Authors' reply

We appreciate these comments about our research and agree that it would be very important to identify, if possible, clinical, neurobiological and/or pharmacogenomic characteristics of patients with depression who are likely to benefit specifically from an antidepressant. We also understand Professor Naudet's scepticism about whether or not more complex statistical models of data analysis can or should be used for the purposes of regulatory review of novel medications. We note that although the concept of benefiter/non-benefiter is similar to that of responder/non-responder, there are fundamental differences. Although response can be calculated for each patient (either there is at least a 50% improvement or not), the benefiter variable cannot. It is the probability of the patient being a benefiter that is estimated, based on all information available for the patient (covariates and outcome variables). For instance, a patient with a baseline Montgomery-Åsberg Depression Rating Scale ¹ score of 30 and a Week 8 score of 5 will have a large probability of being a benefiter, while a patient with a baseline score of 30 and a Week 8 score of 25 will have a low probability of being a benefiter. A patient with a baseline score of 30 and a Week 8 score of 15 has an equal probability of belonging to either group. Although the classification of a patient as a responder or not may seem clear cut, in practice the difference between a non-responder and a responder can be due to a 1-point difference on an assessment scale.

We also think that it is important to point out that treatment with placebo in a randomised controlled trial (RCT) is not the same as no treatment. Beyond the frequent visits and detailed assessments that are part of the study protocol, patients in RCTs must meet specific inclusion criteria, and many are excluded for safety reasons. Thus, they are not representative of the patients seen in normal clinical practice. Patients participating in a placebo-controlled RCT also know that there is a chance that they are receiving placebo, possibly reducing their likelihood of responding, and patients randomised to placebo know that there is a chance that they are receiving active treatment, possibly increasing their chances of responding.

Finally, we do believe that the fundamental finding of our paper, namely that antidepressants convey large clinical benefits for a meaningful subgroup of patients with depression participating in contemporary RCTs, is a valid (‘true’) observation and, therefore, is not dependent on the use of a particular statistical model.