The usefulness of antidepressants in patients with mild or moderate depression has been questioned and therefore Thase et al ^{Reference Thase, Larsen and Kennedy1} re-analysed randomised controlled trial (RCT) data for escitalopram. They calculated that a subgroup of patients (14%) with mild or moderate depression did respond to treatment and not to placebo, but that the subgroup of patients with severe depression who responded to treatment was larger (23%).

The findings from Thase et al ^{Reference Thase, Larsen and Kennedy1} illustrate what is known from the philosophy of science about RCTs: if T causes O in a population, this only implies that T causes O in at least some members of that population. ^{Reference Cartwright2} Reporting average results can be misleading if the population is not causally homogeneous.

Ideally, one should try to identify the specific subgroup of responders. The authors could consider analysing their data further by looking at demographic factors such as age, because younger people are less likely to respond, ^{Reference Tsapakis, Soldani, Tondo and Baldessarini3} and the presence of particular symptoms such as anxiety, which might adversely influence outcome as well, ^{Reference Coryell, Fiedorowicz, Solomon, Leon, Rice and Keller4} if these data are available. However, it is unlikely that one can identify the subgroups of responders with 100% accuracy with these additional data.

More generally, Thase et al's results show that by reporting average results, important findings from RCTs might be missed. Reporting binary outcomes such as number or percentage of patients improved or in remission should be encouraged in psychiatric research, even if the primary outcome variable is considered to be a continuous interval variable. This applies to all RCTs with non-homogeneous populations. Unfortunately, CONSORT requirements do not make this compulsory at the moment. ^{Reference Schulz, Altman and Moher5}