Objective.—To examine whether the distributions of active and placebo response rates change after transformation into therapeutic gains, numbers needed to treat, and therapeutic ratios for 51 published clinical trials of 5-HT1B/1D agonists in the acute treatment of migraine.
Background.—Acute migraine therapies have been compared using meta-analysis. Before pooling the results of noncontemporaneous clinical trials for meta-analysis, reconciliation of active response rates with concurrent placebo controls is done because of fluctuations in placebo response rates. The 3 common methods of reconciliation are to find for each clinical trial the therapeutic gain, the number needed to treat, or the therapeutic ratio.
Methods.—Raw active and placebo response rates were tabulated and displayed as histograms. The distributions of the therapeutic gains, numbers needed to treat, and therapeutic ratios were similarly plotted. These distributions were then compared with normal distributions using chi-squared goodness-of-fit methodology.
Results.—The distribution of active response rates was consistent with a normal distribution (passing a goodness-of-fit test). Placebo response rates were not normally distributed. The distribution of therapeutic gains failed a test of normality (P= .018), as did the numbers needed to treat (P < .001); the skews of these distributions were toward opposite ends of the scale for relative efficacy. The therapeutic ratios also failed a test for normality, but passed a test for log-normal distribution.
Conclusion.—Choice of transformation is a potential source of bias in meta-analysis of acute migraine therapies.