Reported Data on Granisetron and Postoperative Nausea and Vomiting by Fujii et al. Are Incredibly Nice!

We read with interest the recent article of Fujii et al. (1) about granisetron in the prevention of postoperative nausea and vomiting (PONV). With increasing amazement, we noticed that the results reported by Fujii et al. are incredibly nice and we became skeptical when we realized that side effects were almost always identical in all groups. During 1994–1999, 47 articles have been published by Fujii et al. (http://www.nlm.nih.gov/> search “Fujii-Y and granisetron”) (1–47). In 21 articles, the most frequently reported side effect, headache, is given for the overall 24-h period of observation (Table 1). In 13 articles, the frequency of headache was reported to be identical in all groups (5,7–11,15,16,19,20,29,40,43), while this side effect differed, at most, by one patient per group in the remaining 8 papers (4,6,12–14,27,28,32). Surely, assuming that the study drug has no impact on a side effect, one would expect similar results between the groups. However, identical results are still relatively rare as binominal distribution causes a certain variability in reality. Thus, we tested the null hypothesis that the reported identical incidences could have occurred by chance with the alternative hypothesis that an underlying reason lead to such identical results.Table 1: Incidence of Headache per Group in the Investigated Publications and Calculated Probabilities of Obtaining Identical GroupsFirst, the incidence of headache (p) was calculated by the total number of patients with headache (n = 186) divided by the total number of patients (n = 2369) in the 21 studies, resulting in an incidence (with lower to upper 95% confidence intervals) of p = 0.0785 (0.0681 to 0.09). Second, the probability (P) that x out of n patients will experience headache will generally be calculated according to formula for the binominal distribution P = (n!/(x!(n − x)!)) × px × (1 − p)n−x (48). Third, for each study, the probability that all (z) groups have identical results with x patients suffering from headache is calculated by P = Pz. These calculations were done for x between 0 and 10 as the probabilities for x > 10 were too small to have any impact on the calculation (Table 1). Fourth, the joint probability that a study results in any identical number of patients is calculated by the sum of the single probabilities for x = 0, 1, 2, … 10, also given in Table 1. Fifth, the joint probabilities of all 18 separate studies with more than two groups–all three studies with two groups had identical results anyway–was in the range of P = 0.0002 to 0.0548. To make it safe and simple, the hypothesis that 10 of 18 studies will result in identical numbers was tested by applying again the above mentioned formula for the binominal distribution with the highest probability of 0.0548. This resulted in a final probability (Pfinal) that 10 of those 18 studies will have identical results by chance of Pfinal = 6.78 × 10−9! Thus, we have to reject the null hypothesis that the frequency of identical results simply occurred because of the assumption that the incidence of headache is not affected by the intervention, and we have to conclude that there must be an underlying influence causing such incredibly nice data reported by Fujii et al. We thank Horst Fassl, Emeritus Professor of Medical Statistics (Nieder-Olm, Germany) and Herbert Vogt, Professor of Mathematics (Dept. of Applied Mathematics, University of Wuerzburg, Germany) for their statistical advice. Peter Kranke Christian C. Apfel MD Norbert Roewer MD