Comment
Network analyses to rank pharmacological treatments for
generalised anxiety disorder
Many treatments exist for generalised anxiety disorder.
Guidelines, such as the National Institute for Health
and Care Excellence guidelines,1 usually provide a
list of possible treatment options, but they do not
systematically list the treatments in preferential order
with regard to efficacy and tolerability because there are
few head-to-head comparisons of these interventions.
However, it makes a difference for patients when
one treatment reduces the Hamilton Anxiety Scale
(HAM-A), a common self-reported measure of feelings
of anxiety, from 26 points to 10 points, for example,
compared with another that has been shown to achieve
a reduction to 16 points. The network meta-analysis is
a newer method for putting all available treatments
in a ranking order: when treatment B was better than
treatment C and A was better than B, the network meta-
analysis can indirectly conclude that A is better than
C (even if A and C have never been compared directly).
This can conceivably be extended to all available
treatments by creating a network of direct and indirect
head-to-head and placebo comparisons.
In the Lancet, April Slee and colleagues2 report their
comprehensive network meta-analysis on medications
for generalised anxiety disorder and found that
some drugs—eg, duloxetine, pregabalin, venlafaxine,
or escitalopram—were efficacious with relatively
good acceptability. Quetiapine showed the largest
reduction of HAM-A scores but was poorly tolerated,
which is probably the reason why it is not licensed for
generalised anxiety disorder by the US Food and Drug
Administration (FDA) and the European Medicines
Agency. Agomelatine, a new drug which is not licensed
for generalised anxiety disorder, was also found to be
efficacious and well tolerated, but these findings were
based only on small sample sizes.
The strength of Slee and colleagues’ study is that it
also included seven unpublished papers and did not
only look for efficacy but also for acceptability. It is also a
special merit that the authors included 16 Chinese studies
not published in English. For example, mirtazapine for
generalised anxiety disorder was only studied in China.
The network meta-analysis is still a relatively new
method of analysing data from numerous trial settings,
www.thelancet.com Published online January 31, 2019 http://dx.doi.org/10.1016/S0140-6736(18)32180-9 1
which looks attractive but has some pitfalls.3,4 For
example, in traditional meta-analyses, funnel plot
Gary Waters/Ikon Images/Science Photo Library
methods like Duval and Tweedie’s so-called Trim and Fill
method are used to account for publication bias. These
methods have the assumption that small studies with
wide confidence intervals are less often published than
large studies with narrow confidence intervals when the
result was negative. However, these methods cannot be
used for network meta-analyses. Analogous methods
for this technique are still under development.5 Because Published Online
January 31, 2019
Slee and colleagues have identified unpublished trials by http://dx.doi.org/10.1016/
using trial registries and FDA files, however, the risk for S0140-6736(18)32180-9
publication bias in their study may be low. See Online/Articles
http://dx.doi.org/10.1016/
Another method for assessing comparative effective­ S0140-6736(18)31793-8
ness of health-care interventions is conventional meta-
analysis with pre–post comparisons. Instead of looking
for differences between an active treatment and placebo
control, the effect sizes between baseline and endpoint
can be compared over the large number of competing
interventions. By contrast with treated-versus-control
effect sizes, pre–post effect sizes might overestimate
the true effect of a treatment because unspecific effects
such as the placebo effect are also included in the overall
effect. However, when the unspecific effects are similar
in all trials, this method is useful for ranking different
therapies. The pre–post method compares treatments
that were not randomly allocated, which is a potential
limitation. In a recent pre–post meta-analysis of
treatments for all anxiety disorders,6 we also found high
effect sizes for venlafaxine, escitalopram, pregabalin,
benzodiazepines, quetiapine, and hydroxyzine.
Slee and colleagues did not look at psychological
therapies such as cognitive behavioural therapy (CBT),
which are often used in the treatment of generalised
anxiety disorder. A network meta-analysis that included
CBT would be complicated because the technique is
mostly compared with waitlist control groups, which
have a much lower pre–post effect size (Cohen’s d=0·20
in our meta-analysis) than placebo controls (d=1·29).6
Therefore, if placebo and waitlist controls were to be
grouped together in a network meta-analysis, this
discrepancy would lead to an overestimation of CBT
effects. In fact, we have found that with psychological
 Comment
therapies, only 60·4% of the effect size of average drugs
can be achieved.6 Moreover, in a follow-up analysis, we
could not find more enduring effects of psychological
therapies than of medications.7 Therefore, drugs remain
an important treatment option for generalised anxiety
disorder.
In the interest of patients, the most effective and
acceptable interventions should be used. Although
conventional and network meta-analyses have their
pitfalls, future treatment guidelines should make use
of them for the development of efficacy rankings for all
competing interventions.
*Borwin Bandelow, Dirk Wedekind
Department of Psychiatry and Psychotherapy, University of
Göttingen, D-37075 Göttingen, Germany
bbandel@gwdg.de
In the past 3 years, BB has been on the speaker’s board for Hexal, Janssen, Lilly,
and Lundbeck and on the advisory board for Lundbeck and Mundipharma for
the pharmacological treatment of anxiety disorders. DW has been on the
speaker’s board for Mundipharma and Pfizer and on the advisory board for
Servier for the pharmacological treatment of anxiety disorders.
2 www.thelancet.com Published online January 31, 2019 http://dx.doi.org/10.1016/S0140-6736(18)32180-9
1 National Institute for Health and Care Excellence. Anxiety: management of
anxiety (panic disorder, with or without agoraphobia, and generalised
anxiety disorder) in adults in primary, secondary and community care.
London: The British Psychological Society and The Royal College of
Psychiatrists, 2011.
2 Slee A, Nazareth I, Bondaronek P, Liu Y, Cheng Z, Freemantle N.
Pharmacological treatments for generalised anxiety disorder: a systematic
review and network meta-analysis. Lancet 2019; published online Jan 31.
http://dx.doi.org/10.1016/S0140-6736(18)31793-8.
3 Li T, Puhan MA, Vedula SS, Singh S, Dickersin K, Ad Hoc Network
Meta-analysis Methods Meeting Working Group. Network
meta-analysis—highly attractive but more methodological research is
needed. BMC Med 2011; 9: 79.
4 Chaimani A, Salanti G, Leucht S, Geddes JR, Cipriani A. Common pitfalls and
mistakes in the set-up, analysis and interpretation of results in network
meta-analysis: what clinicians should look for in a published article.
Evid Based Ment Health 2017; 20: 88–94.
5 Mavridis D, Welton NJ, Sutton A, Salanti G. A selection model for accounting
for publication bias in a full network meta-analysis. Stat Med 2014;
33: 5399–412.
6 Bandelow B, Reitt M, Rover C, Michaelis S, Gorlich Y, Wedekind D. Efficacy
of treatments for anxiety disorders: a meta-analysis.
Int Clin Psychopharmacol 2015; 30: 183–92.
7 Bandelow B, Sagebiel A, Belz M, Gorlich Y, Michaelis S, Wedekind D.
Enduring effects of psychological treatments for anxiety disorders:
meta-analysis of follow-up studies. Br J Psychiatry 2018; 212: 333–38.