Authors' reply

I agree that the evidence base for strategies for treatment-resistant depression has been poor but it is (slowly) improving (DeBattista, 2006). Dr Vergouwen's suggestion that predictors of remission should be sought scientifically is most welcome. The studies mentioned are some of a number that look at the proportion of patients who respond late when early antidepressant response is disappointing (e.g. Mulsant et al, 2006). I am sure it will not be long before someone performs a meta-analysis yielding more-conclusive results. However, in clinical practice the alternative to continuing a drug which has generated a poor response is most commonly switching to another. However, from an evidence base standpoint this is where things get complex.

When considering analysis of benefit from a switch strategy after a certain number of weeks (say an 8-week v. 4-week switch with follow-up at 24 weeks), the methodology of an ideal trial is not straightforward and hence rare (to the point of invisibility!) in the literature. Three arms are required. Arm 1 includes patients who switch if non-responsive at 4 weeks; arm 2 those who switch if non-responsive at 8 weeks and, equally importantly, arm 3 patients who do not switch and stay on their original antidepressant for the duration of the trial. The third arm establishes how many would continue to enter remission even if initially non-responsive. Comparing switch with maximisation or augmentation or combination strategies would also ideally require a study of similar design. I know of no such studies, and the recruitment of the necessary number of patients with some level of treatment resistance is very difficult. A recent review of combination trials for treatment-resistant depression found only two that were randomised against a drug plus placebo arm (Dodd et al, 2005).

The other important issue is exactly how to separate responders from non-responders (or remitters from non-remitters) (Israel, 2006). In my view, because any definition of response is arbitrary, the threshold taken to define response (20%, 30% or 50% improvement, for example) will affect the success of the switch strategy. The main danger of switching too early is robbing a patient who was on a trajectory of good improvement from continuing successful treatment. The danger of switching too late is leaving a patient with distressing symptoms longer than necessary without effective treatment. In reality, ratings on a depression scale at 4 or 8 weeks after starting treatment will be somewhere between baseline and entirely asymptomatic - thus virtually all patients could be considered ‘partial responders’. Many areas of psychopharmacology are moving towards early identification and treatment. I doubt that treatment-resistant depression will be the exception.