Sir

The report of Rapaport et al (2006a) raises several concerns. Contrary to policy, the data were previously published in extenso (Nemeroff, 2005), but this previous report was not acknowledged. Next, a corrigendum (Rapaport et al, 2006b) belatedly disclosed the authors' financial ties to the sponsor (Janssen). No author may claim ignorance of the journal's policies on prior publication and disclosure of competing interests. Now comes a second corrigendum (Rapaport et al, 2007) that disowns a claim of efficacy. This retraction highlights but one of several unexplained differences in statistical reporting and conclusions between this report and Janssen's 2005 report on ClinicalTrials.gov.

In a subgroup analysis, Rapaport et al (2006a) claimed significant efficacy for risperidone on two measures of relapse prevention. The second corrigendum disowned one of these claims (proportions relapsing) as an ‘error.’ Janssen (2005) never made this claim. The result never approached significance (p=0.4) (Rapaport et al, 2007), yet the authors stated p=0.05 in the text and highlighted p⩽0.05 in Abstract.

The second corrigendum further implied that the efficacy result based on time to relapse remained significant. That claim also differs from Janssen's 2005 report. Janssen reported p=0.0512, calling the result of ‘borderline statistical significance.’ The present authors reported this analysis as p=0.05 and characterized it as ‘statistically and potentially clinically meaningful’ (Rapaport et al, 2006a). Elsewhere, the authors characterized this result as ‘significant’ (Nemeroff, 2005) and ‘significantly delayed (relapse)’ (Nemeroff et al, 2004). These characterizations differ from Janssen's candid statement.

An additional discrepancy involves the toxicity data. Janssen (2005) reported that mean weight gain in the maintenance treatment phase was significantly greater with risperidone than with placebo. Rapaport et al (2006a) reported the same data but without providing the statistical analysis. This omission was not explained. Likewise, they presented no statistical analysis of the proportions who gained 7% or more body weight during long-term treatment. These proportions appear to be 8.3% or 10 of 122 patients on risperidone, 2.6% or 3 of 119 patients on placebo. For this distribution, the Pearson's χ2 result is borderline significant (p=0.0512), as is the Fisher's exact probability test result (p=0.046, one-tailed).

Thus, the risk–benefit profile of long-term risperidone use in refractory depression may not be as favorable as depicted by Rapaport et al (2006a)—higher risk, lower benefit. The only remaining efficacy claim—temporary improvement with short-term risperidone augmentation—rests on uncontrolled data, the weakest level of evidence. The claim that risperidone significantly prevented relapse in a subgroup should be retracted.