Drs Citrome and Ketter appear to appreciate the concern we raised about the limitations of applying the NNT from placebo-controlled studies to the clinical situation (where there is no placebo control condition). However, in their letter they maintain that, ‘Indirect comparisons of effect sizes among different medication choices can be quite helpful in ranking interventions for both efficacy and common tolerability challenges, provided that the studies used for these calculations are similar enough’. We do not disagree; in fact, we quoted Garcia in our paper: ‘to directly compare NNTs one needs to ensure that […] the control or comparisons groups to which the treated group was compared were equivalent’. ^{Reference Garcia1}

Our point in the paper was that insufficient attention is typically paid to the question of whether control conditions are ‘similar enough’, and we believe this point still holds. Although it is not clear from their letter to what type of situation Drs Citrome and Ketter refer, one is likely on firmest ground when comparing NNTs and NNHs for antidepressant medications calculated from placebo-controlled trials of similar methodology and quality. However, even in this optimal case, it has been established that placebo response can vary significantly from trial to trial, and thus the control conditions for two studies may in fact be less similar than one might suppose. ^{Reference Walsh, Seidman, Sysko and Gould2}

Perhaps it would be less problematic to compare the NNTs and NNHs calculated from a comparator trial of two or more antidepressants, because of course in this case there is no issue about the similarity of the studies. The problem is that, to our knowledge, there has not been a consistent finding that one antidepressant has therapeutic superiority or greater tolerability compared with another. One must be careful not to use the NNT and NNH from a single study when that finding has not been replicated, especially since comparator studies are primarily industry-sponsored.

Beyond the specific case of comparing two antidepressant medications, the points made by Citrome and Keller are not relevant to the fundamental thesis of our paper that NNTs calculated from placebo-controlled trials do not inform the clinician's choice whether to prescribe or not prescribe. Additionally, our further point still stands that NNHs and NNTs cannot be applied without significant confounding to decisions of whether to prescribe medications or psychotherapy, since the control conditions for these treatments are usually radically different.