Efficacy and Acceptability of Pharmacological Treat-

ments for Depressive Disorders in Primary Care:

Systematic Review and Network Meta-Analysis
Klaus Linde, MD1 ABSTRACT
Levente Kriston, PhD2 PURPOSE The purpose of this study was to investigate whether antidepressants
Gerta Rücker, PhD3 are more effective than placebo in the primary care setting, and whether there
are differences between substance classes regarding efficacy and acceptability.
Susanne Jamil1
METHODS We conducted literature searches in MEDLINE, Embase, Cochrane Cen-
Isabelle Schumann, MD1 tral Register of Controlled Trials (CENTRAL), and PsycINFO up to December 2013.
Karin Meissner, MD1,4 Randomized trials in depressed adults treated by primary care physicians were
included in the review. We performed both conventional pairwise meta-analysis
Kirsten Sigterman, MD1 and network meta-analysis combining direct and indirect evidence. Main out-
Antonius Schneider, MD1 come measures were response and study discontinuation due to adverse effects.
Institute of General Practice, Technische
1
RESULTS A total of 66 studies with 15,161 patients met the inclusion criteria. In
Universität München, Munich, Germany network meta-analysis, tricyclic and tetracyclic antidepressants (TCAs), selective
2
Department of Medical Psychology, serotonin reuptake inhibitors (SSRIs), a serotonin-noradrenaline reuptake inhibi-
University Medical Center Hamburg- tor (SNRI; venlafaxine), a low-dose serotonin antagonist and reuptake inhibitor
Eppendorf, Hamburg, Germany (SARI; trazodone) and hypericum extracts were found to be significantly superior
3
Institute of Medical Biometry and Statis- to placebo, with estimated odds ratios between 1.69 and 2.03. There were no
tics, University Medical Center Freiburg, statistically significant differences between these drug classes. Reversible inhibi-
Freiburg, Germany tors of monoaminoxidase A (rMAO-As) and hypericum extracts were associated
with significantly fewer dropouts because of adverse effects compared with TCAs,
Institute of Medical Psychology, Ludwig-
4

Maximilians-University Munich, Munich,

SSRIs, the SNRI, a noradrenaline reuptake inhibitor (NRI), and noradrenergic and
Germany specific serotonergic antidepressant agents (NaSSAs).

CONCLUSIONS Compared with other drugs, TCAs and SSRIs have the most solid
evidence base for being effective in the primary care setting, but the effect size
compared with placebo is relatively small. Further agents (hypericum, rMAO-As,
SNRI, NRI, NaSSAs, SARI) showed some positive results, but limitations of the
currently available evidence makes a clear recommendation on their place in
clinical practice difficult.

Ann Fam Med 2015;13:69-79. doi: 10.1370/afm.1687.

INTRODUCTION

E
pidemiological studies indicate that depressive disorders are highly
prevalent in the general population worldwide.1 Most cases are seen
and managed in primary care, and only a small proportion of these
are referred to specialized care.2 Most research findings upon which treat-
ment decisions are made, however, have involved patients cared for by
Conflicts of interest: authors report none. mental health specialists.3 It is not fully clear whether the findings from
trials in specialty settings can be generalized to primary care. There is
some evidence suggesting that primary care patients with depressive
CORRESPONDING AUTHOR disorders are less severely depressed,4 experience a milder course of ill-
Klaus Linde, MD ness,5 have a distinct symptom profile with more complaints of fatigue
Institute of General Practice and somatic symptoms,6 and are more likely to have accompanying physi-
Technische Universität München
Orleansstr. 47 cal complaints7 than are patients referred to specialty mental health care.
D-81667 München, Germany These differences could have an impact on guideline development and
[email protected] management of depression in primary care.

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

69
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

Pharmacological interventions are a cornerstone providing primary care in their respective countries. We
of antidepressant treatment,8 yet there is an ongoing excluded trials that recruited patients from community-
debate as to whether their relatively small effects com- based centers specializing in mental health care. Trials
pared with placebo observed in clinical trials are clini- had to report results of at least 1 of the following out-
cally relevant.9,10 Meta-analyses restricted to primary comes: response to treatment, remission, mean score on
care patients have been performed for some antidepres- a depression scale (posttreatment or change from base-
sant drugs.11-14 Researchers conclude that these treat- line), frequency of adverse effects, or study discontinua-
ments are effective in primary care settings. It is not tion (for any reason or from adverse effects).
possible, however, to determine whether the available Four authors (K.L., K.S., S.J., and K.M.) reviewed
treatment options are comparable (ie, whether some all trials for screening selection and extraction. In the
treatments are superior to others in primary care). Tradi- first screening 1 reviewer excluded clearly irrelevant
tional meta-analyses are restricted to the direct compari- records. In the second screening, 2 reviewers indepen-
son of 2 interventions by pooling data only from trials dently checked all remaining records against inclusion
with similar treatment arms. Network meta-analysis criteria. The full texts of articles were obtained for all
allows for the estimation of relative effects of inter- records that were considered potentially relevant or
ventions that have not been compared directly.15 We unclear and were assessed formally for eligibility by at
systematically reviewed randomized trials of pharmaco- least 2 reviewers independently. Disagreements were
logical treatments of depression in primary care settings. resolved by discussion.
We used conventional and network meta-analysis to
investigate whether there is evidence that in the primary Data Extraction and Assessment of Risk of Bias
care setting antidepressants are more effective than pla- In the first extraction step, at least 2 reviewers inde-
cebo and whether there are differences in efficacy and pendently extracted information on patients, methods,
acceptability between the various substance classes. and results of all included studies using a pretested
form. The Cochrane Collaboration’s tool for assess-
ing risk of bias was used to assess internal validity.18
METHODS As the included studies reported results on efficacy
Details of the methods have been described in our in a highly diverse and often incomplete manner, we
published protocol.16 We also reviewed trials on psy- performed an additional extraction round using a
chological interventions. Because trials of pharmaco- preference approach for extracting data for response
logical and psychological interventions differ greatly and remission (Supplemental Appendix, sTable 2). This
regarding recruitment strategies, patients, control additional extraction was done by 1 reviewer (K.L.),
interventions and outcomes, these trials are analyzed whereas a second reviewer (K.S. or K.M.) cross-
separately and reported in a companion article pub- checked all extracted data and recalculated imputa-
lished in this issue.17 tions. Adequacy of dosages tested was checked against
guideline recommendations.19
Search Strategy and Study Selection
We searched MEDLINE, Embase, Cochrane Central Comparisons and Outcomes
Register of Controlled Trials (CENTRAL), and Psyc- As prespecified in the study protocol, we analyzed
INFO (main search June 2011, last update searches drugs according to their substance class.16 Efficacy
December 2013; see Supplemental Appendix, section end points were response (primary outcome defined as
1, for the complete MEDLINE search strategy). We at least a 50% score reduction on a depression scale)
searched trial registries for unpublished and ongoing and remission (secondary outcome defined as having a
studies. In addition, we screened references from iden- symptom score below a fixed threshold). Patients with
tified trials and published systematic reviews focusing missing data were considered nonresponders or nonre-
on primary care studies of depression treatments11-14 for mitters. In cases where responder and remission data
additional trials. were not reported, we imputed it from available score
We included randomized controlled trials that com- data.20 Acceptability outcomes were discontinuation
pared drugs belonging to different pharmacological (dropout) because of adverse effects (primary accept-
classes with one another or placebo in the treatment ability outcome), discontinuation for any reason, and
of adult patients having prevalent or incident unipolar the number of patients experiencing adverse effects.
depressive disorder. Patients had to be recruited from For our main analysis, we used data after completion of
a primary care setting consisting of family physicians’ treatment. In the rare case in which treatment duration
or general practitioners’ private practices, primary care was longer than 3 months, the measurement point clos-
clinics or networks, internists, or other nonpsychiatrists est to 3 months was used. For the analysis of long-term

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

70
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

effects, we used the measurement closest to 6 months and reuptake inhibitor (SARI), in 10 a noradrenergic
after randomization. and specific serotonergic antidepressive agent (NaSSA),
in 6 a reversible inhibitor of monoamine oxidase A
Statistical Analyses (rMAO-A), in 14 a hypericum extract (St. John’s wort)
We used the odds ratio as the effect measure. Conven- (Table 1, which also displays single agents tested) and
tional meta-analyses of pairwise direct comparisons in 24 a placebo. Initial dosages were always within the
within studies were performed using the inverse vari- range of recommendations19 for starting treatment. In
ance weighted random effects model option in the 11 treatment arms (from 10 trials), however, dosages
Cochrane Informatics and Management Department remained below recommended standard dosage for all
RevMan 5.2 software. For network meta-analyses, or most patients: in all 5 trazodone arms (150 mg in 3
a Bayesian framework (WinBUGS and R interface and 100 to 200 mg in 2 trials), 2 amitriptyline arms (75
R2WinBUGS [http://www.r-project.org/]) following mg and 50 to 100 mg, respectively), 2 mianserin arms
the recommendations of the UK Decision Support (40 mg and 30 to 60 mg, respectively), 1 clomipramine
Unit of the National Institute of Health and Clinical (40 mg), and 1 imipramine arm (mean dose 58 mg). In
Excellence (NICE) was used to combine direct and many trials daily dosages were in the lower range of
indirect evidence.21 To examine for inconsistency, we recommendations for standard dosages. Thirty-eight
compared the results of the consistency model with (58%) studies were restricted to patients with major
that of an inconsistency model using the
deviance information criterion.22 The Figure 1. Study selection process.
potential impact of 4 prespecified (risk of
bias, diagnostic subtype of depression, 18,946 Records identi- 21 Additional records
mean age of participants, and duration fied through electronic identified through
database searching other sources
of treatment) and 2 post hoc defined
covariates (dosage and sample size) was
analyzed using a meta-regression model.23
Sensitivity analyses were performed
10,275 Records after
excluding outlier studies. Funnel plots duplicates removed
were produced for all direct comparisons
with data from at least 5 trials.
9,401 Clearly irrelevant
records excluded
RESULTS
Study Selection and Characteristics 874 Records screened
of Included Studies by a single reviewer

A total of 66 studies with 15,161 patients

met the inclusion criteria (Figure 1; see 609 Further irrelevant
also the Supplemental Appendix, Section records excluded
2 for references and Section 3 for charac-
teristics of individual studies). Four trials
265 Full-text reports
were available as unpublished reports only assessed for eligibility
(3 from a drug company’s trial registry and
1 thesis). The 66 trials included 147 treat-
196 Full-text reports excluded:
ment groups. After the pooling of groups
12 Inadequate study design
in which different dosages of the same
50 Inadequate diagnosis/setting
agent or 2 agents of the same substance
35 Inadequate intervention
class had been tested, 140 treatment 72 Psychological interventions
arms formed the basis of our analyses. In 16 Inadequate comparator group
37 treatment arms patients had received 8 No relevant outcome measure
a tricyclic or tetracyclic antidepressive 3 Study protocol only
agent (TCA), in 37 a selective serotonin
reuptake inhibitor (SSRI), in 6 a serotonin-
66 Studies (in 65 main + 4
noradrenaline reuptake inhibitor (SNRI), additional articles = 69 reports)
in 1 a noradrenaline reuptake inhibitor included in systematic review
(NRI), in 5 a serotonin (5-HT2) antagonist

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

71
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

depression only, whereas 28 (42%) studies also included from attrition seemed low in 21 (32%) studies and bias
patients who had other depressive disorders or did not resulting from the reporting of outcomes was found in
provide details on the exact type of depressive disor- 49 (74%) studies. Overall risk of bias was considered to
ders involved (Table 2). be low in 11 (17%), unclear in 23 (35%), and high in 32
(49%) studies.
Risk of Bias
Seventeen trials (26%) reported the method of gener- Efficacy
ating the allocation sequence, and 14 (21%) reported Fifty-nine (89%) studies provided sufficient data to be
an adequate method of allocation concealment included in the analysis of the main outcome measure
(Supplemental Appendix, sTable 3 for the assessment response. For 17 of the 36 possible comparisons (46%)
of individual trials). All remaining trials provided no there was at least 1 head-to-head trial (Figure 2 for the
information on this issue. In 57 (86%) double-blind tri- network and Table 3 for pooled estimates; Section 4
als without clear indications of unblinding, we consid- of the Supplemental Appendix provides a forest plot
ered the risk of bias to be low; in 9 (14%) trials either with odds ratios of all individual trials). More than 3
no blinding or unblinding seemed likely. Bias resulting comparative trials were available for TCAs vs SSRIs
(n = 18), hypericum extracts vs
placebo (n = 9), TCAs vs placebo
Table 1. Overview of Antidepressant Medications Tested
in the Included Trials and in the Meta-Analysis
(n = 8), SSRIs vs placebo (n = 7)
and SSRIs vs hypericum extracts
Medication Class Trial Groups/Arms and Specific Medication Tested (n = 6). In 6 of the 13 direct com-
TCA, tricyclic and tetracyclic 40 S eparate groups in trials, after pooling of groups in stud- parisons with 2 or more trials,
antidepressants ies with more than 1 TCA, 37 arms in analyses there was statistical heterogene-
14 Amitriptyline (1 pooling of 2 groups with different ity (I2 >40% and/or P <.1 in the
dosages, 1 pooled with dothiepin)
8 Imipramine (1 pooled with desipramine)
χ2 test). TCAs, SSRIs, SNRI,
8 Dothiepin/dosulepin (1 pooled with amitriptyline) and hypericum extracts were
3 Clomipramine significantly superior to placebo,
1 Amineptine but SNRI and NaSSAs were not.
1 Desipramine (pooled with imipramine) There were no trials compar-
1 Doxepin ing NRI, SARI, and rMAO-As
1 Lofepramine with placebo. Funnel plots of
1 Maprotiline comparisons with placebo were
1 Tianeptine
difficult to interpret because of
1 Individualized TCA
SSRI, selective serotonin reup- 38 A
 rms in total, 37 after pooling of 2 arms in a study with
the small numbers of studies per
take inhibitors 2 SSRI 35 arms substance class (Supplemental
12 Paroxetine Appendix, sFigures 3-7 for fun-
9 Fluoxetine nel plots). Visual inspection
5 Sertraline suggested asymmetry for trials
5 Citalopram (1 pooled with escitalopram) comparing hypericum extract
2 Fluvoxamine
and placebo. The only significant
3 Escitalopram (1 pooled with citalopram)
differences between substance
1 Individualized SSRI
SNRI, serotonin-noradrenaline 6 Arms in total
classes indicated superiority of
reuptake inhibitor 6 Venlafaxine TCAs compared with NaSSAs
NRI, noradrenaline reuptake 1 Arm in total and rMAO-As, and of SARI com-
inhibitor 1 Reboxetine pared with NaSSAs.
SARI, serotonin (5-HT2) antag- 5 Arms in total In the network meta-analysis,
onists and reuptake inhibitor 5 Trazodone TCAs, SSRIs, SNRI, SARI (low-
NaSSA, noradrenergic and 9 Arms in total dose trazodone) and hypericum
specific serotonergic antide- 8 Mianserin (2 underdosing)
pressive agents extracts were found to be sig-
1 Mirtazapine
nificantly superior to the pla-
rMAO-A, reversible inhibitors 8 Arms in total, 6 in analyses
of monoaminoxidase A 4 Moclobemide,
cebo, but effects were relatively
4 Minaprine (2 pooling of different dosages) small, with estimated odds ratios
Hypericum, extracts from 15 A
 rms in total, 14 after pooling of 2 arms testing different between 1.69 and 2.03 (Table
Hypericum perforatum L. extracts in 1 study; 12 different extracts 3). There were no significant
(St. John’s wort)
differences between these drug

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

72
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

classes, but 95% credible intervals

Table 2. Characteristics of Included Studies were wide except for the comparison
Publication Year between TCAs and SSRIs. NRI, NaS-
Characteristic 1996 (1971, 2012) SAs, and rMAO-As were not signifi-
Number of patients cantly different from placebo. TCAs,
Total 15,161 SSRIs, and hypericum extracts were
Median (minimum, maximum) 162 (21, 1,385) significantly superior to NaSSAs and
Diagnosis, No. (%) rMAO-As. Hypericum extracts were
Major depression only 38 (58) also more effective than NRI. SNRI
Mixed/unclear 25 (38)
and low-dose SARI were superior to
Minor depression/dysthymia 3 (5)
NaSSAs. We found no evidence of
Classification of depression, No. (%)
Diagnostic and Statistical Manual of Mental Disorders 15 (23)
inconsistency between direct and
(Fourth Edition) indirect comparisons.
Diagnostic and Statistical Manual of Mental Disorders 24 (36) Fixed-effects analyses yielded
(Third Edition)
results very similar to random-effects
International Classification of Diseases, Tenth Revision 6 (9)
analyses, thus providing no evidence
International Classification of Diseases, Ninth Revision 4 (6)
Research Diagnostic Criteria 4 (6)
for heterogeneity. Meta-regression
Not reported/other 13 (20) analyses did not show a significant
Restricted to patients >55 y, No. (%) 8 (12) influence of the type of depression
Median length of treatment in weeks (minimum, maximum) 6 (4, 52) (major depression or not), risk of bias,
Overall risk of bias, No. (%) restriction to elderly patients, timing
High (high risk in 1 or more items) 32 (49) of the outcome measurement, under-
Unclear (no item high risk, <3 low risk) 23 (35) dosing, and sample size on treatment
Low (at least 3 low, no high risk) 11 (17) outcome. Correspondingly, model
Data available for meta-analysis, No. (%)
estimates for patients with major
Response 59 (89)
depression, for studies with adequate
Remission 55 (83)
Total number of patients dropping out 60 (91)
drug dosages, and for large trials were
Number of patients dropping out due to adverse effects 58 (88) very similar to the unadjusted main
Number of patients reporting adverse effects 40 (61) estimates (Supplemental Appendix,
Section 6). The exclusion of outlier
studies did not have any relevant
impact on the findings.
Figure 2. Network for the main efficacy outcome response.
In secondary efficacy analyses
Placebo using remission as the outcome, all
drug classes were superior to placebo
8 9 without significant differences between
7 2 drug classes (Supplemental Appen-
TCA Hypericum
dix, Sections 6 and 7). Only 9 trials
6
18 reported long-term outcomes (>12
3 weeks), and of these, only 1 included a
rMAO-A placebo control group. Point estimates
SSRI
2 were similar (between 1.55 and 1.78)
2 3 for the drugs that could be included
3
in the network (SSRI, TCA, SNRI,
SNRI NaSSA NaSSA, and hypericum) but confi-
2
dence intervals were very wide, and
4 differences compared with placebo
were not statistically significant.
NRI Low-dose SARI

Figures indicate the number of direct comparisons (lines without figure indicate 1 comparison) Acceptability
Hypericum = extract from Hypericum perforatum L.; NaSSA = noradrenergic and specific serotonergic Fifty-eight studies (88%) reported
antidepressive agent (mianserin, mirtazapine); NRI = noradrenaline reuptake inhibitor (reboxetine); rMAO-
A = reversible inhibitor of monoaminoxidase A (moclobemide, minaprine); SARI = serotonin (5-HT2) the number of patients dropping
antagonist and reuptake inhibitor (trazodone); SNRI = serotonin-noradrenaline reuptake inhibitor (venla- out because of adverse effects. The
faxine); SSRI = selective serotonin reuptake inhibitor; TCA = tricyclic and tetracyclic antidepressant.
data for 5 studies could not be used

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

73
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

Table 3. Results of Conventional and Network Meta-Analyses for the Main Efficacy Outcome Response

Low-Dose
Medication TCA SSRI SNRI NRI SARI
TCA
OR (95% CI)a 0.95 (0.84-1.08) 0.83 (0.37-1.89) na 1.43 (0.60-3.42)
No. of studies; I2, %; P valueb 18; 0; .45; 2; 55; 0.14 na 2; 79; .03
OR (95% CI)c 1.04 (0.88-1.21) 1.00 (0.71-1.40) 1.23 (0.73-2.14) 0.97 (0.65-1.47)
SSRI
OR (95% CI)a 1.05 (0.93-1.19) 0.92 (0.67-1.25) 1.18 (0.85-1.65) na
No. of studies;I2, %; P valueb 18; 0; .45 4; 64; .04 1 na
OR (95% CI)c 0.97 (0.82-1.13) 0.96 (0.71-1.29) 1.19 (0.71-2.04) 0.94 (0.63-1.43)
SNRI
OR (95% CI)a 1.20 (0.53-2.72) 1.09 (0.79-1.49) na na
No. of studies; I2, %; P valueb 2; 55; .14 4; 64; .04 na na
OR (95% CI)c 1.00 (0.72-1.41) 1.03 (0.77-1.42) 1.23 (0.67-2.28) 0.97 (0.60-1.63)
NRI
OR (95% CI)a na 0.84 (0.61-1.18) na na
No. of studies; I2, %; P valueb na 1 na na
OR (95% CI)c 0.60 (0.41-0.91) 0.63 (0.41-0.87) 0.61 (0.36-1.02) 0.79 (0.41-1.57)
Low-dose SARI
OR (95% CI)a 0.70 (0.29-1.67) na na na
No. of studies; I2, %; P valueb 2; 79; .03 na na na
OR (95% CI)c 1.03 (0.68-1.55) 1.06 (0.70-1.58) 1.03 (0.61-1.67) 1.26 (0.63-2.47)
NaSSA
OR (95% CI)a 0.26 (0.11-0.60) 0.95 (0.49-1.85) na na 0.47 (0.31-0.73)
No. of studies; I2, %; P valueb 1 3; 60; .08 na na 4; 0; .56
OR (95% CI)c 0.64 (0.46-0.93) 0.67 (0.47-0.95) 0.64 (0.41-0.98) 0.79 (0.42-1.45) 0.63 (0.41-0.93)
rMAO-A
OR (95% CI)a 0.61 (0.43-0.86) na na na na
No. of studies; I2, %; P valueb 3; 0; .79 na na na na
OR (95% CI)c 0.61 (0.41-0.91) 0.63 (0.41-0.97) 0.61 (0.36-1.02) 0.74 (0.37-1.52) 0.59 (0.35-1.04)
Hypericum
OR (95% CI)a 1.15 (0.71-1.85) 1.14 (0.81-1.59) na na na
No. of studies; I2, %; P valueb 2; 24; .25 6; 43; .12 na na na
OR (95% CI)c 1.16 (0.81-1.50) 1.20 (0.96-1.51) 1.16 (0.80-1.68) 1.43 (0.81-2.64) 1.13 (0.72-1.73)

CI = confidence interval; CI = credible interval; hypericum = extract from Hypericum perforatum L.; rMAO-A = reversible inhibitor of monoaminoxidase A (moclobemide,
minaprine); na = not available; NaSSA = noradrenergic and specific serotonergic antidepressive agent (mianserin, mirtazapine); NRI = noradrenaline reuptake inhibitor
(reboxetine); OR = odds ratio; SARI = serotonin (5-HT2) antagonist and reuptake inhibitor (trazodone); SNRI = serotonin-noradrenaline reuptake inhibitor (venlafaxine);
SSRI = selective serotonin reuptake inhibitor; TCA = tricyclic and tetracyclic antidepressant.

Note: Odds ratios >1 indicate more study discontinuation in patients receiving the treatment given in the row heading.
a
Conventional meta-analysis of within-study comparisons with pooled odds ratios.
b
Studies with direct comparisons available, with I2 value and P value from the χ2 test for heterogeneity.
C
From network meta-analysis.

in meta-analyses, however, because there were no significantly more dropouts resulting from adverse
dropouts attributable to side effects in any treatment effects compared with rMAO-As, the NRI was associ-
group. For 16 of the 36 possible comparisons (44%) ated with significantly more than SSRIs, and NaSSAs
there was at least 1 head-to-head trial. In 4 of the 13 were associated with significantly more than the low-
direct comparisons with 2 or more trials, there was dose SARI. In network meta-analysis combining direct
an indication of statistical heterogeneity (Table 4). and indirect evidence (Table 4), TCAs, SSRIs, SNRI,
Because of the often very wide confidence intervals NRI, and NaSSAs were associated with significantly
(caused by the small number of trials and low event more study discontinuations resulting from adverse
rates), findings of conventional meta-analyses of head- effects than was placebo. Attrition was not signifi-
to-head comparisons have to be interpreted with cantly different from placebo with (mostly low-dose)
great caution (Supplemental Appendix, Section 3 for SARI, rMAO-As, and hypericum extracts. rMAO-As
individual study findings). TCAs were associated with and hypericum extracts were associated with signifi-

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

74
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

mary care. Yet, although we compared

groups of similar interventions instead
of single specific interventions, the num-
NaSSA rMAO-A Hypericum Placebo ber of trials in some substance groups
was low. TCAs, SSRIs, and hypericum
3.83 (1.67-8.75) 1.65 (1.16-2.34) 0.87 (0.54-1.41) 1.67 (1.24-2.25) extracts were investigated more often
1 3; 0; .79; 2; 24; 0.25; 8; 25; .23; in the primary care setting than other
1.55 (1.11-2.18) 1.65 (1.10-2.45) 0.86 (0.67-1.10) 1.75 (1.42-2.15) drug classes. Our primary efficacy
analyses suggest that TCAs, SSRIs,
1.05 (0.54-2.03) na 0.88 (0.63-1.23) 1.57 (1.31-1.89)
SNRI, low-dose SARI, and hypericum
3; 60; .08 na 6; 46; .10 7; 0; .83
1.50 (1.06-2.11) 1.60 (1.03-2.54) 0.83 (0.66-1.04) 1.69 (1.40-2.04)
extracts are effective for the treatment
of acute depression, but effects when
na na na 1.94 (0.96-3.93) compared with placebo were modest
na na na 1 in size. With 40% of patients respond-
1.55 (1.02-2.45) 1.65 (0.98-2,80) 0.86 (0.59-1.24) 1.75 (1.24-2.47) ing to placebo, an odds ratio of 1.69
(as found for SSRIs) would mean that
na na na na 53% of patients receiving an antidepres-
na na na na sant respond. The absolute difference
0.94 (0.58-1.54) 1.34 (0.66-2.70) 0.52 (0.32-0.84) 1.42 (0.81-2.49) of 13% more than placebo corresponds
to a number needed to treat between
2.11 (1.37-3.25) na na na
7 and 8 patients. In secondary analyses
4; 0; .56 na na na
using remission as the outcome, NRI,
1.60 (1.07-2.42) 1.70 (0.96-2.89) 0.89 (0.58-1.38) 1.80 (1.17-2.76)
NaSSAs, and rMAO-As were also found
1.07 (0.49-2.34) na 1.34 (0.85-2.10) to be effective. rMAO-As, hypericum
1 na 2; 0; .98 extracts, and low-dose SARI tended to
1.06 (0.65-1.72) 0.55 (0.37-0.82) 1.13 (0.78-1.62) be associated with more favorable results
in the acceptability analyses. The qual-
0.93 (0.43-2.03) na na ity of most of the trials included in our
1 na na review was mediocre or weak. Because
0.94 (0.58-1.54) 0.52 (0.32-0.84) 1.06 (0.67-1.66) there was a small number of studies with
observation periods of longer than 12
na na 2.02 (1.41-2.88)
weeks, reliable comparative analysis of
na na 9; 43; .08
long-term effects was not possible.
1.80 (1.21. 2.67) 1.92 (1.20-3.16) 2.03 (1.63-2.53)
Conventional meta-analyses or ran-
domized trials restricted to primary care
patients have been performed for SSRIs
and TCAs compared with placebo,11,12
SSRIs compared with TCAs,13 and a
variety of newer antidepressants com-
pared with different comparators.14 The
reviews concluded that the reviewed
antidepressants are superior to placebo in
cantly fewer dropouts because of adverse effects com- primary care patients.13,14 The evidence on the relative
pared with TCAs, SSRIs, SNRI, NRI, and NaSSAs. efficacy of antidepressants was sparse and considered
(For results on dropouts for any reasons and patients to be of variable quality but suggested no major differ-
with adverse effects in comparison with placebo, see ences. Regarding acceptability, findings favored SSRIs
the Supplemental Appendix, Section 7). and most newer antidepressants when compared with
TCAs.13,14 Our review includes more primary care-
based trials than the mentioned previous reviews taken
DISCUSSION together. In addition to conventional pairwise meta-
This systematic review shows that a considerable analysis, we performed network meta-analysis, making
number of randomized trials have investigated the efficient use of all available data. Our analyses confirm
short-term (up to 12 weeks) efficacy and acceptability that short-term effects of SSRIs and TCAs are similar
of pharmacological treatments for depression in pri- and that fewer patients receiving SSRIs report adverse

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

75
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

Table 4. Results of Conventional and Network Meta-Analyses for the Main Acceptability Outcome Study
Discontinuation Because of Adverse Effects

Low-Dose
Medication TCA SSRI SNRI NRI SARI
TCA
OR (95% CI)a 1.15 (0.85-1.55) 0.86 (0.46-1.62) na 1.18 (0.46-3.01)
No. of studies; I2, %; P valueb 14; 43; .05 2; 0; .80 na 2; 54; .14
OR (95% CI)c 1.12 (0.84-1.47) 0.89 (0.55-1.37) 0.44 (0.18-1.06) 1.86 (0.92-3.75)
SSRI
OR (95% CI)a 0.87 (0.64-1.17) 0.81 (0.58-1.13) 0.39 (0.28-0.55) na
No. of studies; I2, %; P valueb 14; 43; .05 4; 38; .18 1 na
OR (95% CI)c 0.89 (0.68-1.19) 0.79 (0.55-1.19) 0.39 (0.16-0.90) 1.65 (0.85-3.45)
SNRI
OR (95% CI)a 1.16 (0.62-2.20) 1.24 (0.89-1.74) na na
No. of studies; I2, %; P valueb 2; 0; .80 4; 38; .18 na na
OR (95% CI)c 1.12 (0.73-1.81) 1.26 (0.84-1.83) 0.50 (0.19-1.23) 2.08 (0.95-4.85)
NRI
OR (95% CI)a 2.53 (1.81-3.53) na na
No. of studies; I2, %; P valueb 1 na na
OR (95% CI)c 2.25 (0.95-5.48) 2.54 (1.11-6.37) 2.01 (0.82-5.20) 4.19 (1.59-12.48)
SARI
OR (95% CI)a 0.84 (0.33-2.15) na na na
No. of studies; I2, %; P valueb 2; 54; .14 na na na
OR (95% CI)c 0.54 (0.26-1.08) 0.61 (0.29-1.17) 0.48 (0.21-1.05) 0.24 (0.08-0.63)
NaSSA
OR (95% CI)a 1.97 (0.71-5.51) 1.28 (0.69-2.36) na na 3.55 (1.27-9.90)
No. of studies; I2, %; P valueb 1 3; 31; .23 na na 3; 58; .09
OR (95% CI)c 1.56 (0.88-2.76) 1.75 (1.04-3.02) 1.39 (0.73, 2.76) 0.69 (0.27-1.71) 2.89 (1.56-6.18)
rMAO-A
OR (95% CI)a 0.39 (0.18-0.85) na na na na
No. of studies; I2, %; P valueb 5; 47; .11 na na na na
OR (95% CI)c 0.38 (0.20-0.67) 0.43 (0.21-0.81) 0.34 (0.15-0.71) 0.17 (0.06-0.48) 0.71 (0.28-1.73)
Hypericum
OR (95% CI)a 0.43 (0.17-1.09) 0.68 (0.35-1.30) na na na
No. of studies; I2, %; P valueb 2; 0; 0.89 6; 0; .92 na na na
OR (95% CI)c 0.42 (0.22-0.78) 0.47 (0.24-0.85) 0.38 (0.18-0.75) 0.19 (0.07-0.50) 0.78 (0.31-1.83)

CI = confidence interval; CI = credible interval; hypericum = extract from Hypericum perforatum L.; rMAO-A = reversible inhibitor of monoaminoxidase A (moclobemide,
minaprine); na = not available; NaSSA = noradrenergic and specific serotonergic antidepressive agent (mianserin, mirtazapine); NRI = noradrenaline reuptake inhibitor
(reboxetine); OR = odds ratio; SARI = serotonin (5-HT2) antagonist and reuptake inhibitor (trazodone); SNRI = serotonin-noradrenaline reuptake inhibitor (venlafaxine);
SSRI = selective serotonin reuptake inhibitor; TCA = tricyclic and tetracyclic antidepressant.

Note: Odds ratios >1 indicate more study discontinuation in patients receiving the treatment given in the row heading.
a
Conventional meta-analysis of within-study comparisons with pooled odds ratios.
b
Studies with direct comparisons available, with I2 value and P value from the χ2 test for heterogeneity.
C
From network meta-analysis.

effects, but we did not find any significant differences all settings.25 Hypericum extracts have a high risk of
regarding study discontinuation resulting from adverse adverse interactions with other drugs,26 however, and
effects or for any reasons among the investigated sub- patients with multiple morbid conditions are typically
stance classes. Our findings are in accordance with a excluded from randomized trials. Furthermore, the
conventional meta-analysis of trials mostly performed hypericum meta-analysis25 also found that evidence
in specialist mental health care showing that the of efficacy was considerably more solid in German-
NRI reboxetine has relevant adverse effects and little speaking regions than in other countries, including the
efficacy.24 In the network meta-analysis hypericum United States, which makes interpretation difficult.
extracts showed similar efficacy but better acceptabil- Because of the limited number of trials per single
ity than SSRIs and TCAs. This result is in line with a agents, our review cannot provide estimates of relative
conventional meta-analysis of hypericum trials from efficacy on this level. Network meta-analyses of single

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

76
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

guidelines make recommendations that

are specific to the primary care set-
ting. Generally, for patients with whom
pharmacotherapy is considered, the UK
NaSSA rMAO-A Hypericum Placebo guideline recommends generic SSRIs as
the first choice; the Canadian guideline
0.51 (0.18-1.42) 2.57 (1.18-5.60) 2.33 (.92-5.93) 2.30 (1.10-4.81) recommends all second-generation anti-
1 5; 47; .11 2; 0; .89 7; 24; .25 depressants, and the US and the German
0.64 (0.36-1.14) 2.63 (1.49-5.09) 2.37 (1.28-4.49) 2.47 (1.59-3.84) guidelines are somewhat less explicit by
listing a number of situations in which
0.78 (0.42-1.44) na 1.48 (0.77-2.85) 1.86 (1.16-2.98)
other choices might be preferable. The
3; 31; .23 na 6; 0; .92 8; 0; .64
0.57 (0.33-0.96) 2.34 (1.24-4.84) 2.11 (1.18-4.14) 2.20 (1.44-3.36)
UK guideline explicitly states that anti-
depressants should not be used routinely
na na na 2.96 (0.90-9.73) for persistent subthreshold depressive
na na na 1 symptoms or mild depression. All guide-
0.72 (0.36-1.37) 2.95 (1.41-6.69) 2.66 (1.33-5.57) 2.77 (1.58-4.87) lines agree that there is evidence for the
effectiveness of hypericum extracts for
na na na na mild to moderate depression. But while
na na na na the German and the Canadian guide-
1.45 (0.58-3.74) 5.94 (2.09-17.13) 5.35 (1.99-14.94) 5.58 (2.28-13.66)
lines are supportive of them when the
risk of interactions is taken into account
0.28 (0.10-0.79) na na na
adequately, the UK guideline explicitly
3; 58; 0.09 na na na
0.35 (0.16-0.64) 1.42 (0.58-3.55) 0.78 (0.31-1.83) 1.33 (0.61-2.89)
discourages their use. The US guideline
does not make a recommendation.
na na 3.52 (1.40-8.84) Despite guideline recommendations,
na na 2; 0; .90 evidence from conventional and network
4.10 (1.81-10.38) 3.69 (1.71-8.56) 3.85 (2.05-7.25) meta-analyses, and widespread use, there
is an ongoing discussion of the extent
na na na to which antidepressants have clinically
na na na relevant effects compared with placebo.
0.24 (0.10-0.55) 0.90 (0.35-2.04) 0.94 (0.43-2.05)
In the meta-analyses the effects size
compared with placebo is frequently
na na 0.79 (0.31-1.97)
considered rather small. Yet, findings
na na 5; 0; .48
0.27 (0.11-0.58) 1.11 (0.49-2.83) 1.04 (0.53-2.06)
from published trials tend to overesti-
mate these effects because of publica-
tion and reporting bias. A meta-analysis
of 74 trials of 12 second-generation
antidepressants registered with the US
Food and Drug Administration (FDA)
indicated statistically significant effects
when compared with placebo for all
agents, but meta-analysis of published
trials only would have inflated the effects
second-generation antidepressants have been performed size by 32% on average.34 For our review, we searched
across settings (mainly including trials from specialist several trial registries of manufacturers, but because of
mental health care). One review concluded that sertra- the large number of agents covered in our review, we
line might be the best choice when starting treatment were unable to comprehensively search for unpublished
for moderate and severe depression,27 one favored esci- trials. Owing to our limited resources and that many
talopram,28 and another considered the evidence insuf- primary care trials are performed outside the approval
ficient to recommend a particular agent.29,30 process, we did not search the FDA database. We can-
We compared our findings with recommenda- not rule out that our findings are distorted by publica-
tions of current high-quality clinical practice guide- tion or reporting bias to some extent.
lines from the United States,8 the United Kingdom,31 Another widely cited meta-analysis of FDA-
Canada,32,33 and Germany.19 Currently, none of these registered trials found that compared with placebo,

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

77
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

medication effects became clinically relevant only cal treatments under conditions of routine care and
in patients with very severe depression.9 This find- stepped-care strategies.
ing would be particularly important for primary care, To read or post commentaries in response to this article, see it
as family physicians see many patients with less than online at http://www.annfammed.org/content/13/1/69.
severe depression. The meta-analysis included very
few trials in patients with mild to moderate depression, Key words: depression, depressive disorders, antidepressive agents,
primary health care, review, systematic
however, and used aggregate data to investigate the
association between baseline severity and outcome. A Submitted January 30, 2014; submitted, revised, May 16, 2014;
recent reanalysis of the same data set using different accepted June 13, 2014.
meta-analytic methods did not confirm the influence
of initial severity on efficacy.35 Also, the findings of Funding support: The study was funded by the German Ministry of
2 newer meta-analyses of individual patient data—a Education and Research (grant 01KG1012).

more appropriate method—are contradictory.36,37 In

Disclaimer: The sponsor was not involved in the design, data collection,
our analyses, there were no major differences in tri- analysis, interpretation, writing of the report or decision to submit the
als limited to patients with major depression and trials report for publication.
that included other depressive patients. We could not
investigate the association between baseline severity Acknowledgment: The contribution of Susanne Jamil to this work is
part of her ongoing doctor of medicine thesis at the Medical Faculty of
and outcome in detail because of the multiple different
the Technische Unviversität München, Munich, Germany.
depression scales used in the primary studies.
Overall, despite publication bias and problems with Systematic review registration: 01KG1012 at http://www.gesundheits
methodological quality, findings seem to agree that forschung-bmbf.de/de/2852.php.
antidepressants are significantly more effective than
placebo. Most critical discussions focus on the clini-  upplementary materials: Available at http://www.AnnFamMed.
S
org/content/13/1/69/suppl/DC1/.
cal relevance of effects on the basis of meta-analytical
effect sizes. Still, it should be noted that when com-
paring antidepressive medications with placebo alone, REFERENCES
effect sizes cannot provide sufficient information on 1. Demyttenaere K, Bruffaerts R, Posada-Villa J, et al; WHO World Men-
the clinical relevance of treatment effects in individual tal Health Survey Consortium. Prevalence, severity, and unmet need
clinical contexts, so this issue is likely to remain con- for treatment of mental disorders in the World Health Organization
World Mental Health Surveys. JAMA. 2004;291(21):2581-2590.
troversial.9,38 The results of our analyses indicate that
antidepressants have higher short-term effects when 2. Bijl RV, Ravelli A. Psychiatric morbidity, service use, and need for care
in the general population: results of The Netherlands Mental Health
compared with placebo also in primary care. Survey and Incidence Study. Am J Public Health. 2000;90(4):602-607.
For decision making in individual patients, fam- 3. Peveler R, Kendrick T. Treatment delivery and guidelines in primary
ily physicians should be aware that SSRIs and TCAs care. Br Med Bull. 2001;57:193-206.
have a somewhat more solid evidence base than other 4. Schwenk TL, Coyne JC, Fechner-Bates S. Differences between
substance classes (with SSRIs having a slightly bet- detected and undetected patients in primary care and depressed
psychiatric patients. Gen Hosp Psychiatry. 1996;18(6):407-415.
ter acceptability profile). Further agents (hypericum,
rMAO-A, SNRI, NRI, NaSSA, SARI) showed some 5. Simon GE, VonKorff M. Recognition, management, and outcomes
of depression in primary care. Arch Fam Med. 1995;4(2):99-105.
positive results, but limitations of the currently avail-
6. Williamson PS, Yates WR. The initial presentation of depression in
able evidence make difficult a clear recommenda- family practice and psychiatric outpatients. Gen Hosp Psychiatry.
tion on their place in clinical practice. Differences 1989;11(3):188-193, discussion 216-221.
between substance classes and between single second- 7. Kroenke K, Price RK. Symptoms in the community. Prevalence,
generation antidepressants seem to be relatively minor. classification, and psychiatric comorbidity. Arch Intern Med. 1993;
153(21):2474-2480.
The latter findings come from network meta-analyses
of trials mostly performed in specialized mental health 8. Trangle M, Dieperink B, Gabert T, et al. Major depression in adults
in primary care. Bloomington, MN: Institute for Clinical Systems
care, but our results suggest that results from primary Improvement; 2012.
care trials and other trials are broadly similar. 9. Kirsch I, Deacon BJ, Huedo-Medina TB, Scoboria A, Moore TJ, John-
It must be emphasized that there are very few data son BT. Initial severity and antidepressant benefits: a meta-analysis
from primary care trials regarding long-term effective- of data submitted to the Food and Drug Administration. PLoS Med.
2008;5(2):e45.
ness and acceptability. Future research should priori-
10. Adli M, Hegerl U. Do we underestimate the benefits of antidepres-
tize large, long-term, pragmatic trials and observational sants? Lancet. 2014;383(9926):1361-1362.
studies addressing clinically relevant questions, such as
11. Arroll B, Elley CR, Fishman T, et al. Antidepressants versus placebo
the best management of mild-to-moderate depression for depression in primary care. Cochrane Database Syst Rev. 2009;
and comparison of pharmacological and psychologi- (3):CD007954.

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

78
 P H A R M ACO L O G I C A L T R E AT M EN T F O R D EP R E S S I V E D IS O R D ER S

12. Arroll B, Macgillivray S, Ogston S, et al. Efficacy and tolerability of 26. Mannel M. Drug interactions with St John’s wort: mechanisms and
tricyclic antidepressants and SSRIs compared with placebo for treat- clinical implications. Drug Saf. 2004;27(11):773-797.
ment of depression in primary care: a meta-analysis. Ann Fam Med.
27. Cipriani A, Furukawa TA, Salanti G, et al. Comparative efficacy and
2005;3(5):449-456.
acceptability of 12 new-generation antidepressants: a multiple-
13. MacGillivray S, Arroll B, Hatcher S, et al. Efficacy and tolerability of treatments meta-analysis. Lancet. 2009;373(9665):746-758.
selective serotonin reuptake inhibitors compared with tricyclic anti-
28. Ramsberg J, Asseburg C, Henriksson M. Effectiveness and cost-
depressants in depression treated in primary care: systematic review
effectiveness of antidepressants in primary care: a multiple treat-
and meta-analysis. BMJ. 2003;326(7397):1014.
ment comparison meta-analysis and cost-effectiveness model. PLoS
14. Mulrow CD, Williams JW Jr, Chiquette E, et al. Efficacy of newer One. 2012;7(8):e42003.
medications for treating depression in primary care patients. Am
29. Gartlehner G, Gaynes BN, Hansen RA, et al. Comparative benefits
J Med. 2000;108(1):54-64.
and harms of second-generation antidepressants: background
15. Caldwell DM, Ades AE, Higgins JP. Simultaneous comparison of paper for the American College of Physicians. Ann Intern Med.
multiple treatments: combining direct and indirect evidence. BMJ. 2008;149(10):734-750.
2005;331(7521):897-900.
30. Gartlehner G, Hansen RA, Morgan LC, et al. Comparative benefits
16. Linde K, Schumann I, Meissner K, et al. Treatment of depressive dis- and harms of second-generation antidepressants for treating major
orders in primary care—protocol of a multiple treatment systematic depressive disorder: an updated meta-analysis. Ann Intern Med.
review of randomized controlled trials. BMC Fam Pract. 2011;12:127. 2011;155(11):772-785.
17. Linde K, Sigterman K, Kriston L, et al. Effectiveness of psychologi- 31. National Collaborating Centre for Mental Health. Depression: the
cal treatment for depressive disorders in primary care: systematic treatment and management of depression in adults. http://www.
review and meta-analysis. Ann Fam Med. 2014;12(6):56-68. nice.org.uk/nicemedia/live/12329/45896/45896.pdf. Updated 2010.
18. Higgins JPT, Green S. Cochrane Handbook for Systematic Reviews of 32. Lam RW, Kennedy SH, Grigoriadis S, et al; Canadian Network for
Interventions Version 5.1.0. Updated March 2011. London, England: Mood and Anxiety Treatments. Canadian Network for Mood and
The Cochrane Collaboration; 2008. http://handbook.cochrane.org. Anxiety Treatments (CANMAT) clinical guidelines for the manage-
ment of major depressive disorder in adults. III. Pharmacotherapy.
19. DGPPN, BÄK, KBV, AWMF, AkdÄ, BPtK, BApK, DAGSHG, DEGAM,
J Affect Disord. 2009;117(Suppl 1):S26-S43.
DGPM, DGPs, DGRW: S3-Leitlinie/Nationale Versorgungsleitlinie Uni-
polare Depression - Langfassung. Berlin, Düsseldorf: DGPPN, ÄZQ, 33. Ravindran AV, Lam RW, Filteau MJ, et al; Canadian Network for
AWMF; 2009. http://www.leitlinien.de/nvl/depression/. Mood and Anxiety Treatments. Canadian Network for Mood
and Anxiety Treatments (CANMAT) Clinical guidelines for the
20. Furukawa TA, Cipriani A, Barbui C, Brambilla P, Watanabe N.
management of major depressive disorder in adults. V. Comple-
Imputing response rates from means and standard deviations in
mentary and alternative medicine treatments. J Affect Disord.
meta-analyses. Int Clin Psychopharmacol. 2005;20(1):49-52.
2009;117(Suppl 1):S54-S64.
21. Dias S, Sutton AJ, Ades AE, Welton NJ. Evidence synthesis for
34. Turner EH, Matthews AM, Linardatos E, Tell RA, Rosenthal R. Selec-
decision making 2: a generalized linear modeling framework for
tive publication of antidepressant trials and its influence on appar-
pairwise and network meta-analysis of randomized controlled trials.
ent efficacy. N Engl J Med. 2008;358(3):252-260.
Med Decis Making. 2013;33(5):607-617.
35. Fountoulakis KN, Veroniki AA, Siamouli M, Möller HJ. No role for
22. Dias S, Welton NJ, Sutton AJ, Caldwell DM, Lu G, Ades AE. Evi-
initial severity on the efficacy of antidepressants: results of a multi-
dence synthesis for decision making 4: inconsistency in networks of
meta-analysis. Ann Gen Psychiatry. 2013;12(1):26.
evidence based on randomized controlled trials. Med Decis Making.
2013;33(5):641-656. 36. Fournier JC, DeRubeis RJ, Hollon SD, et al. Antidepressant drug
effects and depression severity: a patient-level meta-analysis. JAMA.
23. Dias S, Sutton AJ, Welton NJ, Ades AE. Evidence synthesis for deci-
2010;303(1):47-53.
sion making 3: heterogeneity—subgroups, meta-regression, bias,
and bias-adjustment. Med Decis Making. 2013;33(5):618-640. 37. Gibbons RD, Hur K, Brown CH, Davis JM, Mann JJ. Benefits from
antidepressants: synthesis of 6-week patient-level outcomes from
24. Eyding D, Lelgemann M, Grouven U, et al. Reboxetine for acute
double-blind placebo-controlled randomized trials of fluoxetine and
treatment of major depression: systematic review and meta-analysis
venlafaxine. Arch Gen Psychiatry. 2012;69(6):572-579.
of published and unpublished placebo and selective serotonin reup-
take inhibitor controlled trials. BMJ. 2010;341:c4737. 38. Cuijpers P, Turner EH, Koole SL, van Dijke A, Smit F. What is the
threshold for a clinically relevant effect? The case of major depres-
25. Linde K, Berner MM, Kriston L. St John’s wort for major depression.
sive disorders. Depress Anxiety. 2014;31(5):374-378.
Cochrane Database Syst Rev. 2008;(4):CD000448.

ANNALS O F FAMILY MEDICINE ✦ WWW.ANNFA MME D.O R G ✦ VO L. 13, N O. 1 ✦ JA N UA RY/FE BRUA RY 2015

79