The ‘delay’ hypothesis has had a long life and has greatly influenced the treatment of depression and research aimed at the development of new ‘more rapidly acting’ drugs. Evidence has gradually accumulated that the hypothesis is inaccurate and its entrance into the lore of clinical practice and textbooks is unwarranted. The hypothesis was derived from studies that did not test it directly and are now viewed as both conceptually and methodologically inadequate. In his editorial, Mitchell (Reference Mitchell2006) makes salient points and, after reviewing some of the current literature, comes to the correct conclusions. However, our earlier results that refuted the hypothesis (Katz et al, Reference Katz, Koslow and Maas1987, Reference Katz, Koslow and Maas1991) and a more recently published study (Reference Katz, Tekell and BowdenKatz et al, 2004), designed to definitively test the ‘delay’ notion, were not referenced. Their omission leaves important gaps.

Mitchell highlights the inadequacy of the ‘outcome’ measures used in early studies and the importance of distinguishing ‘improvement’ from ‘full response’. The central criticism of the early work (e.g. by Reference Gelenberg and ChesenGelenberg & Chesen, 2000) is that most information was derived from clinical trials, not studies designed to accurately estimate onset and time course of changes. Such studies require a placebo control, sensitive measures of behavioural change (beyond the Hamilton Rating Scale for Depression) and frequent early assessment. The meta-analyses of Stassen et al (Reference Stassen, Angst and Delini-Stula1997) came close to achieving these aims. However, no study had met all the necessary requirements until that conducted by our group (Reference Katz, Tekell and BowdenKatz et al, 2004). This measured the major behavioural components of the disorder intensively at 3-day intervals, operationally distinguished improvement and full response and compared pharmacologically different antidepressants (a selective serotonin reuptake inhibitor and a selective noradrenaline reuptake inhibitor) with placebo. In targeting the issue directly, it used appropriate statistical methods to investigate whether algorithms could be developed to predict treatment response from early behavioural changes, problems alluded to by Mitchell. We understand that an editorial cannot provide an exhaustive review but believe the information above will contribute to the important issues addressed.