Single-Site Corticosteroid Injection Is as Effective as Multisite Corticosteroid Injection in the Nonsurgical Treatment of Frozen Shoulder: A Systematic Review With Meta-Analysis of Randomized Controlled Trials

Two authors independently searched the Embase, PubMed, and Cochrane Library databases from inception to June 5, 2022, and the reference lists of published systematic reviews for relevant studies. The search specifics were as follows: ‘‘(((((Multisite) OR (sites)) OR (dual-target)) OR (two targets)) AND (((((((((corticosteroid) OR (glucocorticoid)) OR (triamcinolone)) OR (methylprednisolone)) OR (hydrocortisone)) OR (prednisolone)) OR (cortisone)) OR (dexamethasone)) OR (betamethasone))) AND ((((((((((((((((((((((((Bursitides) OR (Bursitis)) OR (Periarthritis)) OR (Frozen Shoulder)) OR (Frozen Shoulders)) OR (Shoulder, Frozen)) OR (Adhesive Capsulitis of the Shoulder)) OR (Shoulder Adhesive Capsulitis)) OR (Adhesive Capsulitides, Shoulder)) OR (Adhesive Capsulitis, Shoulder)) OR (Capsulitides, Shoulder Adhesive)) OR (Capsulitis, Shoulder Adhesive)) OR (Shoulder Adhesive Capsulitides)) OR (Capsulitis)) OR (Capsulitides)) OR (Pes Anserine Bursitis)) OR (Bursitides, Pes Anserine)) OR (Bursitis, Pes Anserine)) OR (Pes Anserine Bursitides)) OR (Adhesive Capsulitis)) OR (Adhesive Capsulitides)) OR (Capsulitides, Adhesive)) OR (Capsulitis, Adhesive)) OR (Stiff Shoulder)).’’ No language restrictions or study types were imposed.

The same 2 authors independently screened all titles and abstracts for relevance and eligibility. After the screening, chance-adjusted agreement was assessed by kappa value (0-0.20, poor agreement; 0.21-0.40, fair agreement; 0.41-0.60, moderate agreement; 0.61-0.80, good agreement; and 0.81-1.00, perfect agreement). A third author resolved any disagreements. Studies were reviewed if they met the following PICOS (patients, intervention, comparison, outcome, and study type) criteria:

P: Patients with FS;

I: Multisite corticosteroid injection;

C: Single-site corticosteroid injection;

O: Visual analog scale (VAS) score, ROM, American Shoulder and Elbow Surgeons (ASES) score, or Constant–Murley score (at least 1 outcome); and

S: Level I or II study.

The exclusion criteria were as follows: (1) animal study; (2) cell study; (3) short communication or conference abstracts; and (4) intervention that did not involve steroid injections.

The same 2 authors used the Levels of Evidence for Primary Research Question to assess literature quality and the Modified Coleman Methodology Score (MCMS). The MCMS has a scaled potential score ranging from 0 to 100 to evaluate inclusion criteria, sample size calculation, randomization, follow-up, patient analysis, blinding, similarity in treatment, treatment description, group comparability, outcome assessment, description of rehabilitation protocol, clinical effect measurement, and the number of patients treated. A score of 85 to 100 means excellent, 70 to 84 means good, 55 to 69 means fair, and less than 55 means poor. The kappa score, which evaluates the degree of agreement between authors, was calculated.

The Cochrane Collaboration risk-of-bias tool was used to evaluate the risk of bias in the included studies; it contains the following domains: bias of random sequence generation (selection bias), bias of allocation concealment (selection bias), bias of blinding participants and personnel (performance bias), bias of blinding outcome assessment (detection bias), bias of missing outcome data (attrition bias), bias of selective reporting (reporting bias), and other bias. The same 2 authors independently assessed the bias of the included RCTs by scoring them as low, unclear, or high risk. Any discrepancies were resolved by discussion, and the third reviewer made the final decision.

Two same authors independently collected available data from the included studies. The following essential characteristics were collected: author, year, journal, country, male sex, age, duration of symptoms, follow-up, Level of Evidence, inclusion criteria, injection material, injection content, injection site, ultrasonographic guidance, approach, and physiotherapy program. In addition, VAS pain scores, abduction, flexion, internal rotation, external rotation, ASES Assessment Form scores, Constant–Murley Shoulder scores, and complications were extracted as outcome measurements, and 4 time intervals of these measures were analyzed. We contacted the author to obtain missing data and extracted the mean value using Origin software (Version 2021; OriginLab Corp., Northampton, MA) when data were presented in figures.

This meta-analysis was performed with Review Manager, version 5.3 (The Cochrane Collaboration,). Statistical heterogeneity was assessed with I² statistics as follows: 0% < I² < 25%, unimportant heterogeneity; 25% < I² < 50%, moderate heterogeneity; and I² > 50%, important heterogeneity. We used a random-effects model for all comparisons because disease phases increase the risk of heterogeneity. The treatment effects of all continuous were measured by mean differences (MDs) with 95% confidence intervals (95% CIs). Dichotomous were measured by risk ratios and 95% CIs. If the comparisons with more than 1 met eligible intervention groups, the control group was divided into more groups with a smaller sample size that allowed all suitable comparisons to be included. If the outcome measures were reported as the mean and 95% CI, standard deviation (SD) values were estimated using “Finding the Standard Deviation using Confidence Intervals” in the Excel version of the RevMan Calculator (Microsoft, Redmond, WA). When the outcome measures were reported in the mean and standard error of the mean, SD values were estimated with the following formula: SD = standard error of the mean × sqrt(n), where sqrt is the square root and n is the number of participants.

In all analyses, a P value of .05 was considered statistically significant. Data analyses were performed for the following intervals: (1): 3 to 4 weeks; (2): 6 to 8 weeks; (3): 12 to 16 weeks, and (4): 24 to 26 weeks. When the number of included studies was less than 10, publication bias was not considered. To assess the robustness of the effect sizes, we performed a sensitivity analysis by extracting all high heterogeneity results that synthesized more than 2 studies during 4 time intervals

The results of the initial search yielded 260 studies (PubMed = 21, Embase = 126, Cochrane = 113). After the removal of 27 duplicates, 233 studies remained, and 5 were deemed eligible for further screening. Thus, 5 studies were carefully reviewed.^{,  ,  ,  ,}  However, one study was a short communication that did not meet our inclusion criteria, and one additional study was identified from the citation search. Finally, 5 RCTs were included in this review^{,  ,  ,  ,}  (Fig 1). The kappa score was 0.88, indicating perfect agreement.

All of the studies were published in different journals. Of the 5 studies, 2 RCTs^, were from South Korea, and the others were from Norway, India, and Turkey. The minimum follow-up time was 12 weeks. There were three Level I and two Level II studies (Table 1).

All studies included patients with shoulder pain and limited motion. Specifically, one study inclusion criteria were pained with limitation of both active and passive shoulder movements in at least 2 directions (forward flexion <120° or 50% restriction of contralateral external rotation and internal rotation). In addition, patients in two studies were assessed for pain and passive restriction of shoulder motion.^, One study did not report the specific restriction, whereas another study reported that inclusion criteria for patients had lost more than 20% of their shoulder movements in all directions. Cho et al. used 2 different length needles (3 cm and 6 cm) for intra-articular and subacromial injection respectively. One study used a 16-gauge needle, and another study only reported 20-mL needles. Three studies used a 40-mg dose for injection,^{,  ,}  and a 20-mg dose was used in one study. A sham injection was performed in one study. In addition, one RCT used a 40-mg dose for single-site injection and 80 mg for multisite injection. In this study, injection sites include the glenohumeral joint, posteroinferior capsule, subacromial space, posterosuperior capsule, biceps long head, and area around the coracohumeral ligament. Multisite injection was selected for the glenohumeral joint combined with the subacromial space in 2 studies.^, Three injection sites were selected in one RCT, and 4 sites were selected in one study. Ultrasound-assisted injection was reported in 4 studies.^,,, Except for 2 studies^,, reporting 2 approaches of multisite injection, all injection approaches were posterior approaches. Three RCTs reported the combination of physical therapy and injections (Table 2).^,,

According to the MCMS, there were 3 excellent quality studies^,, and 2 good-quality studies.^, Only one study obtained a score for follow-up, reducing the variability among studies. Two studies^, received fair scores in the description of the surgical procedure, and 2 studies^, did not receive any points for postoperative rehabilitation that may hinder the clinical interpretation of the results. Only 1 study obtained a perfect score in assessing outcomes that enhanced the efficacy of the clinical results. However, the scores of this study were reduced in the description of the subject selection process due to the long assessment period and the small number of patients lost to follow-up (Table 3). There was a very good agreement between authors according to the kappa score (0.88).

All 5 studies had a low risk of bias in random sequence generation and allocation concealment. One study was a single-blind clinical study, which increased the risk of performance bias. There was no detailed description of the blinding method used in the process in the 3 studies,^,, and there was an unclear risk of performance bias and detection bias. Pushpasekaran et al. only reported Constant-Murley score, and they did not report total structured values, such as SD or standard error and other outcomes. Thus, this study was rated as having a high risk of attrition bias and reporting bias. Finally, 3 studies did not report the experience of the injectors, indicating that they had unclear risks^,, (Fig 2).

Four studies^,,, reported VAS scores at 3 to 4 weeks, and one study presented the results in figures (Appendix Table 1, available at www.arthroscopyjournal.org). The results revealed that there were no statistically significant differences in VAS scores (MD 1.19 [–0.05 to 2.43], P = .06), and the heterogeneity was high (I² = 90%; P < .00001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data (Fig 3).

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, showed no significant differences between multisite group and single-site group for VAS scores (MD 0.77 [–0.46 to 2.01], P = .22), and the heterogeneity was 85%. (Appendix Figure 1, available at www.arthroscopyjournal.org)

Three studies^,, reported VAS scores at 6 to 8 weeks, and one study presented the results in figures. The results revealed that there were no statistically significant differences in VAS scores (MD 0.38 [–0.66 to 1.41], P = .48), and the heterogeneity was high (I² = 77%; P = .01). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

Four studies^,,, reported VAS scores at 12 to 16 weeks. The results revealed that there were no statistically significant differences in VAS scores (MD 0.54 [–0.10 to 1.17], P = .10), and the heterogeneity was high (I² = 83%; P < .0001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data. When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 3 studies^,, showed no significant differences between multisite group and single-site group for VAS scores (MD 0.20 [–0.08 to 0.48], P = .17), and the heterogeneity was 16%.

Three studies^,, reported VAS scores at 24 to 26 weeks, and one study presented the results in figures. The results revealed that there were no statistically significant differences in VAS scores (MD 0.50 [–1.26 to 2.27], P = .58), and the heterogeneity was high (I² = 87%; P = .006). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

Three studies^,, reported abduction at 3 to 4 weeks (Appendix Table 2, available at www.arthroscopyjournal.org). The results revealed that the multisite group had better abduction than the single-site group (MD –15.66 [–30.03 to –1.28], P = .03), and the heterogeneity was high (I² = 83%; P = .0006). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data (Fig 4).

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, showed no significant differences between multisite group and single-site group for abduction (MD –11.07 [–26.20 to 4.07], P = .15), and the heterogeneity was 80% (Appendix Figure 2, available at www.arthroscopyjournal.org).

Two studies^, reported abduction at 6 to 8 weeks. The results revealed that there were no statistically significant differences in abduction (MD –6.65 [–16.38 to 3.07], P = .18), and the heterogeneity was high (I² = 63%; P = .07). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

Three studies^,, reported abduction at 12 to 16 weeks. The results revealed that there were no statistically significant differences in abduction (MD –13.35 [–28.61 to 1.90], P = .09), and the heterogeneity was high (I² = 85%; P = .0001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data. When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, showed no significant differences between multisite group and single-site group for abduction (MD –5.68 [–12.34 to 0.97], P = .09), and the heterogeneity was 4%.

Two studies^, reported abduction at 24 to 26 weeks. The results revealed that there were no statistically significant differences in abduction (MD –15.11 [–51.44 to 21.23], P = .42), and the heterogeneity was high (I² = 91%; P = .0007). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

Four studies^,,, reported flexion at 3 to 4 weeks, and one study presented the results in figures (Appendix Table 3, available at www.arthroscopyjournal.org). The results revealed that there were no statistically significant differences in flexion (MD –12.21 [–24.49 to 0.08], P = .05), and the heterogeneity was high (I² = 85%; P = .0002). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data (Fig 5).

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, showed no significant differences between multisite group and single-site group for flexion (MD –7.93 [–20.11 to 4.25], P = .20), and the heterogeneity was 79% (Appendix Figure 3, available at www.arthroscopyjournal.org).

Four studies^,,, reported flexion at 6 to 8 weeks, and one study presented the results in figures. The results revealed that there were no statistically significant differences in flexion (MD –11.55 [–24.69 to 1.60], P = .09), and the heterogeneity was high (I² = 88%; P < .0001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, showed no significant differences between multisite group and single-site group for flexion (MD –5.68 [–14.61 to 3.13], P = .20), and the heterogeneity was 63%.

Four studies^,,, reported flexion at 12 to 16 weeks, and one study presented the results in figures. The results revealed that there were no statistically significant differences in flexion (MD –8.19 [–21.17 to 4.89], P = .22), and the heterogeneity was high (I² = 86%; P < .0001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, showed no significant differences between multisite group and single-site group for flexion (MD –2.51 [–12.50 to 7.47], P = .09), and the heterogeneity was 69%. Two studies^, reported flexion at 24 to 26 weeks, and one study presented the results in figures.

Four studies^,,, reported external rotation at 3-4 weeks, and one study presented the results in figures (Appendix Table 4, available at www.arthroscopyjournal.org). The results revealed that there were no statistically significant differences in external rotation (MD –7.85 [–16.87 to 1.17], P = .09), and the heterogeneity was high (I² = 90%; P < .00001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data (Fig 6).

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, favor multisite group for external rotation (MD –11.31 [–18.71 to –3.92], P = .003), and the heterogeneity was 76% (Appendix Figure 4, available at www.arthroscopyjournal.org).

Four studies^,,, reported external rotation at 6-8 weeks, and one study presented the results in figures. The results revealed that there were no statistically significant differences in external rotation (MD –7.83 [–18.46 to 2.79], P = .15), and the heterogeneity was high (I² = 93%; P < .00001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, favor multisite group for external rotation (MD –11.76 [–20.71 to –2.81], P = .010), and the heterogeneity was 85%.

Four studies^,,, reported external rotation at 12 to 16 weeks, and one study presented the results in figures. The results revealed that there were no statistically significant differences in external rotation (MD –6.95 [–18.04 to 4.14], P = .22), and the heterogeneity was high (I² = 92%; P < .00001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, favor multisite group for external rotation (MD –11.19 [–20.30 to –2.08], P = .02), and the heterogeneity was 84%.

Two studies^, reported external rotation at 24 to 26 weeks, and one study presented the results in figures.

Two studies^, reported internal rotation at 3-4 weeks (Appendix Table 5, available at www.arthroscopyjournal.org). The results revealed that there were significant differences in internal rotation in favor of the multisite treatment (MD –12.80 [–19.26 to –6.34], P = .0001), and the heterogeneity was high (I² = 63%; P = .07). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data (Fig 7).

Two studies^, reported internal rotation at 6 to 8 weeks. The results revealed that there were significant differences in internal rotation in favor of the multisite treatment (MD –12.10 [–19.83 to –4.37], P = .002), and the heterogeneity was high (I² = 79%; P = .008). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

Two studies^, reported internal rotation at 12 to 16 weeks. The results revealed that there were significant differences in internal rotation in favor of the multisite treatment (MD –11.06 [–19.11 to –3.01], P = .007), and the heterogeneity was high (I² = 78%; P = .010). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.

Three studies^,, reported ASES scores at 3 to 4 weeks (Appendix Table 6, available at www.arthroscopyjournal.org).). The results revealed that there were significant differences in ASES scores in favor of the multisite treatment (MD –10.13 [–19.54, –0.72], P = .03), and the heterogeneity was high (I² = 87%; P < .00001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data (Fig 8).

When we excluded the study that caused the greatest heterogeneity due to bias, the pooled analysis of 2 studies^, showed no significant differences between multisite group and single-site group for ASES scores (MD –6.79 [–15.24 to 1.66], P = .12), and the heterogeneity was 80% (Appendix Figure 5, available at www.arthroscopyjournal.org).

Three studies^,, reported ASES scores at 6 to 8 weeks. The results revealed that there were no statistically significant differences in ASES scores (MD –7.46 [–17.45 to 2.53] P = .14), and the heterogeneity was high (I² = 88%; P < .00001). The result suggests that current statistics are underpowered, which makes it hard to draw strong inferences from the available data.