Efficacy of drug treatments for generalised anxiety disorder: systematic review and meta-analysisBMJ 2011; 342 doi: https://doi.org/10.1136/bmj.d1199 (Published 11 March 2011) Cite this as: BMJ 2011;342:d1199
All rapid responses
Re: Efficacy of drug treatments for generalised anxiety disorder: systematic review and meta-analysis
I was intrigued to read these comments on the original paper. Network meta-analysis techniques have evolved since the time our study was published, and it seems reasonable to state that the optimal approach for interpreting data is still evolving. Some of the approaches to data re-analysis adopted by Cipriani and colleagues (i.e. Higgins et al. 2012; Chaimani et al. 2013) were not clearly established at the time that the original article appeared. The original article cautioned against the finding relating to efficacy of fluoxetine (noting the rather limited overall evidence base, highlighting that there was only a single published study for fluoxetine and acknowledging it was an outlier, and stating '...the limited evidence on which the fluoxetine analysis was based casts doubt on the robustness of this finding'), and the re-analysis by Cipriani et al. confirms the original cautious statement.
Competing interests: All potential conflicts of interest were reported in the 2011 paper: there are no new potential conflicts of interest to declare.
We have already raised the issue about the inclusion in this network meta-analysis of the 3-arm study by Silverstone and colleagues, which compared fluoxetine, venlafaxine and placebo (Silvestone 2001). See http://www.bmj.com/content/342/bmj.d1199/rapid-responses. We further add some comments on the methodology employed.
In the network meta-analysis (Baldwin 2011) it is reported that the authors “tested the validity of the mixed treatment model by comparing the consistency of results between the mixed treatment meta-analyses and the direct comparison meta-analyses.” It has been previously suggested that this is not a valid method for assessing inconsistency because the network estimates (“mixed treatment meta-analyses”) are a combination of the direct (“direct comparison meta-analyses”) and indirect estimates and consequently they are not expected to differ much, even in the presence of substantial inconsistency (Nikolakopoulou 2014). As network meta-analysis technique is quickly gaining popularity in the literature (but quality is still variable – Nikolakopoulou 2014), we re-analyzed the 27 studies included in this network following the methods reported in the original article and compared our findings with the published results. We summarise here some of the main results we found and the full report with all our comments will be submitted soon to a peer-reviewed scientific journal.
In our re-analysis, both the design-by-treatment interaction test (Higgins 2012) and the loop-specific approach (Salanti 2009; Bucher 1997; Chaimani 2013) did not reveal statistically significant inconsistency, however the large uncertainty in the estimation of the ratio of odds ratios (ROR) between direct and indirect estimates in 3 out of 6 loops would have deserved further exploration as they can’t exclude the possibility of substantial inconsistency. It is probably not by chance that the loop including Silvestone 2001 study (fluoxetine, placebo, venlafaxine) had the largest upper confidence interval (CI) limit (ROR=2.51, 95% CI 1.00 to 11.43) and this reinforced our concerns about the inclusion of this study in the network (and therefore, the validity of the overall results from the published paper).
According to our re-analysis, in terms of response, all drugs - except fluoxetine, quetiapine and paroxetine - appeared to be statistically significantly more effective than placebo. However, no important differences existed between the ten interventions in terms of efficacy. So, it is very likely that the advantage of fluoxetine in the hierarchy of the published report was the consequence of using an inappropriate ranking measure that did not properly account for statistical uncertainty. In fact, we ranked efficacy of treatments using the surface under the cumulative ranking curve percentages (the so-called SUCRA - Salanti 2011, Chaimani 2013) and the mean ranks, and we found different results: lorazepam ranked first in terms of response (75.8%) and escitalopram for remission (80.8%) (Table 1). Fluoxetine was still among the drugs with potentially better efficacy profile, however the small differences among treatments suggest that the most sensible and appropriate conclusion from Baldwin 2011 data would have been that there is too large uncertainty around the hierarchy of treatments for generalized anxiety disorder.
Table 1. SUCRA values for response, withdrawals and remission.
Treatment Response Withdrawals Remission
Placebo 5.0% 91.9% 1.2%
Escitalopram 53.3% 51.0% 80.8%
Tiagabine 22.9% 39.4% 17.4%
Duloxetine 61.1% 18.3% 39.0%
Paroxetine 51.2% 30.7% 56.6%
Pregabalin 52.3% 65.5% -
Venlafaxine 60.1% 32.9% 64.8%
Lorazepam 75.8% 16.6% -
Fluoxetine 67.6% 71.4% 80.3%
Sertraline 60.3% 82.4% 60.0%
Quetiapine 40.5% 91.9% -
We hope that these comments will help researchers and clinicians recognize the complexity of conducting a high-quality network meta-analysis, which require a multi-disciplinary team with clinical and technical expertise to adequately cover each step of the research project, including skills in literature search, data extraction, and statistical analysis.
As for all systematic review and standard meta-analyses, results from network meta-analyses should be replicable. Published papers must include all of the information that readers need to completely understand how the study was conducted, independently assess the validity of the analyses and reach their own interpretations. The availability of the review protocol and the codes for statistical analyses should become soon a mandatory requirement for all network meta-analyses.
Finally, we would like to thank Gillian Sibbring and Complete Clarity for providing us with the original dataset and some of the codes they used for the analyses of Baldwin 2011 paper.
Andrea Cipriani,1 Anna Chaimani,2 Georgia Salanti,2 Stefan Leucht,3 John R. Geddes,1
1 Department of Psychiatry, University of Oxford, UK; 2 Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina, Greece; 3 Department of Psychiatry and Psychotherapy TU-München, Munich, Germany;
Baldwin D, Woods R, Lawson R, Taylor D. Efficacy of drug treatments for generalised anxiety disorder: systematic review and meta-analysis. BMJ 2011;342:d1199.
Bucher HC, Guyatt GH, Griffith LE, Walter SD. The results of direct and indirect treatment comparisons in meta-analysis of randomized controlled trials. J Clin Epidemiol 1997;50:683–91.
Chaimani A, Higgins JPT, Mavridis D, Spyridonos P, Salanti G. Graphical tools for network meta-analysis in STATA. PLoS One 2013;8:e76654.
Dechartres A, Altman DG, Trinquart L, Boutron I, Ravaud P. Association between analytic strategy and estimates of treatment outcomes in meta-analyses. JAMA 2014;312:623–30.
Higgins JPT, Jackson D, Barrett JK, Lu G, Ades AE, White IR. Consistency and insconsistency in network meta-analysis: concepts and models for multi-arm studies. Res Synth Meth 2012;3:98–110.
Nikolakopoulou A, Chaimani A, Veroniki AA, Vasiliadis HS, Schmid CH, Salanti G. Characteristics of Networks of Interventions: A Description of a Database of 186 Published Networks. PLoS ONE 2014;9:e86754.
Salanti G, Marinho V, Higgins JP. A case study of multiple-treatments meta-analysis demonstrates that covariates should be considered. J Clin Epidemiol 2009;62:857-64.
Salanti G, Ades AE, Ioannidis JP. Graphical methods and numerical summaries for presenting results from multiple-treatment meta-analysis: an overview and tutorial. J Clin Epidemiol 2011;64:163–71.
Silverstone PH, Salinas E. Efficacy of venlafaxine extended release in patients with major depressive disorder and comorbid generalized anxiety disorder. J Clin Psychiatry 2001;62:523–9.
Competing interests: No competing interests
In my practice GAD is common;patients mostly presenting with somatic
symptoms expecting early improvement in symptoms. Unfortunately Fluoxetin
as well as Sertralin fail in giving early relef as compared to
benzodiazepines which however cannot be continued for long for obvious
reasons. The drug Fluanxol however gives early relif in these cases in
varying doses depending on relapses.
Problem with Fluanxol in treatment of GAD is that there is scanty
authoritative references in literature and this study fails to mention it.
Competing interests: No competing interests
Thank you for your considered comments. We are aware of potential
biases arising from the data included in our study. We either have to make
the best use of the data available to us, or ignore it.
We outlined the methodological limitations of our analysis with
respect to the fluoxetine data in the manuscript discussion. However, in
addition to our acknowledgment of the Silverstone and Salinas study's
limitation pertaining to the small sample size, we perhaps ought to have
highlighted further limitations inherent in its study design.
To your second point regarding the inconsistency between the results
of the direct and indirect analyses, the indirect comparisons carried an
influence in our analysis that perhaps may be more than typical, owing to
the characteristics of the network (figure 1), with most studies comparing
active treatments versus placebo.
Further to this, the variability that was apparent in placebo
responses (Piercy et al 1996 - ref 63 from paper) would mean that the odds
ratio through indirect comparison (via placebo) would be of a relatively
large magnitude, hence having an increased weight in the mixed treatment
We acknowledged the relatively small number of studies for some
active treatments in the manuscript and the effect of this, on fluoxetine
ranking, is reflected in Figure 5, where as well as being ranked first
(most likely to be the best treatment) it also ranks relatively highly in
being the worst treatment (exemplified for withdrawals due to AEs).
We would urge any further work in this area to consider this point
closely, by comparing direct and indirect models using analytical methods.
Guobing Lu, A. E Ades. Journal of the American Statistical
Association. June 1, 2006, 101(474): 447-459).
Competing interests: All potential competing interests are described in the original manuscript.
Just for clarification, in the full-text version of our manuscript
published online, sertraline is ranked first, in
table 1 and in figures 4 and 5. This means that sertraline was associated
with the fewest number of withdrawals due to adverse events. Ranking
highest for withdrawals, does not mean that it was associated with the
highest number of withdrawals. It pays to read the full version as well as
the pico summary. We hope this resolves any misunderstanding.
Competing interests: All potential competing interests are described in the original publication.
This relates to the version in the Journal.
In the written text Sertraline is mentioned as the first for
tolerability (49.3%). However, the table ranks Sertraline first for
"percentage of patients withdrawing from the study because of adverse
events" - and indeed has the highest number in brackets (again, 49.3%).
One of these has to be wrong.
Competing interests: No competing interests
The results of this Multiple-Treatments Meta-analysis (MTM), which
appraised the evidence for comparative efficacy and tolerability of drug
treatments and placebo in patients with generalised anxiety disorders,
suggest that fluoxetine has the highest probability to be the best
treatment in terms of response and remission. As reported in the
accompanying editorial, this conclusion is based on 33 patients who
received fluoxetine in one three-arm study comparing fluoxetine with
venlafaxine and placebo. More precisely, the data were taken from a post-
hoc subgroup analysis of a study that was prospectively designed to
demonstrate the efficacy of venlafaxine in outpatients with depression and
anxiety. Following completion of the study, a subset of patients who had a
recorded formal DSM-IV diagnosis of generalized anxiety disorder was
identified in the dataset and analyzed. The implications of this design
include that patients in this subgroup analysis had a diagnosis of major
depressive disorder in addition to that of generalized anxiety disorder.
Moreover, random allocation was not stratified by comorbid diagnosis of
generalized anxiety disorder, and the groups were not similar at baseline
in terms of socio-demographic and clinical characteristics.
Clearly, we are all well aware of the problems associated with
subgroup analyses in clinical trials, but what strikes here is that the
results of this subgroup analysis were not in favour of fluoxetine over
venlafaxine: the adjusted mean difference from placebo (HAM-A total score)
was 2.5 (with a 95% confidence interval (95% CI) between -1.7 and 6.7) for
fluoxetine and 4.5 (95% CI 0.2 to 8.7) for venlafaxine. In terms of
responders, 59% responded to venlafaxine and 45% to fluoxetine. Remission
rate was 12/33 (36%) for fluoxetine and 10/32 (31%) for venlafaxine.
In addition to raising concerns on the inclusion of the fluoxetine
study, we wonder how the MTM methodology has been able to rank fluoxetine
as first treatment on the basis of this direct evidence. We had experience
of using the MTM approach in the field of antidepressants for major
depression and we found that combining direct and indirect evidence
increased precision without materially conflicting results, which does not
seem the case in the present analysis.
Competing interests: No competing interests