Thirty-one eligible RCTs were identified; 11 trials were sham controlled, 10 had a no acupuncture control, and 11 had three arms. The IPD meta-analyses included 29 trials, with 17,922 participants; 14,597 compared acupuncture with no acupuncture and 5,230 compared acupuncture with sham control.
Compared with no acupuncture, acupuncture was significantly superior, with pooled standardised mean differences (random-effects model) of 0.57 standard deviations (95% CI 0.29 to 0.85; six RCTs) for osteoarthritis, 0.38 standard deviations (95% CI 0.22 to 0.55; five RCTs) for chronic headache, and 0.51 standard deviations (95% CI 0.36 to 0.67; seven RCTs) for back or neck pain.
Compared with sham treatment, acupuncture was significantly superior. The pooled standardised mean differences were 0.37 (95% CI 0.03 to 0.72; five RCTs) for osteoarthritis, 0.15 (95% CI 0.05 to 0.24; four RCTs) for chronic headache, 0.52 (95% CI 0.14 to 0.90; eight RCTs) for back or neck pain, and 0.62 (95% CI 0.46 to 0.77; three RCTs) for shoulder pain.
The authors stated that with baseline pain scores of 60 out of 100 and a standard deviation of 25, these effect sizes corresponded to response rates (pain reduction of 50% or more) of approximately 30% for no acupuncture, 42.5% for sham acupuncture and 50% for true acupuncture.
Statistical heterogeneity was significant for five of the seven comparisons in the primary analysis. Removal of three RCTs that strongly favoured acupuncture, reduce the heterogeneity for the acupuncture versus sham comparison. The authors stated that heterogeneity for the acupuncture versus no acupuncture comparison was largely explained by differences between the control groups. Pre-specified sensitivity analyses showed that the inclusion of summary data from RCTs for which IPD were not available or which were published too late for inclusion had little impact on the results. Other sensitivity analyses were reported. No evidence of significant publication bias was found.