VALUE IN HEALTH ] (2017) ]]]–]]]

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/jval

Aripiprazole Lauroxil Compared with Paliperidone Palmitate in

Patients with Schizophrenia: An Indirect Treatment
Comparison
Chris Cameron, PhD1,*, Jacqueline Zummo, MPH, MBA2, Dharmik N. Desai, PharmD2,
Christine Drake, BScN1, Brian Hutton, PhD1,3, Ahmed Kotb, MSc4, Peter J. Weiden, MD2
1
Cornerstone Research Group, Inc., Burlington, Ontario, Canada; 2Alkermes, Inc., Waltham, MA, USA; 3Ottawa Hospital Research
Institute, Ottawa, Ontario, Canada; 4Royal College of Surgeons in Ireland, Dublin, Ireland

AB STR A CT

Background: Aripiprazole lauroxil (AL) is a long-acting injectable compared with two efficacious doses of PP (156 mg and 234 mg
atypical antipsychotic recently approved for treatment of schizophre- monthly). All four active-treatment conditions were associated
nia on the basis of a large-scale trial of two doses of AL versus placebo. with comparable reductions in acute symptoms (Positive and
There are no direct-comparison studies with paliperidone palmitate Negative Syndrome Scale) versus placebo and were of similar
(PP; long-acting antipsychotic used most often in acute settings) for magnitude (range of mean difference –8.12 to –12.01, with over-
the acute psychotic episode. Objectives: To indirectly compare effi- lapping 95% credible intervals). Between-group comparisons of
cacy and safety of the pivotal AL study with all PP studies meeting active-treatment arms were associated with summary estimates
indirect comparison criteria. Methods: Systematic searches of MED- of magnitude near 0. No clinically meaningful differences in
LINE, Embase, Cochrane CENTRAL, PsycINFO, ClinicalTrials.gov, selected safety or tolerability parameter incidence were found
International Clinical Trials Registry Platform, and gray literature between active treatments. Conclusions: These results suggest
were performed to identify randomized controlled trials of PP with that both AL and PP are effective for treatment of adults experi-
similar designs to the AL trial. Bayesian network meta-analysis encing acute exacerbation of schizophrenia.
compared treatments with respect to symptom response and Keywords: aripiprazole lauroxil, indirect comparison, paliperidone
tolerability issues including weight gain, akathisia, parkinsonism, palmitate, schizophrenia.
and likelihood of treatment-emergent adverse events. Results:
Three appropriate PP studies were identified for indirect compar- Copyright & 2017, International Society for Pharmacoeconomics and
ison. Both doses of AL (441 mg and 882 mg monthly) were used and Outcomes Research (ISPOR). Published by Elsevier Inc.

treating acutely ill patients with LAI formulations of antipsychotics

Introduction
is to stabilize the acute symptoms with an antipsychotic that has a
Schizophrenia is a persistent and disabling disorder character- long-acting option. Initiation of the LAI formulation during an acute
ized by distortions in thoughts, perceptions, emotions, language, psychotic episode would have the dual goals of treating the acute
disease awareness, and behavior [1,2]. The course of schizophre- psychotic episode while simultaneously converting the patient to
nia varies and is often phasic in nature, with intervals of stability the long-acting formulation for ongoing maintenance therapy after
and relapse [3]. Some patients experience acute psychotic epi- the acute psychotic episode has resolved.
sodes followed by periods of relative stability, whereas others Although it is beyond the scope of this report to go into detail
remain persistently symptomatic. One of the fundamental about the relative clinical value of LAIs over their oral counter-
aspects of long-term treatment after the acute psychotic episode parts, in general, they are recommended as a more consistent
resolves is that continuous treatment with antipsychotic medi- and reliable way to administer ongoing antipsychotic treatment.
cation is needed to prevent or delay future relapse [4]. Therefore, Furthermore, relative to oral medication, LAIs can be helpful in
pharmacologic treatment of the acute psychotic episode should reducing the likelihood of inadvertent gaps in taking medication
consider the impact of the current treatment for the future as and may lengthen the duration of time during which the patient
well as for the current episode. remains on uninterrupted therapy. Also, in marked contrast to
Some second-generation antipsychotics are available as long- oral medication, the patient’s adherence status is always known
acting injectable (LAI) formulations. The general strategy for and therefore provides the clinician with accurate and real-time

* Address correspondence to: Chris Cameron, Cornerstone Research Group, Inc., Suite 204, 3228 South Service Road, Burlington, Ontario,
Canada L7N 3H8.
E-mail: ccameron@cornerstone-research.com.
1098-3015$36.00 – see front matter Copyright & 2017, International Society for Pharmacoeconomics and Outcomes Research (ISPOR).
Published by Elsevier Inc.
http://dx.doi.org/10.1016/j.jval.2017.03.010
2 VALUE IN HEALTH ] (2017) ]]]–]]]

information on those patients who may decide to stop their Outcomes

medication. Studies that reported information pertaining to changes in
As more LAIs become available, the need to evaluate the PANSS total score from baseline at approximately 12 weeks were
relative benefits and risks between them is underscored. There of interest. In addition, those reporting on the occurrence of
are few direct comparisons of the relative efficacy and tolerability weight gain of more than 7%, akathisia (i.e., sensation of rest-
of starting a long-acting formulation during the acute episode. lessness and a sense of need for continuous motion), and treat-
Network meta-analysis (NMA) is an increasingly used method [5] ment-emergent, nonakathisia, extrapyramidal symptom (EPS)-
for indirect comparison of two or more interventions that have related adverse events (AEs) and treatment-emergent adverse
not been compared in head-to-head clinical trials [6]. NMA events (TEAEs; e.g., pain at injection site, myalgia, dizziness,
methods have been used in psychiatry by researchers to evaluate insomnia, headache, anxiety, agitation, nausea, vomiting, con-
oral antipsychotic drugs for schizophrenia [7], treatments for stipation, and suicidal ideation) were reviewed.
bipolar disorder [8], and monotherapies for acute mania [9]. In the
present NMA, an indirect comparison was performed between
the newest US Food and Drug Administration–approved, second-
Study design
To allow for reliable indirect comparisons, only studies with
generation, monthly LAI antipsychotic, aripiprazole lauroxil (AL;
design features similar to those of Meltzer et al. [14] were
Aristadas, Alkermes, Inc., Waltham, MA), and the most widely
selected. In particular, the Meltzer study was the pivotal phase
used second-generation, monthly LAI, paliperidone palmitate (PP;
III randomized controlled trial (RCT) that recruited acutely ill
Invega Sustennas, Janssen Pharmaceuticals, Inc., Titusville, NJ),
subjects entering the hospital, and an AL initiation regimen
in the United States.
(including a 3-week cotreatment period with oral aripiprazole)
was used to stabilize these patients. To reduce bias, studies using
different study designs (e.g., randomized withdrawal designs,
phase II) and/or having a much longer treatment period (e.g., 52
Methods
weeks) or not reporting 12-week data were not included in the
indirect comparison. Finally, for PP studies, we excluded any PP
Study Eligibility Criteria dose arm that would currently be considered subtherapeutic for
A key assumption underpinning NMA is transitivity of the studies acutely ill patients.
[6,10–13]. That is, all patients could have been randomized to any
of the trials that are included in the study. To assess transitivity,
Approach to Literature Search for Eligible Studies
one can collect information from identified studies, such as
design and patient characteristics, and carefully consider A bibliographic database search was conducted in MEDLINE
whether the studies appear similar enough to be compared (At http://ospguides.ovid.com/OSPguides/medline.htm, I see the
[6,10–13]. In accordance with these principles, significant efforts following: Epub Ahead of Print, In-Process & Other Non-Indexed
were undertaken to critically appraise the literature and to Citations Ovid MEDLINE Daily, and Ovid MEDLINE 1946 to
identify and analyze studies that were most similar to the AL present), Embase, Cochrane CENTRAL, and PsycINFO (OVID plat-
pivotal phase III trial by Meltzer et al. [14] in terms of patient and form) from 2000 to November 11, 2016. The search strategies were
study characteristics. Eligibility criteria were established using designed by one information specialist and independently peer-
the following Population Interventions Comparators Outcomes reviewed by a second using the Peer Review for Electronic Search
Study Design (PICOS) framework criteria. Strategies tool [15]. Searches were developed in MEDLINE and
adapted to each database using a combination of subject heading
and text word terms for “aripiprazole lauroxil” or “paliperidone
palmitate” and “schizophrenia.” Results were limited to RCTs
Population
published in English from 2000 onward. A gray literature search
The population comprised adults with a diagnosis of schizophre-
for health technology assessments, regulatory information, and
nia according to the Diagnostic and Statistical Manual of Mental
government documents was conducted. This included an exten-
Disorders, 4th Edition, Text Revision, criteria and who were
sive search of the International Clinical Trials Registry Platform
experiencing an acute exacerbation. To allow comparison with
and ClinicalTrials.gov, which was performed using the Advanced
the study by Meltzer et al. [14], identified studies were critically
Search feature and broad search terms to reduce the risk of
appraised to ensure that patients among included studies were
missing relevant studies. The Canadian Agency for Drugs and
similar to those in the study by Meltzer et al. [14] in terms of
Technologies in Health gray literature tool, Grey Matters: a practical
baseline Positive and Negative Syndrome Scale (PANSS) total
tool for searching health-related grey literature (http://www.cadth.ca/
scores (the PANSS is the most widely used symptom severity
resources/grey-matters), was consulted for this search. See
rating scale for schizophrenia trials and was the primary out-
Appendix 1 in Supplemental Materials found at http://dx.doi.
come in Meltzer et al. [14]), mean age, and Clinical Global
org/10.1016/j.jval.2017.03.010 for a full description of the search
Impression-Severity (CGI-S) Scale scores. Studies were not eligible
strategy and resources used.
for inclusion if they differed significantly from Meltzer et al. [14]
(see Appendix 3 in Supplemental Materials found at http://dx.doi.
org/10.1016/j.jval.2017.03.010). Study Selection Process
Two reviewers independently reviewed citation titles and
abstracts identified to assess study eligibility. Citations consid-
Interventions/comparators ered to describe potentially eligible articles were independently
Studies evaluating PP 156 mg and 234 mg monthly were of reviewed in full-text form for formal inclusion in the final review.
interest for comparison. These doses were selected because they Disagreements between reviewers were resolved during a con-
are the most widely used doses of PP and fall within the sensus meeting. A modified Preferred Reporting Items for Sys-
recommended dosing range for schizophrenia and/or schizo- tematic Reviews and Meta-Analyses (PRISMA) flow diagram
affective disorder. Placebo was also included as an eligible documenting the process of study selection was also prepared
comparator, given its potential role as an important source of (see Appendix 2 in Supplemental Materials found at http://dx.doi.
indirect evidence for NMAs. org/10.1016/j.jval.2017.03.010).
 VALUE IN HEALTH ] (2017) ]]]–]]] 3

Data Collection from Included Studies were used to adjust for the relation between baseline PANSS total
score and change scores in the drug and placebo groups after a
A standardized Excel-based form for data extraction was created
homogeneous set of trials was identified. The adjusted changes
to capture all relevant information from studies included in the
in PANSS scores were subsequently analyzed with WinBUGS,
review, and data were gathered independently by two reviewers.
adjusting for baseline PANSS total score in the NMA model. We
The following data elements were systematically collected from
conservatively applied a constant decrement to PP (without a
each study: aspects of study characteristics (ClinicalTrials.gov
variance) so as to not penalize PP for the adjustment.
identifier [National Clinical Trial (NCT) number], related publica-
A sensitivity analysis was conducted to explore the impact of
tions and authors, study start and completion dates, and study
incorporating a phase II PP study [18], comparing placebo and PP
sponsor), patient descriptors (eligibility criteria, sex distribution,
156 mg, into the change in PANSS score NMA (see Appendix 11 in
average and range of patient age, average baseline PANSS total
Supplemental Materials found at http://dx.doi.org/10.1016/j.jval.
score, average baseline CGI-S score, and average baseline weight
2017.03.010). The checklist from the PRISMA extension statement
in kilograms), information regarding the interventions studied
for NMA was used to guide the preparation of this article [19]. A
(dosing, timing, and method of administration of interventions),
completed checklist is provided in Appendix 6 in Supplemental
and study design and conduct considerations (duration of follow-
Materials found at http://dx.doi.org/10.1016/j.jval.2017.03.010.
up; techniques for randomization, allocation concealment, and
blinding; additional design details; and patient attrition).

Results
Data Analysis and Reporting of Findings
For all outcomes, we performed Bayesian NMAs according to
Extent of Evidence Identified
well-established methods outlined by the National Institute for
Health and Care Excellence [10–12]. Outcomes included a change A total of 1318 citations were identified by the electronic liter-
from baseline in PANSS total score, occurrence of weight gain of ature search of MEDLINE, Embase, Cochrane CENTRAL, PsycINFO,
more than 7%, TEAEs, akathisia, and EPS. Between-group differ- ClinicalTrials.gov, International Clinical Trials Registry Platform,
ences for change in PANSS total score were expressed as mean and the gray literature. No additional studies were identified via
differences with 95% credible intervals (CrIs), and comparisons of inspection of bibliographies of included studies. A total of 1318
binary outcomes were expressed as odds ratios (ORs) with 95% unique citations were screened after removal of duplicates. Of
CrI. Estimates with 95% CrI that excluded the null value of 1 for these records, 1268 were excluded because of irrelevance, and 50
binary outcomes (or 0 for continuous outcomes) were considered were subjected to full-text screening. This resulted in four studies
to reflect differences between interventions. Both fixed-effects that met eligibility criteria [14,20–22]. Appendix 2 in
(FE) and random-effects (RE) NMA models were used. The Supplemental Materials reports the modified PRISMA flow dia-
presence of single-study connections between interventions in gram outlining reasons for exclusion of studies during full-text
the treatment network for some of the outcomes (EPS-related AEs screening, with a listing of rationales for excluded studies
and akathisia) precluded the use of RE models. To assess whether provided in Appendix 3 in Supplemental Materials.
models were adequate fits to the data, we compared the posterior
mean of the residual deviance from each NMA to the correspond- Study Characteristics
ing number of unconstrained data points (approximately equal if Meltzer et al. (NCT01469039) [14] was sponsor-funded by
fit is adequate). The deviance information criterion (DIC) was also Alkermes, Inc., and Pandina et al. (NCT00590577) [20] was
assessed to examine the model fit between the FE and RE models, sponsor-funded by Janssen Pharmaceuticals, Inc. Nasrallah
where a lower DIC would imply a better-fitted model if the et al. (NCT00101634) [21] and Gopal et al. (NCT00147173) [22] were
difference in DIC between the two models was 3 or more. We both sponsor-funded by Johnson & Johnson Pharmaceutical
also estimated the probability of each intervention being the best Research and Development. All four studies used a phase III,
(referred to as p[best]) and inspected each treatment’s surface parallel-group study design. The studies of Meltzer et al. [14],
under the cumulative ranking (SUCRA) curve value [16]. For Pandina et al. [20], and Gopal et al. [22] were scored to be
interpretation, both p[best] and SUCRA values range from 0% to adequate for double blinding and allocation concealment (see
100%, and values closer to 100% are preferred. NMAs were Appendix 4 in Supplemental Materials found at http://dx.doi.org/
performed using WinBUGS version 1.4.3 (MRC Biostatistics Unit, 10.1016/j.jval.2017.03.010), but there was a high risk of bias for
Cambridge, UK) and NetMetaXL version 1.6.1 software (Corner- double blinding and allocation concealment for the study by
stone Research Group Inc. , Burlington ON) . Three chains were fit, Nasrallah et al. [21]. Attrition was high but similar in all four
and analyses used burn-in samples of 40,000 iterations or more studies (see Appendix 4 in Supplemental Materials). The treat-
and subsequent sampling iterations of 40,000 or more. Trace ment period among studies ranged from 12 to 13 weeks.
plots and Gelman-Rubin plots were reviewed to assess model
convergence, and autocorrelation was assessed. Forest plots
summarizing all estimated pairwise differences between inter- Intervention Characteristics
ventions, along with corresponding 95% CrIs, were prepared to Including the placebo arms, a total of four different intervention
present all findings. To assess inconsistency in the change from groups were identified in the PP studies and three in the AL study.
baseline PANSS score, deviance and DIC statistics were compared There were differences in the dosing frequency and oral co-
in fitted consistency and inconsistency models [12]. The posterior administration of AL and PP. The PP strategy was to follow the
mean deviance of the individual data points in the consistency initial injection with a second injection 1 week later, with the
model was also plotted against their posterior mean deviance in third and fourth injections in 4-week intervals. In Pandina et al.
the inconsistency model to identify any loops in which incon- [20], the initiation dose of PP on day 1 (234 mg intramuscularly
sistency was present (see Appendix 10 in Supplemental Materials [IM]) was followed by 156 mg, 234 mg, or 39 mg IM injections on
found at http://dx.doi.org/10.1016/j.jval.2017.03.010) [12]. days 8, 36, and 64 (Table 1). In Gopal et al. [22] and Nasrallah et al.
We also conducted a covariate-adjusted NMA on the basis of a [21], the initiation dose of PP on day 1 was the same dose
published individual patient data (IPD) meta-analysis for PANSS administered on subsequent days. In Gopal et al. [22], doses of
total score [17]. In brief, given that baseline PANSS score is an 78 mg, 156 mg, and 234 mg were examined. In Nasrallah et al.
effect modifier [17], equations developed by Furukawa et al. [17] [21], doses of 39 mg, 78 mg, and 156 mg were examined. All arms
 4
Table 1 – Study characteristics.
Characteristic Study, NCT no. (other study ID nos.)

Meltzer et al. [14], NCT01469039 Pandina et al. [20], NCT00590577 Gopal et al. [22], NCT00147173 Nasrallah et al. [21],
(CR012550, R092670PSY3007) NCT00101634

Trial duration 12 13 13 13
(wk)
Study design Phase III, multicenter, randomized, double-blind, Phase III, multicenter, randomized, double- Phase III, multicenter, randomized, double- Phase III, multicenter,
brief parallel-group, placebo-controlled study blind, parallel-group, placebo-controlled blind, parallel-group, placebo-controlled randomized, double-blind,
summary designed to evaluate AL in patients with study of three doses of PP or placebo study designed to confirm the efficacy parallel-group, placebo-
schizophrenia experiencing an acute consisting of two periods: 7-d screening and safety of three fixed doses of PP or controlled study designed to

VALUE IN HEALTH ] (2017) ]]]–]]]

exacerbation episode. After up to 10 d for period and 13-wk double-blind period. placebo. There was a 7-d screening evaluate the efficacy and safety
screening, patients were randomized to a 12-wk period and 13-wk double-blind of three fixed doses of PP or
treatment period and an 8-wk follow-up period. treatment period. placebo. There was a 7-d
screening period and a 13-wk
double-blind treatment period.
Severity of
DSM-IV diagnosis of schizophrenia for Z2 y before
DSM-IV diagnosis of schizophrenia for
DSM-IV diagnosis of schizophrenia for at
DSM-IV diagnosis of schizophrenia
illness at screening at least 1 y before screening least 1 y before screening for at least 1 y before screening
study entry
Experiencing an acute exacerbation or relapse with
Experiencing an acute exacerbation of Screening and baseline PANSS scores Z70 Screening and baseline PANSS
onset o2 mo before screening symptoms and r120 scores Z70 and r120

Outpatient or hospitalized for o2 wk for the Screening PANSS score Z70 and r120, and
current exacerbation baseline score Z60 and r120
Screening and baseline PANSS scores Z70 and
r120
Study arms Arm 1: AL IM Arm 2: AL IM Arm 3: Arm 1: PP Arm 2: PP Arm 3: Arm 1: PP Arm 2: PP Arm 3: Arm 1: PP 156 mg Arm 2:
injection, injection, Placebo IM 234 mg IM 234 mg IM Placebo 156 mg 234 mg Placebo IM injection Placebo
441 mg days 882 mg days injection injection injection IM IM IM IM days 1, 8, 36, IM
1, 29, and 57 1, 29, and 57 days 1, 29, day 1 and days 1, 8, injection injection injection injection and 64 injection
and 57 156 mg IM 36, and 64 days 1, 8, days 1, days 1, 8, days 1, 8, days 1, 8,
Oral Oral Oral days 8, 36, 36, and 8, 36, 36, and 36, and 36, and
aripiprazole aripiprazole placebo and 64 64 and 64 64 64 64
15 mg/d 15 mg/d from day 1
from day 1 from day 1 to 21
to 21 to 21
A fourth arm of PP 234 mg IM day 1 and 39 mg A fourth arm of PP 78 mg IM days 1, 8, 36, Third (78 mg) and fourth (39 mg)
IM days 8, 36, and 64 was excluded. and 64 was excluded. arms of PP IM days 1, 8, 36, and
64 were excluded.

continued on next page

 VALUE IN HEALTH ] (2017) ]]]–]]] 5

in all of the PP studies had no oral supplementation. The arms

AL, aripiprazole lauroxil; CGI-S, Clinical Global Impression-Severity Scale; DSM-IV, Diagnostic and Statistical Manual of Mental Disorders, 4th edition; ID, identification; IM, intramuscular; NCT,
Bulgaria, Romania, and Russia
with 39-mg and 78-mg doses were excluded because these doses

Pharmaceutical Research and

None during the study period.

are considered subtherapeutic for the treatment of acutely ill

United States, South Africa,
patients, and including these arms would bias the efficacy

40.8 ⫾ 11.3
comparison in favor of AL (Table 1).

90.9 ⫾ 12

81.4 ⫾ 22
Johnson & Johnson
The AL strategy started with the initial injection dose remain-

NR
67
Development ing the same for all future injections per a 4-week injection
interval. All subjects receiving AL also received a 21-day oral
supplementation after the first injection, which was then dis-
continued for the rest of the study. Thus, AL 441 mg IM was given
on day 1 along with 15 mg oral aripiprazole for 21 days, followed
by AL 441 mg on days 29 and 57, and AL 882 mg IM was given on
day 1 along with 15 mg oral aripiprazole for 21 days, followed by
882 mg on days 29 and 57. All of the studies included a placebo
United States, Malaysia, the Republic of

group (Table 1).

Johnson & Johnson Pharmaceutical

Figure 1 presents a network graph summarizing the structure

of the treatment network. As mentioned, we excluded the 39-mg
Korea, Taiwan, and Ukraine
None during the study period.

Research and Development

and 78-mg PP doses, given that they are not used in clinical
84.4 ⫾ 24.6
40 ⫾ 10.8

91 ⫾ 11.8

practice.
NR
69

Patient Characteristics
The four included studies enrolled a total of 1589 patients (400
for all AL doses, 576 for the PP doses used in this comparison,
and 613 for the combined placebo groups). Pertinent demo-
graphic characteristics, including mean age, sex distribution,
baseline mean PANSS total scores, CGI-S scores, and body
National Clinical Trial; NR, not reported; PANSS, Positive and Negative Syndrome Scale; PP, paliperidone palmitate.

weight, are provided in Table 1. The percentage of male patients

None during the study period; lorazepam was

(66%–69%) and average patient age (39.3–40.8 years) were sim-

allowed throughout the duration of the

ilar between the studies. Baseline mean PANSS total score

United States, Europe, and Asia

ranged from 87.1 to 92.8, and mean CGI-S scores ranged from
Janssen Pharmaceuticals, Inc.

4.7 to 4.9. Figure 2 presents a comparison of several study and

patient characteristics, including identification of a small
imbalance in baseline PANSS total score, for the four studies.
39.3 ⫾ 10.6

87.1 ⫾ 10.9

4.7 ⫾ 3.6

This variation was addressed by conducting a second NMA that

NR
66

adjusted for baseline PANSS total score using a published IPD

study for all arms.

analysis [17].

Findings from Network Meta-Analyses

Summaries for all NMAs are provided hereafter for each outcome.
When available, analyses were based on data from the four RCTs
and a total of 1589 patients. For reported outcome data, we
present findings from the RE model here, with findings from
throughout the duration of the study for all arms.
daily from day 1 to 21; the placebo injection arm

the FE models presented in Appendix 8 in Supplemental

Both active arms received oral aripiprazole 15 mg

Materials found at http://dx.doi.org/10.1016/j.jval.2017.03.010.

antipsychotic was allowed for the remaining

Findings from the FE models are, however, presented here for

received placebo tablets; no further oral

study duration; lorazepam was allowed

EPS-related AEs and akathisia because data were not available for
United States, Europe, and Asia

two of the PP studies (Table 2). Model fit was found to be adequate

for all analyses, with the number of unconstrained data points
Alkermes, Inc.

being of comparable magnitude or less to that of the posterior

92.8 ⫾ 10.8
39.7 ⫾ 11

80 ⫾ 18.5
4.9 ⫾ 0.6

mean residual deviance for the analysis. There were no issues of

68

convergence or autocorrelation. A summary of model fit for each

analysis is provided in Appendix 5 in Supplemental Materials
found at http://dx.doi.org/10.1016/j.jval.2017.03.010.

Change in PANSS total score

Results from the primary NMA are presented in Figure 3. All four
regimens were associated with reductions in PANSS total score
compared with placebo and were of similar magnitude (range of
mean differences –8.12 to –12.01, with overlapping 95% CrIs).
Between-group comparisons of the four effective-treatment arms
score, mean
supplement

Study sponsor

mean ⫾ SD

mean ⫾ SD
Age (y), mean

Sex: Male (%)

(AL 441 mg and 882 mg; and PP 156 mg and 234 mg) were
CGI-S score,

Weight (kg),
PANSS total
Study sites
Use of oral

associated with summary estimates of magnitude near 0. The

⫾ SD

⫾ SD

results of the NMA conducted to adjust for the slight difference in

baseline PANSS total score (using IPD equations developed by
Furukawa et al. [17]) are also presented in Figure 3. Only minor
6 VALUE IN HEALTH ] (2017) ]]]–]]]

Fig. 1 – Evidence network for eligible randomized controlled trials. Note. Treatment nodes are sized to reflect the proportionate
number of patients randomized to each treatment in the network. All pairwise comparisons were based on single studies. AL,
aripiprazole lauroxil; IM, intramuscular; PP, paliperidone palmitate.

shifts in the summary estimates for all comparisons were Weight gain
observed. The sensitivity analysis conducted to examine the Because the more than 7% increase in body weight is a standard
impact of inclusion of the phase II PP study [18] on change in regulatory metric, it was available for all publications and is used
PANSS total score revealed only minor changes in the summary here as the outcome variable. Figure 4A presents a graphical
estimates for all comparisons (see Appendix 11 in Supplemental summary of estimated pairwise comparisons with corresponding
Materials). 95% CrIs for weight gain of more than 7% across studies.
Results from the pairwise meta-analyses compared favorably Comparisons of each active regimen versus placebo were asso-
with estimates from the NMA (see Appendix 9 in Supplemental ciated with ORs of 1.74, 1.55, 3.47, and 3.14 for AL 441 mg, AL 882
Materials found at http://dx.doi.org/10.1016/j.jval.2017.03.010). Using mg, PP 156 mg, and PP 234 mg, respectively, with 95% CrIs for PP
RE consistency models as the reference case, RE inconsistency 156 mg and 234 mg excluding the null value of 1. No differences
models were found to be associated with DIC values of comparable between active treatments were observed.
magnitude (DIC values of 36.95 and 36.83 for consistency and
inconsistency models, respectively). A scatterplot of the posterior
mean deviance for the consistency model (primary NMA) versus
Treatment-emergent adverse events
Figure 4B presents a graphical summary of estimated pairwise
the inconsistency model is presented in Appendix 10 in
comparisons with corresponding 95% CrIs for TEAEs across
Supplemental Materials. Inspection did not identify any concerns
studies. There were no differences for the comparisons between
regarding inconsistency between direct and indirect evidence.
each active treatment and placebo. Similarly, there were no
Corresponding values for p[best] and SUCRA were both low-
differences between the active treatments.
est for placebo (p[best] ¼ 0%, SUCRA ¼ 0%), and values for both
were highest for AL 882 mg (p[best] ¼ 55.6%, SUCRA ¼ 83.7%)
in the unadjusted NMA. The findings remained the same for Extrapyramidal symptoms
placebo in the adjusted analysis, and those for AL 882 mg Figure 4C presents a graphical summary of estimated pairwise
decreased slightly (p[best] ¼ 50.4%, SUCRA ¼ 80.1%) but comparisons with corresponding 95% CrIs for treatment-emer-
remained the most favorable of the four active-treatment arms. gent, EPS-related AEs (these excluded akathisia-related terms,
More detailed findings related to the probability of being ranked which were analyzed separately as mentioned in the next
best, second best, third best, and fourth best are reported in section). There were no differences between the active treat-
Appendix 7 in Supplemental Materials found at http://dx.doi. ments and placebo. Similarly, there were no differences between
org/10.1016/j.jval.2017.03.010. the active treatments.
 VALUE IN HEALTH ] (2017) ]]]–]]] 7

Fig. 2 – Graphical exploration of baseline heterogeneity between trials. CGI-S, Clinical Global Impression-Severity Scale;
PANSS, Positive and Negative Syndrome Scale.

Akathisia different oral atypical or second-generation antipsychotics for clini-

Figure 4D presents a graphical summary of estimated pairwise cians to select from, not including the older, so-called conventional
comparisons with corresponding 95% CrIs for akathisia. Com- or first-generation antipsychotics. Because many of these agents
parisons of each active regimen versus placebo suggested have been available and studied for some time, it has been possible
increases in the risk of akathisia with AL (OR 2.96 and 2.95 for to conduct NMAs on most of these agents.
the 441-mg and 882-mg doses, respectively; 95% CrIs excluded There are fewer LAI choices available, and many of these
the null value of 1, meaning that subjects in both AL arms were options are assumed to have efficacy and tolerability profiles
more likely to have an akathisia AE compared with subjects that reflect the active antipsychotic moiety released into the
receiving placebo). For PP, the comparison CrIs with placebo plasma. LAIs currently available for the treatment of schizo-
were not associated with increases (OR 0.99 and 1.14 for the phrenia include those with an active moiety that is one of the
156-mg and 234-mg doses, respectively, with 95% CrIs that high-potency first-generation agents (haloperidol decanoate and
included the null value of 1). In comparisons between regimens, fluphenazine decanoate for haloperidol and fluphenazine,
there were no differences between the AL and PP doses. respectively) or second-generation agents (risperidone micro-
spheres for risperidone, PP for paliperidone, olanzapine
pamoate for olanzapine, and aripiprazole monohydrate and
AL for aripiprazole). Given the growing number of treatment
Discussion
options available as a long-acting formulation, comparisons
among the various LAIs are needed to identify treatment
Summary of Findings options with the best balance of benefits and risks to optimize
LAI medications for the management of schizophrenia can offer patient care.
various important benefits for patients, including reduced fre- Given the current lack of head-to-head RCTs comparing LAIs,
quency of administration, better treatment adherence, and reduced systematic review methods incorporating novel evidence-
risk of rapid relapse. In the United States, there are as many as 14 synthesis approaches are needed. However, there exists a high

Table 2 – Available outcomes for eligible randomized controlled trials.

Outcome Study, NCT no. (other study ID nos.)

Meltzer et al. [14], Pandina et al. [20], Gopal et al. [22], Nasrallah et al. [21],
NCT01469039 NCT00590577 NCT00147173 NCT00101634
(CR012550, R09267OPSY3007)

Change in PANSS total All arms All arms All arms All arms
score from baseline
Weight gain 47% All arms All arms All arms All arms
TEAE All arms All arms All arms All arms
EPS* All arms All arms NR NR
Akathisia All arms All arms NR NR
AE, adverse event; EPS, extrapyramidal symptom; ID, identification; NCT, National Clinical Trial; NR, not reported; PANSS, Positive and
Negative Syndrome Scale; TEAE, treatment-emergent adverse event.
* Treatment-emergent, nonakathisia, EPS-related AEs.
8 VALUE IN HEALTH ] (2017) ]]]–]]]

Treatment Comparison Mean Difference (95% CrI)

Favors First Treatment Favors Second Treatment

Fig. 3 – Forest plot with unadjusted and adjusted changes in PANSS total score from random-effects network meta-analyses.
Note. PP study subjects in Pandina et al. [20] were initially treated with 234 mg on day 1; all AL study subjects were initially
treated with 15 mg oral aripiprazole for 3 weeks. AL, aripiprazole lauroxil; CrI, credible interval; IPD, individual patient data;
PANSS, Positive and Negative Syndrome Scale; PL, placebo; PP, paliperidone palmitate.

degree of both clinical and methodologic heterogeneity among discontinuation. Similarly, consistent with observations among
trials of injectables in schizophrenia [23–25]. Therefore, it is oral antipsychotic treatments, PP was associated with a greater
imperative that careful assessment of candidate studies is per- risk of weight gain compared with placebo. We found no
formed before their combination within an indirect comparison/ differences in TEAEs and treatment-emergent, nonakathisia,
NMA. To ensure that the indirect comparisons conducted in the EPS-related AEs.
present study were based on studies enrolling comparable
patient populations and using similar designs, we limited our
review to a homogeneous set of trials including either AL or PP. Strengths and Limitations
AL is a novel LAI recently approved by the US Food and Drug The present analysis adhered to best practices for the conduct of
Administration for the treatment of schizophrenia, whereas PP is NMAs [10,11] and followed PRISMA guidelines for extension of
the most widely used LAI in the United States. NMAs [19]. The main strengths of the present analysis relate to
After a systematic search of ClinicalTrials.gov as well as its rigor with respect to assessment of study heterogeneity.
the published literature, we identified one published study of Schizophrenia is a complex condition with significant patient
AL [14] and three published studies of PP [20–22] with similar heterogeneity along the disease continuum. Rather than naively
patient and study characteristics. All included studies were pooling all studies comparing these agents, we opted to carefully
placebo-controlled, had similar baseline total PANSS scores, assess identified studies and compare only those that were
considered multiple drug doses, and were approximately similar in terms of patient characteristics and study design. This
12 weeks in duration. No RCTs involving a direct comparison approach resulted in some limitations that warrant considera-
between the two agents were found; therefore, any inferences tion. Given that rigorous criteria were applied to ensure that the
drawn in the present report are based on findings from the assumptions underpinning NMA were not violated, the number
indirect comparison/NMA. Despite the absence of such head-to- of eligible studies was restricted. We were therefore left with only
head comparisons, studies have shown that inferences drawn four RCTs to inform the indirect comparisons of interest. Because
from indirect comparisons are similar to those from direct of the restrictive inclusion/exclusion criteria, we identified PP
evidence, particularly when careful attention is taken to assess studies that yielded similar effect estimates for most outcomes,
heterogeneity among candidate studies [13,26]. lending credibility to our analyses and conclusions. Nonethe-
Our indirect comparison found no differences in total PANSS less, in reviewing the characteristics of the eligible trials, we
scores between AL and PP. Findings were consistent between identified a small imbalance in baseline PANSS total score
the unadjusted analysis and adjusted analysis, in which we between the trials; this represented a threat to the transitivity
accounted for the slightly higher baseline PANSS total score in assumption required for a valid NMA. To address this concern,
Meltzer et al. [14]. In general, the overall safety profiles for we conducted a secondary analysis incorporating IPD, which
AL and PP were similar to those observed for oral forms. We accounted for variations in baseline PANSS total score. This
found that AL was associated with an increase in akathisia secondary analysis produced findings that were similar to those
relative to placebo, a finding consistent with the known safety in our primary analysis for this outcome, providing more
profile of antipsychotic medications. Nevertheless, there were confidence in our findings for PANSS total score. Finally, the
no differences between AL and PP. Furthermore, when use of Meltzer et al. [14] as the anchor study necessitated
akathisia was reported in Meltzer et al. [14], it tended to occur comparison with studies with shorter treatment periods (e.g.,
early in treatment and was generally mild to moderate in 12 weeks) and intermediate outcomes such as PANSS total
severity, unrelated to dose, and rarely resulted in treatment score. Follow-up analyses using methodologically rigorous
 VALUE IN HEALTH ] (2017) ]]]–]]] 9

Fig. 4 – Forest plots with odds ratios (95% CrI) for other AE outcomes. Random-effects results are presented for weight gain of
more than 7% and TEAEs. Fixed-effects results are presented for EPS and akathisia. Note. PP study subjects in Pandina et al.
[20] were initially treated with 234 mg on day 1; all AL study subjects were initially treated with 15 mg oral aripiprazole for
21 d. †Treatment-emergent, nonakathisia, EPS-related AEs. ‡When akathisia was reported, it tended to occur early in
treatment and was generally mild to moderate in severity, not dose related, and rarely resulted in discontinuation of PP. AE,
adverse event; AL, aripiprazole lauroxil; CrI, credible interval; EPS, extrapyramidal symptoms; PL, placebo; PP, paliperidone
palmitate; TEAE, treatment-emergent adverse event.

nonrandomized study designs should be conducted to validate

findings from our analysis and investigate how treatments Supplemental Materials
compare in terms of other important outcomes such as hospi-
talizations and relapse. Supplemental material accompanying this article can be found in
the online version as a hyperlink at http://dx.doi.org/10.1016/j.
jval.2017.03.010 or, if a hard copy of article, at www.valueinhealth
journal.com/issues (select volume, issue, and article).
Conclusions
The present NMA suggests that AL is associated with similar
reductions in PANSS total score compared with PP in patients
with schizophrenia experiencing an acute exacerbation. No R EF E R EN C ES
differences in TEAEs, EPS, akathisia, or weight gain were found
between AL and PP. These results suggest that clinicians can
consider either AL or PP when treating adults experiencing an
acute exacerbation of schizophrenia. [1] National Institute of Mental Health. Schizophrenia. 2015. Available from:
http://www.nimh.nih.gov/health/publications/schizophrenia-booklet-12-
2015/index.shtml. Accessed February 21, 2017.
[2] American Psychiatric Association. Diagnostic and Statistical Manual of
Mental Disorders (5th ed). Arlington, VA: American Psychiatric
Acknowledgments Association, 2013.
[3] Lieberman JA, Malaspina D, Jarskog LF. Preventing clinical deterioration
We thank Maureen Rice (Cornerstone Research Group, Burling- in the course of schizophrenia: the potential for neuroprotection. CNS
ton, Ontario, Canada) for conducting the search. We also thank Spectr 2006;11(Suppl):1–13.
Lauren P. Baker (Catalysis Biomed LLC Seattle, Washington, [4] Buchanan RW, Kreyenbuhl J, Kelly DL, et al. Schizophrenia Patient
Outcomes Research Team (PORT). The 2009 schizophrenia PORT
United States) for editorial services and Cheryl Druchok (Corner-
psychopharmacological treatment recommendations and summary
stone Research Group, Burlington, Ontario, Canada) for assis- statements. Schizophr Bull 2010;36:71–93.
tance with manuscript revisions. We thank two anonymous [5] Nikolakopoulou A, Chaimani A, Veroniki AA, et al. Characteristics of
reviewers for helpful comments/suggestions on earlier drafts of networks of interventions: a description of a database of 186 published
networks. PLoS One 2014;9:e86754.
the manuscript.
[6] Caldwell DM, Ades AE, Higgins JP. Simultaneous comparison of
Source of financial support: This study was funded by multiple treatments: combining direct and indirect evidence. BMJ
Alkermes, Inc. (Waltham, MA). 2005;331:897–900.
10 VALUE IN HEALTH ] (2017) ]]]–]]]

[7] Leucht S, Cipriani A, Spineli L, et al. Comparative efficacy and [18] Kramer M, Litman R, Hough D, et al. Paliperidone palmitate, a potential
tolerability of 15 antipsychotic drugs in schizophrenia: a multiple- long-acting treatment for patients with schizophrenia: results of a
treatments meta-analysis. Lancet 2013;382:951–62. randomized, double-blind, placebo-controlled efficacy and safety
[8] Miura T, Noma H, Furukawa TA, et al. Comparative efficacy and study. Int J Neuropsychopharmacol 2010;13:635–47.
tolerability of pharmacological treatments in the maintenance [19] Hutton B, Salanti G, Caldwell DM, et al. The PRISMA extension
treatment of bipolar disorder: a systematic review and network meta- statement for reporting of systematic reviews incorporating network
analysis. Lancet Psychiatry 2014;1:351–9. meta-analyses of health care interventions: checklist and explanations.
[9] Cipriani A, Barbui C, Salanti G, et al. Comparative efficacy and Ann Intern Med 2015;162:777–84.
acceptability of antimanic drugs in acute mania: a multiple-treatments [20] Pandina GJ, Lindenmayer JP, Lull J, et al. A randomized, placebo-
meta-analysis. Lancet 2011;378:1306–15. controlled study to assess the efficacy and safety of 3 doses of
[10] Dias S, Sutton AJ, Ades AE, Welton NJ. Evidence synthesis for decision paliperidone palmitate in adults with acutely exacerbated
making 2: a generalized linear modeling framework for pairwise and schizophrenia. J Clin Psychopharmacol 2010;30:235–44.
network meta-analysis of randomized controlled trials. Med Decis [21] Nasrallah HA, Gopal S, Gassmann-Mayer C, et al. A controlled,
Making 2013;33:607–17. evidence-based trial of paliperidone palmitate, a long-acting injectable
[11] Dias S, Welton NJ, Sutton AJ, Ades AE. Evidence synthesis for decision antipsychotic, in schizophrenia. Neuropsychopharmacology
making, 1: introduction. Med Decis Making 2013;33:597–606. 2010;35:2072–82.
[12] Dias S, Welton NJ, Sutton AJ, et al. Evidence synthesis for decision [22] Gopal S, Hough DW, Xu H, et al. Efficacy and safety of paliperidone
making, 4: inconsistency in networks of evidence based on randomized palmitate in adult patients with acutely symptomatic schizophrenia:
controlled trials. Med Decis Making 2013;33:641–56. a randomized, double-blind, placebo-controlled, dose-response study.
[13] Jansen JP, Naci H. Is network meta-analysis as valid as standard Int Clin Psychopharmacol 2010;25:247–56.
pairwise meta-analysis? It all depends on the distribution of effect [23] Alphs L, Schooler N, Lauriello J. How study designs influence
modifiers. BMC Med 2013;11:159. comparative effectiveness outcomes: the case of oral versus long-
[14] Meltzer HY, Risinger R, Nasrallah HA, et al. A randomized, double- acting injectable antipsychotic treatments for schizophrenia.
blind, placebo-controlled trial of aripiprazole lauroxil in acute Schizophr Res 2014;156:228–32.
exacerbation of schizophrenia. J Clin Psychiatry 2015;76:1085–90. [24] Kane JM, Kishimoto T, Correll CU. Assessing the comparative
[15] Canadian Agency for Drugs and Technologies in Health. PRESS—Peer effectiveness of long-acting injectable vs. oral antipsychotic
Review of Electronic Search Strategies: 2015 Guideline Explanation and medications in the prevention of relapse provides a case study in
Elaboration (PRESS E&E). Ottawa, ON: Canadian Agency for Drugs and comparative effectiveness research in psychiatry. J Clin Epidemiol
Technologies in Health, 2016. 2013;66:S37–41.
[16] Salanti G, Ades AE, Ioannidis JP. Graphical methods and numerical [25] Leucht C, Heres S, Kane JM, et al. Oral versus depot antipsychotic drugs
summaries for presenting results from multiple-treatment meta- for schizophrenia—a critical systematic review and meta-analysis of
analysis: an overview and tutorial. J Clin Epidemiol 2011;64:163–71. randomised long-term trials. Schizophr Res 2011;127:83–92.
[17] Furukawa TA, Levine SZ, Tanaka S, et al. Initial severity of schizophrenia [26] Song F, Altman DG, Glenny AM, Deeks JJ. Validity of indirect comparison
and efficacy of antipsychotics: participant-level meta-analysis of 6 for estimating efficacy of competing interventions: empirical evidence
placebo-controlled studies. JAMA Psychiatry 2015;72:14–21. from published meta-analyses. BMJ 2003;326:472.