Statistical reanalysis of a randomized trial of acupuncture for pain reveals positive effects as well as adverse treatment interactions on pain, attrition, and mortality

Objective: The aim of this study was to determine, using analysis of covariance, whether a statistical reanalysis of a previously published study on neuropathic pain would reveal undetected significant effects of acupuncture and amitriptyline on pain, attrition, and mortality in HIV-infected patients. Background: Shlay et al published an article in the Journal of the American Medical Association (1988) reporting that neither acupuncture nor amitriptyline had effects on pain in HIV-infected patients. However, they failed to perform a factorial analysis of variance (ANOVA) or covariance (ANCOVA) reflective of their core research design. Instead, research design problems necessitated the use of a relatively insensitive statistic. Methods: The originally planned study employed a completely crossed 2 × 2 design involving acupuncture and amitriptyline and their controls. Reanalyses performed on the raw data involved ANCOVA and Pearson chi-square tests. Subjects and Setting: The factorial option consisted of 125 HIV-infected men with peripheral neuropathic pain, being treated at health clinics in 10 different cities. Outcome Measures: Outcome measures were pain intensity, global pain relief, attrition, and mortality. Results: In contrast to the originally reported findings, the interactions of amitriptyline and acupuncture over time on pain intensity and pain relief were statistically significant. There were also significant effects for acupuncture and amitriptyline on attrition and mortality, particularly when baseline health was poor. Acupuncture by itself was associated with greater pain relief, whereas the combination of acupuncture and amitriptyline was associated with a reduced level of pain relief. Acupuncture without amitriptyline was associated with substantially reduced attrition and, importantly, decreased mortality. Effects involving acupuncture tended to be magnified in patients in poor health. The combination of acupuncture and amitriptyline resulted in an adverse treatment interaction on mortality, especially in patients in poor health (53% death rate for the combination of acupuncture and amitriptyline vs 11% death rate for acupuncture only). Conclusions: Trials of acupuncture and other treatments should use efficient statistical techniques to assure detection of significant effects. Interactions involving various combinations of acupuncture and amitriptyline, which were undetectable in the original analytical approach, revealed previously undetected beneficial, as well as adverse, effects.