Amsterdam and colleagues ^{Reference Amsterdam, Luo and Shults1} frankly acknowledge the multiple methodological deficiencies of their paper. However, this does not prevent them from drawing unfounded conclusions. In his editorial, Thase ^{Reference Thase2} makes the same error in accepting conclusions that are clearly not warranted by the data presented. What is most striking about Amsterdam et al's paper is that the results of lithium, fluoxetine and placebo are equally poor. The study is invalidated by the small sample size, the use of an ‘enriched’ sample and an inadequate analysis of the emergence of hypomania.

In the fluoxetine group, only 11 out of 28 participants completed the study (36%). Only 4 patients with rapid-cycling bipolar disorder completed this arm (33%). If we calculate this on the basis of the 42 patients in the rapid-cycling group who entered the first phase of the study, this amounts to a 10% completion rate! In the lithium arm, only 5 out of 26 participants completed the study (20%).

It is not possible to justify the conclusion that antidepressant monotherapy has a place in the treatment of rapid-cycling bipolar disorder based on these small numbers. In fact, it is potentially dangerous to do so. There is evidence that antidepressants, by accelerating the course of bipolar disorder and precipitating mixed states, can lead to protracted morbidity and increased risk of suicide. ^{Reference Eppel3}