Anda di halaman 1dari 13

CNS Drugs (2016) 30:2739

DOI 10.1007/s40263-015-0308-1

SYSTEMATIC REVIEW

A Systematic Review and Network Meta-Analysis to Assess


the Relative Efficacy of Antipsychotics for the Treatment
of Positive and Negative Symptoms in Early-Onset Schizophrenia
Rebecca C. Harvey1 Anthony C. James2 Gemma E. Shields1

Published online: 22 January 2016


Springer International Publishing Switzerland 2016

Abstract
Introduction Early-onset schizophrenia (EOS) is a serious debilitating disorder with considerable morbidity and a
reduced life expectancy; therefore, early diagnosis and
effective treatments are particularly important. Negative
symptoms are more prominent in adolescents and children
(compared with adults), and are key predictors of worse
functional and clinical outcomes in EOS. Therefore, this
study aimed to explore the relative efficacy of antipsychotics used in the treatment of EOS, with a focus on
studies reporting effectiveness using the Positive and
Negative Syndrome scale (PANSS), a scale that includes
an overall symptom measure, in addition to separate subscales for positive and, importantly, negative symptoms.

Electronic supplementary material The online version of this


article (doi:10.1007/s40263-015-0308-1) contains supplementary
material, which is available to authorized users.
& Rebecca C. Harvey
bharvey@bresmed.co.uk
1

BresMed Health Solutions, North Church Business House,


84 Queen Street, Sheffield S1 2DW, UK

Highfield Unit Oxford, Warneford Hospital, Oxford, UK

Methods A systematic literature review was conducted


using the MEDLINE and Cochrane Central Register of
Controlled Trials databases to identify trials conducted in
children and adolescents with schizophrenia, and symptom
control was reported using the PANSS. A Bayesian random-effects network meta-analysis was performed, synthesising data for a number of outcomes, including mean
change from baseline in PANSS scores, treatment discontinuation and weight gain.
Results Eleven studies were included in the evidence
synthesis, comprising 1714 patients across eight active
interventions (aripiprazole, haloperidol, molindone, olanzapine, paliperidone, quetiapine, risperidone and ziprasidone) and placebo. All treatments showed a greater
reduction in total PANSS scores vs placebo; however, only
three interventions (molindone, olanzapine and risperidone) were associated with a statistically significant
reduction in total PANSS scores at 6 weeks vs placebo.
Haloperidol had the greatest reduction vs placebo; however, this result was not statistically significant [mean
difference, -15.6, 95 % credible interval (-35.4, 4.1)].
Haloperidol, olanzapine and risperidone showed a statistically significant reduction in positive PANSS scores vs
placebo; however, whilst all interventions showed a trend
of reduction in negative PANSS scores vs placebo, no
comparisons were statistically significant.
Conclusions Many of the treatments are efficacious in
controlling symptoms, and all showed a trend of superiority
vs placebo for total, positive and negative PANSS scores,
although only olanzapine and risperidone yielded statistically significant results vs placebo for both total and positive PANSS scores. Varying results for discontinuation
and weight gain demonstrate a need to balance efficacy
with side-effect profiles.

28

Key Points
The analysis demonstrates positive outcomes in
terms of symptom control, using the Positive and
Negative Syndrome Scale outcome scale, although
there were no statistically significant results between
treatments for managing negative symptoms.
Rates for discontinuation and weight gain highlight a
need to balance treatment efficacy with side-effect
profiles.
Evidence for the comparative efficacy and safety of
antipsychotic use in the paediatric population is
limited; large randomised controlled trials are
needed to verify the results of indirect treatment
comparisons.

1 Introduction
Early-onset schizophrenia (age of onset before 18 years)
(EOS) is often a severe and debilitating psychotic disorder
with considerable impairments in psychological, social,
educational and occupational functioning and a reduced
life expectancy. Schizophrenia is estimated to affect
1.61.9 per 100,000 in the child population, with a steep
rise during adolescence [1, 2]. EOS poses a heavy burden
on the patient, family and health services, accounting for
24.5 % of all adolescent psychiatric admissions (with an
overall admission rate of 0.46 per 1000 for this age range in
England and Wales) [3].
EOS appears to be a more severe disorder than adult-onset
schizophrenia [4]. In particular, it has been linked to a higher
burden of negative symptoms, in addition to reduced cognitive functioning, lower educational attainment and poorer
response to treatment [5]. Furthermore, treatment can be
problematic in children and adolescents, although a recent
longitudinal study from Melbourne, with an early-intervention service, found advantageous outcomes with early
treatment [6]. Psychological interventions have a positive
impact if delivered before the onset of psychosis in individuals with attenuated or transient psychotic symptoms, and
are recommended as a first line of treatment for psychosis in
this age group [7, 8]. Antipsychotic (AP) medication is the
mainstay of treatment of psychosis; however, there is limited
evidence of efficacy for APs in this age group [8, 9]. Systematic reviews indicate efficacy, but with small effect sizes.
For second-generation antipsychotics (SGAs), there are
considerable problems with weight gain, metabolic and
cardiac effects, and a risk of diabetes mellitus [10]. Overall,
the risk of side effects associated with AP treatment for

R. C. Harvey et al.

schizophrenia in this age group is greater than that in adults


[1012]. This has led to concerns about a potential public
health crisis, with the increased prescription of APs resulting
in increased risks of cardiovascular mortality following
treatment [13, 14].
Hence, there is a pressing need to systematically evaluate treatments and their side effects. A previous study by
Leucht et al. compared the efficacy and tolerability of AP
therapies in the adult schizophrenia population [15]. That
study found that APs differed substantially in side effects,
with only small but robust differences in efficacy. However, conclusions reached in the adult population may not
necessarily apply to the adolescent and child population,
who are undergoing major physical growth, including
crucial brain and cognitive changes.
The objective of this review was to explore the relative
efficacy of AP therapies for EOS. While looking at
symptom improvement overall is helpful, there is little
correlation between positive and negative symptoms [16].
Therefore, when studies use a symptom measure that
summarises both aspects, it means that we cannot infer the
impact of intervention on each symptom group. The
validity of distinguishing between positive and negative
symptoms has been explored and confirmed as a useful tool
for monitoring disease severity and predicting outcomes
[17]. In addition, the literature recognises that negative
symptoms are more prominent in adolescents and children
compared with adults, and are key predictors of worse
functional and clinical outcomes in EOS [1821]. Therefore, we focused on studies reporting effectiveness using
the Positive and Negative Syndrome Scale (PANSS), a
scale that includes an overall symptom measure in addition
to separate subscales for positive and negative symptoms.
A systematic literature review (SLR) was conducted to
identify trials conducted in children and adolescents with
schizophrenia that report symptom control and effectiveness
using the PANSS. Following this, a network meta-analysis
(NMA) synthesising data extracted from the final list of
relevant studies was conducted to broaden the evidence base
by providing a different method of comparing pharmacological interventions. The analysis aimed to assist clinicians
in making decisions on therapeutic approaches to EOS (in
particular, which treatments may be efficacious for patients
with predominantly positive or negative symptoms), and the
results highlight the limitations in the current evidence base.

2 Materials and Methods


2.1 Literature Search
An SLR was conducted in January 2015 using two global
electronic databases (MEDLINE and the Cochrane Central

Relative Efficacy of Antipsychotics for the Treatment of Early-Onset Schizophrenia

29

Register of Controlled Trials), to identify evidence of the


effectiveness of interventions for EOS. Searches were
limited to English language studies, with no restriction
placed on date. Results were limited to clinical trials, with
other types of publication (e.g. reviews and letters) excluded from initial searches.
Searches identified trials conducted in the EOS population. Common search terms included disease-specific
terms, e.g. schizophrenia; trial terms, e.g. randomisation; and age-group terms, e.g. adolescent. Both freetext and expanded medical subject headings were used to
construct search terms. Strategies varied between the
databases as a result of design and search capabilities.
Spelling variants, synonyms and abbreviations were
included to capture all potentially relevant citations. A full
copy of the search strategy is presented in the electronic
supplementary material. Reference lists of key papers were
checked for articles that the initial search had failed to
identify.

2.3 Data Extraction and Quality Assessment

2.2 Inclusion Criteria

Because of the availability of multiple interventions used to


treat EOS, it is of interest to determine which treatment is
the most efficacious. An NMA enables relative efficacy to
be estimated between any comparators identified in the
SLR, despite the absence of head-to-head trials. Indirect
comparisons may be useful to compare competing interventions that have not been compared in a head-to-head
RCT. NMA is an extension of pairwise meta-analysis,
relying upon the use of both direct and indirect evidence.
This combination of evidence produces more precise estimates of relative effectiveness than considering only direct
trial data (where such data exist). Not all studies are
required to be placebo controlled for inclusion within an
NMA. Moreover, synthesis of data from studies comparing
active treatments may provide feedback loops, which
assist with checking consistency between direct and indirect evidence.
A Bayesian NMA, using Markov chain Monte Carlo
methods was performed to synthesise the evidence base
identified from the SLR. A random-effects model was fitted
to the data, which was selected because schizophrenia, by
its very nature, is expected to occur in a very heterogeneous population. The statistical between-study heterogeneity that is likely to be present within this population
will be accounted for by the use of the random-effects
model.
Analyses were performed using statistical software,
WinBUGS (version 1.4.3) and R (version 3.1.1). The
models fitted, measuring arm-specific change from baseline
effects, were proposed by Dias et al. in the National
Institute for Health and Care Excellence (NICE) Decision
Support Unit (DSU) Technical Support Document (TSD) 2
[23]. Vague prior distributions were assigned to parameters

Following database searches, primary screening was conducted by reviewers with titles and abstracts of all identified citations against a set of explicit eligibility criteria. For
citations that appeared to meet the inclusion criteria based
on abstract screening, full articles were obtained and
assessed against the full eligibility criteria. Two reviewers
independently conducted primary (title and abstract) and
secondary (full-text) screening before comparing results.
Any disagreements were discussed and settled with the
assistance of a third reviewer.
Eligibility studies (a) focused on patients aged 18
years and below with a diagnosis of schizophrenia,
schizoaffective disorder or schizophreniform disorder;
(b) conducted two or more arm trials in the disease area;
(c) reported mean (and standard deviation or error)
baseline and changes over time in PANSS symptom
scores (total, positive or negative), where reductions in
PANSS scores are indicative of symptom improvements;
and (d) reported data between 6- and 12-week endpoints
for comparisons to be justified. Where there was evidence that the studies reported results from the same
primary study, the paper reporting these data within the
most relevant time scale (i.e. closest to a 6-week endpoint for primary outcome data) was chosen, or if time
points were equal, the most recently published paper was
chosen, to avoid double counting the data.
Because of the paucity of randomised controlled trials
(RCTs) in this population, non-RCT designs were included
to allow for a broader treatment comparison. A sensitivity
analysis excluding non-randomised studies was planned at
the protocol stage.

Specific data were extracted using a pre-defined and


pilot-tested data extraction form, including quantitative
results. Standard errors/deviations were extracted where
available. Information on study design and patient characteristics were also extracted to assess heterogeneity in
population characteristics and study designs. Studies
were reviewed for quality and bias risk using the Critical
Appraisal Skills Programme critical appraisal RCTs
checklist [22].
The data extraction process was completed independently by two reviewers. Extracted data were then crosschecked, and any disagreements were discussed and
resolved.
2.4 Data Analysis
2.4.1 Synthesis of Evidence

30

of interest and were in line with those specified in the NICE


DSU TSD 2 [23]. Analyses were based on 10,000 samples
after a burn-in of 50,000 was discarded. This burn-in was
selected after assessing the BrooksGelmanRubin statistic
for two chains with varying initial values [23]. Convergence was observed after approximately 50,000 iterations.
A thinning interval of 100 was chosen to reduce the
autocorrelation between samples and to ensure that the
chain was mixing well and was representative of the posterior distribution. Overall goodness of fit of the model was
assessed by examining the total residual deviance. The
total residual deviance of a well-fitting model should be
similar to the number of unique data points included within
the analysis [23]. By adopting a Bayesian approach,
treatment ranking probabilities may be estimated. These
show the probability of each treatment attaining each
possible rank within the network (i.e. first, second, third
and so on) [24].
2.4.2 Outcome Measures
The primary efficacy measure was mean change from
baseline to 6 weeks in total PANSS scores (a symptom
severity scale) [25]. Secondary outcomes were mean
change from baseline to 6 weeks in positive and negative
subscale PANSS scores, weight, odds of all-cause treatment discontinuation and odds of discontinuation because
of adverse events. Studies reporting results at any time
between 8 and 12 weeks post-baseline were used as proxy
where 6-week data were not reported; an approach taken by
Leucht et al., who considered it to be clinically plausible
[15].
2.4.3 Assumptions Regarding the Data
Where a study did not report a measure of variation
around the change in PANSS score, the following
assumptions were made to retain the study in the analysis.
If the study reported a variance of the baseline (Vb) and
endpoint (Ve) PANSS scores, the variance of the change
(VD) in PANSS scores was calculated using the following
equation:
p
V D V b V e  2q V b V e ;
where the value of q (within-patient correlation) is rarely
reported, and a conservative correlation of 0.5 was used
to reflect the lack of independence between baseline and
endpoint scores. This is detailed further in the NICE
DSU TSD 2 [23]. For studies that did not a report a
variance of either the change from baseline or the endpoint, the average (mean) value from all other studies
was imputed.

R. C. Harvey et al.

Different doses of the same intervention were not distinguished in the analysis. For trials that consider the same
intervention at different doses, data were pooled, weighting
by the sample size in each treatment arm to estimate an
overall effect for this treatment within a study.
Sensitivity analyses were performed to exclude particular studies from the NMA based on any anomalies identified in the trial. This included the removal of the one nonrandomised trial that was identified to examine the impact
on results.
2.4.4 Inconsistency of Evidence
An underlying assumption of NMAs is that direct and
indirect sources of evidence are estimating the same
treatment effects. When this assumption is violated,
inconsistency may be present, which is a discrepancy
between direct and indirect evidence. Inconsistency can
only be checked where both direct and indirect data are
present for any pairs of treatments, i.e. closed loops exist
within the network (not arising from three-arm trials).
Potential inconsistency was evaluated using a node-splitting approach, which is only possible within a Bayesian
framework [26].
2.4.5 Presentation of Results
Results are presented in the form of forest plots showing
relative efficacy of all interventions vs placebo. Additionally, all pairwise treatment comparisons are presented
in tabular format within the supplementary material to
show the relative efficacy between all interventions
included in the network. Point estimates [mean difference (MD) for continuous outcomes, odds ratios (OR)
for binary outcomes] and 95 % credible intervals (CrI)
are presented. Because a reduction in PANSS scores
(total, positive and negative), lower discontinuation rates
and lower weight change are favourable, point estimates
lying to the right of the line of no difference favour the
intervention over placebo. When the 95 % CrI lies
exclusively to the left or right of this line of no difference, this indicates a statistically significant result vs
placebo. Treatment ranking probabilities are presented in
the form of rankograms, which show the probability of
attaining each possible rank. The vertical axis represents
the probability, and the horizontal axis shows the possible ranks in ascending order (equal to the number of
treatments in the network). Lines showing a decreasing
trend indicate more efficacious treatments as they indicate a reducing probability of being ranked lower. All
outcomes are displayed on one figure and are identified
by the line type.

Relative Efficacy of Antipsychotics for the Treatment of Early-Onset Schizophrenia

3 Results
3.1 Search Results
In total, 1613 potentially relevant citations were identified
through database searches. Primary title and abstract
screening resulted in 1174 citations being excluded on the
basis that they clearly did not meet the eligibility criteria of
the review. Of the resulting 439 citations accessed in full,
11 were publications of studies meeting the inclusion criteria (Fig. 1).
3.2 Study Characteristics and Quality Assessment
Key characteristics of the included studies are presented in
Table 1. Generally, the studies were quite small, although
this varied, from 7 to 190 patients per treatment arm.
Baseline patient characteristics (such as age and proportion

31

of male individuals) varied across studies, demonstrating


heterogeneity in the study population.
All but one of the studies randomised patients to treatment arms [2837]. The risk of bias owing to a lack of
blinding affected two studies [27, 28], and may affect a
third in which the blinding method was unclear [34].
However, the majority of studies satisfactorily reported
blinding and methods, reducing the risk of bias [2933, 35
37]. There were signs of selective reporting as it was
unclear whether two trials reported all outcomes measured
[29, 32]. Potential bias was suggested by differences
between the baseline characteristics of groups in three
studies, which reported some minor differences between
groups [27, 30, 34]. Five of the studies did not have significant differences in the drop-out rates between arms [30,
31, 3537]; five studies demonstrated differences between
the drop-out rates between arms, but only two studies
reported and explained differences in rates between the

Identification

Fig. 1 Flow diagram of search

Records identified through database searching (n=1,715)


Medline (n=383)
Cochrane Central Register of Controlled Trials (n=1,332)

Eligibility

Screening

Records after duplicates removed (n=1,613)

Records screened (n=1,613)

Records excluded (n=1,174)

Full-text articles assessed for

Full-text articles excluded

eligibility (n=439)

(n=428)
Outcome (n=21)
Study design (n=26)

Included

Population (n=336)

Studies included in synthesis


(n=11)

Overlapping patient
population (n=8)
Duplicate (n=6)
Lack of detail (n=26)
Unable to access (n=5)

32

R. C. Harvey et al.

Table 1 Study characteristics


References

Study type

Gothelf et al. (2003) [27]

Controlled trial

Mozes et al. (2006) [28]

Findling et al. (2008) [29]

Sikich et al. (2008) [30]

Haas et al. (2009) [31]

Haas et al. (2009) [32]

Randomised controlled trial

Randomised, double-blind, placebocontrolled trial


Randomised, double-blind, controlled
trial

Randomised controlled trial

Randomised, double-blind, placebocontrolled trial

Time point
post-baseline
(week)
8

12

Treatment arm
(mg/day)

Age,
years
(mean)

Male
patients,
%

Baseline
total
PANSS

Haloperidol
(515)

17.1

71.4

86.1

Olanzapine
(1020)

17

16.8

68.4

71.6

Risperidone
(0.56)

15

17.0

52.9

90.2

Olanzapine
(2.520)

12

11.5

41.7

92.75

Risperidone
(0.254.5)

13

10.7

38.5

93.85

Aripiprazole (10)

99

15.6

45

93.6

Aripiprazole (30)

97

15.4

63.7

94.0

Placebo

98

15.4

61

94.6

Molindone
(10140)
Olanzapine
(2.520)

40

NR

58

99.7

35

NR

71

100.3

Risperidone
(0.56)

41

NR

66

103.3

Risperidone
(0.150.6)
Risperidone
(0.56)

131

NR

NR

96.4

124

NR

NR

93.3

Risperidone
(13)

54

15.7

30

95.4

Risperidone
(46)
Placebo

50

15.7

37

93

54

15.5

35

93.2

72

16.1

51

95.3

Kryzhanovskaya et al.
(2009) [33]

Randomised, double-blind, placebocontrolled trial

Olanzapine
(2.520)
Placebo

35

16.3

24

95.5

Swadi et al. (2010) [34]

Randomised controlled trial

Quetiapine
(100800)

11

NR

NR

89

Risperidone
(0.56)

11

NR

NR

87.09

Paliperidone
(1.5)

54

15.1

30

91.6

Paliperidone
(36)

48

15.3

31

90.6

Paliperidone
(612)

47

15.5

33

91.5

Singh et al. (2011) [35]

Findling et al. (2012) [36]

Randomised, double-blind, placebocontrolled trial

Randomised, double-blind, placebocontrolled trial

Randomised, double-blind, placebocontrolled trial

Placebo

51

15.7

23

90.6

Quetiapine (400)

73

15.45

43

96.2

Quetiapine (800)

74

15.45

44

97.0

Placebo
Findling et al. (2013) [37]

Ziprasidone
(40160)
Placebo

NR not reported, PANSS Positive and Negative Syndrome Scale

73

15.34

42

96.7

190

15.24

56.48

88.9

88

15.43

68.69

87.4

Relative Efficacy of Antipsychotics for the Treatment of Early-Onset Schizophrenia

33

Fig. 2 Network of evidence.


Solid lines represent two-arm
studies, dashed lines represent
three-arm studies; node size is
proportional to the number of
patients treated with
intervention, Asterisk study only
included in total Positive and
Negative Syndrome Scale
analysis; Double asterisk study
included in all-cause
discontinuation analysis

arms [28, 33], while in the remaining three studies, this was
unclear [27, 32, 34].
3.3 Data Analysis
3.3.1 NMA Evidence Base
Eleven studies were included in the primary analysis
(change from baseline total PANSS scores) as shown by
the network of evidence (Fig. 2). All studies except Swadi
(2010) were included in the analyses of change from
baseline positive PANSS scores, negative PANSS scores
and weight change (denoted * in Fig. 2) [34]. All-cause
discontinuation was less frequently reported; seven studies
were included in this analysis (denoted ** in Fig. 2). For
all analyses except all-cause discontinuation, eight active
treatments and placebo were included within the network
of evidence (Fig. 2).
3.3.2 Assessment of Efficacy
Relative treatment effects vs placebo for change from
baseline to 6 weeks in total PANSS scores are shown by the
forest plot (Fig. 3a). There was evidence to suggest that
three interventions (molindone, olanzapine and risperidone)
were associated with a statistically significant reduction in

total PANSS scores at 6 weeks vs placebo. The most


effective intervention was haloperidol, with an estimated
reduction of 15.6 in total PANSS scores vs placebo; however, the 95 % CrI for haloperidol showed a great amount of
uncertainty around the true relative effect vs placebo. This
statistically non-significant result may be attributed to the
positioning of haloperidol in the network as it is not directly
connected with placebo (Fig. 2). In addition, this result may
be owing to the presence of only one study (which was also
non-randomised) evaluating the effects of haloperidol. The
treatment ranking probabilities are shown in Fig. 4.
Haloperidol has the greatest probability of being the best
treatment (prob = 0.49), followed by molindone
(prob = 0.25). All remaining treatments have probabilities
of less than 0.13. While the point estimate for haloperidol
showed a statistically non-significant reduction vs placebo,
the ranking probabilities are being driven by the distribution
of the relative treatment effects and take into account the
uncertainty shown by the 95 % CrI.
Relative treatment effects vs placebo for change from
baseline to 6 weeks in positive and negative subscale
PANSS scores are shown by the forest plots (Fig. 3b, c).
Haloperidol, olanzapine and risperidone showed a statistically significant reduction in positive PANSS scores vs
placebo as the 95 % CrIs lie exclusively to the right of
zero. While all interventions showed a trend of a greater

34

R. C. Harvey et al.

Fig. 3 Forest plots for all outcomes. AE adverse events, CrI credible interval, MD mean difference, OR, odds ratio, PANSS Positive and
Negative Syndrome Scale

Relative Efficacy of Antipsychotics for the Treatment of Early-Onset Schizophrenia

35

Fig. 4 Treatment ranking probabilities

reduction in negative PANSS scores vs placebo, no comparison was statistically significant. Haloperidol and
molindone showed the greatest reduction in negative
PANSS scores vs placebo (MD = -3.42); however, the
respective 95 % CrIs spanned zero. Treatment ranking
probabilities (Fig. 4) show that haloperidol had the greatest
probability of being ranked the best treatment in the network for both positive and negative PANSS scores, with
probabilities of 0.85 and 0.40, respectively.
Relative treatment effects vs placebo for change from
baseline to 6 weeks in weight are shown by the forest plot
(Fig. 3d). Olanzapine, quetiapine and risperidone showed a
statistically significant weight gain vs placebo. Haloperidol, molindone and ziprasidone showed a trend of reduced
weight gain vs placebo; however, there was no statistically
significant difference between these treatments and placebo. The treatment ranking probabilities (Fig. 4) show
that haloperidol and molindone have the greatest probabilities of being the best treatments (prob = 0.50 and
prob = 0.45, respectively). All remaining treatments each
had probabilities of less than 0.03.
Odds ratios of all treatments vs placebo for all-cause
discontinuation are shown in the forest plot (Fig. 3e). One
treatment (risperidone) showed statistically significant
reduced odds of discontinuing treatment because of any
reason vs placebo. All treatments showed a trend of lower
odds of discontinuing because of any reason vs placebo; for
risperidone, this is a statistically significant result
[OR = 0.48, 95 % CrI (0.25, 0.84)]. The treatment with
the lower odds vs placebo was haloperidol. However, there
is much uncertainty around this estimate [OR = 0.19,
95 % CrI (0.01, 2.42)]. The treatment ranking probabilities

(Fig. 4) show that haloperidol has the greatest probability


of being the best treatment (prob = 0.67), followed by
quetiapine (prob = 0.16). All remaining treatments each
had probabilities of less than 0.08 of being ranked as the
best treatment. Discontinuation because of adverse events
was also investigated, and all treatments showed a trend of
increased odds vs placebo. However, because of low event
rates in the placebo arm of studies and the presence of only
nine studies estimating eight treatment effects, there is
considerable uncertainty in the results (Fig. S1 in the
Electronic Supplementary Material).
Haas et al. investigated the effects of risperidone at two
doses [31]. The authors stated that one of these doses
(0.150.6 mg/day) was equivalent to a control arm; hence,
in the following analyses, this arm was reclassified as
placebo vs risperidone to allow this study to be included in
the analyses. Removing this study in a sensitivity analysis
performed on total PANSS scores yielded almost identical
results (additional uncertainty was observed in the pairwise
comparisons), which are presented in the Electronic Supplementary Material (Fig. S2A), A second sensitivity
analysis was performed, removing the one identified nonrandomised study (Gothelf) from the analysis [27]. Consequently, haloperidol was omitted from the network.
Results for all remaining treatment comparisons were
almost identical to those when including this study and are
presented in the Electronic Supplementary Material
(Fig. S2B). Despite NMAs relying on the presence of
RCTs, this study was included in the analysis to allow
relative efficacy to be estimated between haloperidol and
other comparators, as it was the only evidence identified
for this treatment.

36

R. C. Harvey et al.

Table 2 Goodness-of-fit statistics


Outcome

Total
residual
deviance

No. of data points


included in the
model

Total PANSS score

26.1

24

Positive PANSS score

20.6

22

Negative PANSS score


Weight gain

20.8
20.0

22
22

All-cause discontinuation

16.0

15

Sensitivity analysis: total PANSS


score (removal of Haas et al.
[31])

19.9

22

PANSS Positive and Negative Syndrome Scale

For completeness, all pairwise treatment comparisons for


each outcome are presented in the Electronic Supplementary
Material (Figs. S3S5). These show the relative treatment
effects for all treatments compared with one another.
Goodness of fit of the model measured by the total
residual deviance is presented in Table 2. The number of
individual data points included in each model is comparable to the total residual deviance obtained from each
analysis. As such, the models appear to be a reasonable fit
to the data for each outcome. Removal of Haas et al. in a
sensitivity analysis did not improve model fit according to
the total residual deviance [31].
There is evidence of mild-to-moderate heterogeneity in
the effects of interventions between studies. This result is
expected because of the heterogeneous population under
investigation in these analyses. Inconsistency was checked
for the primary outcome (total PANSS) using a nodesplitting approach. Only three pairwise treatment comparisons could be evaluated for inconsistency, relying on the
presence of feedback loops within the network of evidence.
No statistically significant inconsistencies were detected
between the direct and indirect evidence.

4 Discussion
This study, using an NMA, aimed to provide a comparison
of the efficacy profiles of different AP medications for the
treatment of EOS, with data from both direct and indirect
trial comparisons. Changes in PANSS scores can demonstrate the efficacy of a treatment in maintaining symptom
control; however, rather than looking at total PANSS
scores, we chose to first analyse the positive and negative
symptom scores. These are useful clinical indicators, particularly in view of the poor treatment response for negative symptoms to date. While acknowledging the
uncertainty in the results, haloperidol showed a trend of
being the most efficacious intervention for reducing

positive PANSS scores followed by olanzapine and


risperidone (all comparisons vs placebo were statistically
significant). Furthermore, both risperidone and olanzapine
showed positive effects in treating negative symptoms,
which fell just short of statistical significance, and were
considerably greater than the effects of other medications.
Considerable concern has been expressed by clinicians,
academics and organisations such as NICE over the side
effects of SGAs, in particular weight gain, which can have
a far greater impact on the adolescent population [10, 38].
Weight gain is a risk factor not only for obesity, which by
itself is associated with a reduced quality of life, but also
for metabolic syndrome and diabetes [39, 40]. However,
sufficient data were available from 11 trials to allow
analysis. Three treatments showed a statistically significant
weight gain at 6 weeks vs placebo. Olanzapine showed the
greatest weight gain vs placebo, whereas molindone and
haloperidol showed the least likelihood to cause weight
gain, and indeed showed a trend of reduced weight gain
compared with placebo. These findings are reflected in
recent clinical guidelines, as olanzapine has been recommended only as a second-line treatment option because of
the significant weight gain associated with treatment [41].
Data for all-cause discontinuation were analysed as a
proxy measure for tolerability. Risperidone showed statistically significant lower odds of discontinuing treatment
when compared with placebo. All other treatments showed
a trend of lower odds of discontinuing when compared with
placebo, although results were not statistically significant.
All active treatments were found to be associated with
treatment discontinuation because of adverse events;
olanzapine was associated with the highest odds of discontinuation because of adverse events. However, uncertainty around this estimate means that it should be
interpreted with caution.
The finding from this study that haloperidol, a firstgeneration antipsychotic (FGA), is the most efficacious AP
raises some interesting questions, particularly in light of the
serious weight and cardio-metabolic side effects associated
with the use of SGAs; these problems have previously been
described as an emerging health crisis [42]. A Cochrane
review found no difference in efficacy between FGAs and
SGAs but notable and important differences in side-effect
profiles [12]. Given the higher rates of extrapyramidal side
effects (EPSE) with FGAs, some of which are serious longterm complications, such as tardive dyskinesia, it may
nonetheless be appropriate to reconsider the use of FGAs
as part of a treatment plan. For instance when weight gain
is evident or there is an increased risk of side effects such
as diabetes, abnormal lipid levels or impaired glucose
tolerance.
Comparison of our results with the NMA of APs
conducted in adults is informative [15]. The study

Relative Efficacy of Antipsychotics for the Treatment of Early-Onset Schizophrenia

conducted in the adult population was much more comprehensive and included 212 trials, with data for 43,049
participants, which is considerably more than the 11
studies and 1714 participants available here. This highlights, once again, the poor evidence base for APs in the
adolescent population. However, the interesting suggestion coming from both studies is to consider broadening
the range of medications used, depending on the clinical
needs of the patient, rather than limiting choice to the
dichotomous classification of APs into first- and secondgeneration groupings.
There are a number of limitations to the study. The
usefulness of existing trials completed in the paediatric
population is often restricted by their quality, small sample
numbers and heterogeneity among the study populations
[43]. Trials demonstrate efficacy of symptom control;
however, there is a paucity of data on the side effects of
treatment and longer-term health impacts, demonstrating a
need for more research. The potential bias identified in the
studies reduces the reliability of the data used to form the
network, and some of the bias (e.g. signs of selective bias
and differences in baseline characteristics) may lead to
incorrect interpretations around treatment effect. Small
sample numbers in the trials may be due to hesitation (in
particular ethical considerations) in conducting clinical
trials in a younger populations [44]. The literature search
was not restricted to RCTs, acknowledging the lack of data
in this population, and thus widened the evidence base,
which places the study at a higher risk of bias. Additionally, NMAs are susceptible to publication bias, with
detrimental effects of publication bias being identified in
previous psychiatric studies [45].
There is considerable uncertainty in the relative treatment effects, most likely because of the limited number of
studies included in the network. Many of the treatment
comparisons are informed by only one trial. It has previously been noted that there is often a substantial response
in the placebo arm of trials for psychiatric interventions,
which may therefore limit the reliability of NMA results
[45]. Additionally, a non-RCT was included in the analysis
to allow comparisons to be made with haloperidol. Inclusion of the study (Gothelf et al.) enabled relative effects vs
haloperidol to be estimated [27]. While a sensitivity analysis excluding this study was performed, with results
remaining consistent, NMAs do rely upon the synthesis of
RCTs only. Another limitation with the evidence base is
the assumption that risperidone (0.150.6 mg) is equivalent
to placebo. This assumption was made so this trial could be
included within the analyses, thus providing more evidence
on the higher dose of risperidone, given the paucity of data
in this evidence base. In addition, as the typical starting
dose is 0.5 mg, it is stated that this low dose is seen as a
pseudo-placebo [46].

37

The study also includes few outcomes owing to limitations in the data (a lack of reporting of all possible
outcomes) and research constraints. An increase in the
number of studies available should report a greater
number of outcomes (e.g. EPSE), which will allow for
more in-depth research. In addition, this review focused
on the PANSS outcome scale for assessing response to
treatment as this allowed us to differentiate between the
positive and negative symptoms in schizophrenia. However, this meant that some treatments could not be
included because trials studying these treatments did not
use PANSS as the tool to measure symptom control, and
they therefore did not meet the inclusion criteria (e.g. a
trial was identified with clozapine, but had to be
excluded as the outcome was not reported using
PANSS). There are a number of other scales, including
the Brief Psychiatric Rating Scale and the Clinical
Global Impression. To widen the scope of the research,
it would have been possible to widen the criteria and to
make assumptions around the relationship between outcome measures. However, this would have prevented the
separate analysis of the positive and negative subscales,
and would have relied on using the standardised MD,
which also has limitations (e.g. outcome measures would
need to be deemed similar enough to combine and
interpretability of results could be challenging). As it
stands, decision makers need to carefully consider the
treatments available in their locality and whether the
study is reflective of the commonly used treatments.
Finally, this study focuses on pharmacological treatments. There is an emerging evidence base for the effectiveness of psychological interventions, although no studies
were identified during the review that focused on these
alternative treatments.

5 Conclusion
Evidence for the comparative efficacy and safety of AP use
in the paediatric population is limited, and results should be
interpreted cautiously. The evidence does demonstrate
favourable outcomes in terms of symptom control, using
the PANSS outcome scale; however, it does also highlight
a lack of statistically significant effects on negative
symptoms, increased weight gain and a lack of quantifiable
data to assess the risk of other side effects of treatment
more comprehensively.
Clinical strategy will involve a trade-off between the
efficacy of an AP and the likely side effects resulting from
treatment. This has to be seen as a major consideration, not
only because of the serious side effects seen with SGAs
(predominantly weight gain and associated metabolic
effects), and with FGAs (EPSE, including tardive

38

R. C. Harvey et al.

dyskinesia), but also because of the high rate of discontinuation rates for APs in general.
Acknowledgments We thank Rachel Brown, a Clinical Lead
Pharmacist, for advising on the dosing of interventions included in the
analyses.
Author contributions G E. Shields and R. C. Harvey designed and
carried out the research, with A. C. James providing clinical validation. G. E. Shields undertook the systematic review, with R. C. Harvey acting as the second reviewer. R. C. Harvey undertook the
statistical analysis. G. E. Shields, R. C. Harvey and A. C. James wrote
the first draft of the manuscript. All authors contributed to and have
approved the final manuscript.

Compliance with Ethical Standards


Conflicts of interest R C. Harvey, G. E. Shields and A. C. James
state that they have no conflicts of interest.
Funding

No funding was received for this research.

References
1. Burd L, Kerbeshian J. A North Dakota prevalence study of
schizophrenia presenting in childhood. J Am Acad Child Adolesc
Psychiatry [Internet]. 1987;26:34750. Available from: http://
www.ncbi.nlm.nih.gov/pubmed/3496327. Cited 3 Dec 2015.
2. Thomsen PH. Schizophrenia with childhood and adolescent
onset: a nationwide register-based study. Acta Psychiatr Scand
[Internet]. 1996;94:18793. Available from: http://doi.wiley.com/
10.1111/j.1600-0447.1996.tb09847.x. Cited 3 Dec 2015.
3. James A, Clacey J, Seagroatt V, Goldacre M. Adolescent inpatient psychiatric admission rates and subsequent one-year mortality in England: 19982004. J Child Psychol Psychiatry.
[Internet]. 2010;51:1395404. Available from: http://www.ncbi.
nlm.nih.gov/pubmed/20738446. Cited 3 Dec 2015.
4. Clemmensen L, Vernal DL, Steinhausen H-C. A systematic
review of the long-term outcome of early onset schizophrenia.
BMC Psychiatry [Internet]. 2012;12:150. Available from: http://
www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3521197
&tool=pmcentrez&rendertype=abstract. Cited 3 Dec 2015.
5. Andreasen NC. Positive and negative symptoms in schizophrenia.
Arch Gen Psychiatry [Internet]. American Medical Association;
1990;47:615. Available from: http://archpsyc.jamanetwork.com/
article.aspx?articleid=495059&resultclick=1. Cited 8 Dec 2015.
6. Amminger GP, Henry LP, Harrigan SM, Harris MG, AlvarezJimenez M, Herrman H, et al. Outcome in early-onset
schizophrenia revisited: findings from the Early Psychosis
Prevention and Intervention Centre long-term follow-up study.
Schizophr Res [Internet]. 2011;131:1129. Available from: http://
www.ncbi.nlm.nih.gov/pubmed/21741219. Cited 10 Nov 2015.
7. Stafford MR, Jackson H, Mayo-Wilson E, Morrison AP, Kendall
T. Early interventions to prevent psychosis: systematic review
and meta-analysis. BMJ [Internet]. 2013;346:f185. Available
from: http://www.bmj.com/content/346/bmj.f185. Cited 4 Nov
2015.
8. Kendall T, Hollis C, Stafford M, Taylor C. Recognition and
management of psychosis and schizophrenia in children and
young people: summary of NICE guidance. BMJ [Internet].
2013;346:f150. Available from: http://www.ncbi.nlm.nih.gov/
pubmed/23344308. Cited 3 Dec 2015.

9. McClellan J, Stock S. Practice parameter for the assessment and


treatment of children and adolescents with schizophrenia. J Am
Acad Child Adolesc Psychiatry [Internet]. 2013;52:97690.
Available from: http://www.ncbi.nlm.nih.gov/pubmed/23972700.
Cited 18 Nov 2015.
10. Correll CU. Addressing adverse effects of antipsychotic treatment in young patients with schizophrenia. J Clin Psychiatry
[Internet]. 2011;72:e01. Available from: http://www.ncbi.nlm.
nih.gov/pubmed/21272508. Cited 3 Dec 2015.
11. Fraguas D, Correll CU, Merchan-Naranjo J, Rapado-Castro M,
Parellada M, Moreno C, et al. Efficacy and safety of secondgeneration antipsychotics in children and adolescents with psychotic and bipolar spectrum disorders: comprehensive review of
prospective head-to-head and placebo-controlled comparisons.
Eur Neuropsychopharmacol [Internet]. 2011;21:62145. Available from: http://www.ncbi.nlm.nih.gov/pubmed/20702068.
Cited 3 Dec 2015.
12. Kumar A, Datta SS, Wright SD, Furtado VA, Russell PS.
Atypical antipsychotics for psychosis in adolescents. Cochrane
Database Syst Rev [Internet]. 2013;10:CD009582. Available
from: http://www.ncbi.nlm.nih.gov/pubmed/24129841. Cited 3
Dec 2015.
13. Olfson M, Blanco C, Liu S-M, Wang S, Correll CU. National
trends in the office-based treatment of children, adolescents, and
adults with antipsychotics. Arch Gen Psychiatry [Internet].
American Medical Association; 2012;69:124756. Available
from:
http://archpsyc.jamanetwork.com/article.aspx?articleid=
1263977. Cited 8 Dec 2015.
14. Burghardt KJ, Ellingrod VL. Detection of metabolic syndrome in
schizophrenia and implications for antipsychotic therapy: is there a
role for folate? Mol Diagn Ther [Internet]. 2013;17:2130.
Available from: http://www.pubmedcentral.nih.gov/articlerender.
fcgi?artid=4077272&tool=pmcentrez&rendertype=abstract. Cited
5 Nov 2015.
15. Leucht S, Cipriani A, Spineli L, Mavridis D, Orey D, Richter F,
et al. Comparative efficacy and tolerability of 15 antipsychotic
drugs in schizophrenia: a multiple-treatments meta-analysis.
Lancet [Internet]. Elsevier; 2013;382:95162. Available from:
http://www.thelancet.com/article/S0140673613607333/fulltext.
Cited 11 Jul 2015.
16. Addington J, Addington D. Positive and negative symptoms of
schizophrenia. Schizophr Res [Internet]. Elsevier; 1991;5:519.
Available from: http://www.schres-journal.com/article/0920996491
90053T/fulltext. Cited 8 Dec 2015.
17. Blanchard JJ, Cohen AS. The structure of negative symptoms within
schizophrenia: implications for assessment. Schizophr Bull [Internet]. 2006;32:23845. Available from: http://schizophreniabulletin.
oxfordjournals.org/content/32/2/238.full. Cited 8 Dec 2015.
18. Peralta V, Cuesta MJ. How many and which are the psychopathological dimensions in schizophrenia? Issues influencing
their ascertainment. Schizophr Res [Internet]. Elsevier;
2001;49:26985. Available from: http://www.schres-journal.
com/article/S0920996400000712/fulltext. Cited 8 Dec 2015.
19. Rothberg MB, He S, Rose DN. Management of influenza symptoms
in healthy adults. J Gen Intern Med [Internet]. 2003;18:80815.
Available from: http://www.pubmedcentral.nih.gov/articlerender.
fcgi?artid=1494927&tool=pmcentrez&rendertype=abstract. Cited
17 Nov 2015.
20. Daz-Caneja CM, Pina-Camacho L, Rodrguez-Quiroga A, Fraguas D, Parellada M, Arango C. Predictors of outcome in earlyonset psychosis: a systematic review. npj Schizophr. [Internet].
Nature Publishing Group; 2015;1:14005. Available from: http://
www.nature.com/articles/npjschz20145. Cited 9 Dec 2015.
21. Vyas NS, Hadjulis M, Vourdas A, Byrne P, Frangou S. The
Maudsley early onset schizophrenia study: predictors of psychosocial outcome at 4-year follow-up. Eur Child Adolesc

Relative Efficacy of Antipsychotics for the Treatment of Early-Onset Schizophrenia

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

Psychiatry [Internet]. 2007;16:46570. Available from: http://


www.ncbi.nlm.nih.gov/pubmed/17896122. Cited 9 Dec 2015.
Critical Appraisal Skills Programme. CASP randomised controlled trial checklist [Internet]. Available from: http://www.caspuk.net/#!casp-tools-checklists/c18f8. Cited 3 Jan 2016.
Dias S, Sutton AJ, Ades AE, Welton NJ. Evidence synthesis for
decision making 2: a generalized linear modeling framework for
pairwise and network meta-analysis of randomized controlled trials.
Med Decis Making [Internet]. 2013;33:60717. Available from:
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=370
4203&tool=pmcentrez&rendertype=abstract. Cited 3 Dec 2015.
Salanti G, Ades AE, Ioannidis JPA. Graphical methods and
numerical summaries for presenting results from multiple-treatment meta-analysis: an overview and tutorial. J Clin Epidemiol
[Internet]. 2011;64:16371. Available from: http://www.ncbi.
nlm.nih.gov/pubmed/20688472. Cited 3 Dec 2015.
Kay SR, Fiszbein A, Opler LA. The Positive and Negative
Syndrome Scale (PANSS) for schizophrenia. Schizophr Bull
[Internet]. 1987;13:26176. Available from: http://www.ncbi.
nlm.nih.gov/pubmed/3616518. Cited 14 Jan 2015.
Dias S, Welton NJ, Sutton AJ, Caldwell DM, Lu G, Ades AE.
Evidence synthesis for decision making 4: inconsistency in networks of evidence based on randomized controlled trials. Med
Decis Making [Internet]. 2013;33:64156. Available from: http://
www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3704208
&tool=pmcentrez&rendertype=abstract. Cited 3 Dec 2015.
Gothelf D, Apter A, Reidman J, Brand-Gothelf A, Bloch Y, Gal
G, et al. Olanzapine, risperidone and haloperidol in the treatment
of adolescent patients with schizophrenia. J Neural Transm [Internet]. 2003;110:54560. Available from: http://www.ncbi.nlm.
nih.gov/pubmed/12721815. Cited 3 Dec 2015.
Mozes T, Ebert T, Michal S-E, Spivak B, Weizman A. An openlabel randomized comparison of olanzapine versus risperidone in
the treatment of childhood-onset schizophrenia. J Child Adolesc
Psychopharmacol [Internet]. 2006;16. Available from: https://
cpnp.org/resource/reference/150299. Cited 3 Dec 2015.
Findling RL, Robb A, Nyilas M, Forbes RA, Jin N, Ivanova S,
et al. A multiple-center, randomized, double-blind, placebocontrolled study of oral aripiprazole for treatment of adolescents
with
schizophrenia.
Am
J
Psychiatry
[Internet].
2008;165:143241. Available from: http://www.ncbi.nlm.nih.
gov/pubmed/18765484. Cited 3 Dec 2015.
Sikich L, Frazier JA, McClellan J, Findling RL, Vitiello B, Ritz
L, et al. Double-blind comparison of first- and second-generation
antipsychotics in early-onset schizophrenia and schizo-affective
disorder: findings from the treatment of early-onset schizophrenia
spectrum disorders (TEOSS) study. Am J Psychiatry [Internet].
2008;165:142031. Available from: http://www.ncbi.nlm.nih.
gov/pubmed/18794207. Cited 3 Dec 2015.
Haas M, Eerdekens M, Kushner S, Singer J, Augustyns I, Quiroz
J, et al. Efficacy, safety and tolerability of two dosing regimens in
adolescent schizophrenia: double-blind study. Br J Psychiatry
[Internet]. 2009;194:15864. Available from: http://www.ncbi.
nlm.nih.gov/pubmed/19182179. Cited 3 Dec 2015.
Haas M, Unis AS, Armenteros J, Copenhaver MD, Quiroz JA,
Kushner SF. A 6-week, randomized, double-blind, placebo-controlled study of the efficacy and safety of risperidone in adolescents with schizophrenia. Mary Ann Liebert, Inc., Rochelle, NY;
2009; Available from: http://online.liebertpub.com/doi/abs/10.
1089/cap.2008.0144. Cited 3 Dec 2015.
Kryzhanovskaya L, Schulz SC, McDougle C, Frazier J, Dittmann
R, Robertson-Plouch C, et al. Olanzapine versus placebo in
adolescents with schizophrenia: a 6-week, randomized, doubleblind, placebo-controlled trial. J Am Acad Child Adolesc Psychiatry [Internet]. 2009;48:6070. Available from: http://www.
ncbi.nlm.nih.gov/pubmed/19057413. Cited 3 Dec 2015.

39

34. Swadi HS, Craig BJ, Pirwani NZ, Black VC, Buchan JC, Bobier
CM. A trial of quetiapine compared with risperidone in the
treatment of first onset psychosis among 15- to 18-year-old
adolescents. Int Clin Psychopharmacol [Internet]. 2010;25:16.
Available from: http://www.ncbi.nlm.nih.gov/pubmed/19809337.
Cited 10 Nov 2015.
35. Singh J, Robb A, Vijapurkar U, Nuamah I, Hough D. A randomized, double-blind study of paliperidone extended-release in
treatment of acute schizophrenia in adolescents. Biol Psychiatry
[Internet]. 2011;70:117987. Available from: http://www.ncbi.
nlm.nih.gov/pubmed/21831359. Cited 3 Dec 2015.
36. Findling RL, McKenna K, Earley WR, Stankowski J, Pathak S.
Efficacy and safety of quetiapine in adolescents with
schizophrenia investigated in a 6-week, double-blind, placebocontrolled trial. J Child Adolesc Psychopharmacol [Internet].
New Rochelle: Mary Ann Liebert, Inc.; 2012;22:32742. Available from: http://online.liebertpub.com/doi/abs/10.1089/cap.
2011.0092. Cited 22 Oct 2015.
37. Findling RL, Cavus I, Pappadopulos E, Vanderburg DG, Schwartz
JH, Gundapaneni BK, et al. Efficacy, long-term safety, and tolerability of ziprasidone in children and adolescents with bipolar disorder. J Child Adolesc Psychopharmacol [Internet]. 2013;23:54557.
Available from: http://www.pubmedcentral.nih.gov/articlerender.
fcgi?artid=3804078&tool=pmcentrez&rendertype=abstract. Cited
3 Dec 2015.
38. National Institute for Health and Care Excellence. Psychosis and
schizophrenia in children and young people: recognition and
management. In: NICE guidelines CG155. [Internet]. 2013.
Available from: https://www.nice.org.uk/guidance/cg155. Cited
14 Aug 2015.
39. Newcomer JW. Second-generation (atypical) antipsychotics
and metabolic effects: a comprehensive literature review.
CNS Drugs [Internet]. 2005;19 Suppl 1:193. Available from:
http://www.ncbi.nlm.nih.gov/pubmed/15998156. Cited 29 Nov
2015.
40. Allison DB, Mackell JA, McDonnell DD. The impact of weight
gain on quality of life among persons with schizophrenia. Psychiatr Serv [Internet]. 2003;54:5657. Available from: http://
www.ncbi.nlm.nih.gov/pubmed/12663847. Cited 3 Dec 2015.
41. Buchanan RW, Kreyenbuhl J, Kelly DL, Noel JM, Boggs DL, Fischer
BA, et al. The 2009 schizophrenia PORT psychopharmacological
treatment recommendations and summary statements. Schizophr Bull
[Internet]. 2010;36:7193. Available from: http://schizophreniabulle
tin.oxfordjournals.org/content/36/1/71.full. Cited 8 Dec 2015.
42. Reeves GM, Keeton C, Correll CU, Johnson JL, Hamer RM,
Sikich L, et al. Improving metabolic parameters of antipsychotic
child treatment (IMPACT) study: rationale, design, and methods.
Child Adolesc Psychiatry Ment Health [Internet]. BioMed Central Ltd; 2013;7:31. Available from: http://www.capmh.com/
content/7/1/31. Cited 3 Dec 2015.
43. Mattai AK, Hill JL, Lenroot RK. Treatment of early-onset
schizophrenia. Curr Opin Psychiatry [Internet]. 2010;23:30410.
Available from: http://www.ncbi.nlm.nih.gov/pubmed/20502331.
Cited 3 Dec 2015.
44. Field M, Behrman R. The ethical conduct of clinical research
involving children [Internet]. National Academies Press; 2004.
Available from: https://books.google.com/books?hl=en&lr=&id=
El-bAgAAQBAJ&pgis=1. Cited 3 Dec 2015.
45. Yildiz A, Vieta E, Correll CU, Nikodem M, Baldessarini RJ.
Critical issues on the use of network meta-analysis in psychiatry.
Harv Rev Psychiatry [Internet];22:36772. Available from: http://
www.ncbi.nlm.nih.gov/pubmed/25377612. Cited 3 Dec 2015.
46. Gerlach M, Warnke A, Greenhill L. Psychiatric drugs in children
and adolescents: basic pharmacology and practical applications
[Internet]. Springer; 2014. Available from: https://books.google.
com/books?id=AmQlBAAAQBAJ&pgis=1. Cited 3 Dec 2015.