) of the 3 studies that compared ketorolac 60 mg with methylprednisolone 40 mg produced ESs favoring SCIs, small and large, respectively, whereas the other (Siddique et al ⁵⁴ ) produced a moderate ES for SNIs. A large ES for SNIs was seen in Çift, ¹⁷ which compared tenoxicam with methylprednisolone. The lone study (Yu et al ⁶⁶ ) comparing lornoxicam with triamcinolone resulted in a small ES favoring SCIs (Table 2).

Table 2.

MD ± SD and effect sizes (Cohen d) for the VAS, the CMS, and the UCLA

Study (Author, Year, Reference)	SNI (MD ± SD)	SCI (MD ± SD)	Effect Size (Cohen d)
VAS
Min et al (2013) 38	1.83 ± 2.25	0.90 ± 1.86	0.45	Small for SNI
Çift (2015) 17	5.20 ± 0.60	2.60 ± 1.80	1.94	Large for SNI
Taheri et al (2017) 59	4.15 ± 1.85	4.55 ± 2.44	-0.18	Small for SCI
Yu et al (2018) 66	3.70 ± 1.65	3.50 ± 1.65	0.12	Small for SNI
Abolhasani et al (2019) 1	4.75 ± 2.13	4.51 ± 1.97	0.12	Small for SNI
Goyal et al (2020) 25	4.05 ± 1.24	5.25 ± 1.12	-1.02	Large for SCI
Kim et al (2021) 31	5.70 ± 1.84	5.60 ± 1.75	0.06	Small for SNI
Siddique et al (2021) 54	3.54 ± 1.64	2.41 ± 1.47	0.73	Moderate for SNI
CMS
Karthikeyan et al (2010) 30	8.63 ± 34.65	22.76 ± 30.44	-0.43	Small for SCI
Aksakal et al (2017) 3	13.50 ± 62.24	31.30 ± 36.90	-0.35	Small for SCI
Taheri et al (2017) 59	31.59 ± 29.60	23.00 ± 23.91	0.32	Small for SNI
Siddique et al (2021) 54	34.90 ± 12.85	24.68 ± 10.91	0.86	Large for SNI
UCLA
Min et al (2013) 38	7.15 ± 6.10	2.13 ± 6.14	0.82	Large for SNI
Aksakal et al (2017) 3	0.55 ± 0.91	1.30 ± 1.81	-0.52	Moderate for SCI
Yu et al (2018) 66	9.50 ± 3.78	9.70 ± 3.08	-0.06	Small for SCI
Kim et al (2021) 31	16.40 ± 5.48	15.30 ± 5.10	0.21	Small for SNI

CMS, Constant-Murley scale; MD, mean difference; SCI, subacromial corticosteroid injection; SNI, subacromial nonsteroidal anti-inflammatory injection; VAS, visual analog scale; UCLA, University of California Los Angeles.

The CMS was utilized to evaluate change in function in 4 studies. One resulted in a large ES, whereas the other 3 were small. The studies by Taheri et al ⁵⁹ and Siddique et al ⁵⁴ reported ESs favoring NSAIDs, small and large, respectively. The studies by Karthikeyan et al, ³⁰ which utilized tenoxicam 20 mg and methylprednisolone 40 mg, and Aksakal et al, ³ which compared lornoxicam with betamethasone, demonstrated small ESs for SCIs (Table 2).

Four studies reported the UCLA as a measure of function as well. Min et al ³⁸ and Kim et al ³¹ produced ESs in favor of SNIs, large and small, respectively. The papers by Aksakal et al ³ and Yu et al ⁶⁶ reported ESs for SCIs, moderate and small, respectively (Table 2).

Meta-analysis and Heterogeneity

For the VAS, 8 papers were initially included, with a cumulative random ES of 0.26 (P = 0.38) and a considerable amount of heterogeneity (I² = 0.90). After the removal of potential outliers, a new forest plot was generated for 7 ESs (Figure 3) and the cumulative random ES equaled 0.05 (P = 0.83), with a substantial, but now acceptable, amount of heterogeneity (I² = 0.81). This indicates a small ES favoring NSAIDs that is not statistically significant. In addition, the 95% CI crosses 0, supporting the possibility of no difference.

Figure 3.

Metaanalysis of the visual analog scale (VAS) (top), the Constant-Murley scale (CMS) (middle) and the University of California Los Angeles (UCLA) (bottom) represented as forest plots that show effect sizes (ESs) as Cohen d. The ESs are characterized by the standard interpretation (0.2 = small, 0.5 = moderate, and 0.8 = large), ²⁹ with right or positive numbers reflecting a preference for subacromial nonsteroidal anti-inflammatory injection (SNIs), whereas left or negative numbers reflect a preference for subacromial corticosteroid injections (SCIs).

Four papers were included for the CMS, and heterogeneity was substantial but acceptable (I² = 0.87). The cumulative random ES was 0.12 (P = 0.71), indicating a small ES favoring NSAIDs that is statistically insignificant. In addition, the 95% CI crosses 0, supporting the possibility of no difference (see Figure 3).

The UCLA analysis also included 4 studies with substantial but acceptable heterogeneity (I² = 0.71), and a cumulative random ES of 0.07 (P = 0.79). This indicates a small ES favoring NSAIDs that is not statistically significant. In addition, the 95% CI crosses 0, supporting the possibility of no difference (see Figure 3).

Metaregression

Among the VAS studies, weak and very weak inverse relationships between ES and percentage female (R² = 0.300) or age (R² = 0.035) were noted, respectively. Within the CMS analysis, no relationship was found with percentage female (R² = 0), and a strong inverse relationship was found with age (R² = 1). For the UCLA, a strong inverse relationship was seen with percentage female (R² = 1), while no relationship was seen with age (R² = 0) (Figure 4).

Figure 4.

Metaregression of the visual analog scale (VAS) (top), the Constant-Murley scale (CMS) (middle), and the University of California Los Angeles (UCLA) (bottom) for both percentage female and average age (years), represented in bubble plots.

Publication Bias

A slight asymmetric distribution toward positive Cohen d is seen within the VAS analysis, whereas symmetry is seen within the CMS and UCLA funnel plots. In addition, 2 studies (Siddique et al, ⁵⁴ Goyal et al ²⁵ ) for the VAS, 2 for the CMS (Siddique et al, ⁵⁴ Karthikeyan et al ³⁰ ), and 2 (Aksalkal et al, ³ Min et al ³⁸ ) for the UCLA were outside the 95% pseudo CI. However, for each scale, 1 was left of the negative margin and the other right of the positive margin (Figure 5).

Figure 5.

Publication bias analysis of the visual analog scale (VAS) (top), the Constant-Murley scale (CMS) (middle), and the University of California Los Angeles (UCLA) (bottom), represented by funnel plots, where right or positive numbers reflect a preference for subacromial nonsteroidal anti-inflammatory injection (SNIs), whereas left or negative numbers reflect a preference for subacromial corticosteroid injections (SCIs).

Cost Analysis

The cost of an in office subacromial injection without ultrasound guidance (CPT 20610) is $66.44, while the cost with ultrasound guidance (CPT 20611) is $102.09. Injectable ketorolac 15 mg (HCPCS J1885) is $0.47, methylprednisolone 40 mg (HCPCS J1030) is $5.91, betamethasone 3 mg (HCPCS J0702) is $6.79, and triamcinolone 10 mg (HCPCS J3301) is $1.12. Tenoxicam and lornoxicam were not listed under these guidelines (Table 3).

Table 3.

Cost of a subacromial injection, and the NSAIDs or corticosteroids used in them based on the most recent Medicare data^14,15

Procedure a	CPT Code	Price (USD)
Major joint or bursa injection or aspiration without ultrasound guidance	20610	66.44
Major joint or bursa injection or aspiration with ultrasound guidance	20611	102.09
Medication b	HCPCS code	Price (USD)
Ketorolac 15 mg	J1885	0.47
Tenoxicam	Not available	Not available
Lornoxicam	Not available	Not available
Methylprednisolone 40 mg	J1030	5.91
Betamethasone 3 mg	J0702	6.79
Triamcinolone 10 mg	J3301	1.12

NSAID, nonsteroidal anti-inflammatory drug; USD, US dollars.

^aNational payment amount for procedures performed in a non-facility setting.

^bPayment limit amounts.

Discussion

SIS is the most common cause of shoulder pain seen in primary care and orthopaedic clinics. ³⁹ It often affects those older than 40 years, ²⁴ which is consistent with the mean age of patients in this review. The studies included had an overall high CONSORT score and low Cochrane risk of bias, which increases the confidence in our results and their interpretation. The results suggest that SNIs are as effective as SCIs for pain reduction and improving function in patients with SIS. In addition, SNIs appear to be the cheaper option and cause no major difference in side effects.

No consensus exists on the optimal medication for subacromial injections in patients with SIS; thus, there is considerable variability among clinicians performing these injections. The relative potency of many injectable corticosteroids has been determined chemically (Table 4), ³⁶ and based on this measure, each study used an equivalent dose of corticosteroid for their SCI group, despite using different medications. The only exception was in the study of Abolhasani et al, ¹ in which 1 of the 2 corticosteroid groups utilized triamcinolone 80 mg, which has double the potency of the other medications. Thus, the results from that cohort were omitted from our analysis. This equivalence is further supported clinically by several trials that demonstrated similar pain reduction between these corticosteroids in patients with knee osteoarthritis.^5,13,49,61 Unfortunately, no such comparison has been developed for injectable NSAIDs. However, 1 study showed that 8 mg of lornoxicam and 30 mg of ketorolac caused equivocal reduction in VAS among patients undergoing abdominal surgery, ⁴² and another study demonstrated no significant difference in pain control between ketorolac 30 mg and 60 mg following knee surgery. ⁵⁵ It has also been shown that VAS scores were reduced by the same amount in patients with renal colic who received either 20 mg of tenoxicam or8 mg of lornoxicam. ¹⁶ Based on these results, the studies used comparable doses of NSAIDs.

Table 4.

Potency and relative dosage of injectable steroids ³⁶

Corticosteroid	Potency	Equivalent Dosing (mg)	Injectable (mg/mL)
Hydrocortisone	1	20	200
Betamethasone	25-35	0.6-0.8	6-8 a
Triamcinolone	5	4	40
Methylprednisolone	5	4	40

^aBetamethasone is usually administered as a formula containing 3 mg of betamethasone sodium phosphate with 3 mg of betamethasone acetate or 5 mg of betamethasone dipropionate with 2 mg of betamethasone sodium phosphate.

Similarly, no consensus exists on the ideal follow-up period postinjection. Four to 6 weeks was chosen as it has been suggested as an optimal time frame to determine the success of musculoskeletal injections,^21,45 and because each study included reassessed their participants at 4 or 6 weeks, allowing for consistent follow-up data.

The VAS is a tool developed to help assess pain and is highly reliable for acute and chronic musculoskeletal pain.^6,7,9 Our VAS meta-analysis produced a small ES favoring NSAIDs, but with a high P value (0.83); thus, the results are both clinically and statistically insignificant. The study by Cift ¹⁷ was removed as a potential outlier, with unclear allocation concealment methods as a potential cause for biased data, resulting in acceptable heterogeneity. This suggests that SNIs are similar to SCIs in pain relief for SIS. In addition, the metaregression suggests that a difference in treatment effect may exist based on gender, specifically that women do not respond as well to SNIs as men. To a lesser extent, a similar relationship may exist with age, in which older patients do not respond as well to SNIs as younger patients. However, the small number of studies in our regressions limit the power of these analyses, and further studies need to be conducted to draw stronger conclusions. The asymmetry toward positive ESs in our VAS funnel plot is concerning for underlying publication bias; however, as fewer than 10 studies were included, this could be due to chance, and firm conclusions cannot be drawn. ⁵⁶ The only large ES for SNIs was produced by tenoxicam, ¹⁷ which may signify that it is a superior NSAID for pain relief in SIS. In addition, both studies^31,38 comparing ketorolac to triamcinolone resulted in small ESs toward SNIs, while the majority^59,25,55 comparing ketorolac with methylprednisolone produced ESs favoring SCIs. This could mean that methylprednisolone is better than triamcinolone for pain control in SIS.

Tools such as the CMS and the UCLA combine subjective and objective measures to determine overall shoulder functionality. ⁶ Both have high construct validity, with the interrater reliability of the CMS being very good. ⁶⁴ Our meta-analyses of both the CMS and UCLA data produced ESs that were small toward NSAIDs, but with high P values (0.71 and 0.79); therefore, the results are clinically and statistically insignificant. The study by Siddique et al ⁵⁴ was explored as a possible outlier in both analyses. However, no potential sources of bias were determined by Cochrane analysis, and it was the most robust study, having the largest number of participants. In addition, there was an acceptable heterogeneity for both scales; thus, the study was not removed. Taken together, these analyses suggest that SNIs are similar to SCIs in improving function in patients with SIS. In addition, the metaregression for the CMS suggests no relationship exists for gender, but one for age, the same as seen with VAS, does. Meanwhile, the regression for UCLA suggests the opposite, that no relationship exists with age, but one for gender, the same as seen with VAS, does. However, as with our VAS regressions, the small number of studies included make these conclusions relatively weak. Funnel plots for the CMS and the UCLA were symmetric, suggesting no publication bias, but as fewer than10 studies were included, this could be due to chance alone. ⁵⁶ In terms of ESs, both studies^55,59 in the CMS analysis that compared ketorolac with methylprednisolone produced ESs favoring SNIs, which contrasts the results seen with the VAS. This could indicate that methylprednisolone is more effective than ketorolac for pain reduction, but less effective for improving function. Meanwhile, in the UCLA analysis, the studies by Min et al ³⁸ and Kim et al ³¹ also favored SNIs, as seen in the VAS analysis, suggesting that ketorolac is superior to triamcinolone for both pain reduction and improving function. Interestingly, all studies utilizing oxicams^3,30,66 for the CMS or UCLA favored SCIs. This contrasts with the VAS analysis for tenoxicam but is in line with the results seen for lornoxicam, suggesting that tenoxicam may be a better choice than corticosteroids for decreasing pain but a poorer choice when trying to improve functionality. Lornoxicam is a poor choice in general.

As per the most recent Medicare reimbursement guidelines,^14,15 both SNIs and SCIs are billed with the same procedure code, assuming they are performed in the same fashion, with ultrasound guidance or without; thus, the medication price dictates cost differences. Certain injections use dosages higher than the billed amount; therefore, multiple instances of that code are required for proper cost determination. For example, for an SNI of ketorolac 30 mg, 2 instances of ketorolac 15 mg would be billed, for a total cost of $0.94 instead of $0.47. Even when accounting for these differences, NSAIDs are cheaper than corticosteroids for Medicare patients, and likely those with other insurances as well.

No significant difference in side effects were present among the 10 RCTs. While little is known about the side effect profile of SNIs, these results contrast with current literature on SCIs, which report many complications.^53,58 However, the lack of effects or difference seen may be due to (1) the short follow-up period or (2) a lack of protocols being implemented to identify complications, such as monitoring blood glucose or renal function.

There are 3 previously published systematic reviews comparing nonsurgical treatment modalities in SIS.^20,57,62 Our review builds on previous ones for several reasons. First, none of the previous reports exclusively compared SNIs with SCIs, as trials utilizing injectable, oral, or topical NSAIDs versus SCIs were included. Second, these 3 reviews predate 7 of the 10 RCTs included in this paper, permitting a more comprehensive analysis of the question, including a metaanalysis. Third, to the best of our knowledge, we are the first to conduct a cost analysis of these injections.

This systematic review has significant limitations. First, heterogeneity exists in the medications used by the studies included. While equivalence can be shown chemically and clinically, trials that simultaneously evaluate several NSAIDs and corticosteroids would be ideal to discern the optimal medication and dosage. Second, inference on the efficacy of SNIs versus SCIs over longer periods of time cannot be made as this review is focused on 4 to 6 weeks postinjection. Finally, we assume a normal distribution of population for the studies, which is highly unlikely; larger-scale trials could minimize this assumption and the likely resulting bias.

Conclusion

This systematic review and meta-analysis suggest that SNIs are as effective as SCIs for short-term pain reduction and improving function in patients with SIS. In addition, the cost analysis demonstrated that SNIs are cheaper than SCIs when prices and dosing are standardized. Finally, there were no major detected differences in complications between these injections. Therefore, SNIs should be thought of as a viable, cost effective alternative to SCIs when injection therapy is being considered for a patient with SIS.