Randomized controlled trials (RCTs) of acupuncture, which were classified as a “needle stimulation” (penetration of the skin and the resultant de-qi sensation), regardless of the methods of stimulation or type of needles (e.g., electroacupuncture (EA), pharmacopuncture, thread-embedding therapy, and acupotomy), stimulating points (e.g., auricular points, scalp acupuncture points, and tender points), duration, or number of treatments were included. There were no restrictions by language, date of publication, settings, or locations of the study. Patients who had been diagnosed with RCDs based on clinical assessments with or without radiological evidence (e.g., ultrasound, magnetic resonance imaging) were included. The trials of patients with pain in the shoulder or upper extremity due to pathologies not related to the rotator cuff, such as fractures, osteoarthritis, adhesive capsulitis, or hemiplegic shoulders were excluded. There were no restrictions on age, sex, or ethnic origin. Control interventions included active treatments other than different types of acupuncture or acupoint-related treatments, placebo, or no treatment. Quasi-randomized or non-RCTs were excluded.

The collected data was classified for all the time points of assessment as short-term (up to 3 months post-randomization), intermediate-term (up to 6 months post-randomization), and long-term (more than 6 months post-randomization). The primary and secondary outcomes are presented in Table 1.

There were 12 electronic databases (Table 2) searched up to November 30^th, 2019. There were 3 trial registries checked, and references of the trials, and reviews. Study protocols or conference abstracts, were tracked to determine whether they had been published. The search strategies applied to each database are shown in Appendix A.

One reviewer searched for the studies. Two reviewers independently assessed eligibility of studies. Disagreements between the two review authors were discussed with a third reviewer. One reviewer extracted the relevant data other than information on risk of bias assessment. The details of the intervention were summarized based on the revised Standards for Reporting Interventions in Clinical Trials of Acupuncture (STRICTA) [13]. Two reviewers independently assessed the risk of bias, according to the Cochrane Handbook for Systematic Reviews of Interventions [14].

Dichotomous outcomes were presented as risk ratios (RR), and continuous outcomes as the mean difference (MD) or standardized mean differences (SMD) with 95% confidence intervals. Multiple acupuncture groups in a trial were combined for pairwise comparison. In case of multiple outcome measurements within the same time frame (short, intermediate, or long term), the last measurement data within that time frame was used. An attempt to contact the corresponding author was made in each study to obtain missing or unclear information; missing values were not imputed other than sensitivity analysis. For the assessment of heterogeneity, we calculated the I² statistic. An I² rate of less than 40% was categorized as low, 30–60% as moderate, 50–90% as substantial, and 75–100% as considerable heterogeneity [14]. RevMan software (Review Manager Version 5.4 for Windows, The Nordic Cochrane Centre, Copenhagen) was used to perform a meta-analysis. A random-effects model was used because substantial clinical heterogeneity amongst the included studies was expected. The pooled effect estimates of all the comparisons were reported, regardless of the observed heterogeneity. A predefined subgroup analysis was performed for (1) diagnosis of the study population (i.e., RC tear or without tear), (2) type of acupuncture stimulation (i.e., manual, electrical, or other stimulation techniques, such as pharmacoacupuncture, acupotomy, or thread-embedding therapy), where meta-analyses of primary outcomes included at least 6 studies. Sensitivity analyses were performed to determine whether the results of the trials were affected by trials where the radiological diagnosis of an RCD was not mentioned in the participant eligibility criteria, or the measures of variance were missing, and by different methods of analysis (random-effects model or fixed-effect model). Where the data of variance were missing, sensitivity analysis was conducted by imputing the largest standard deviation within the same meta-analysis.

The Grading of Recommendations Assessment, Development and Evaluation approach was adopted to assess the certainty of the evidence for each main outcome, and for reporting adverse events for the following domains: the risk of bias across studies, inconsistency, indirectness, imprecision, and publication bias [15]. The methods and recommendations described in the Cochrane Handbook for Systematic Reviews of Interventions were used for the judgments [14].

There were 3,686 records identified from English, Chinese, and Korean databases. After the removal of duplicates, the titles and abstracts of 2,654 records were screened. After screening 297 full-text articles, 28 RCTs were included which in total had 2216 participants in the qualitative syntheses, and 27 RCTs were included in the meta-analyses (Fig. 1). Contacting corresponding authors to verify the eligibility of the studies was not possible due to the absence of contact information in the published studies [16,17] or unanswered emails [18–21]. One RCT studying the feasibility of the Park sham needle was excluded because the purpose of the study was to assess the feasibility of the needle and it did not provide clinical outcome results [22]. One study was excluded from the meta-analysis due to errors on the sample size and an absence of reporting on variance of the outcomes [23]. One study was ongoing and was therefore not included in the this current review [24].

All studies were parallel-group RCTs. Seven trials had multiple acupuncture arms. Studies were conducted in China (n = 23), Spain (n = 2), Sweden (n = 1), Germany (n = 1), and Greece (n = 1). The number of participants in each study ranged from 30 to 178. The trials included patients with RC (i.e., supraspinatus) tendinitis (n = 9) [23,25–32], RC tear (n = 3) [33–35], RC injury (n = 5) [36–40], impingement syndrome (n = 7) [41–47], subacromial pain syndrome (n = 1) [48], subacromial bursitis (n = 1) [49], and nonspecific shoulder pain (n = 2) [50,51]. Two studies with nonspecific shoulder pain were included, since the eligibility criteria for the trials were deemed consistent with a clinical or radiological diagnosis of RCD [50,51]. Of the 28 trials, 13 studies used both clinical and radiological criteria for the diagnosis of RCD [23,27,29,33–36,39–41,43,49,51]. The types of acupuncture conducted in the 28 studies included manual acupuncture (n = 7) [25,27,33,40,45,49,50], dry needling of trigger points (n = 2) [48,51], a mixture of manual acupuncture and EA (n = 6) [28,34–37,46], EA (n = 2) [23,44], acupotomy (n = 10) [26,29–32,38,39,41–43], warm needle acupuncture (n = 1) [47], and fire needle acupuncture (n = 1) [30]. One RCT used two different styles of acupuncture arms (i.e., fire needle acupuncture, acupotomy, and fire needle acupuncture plus acupotomy) [30]. The details on the acupuncture characteristics of the included studies are presented in Table 3. A total of 15 trials reported both pain and shoulder function using validated scales. The most frequently used measurement scales for functional status of the shoulder were the Constant-Murley score [52] (n = 10) and the University of California-Los Angeles (UCLA) shoulder rating scale [53] (n = 9). Seventeen trials from the 23 Chinese studies reported the proportion of improved participants after treatment [25,26,28,30–33,37,39–44,46,47,49]. One trial reported a successful outcome defined by a 50% improvement in the disabilities of the arm, shoulder, and hand score [48]. Quality of life was reported in one trial using subdomains of the Short-Form 36 (SF-36) Health Survey [50]. Fourteen trials reported whether adverse events occurred [27,28,30–33,35,38,43,45,47–50]. All trials reported outcomes within 12 weeks post-randomization, and 11 trials also assessed outcomes after more than 12 weeks [23,26,27,29,41,42,45,48–51]. Further details on the characteristics of the included studies are presented in Table 4 and Appendix B.

The results of the assessment and the rationale for the decisions are presented in the risk of bias summary (Fig. 2) and the risk of bias tables (Appendix B). Random sequence generation in 14 studies, and allocation concealment in 23 studies were rated as having unclear risk of bias due to insufficient information. Except for the one, patient-blind sham-controlled trial which was rated as having low risk of performance bias [27], the remaining 27 trials were rated as having high risk of bias, because they were open trials. The blinding of the outcome assessment was reported in only 7 trials [23,26,27,45,48,50,51], and the rest of the trials had no descriptions. Most of the trials were rated as having low risk of attrition bias, because they reported no missing outcome data. Twenty-four studies were rated as having unclear risk of reporting bias due to the lack of a study protocol. There were 15 trials rated that did not provide radiological evidence of RCD as having unclear risk of other biases related to the misclassification of RCDs [25,26,28,30–32,37,38,42,44–48,50].

Forest plots based on the analyses are presented in Appendix C. The results of secondary outcomes of each comparison are presented in Table 5. “Summary of findings” tables are presented in Appendix F.

Manual acupuncture was superior to non-penetrating sham acupuncture in terms of shoulder function as measured using the modified Constant-Murley score at 3 months (n = 52; MD 17.30; 95% CI 7.79 to 26.81) [27].

Acupotomy treatment of RCDs showed significant reduction of pain intensity measured using the numerical rating scale compared with no treatment at 3 months (n = 52; MD −3.45; 95% CI −4.06 to −2.84) [29].

Seven studies of acupuncture versus non-pharmacological interventions were included in the analysis [25,26,36,39,44,45,50]. Non-pharmacological interventions included in this comparison were extracorporeal shock wave therapy (ESWT) [25], ultrasound therapy [45], electrotherapy [39], exercises[36,44,50] and manual therapy [26]. Acupuncture showed favorable results on pain reduction measured using VAS scores compared with ESWT, exercise, or electrotherapy combined at 3 months (5 studies; n = 280; MD −1.55; 95% CI −3.05 to −0.04; I² = 97%) [25,36,39,44,50], and on shoulder function measured using several scales compared with the exercise, ultrasound therapy, or electrotherapy combined (5 studies; n = 305; SMD 1.56; 95% CI 0.58 to 2.54; I² = 93%) [36,39,44,45,50]. Four studies reported the proportion of improved participants measured within 12 weeks [25,26,39,44]. There was no difference between the acupuncture and ESWT, manual therapy, electrotherapy, or exercise combined (4 studies; n = 219; RR 1.07; 95% CI 0.89 to 1.29; I² = 69%) [25,26,39,44].

Six studies of acupuncture versus pharmacological interventions were included in the analysis [30,41–43,47,49]. Pharmacological interventions included in this comparison were loxoprofen 180 mg per day [30], topical application of diclofenac gel [47], and local corticosteroid (i.e., prednisolone or triamcinolone) or lidocaine injection [41–43,49]. A significant reduction of pain measured using VAS scores compared with the loxoprofen or diclofenac gel combined group was observed in the acupuncture group (2 studies; n = 212; MD −1.29; 95% CI −1.49 to −1.10; I² = 0%) [30,47]. Acupuncture was superior to loxoprofen, diclofenac gel, or local corticosteroid injection combined in terms of shoulder function measured using several scales within 12 weeks (4 studies; n = 462; SMD 0.71; 95% CI 0.39 to 1.03; I² = 60%) [30,41,43,47]. Acupuncture showed favorable results compared with the loxoprofen, diclofenac gel, or local corticosteroid injection combined in terms of the proportion of improved participants within 12 weeks (6 studies; n = 644; RR 1.17; 95% CI 1.02 to 1.35; I² = 76%) [30,41–43,47,49].

Eight studies of acupuncture plus non-pharmacological interventions versus non-pharmacological interventions alone were included in the analysis [28,33–35,37,46,48,51]. Non-pharmacological interventions included in this comparison were manual therapies [33–35,51], exercise [48], ESWT [28], and a combination of the physical therapies (i.e., electrotherapy, exercise, massage [37,46]. Five studies reported pain intensity measured within 12 weeks. One study [48] reported both current pain and the worst shoulder pain experienced in the past week; the current intensity of pain was used for the analysis because the other four studies evaluated the overall pain currently experienced. There was significant reduction of pain with acupuncture treatment than with the manual therapies, exercise, or ESWT combined (5 studies; n = 431; MD −1.53; 95% CI −2.19 to −0.88; I² = 90%) [28,34,35,48,51]. All eight studies reported shoulder function within 12 weeks which was measured using several instruments. One study [34] reported both the UCLA shoulder rating scale score and the Shoulder Pain and Disability Index score for the assessment of shoulder function. The UCLA shoulder rating scale was chosen for the analysis, because it was also used in other two studies. The acupuncture group showed favorable results compared with the control group (8 studies; n = 557; SMD 1.74; 95% CI 0.91 to 2.57; I² = 94%). Use of the Shoulder Pain and Disability Index score in the study [34] did not change the direction of effect (SMD 1.36; 95% CI 0.80 to 1.92; I² = 88%) although the magnitude of effect was diminished. Modest benefits of acupuncture treatment were observed in terms of the proportion of improved participants within 12 weeks compared with ESWT, manual therapy, or physical therapies combined (4 studies; n = 163; RR 1.15; 95% CI 1.03 to 1.28; I² = 0%) [28,33,37,46].

Pharmacological interventions included in this comparison of acupuncture plus pharmacological interventions versus pharmacological interventions alone were local triamcinolone or procaine injection [31,32,40], oral analgesics and non-steroidal anti-inflammatory drugs [23], and oral Chinese herbal medicine (i.e., Xujin Jiegu liquid; 续筋接骨液) [38]. Of the five relevant studies [23,31,32,38,40], one study [23] could not contribute to the meta-analysis due to missing information on the sample size and the variance of the effects. In the study, the mean VAS score of shoulder pain measured at 3 weeks was 4.6 in the control group and 1.5 in the acupuncture group [23]. There was a modest reduction of pain in the acupuncture group compared with the local triamcinolone or procaine injection or Xujin Jiegu liquid combined group measured within 12 weeks (4 studies; n = 369; MD −0.98; 95% CI −1.94 to −0.01; I² = 89%) [31,32,38,40].

Four studies [23,31,32,38] reported shoulder function measurements within 12 weeks. One study [23] which was excluded from the meta-analysis, reported the mean Instrumental Activities of Daily Living score as 6.8 in the control group and as 7.4 in the acupuncture group (a higher score indicates a better outcome). There was a significant benefit of acupuncture compared with the local triamcinolone injection or Xujin Jiegu liquid combined in terms of shoulder function measured within 12 weeks (3 studies; n = 219; SMD 0.84; 95% CI 0.19 to 1.49; I² = 80%) [31,32,38]. There was no between-group difference in the proportion of improved participants within 12 weeks compared with local triamcinolone or procaine injection (3 studies; n = 309; RR 1.05; 95% CI 0.94 to 1.16; I² = 0%) [31,32,40].

Three subgroup analyses were available, including acupotomy versus non-acupotomy, RCD without tear versus RCD with tear, and manual acupuncture versus EA. There was no evidence against the null hypothesis that the effects of the subgroups were not different. Three sensitivity analyses were conducted, one of which showed a marked but not statistically significant increase in the effect size when the analysis was confined to studies with both clinical and radiological assessments of RCD in the comparison of acupuncture with non-pharmacological interventions (Appendix E.1). The same sensitivity analysis for the comparison of acupuncture plus non-pharmacological interventions with non-pharmacological interventions alone showed an increase in the effect size with statistical significance (Appendix E.3). In a fixed-effect model for short-term pain in a comparison of acupuncture plus non-pharmacological interventions with non-pharmacological interventions alone, the magnitude and precision of the effect estimates were increased compared with random-effects model (Appendix E.2). Assessment of small study effects were unavailable due to the insufficient number of studies.

Adverse events of manual acupuncture were reported in one study [27] including headaches (n = 3), dizziness (n = 2), loss of strength in the legs (n = 1), and an inflammatory reaction (n = 1) in the acupuncture group. There were 4 studies with a manual acupuncture group which reported that no adverse events occurred [33,45,49,50]. One study reported itchiness and redness of the skin after fire-needling (n = 6), and dizziness during acupotomy (n = 2) [30]. Four trials which had an acupotomy group reported that no adverse events occurred [31,32,38,43]. One trial with a dry needling group [48], and two trials with EA groups [28,35] reported that no adverse events occurred. One study reported that they had planned to assess adverse events, but did not report the adverse events in the results section [47]. The rest of the studies did not provide safety data.

The certainty of the evidence for the primary outcome (i.e., pain, shoulder function, and the proportion of improved participants) was determined. The certainty of evidence of acupuncture versus sham acupuncture for short-term shoulder function was downgraded once from high to moderate for imprecision due to the small number of participants. The certainty of evidence of other comparisons downgraded three times from high to very low due to serious concerns on the risk of bias, indirectness, imprecision and/or inconsistency (Appendix F).

There were 28 RCTs involving 2216 participants included in this review, that evaluated the effectiveness and safety of acupuncture treatments for RCDs. Most studies compared acupuncture either alone or in combination with active comparators, or with an active comparator alone, implying the current evidence has been focused on evaluating the additive effects or comparative effectiveness of acupuncture compared with other interventions. The certainty of evidence for the primary outcomes in the studies in this review were mostly very low (except for one outcome) due to the high/unclear risk of bias, substantial heterogeneity across studies, imprecision of the estimates, and potential indirectness from different study contexts, which indicated a careful interpretation of the effect estimates in the review. All trials reported short-term outcomes and around a half provided intermediate-term outcomes, which indicated the absence of evidence of acupuncture for long-term results in patients with RCDs. The reported adverse events were minor and not frequent, although incomplete reporting of safety data was prevalent in included studies of this review. Overall, very-low certainty of evidence was observed on the short-term benefits of acupuncture either alone or combined with active comparisons as compared with the same active comparison alone.

The participants in the included trials were patients from outpatient clinics. Most of the trials were conducted and published in China, and the findings of this review may not be applicable in different settings. Clinical characteristics such as availability of radiological confirmation of RCDs, severity and duration of symptoms before the study enrolment, types of acupuncture, and control interventions and outcome measures were heterogeneous. Therefore, the findings of this review provide only rough information on the evidence of acupuncture for RCDs. Non-penetrating types of acupuncture such as laser acupuncture or acupressure were not included in this review, and evaluating evidence requires future studies. As the review does not provide evidence for effects of combination interventions such as acupuncture plus cupping, moxibustion, or herbal medicine, the evidence may have limited application in Korea or some East Asian countries where combinations of treatments are commonly practiced in clinic [54].

Most studies have focused on outcomes in pain, shoulder function, or other symptoms. Therefore, it remains unclear whether acupuncture is beneficial for improving quality of life or other patient-reported outcomes such as satisfaction, which were rarely addressed in the studies of this review. Incompleteness of reporting on safety did not allow a comprehensive assessment of the harmful effects of acupuncture.

A substantial number of the included RCTs did not report the method of randomization or details of allocation concealment. Despite best efforts to obtain further information on details of random allocation process, this was not possible (absence of the e-mail address or response). Therefore, risk caused by including non-randomized trials cannot be excluded in this review which may have biased the review results towards the overestimation of the effects. The risk of attrition bias was rated as low when the number of participants enrolled and those analyzed were same, although intention-to-treat analyses could not be guaranteed. This may have biased results in this review, although the direction of bias on the effect estimates seems unclear. The proportion of improved participants was analyzed as a primary outcome because it was frequently reported in the Chinese trials included in this review, and was deemed as an important outcome, although it was not pre-defined at the protocol stage. This ad-hoc analysis may have potentially introduced the risk of chance findings. Therefore, the results should be interpreted with caution. Pooled effect estimates were sought regardless of the heterogeneity, against the study protocol and the standard guidance [14]. This was because the pooled effect estimates may have provided exploratory information for further research, although the considerable heterogeneity, and risk of bias should be considered when interpreting the pooled effects. The existence of small study effects could not be assessed due to insufficient number of the studies in the same meta-analysis. Therefore, it remains unclear whether publication bias affected our findings.

One systematic review of acupuncture for RCD was identified [11], along with four systematic reviews related to acupuncture for shoulder pain [7–10]. The search date and range of databases searched were different between studies. Few or none of the Chinese studies were analyzed in the previous reviews. This review reflected studies with a wide range of acupuncture-related interventions. The eligibility criteria was similar in one review [11], but it excluded studies that administered analgesics as co-interventions, whereas this current review did not set any limitations on the co-interventions other than acupoint-related interventions. Similarities between the findings of this current review and previous reviews include high/unclear risk of bias in the included studies, paucity of longer-term outcome assessment and of patient-relevant outcomes including quality of life and patients’ satisfaction, and incomplete reporting on safety of acupuncture, which may indicate little progress in the quality of methodology in trials on this topic over the years.

Based on the overall very low-certainty of the evidence, observed short-term benefits of acupuncture should be interpreted with caution. Incompleteness of reporting on safety of acupuncture should be considered, especially for interventions with potentially higher risks such as acupotomy or warm/fire needle acupuncture. Whether acupuncture can yield a long-term improvement in the symptoms or shoulder function of patients with RCDs remains unclear. Any particular types of acupuncture or regimens of interventions cannot be endorsed from this current review, although details of acupuncture treatments might be referred to as a source of information for clinical practice.

High-quality RCTs with adequate power to give meaningful results, and in various settings should be conducted in the future. Transparent reporting of the methods and results are required for valid interpretation of study findings. Bias associated with the randomization process, outcome assessor blinding, incomplete outcome reporting, selective outcome reporting, and misclassification due to absence of radiological assessment of RCDs should be addressed in designing future trials. Evidence on the role of acupuncture for long-term recovery, quality of life, and patients’ satisfaction is needed. Use of validated and standardized outcomes across studies may enable meta-analysis of the data, which would enable the examination of small study effects. Information on safety of interventions should be explicitly reported.