pharmacoepidemiology and drug safety 2014; 23: 299–308

Published online 24 January 2014 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.3568

ORIGINAL REPORT

Comparative effectiveness of monotherapy with mood stabilizers

versus second generation (atypical) antipsychotics for the treatment
of bipolar disorder in children and adolescents†
Hua Chen1*, Sonam Mehta1, Rajender Aparasu1, Ayush Patel1 and Melissa Ochoa-Perez2
1
University of Houston College of Pharmacy, Houston, TX, USA
2
Legacy Community Health Services, Houston, TX, USA

ABSTRACT
Objective This study compared the effectiveness and safety of second generation (atypical) antipsychotic (SGA) versus traditional mood
stabilizers (MS) in children and adolescents with bipolar disorder.
Methods The study was a retrospective cohort study on 5 years (2003–2007) of Medicaid claims data from four geographically diversified
states. Children and adolescents aged 6–18 years who initiated a new treatment episode for bipolar disorder on either an SGA or an MS were
followed for 12 months to compare the effectiveness and safety between the two therapeutic categories for pediatric bipolar disorder (PBD).
The outcome measures were psychiatric hospital admission, all cause medication discontinuation and treatment augmentation. Potential
selection bias caused by unobserved confounding was addressed with instrumental variable methods, using physician prescribing preference
and year of cohort entry as the instruments. Sensitivity analysis was conducted to test the robustness of findings against the uncertainties on
PBD diagnosis.
Results Of the 7423 bipolar children and adolescents identified, 66.60% started treatment on SGA, whereas 33.40% initiated on MS.
Patients who initiated on MS and SGA had comparable risk of psychiatric hospital admission (HR = 1.172, 95%CI: 0.827–1.660). However,
as compared with those who initiated on MS, patients who initiated on SGA were less likely to discontinue the treatment (HR = 0.634, 95%
CI: 0.419–0.961) and less likely to receive treatment augmentation (HR = 0.223, 95%CI: 0.103–0.484).
Conclusion As compared with MS monotherapy, SGA monotherapy could be a more effective and safer treatment option for PBD.
Copyright © 2014 John Wiley & Sons, Ltd.

key words—mood stabilizers; atypical antipsychotics; pediatric bipolar disorder; comparative effectiveness; pharmacoepidemiology

Received 2 May 2013; Revised 3 December 2013; Accepted 12 December 2013

INTRODUCTION Parameters for treatment of early-onset bipolar disor-

Bipolar disorder is one the most devastating of all
psychiatric disorders in terms of risk for suicide, need
for hospitalization and suffering. The treatment of
adolescent or juvenile patients with bipolar disorder
is modeled after treatments provided to adults. The
Child and Adolescent Bipolar Foundation expert
consensus treatment guidelines and the American
Academy of Child and Adolescent Psychiatry Practice
*Correspondence to: H. Chen, Department of Pharmaceutical Health Outcomes
and Policy, College of Pharmacy, University of Houston, 1441 Moursund St.,
Houston, TX 77030-3407, USA. E-mail: hchen25@Central.UH.EDU
†
The manuscript is based on an unfunded research. The abstract of this manu-
th
script was presented at the AACAP 59 annual meeting on 25 Oct 2012 at
San Francisco.
der indicate medications as the central components of
first-line intervention for pediatric bipolar disorder
(PBD).1,2 Monotherapy with either mood stabilizers
(MS) or second generation (atypical) antipsychotic
medications (SGA) is recommended as the first-line
treatment for pediatric bipolar I disorder.
Despite of the similar guideline status, SGA have
gained its popularities over MS in recent years.3,4
The preference is supported by emerging data from
child and adolescent studies.5–8 There were three
randomized double blind clinical trials (RCTs) and
one medical chart review that compared the effective-
ness and safety of valproate versus SGA (risperidone,
quatiapine) in pediatric samples with bipolar disorders.
Both the RCTs and the medical chart review showed

Copyright © 2014 John Wiley & Sons, Ltd.

300 h. chen et al.
that SGA were associated with more rapid onset of
effect, and of these, three reported that SGA were
also associated with greater manic symptom reduction
than valproate.
Bipolar disorder is a life-long disease associated
with frequent recurrences that requires long-term
management using pharmaceuticals.9 The existing stud-
ies, however, all had sample sizes of 50 or less and short
follow-up periods ranging from 4 to 12 weeks. Only
patients who met diagnostic criteria for acute mania
were included in these studies despite that these drugs
are also used in the treatment of other subtypes of
bipolar disorder.10 The short follow-ups and small
patient samples do not allow sufficient time and scope
to assess the magnitude of drug effect in long-term bipo-
lar management in real world pediatric populations.
Therefore, the objective of this study was to determine
the long-term comparative effectiveness and safety of
MS versus SGA for the treatment of PBD.
METHODS
Data source
The 2003 to 2007 Medicaid Analytic eXtract (MAX)
requested from the Center for Medicare and Medicaid
Services was used. The MAX data include payments
and utilization for services provided under fee-for-service
arrangements in the Medicaid and Medicaid expansion
State Children’s Health Insurance Program. In this
study, data from four states, namely Texas, New York,
California and Illinois, were selected because of avail-
ability of large pediatric population, and geographic
and demographic diversity.
Pharmacotherapy for bipolar disorder
The pharmacotherapy of bipolar disorder that was
investigated in the study included atypical antipsychotics
(risperidone, aripiprazole, olanzapine, quetiapine,
clozapine and ziprasidone) and MS (lithium, sodium
divalproex/valproate, carbamazepine, oxcarbazepine,
lamotrigine and gabapentin). Of the MS included in the
analysis, carbamazepine, gabapentin, oxcarbazepine and
lamotrigine are not officially approved by the FDA for
use in bipolar disorder. These medications were included
in the analysis because their use in bipolar disorder has
been studied and published in the medical literature.
Study population
Children and adolescents with bipolar disorder were
identified based on the following algorithm: (1) patients
aged between 6 and 18 years (children and adolescents);
(2) those who received a minimum of two diagnoses of
Copyright © 2014 John Wiley & Sons, Ltd.
bipolar disorder (International Classification of Dis-
eases, 9th version, Clinical Modification [ICD-9-CM]:
(i) manic/mixed type: 296.0, 296.1, 296.4 or 296.6x;
(ii) depressive type: 296.5; and (iii) unspecified: 296.7
or 296.8) on different service dates between 2003 and
2007; and (3) filled at least one prescription for an MS
or SGA during the same period. We then constructed
18-month episodes of bipolar care. Each episode was
identified by a paid claim for an MS or an SGA prescrip-
tion and a bipolar diagnosis within 30 days of the pre-
scription date. On the basis of index medications,
patients were categorized to either MS or SGA group,
and the class effect of SGA was compared with MS.
We defined a 6-month pretreatment period prior to this
index prescription, during which there were neither
evidence of pharmacotherapy for bipolar disorder nor
an indicator of other bipolar-related care. The second
12-month period during each 18-month episode defined
the follow-up period, during which we looked for
indicators of treatment failure. Patients who received
diagnoses of schizophrenia (ICD-9-CM: 295.0–295.9)
or epilepsy (ICD-9-CM: 345.xx) were excluded from
the cohort to make sure that the prescriptions were used
for bipolar disorder.
Measurements and outcomes
Psychiatric hospital admission, all cause medication
discontinuation and augmentation were selected as
the indexes of diminishing in drug effectiveness. We
define SGA or MS discontinuation for any cause as a
gap of ≥30 days in the prescription record of the index
medication. A medication gap can have resulted from
either discontinuing the index medication or switching
to another drug for bipolar disorder. Psychiatric
hospital admission, which is a marker of treatment
nonresponse or clinical relapse, is defined as any inpa-
tient treatment episode with a primary discharge diag-
nosis of a mental disorder (ICD-9-CM: 290–319)
following the first use of the SGA or MS medica-
tion. All cause medication discontinuation integrates
physician, patient and family perceptions of medi-
cation efficacy, tolerability and safety into a single
global, clinically meaningful effectiveness mea-
sure.11,12 Psychiatric hospital admission is also a
clinically important and financially significant event
that commonly indicates that the patient’s symp-
toms cannot be safely managed outside of an insti-
tutional setting.
Although discontinuation and psychiatric hospital
admission are well-defined indicators of treatment
failure, these measures may not be sensitive enough
to catch subtle variations in patients’ response to
Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds
 mood stabilizers vs. atypical antipsychotics in bipolar youth
treatments. Literature suggests that treatment partial
response is more common than nonresponse in the
course of bipolar management. Approximately 50%
of adults and adolescents with mania do not fully
respond to treatment with a single MS,13and augmen-
tation with another agent/combination therapy is often
necessary.1,13 Therefore, treatment augmentation,
defined as the addition of an antidepressant, antipsy-
chotic or an MS other than the index medication
before the treatment discontinuation was also selected
as an outcome variable.
Statistical analysis
Cox proportional hazards analysis was conducted to
assess the association between the treatments (SGA
vs. MS) and the time to events. Patients were censored
if they had a medical hospitalization or reach the end
of the study. Covariates for adjustment in the model
included patient demographics (age, gender, race,
geographic location by state and Medicaid eligibility
[foster care, poverty and disability]), bipolar subtypes
(manic type, mixed type, depressive type and
unspecified) and psychotherapy presented during the
6 months pre-index period and common comorbidities
of PBD including attention deficit hyperactivity
disorder (ADHD), oppositional defiant disorder
(ODD), conduct disorder, anxiety disorders and sub-
stance use disorder. Because of the competing nature
of treatment discontinuation and augmentation, the
analysis on time to augmentation was only conducted
in patients who continued the initial treatment during
the follow-up period.
Instrumental variable analysis. Mood stabilizers and
SGA are the first-line treatment recommend by treat-
ment guidelines for PBD. In the absence of perfect
information on why a given individual received one
drug versus another, we cannot exclude the possibility
that the treatment selection is relevant to a patient’s
illness profile, such as family history and disease
severity, for which accurate measurements could be
difficult to obtain from the claims data. Instrumental
variable (IV) analysis is an econometric method used
to remove the effects of hidden bias in observational
studies. The analysis requires variables that can be
used as “instruments,” such that the instruments
meet the following criteria: The instrument is statisti-
cally correlated with the treatment, and the instrument
does not cause the outcome except through the
treatment variable.
Copyright © 2014 John Wiley & Sons, Ltd.
301
In this study, we identified two variables that can be
used as instruments, namely “physician prescribing
preference” and “year of cohort entry.” Physician
prescribing preference was operationalized as the
treatment prescribed to the previous eligible patient
of the same physician.14 According to this approach,
if the last new prescription written by a physician to
a bipolar child was for an MS, then for the next
patient, the physician was classified as an “MS
prescriber.” Otherwise, he was classified as an “SGA
prescriber.” “Year of cohort entry” was defined as
the year that the index medication was initiated. These
two variables had been established in the literature as
appropriate instruments of treatment administration
for psychotropic drugs.14–16
Apart from these conceptual justifications and vali-
dation of these instruments from the past literature,
we also performed specification tests to validate the
instruments. First, the predictive power of the IVs on
the treatment variable was tested by running a regres-
sion of the treatment variable on the instruments and
other covariates in the model. Second, the F-statistic
for joint significance of the instruments in first-stage
regression was calculated. The statistic is one of the
commonly used diagnostics for weak instruments.
Third, the endogeneity test was performed to establish
the presence of unobserved confounding.
The IV analysis that was used in this study is the
“two-stage residual inclusion” (2SRI) estimation.17
In the first step of the 2SRI model, a linear proba-
bility analysis was conducted to estimate the
predicted value of treatment assignment using the
two IVs and the covariates. The predicted value of
treatment assignment through this model was
further used to calculate the residuals. In the second
step, a Cox proportional hazard regression of the
outcome variable was performed on the original
treatment variable, all the covariates (not including
the two IVs) and the residuals estimated from the
first step linear probability model. Boot-strapping
was used to estimate standard errors for the
treatment effect because the original standard errors
were unreliable because of the inclusion of an
estimated residual.
Sensitivity analysis. One of the major concerns on the
research of PBD is that mental health practitioners
disagree about the symptoms of bipolar disorder in
youth and how they differ from those in adults. In
adolescents, where the presentation of the disorder
more closely resembles that seen in adults, there is
much less controversy. To test the robustness of
Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds
302 h. chen et al.
findings against these uncertainties on PBD diagnosis,
the analysis on time to psychiatric hospital admission,
time to discontinuation and time to augmentation
was also conducted in a subgroup of adolescents (aged
12 years or older).
SAS version 9.2 (SAS institute Inc., Cary, NC,
USA) was used for analyses. This study was
approved by the Institutional Review Board of the
University of Houston.
Figure 1. Cohort assembly
Copyright © 2014 John Wiley & Sons, Ltd.
RESULTS
Patient characteristics
As presented in Figure 1, a total of 7423 bipolar
children and adolescents initiated the treatment on
either MS or SGA were identified after applying the
inclusion and exclusion criteria. The most frequently
prescribed SGA were risperidone, quetiapine and
aripiprazole. The most frequently prescribed MS were
divalproex sodium, oxcarbazepine and lithium.
Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds
 mood stabilizers vs. atypical antipsychotics in bipolar youth
The mean age of the sample was 12.73 ± 3.38 years;
57% of the patients were adolescents (>12 years of
age), and 54% were males. Whites comprised about
46% of the study population and around 41% of the
patients belonged to Texas. Approximately 17% of
the patients were in foster care. Use of psychotherapy
during the pre-index period was observed in 40% of
the patients. ADHD was the most common comorbid-
ity that was present in 35% of the patients.
Descriptive analysis
Of the 7423 patients identified, 4944 (66.60%)
initiated on SGA, whereas 2479 (33.40%) initiated
on MS. Table 1 presents the distribution of treatment
groups by patient demographics and comorbidities.
SGA recipients were significantly different from MS
recipients in all variables except for the likelihood of
having some comorbid mental disorders such as
oppressive compulsive disorder, conduct disorder and
substance use disorder. SGA were used more often in
male than in female, in children than in adolescents
and in New York than in other states, and less used
in Whites than non-Whites. SGA recipients were more
303
likely to have comorbid ADHD and ODD, and less
likely to receive psychotherapy compared with the
MS recipients.
The median time to hospitalization was 187 days for
patients initiated on SGA and 186 days for those initi-
ated on MS. The median days for time to discontinua-
tion were 147 days for the MS recipients and 163 days
for the SGA recipients. Among those who continued
the treatments and had no psychiatric hospital admis-
sion during the 1-year follow-up period, 15% of the
patients initiated on MS, and 10.92% of the patients
initiated on SGA received augmentation with other
psychotropic drugs. The median time to augmentation
was 102 days for the SGA recipients and 95 days for
the MS recipients.
Instrument variable analysis
According to the operation definition of physician
prescribing preference, the chronologically first
patient for each physician was excluded so that each
patient had a physician prescribing preference
defined. As a result, 2227 (30%) out of 7423 patients
were excluded.

Table 1. Descriptive statistics of patient demographics and comorbidities (*: p < 0.05)
MS recipients (%N) SGA recipients (%N)
Variable p value
N = 7423 2479 4944

Age 6–12 years (3131) 877 (28.00%) 2254 (72.00%) <0.0001*

13–18 years (4292) 1602 (37.33%) 2690 (62.67%)
Gender Males (4012) 1173 (29.24%) 2839 (70.76%) <0.0001*
Females (3411) 1306 (38.29%) 2105 (61.71%)
Race Whites (3351) 1198 (35.76%) 2153 (64.24%) 0.0270*
Blacks (1940) 613 (31.58%) 1327 (68.42%)
Others (2131) 668 (31.35%) 1463 (68.65%)
State California (1349) 460 (34.10%) 889 (65.90) <0.0001*
Illinois (2280) 910 (39.91%) 1370 (60.09%)
New York (715) 197 (27.60%) 518 (72.40%)
Texas (3080) 912 (29.61%) 2168 (70.39%)
Type of index bipolar diagnosis Unspecified (4114) 1289 (31.34%) 2825 (68.66%) 0.0060*
Mania (929) 335 (36.07%) 594 (63.93%)
Depression (817) 261 (31.95%) 556 (68.05%)
Mixed (1563) 594 (37.99%) 969 (62.01%)
Type of Medicaid eligibility Disability (2152) 600 (24.21%) 1552 (31.40%) <0.0001*
Poverty (3504) 1247 (50.30%) 2257 (45.65%)
Foster care (1767) 632 (24.49%) 1135 (22.95%)
Psychotherapy Yes (2991) 1078 (36.03%) 1913 (63.97%) 0.0068*
No (4432) 1401 (31.61%) 3031 (68.39%)
Attention deficit hyperactive disorder Yes (2579) 755 (29.28%) 1824 (70.72%) 0.0002*
No (4844) 1724 (35.60%) 3120 (64.40%)
Oppressive compulsive disorder Yes (717) 225 (31.36%) 492 (68.64%) 0.403
No (6706) 2254 (33.61%) 4452 (66.39%)
Conduct disorder Yes (1094) 405 (37.02%) 689 (62.98%) 0.0593
No (6329) 2074 (32.77%) 4255 (67.23%)
Anxiety disorder Yes (965) 371 (38.46%) 594 (61.54%) 0.0141
No (6458) 2108 (32.64%) 4350 (67.36%)
Substance use disorder Yes (394) 144 (36.56%) 250 (63.44%) 0.3478
No (7029) 2335 (33.22%) 4694 (66.78%)

Copyright © 2014 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds
304
Table 2 presents the first-stage linear probability
model for the selection between SGA and MS as
a function of the IVs and all other covariates. The
two IVs are the strongest predictor of treatment
selections. The IV coefficients are of expected
signs; physicians who previously prescribed SGA
to children and adolescents with bipolar disorder
were at a higher chance of prescribing SGA again
to the next patient with PBD. Likewise, the more
recent pediatric bipolar patients were treated, the
higher the chances for these patients to receive
SGA. The F-statistic for the joint significant of the
IVs is 127. The diagnostic is well above the widely
used rule of thumb (i.e., 10) suggested by Staiger
and Stock—leading to a rejection of the null
hypothesis of weak instrumentation.
Tables 3–5 present the results of Cox propor-
tional hazard analysis on outcomes with and with-
out using IVs. The existence of unobserved
confounding (endogeneity) was tested in this step
(second stage) of the IV analysis. If the hazard ratio
of the residual estimated from the first-stage linear
probability model is not statistically significant, this
is taken as a suggestion that the selection between
SGA and MS in the study population is not affected
by unobserved confounding, and then the Cox
estimator without using IVs will be more efficient,
whereas a statistically significant hazard ratio for
h. chen et al.
the residual confirms the existence of endogeniety
and the need of analyses using IV.
Results of the second-stage IV analysis for the
outcomes of interest show that the residual was not
statistically significant for the analysis of time to
hospitalization, but it was significant in the analysis
of time to discontinuation and augmentation.
Therefore, for the analysis of time to hospitalization,
the findings of the traditional Cox model were
interpreted, whereas for time to discontinuation and
time to augmentation, findings of the IV analysis were
interpreted.
As presented in Tables 3–5, children and adoles-
cents with bipolar disorder initiated on MS and SGA
had comparable risk of psychiatric hospital admission
(HR = 1.172, 95%CI: 0.827–1.660). As compared with
those who initiated on MS, patients who initiated on
SGA were less likely to discontinue the treatment
(HR = 0.634, 95%CI: 0.419–0.961) and less likely to
receive treatment augmentation (HR = 0.223, 95%CI:
0.103–0.484).
Sensitivity analysis
As presented in Tables 3–5, sensitivity analysis on
a subcohort of adolescents led to results that were
consistent with the findings of the main analyses.
For adolescents 12 years and older, the risk of

Table 2. First-stage linear probability model for the selection between second generation (atypical) antipsychotics and mood stabilizers (*: p < 0.05)
Variables Parameter estimates (p value)

Gender Male Reference

Female 0.063 (<0.0001)*
Age Children (6–12 years) Reference
Adolescents (13–18 years) 0.066 (<0.0001)*
Race Others Reference
Whites 0.012 (0.534)
Blacks 0.029 (0.170)
State Texas Reference
California 0.042 (0.074)
Illinois 0.057 (0.004)*
New York 0.011 (0.686)
Type of index bipolar disorder Unspecified Reference
Mania 0.027 (0.258)
Depression 0.008 (0.757)
Mixed 0.03 (0.124)
Year of cohort entry 0.035 (<0.0001)*
Medicaid eligibility Disability Reference
Poverty 0.028 (0.137)
Foster 0.030 (0.170)
Attention deficit hyperactive disorder Yes vs. no 0.020 (0.233)
Oppressive compulsive disorder Yes vs. no 0.014 (0.576)
Conduct disorder Yes vs. no 0.035 (0.105)
Anxiety disorder Yes vs. no 0.015 (0.499)
Substance use disorder Yes vs. no 0.044 (0.191)
Psychotherapy Yes vs. no 0.049 (0.005)*
Physician preference Mood stabilizer Reference
Atypical antipsychotic 0.318 (<0.0001)*

Copyright © 2014 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds
 mood stabilizers vs. atypical antipsychotics in bipolar youth 305
Table 3. Survival analysis on psychiatric hospital admission (*: p < 0.05)
Hazard ratio (95%CI)

Main analysis—all Main analysis—all Sensitivity analysis— Sensitivity analysis—

Variables patients (without IV) patients (with IV) adolescents (without IV) adolescents (with IV)

Index medication (SGA vs. MS) 1.172 (0.827–1.660) 0.455 (0.164–1.261) 1.168 (0.772–1.766) 0.368 (0.094–1.436)
Age (adolescents vs. children) 1.308 (0.905–1.891) 1.236 (0.825–1.854) – –
Gender (female vs. male) 1.627 (1.156–2.290)* 1.543 (1.069–2.226)* 1.595 (1.047–2.428)* 1.474 (0.938–2.318)
Race
Whites vs. others 0.827 (0.555–1.230) 0.832 (0.551–1.259) 0.947 (0.574–1.561) 0.974 (0.576–1.646)
Blacks vs. others 0.703 (0.451–1.096) 0.708 (0.448–1.119) 0.878 (0.509–1.514) 0.884 (0.514–1.521)
Medicaid eligibility
Foster care vs. disability 0.733 (0.463–1.162) 0.693 (0.412–1.165) 0.662 (0.376–1.166) 0.616 (0.330–1.151)
Poverty vs. disability 0.693 (0.460–1.046) 0.673 (0.437–1.035) 0.716 (0.436–1.174) 0.698 (0.402–1.211)
State Medicaid program
California vs. Texas 0.626 (0.345–1.138) 0.609 (0.314–1.182) 0.741 (0.371–1.483) 0.712 (0.336–1.506)
Illinois vs. Texas 1.384 (0.933–2.051) 1.344 (0.897–2.013) 1.306 (0.812–2.101) 1.223 (0.744–2.008)
New York vs. Texas 0.537 (0.251–1.146) 0.495 (0.213–1.148) 0.510 (0.196–1.328) 0.383 (0.026–5.729)
Type of index bipolar disorder
Mania vs. unspecified 0.939 (0.553–1.594) 0.817 (0.522–1.280) 0.886 (0.470–1.671) 0.92 (0.537–1.575)
Depression vs. unspecified 1.216 (0.749–1.972) 0.890 (0.511–1.552) 1.129 (0.627–2.033) 0.843 (0.424–1.674)
Mixed vs. unspecified 0.868 (0.567–1.329) 1.171 (0.702–1.953) 0.985 (0.594–1.635) 1.080 (0.585–1.993)
Comorbidities
Attention deficit hyperactive disorder 0.88 (0.602–1.285) 0.892 (0.591–1.347) 1.377 (0.876–2.163) 1.384 (0.864–2.219)
Oppressive compulsive disorder 1.532 (0.967–2.428) 1.567 (0.959–2.561) 1.731 (1.008–2.97)* 1.747 (0.979–3.119)
Conduct disorder 0.998 (0.629–1.581) 0.935 (0.593–1.475) 0.978 (0.559–1.711) 0.868 (0.480–1.570)
Anxiety disorder 0.949 (0.594–1.517) 0.885 (0.546–1.435) 0.786 (0.434–1.425) 0.731 (0.378–1.417)
Substance abuse disorder 1.551 (0.889–2.707) 1.553 (0.827–2.920) 1.649 (0.940–2.894) 1.689 (0.878–3.250)
Psychotherapy 1.247 (0.875–1.778) 1.193 (0.813–1.75) 1.318 (0.851–2.04) 1.234 (0.79–1.928)
Residual – 2.925 (0.953–8.985) – 3.705 (0.833–16.477)

IV, instrumental variable; MS, mood stabilizer; SGA, second generation (atypical) antipsychotic.

Table 4. Survival analysis on time to discontinuation (*: p < 0.05)

Hazard ratio (95%CI)

Main analysis—all Main analysis—all Sensitivity analysis— Sensitivity analysis—

Variables patients (without IV) patients (with IV) adolescents (without IV) adolescents (with IV)

Index medication (SGA vs. MS) 0.998 (0.888–1.122) 0.634 (0.419–0.961)* 1.027 (0.883–1.194) 0.454 (0.251–0.822)*
Age (adolescents vs. children) 0.981 (0.872–1.103) 0.951 (0.842–1.073) – –
Gender (female vs. male) 0.962 (0.859–1.078) 0.932 (0.825–1.052) 0.954 (0.821–1.109) 0.897 (0.755–1.064)
Race
Whites vs. others 0.882 (0.770–1.01) 0.884 (0.771–1.012) 1.041 (0.866–1.253) 1.049 (0.872–1.263)
Blacks vs. others 1.112 (0.960–1.287) 1.115 (0.958–1.298) 1.189 (0.971–1.457) 1.202 (0.968–1.493)
Medicaid eligibility
Foster care vs. disability 0.662 (0.376–1.166) 0.518 (0.436–0.615) 0.640 (0.518–0.791) 0.605 (0.487–0.752)
Poverty vs. disability 0.716 (0.436–1.174) 0.728 (0.638–0.831) 0.799 (0.670–0.952) 0.779 (0.653–0.929)
State Medicaid program
California vs. Texas 1.138 (0.960–1.350) 1.131 (0.962–1.33) 1.316 (1.054–1.643)* 1.293 (1.019–1.64)*
Illinois vs. Texas 1.044 (0.906–1.202) 1.021 (0.883–1.180) 1.037 (0.860–1.251) 0.985 (0.802–1.208)
New York vs. Texas 1.195 (0.981–1.456) 1.174 (0.965–1.427) 1.253 (0.963–1.631) 1.17 (0.888–1.543)
Type of index bipolar disorder
Mania vs. unspecified 1.032 (0.87–1.223) 1.02 (0.857–1.213) 1.024 (0.827–1.268) 1.006 (0.811–1.247)
Depression vs. unspecified 1.00 (0.834–1.198) 0.988 (0.819–1.193) 1.063 (0.847–1.333) 1.038 (0.821–1.313)
Mixed vs. unspecified 1.043 (0.908–1.198) 1.018 (0.882–1.176) 0.992 (0.820–1.200) 0.946 (0.768–1.165)
Comorbidities
Attention deficit hyperactive disorder 1.038 (0.922–1.17) 1.039 (0.919–1.174) 1.173 (0.992–1.387) 1.169 (0.984–1.389)
Oppressive compulsive disorder 0.981 (0.816–1.179) 0.999 (0.831–1.202) 0.949 (0.739–1.219) 0.974 (0.75–1.264)
Conduct disorder 0.978 (0.837–1.143) 0.951 (0.812–1.114) 1.032 (0.838–1.273) 0.962 (0.777–1.192)
Anxiety disorder 0.925 (0.784–1.091) 0.904 (0.77–1.061) 0.937 (0.758–1.157) 0.894 (0.714–1.119)
Substance abuse disorder 1.160 (0.914–1.472) 1.174 (0.91–1.515) 1.143 (0.895–1.46) 1.176 (0.905–1.527)
Psychotherapy 1.055 (0.931–1.196) 1.031 (0.906–1.174) 1.095 (0.924–1.296) 1.042 (0.861–1.261)
Residual – 1.669 (1.087–2.565)* – 2.487 (1.34–4.613)*

IV, instrumental variable; MS, mood stabilizer; SGA, second generation (atypical) antipsychotic.

Copyright © 2014 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds
306 h. chen et al.
Table 5. Survival analysis on time to treatment augmentation (*: p < 0.05)
Hazard ratio (95%CI)

Main analysis—all Main analysis—all Sensitivity analysis— Sensitivity analysis—

Variables patients (without IV) patients (with IV) adolescents (without IV) adolescents (with IV)

Index medication (SGA vs. MS) 0.633 (0.490–0.817)* 0.223 (0.103–0.484)* 0.628 (0.442–0.892)* 0.257 (0.087–0.763)*
Age (adolescents vs. children) 0.713 (0.546–0.932)* 0.665 (0.505–0.876)* - -
Gender (female vs. male) 1.115 (0.860–1.446) 1.03 (0.783–1.357) 1.408 (0.966–2.052) 1.284 (0.868–1.900)
Race
Whites vs. others 1.261 (0.940–1.691) 1.27 (0.962–1.677) 1.582 (1.027–2.436)* 1.584 (1.044–2.402)*
Blacks vs. others 0.592 (0.400–0.876)* 0.589 (0.397–0.872)* 0.75 (0.427–1.317) 0.73 (0.415–1.284)
Medicaid eligibility
Foster care vs. disability 0.571 (0.412–0.79) 0.552 (0.397–0.766) 0.746 (0.47–1.182) 0.735 (0.463–1.170)
Poverty vs. disability 1.056 (0.761–1.466) 1.001 (0.708–1.414) 1.196 (0.736–1.943) 1.139 (0.679–1.910)
State Medicaid program
California vs. Texas 0.715 (0.482–1.061) 0.705 (0.478–1.041) 0.795 (0.464–1.363) 0.784 (0.45–1.366)
Illinois vs. Texas 0.811 (0.584–1.128) 0.763 (0.543–1.074) 0.77 (0.489–1.211) 0.729 (0.449–1.183)
New York vs. Texas 1.032 (0.673–1.583) 0.962 (0.618–1.499) 1.257 (0.713–2.217) 1.147 (0.633–2.079)
Type of index bipolar disorder
Mania vs. unspecified 0.881 (0.599–1.296) 0.67 (0.461–0.973)* 0.787 (0.468–1.323) 0.572 (0.334–0.978)*
Depression vs. unspecified 0.771 (0.485–1.227) 0.85 (0.572–1.263) 0.779 (0.43–1.414) 0.754 (0.429–1.327)
Mixed vs. unspecified 0.716 (0.507–1.012) 0.727 (0.447–1.18) 0.620 (0.369–1.040) 0.712 (0.362–1.399)
Comorbidities
Attention deficit hyperactive disorder 0.88 (0.668–1.159) 0.897 (0.673–1.194) 1.067 (0.698–1.631) 1.071 (0.686–1.672)
Oppressive compulsive disorder 0.933 (0.600–1.45) 0.987 (0.617–1.58) 0.779 (0.414–1.466) 0.78 (0.380–1.599)
Conduct disorder 1.087 (0.765–1.545) 1.012 (0.705–1.453) 1.375 (0.850–2.223) 1.262 (0.756–2.107)
Anxiety disorder 0.919 (0.634–1.331) 0.882 (0.592–1.313) 0.751 (0.441–1.28) 0.705 (0.408–1.219)
Substance abuse disorder 1.21 (0.665–2.201) 1.215 (0.640–2.306) 1.269 (0.693–2.327) 1.311 (0.662–2.596)
Psychotherapy 0.998 (0.754–1.321) 0.951 (0.709–1.276) 0.923 (0.618–1.378) 0.893 (0.597–1.337)
Residual – 3.234 (1.426–7.334)* – 2.719 (0.883–8.371)

IV, instrumental variable; MS, mood stabilizer; SGA, second generation (atypical) antipsychotic.

psychotropic hospitalization was comparable be- Bipolar disorder is characterized by a cycling

tween the MS and SGA (HR = 1.168, 95%CI:
0.772–1.766). However, the MS recipients were
associated with higher risk of treatment discontinu-
ation (HR = 0.454, 95%CI: 0.251–0.822) and aug-
mentation (HR = 0.628, 0.442–0.892) as compared
with the SGA recipients.
DISCUSSIONS
Consistent with the findings of several short-term
pediatric trials for acute mania, our analysis suggests
that SGA might also be more effective and better toler-
ated than traditional MS on the maintenance stage of
the treatment for PBD. Although the two study groups
had comparable risk of hospital admission, children
and adolescents who initiated treatment on an
SGA were less likely to discontinue and also less
likely to receive regimen augmentation than those
who initiated treatment on an MS during the 12-month
follow-up period. The more favorable long-term
outcomes associated with SGA versus MS could be
explained by its better antidepressant effect and less
intensive monitoring requirement as compared with
the traditional MS.
between depression and mania. Depressive symp-
toms are often more persistent and debilitating.18
A 4-year longitudinal study of 86 youth diagnosed
with DSM-IV bipolar I disorder found that they
spent nearly half of the time meeting criteria for
major depression, minor depression or dysthymia.19
Another 2-year follow-up study of children and
adolescents with bipolar spectrum disorders (bipolar
I disorder; bipolar II disorder; and bipolar disorder,
not otherwise specified) showed that approximately
60% of the observation time, bipolar disorder
youths experienced syndromal and subsyndromal
bipolar disorder symptoms, particularly depressive
and mixed symptoms and repeated changes in
symptom polarity.18 Therefore, successful long-term
management of PBD requires a medication that
treats both mania and depression, without
neglecting or exacerbating one phase for the sake
of managing the other.
There is no traditional MS that possesses a similar
degree of efficacy in treating both the manic and
depressive phases of bipolar disorder. Most of
these drugs are purely antimanic agents except for
lamotrigine and lithium.2 However, lamotrigine is
only approved by FDA for use in those over the

Copyright © 2014 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds
 mood stabilizers vs. atypical antipsychotics in bipolar youth
age of 16 years. It is not commonly used in children
because of the increased risk of fatal side effects,
such as Stevens–Johnson syndrome in the young
age group. Lithium, despite with an indication for
PBD, was used less often in recent years because
of its narrow therapeutic windows and intensive
monitoring requirement. Moreover, literature shows
that diligent monitoring on serum levels and renal
function might also have caused a higher rate and
earlier discontinuation when lithium is used outside
of controlled trials.20
Although the antidepressant effect of antipsychotics
has not been tested in pediatric population, this effect
of antipsychotics has been confirmed by RCTs using
adult samples with unipolar depression. In 2008, the
FDA approved the use of aripiprazole in combination
with antidepressant medication for the treatment of
major depression in adults.21 In November, coinciding
with the FDA approval of this supplemental New
Drug Application for aripiprazole, an RCT demon-
strated increased antidepressant effect from the addi-
tion of risperidone to antidepressant monotherapy.22
The data specific to patients with bipolar depression
are very sparse, but two reports with olanzapine
involving a total of 18 adult patients found that 14
had positive response.23,24
In our study cohort, SGA were used predominantly
more than MS. Systematic reviews on pediatric data
suggest that SGA cause more weight gains than
MS in youth.25 Age-inappropriate weight gain is of
particular concern in children and adolescents due to
its association with glucose and lipid abnormalities
and cardiovascular morbidity/ mortality.26,27
Long-term comparative effectiveness studies are
necessary to justify the use of SGA in children and
adolescents. RCTs remain the gold standard for
developing such information; however, this study
design is not always feasible, practical or sufficiently
timely. The present findings may help extend and
complement previous short-term observations on the
effect of SGA and MS in RCTs. It provides evidence
that SGA not only had a better anti-manic effect than
MS in the first few weeks of treatment for PBD, but
also could have more favorable long term safety and
effectiveness profiles as compared to MS.
Limitations
A number of empirical limitations might have
affected the applicability of the study finding. First,
children and adolescents with bipolar disorder
Copyright © 2014 John Wiley & Sons, Ltd.
307
were identified solely based on diagnostic claims.
Although there is increasing consensus on the exis-
tence of PBD, debates regarding the boundary of
the disease continue. The absence of structured di-
agnostic evaluation of the subjects may have led
to misdiagnosis. The most controversy regarding
the diagnosis resides on the existence of the disease
in young age groups and the symptom overlapping
between ADHD and bipolar disorders in the young
age group. However, sensitivity analysis performed
in adolescents, a subcohort in whom bipolar diagno-
sis is more stable and less controversial, also
showed that SGA was associated with lower risk
of treatment discontinuation and augmentation than
MS. The consistent findings between the main
analysis and the sensitivity analysis suggest the
robustness of our findings against the uncertainties
on diagnosis. Second, only patients who initiated a
new treatment episode on either an SGA or an MS were
included in the analysis. Patients who had pre-index uti-
lization of SGA or MS and patients who had pre-index
psychiatric hospital admissions and emergency room
visits were excluded because of the lack of information
on inpatient drug utilization in the data. Excluding these
patients diminished the differences between the study
groups and ensured comparability. However, the
approach was also likely to exclude the severe cases
and therefore affect the generalizability of the study
findings. Lastly, because of the lack of clinical details
in claims data, we were not able to explore specific
reasons behind the variation of the outcome measure
used in the study. For instance, we cannot answer
whether treatment discontinuation was caused by ineffi-
cacy or intolerability, and whether the intolerability was
owing to weight gain, extrapyramidal signs or sedation.
More research is needed to understand the reasons of
discontinuation and develop strategies that help to
enhance continuity of treatment.
CONCLUSION
As compared with MS, SGA were associated with
longer time to discontinuation and augmentation in
the treatment of bipolar disorder in children and
adolescents. The finding supports the use of SGA
as initial treatment for PBD.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds
308 h. chen et al.
KEY POINTS
• Pediatric bipolar disorder is increasingly diag-
nosed and treated in children and adolescents.
Traditional mood stabilizers (MS) and second
generation antipsychotics (SGA) are the first-line
treatment recommended by treatment guidelines.
As compared with patients initiated on MS, those
who initiated on SGA were less likely to discon-
tinue the treatment and also less likely to receive
treatment augmentation during the 12-month
follow-up period. Therefore, as compared with
MS, SGA could be a more effective and safer
treatment option for pediatric bipolar disorder.
• Instrumental variable (IV) analysis may offer a
useful approach to the problem of unmeasured
confounding presented in comparative drug effec-
tiveness research using administrative claims data.
IV analyses have the potential to fill important
gaps in evidence-based medicine. As compared
with randomized clinical trials, observational
studies offer an expansion in generalizability to
under-represented populations in clinical trials,
such as children and adolescents.
ETHICS STATEMENT
The CMS data used in the study was granted by the
Center for Medicaid and Medicare services through
DUA#21522. The study was also approved by the
University of Houston Committees for the Protection
of Human Subjects.
REFERENCES
1. Kowatch RA, Fristad M, Birmaher B, Wagner KD, Findling RL, Hellander M.
Treatment guidelines for children and adolescents with bipolar disorder. J Am
Acad Child & Adolescent Psychiatry 2005; 44: 213–235.
2. McClellan J, Kowatch R, Findling RL. Practice parameter for the assessment and
treatment of children and adolescents with bipolar disorder. J Am Acad Child
Adolesc Psychiatry 2007; 46: 107–125.
3. Moreno C, Laje G, Blanco C, Jiang H, Schmidt AB, Olfson M. National trends in
the outpatient diagnosis and treatment of bipolar disorder in youth. Arch Gen
Psychiatry 2007; 64: 1032.
4. Potter MP, Liu HY, Monuteaux MC, et al. Prescribing patterns for treatment of
pediatric bipolar disorder in a specialty clinic. J Child Adolescent Psychophar-
macology 2009; 19: 529–538.
5. Macmillan CM, Withney JE, KorndÖRfer SR, Tilley CA, Mrakotsky C,
Gonzalez-Heydrich JM. Comparative clinical responses to risperidone and
Copyright © 2014 John Wiley & Sons, Ltd.
divalproex in patients with pediatric bipolar disorder. J Psychiatr Pract 2008;
14: 160–169.
6. Delbello MP, Kowatch RA, Adler CM, et al. A double-blind randomized pilot
study comparing quetiapine and divalproex for adolescent mania. J Am Acad
Child Adolesc Psychiatry 2006; 45: 305.
7. Pavuluri MN, Henry DB, Findling RL, et al. Double-blind randomized trial of
risperidone versus divalproex in pediatric bipolar disorder. Bipolar Disord
2010; 12: 593–605.
8. Geller B, Luby JL, Joshi P, et al. A randomized controlled trial of risperidone,
lithium, or divalproex sodium for initial treatment of bipolar I disorder, manic
or mixed phase, in children and adolescents. Arch Gen Psychiatry 2012; 69:
515–528.
9. Lin PI, McInnis MG, Potash JB, et al. Clinical correlates and familial aggrega-
tion of age at onset in bipolar disorder. Am J Psychiatr 2006; 163: 240–246.
10. Carlson GA, Jensen PS, Findling RL, et al. Methodological issues and controver-
sies in clinical trials with child and adolescent patients with bipolar disorder:
report of a consensus conference. J child Adolescent Psychopharmacology
2003; 13: 13–27.
11. Swartz MS, Perkins DO, Stroup TS, McEvoy JP, Nieri JM, Haak DC. Assessing
clinical and functional outcomes in the Clinical Antipsychotic Trials of Interven-
12.
tion Effectiveness (CATIE) schizophrenia trial. Schizophr Bull 2003; 29: 33.
Lieberman JA, Stroup TS, McEvoy JP, et al. Effectiveness of antipsychotic
drugs in patients with chronic schizophrenia. N Engl J Med 2005; 353:
1209–1223.
13. Kowatch RA, Sethuraman G, Hume JH, Kromelis M, Weinberg WA. Combina-
tion pharmacotherapy in children and adolescents with bipolar disorder. Biol
14.
Psychiatry 2003; 53: 978–984.
Brookhart MA, Wang P, Solomon DH, Schneeweiss S. Evaluating short-term
drug effects using a physician-specific prescribing preference as an instrumental
variable. Epidemiology (Cambridge, Mass) 2006; 17: 268.
15. Brookhart MA, Rassen JA, Wang PS, Dormuth C, Mogun H, Schneeweiss S.
Evaluating the validity of an instrumental variable study of neuroleptics: can
between-physician differences in prescribing patterns be used to estimate
treatment effects? Med Care 2007; 45: S116–S122.
16. Rassen JA, Brookhart MA, Glynn RJ, Mittleman MA, Schneeweiss S. Instru-
mental variables II: instrumental variable application—in 25 variations, the
physician prescribing preference generally was strong and reduced covariate
imbalance. J Clin Epidemiol 2009; 62: 1233.
17. Terza JV, Basu A, Rathouz PJ. Two-stage residual inclusion estimation: address-
ing endogeneity in health econometric modeling. J Health Econ 2008; 27: 531.
18. Birmaher B, Axelson D, Strober M, et al. Clinical course of children and adoles-
cents with bipolar spectrum disorders. Arch Gen Psychiatry 2006; 63: 175.
19. Wozniak J, Biederman J, Kwon A, et al. How cardinal are cardinal symptoms in
pediatric bipolar disorder? An examination of clinical correlates. Biol Psychiatry
2005; 58: 583–588.
20. Calabrese JR, Shelton MD, Rapport DJ, et al. A 20-month, double-blind, main-
tenance trial of lithium versus divalproex in rapid-cycling bipolar disorder. Am J
Psychiatr 2005; 162: 2152–2161.
21. US Food and Drug Administration approves Abilify (aripiprazole) as the first
medication for add-on treatment of major depressive disorder (MDD). November
20, 2007. (Accessed December 20, 2007, 2007, at http://newsroom.bms.com/in-
dex.php?s=press_releases&item=324.)
22. Mahmoud RA, Pandina GJ, Turkoz I, et al. Risperidone for treatment-refractory
major depressive disorder: a randomized trial. Ann Internal Med 2007; 147:
593–602.
23. Rothschild AJ, Bates KS, Boehringer KL, Syed A. Olanzapine response in
psychotic depression. J Clin Psychiatry 1999; 60.
24. Zarate Jr CA, Narendran R, Tohen M, et al. Clinical predictors of acute response
with olanzapine in psychotic mood disorders. J Clin Psychiatry 1998; 59: 24.
25. Correll CU, Sheridan EM, DelBello MP. Antipsychotic and mood stabilizer
efficacy and tolerability in pediatric and adult patients with bipolar I mania: a
comparative analysis of acute, randomized, placebo-controlled trials. Bipolar
Disord 2010; 12: 116–141.
26. Freedman DS, Khan LK, Serdula MK, Dietz WH, Srinivasan SR, Berenson GS.
The relation of childhood BMI to adult adiposity: the Bogalusa Heart Study.
Pediatrics 2005; 115: 22–27.
27. Bhargava SK, Sachdev HS, Fall CHD, et al. Relation of serial changes in
childhood body-mass index to impaired glucose tolerance in young adulthood.
N Engl J Med 2004; 350: 865–875.
Pharmacoepidemiology and Drug Safety, 2014; 23: 299–308
DOI: 10.1002/pds