Statistical reanalysis of four recent randomized trials of acupuncture for pain using analysis of covariance

OBJECTIVES: Acupuncture has been promoted for the treatment of chronic pain. Though many randomized trials have been conducted, these have been criticized for deficiencies of methodology, acupuncture technique, and sample size. Somewhat less emphasis has been placed on methods of statistical analysis. This paper describes 4 recent randomized trials of acupuncture for musculoskeletal or headache pain. Each trial used statistical methods that did not adjust for baseline pain scores and were thus of suboptimal power. The objective of this study is to reanalyze the trials using analysis of covariance (ANCOVA). METHODS: Raw data for the 4 trials were obtained from the original authors. Data were reanalyzed by ANCOVA. RESULTS: For 2 trials--acupuncture versus placebo for chronic headache and acupuncture versus transcutaneous electric nerve stimulation for back pain--reanalysis did not change the conclusion of no difference between groups, but showed that clinically significant differences between groups could not ruled out. Reanalysis of a trial of acupuncture versus placebo for shoulder pain slightly strengthened the evidence of acupuncture effectiveness. Reanalysis of the fourth trial, which compared acupuncture to placebo acupuncture and massage for neck pain, reversed the results of the original paper: reanalysis found acupuncture to be effective and that its effectiveness could not be ascribed to a placebo effect. DISCUSSION: Future trials of acupuncture and other modalities for pain should use efficient statistical methods. ANCOVA is more efficient than unadjusted analysis where used appropriately.