For shoulder pain, the large pooled analyses and the systematic reviews of acupuncture point in different directions. The individual patient data meta-analyses of the Acupuncture Trialists’ Collaboration found an effect against sham for shoulder pain about three times the size of the one for back and neck pain, osteoarthritis or headache. The Cochrane review of 2005, by Green, Buchbinder and Hetrick, concluded that its nine trials could neither support nor refute acupuncture, and the systematic reviews since have been small, split by diagnosis and rated low or very low in certainty.
Shoulder pain covers several diagnoses, and the Cochrane review reported rotator cuff disease, adhesive capsulitis and mixed diagnoses separately. Adhesive capsulitis, or frozen shoulder, is a painful loss of movement in the joint, and its symptoms recover after one to four years, according to a 2020 review by Ben-Arie and colleagues. Shoulder pain after a stroke has a separate literature, covered in Stroke recovery.
The Cochrane review
Green and colleagues searched to December 2003 for trials of acupuncture in adults with shoulder pain lasting more than three weeks, and found nine of varying quality. All described the acupuncture poorly and used different placebos. Two trials in rotator cuff disease could be pooled: they found no significant short-term difference from placebo, which the authors said could reflect samples too small to detect one. On the Constant-Murley score, a 0 to 100 measure of shoulder function, acupuncture beat placebo by 17.3 points at four weeks; by four months the difference was 3.5 points, still statistically significant but unlikely to matter to patients. A small pilot study found traditional and ear acupuncture each added some benefit to mobilisation, and the one trial that assessed adverse events found no difference from placebo. The review has not been updated since; its last listed change, in 2008, converted it to a new format.
The individual patient data meta-analyses
The Acupuncture Trialists’ Collaboration obtained raw data on each participant from trials with adequate concealment of allocation, meaning the people enrolling patients could not know or influence the next assignment. Its 2012 analysis pooled three shoulder pain trials against sham for an effect of 0.62 standard deviations. With the outlying trials by Vas and colleagues set aside, too few shoulder trials remained to pool, and the other conditions came to 0.15 to 0.23. In the analysis’s check for publication bias, smaller trials showed larger effects against sham; the authors attributed this partly to the shoulder trials, which were small with large effects, and the asymmetry was no longer significant once the shoulder trials and the Vas trials were excluded. The 2018 update had four shoulder trials against sham, with an effect of 0.57, and 0.58 from three once the trials by Vas and colleagues were excluded; it found no shoulder trial that compared acupuncture with no acupuncture. Without the Vas trials, the effect for musculoskeletal pain, osteoarthritis and headache was 0.16 to 0.19, and the authors called the shoulder effect clearly much greater. Chronic pain covers these analyses in full.
The GRASP trial
The largest single trial, GRASP, published in 2010 by Molsberger and colleagues, randomised 424 outpatients in Germany with shoulder pain of at least six weeks and a pain score of at least 50 on a 0 to 100 scale. They received Chinese acupuncture, sham acupuncture or conventional orthopaedic treatment, 15 sessions over six weeks from 31 office-based orthopaedists trained in acupuncture, and the patients did not know which needling they had. A responder was anyone whose pain fell by at least half. Three months after treatment ended, 65% of the acupuncture group, 24% of the sham group and 37% of the conventional treatment group were responders; at the end of treatment the figures were 68%, 40% and 28%. Acupuncture differed significantly from both other groups at both times.
Reviews by diagnosis
| Review | Condition | Trials (participants) | Controls | Finding |
|---|---|---|---|---|
| Ben-Arie et al. 2020 | Frozen shoulder | 13 studies | Medication, physiotherapy, sham acupuncture, other acupuncture | Less pain, better Constant-Murley function and forward flexion; no difference in abduction or external rotation. Evidence very low |
| An et al. 2024 | Shoulder impingement | 5 trials | One sham; four compared exercise or physiotherapy with and without needling | Pain SMD −0.50, function SMD −0.57, from small, varied trials |
| Trinh et al. 2025 | All shoulder pain except after stroke | 10 trials (748) | Acupuncture alone against non-active controls | Mixed results across trials; limited, low-quality evidence |
SMD is the standardised mean difference, the measure reviews use to pool trials that scored the same outcome on different scales; a negative value favours acupuncture. In An’s review, three of the five trials needled myofascial trigger points, the technique more often called dry needling. Ben-Arie’s review found LI15 and TB14, which the review names Jian Yu and Jian Liao, the points its trials used most, and two of its trials included ST38 (Tiaokou) on the lower leg.
What the guidelines say
A 2025 review of 17 guidelines on acupuncture for chronic musculoskeletal pain, published between 2014 and 2024, found shoulder pain the condition they addressed most: 14 of their 35 recommendations, of which 13 supported acupuncture. The only two strong recommendations in the whole set were for shoulder pain. Among the guidelines devoted to shoulder pain were a Korean medicine guideline on manual acupuncture from 2017 and a traditional Chinese medicine guideline on frozen shoulder from 2023. NICE has no guideline on shoulder pain. Its Clinical Knowledge Summary on frozen shoulder, last revised in November 2022, lists acupuncture among the treatments a physiotherapist may give during a course of physiotherapy. Clinical guidelines sets these beside other bodies’ positions.
What the reviews could not settle
The large effect the individual patient data analyses found rests on four small sham-controlled trials, and the Collaboration’s own bias checks flagged shoulder pain as the indication where small trials had large effects. GRASP, the largest trial, found a wide gap between acupuncture and sham, but the Cochrane review, which predates it, found too few comparable trials to pool, and no Cochrane update has assessed GRASP alongside the earlier evidence. The later reviews each include few trials, and several of those trials added acupuncture to physiotherapy or exercise and compared the result with physiotherapy or exercise alone, a design that cannot separate needling from the extra attention. Frozen shoulder improves on its own over one to four years, so only a comparison group can show what a treatment added.
AcuiQ’s shoulder pain page lists the protocols individual studies prescribed, each with its citation, and the frozen shoulder, adhesive capsulitis and chronic shoulder pain pages do the same for the terms those studies used.