Andrea Cipriani from the University of Oxford and colleagues have recently published the largest and most comprehensive meta-analysis of the efficacy of antidepressants in the Lancet (Cipriani et al., Lancet 2018). The paper is worth dealing with more intensively.
The authors examined all clinical (short-term) trials that compared 21 antidepressants since their approval until January 2016, either with each other or with placebo. With these studies, they performed a network meta-analysis, totaling more than 116,000 patients. The primary endpoint was response rate.
The main results of the analysis are:
All 21 antidepressants were significantly superior to placebo.
However, there were significant differences in the odds ratios (OR, which represents the chance to reach a response with the antidepressant relative to the chance to reach a response with placebo) between the different compounds. If the antidepressants are ranked, amitriptyline is at the top (OR 2.13) and reboxetine at the bottom (OR 1.37). That means, with amitriptyline a response to therapy is 2.13 times more likely than with placebo. In detail, the results are as follows:
Acceptance of therapy was measured as the dropout rate. Here, two compounds were better than placebo, agomelatine and fluoxetine. The only drug that resulted in a higher dropout rate than placebo was clomipramine. All other antidepressants were not different from placebo in this regard.
Looking at the head-to-head comparisons of the individual antidepressants with each other, there were only a few significant differences. In some of these comparisons, there were superiorities for agomelatine, amitriptyline, escitalopram, mirtazapine, paroxetine, venlafaxine and vortioxetine over other antidepressants, whereas fluoxetine, fluvoxamine, reboxetine and trazodone were inferior in these comparisons.
However, these meta-analytical comparisons should not be overstated. For example, agomelatine and vortioxetine are nearly equal in the ranking above, but in the only direct comparative study of the two against each other, vortioxetine was significantly superior to agomelatine (Montgomery et al., Hum Psychopharmacology 2014). In this context, it seems particularly surprising that the German Federal Joint Committee (G-BA) in its benefit assessment of vortioxetine had excluded agomelatine as a reference substance with the following reasoning (see my post in my old Mind-and-Brain blog, only in German): “Agomelatine is not considered as a suitable comparator in view of the inconsistent data available with evidence of inferiority to SSRIs in acute therapy efficacy.” This raises the question: what does the G-BA know what the authors of the present meta-analysis do not know?
Other results of the meta-analysis are remarkable:
smaller and older studies presented larger effects of the active interventions versus placebo
within the head-to-head comparisons, when a treatment was the novel or experimental drug of comparison, it appeared to be significantly more effective than when that same treatment was the older or control drug of comparison
response to the same antidepressant was on average smaller and dropouts more likely to occur in placebo controlled trials than in head-to-head studies
There are questions left:
Why are antidepressants, with the exception of fluoxetine, not effective in children and adolescents, but in adults? Is neurobiology really so much different? Why don’t we have an idea of how to explain this differential effectiveness?
Why is desvenlafaxine so much less effective (at least in the ranking) than venlafaxine? When venlafaxine is administered, its active metabolite desvenlafaxine most likely contributes much more to the antidepressant activity than the parent.
„These data suggest that antidepressants may be less effective than their wide marketing suggests. Short-term benefits are small and long-term balance of benefits and harms is understudied. I discuss how the use of many small randomized trials with clinically non-relevant outcomes, improper interpretation of statistical significance, manipulated study design, biased selection of study populations, short follow-up, and selective and distorted reporting of results has built and nourished a seemingly evidence-based myth on antidepressant effectiveness and how higher evidence standards, with very large long-term trials and careful prospective meta-analyses of individual-level data may reach closer to the truth and clinically useful evidence.“
Ioannidis also wrote what we need in the future:
“To settle this important matter with robust evidence, we need large trials with 100-fold more patients than what has been the norm to-date. These trials should be linked to long-term follow-up registries with thorough recording of long-term outcomes for both efficacy and harms. Efficacy outcomes should include long-term management and evolution of depression, suicide and major life events. These trials should be designed in a way that they do not simply have power to show a nominally statistically significant benefit (of debatable clinical importance), but for demonstrating whether the benefit is conclusively within the range of clinical importance and has a long-term portend.”