We compliment Dr Flores and colleagues for the detailed descriptions of methodology and statistical analysis in their report on mifepristone for psychotic major depression (PMD) (Flores et al, 2006). However, similar to our concerns about earlier trials of mifepristone for this disorder (Rubin and Carroll, 2004), we have reservations about the authors’ interpretations of their data. First, we question the analysis of mifepristone’s efficacy. Only the number of individuals meeting a 50% reduction in the Brief Psychiatric Rating Scale’s Positive Symptom Subscale (BPRS PSS) was reported as significantly different between drug and placebo (reported w 1⁄4 3.968, 1 df, p1⁄4 0.046). The authors used an uncorrected w test to obtain this result, even though the frequency is less than five in one of the cells of the 2 2 distribution. When the more appropriate w test with Yates’ correction is applied, the result is not significant (w1⁄4 2.539, 1 df, p1⁄4 0.111). An alternative statistical test is Fisher’s exact probability, which, with a two-tailed test, also is nonsignificant (p1⁄4 0.109). Thus, there is not even a nominally significant difference in the BPRS PSS between drug and placebo. In addition, by ANOVA, the authors found no main effect of medication on BPRS PSS scores (Flores et al, 2006). This negative finding casts further doubt on the validity of the authors’ claim of efficacy using the categorical criterion of 50% reduction of BPRS PSS scores. The conclusion that ‘ymifepristone was found to be significantly more effective than placebo in reducing psychotic symptoms’ therefore is not supported. The second problem is that three outcome measures (the BPRS PSS, the full BPRS, and the Hamilton Depression Scale) were each analyzed by two different statistical methods, w for numbers of subjects achieving a certain percentage reduction on each scale, and repeated measures ANOVA on the raw scale scores at Day 0 and Day 8. None of the other five statistical tests yielded a significant result. The aforementioned reported significance level of 0.046, even if it were valid, would represent a true significance for only one test and would become clearly nonsignificant when corrected for the actual number of statistical tests performed. Thus, by conventional statistical standards, the authors failed to demonstrate efficacy of mifepristone in reducing psychotic symptoms. Third, there is reason to question the generalizability of these less than robust clinical trial results to clinical practice. The prevailing construct of PMD describes a severe illness (American Psychiatric Association, 2000) with significant suicide risk and functional incapacity that requires close clinical monitoring. Indeed, the company that aims to bring mifepristone to market for treatment of PMD (Corcept Therapeutics Inc.) has stated its intention to target in-patient psychiatric services that offer electroconvulsive therapy (ECT) as the company’s primary market (Corcept Therapeutics Inc., 2004). In the present study, however, an unstated number of subjects was recruited by advertisement rather than through clinical pathways; subjects waited up to 4 weeks between screening and study participation; and, when the study treatment was finally scheduled, 16% of the enrolled sample (one in six cases) no longer met diagnostic criteria for major depressive disorder with psychotic features. These characteristics suggest that many patients had relatively mild or transient symptoms, and that they were generally not in clinical crisis. As well, the authors failed to demonstrate baseline overactivity or dysregulation of the hypothalamo-pituitary-adrenal (HPA) axis, a salient feature of PMD that provides the rationale for trials of a glucocorticoid receptor blocker like mifepristone. Fourth, in their analyses of efficacy, the authors erroneously claimed that they performed an ‘(e)xamination of the clinical utility of mifepristone in the treatment of psychotic symptomsy’ They did not examine clinical utility. Rather, they examined a surrogate psychometric measure of psychotic symptom severity (BPRS PSS). This instrument is an ordinal subscale that lacks proven ratio properties, so the clinical utility of a 50% reduction of symptoms is uncertain. It is clear that the patients still had a notable mean psychotic symptom load at the end of 8 days Online publication: 21 June 2006 at http://www.acnp.org/citations/ Npp062106060143/default.pdf Received 1 March 2006; accepted 9 June 2006 *Correspondence: Dr BJ Carroll, Pacific Behavioral Research Foundation, 100 Del Mesa Carmel, PO Box 223040, Carmel, CA 93922-3040, USA, Tel: + 1 831 626 1467, Fax: + 1 831 626 6963, E-mail: bcarroll@redshift.com Neuropsychopharmacology (2006) 31, 2793–2794 & 2006 Nature Publishing Group All rights reserved 0893-133X/06 $30.00