Statistical Alchemy for Drug Treatment of Generalized Anxiety Disorder: A Commentary on the Meta-Analysis by Baldwin et al. [BMJ 2011;342:d1199]

late well in the meta-analysis [2] , probably since this categorization is likely to be affected by the different characteristics of the trials, maximizing the inherent tendency of meta-analyses to violate the internal validity of the placebo comparison. Another very questionable procedure was to infer tolerability from withdrawals that may have been due to a number of reasons. This choice of outcome measures excluded a great number of studies concerned with older and low-cost medications, such as benzodiazepines, despite a large body of evidence pointing to their efficacy in generalized anxiety disorder [7] . Only two trials included a benzodiazepine (lorazepam), and no trial used a tricyclic. Berney et al. [8] performed a systematic review (up to 2006) of controlled trials on anxiety disorders that compared antidepressant drugs with benzodiazepines, and could identify only one trial with newer antidepressants. They concluded that the major change in prescribing patterns from benzodiazepines to newer antidepressants in anxiety disorders occurred without any comparative evidence [8] . A major driving force behind the change was the risk of dependence with benzodiazepines, even though withdrawal symptoms frequently occur with newer antidepressants upon tapering and discontinuation [9] , even in optimal conditions, and do not necessarily subside within a few weeks [10] . A very recent report has raised preliminary but alarming data on In the mid-1990s, Alvan R. Feinstein [1] compared meta-analyses to the alchemy that existed before modern scientific chemistry. The analogy was the hope to convert existing things into something better (changing base metals into gold) and the work with material that was heterogeneous and poorly identified. It was difficult to foresee in those days, however, the type of ‘grand prix’ model of meta-analysis applied by Baldwin et al. [2] to the drug treatment of generalized anxiety disorder, which was commented upon in an editorial by Furukawa [3] , who had also authored a similar type of meta-analysis [4] . Baldwin et al. [2] conducted a systematic review of randomized controlled drug trials for generalized anxiety disorder. Trials were selected if they could allow determination of ‘response’ (the proportion of patients who experienced a reduction of at least 50% of their baseline score on the Hamilton anxiety scale) and ‘remission’ (the proportion of patients below a certain cutoff point on the scale). The authors failed to justify the choice of outcome measures, which have two major disadvantages: in generalized anxiety disorder we are dealing with a decrease in symptom intensity more than with a simple presence or absence [5] , and categorization is hindered by the availability of different versions and translations of the Hamilton scale [6] . Indeed, unlike what is commonly found in individual trials, response and remission did not correReceived: April 18, 2011 Accepted after revision: April 27, 2011 Published online: June 3, 2011