To the Editor: We enjoyed reading the article by Dr. Meltzer and colleagues. It is interesting to see novel approaches to treating psychosis. However, there are a few points that we feel are worth mentioning.

The results of Dr. Meltzer and colleagues show that the serotonin 5-HT_2A/2C antagonist SR46349B is helpful in treating the negative symptoms of schizophrenia. This indeed is true and has been shown with another 5-HT_2A/2C antagonist. In 2001, Berk and colleagues (1) showed that mirtazapine’s 5-HT_2A/2C antagonism (2) improves negative symptoms in schizophrenic patients already taking haloperidol. It would have been supportive of the authors’ findings if this had been discussed.

Dr. Meltzer and colleagues used the Cochran-Mantel-Haenszel test for improved efficacy on the Barnes Rating Scale for Drug-Induced Akathisia; however, they used Fisher’s exact test to compare the incidence of adverse events of active treatments to placebo. Both of these tests are used to compare dichotomous nominal variables but not with the same result. The Cochran-Mantel-Haenszel test typically creates a p value that is different from the p value for Fisher’s exact test (3, 4). The authors attempted to use this difference in statistics to minimize the adverse events listed.

It would have been worthwhile for the authors to discuss the rate of "extrapyramidal disorders" as an adverse event for the NK₃ antagonist (SR142801). With the Cochran-Mantel-Haenszel test, the p value was <0.05, but with Fisher’s exact test, it only approached significance (p=0.09).

Sometimes the difference between the Cochran-Mantel-Haenszel test and Fisher’s exact test does not matter. For example, the authors did not mention that the NK₃ antagonist and the 5-HT_2A/2C antagonist not only had more completers but that it was statistically significant over placebo; the p values were <0.05 for both tests. Another example that produced similar p values (<0.05) was that haloperidol was the only active treatment out of five to have fewer patients drop out because of lack of efficacy.

It seems the difference between the two tests matters only in cases with values close to significance. We encourage the authors to consider using only one test when comparing two different variables of the same type—one test for dichotomous nominal variables, one for ordinal, and one for interval. All statistics mentioned were derived from SPSS version 11.5 (SPSS, Chicago) and comprehensive meta-analysis.