University of Southern California Dissertations and Theses

Assessment of the impact of second-generation antipscyhotics in Medi-Cal patients with bipolar disorder using panel data fixed effect models

(USC Thesis Other)

Assessment of the impact of second-generation antipscyhotics in Medi-Cal patients with bipolar disorder using panel data fixed effect models

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

ASSESSMENT OF THE IMPACT OF SECOND-GENERATION
ANTIPSCYHOTICS IN MEDI-CAL PATIENTS WITH BIPOLAR DISORDER
USING PANEL DATA FIXED EFFECT MODELS

by

Jinhee Park

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PHARMACEUTICAL ECONOMICS AND POLICY)

August 2009

Copyright 2009 Jinhee Park

ii
Acknowledgments

It has been a long journey and I am grateful to a number of people who helped me
complete the journey. First of all, I am deeply grateful for the tremendous support and
guidance of my graduate adviser, Dr. Jeff McCombs. I would like to express my
gratitude to the other committee members, Dr. Jeonghoon Ahn and Dr. Cheng Hsiao,
for their guidance and valuable suggestions on my research. I also would like to thank
other faculty members at the Department of Pharmaceutical Economics and Policy for
their guidance and exceptional graduate education they provided me and fellow
graduate students for their encouragement and friendship. I would like to
acknowledge my coworkers at GlaxoSmithKline for their support and friendship. I
particularly want to thank Ron Cantrell and Julie Priest for their encouragement and
support. Most of all, I would like to thank my family - my husband Jongwon and my
daughter Seungyeon who gave me strength to go on through the toughest times for
their love and patience. My sister and best friend Hyunju and my mom have been the
greatest supporter throughout my graduate studies and I wouldn’t have done it without
their love and help. Lastly I want to thank my late father. He saw me walk at the
graduation commencement and even though he didn’t say it, I could tell how proud he
was of his daughter. Dad, this is the end of the journey and I’m finally here.

iii

Table of Contents

Acknowledgements ii

List of Tables v

Abstract ix

Chapter 1: Introduction 1
Clinical Background 2
Definition of Disease 2
Classification of Bipolar Disorder 3
Epidemiology and Course of Bipolar Disorder 4
Psychotic Symptoms 5
Drug Therapy for Bipolar Disorder 6
Acute Treatment 6
Maintenance Treatment 7
Limitation of Current Pharmacotherapy 8
Efficacy of Antipsychotics in Bipolar Disorder 9
Long-term Treatment with Antipsychotics 14
Outcomes of Drug Therapy in Patient with Bipolar Disorder 15
Risk of Suicide 15
Hospitalization 16
Diabetes 17
Social and Economics Impact of Bipolar Disorder 20
Treatment Effect of Second Generation Antipsychotics 23
Drug Therapy in the Medi-Cal population 25
Economic Treatment Effect of Second Generation Antipsychotic
Therapy in Medi-Cal Patients with Bipolar Disorder 27
Use of Claims Database in Outcomes Research 29
Self-Selection Problem in Claims Database 30
Econometric Tools for Estimating Unbiased Treatment Effect 31
Panel Data Approach 34
Research Objectives

Chapter 2: Research Design and Methods 39
Data Source 39
Study Population 39
Type of Antipsychotics Included in the Study 41
Data Structure 41
iv
Outcome Measures 42
Healthcare Utilization 42
Economic Outcomes 43
Antipsychotic Medication Utilization Patterns 43
Econometric Models 44
Fixed Effect/First Difference Models for Healthcare Costs 44
Fixed Effect Poisson Estimation for Count Variables 45
Multivariate Models 47
Statistical Software 51

Chapter 3: Results 52
Demographic Characteristics 52
Comparisons of First-Generation vs. Second-Generation Antipsychotics 52
Description of Study Population 52
Factors Affecting Treatment Decision 56
Duration of Therapy 59
Switching to and/or Augmenting with Other Antipsychotics 60
Comparison of Healthcare Costs 61
Descriptive Statistics 61
Panel Data Fixed-Effect Model for Medical Costs 62
Comparison of Healthcare Utilizations 69
Panel Data Fixed-effect Poisson Regression Models 70
Comparisons with Cross Sectional Analysis 72
Second-Generation Head to Head Comparison 86
Descriptive Statistics of Study Population 86
Baseline Healthcare Utilizations 87
Factors Affecting Treatment Decision 88
Switching to and/or Augmentation with Other Antipsychotics 89
Descriptive Statistics of Post-Intervention Healthcare Costs and
Utilization 91
Panel Data First Difference Models for Healthcare Costs 93
Panel Data Fixed Effect Poisson Regression Models 98
Comparison with Cross Sectional Analyses 101

Chapter 4: Discussion and Conclusion 114

References 126

v

List of Tables

Table 1: Demographic Characteristic of Study Sample 53

Table 2 Baseline Health Care Costs by Drug Type 55

Table 3: Baseline Health Care Utilization by Drug Type 56

Table 4: Logistic regression model for the selection of antipsychotic medications 58

Table 5: Duration of Therapy and Rate of Switch/Augmentation by Drug Type 59

Table 6: Likelihood of Switch/Augmentation: Logistic Regression 61

Table 7: Health Care Costs by Drug Type – 1 Year Post 62

Table 8: Panel Data First Difference Model: Total Healthcare Costs 63

Table 9: Panel Data First Difference Model: Prescription Costs 64

Table 10: Panel Data First Difference Model: Total Medical Costs 64

Table 11: Panel Data First Difference Model: Hospitalization Costs 65

Table 12: Panel Data First Difference Model: Outpatient Costs (Part B) 66

Table 13: Panel Data First Difference Model: Long-term Care Costs 66

Table 14: Panel Data First Difference Model: Community Mental Healthcare
Facility Costs 67

Table 15: Panel Data First Difference Model: Other Costs 67

Table 16: Panel Data First Difference Model: Impact of the Duration of
Therapy on Total Healthcare Costs 68

Table 17: Panel Data First Difference Model: Total Healthcare Costs
Excluding patients with extended antipsychotic therapy 69

Table 18: Descriptive Statistics of Medical Utilization Patterns 70

vi
Table 19: Fixed-Effect Poisson Model – Hospitalization rate 71

Table 20: Fixed-Effect Poisson Model –Length of Hospital Stays 71

Table 21: Fixed-Effect Poisson Model –Length of LTC Stays 72

Table 22: Fixed-Effect Poisson Model – Number of Suicide Attempts 72

Table 23: Cross Sectional Analysis – Total Healthcare Costs 73

Table 24: Cross Sectional Analysis – Pharmacy Costs 74

Table 25: Cross Sectional Analysis – Medical Costs 75

Table 26: Cross Sectional Analysis – Outpatient costs 76

Table 27: Cross Sectional Analysis – Hospitalization Costs 77

Table 28: Cross Sectional Analysis – CMHC costs 78

Table 29: Cross Sectional Analysis – Long Term Care Costs 79

Table 30: Cross Sectional Analysis – Other Costs 80

Table 31: Cross Sectional Analysis – Number of Hospitalization 81

Table 32: Cross Sectional Analysis – Days of Hospital stays 82

Table 33: Cross Sectional Analysis – Days of Long-term Care stays 83

Table 34: Cross Sectional Analysis – Number of Suicide Attempts 84

Table 35: Summary of the Panel data and Cross Sectional Models 85

Table 36: Demographic Characteristics of Study Sample – Head to Head
Comparisons 86

Table 37: Baseline Healthcare costs by drug 87

Table 38: Baseline Medical Utilization by drug 88

Table 39: Multinomial Logit Model for Selection of Second-Generation
Antipsychotic Medication 89

vii
Table 40: Duration of Therapy and Rate of Switch or Augmentation by Drug 90

Table 41: Logistic Regression Model – Switch/Augmentation 91

Table 42: Descriptive Statistics of Healthcare Costs by Drug – Year 2 92

Table 43: Descriptive Statistics of Healthcare Utilization by Drug – Year 2 93

Table 44: Head to Head Comparison Panel Data First Difference Model:
Total Healthcare Costs 94

Table 45: Head to Head Comparison Panel Data First Difference Model:
Prescription Costs 94

Table 46: Head to Head Comparison Panel Data First Difference Model:
Medical Costs 95

Table 47: Head to Head Comparison Panel Data First Difference Model:
Long-term Care Costs 96

Table 48: Head to Head Comparison Panel Data First Difference Model:
Hospitalization Costs 96

Table 49: Head to Head Comparison Panel Data First Difference Model:
Outpatient Services Costs (Part B costs) 97

Table 50: Head to Head Comparison Panel Data First Difference Model:
Community Mental Health Facility Costs 97

Table 51: Head to Head Comparison Panel Data First Difference Model:
Other Costs 98

Table 52: Head to Head Comparison Panel Data Fixed Effect Poisson Model:
Number of Hospitalizations 99

Table 53: Head to Head Comparison Panel Data Fixed Effect Poisson Model:
Days of Hospital Stays 99

Table 54: Head to Head Comparison Panel Data Fixed Effect Poisson Model:
Days of Long-term Care Facility Stays 100

Table 55: Head to Head Comparison Panel Data Fixed Effect Poisson Model:
Number of Suicide Attempts 100

viii
Table 56: Cross Sectional Head to Head Comparisons – Pharmacy Costs 101

Table 57: Cross Sectional Head to Head Comparisons – Medical Costs 102

Table 58: Cross Sectional Head to Head Comparisons - Total Healthcare Costs 103

Table 59: Cross Sectional Head to Head Comparisons – Long Term Care Costs 104

Table 60: Cross Sectional Head to Head Comparisons – Part B Costs 105

Table 61: Cross Sectional Head to Head Comparisons – CMHC Costs 106

Table 62: Cross Sectional Head to Head Comparisons – Hospitalization Costs 107

Table 63: Cross Sectional Head to Head Comparisons – Other Costs 108

Table 64: Cross Sectional Head to Head Comparison – Number of
Hospitalization 109

Table 65: Cross Sectional Head to Head Comparison – Days of Hospital Stays 110

Table 66: Cross Sectional Head to Head Comparison – Days of
Long-term Care Stays 111

Table 67: Cross Sectional Head to Head Comparison – Number of Suicide
Attempts 112

Table 68: Summary of the Panel Data and Cross Sectional Models 113

ix
Abstract

While the comparative efficacy of second-generation antipsychotics (SGAs)
vs. first generation antipsychotics (FGAs) has been well documented, the treatment
effect of SGAs in bipolar disorder has not been directly evaluated in a real-world
setting. Panel data fixed effects models provide a simple yet powerful tool in
measuring treatment effects. The objectives of this study were to investigate the
treatment effect of SGAs in bipolar patients enrolled in the fee-for-service (FFS)
Medi-Cal program using panel data fixed effect models. The retrospective paid claims
files from the Medi-Cal program was used. Patients were included if they had at least
1) one claim(s) for any bipolar medications and 2) one medical claim(s) with ICD-9
codes 296.40-296.99. The population was further defined as: 1) patients who initiated
a new antipsychotic therapy between 2000 and 2001 (index date); 2) were 18 ≤ age at
the index date; and 3) were continuously eligible for Medi-Cal during the 2-year pre-
and post-index date period. Fixed effects panel data linear models were fitted for
continuous treatment outcomes while fixed effects Poisson regression models were
used for count variables. The results show that compared to FGAs, SGA use was
associated with a significantly larger increase in total healthcare costs (2366.0,
p=0.028) as well as pharmacy costs (770.9, p <0.0001). The SGA group also had a
larger increase in medical costs compared to the FGA group (1687.0, p=0.09). These
results suggest that the anticipated improvements in medical utilizations for the SGAs-
treated patients were not realized resulting in a greater increase in healthcare costs in
x
the SGA group. The results form the fixed effect Poisson analyses were consistent
with this conclusion, indicating that SGA use did not have significantly positive
impact on the utilization of the medical services examined. When 3 SGAs were
compared head-to-head, there was no significant difference in outcomes between
patients treated with olanzapine and risperidone. The quetiapine group had a
significantly lower pharmacy costs compared to the olanzapine group. However, the
savings in pharmacy costs did not translate into savings in the total healthcare costs
primarily due to the increase in medical costs in the quetiapine group.
1
CHAPTER 1 INTRODUCTION
Traditionally, the goal of antipsychotic therapy in patients with bipolar
disorder was to control acute manic symptoms and it was recommended that
antipsychotic therapy should be reconsidered once the manic symptoms are controlled.
However, studies have shown that many patients with bipolar disorders are on
antipsychotic therapy for an extended time period and in many cases, second
generation antipsychotics became a part of the maintenance therapy regimen (APA
2002).
Although many studies have documented the efficacy of conventional
antipsychotics as an acute treatment for bipolar mania, conventional antipsychotics
have a high prevalence of severe side effects such as extrapyramidal symptoms (EPSs)
and tardive dyskinesia which may result in poor compliance (Nemeroff 2000, APA
2002 ). Second-generation antipsychotics show similar efficacy in controlling manic
symptoms as conventional antipsychotics but have a favorable side-effect profile.
Since the introduction of second-generation antipsychotics, there has been a growing
trend of using these drugs as both first line therapy in bipolar mania and as an
extended maintenance therapy in bipolar disorder. In the California Medicaid
program (Medi-Cal), this trend has been accelerated by the removal of prior
authorization requirements on second-generation antipsychotics in October 1997. It
was hypothesized that the use of second-generation antipsychotics would decrease
utilization of other costly medical services including office visits, hospitalizations,
emergency room visits, and institutionalization in long-term care facilities, thus
2
reducing the total health care costs of treating patients with bipolar disorder.
However, the economic treatment effect of second-generation antipsychotics in
bipolar disorder has not been directly evaluated. This is of particular concern to Medi-
Cal since second-generation antipsychotics have become the most costly medications
in the Medi-Cal pharmacy benefit program (CDHS 2004).
The objective of this study is to investigate the treatment effect of second-
generation antipsychotics in bipolar patients enrolled in the fee-for-service (FFS)
Medi-Cal program. Specifically, the economic outcomes of the atypical
antipsychotics in comparison to the conventional antipsychotics will be evaluated
controlling for the factors that may affect the outcomes of antipsychotic treatment.
Also the economic outcomes of different second-generation antipsychotics will be
compared.

CLINICAL BACKGROUND
Definition of Disease
Bipolar disorder, also known as manic-depressive disorder, is a recurrent mood
disorder characterized by cycles of depression and mania

(Nemeroff 2000). The
pathophysiolology of bipolar disorders remains largely unknown and it is widely
believed that multiple factors work simultaneously to produce the illness. There are
several hypotheses that have been the main focus of research. One of the most widely
accepted hypotheses involves neurotransmitters such as dopamine and serotonin.
Evidence from pharmacologic research on bipolar disorder, specifically the fact that
3
drugs that affect the neurotransmitters are effective in controlling symptoms, supports
this hypothesis. However, it is still not clear whether affective disorders are caused by
the imbalance in the concentration of the different neurotransmitters or other some
other mechanisms regarding these neurotransmitters

(Muller-Oerlinghausen 2002).
Genetic factors also play a role in bipolar disorder. Numerous studies have
reported increased risk of bipolar disorders in family members of a bipolar patient. It
is hypothesized that the genes associated with the neurotransmitter system can be
inherited and bipolar disorder is triggered when these genetic factors interact with
other environmental or developmental factors.

Classification of Bipolar Disorders
Bipolar disorders can be categorized based on the patterns of symptoms
experienced by the patient. According to Diagnostic and Statistical Manual of Mental
Disorders [4
th
Edition (DSM-IV)], Bipolar I disorders is defined as having one or more
full-blown manic or mixed episodes, with a history of one or more major depressive
episodes (APA DSM 2000). Patients with Bipolar II disorder have recurrent major
depressive episodes with one or more of hypomanic episode. Cyclothymic disorder is
characterized by numerous periods of mild hypomanic or depressive symptoms that do
not meet the criteria full-blown manic or major depression lasting for at least 2 years
with no symptom-free intervals lasting longer than 2 months. Patients with disturbed
mood such as mania, hypomania, or depression but whose symptoms do not meet the
4
criteria for other bipolar categories are diagnosed with bipolar disorder not otherwise
specified.

Epidemiology and Course of Bipolar Disorder
The lifetime prevalence of bipolar disorder is estimated at between 1% and
6.4% of the population (Kessler 1994, Narrow 2002, Judd 2003). Many patients
experience their first episode between age 15 and 24 years and the first episode may
consist of any symptom of bipolar disorder – manic, hypomanic, mixed, or depression.
Bipolar disorder is a highly recurrent illness and patients may have more than 10
episodes of mania or depression during their lifetime (Muller-Oerlinghausen 2002).
The duration of episodes and intervals between the episodes varies among individuals,
and the pattern of illness stabilizes after the fourth or fifth episode (APA Practice
Guideline 2002). Most patients experience multiple episodes at an average of 0.4 to
0.7 episodes per year, with each lasting three to six months (Angst 2002).
Bipolar patients without proper education about their illness often consider the
symptoms of a manic episode such as elevated self-esteem and grandiosity to be
desirable and deny having an illness. The desirability of the manic symptoms and
denial of illness often result in patients seeking professional care primarily when they
have depressive symptoms. Also, many bipolar patients fail to report prior manic
episodes, thus delaying proper diagnosis and treatment of the illness (APA Practice
Guidelines 2002).

5
Psychotic Symptoms
Patients with bipolar disorder often have psychotic symptoms such as
hallucinations and delusions, especially during their manic episodes. Some patients
who experience a manic episode may show catatonic symptoms as well. Other
symptoms of manic episodes include inflated self-esteem, grandiosity, euphoria,
increased energy, and agitation. Patients may also become unusually aggressive and
make poor judgments often leading to reckless behavior. Symptoms of depressive
episodes include lasting feeling of hopelessness, pessimism, feelings of guilt, or
helplessness.
Increased rate of substance abuse has been reported in bipolar patients as well
as in other mental disorders. Patients with Bipolar I disorders have a 46% lifetime
prevalence of alcohol-related disorders, compared to a prevalence of 14% in the
general population. Also, Bipolar I and Bipolar II patients have much higher
prevalence of other types of substance abuse, 61% and 48% respectively, compared to
6% in the general population (Brown 2001, Regier 1990).
Bipolar disorder is associated with increased risk of suicide, with the lifetime
risk of suicide ranging from 8% to 20%. In the general population, the lifetime suicide
prevalence is less than 0.5% (Bostwick 2000, Goodwin 2003). In general, the risk of
suicide is associated with the severity of depressive symptoms in a patient. Patients
with a depressive episode or a mixed episode have higher risk of suicide compared to
those with a manic episode.
6
Bipolar disorder also causes substantial psycho-social morbidity, frequently
affecting patients’ relationship with others as well as their occupation and other
aspects of life. Bipolar disorder seriously impairs patients’ interpersonal skills and
personal and professional judgment. Bipolar patients tend to have a substantially
higher divorce rate and are much more likely to have difficulties with maintaining
their occupational status (APA Practice Guidelines 2002).

DRUG THERAPY FOR BIPOLAR DISORDER
Currently, no cure for bipolar disorder exists. However, treatment can
significantly decrease the associated morbidity and mortality and improve patient
functioning. Since bipolar disorder is an episodic illness with a variable course,
specific treatment decisions are made based on the characteristics of the episode at the
time of treatment and the course of illness. Four medication classes are frequently
used for bipolar disorders: mood stabilizers, antipsychotics, antidepressants, and
anticonvulsant (APA Practice Guidelines 2002).

Acute Treatment
The primary goal of treatment of an acute manic or depressive episode is
symptom control to allow a return to normal level psycho-social functioning. The
first-line treatment for manic or mixed episodes is the initiation of an antipsychotic
therapy in conjunction with the mood stabilizer lithium or valproic acid. For milder
manic symptoms, monotherapy of any of these agents can be used. For bipolar
7
depression, the first-line therapy consists of either lithium or an anticonvulsant such as
lamotrigine. In general, standard antidepressants such as SSRIs are recommended as
an add-on therapy to these agents. Monotherapy using an antidepressant is not
recommended in bipolar patients due to a concern that antidepressant monotherapy
might induce mania or rapid cycling (APA Practice Guidelines 2002, Sachs 2000).
However, many bipolar patients receive antidepressant monotherapy due to the
presentation of the patient during the depression phase and the under-reporting of
previous manic symptoms by patients (Shi, McCombs, Thiebaud 2004)

Maintenance Treatment
Patients need to be on maintenance pharmacotherapy in order to prevent
relapse, reduce sub-threshold symptoms, and reduce the risk of suicide following
remission of an acute episode. Lithium monotherapy provides the best empirical
evidence as an effective maintenance treatment. Options for the maintenance
treatment also include monotherapy of various anticonvulsant such as valproate,
lamotrigine, and carbamazepine and combination therapy of these agents with lithium.
Some of the atypical antipsychotics, notably olanzapine, have also been approved as
the maintenance therapy. Current treatment guidelines recommend that once the
remission of acute mania is accomplished, any antipsychotic therapy should be
reassessed and tapered toward discontinuation (APA Practice guidelines 2002, Sachs
2000). In practice, however, many patients continue to be on an antipsychotic for
months after the remission of acute mania (Sernyak 1994, Keck 1996). It is also
8
recommended that antidepressant treatment be reevaluated and discontinued after
patients enter the maintenance phase.

Limitation of Current Pharmacotherapy
Despite the wide range of medications available for the treatment of bipolar
disorder, the current drug therapy is sub-optimal in many bipolar patients. Studies
have shown that a large proportion of patients who are treated with lithium do not
respond to the treatment. The initial response rate of lithium in bipolar population is
reported to be approximately 50%. Up to 55% of the lithium responders will develop
resistance to lithium therapy after 3 years of treatment (Calabrese 1995, Calabrese
1996). Patients with severe manic symptoms, rapid cycling bipolar, or a history of
frequent manic or mixed episodes often respond poorly to lithium treatment. Lithium
is also associated with poor response in the patients with other comorbid conditions
such as substance abuse or schizophrenia (Swann 1999, Bowden 1995). Slow onset of
action can be also of concern in lithium therapy. Lithium has been shown to reduce
manic symptoms within first 10 days of treatment, with a usual range of response
between 5 to 21 days. In contrast, valproate shows antimanic effect within first 5 days
of treatment (Keck 1996). In many cases, the delay in onset of antimanic action is
associated with prolonged hospitalization and has significant economic implications.
Moreover, long-term compliance to lithium therapy has been reported to be poor
(Keck 1997, Weiss 1998).
9
Valproate is preferable to lithium when rapid onset of antimanic action is
needed. Valproate has been shown to have comparable antimanic efficacy to lithium
in several randomized clinical trials and better antimanic efficacy in the patients with
mixed episode (Bowden 1994). A recent study documented similar efficacy of lithium
and valproic acid as a maintenance therapy measured as the time to recurrence of any
mood episode (Bowden 2000). However, little evidence exists to support the efficacy
of valproate in the depression. Other anticonvulsants frequently used in bipolar
disorder as a mood stabilizer have very limited efficacy and safety information
available compared to lithium and valproate (lamotrigine) or appear to be less potent
than lithium (carbamazepine).

Efficacy of Antipsychotics in Bipolar Disorder
First generation antipsychotics such as haloperidol and chlorpromazine have
been widely used as a short-term treatment for acute mania. Antipsychotics have
faster onset of action compared to lithium and studies have shown that typical
antipsychotics such as haloperidol are superior to the mood stabilizer alone in
relieving the symptoms of acute mania (Garfinkel 1980, Kane 1988). However, first
generation antipsychotics are associated with increased risk of exrapyramidal
symptoms and tardive dyskinesia that have been shown to negatively affect patient
compliance in long-term therapy (Perkins 1999).
Second generation antipsychotics have a more favorable side effect profile and
better on overall functioning than conventional antipsychotics (APA Practice
10
Guidelines 2002). Second generation antipsychotics include clozapine, olanzapine,
risperidone, quetiapine, and ziprasidone. Evidence from clinical trials indicates that
second generation antipsychotics are more effective than moodstabilizer monotherapy
and at least as effective as the typical antipsychotics in treatment of mania. Several
randomized controlled clinical trials have found the evidence of efficacy of olanzapine
in the treatment of acute mania or mixed episodes either as a monotherapy or as a
combination with lithium. In a 4-week randomized clinical trial, a total of 115
patients with diagnosis of bipolar disorder, manic or mixed, were randomized to either
olanzapine or placebo (Tohen 2000). The authors found that patients treated with
olanzapine had a statistically significantly greater improvement measured as Young
Mania Rating Scale (YMRS) than the placebo group. Response rate, defined as at
least 50% improvement from baseline score, was significantly higher in the olanzapine
group than in the placebo group. Also, the olanzapine-treated patients showed a
higher rate of euthymia, defined as the proportion of patients whose YMRS score was
greater than 12 at the end point of the study. Emergence of EPSs was low and was not
significantly different in the two groups. However, the olanzapine-treated group
showed a significantly larger mean weight gain compared to the placebo group.
In a 3-week, randomized, double-blind trial that compared olanzapine with
divalproex in the treatment of acute mania, olanzapine was associated with higher
response rate and greater improvement of severity of manic symptoms (Tohen 2002a).
The mean YMRS score decreased by 13.4 in the olanzapine-treated group, whereas
the divalproex-treated group had a mean decrease of 10.4. The olanzapine group had
11
a response rate (defined as 50% or more reduction in YMRS) of 54.4%, and remission
rate (endpoint YMRS 12 or less) of 47.2%, while in the divalproex group, the same
rates were 42.3% and 34.1%, respectively. Somnolence, dry mouth, tremor, speech
disorder, and sleep disorder occurred more often in the olanzapine group, while more
patients in the divalproex group experienced nausea and diarrhea. In addition,
olanzapine was associated with higher rate of weight gain. Tohen et al. also
investigated the efficacy of olanzapine in combination with mood stabilizers in
patients who are nonresponsive to moodstabilizer monotherapy. olanzapine
combination therapy significantly improved YMRS compared to the patients
continuing mood stabilizer monotherapy . Time to remission of symptoms was
significantly shorter in the combination therapy group (Tohen 2002b).
Olanzapine has a similar efficacy in controlling manic symptoms when
compared to a typical antipsychotic. When compared with haloperidol in the bipolar
type schizoaffective disorder patients, olanzapine was associated with significantly
better improvement in the depressive symptoms. Manic and mixed symptoms
improved in both olanzapine and haloperidol groups, and the rates were not
statistically significantly different in the two groups although the olanzapine group
generally had higher improvement rates. Emergent EPS was significantly less likely
in the olanzapine-treated group than in the haloperidol group (Tohen 2001).
Risperidone has also been studied for the treatment of bipolar disorder. Vieta
et al. studied the efficacy and safety of risperidone in the treatment of bipolar and
schizoaffective disorders in a 6-month, multicenter, open label study (Vieta 2001).
12
Risperidone was added to any previous mood stabilizer that the patients were taking
and the outcomes after 6-month study period were compared to the baseline. Addition
of risperidone to the treatment regimen produced significant improvement of
outcomes measured as the YMRS, Hamilton Rating Scale for depression (HAM-D),
the Positive and Negative Syndrome Scale (PANSS), and the Clinical Global
Impressions scale (CGI). The severity of adverse events, related to EPS was measured
by the UKU Side Effect Rating Scale and UKU score decreased significantly at the
endpoint of the study compared to the baseline. However, the validity of this study
may be limited since there was no valid comparison group and therefore, the treatment
reported in the study might have been confounded with natural disease process.
Sachs et al. compared the efficacy of risperidone and haloperidol in
combination therapy with a mood stabilizer in bipolar patients with an acute manic
episode. In this three-week, double-blind, placebo-controlled study, risperidone was
as efficacious as haloperidol in combination therapy with a mood stabilizer. Both
groups had significant improvement of symptoms measured as YMRS and CGI
severity score compared to the placebo group. Risperidone had more favorable
adverse effect profile compared to haloperidol

(Sachs 2002)
Literature on the efficacy of quetiapine in bipolar disorder is limited compared
to olanzapine and risperidone. A 12-month open-label study that compared the
efficacy and tolerability of quetiapine compared to a mood stabilizer in the treatment
of bipolar disorder was conducted by the Altamura group (Altamura 2003). Patients
who had mood episodes were randomly assigned to one of the two groups of open-
13
label treatment, with quetiapine or classical mood stabilizer at flexible doses. The
authors found that quetiapine was is equally efficacious as lithium or valproate in
treating mood disorders. It should be noted that the results from this study might have
limited validity due to the small sample size (n=28) and the fact that the treatment was
not blinded.
Other second-generation antipsychotics have not been studied intensively for
the treatment of bipolar disorders and the information on the efficacy of these agents
are somewhat limited. Keck et al. investigated efficacy and safety of ziprasidone in
the treatment of acute bipolar mania in a placebo-controlled randomized trial (Keck
2003). The ziprasidone group experienced a significantly larger decrease in both
positive and negative syndrome scale compared to the placebo group. Adverse events
that occurred significantly more often in the ziprasidone group include somnolence,
headache, dizziness, hypertonia, nausea, and akathisia. The emergence of EPSs was
not significantly different between the two groups. However, the drop out rate of the
patients was very high in this study and the results of this study should be interpreted
with caution.
Few studies have been conducted to compare the efficacy and safety across
different second-generation antipsychotics in bipolar disorder. One naturalistic study
compared the efficacy and tolerability of clozapine, risperidone, and olanzapine in the
treatment of bipolar disorder (Guille 2000). Efficacy was measured as the changes in
Clinical Global Impressions (CGI) based on the clinical data prepared from chart
review of 42 patients. Comparisons of the CGI changes indicated equivalent efficacy
14
in the three antipsychotics. Levels of EPSs and other side effects except weight gain
were similar in all groups. Weight gain of greater 10 lbs was significantly greater in
patients treated olanzapine. While this study was the first head-to-head comparison of
the atypical antipsychotics in bipolar disorder, methodological flaws limit the validity
of the study. First small sample size limits the validity and generalizability of the
results. Second, factors that influence the treatment selection and outcomes were not
controlled, such as severity of illness, comorbid mental or physical conditions,
concomitant drug use pattern, and duration of antipsychotic therapy.

Long-term Treatment with Antipsychotics
In general, long-term antipsychotic treatment is not recommended in most
bipolar treatment guidelines primarily because of the side effects associated with long-
term antipsychotic therapy. Bipolar patients have been reported as having a greater
risk for developing adverse events of antipsychotic therapy such as tardive dyskinesia
and parkisonism (Kane 1982, Mukherjee 1986). However, in practice, many patients
continue to take antipsychotics for months after the remission of symptoms of manic
episodes. Sernyak et al. found that all 40 of the bipolar acute manic patients who had
been treated with adjunctive antipsychotic treatment during hospitalization were still
receiving antipsychotics at discharge and 95% of them were receiving antipsychotics
at 6 month follow-up (Sernyak 1994). Keck et al. reported that 68% of the bipolar
manic patients continued to receive antipsychotics 6 months after discharge. Factors
that are associated with maintenance antipsychotic treatment at 6-month follow up
15
include being male, medication noncompliance in the month prior to the index
hospitalization, severity of manic symptoms, and prescription of antipsychotics at the
time of discharge (Keck 1996). These studies indicate that antipsychotics have
frequently been used as a “mood stabilizer” in clinical practice. Despite the frequent
utilization of antipsychotics in maintenance phase, only olanzapine has been approved
as a long-term therapy.

Outcomes of Drug Therapy in Patient with Bipolar Disorder
Risk of Suicide
Bipolar patients have substantially higher risk of suicide compared to the
general public. The high rate of suicide is associated with substantial economic
burden to the society. One study calculated that the increased suicide rate in the
bipolar population account for approximately 17% of $37 billion indirect costs of
bipolar illness to society in the US in 1990 (Wyatt 1995). A considerable literature
documents the anti-suicide effect of long-term lithium therapy in bipolar disorder. A
recent meta-analysis of 33 studies showed that long-term lithium therapy was
associated with 13-fold decrease in rates of suicide or attempted suicide compared to
no lithium maintenance treatment (annual rate of 2.97% vs. 0.197%). The lowered
rate of suicide during maintenance lithium therapy nonetheless is substantially higher
than the base rate of general public (Baldessarini 2001).
The effect of other agents used in bipolar disorder on the risk of suicide has not
been addressed. In a recent study, Goodwin et al. evaluated the anti-suicide effect of
16
divalproex in comparison to lithium (Goodwin 2003). After controlling for age, sex,
health plan, year of diagnosis, comorbid medical and psychiatric conditions, and
concomitant psychotropic drug use, divalproex therapy was associated with
significantly greater risk of suicide mortality, suicide attempts resulting in
hospitalization, and suicide attempts or suicide behavior resulting in ER visits but not
hospital admission. So far, no information is available on the anti-suicide effect of
second-generation antipsychotics. As the interest in second-generation antipsychotics
as a potential mood stabilizer increases, it is appropriate to assess the anti-suicide
effect of these agents.

Hospitalization
As with other chronic mental illness, one of the one of the main goals of drug
treatment in bipolar disorder is to reduce utilization of other costly medical services
such as hospitalization. This is particularly true for patients with bipolar disorder who
may require frequent hospitalization due to the acute and cyclical nature of the disease
especially in the manic phase. Bipolar patients have higher hospital admission rate
compared to patients with other mental disorders (Bartel 2000, Peele 2003).
Compared to patients with unipolar depression, bipolar patients have been reported to
be three times more likely to be hospitalized over a six month period (Bartels 2000).
Higher hospital admission rates are translated to higher medical costs in this
population. In fact, hospitalization has been reported as the major driver of the
medical costs in patients with bipolar disorder. Hospitalization and nursing home care
17
accounted for 59% of $7.6 billion direct costs of bipolar disorder in the US in 1990
(Wyatt 1995). Peele et al. reported that privately insured bipolar patients had overall
annual hospital admission rate of 39.1 %, which surpasses even the 35.2% hospital
admission rate for patients with a substance abuse diagnosis. Accordingly, annual
expenditure per patient for bipolar patients was much higher than patients with other
mental disorders including substance abuse, anxiety disorders, and attention deficit
hyperactivity disorder (Peele 2003).
It has been well documented that long-term lithium treatment reduces the risk
of hospital admission and duration of hospital stay (Maj 1998, Scott 2002). However,
few studies evaluated the association between second-generation antipsychotic therapy
and hospital use. Considering the growing importance of second generation
antipsychotics in the treatment of bipolar disorder, the impact of antipsychotic therapy
on delay in future hospitalization will be assessed.

Diabetes
Atypical antipsychotic therapy is that these agents may be associated with
induction or exacerbation of type 2 diabetes mellitus (Regenold 2002). This is
particularly of significance in bipolar disorder since 1) bipolar patients have been
reported to have significantly higher prevalence of obesity and type 2 diabetes
independent of psychotropic drug use and 2) pharmacotherapy of bipolar usually
include other psychotropic drugs that have been reported to be associated with weight
gain and therefore might exacerbate type 2 diabetes (Regenold 2002).
18
There is a considerable body of literature that evaluates the impact of atypical
antipsychotics on the onset of new diabetes or exacerbation of existing diabetes in
patients with schizophrenia. It is not clearly understood, however, how the atypical
antipsychotics induce type 2 diabetes. It was first believed that weight gain, which is a
primary side effect of some atypical antipsychotics, eventually lead to type 2 diabetes.
However, recent studies report that weight gain was not a significant factor for onset
of diabetes in patients treated with atypical antipsychotics, suggesting that a
mechanism other than obesity may be involved in the development of type 2 diabetes
in this population (Henderson 2000, Meyer 2002).
Gianfrancesco et al. examined the differential effects of risperidone,
olanzapine, clozapine, and conventional antipsychotics on type 2 diabetes using claims
date from a large health plan (Gianfrancesco 2002). A logistic regression model was
used to control for age, gender, concurrent use of antipsychotics. The authors reported
increased risk of acquiring or exacerbating type 2 diabetes in patients treated with
olanzapine, clozapine and some conventional antipsychotics. risperidone patients
were not significantly different from the control group (untreated patients). However,
this study does not control for severity of illness, comorbidity, previous treatment
history, baseline weight, family history of diabetes, and other factors some of which
were not observable to the researchers and therefore, the results should be interpreted
with caution.
The Koro group assessed the effect of olanzapine and risperidone on risk of
diabetes among patients with schizophrenia (Koro 2002). This population based
19
nested case-control study showed a significantly increased risk of developing diabetes
of olanzapine patients compared to non-users of antipsychotics and those taking
conventional antipsychotics. Patients taking risperidone had a non-significant
increased risk of developing diabetes than non-users of antipsychotics and those taking
conventional antipsychotics. However, the information on severity of illness, family
history, duration of therapy all of which can influence both choice of drug and the
development of diabetes were not included or controlled for in the model.
Antipsychotic drug use is defined as the antipsychotic used between the beginning of
the study period and index date which is the onset of diabetes, and therefore allowing
for concurrent use of other antipsychotic medications.
A study by Sernyak et al. examined the association of diabetes mellitus with
use of atypical antipsychotics using a large national sample from the Veterans Health
Administration of the Department of Veterans Affairs (Sernyak 2002). Patients were
categorized into five groups based on age: under 40, 40-49, 50-59, 60-69, and over 70
years and logistic regression was utilized to calculate the odds ratio within each of the
age strata for the association of atypical antipsychotic prescription and the diagnosis of
diabetes. Patients were assigned to each group based on the type of antipsychotics
used within the last two weeks of the 4-month study period. The diagnosis of diabetes
was also identified within the same 4-month period. Thus, it is possible that the
diagnosis of diabetes could have occurred earlier than the prescription of
antipsychotics. In addition, some of the important confounding factors were not
controlled in the model including weight, family history, concurrent drug use, and
20
comorbidities. The results indicate that patients who received atypical antipsychotics
were more likely to develop diabetes than those who received typical antipsychotics.
For patients less than 40 years of age, all of the atypical agents including olanzapine,
risperidone, quetiapine, and clozapine were associated with a significantly increased
incidence of diabetes. For older patients, the incidence of diabetes was significantly
increased in the olanzapine, quetiapine, and clozapine groups but not in the
risperidone group.
Current evidence indicates an association of diabetes with use of atypical
antipsychotics primarily in schizophrenic patients. Question still remains whether this
is a common class effect of the second-generation antipsychotics or they truly are
different in terms of impact on diabetes. Studies support either of the theory suffer
from methodological flaws that put the validity of the results in question. Also, the
impact of second-generation antipyschotics in bipolar population has not been studied
exclusively and additional study in this population is warranted.

SOCIAL AND ECONOMIC IMPACT OF BIPOLAR DISORDER
Bipolar disorder is associated with intensive health care utilization, decreased
productivity, high rate of unemployment, and a high risk of suicide, all of which can
have substantial social and economic burden to society. The World Health
Organization (WHO) reported bipolar disorder as the 6
th
leading cause of disease
burden at ages 15-44 years worldwide, as measured by the Disability Adjusted Life
Years (DALYs) (Murray 1996). The disability and productivity loss affect the
21
employment status of many bipolar patients. In the United States, 37% of bipolar
patients are unemployed, only 43% of bipolar patients are employed 6 months after
discharge from a psychiatric hospital and only 21% were functioning at an expected
level of employment (Dion 1988, Woods 2000).
As discussed earlier, bipolar patients are at substantially higher risk of death by
suicide than the general population. In fact, bipolar patients have been reported as
having the highest risk of suicide among any illness other than depression (Woods
2000). Bipolar disorder also imposes stress on the relationship of the patients with
their spouse and other family members. Bipolar patients are less likely to have ever
married and among married bipolar patients, divorce rate is much higher than the
general population. Family and caregivers of bipolar patients often experience burden
and distress, which result in decreased quality of life and lost productivity (Lish 1994).
The total cost of illness to society in the United States was estimated at $45
billion in 1991 (Wyatt 1995). Direct costs of the disease accounted for about 17% of
the total costs and the remaining 83% resulted from other indirect costs including costs
of productivity loss and unemployment of the patients and caregivers and suicide.
Institutional care (hospitalization and nursing home care) was the major driver of the
medical costs accounting for 59% of the direct costs of $7.6 billion. Simon et al.
assessed the direct health care costs of bipolar patients in a privately insured
population (Simon 1999). The annual medical costs of bipolar patients were
compared to those of age and gender matched control groups – the general outpatient
sample, diabetic patients, and patients with unipolar depression. Bipolar patients had
22
annual health care cost of $3,416 (SD=$6,862) which was significantly higher than
costs for general outpatient population ($1,462 (SD=$4,469)), for depressed patients
($2,570 (SD=$7,597)), and for diabetic patients ($3,083 (SD=$6,544)). Stender et al.
compared the number of medical/pharmaceutical encounters and costs of 3,120 bipolar
patients to those of age and gender matched randomly selected comparison group from
the same claim file from a private insurer in the US (Stender 2002). The bipolar group
consumed a total of $2,803,673 ($899 per person) on drugs with mean number of 15
central nervous system (CNS) drugs per person, while the comparison group spent
$461,354 ($148 per person) on drugs with a mean of 1 CNS related drug. In terms of
medical costs, the bipolar group had $16,230,840 ($5,202 per person), compared to
$4,174,797 ($ 1,338 per person) of the comparison group.
In a more recent study, Li et al. reported that the bipolar patients enrolled in
California Medicaid (Medi-Cal) program consumed $10,500 per patient per year,
primarily for ambulatory care ($4,737) and hospital services ($4,648) (Li 2002).
Multivariate analyses showed that drug utilization patterns had a significant impact on
the total health care costs in this population. Mood stabilizer use was associated with
significant reduction in healthcare costs of $5,044. Delays in treatment, measured as
the time between the first observed diagnosis of bipolar disorder and the initiation of
drug therapy, was associated with an increase in total costs of $5,565 relative to other
patients who were treated without delay. These results suggest that bipolar patients
who do not receive optimal drug therapy experience more relapses in symptoms and
therefore incur higher health care costs.
23
TREATMENT EFFECT OF SECOND GENERATION ANTIPSYCHOTICS
Randomized clinical trials measure efficacy of alternative treatment
approaches under tightly controlled clinical condition. However, due to the
complexity of bipolar disorder, growing interests exists in the treatment effect of
second-generation antipsychotics in real world clinical practice (Vieta 2001). For
example, patients with bipolar disorder typically take multiple drugs making it
extremely difficult to project the real world efficacy of the drugs based on their
efficacy measured in randomized clinical trials where concomitant drug use is
restricted.
In addition, randomized clinical trials maximize compliance whereas in
practice, pharmacotherapy of bipolar disorder often fails due to noncompliance with
treatment. The rate of noncompliance in patients with bipolar disorder has been
documented to be as high as 41% (Berk 2003). Li et al showed that only 5.5% of
Medi-Cal patients with bipolar disorder used a mood-stabilizer consistently for 1 year
(Li 2002). It has been reported that among bipolar patients taking either a mood
stabilizer or an anticonvulsant, 47% acknowledged some form of noncompliance in
the preceding 2 year (Scott 2002). The rate of compliance is even lower among
patients with higher risk, often resulting in hospitalization. Keck et al. reported that
64% of the patients who were hospitalized for acute mania were noncompliant with
their pharmacologic regimen in the month prior to admission (Keck 1996).
Noncompliance stems from various factors including medication side effects,
perceived lack of efficacy, lack of insight, denial of illness, and cost of medication
24
(Keck 1997, APA Practice guideline 2002, Weiss 1998). Some bipolar patients are
also reluctant to give up the experience of hypomania and mania. Euphoria,
grandiosity, and elevated ability to focus during the manic episode may be very
desirable and patients recall only these aspects of mania, often denying other
devastating features. Conversely, bipolar patients who are prescribed with certain
types of drugs have been reported to use more medication than prescribed (Weiss
1998). The high rate of noncompliance to their prescription drug regimen in the
bipolar population implies that the efficacy established through randomized clinical
trials may not be realized in practice.
Clinical trials often focus on a specific subgroup of the entire bipolar
population. For example, the majority of second-generation antipsychotic clinical
trials focused only on patients with a manic or mixed episode, and the impact of this
category of drug on bipolar depressive patients is not clear. Clinical trials typically
exclude patients with other psychiatric comorbidities, while a large proportion of
bipolar patients suffer from various comorbidities such as substance abuse (Tohen
2000, Tohen 2002a, Sachs 2002). Theses limitations in applicability of information
from randomized clinical trials substantiate the necessity to assess the effectiveness as
well as the efficacy of a new class of drug therapy.

Drug therapy in the Medi-Cal Population
Medicaid is a public health insurance system that is funded jointly by the
federal and state governments. The California Medicaid (Medi-Cal) program provides
25
comprehensive health care coverage including outpatient and inpatient medical
services and prescription benefits to the poor and disabled people in the state of
California. Medicaid programs including Medi-Cal generally cover a broad spectrum
of prescription drugs. However, the explosive increase in the expenditures in
prescription drugs started in the 1980s prompted many state Medicaid programs to
implement various restrictions on prescription benefits including formularies,
requiring of prior authorization, prescription limits and cost sharing options.
The intended impact of prior authorization was to contain pharmacy benefit
expenditures by preventing over-utilization and/or misuse of costly or potentially toxic
drugs when less expensive or safer alternatives are available. However, it is also has
been hypothesized that restricting access to prescription medications may induce
underutilization of necessary drugs, resulting in poor health outcomes and increased
medical costs for outpatient visits, emergency room (ER) visits, and hospitalization.
Prior to October 1997, Medi-Cal required prior-authorization for all second-
generation antipsychotics. However, the evidence of more favorable side effect
profile and superior efficacy of atypical antipsychotics compared to the conventional
medication increased over time especially for patients with schizophrenia. A growing
number of publications supported the use of the second-generation antipsychotics as
the first line therapy for patients with schizophrenia rather than as an alternative to
conventional antipsychotics in case of treatment failure. In addition, studies
documented that the treatment outcomes in the patients with schizophrenia and bipolar
disorder enrolled in Medi-Cal were sub-optimal (McCombs 1999, McCombs 2000, Li
26
2002). These factors led Medi-Cal to lift the prior authorization requirement allowing
unrestricted access to the second-generation antipsychotics in October of 1997.
Results of studies that evaluated the impact of open access to atypical
antipsychotics on the patients with schizophrenia indicate that the open access was
associated with an immediate increase of number of treatment episodes initiated per
month. The increase was not only due to the substitution effect in those patients who
were previously treated with typical antipsychotics but also due to the access effect
that was observed in the patients who were not on drug therapy prior to open access
(McCombs 2004). These studies also documented the impact of open access on health
care costs in patients with schizophrenia. Open access was associated with significant
increase in total health care costs over the first post-treatment year for patients re-
initiating drug therapy (McCombs 2004). Most of the increase in costs was due to
increased use of ambulatory care and psychiatric hospital services. Prescription drug
costs also increased significantly. For switching/augmenting episodes, open access
was associated with a small but statistically insignificant decrease in total costs for
patients with schizophrenia.
Although the primary beneficiaries of open access were the patients with
schizophrenia, open access was also extended to patients with other severe mental
disorders including bipolar disorder. The impact of open access in the bipolar patients
was documented by Narayan et al. The authors reported a modest increase in the
number of restarting episodes with atypical antipsychotics after open access. As in the
patients with schizophrenia, open access was associated with a significant increase in
27
total health care costs over the first post-treatment year in the bipolar patients who re-
initiate drug therapy (Narayan 2006).

Economic Treatment Effect of Second Generation Antipsychotic Therapy in
Medi-Cal Patients with Bipolar Disorder

As discussed earlier, patients with bipolar disorder are likely to be on
antipsychotic therapy for an extended period and antipsychotics are one of major drug
therapies available for this population. While the use of second generation
antipsychotics has increased dramatically in patients with bipolar disorder, economic
effectiveness of second-generation antipsychotics in this population has not been
determined. The study by Narayan et al. evaluated the economic impact of open
access in patients with bipolar disorder. However, it did not evaluate the economic
outcomes across different antipsychotics for the treatment of bipolar disorder
(Narayan 2006).
The assessment of economic outcomes across different antipsychotics has
particular relevance to Medi-Cal as two of the second-generation antipsychotics,
olanzapine and risperidone, have become the most costly medications in the program
after open access in October 1997. In the first 6 months of 2004, for example, the
total expenditures for these two medications exceeded $200 million (CDHS 2004).
Not surprisingly the expenditures for atypical antipsychotics have increased in the
Medicaid programs in general. Medicaid expenditures for antipsychotics almost
tripled between 1995 and 2001 (Lewin Group 2003, Duggan 2003).
28
It has been argued that the sharp increase in drug expenditure may have been
offset by savings from reduced health care utilization in other sectors such as
hospitalization or long-term care use. The results of studies that evaluated the
economic impact of atypical antipsychotics in patients with schizophrenia generally
indicate no difference in total treatment cost or cost-effectiveness of atypical
antipsychotics compared to conventional ones. Rosenheck et al. compared the total
costs to VA when patients with schizophrenia were randomly assigned to either
olanzapine or haloperidol. The authors reported that the VA patients who were treated
with olanzapine had significantly higher VA costs compared to the patients who were
treated with haloperidol (Rosenheck 2003). However, there was no significant
difference in terms of societal costs (non-VA costs and non-health care costs included)
between the two treatment groups. Gibson et al. analyzed the total annual health care
costs of Michigan Medicaid patients with schizophrenia after the initiation of
antipsychotic therapy (Gibson 2004). There was no significant difference in total
health care costs between patients who initiated therapy with olanzapine, risperidone,
or haloperidone. Other studies that compared individual atypical antipsychotics with
typical antipsychotics show similar results (Gianfrancesco 2002, Tilden 2002).
Considering the extensive use of antipsychotics in this population and growing
trend of using second generation antipsychotics as a maintenance therapy, the
economic outcomes of second generation antipsychotics should be evaluated. Most of
economic studies of second generation antipsychotics focus on the schizophrenia
population, and the results from these studies might not be extended directly to the
29
patients with bipolar disorder. Second generation antipsychotics are used differently
in patients with bipolar disorder and in order to evaluate the economic treatment effect
of second generation antipsychotics in this population, it is important to specify the
econometric models that can address the characteristics of drug therapy for bipolar
disorder.
Specifically, the economic impact of second generation antipsychotics in the
bipolar population enrolled in Medi-Cal will be measured using paid claims database
from Medi-Cal. To accurately measure the economic treatment effect of second
generation antipsychotics in bipolar patients enrolled in Medi-Cal, it is important to
identify the potential issues in using paid claims database in outcomes research and
econometric methods that can address those issues.

USE OF CLAIMS DATABASE IN OUTCOMES RESEARCH
Retrospective claims database has been used extensively in outcomes research.
Retrospective claims data analysis can provide valuable insight to the treatment
outcomes achieved using alternative treatments in a real world setting. Additionally,
outcomes research using large-scale claims database provides superior generalizability
compared to the randomized clinical trials of alternative therapies with relatively small
number of patients. Having large sample is of particular importance in studying rare
clinical events such as hospitalization or attempted suicides. Claims database
typically includes all the direct medical costs paid for by the insurer/employer and
therefore can also be effectively used for measuring economic outcomes of alternative
30
treatment in certain population. However, claims data, by nature, were not created for
the purposed research, and therefore it is important to recognize the potential problems
of claims data analysis and to reduce any bias that may results from these problems.

Self-Selection Problem in Claims Database
Any research proposing to use retrospective claims data to evaluate the
outcomes of alternative treatments or programs must address the challenges created by
the endogeneity of treatment selection. Unlike randomized clinical trials where
treatment selection is an exogenous and random process, the selection of alternative
drug therapy is not a random process in a real world setting and often patients (or
physicians) self-select into treatment. Since patients self-select into a certain
treatment option based on their characteristics and preferences that are often
unobservable, measuring treatment effect using claims database introduces a bias that
should be resolved through proper econometric methods.
The treatment outcomes associated with alternative antipsychotics can be
influenced by various factors that are also related to treatment selection. These factors
may include severity of illness, comorbidities, and family history as well as the
efficacy of the medication. Treatment self-selection bias becomes an acute issue when
these factors cannot be observed by the researcher. When unobserved factors that
affect the treatment selection are also associated with the outcomes of the treatment,
using simple OLS regression to estimate treatment effects will result in inconsistent
estimates. For example, assume that the severity of manic symptoms of patients is
31
negatively correlated with second generation antipsychotic use and positively
correlated with hospitalization risk. Without adjusting for the severity of illness,
standard OLS regression will overestimate the true impact of second generation
antipsychotics on hospitalization risk.

Econometric Tools for Estimating Unbiased Treatment Effect
One of the standard econometric methods frequently applied to measure
unbiased treatment effects involves using instrumental variables. An instrumental
variable, by definition, is correlated with the treatment selection but is uncorrelated
with the outcomes. Thus, when a valid instrument for the treatment selection is
available, instrumental variables method adds exogeneity to otherwise endogenous
treatment selection and recovers a consistent estimator for the local average treatment
effect. However, availability of a valid instrumental variable is limited in many
empirical studies. Also, when the instrument was not highly correlated with the
treatment selection or has a significant correlation with the idiosyncratic error, then the
instrumental variable method can introduce additional bias. Other approaches to
consistently estimate the treatment effect based on the assumption of ignorability of
treatment given observed covariates include propensity scoring method. However, in
many empirical studies, this assumption may not be satisfied.

32
Applying various econometric tools to cross sectional data depends on the
availability and quality of instrumental variables and in many cases involves
constraining assumptions. Cross sectional data structure has some limitations in
comparing treatment outcomes using retrospective claims data. First, it is difficult to
construct a model that incorporates complicated treatment drug use patterns using
cross sectional data. This is particularly acute for bipolar patients who typically use
combinations of medications depending upon the phase of their illness. Also, many
bipolar patients have other mental or physical comorbidities that can influence the
outcomes of drug treatment. With cross sectional data, this kind of information cannot
be easily analyzed. Second, cross sectional data does not utilize the information from
the data set as efficiently as panel data, particularly in terms of resolving issues of
omitted variables problem. When cross sectional data is used, the remedies for
omitted variable/selection bias problems are limited to the conventional methods such
as instrumental variables, whereas other remedies become available with panel data
(Wooldridge 2002, Hsiao 2003).
One approach to incorporating the complex drug use patterns observed in
patients with bipolar disorder is to focus on drug treatment episodes rather than the
individual patient. In this data set-up, observations are defined in terms of drug
treatment episodes rather than an individual patient. Drug treatment episodes start
when patients initiate drug therapy or reinitiate drug therapy after a break in therapy,
or when they switch to or augment with different medication. Outcomes can then be
measured and compared across different treatment groups on a per episode basis. For
33
example, McCombs et al. compared the post 12-month treatment costs associated with
treatment episodes using different antipsychotics (McCombs 2004). Different types of
episodes were defined based on the duration of break in therapy and overlaps with
other related medications. If patients start or reinitiate their drug therapy after a 15 or
more days of break in therapy, then the episode is defined as a restarting episode and
patients are followed as long as they are on related medications without a major break
in therapy. Since patients are followed as long as they are on drug therapy, patients
can switch to or augmented with other medications within a restarting episode.
Switching episodes are defined when patients initiate an additional drug therapy while
the original drug is discontinued within 30 days of the initiation of the second
medication. Augmenting episodes are identified when patients initiate a second
medication while the original medication(s) is maintained for more than 60 days.
Since episodes starts when drug therapy of the patient is initiated, the episode
level data shares a common ground with a randomized clinical trial. However, unlike
randomized clinical trials, where an observation terminates when the patient
discontinues a particular drug of interest, patients can have multiple observations as
they reinitiate therapy. The use of multiple drug treatment episodes per patient
provides more flexibility to model complicated drug use patterns. Also having
multiple observations per patient enables researchers to incorporate additional
information in the model and therefore potentially provides a better solution to the
selection bias problem. However, there are disadvantages of using an episode level
dataset when measuring treatment outcomes. When measuring the economic
outcomes of drug therapy over a certain period, for example, researchers must deal
with overlapping drug treatment episodes and correlation of error terms between
observations. Although the latter problem can be readily dealt with statistically, the
overlapping problems may not be easily resolved.

Panel Data Approach
Panel data provides better solution to the omitted variable problem by
effectively dealing with unobserved heterogeneity. When panel data is available,
unlike cross sectional data, the treatment self-selection problems can be dealt with
more effectively without an instrumental variable or a strong assumption such as the
assumption of observability. Also, panel data does not have the overlapping problems
of the episode level data of episode level data and the correlation of error terms can be
easily dealt with.
A panel data set follows individuals over a period of time and therefore
provides multiple observations per individual, whereas a cross sectional data set
provides a summary snap shot of individuals in the dataset. Thus, panel data can
provide better tools to account for multiple treatment attempts over time and analyze
the outcomes associated with complex drug use patterns. More importantly, panel
data structure enables researcher to build a model that explicitly incorporates a time-
constant, unobserved effect, providing effective solution to the selection bias/omitted
variables problem. Specifically, in the following simple linear model, β cannot be
34
consistently estimated if the unobserved heterogeneity, is correlated with other
explanatory variables, .
i
c
it
x

it i it it
u c x y + + = β 1.1

In the above model, if the Cov ) c x
j
, ( ≠ 0 and , being unobserved, is placed in the
idiosyncratic error term, then the linear estimate of
c
β is inconsistent because the
standard assumption Cov ) = 0 is violated. Within panel data structure,
unobserved heterogeneity can be controlled by either eliminating by demeaning
(fixed effect estimator) or first differencing (fixed difference estimator) or putting it in
the error term and estimating
j j
u x , (
i
c
β based on the consistent estimator of the variance of
error and (random effect estimator).
i
c
Conceptually, the key difference between the random effect model and the
fixed effect model lies on the assumption of the correlation between the unobserved
component and the explanatory variables in the model (Wooldridge 2002, Hsiao
2003). The random effect model assumes orthogonality between and , whereas
the fixed effect estimator does not require the orthogonality to be consistent. In other
words, the random effect estimator requires the following assumption on relation
between two variables to be satisfied to be consistent while fixed effect estimators
allow for a arbitrary correlation between and .
i
c
it
x
i
c
it
x
0 ) ( ) ( = =
i i i
c E x c E 1.2
35

In the current study this assumption of orthogonality is unlikely to be satisfied.
Specifically, the unobserved characteristics of patients are also correlated with
covariates in the model. For example, a patient or a physician decides to use a
particular medication after he observes his severity of illness, which is part of the
unobserved heterogeneity. The implication of this is that the random effect estimator
may not be used in this type of analysis.
Fixed effect estimators (FE) involve transforming equation 1.1. to eliminate
the unobserved heterogeneity, . Transformation can be obtained by averaging
equation 1.1 over the entire time period to get
i
c

it i i i
u c x y + + = β 1.3.

where , ,
11
1 1
∑∑
==
− −
= =
T
t
T
t
it i it i
x T x y T y and
∑
=
−
=
T
t
it i
u T u
1
1
. Subtracting equation 1.3
from equation 1.1 for each time period t eliminates and we get
i
c

it it i it i it
u u x x y y − + − = − β ) ( 1.4

Equation 1.3 can be estimated by pooled OLS. Similarly, can be eliminated using
first differencing transformation. First differencing involves lagging equation 1.1 one
period and subtracting 1.1 so that we can get
i
c
36

1 , 1 , 1 ,
) (
− − −
− + − = −
t i it t i it t i it
u u x x y y β 1.5

Again, equation 1.5 can be consistently estimated using pooled OLS. Fixed effect
estimator with demeaning or first differencing can effectively eliminate the
unobserved heterogeneity without the orthogonality assumption between unobserved
heterogeneity and the explanatory variables.
As the above linear regression example shows, panel data estimators utilize
data more efficiently by creating more data points, enabling researcher effectively deal
with the unobserved heterogeneity. If it can be reasonably assumed that all the factors
affect the selection are either observed or included in the time-invariant unobserved
heterogeneity, then taking care of unobserved heterogeneity using conventional panel
data approach provides a solid solution for the selection bias problem. If, however,
time-varying unobserved explanatory variables are assumed to exist, taking care of
time-invariant unobserved heterogeneity through panel data still leaves
contemporaneous correlation between the explanatory variable of interest (drug
choice, in this study) and idiosyncratic error. The potential simultaneous presence of
unobserved heterogeneity and endogeneity of selection necessitate econometric tools
that control for both of these problems.
In the current study, unobserved factors that affect the selection of drug and
outcomes simultaneously may include severity of illness, family history, prescribing
pattern of the provider, and/or previous experience of patients with various
37
38
antipsychotics. If the study period does not span over an extended period of time, it
can be reasonably assumed that these factors are time-invariant and therefore using
panel data models with fixed effect approach can produce consistent estimate of
treatment effect of second generation antipsychotics.

RESEARCH OBJECTIVES
The first objective of this study is to evaluate the treatment effect of the
second-generation antipsychotics on health care utilization and total healthcare costs
compared to the typical first-generation antipsychotics in the Medi-Cal population
with bipolar disorder. The second objective of this study is to compare the economic
outcomes across different second generation antipsychotics. Panel data models will be
used to address the potential econometric problems arise in measuring the treatment
effect of using retrospective claims database. Models based on cross sectional data
will be also fitted and compared to the panel data models.

39
CHAPTER 2 RESEARCH DESIGN AND METHODS
DATA SOURCE
The retrospective paid claims files from the fee-for-service Medi-Cal program
will be used. The data includes information on all the paid claims for services
delivered between January 1999 and December 2002. Claims made by patients were
retrieved and linked by encrypted patient identifiers.
Information available from the Medi-Cal paid claim files includes
demographic information of the patient, information of the physician, information on
the claims of medical and pharmacy services made by the patients. Demographic
information available from the data includes age, gender, race, county of residence,
and Medicare eligibility. As for the medical services, type of services, date of service,
amount paid, and days of services (e.g. hospital days) are included. Pharmacy claims
was used to derive the information on the type of medication dispensed, date of the
prescription fill, strength, dose, and days of supply, and costs.

STUDY POPULATION
California Medicaid patients with bipolar disorder were identified based on
two criteria: 1) at least one claim with diagnosis of bipolar identified with
corresponding ICD-9 codes (296.40-296.99) and 2) at least one filled prescription for
mood stabilizer, antipsychotic, antidepressant, or anticonvulsant. Study population
was selected using following steps. First, patients who had an antipsychotic
medication between 2000 and 2001 were selected. The date of any first antipsychotic
40
medication fill was identified as the index date. Once the index date was identified,
patients who were continuously eligible for the Medicaid benefits prior and post 12
months of the index date were selected. Since patient eligibility information was not
available, patients were assumed to have lost eligibility if they did not have any claims
for either medical or pharmacy services for any six consecutive months during the two
year period and were excluded from the analysis. This was based on an assumption
that patients with a severe chronic condition such as bipolar disorders can be
reasonably assumed to utilize healthcare services at least once in six months and if
they do not have any pharmacy or medical claims within a six month period, it is due
to a loss of eligibility. Patients who were younger than 18 years at the index date
were excluded from the study.
Patients were further excluded from the study if they had any claims of
antipsychotics in the one year pre index date period. This was done to minimize the
differences in characteristics of patients in different treatment groups. Unlike first-
generation antipsychotic medications, the second-generation antipsychotic
medications are also prescribed for indications other than acute manic symptoms and
comparing any second-generation antipsychotic use with first-generation antipsychotic
may limit the clinical interpretation of the results. Limiting the study population to
those who initiated the antipsychotic therapy (genuine initiation or re-start after a gap
≥ 360 days) may limit the study population to those who have new onset of acute
manic symptoms and therefore ensuring the comparability of the two populations.

41
TYPE OF ANTIPSYCHOTICS INCLUDED IN THE STUDY
Following antipsychotic medications were categorized as typical
antipsychotics: chlorpromazine, fluphenazine, fluphenazine decanoate, haloperidol,
haloperidol decanoate, loxapine, molindone, perphenazine, pimozide,
prochlorperazine, and serentil. Second generation antipsychotic medications include
olanzapine, risperidone and quetiapine. Clozapine and ziprasidone were not included
in the analysis due to the small number of patients on these medications. Apriprazole
was not available on the market at the time of the study.

DATA STRUCTURE
A yearly panel dataset that spans from January 2000 to December 2001 was
created and therefore each patient has 2 observations. Each observation has
information of the drug therapy within the year, other health care services use, and
time variant patients characteristics such as comorbidities. Total health care costs
were calculated based on prescription drug costs, hospitalization costs, Medicare Part
B costs, long-term care costs, community mental health center costs, and other costs
that are covered by Medi-Cal. Hospitalization costs were calculated based on the
number of hospital days incurred within each period using the average of per-diem
cost reported for inpatient services by Medi-Cal ($1053 per hospital day, Annual
Statistical report 2003). Long-term care costs and rehabilitation center costs were
calculated in the same manner ($270 and $634 per day, respectively). Many patients
with chronic mental disease are dually eligible for Medicare and Medicaid and for
42
Medicare Part B costs, what appears in the dataset does not reflect the actual
healthcare costs incurred by the dually eligible patients. In order to address this
problem, health care costs for services that are covered by Medicare Part B are inflated
five times for the dually eligible patients. Dual eligibility is defined based on the dual
eligibility indicator field from Medi-Cal data and also on the age of patient and a
patient was defined to be dually eligible if either a patient is dually eligible based on
the dual eligibility indicator or the patient is over 65 of age at the time of service.

OUTCOME MEASURES
Following clinical and economic outcomes will be measured and compared between
the two classes (typical and second-generation antipsychotics). In addition, the same
set of outcomes will be evaluated across different second-generation antipsychotics for
head-to-head comparisons. The outcomes will be compared using panel data analyses
as well as simple univariate analyses. Univariate analyses included parametric and
nonparametric tests based on the distributional characteristics of the outcome
variables.

Healthcare Utilization
The impact of second generation antipsychotics on health care utilization will
be evaluated using following outcome measures: mean number of hospitalizations per
year, mean days of hospital stays per year, mean number of suicide attempts per year
and mean days of long-term care facility stays per year.
43

Economic Outcomes
Economic outcomes are measured as the health care costs incurred during the
one year period after the initiation of antipsychotic therapy. Health care costs are
defined as the amount Medi-Cal paid for the services rendered to the patients and
calculated from the paid claims for each service. Total health care costs as well as
itemized healthcare costs that are broken down by type of services will be analyzed.
Itemized healthcare costs include pharmacy costs, total hospitalization costs, long-
term care costs, ambulatory care costs and net costs defined as total healthcare costs -
pharmacy costs.

Antipsychotic Medication Utilization Patterns
The duration of antipsychotic medication therapy will be calculated. First, the
duration of uninterrupted therapy after the initiation of therapy will be calculated and
compared between the study groups. If there was an overlap due to early refills, then
the overlapped days were added to the total duration. In addition days covered with an
antipsychotic medication in the year following the initiation of therapy will be
calculated.
In addition, rate of switching to or augmenting with other antipsychotic
medication(s) within 4 weeks of the initiation of therapy was evaluated and compared
across study groups.

ECOMETRIC MODELS
Selection of different antipsychotic classes was modeled using logistic
regression controlling for confounding factors. Similarly, multinomial logit model
was used to estimate the impact of different patient characteristics on selection of
different second-generation antipsychotic medications. Logistic regression was used
to compare the rate of switch to and augmentation with different antipsychotic
medication different antipsychotic classes.
Different panel data approaches were taken based on the distribution of the
outcomes variables of interest. Basic linear fixed effects panel data models were fitted
for continuous variables such as total and itemized healthcare costs. Fixed effects
Poisson regression models were used for count variables such as the number of
hospitalization and days of hospital stays.

Fixed Effect/First Difference Models for Healthcare Costs
With two time periods as in the current study, fixed effect panel data analyses
produce identical results as first differencing estimation, and therefore can be
interpreted as the difference-in-difference estimator of the average treatment effect of
different antipsychotic medication therapy. Let be the binary indicator for
different antipsychotic medication class for year 1 (t=1) and year 2 (t=2), other
exogenous variables, outcome measures, idiosyncratic errors. is unobserved
it
therapy
it
Z
it
y
it
u
i
c
44
individual heterogeneity. The basic linear regression model for outcome can be
written as
it
y

it i it it it
u c Therapy z y + + + = γ β 2.1

First differencing of above equation produces

it it it it
u Therapy z y Δ + Δ + Δ = Δ γ β 2.2

Because we selected patients with no filled prescription for antipsychotic medications
in year 1, = 0 for all i . In the second period, = 1 for the patients
on second generation and = 0 for those who were on typical antipsychotic
medications. Therefore, when T=2 equation 2.2 can be rewritten as
1 i
Therapy
2 i
Therapy
2 i
Therapy

it it it it
u Therapy z y Δ + + Δ = Δ γ β 2.3

In other words, the treatment effect of the second-generation antipsychotic
medications can be estimated by regressing the changes in outcomes on changes in the
set of exogenous variables and the treatment indicator.

45

Fixed Effect Poisson Estimation for Count Variables
Some of the outcome variables such as the number of hospitalization or days of
hospital stay are nonnegative integers limiting the applicability of conventional linear
models. To fit models for count data such as these, the Poisson distribution can be
commonly used. As with linear panel data models, the panel data approach provides
unique opportunity to control for unobserved heterogeneity. In this research, the panel
data method with count data models proposed by Hausman, Hall and Griliches was
employed to address the issues around unobserved heterogeneity and its relationship
with (Hausman 1984). Specifically, as with the linear models, fixed effects
analyses was done to allow for random dependence between estimation of the
treatment effect given unobserved heterogeneity. Let be the outcome variable of
interest for individual i at time (i.e. number of event for individual i at time ).
Hausman et al. used the following specification
i
X
it
n
t t

) exp( ) (
i it it it
c x n E + = = β λ 2.4
)) (exp( ~
i it it
c x Poisson n + β 2.5

where β is a vector of parameters and are explanatory variables. represents
unobserved individual heterogeneity. Let , then since is Poisson
distributed, is also Poisson distributed with parameter
it
x
i
c
∑
=
=
T
t
it i
n n
1
it
n
i
n
∑
=
=
i
T
t
it i
1
λ λ . Also, if two
46
independent random variables are Poisson distributed with means
a
μ and
b
μ
respectively and
b a
K μ μ + = , then the random variable given K has binomial
distribution with probability parameter
b a
a
μ μ
μ
+
(Arellano 2001). Expanding this to
multiple random variables drawn from equation 2.5 where ,
then it follows
iT i i
n n n K ,. ,
2 1 ∑
=
=
T
t
it i
n n
1

)) , ( , ), , ( , ( ~ , ,
1
β β
i T i i i i i it
x p x p n l nultinomia c x n n K 2.6
where
∑
=
+
+
≡
T
t
i it
i it
i T
c x
c x
x p
1
) exp(
) exp(
) , (
β
β
β 2.7

Individual heterogeneity drops out in equation 2.7 and therefore equation 2.6
becomes multinomial distribution conditional only on and . Since equation 2.7
follows multinomial logit specification,
i
c
i
n
i
x
β can be estimated by conditional maximum
likelihood methods. Specifically, the fixed effect Poisson estimator maximizes the
following log likelihood function:
β
ˆ

2.8 () ( )
∑∑ ∑ ∑∑
== =
− −
==
− + Γ =
N
i
T
t
T
s
x x
it
N
i
T
t
it
is it
e n n L
11 1
) (
11
ln 1
β
β

47
48
Multivariate Models
The model that evaluates the treatment effect of second generation
antipsychotics in the Medi-Cal bipolar population should incorporate a variety of
explanatory variables that describe the complex treatment patterns and comorbidities.
Demographic characteristics such as age, gender, race, eligibility categories, and rural-
urban indicator were included in the selection model. The Charlson index was
calculated as a measure of risk-adjustment (Charlson 1987). Specifically, the Deyo
approach based on ICD-9 codes was implemented. Other psychiatric comorbid
conditions that may have direct implications for the outcomes of the antipsychotic
therapy such as schizophrenia and substance abuse were also included in multivariate
modeling.
Pharmacotherapy of bipolar disorder is very complicated and it is important to
incorporate the complex treatment pattern into the model so that the treatment effect of
the antipsychotics can be accurately estimated. Polypharmacy is a common practice in
bipolar disorder. Preliminary data analysis shows that about 60% of antipsychotic
treatment episodes were combination therapy with either a mood stabilizer or an
antidepressant. Failure to control for the simultaneous use of antidepressant and mood
stabilizers may result in a biased estimate of the treatment effect of antipsycthotics.
Variables that capture various treatment patterns will be created and included in the
model. These include indicators for the concurrent use of mood stabilizer and
antidepressant.

Specifically, the following models will be estimated.

Model 1 Comparison between Typical vs. Second-generation Antipsychotics

it i it it it
u c AAP z y + + + = γ α 2.1

where t corresponds to quarters in the study period. is the cost variables including
total healthcare costs and net costs (total costs – medication costs). is a vector of
exogenous variables including time variant covariates such as age, comorbidities,
indicators for other types of drug use including mood stabilizers and antidepressants,
and time dummies and
it
y
it
z
β is a vector of corresponding parameters. The reference
group will be typical antipsychotics and the parameter of interest is the dummy
variable for the second generation antipsychotic use in time and t γ . is
unobserved heterogeneity and the is the idiosyncratic error.
i
c
it
u

Model 2 Comparison of the Impact of Duration of therapy between Typical vs.
Second-generation Antipsychotics

Second model to be estimated involves the economic treatment effect of
second-generation antipsychotics controlling for the duration of therapy. As
mentioned earlier, studies have found that patients stay on antipsychotic therapy even
after manic symptoms are controlled. With a favorable side-effect profile and an
indication as a maintenance therapy, patients who are treated with second-generation
49
antipsychotics are much more likely to have antipsychotic therapy extended beyond
the initial manic phase (Keck 1996, Gianfrancesco 2008). However, the impact of the
extended length of therapy has not been clearly understood. It can be argued that the
extended antipsychotic therapy benefit patients by extending the stabilization of manic
symptoms and by preventing recurrence of manic episodes, i.e. being used as a mood
stabilizer. However, few studies have shown significant clinical or economic benefits
of second-generation antipsychotics over a mood stabilizer as the maintenance therapy
in a real world setting, undermining the argument for benefits of extended
antipsychotic therapy. In order to address the potential differences in the use of
different classes of antipsychotic medications beyond the manic phase, the duration of
therapy will be controlled for second model. By controlling for the duration of
therapy, the comparative treatment effects without the confounding effect of the
extended therapy can be estimated. The model can be written as following:

it i it it t i it it
u c DUR TAP DUR AP y z y + + + + + =
−
_ _
1 ,
γ β α 2.2

where is the duration of second generation antipsychotic use in
time t . indicates the duration of typical antipsychotic use in time t .
it
DUR AAP _
it
DUR TAP _

Model 3 Second-generation Head-to-Head Comparison
The economic outcomes will also be compared among three different second
generation antipsychotics using following model:
50

it i it it t i it it
u c QUET RISP y z y + + + + + =
−
ϕ γ β α
1 ,
2.3

where and indicate use of risperidone and quetiapine in time
respectively, and the reference group is the use of olanzapine in time .
it
RISP
it
QUET t
t
The last model compares the economic treatment effect of different second
generation antipsychotics.

Model 4 Second-generation Head-to-head Comparison Controlling for Duration
of Therapy

it i it it it t i it it
u c dur AAP QUET RISP y z y + + + + + + =
−
δ ϕ γ β α _
1 ,
2.4

where and indicate use of risperidone and quetiapine in time
respectively. indicates the duration of each second-generation
antipsychotic therapy in time respectively.
it
RISP
it
QUET t
it
dur AAP _
t

Statistical Software
SAS® statistical software version 8.2 (SAS, Cary NC) and STATA version 9
SE (STATA Corp. College Station TX) were used for the process of data and the
estimation of the econometric models.

51
52

CHAPTER 3 RESULTS
DEMOGRAPHIC CHARACTERISTICS
There were 33,128 patients with bipolar disorders with at least one claim of an
antipsychotic medication between January 2000 and December 2001. The first fill
date of antipsychotic medication in this time period is identified as index date.
Patients who did not have an antipsychotic medication in the previous year were
selected and included for further identification of the study population. Of the 33,128
patients initially identified, 8,324 patients did not have an antipsychotic prescription
fill during the previous 1 year period and therefore were entered into further analyses.
Patients were further screened for age (age ≥ 18) and continuous eligibility for the
Medi-Cal program during the two year observation period. After the application of
these exclusion criteria, 2,761 patients were identified as the bipolar patients who
initiate a new antipsychotic therapy and therefore entered into the final analyses.

COMPARISONS OF FIRST-GENERATION vs. SECOND-GENERATION
ANTIPSYCHOTICS

Description of the Study Population
Table 1 shows the descriptive statistics of demographic characteristics of the study
population. Of the 2,761 patients, 61.7% was female and the mean age was 45.8 (SD
13.9). The ethnic make-up of the study population included: 59.7% Caucasian, 14.7%
African American, 4.3% Hispanic, 1.6% Asian and 19.8% from other races.
The majority of the study population resides in an urban counties (66.9%),
followed by mixed (22.6%) and rural (10.5%) counties. Seventy five percent of the
study population was eligible for the Medi-Cal program due to disability. Other
eligibility criteria included Old Age Assistance (OAA, 2.5%), Aids to Families with
Dependent Children (AFDC, 15.6%), blindness (0.4%) and others (6.8%).

Table1. Demographic Characteristic of Study Sample

Characteristics Atypical
(n =1835 )
Typical
(n =926 )
p All
(n = 2761)

45.8
(SD =13.8)
190 (6.9%)
404 (14.6%)
755 (27.3%)
783 (28.4%)
406 (14.7%)

44.9
(SD= 13.8)
(7.9%)
226(16.1%)
426 (30.3%)
375 (26.7%)
184 (13.1%)
85 (6.04%)

47.3
(SD= 13.8)
26 (5.0%)
69 (13.2%)
153 (29.3%)
138 (26.4%)
83 (15.9%)
54 (10.3%)

< 0.0001

0.03
0.12
0.66
0.91
0.11
0.001
Age
Mean

18-24
25-34
35-44
45-54
55-64
223 (8.1%) Over 65
Gender
Male

746 (40.7%)
1,089 (59.3%)

311 (33.6%)
615 (66.4%)

0.0003

1,057 (38.3%)
Female 1,704 (61.7%)
Residence
Urban
Rural

1,835 (69.4%)
175 (9.5%)
387 (21.1%)

574 (61.9%)
114 (12.3%)
238 (25.7%)

0.05
0.11
0.37

1,847 (66.9%)
625 (22.6%)
289 (10.5%) Mixed

1,109 (60.4%)
258 (14.0%)
89 (4.9%)
37 (2.0%)
342 (18.6%)

536 (57.8%)
147 (15.9%)
30 (3.2%)
8 (0.9%)
205 (22.1%)

0.03
0.008
0.14
0.2
0.09

1,645 (59.6%)
405 (14.7%)
119 (4.3%)
45 (1.6%)
547 (19.8%)
Race
White
Black
Hispanic
Asian
Others

0.028
0.0008
0.3
0.23

262 (18.6%)
26(1.9%)
2(0.14%)
1,006 (71.5%)
53
1
Aid to Families with Dependent Children

111 (7.9%)

75 (14.3%)
24 (4.6%)
2 (0.4%)
391 (74.8%)
31 (5.9%)
Eligibility
AFDC
1
OAA
2
Blind
Disabled
Other aid 0.14

431 (15.6%)
(2.6%)
4 (0.2%)
1397(72.4%)
142 (7.4%)
2
Old Age Assistance

54

Table 1 also compares the demographic characteristics by the class of index
antipsychotic medication. For descriptive statistics, student t-tests were used for
continuous variables such as age. Chi-square tests were used for categorical variables
such as race and gender. Patients who were on typical antipsychotics were
significantly older than their peers in the second generation antipsychotic group (47.3
vs. 44.9, p < 0.0001). The proportion of female patients was higher in the typical
antipsychotic group than in the second-generation group (59.3% vs. 66.4%). In the
second-generation group, 60.4% was Caucasian, 14.0% was African-American, 4.9%
was Hispanic, 2.0% was Asian, and 18.6% were other races. Among the patients in
the typical antipsychotic group, 57.8% was Caucasian, 15.9% was African-American,
3.2% was Hispanic, 0.9% was Asian, and 22.1% were other races. The proportion of
Caucasian in the second-generation group was significantly higher than in the typical
group (p = 0.03) whereas the proportion of African-American was significantly higher
in the typical group then in the second-generation group (p=0.008). The Medi-Cal
eligibility criteria between the two groups were also different. In the second-
generation group, 1.9 % of the patients were eligible for Medi-Cal due to OAA
whereas OAA accounted for 4.6% of the patients in the other group (p=0.008).
Patients who were eligible for Medi-Cal due to AFDC accounted for 18.6% in the
second-generation group and 14.3% in the typical antipsychotics group (p=0.03). In
both groups, the majority of the patients were eligible due to the disability (71.5% in
the second-generation group and 74.8% in the typical group, p =0.23).
55
Table 2 presents the baseline healthcare costs (total and itemized). Patients in
the typical antipsychotic group had a significantly higher total healthcare costs
compared to patients in the second-generation group ($18,191 vs. $13,648, p <
0.0001). When broken down by the type of service, patients in the typical
antipsychotic group had a significantly higher outpatient costs ($1511 vs. $1296, p =
0.04), long-term care costs ($ 6411 vs. $3432, p = 0.0001), and other costs ($365 vs.
$233, p = 0.008) compared to the patients in the second-generation group. Patients in
the second-generation group had slightly higher prescription drug costs, community
mental health center costs, rehabilitation costs; however, none of them were
statistically significantly different.

Table 2 Baseline Health Care Costs by Drug Type

Health Care Costs Second
Generation
Typical p*

All
Long Term Care

$3432
(SD=16817)
$ 6411
(SD=22707)
0.0001 $ 4431
(SD=19045)
Hospitalization $2685
(SD=14366)
$3785
(SD=23617)
0.20 $3049
(SD=17964)
Rehabilitation $ 14
(SD=207)
$ 7
(SD=102)
0.31 $ 12
(SD=179)
Psychologist $ 3
(SD =37)
$ 1
(SD=14)
0.21 $ 3
(SD=32)
Prescription Costs $ 3539
(SD=8379)
$ 3486
(SD=4738)
0.86 $3521
(SD=7360)
Community
Mental Health
Center
$ 2555
(SD=5125)
$ 2417
(SD=5338)
0.51 $ 2509
(SD=5197)
Other Costs $ 233
(SD=1019)
$ 365
(SD=1563)
0.008 $277
(SD=1230)
Outpatients $ 1296
(SD=2651)
$ 1511
(SD=2746)
0.04 $ 1368
(SD=2685)
Total Health Care
Costs
$ 13648
(SD=24535)
$ 18191
(SD=33685)
<.0001 $15172
(SD=28016)
* based on student t-tests
56
Table 3 shows the health care utilization patterns including in the previous year
by drug type. Patients in the typical antipsychotic medication group had significant
more days of long-term care stays compared to their peers in the second-generation
group (23.66 vs. 12.83, p = 0.0002). Patients initiated an second-generation therapy
had a significantly greater number of suicide attempts in the baseline year (0.09 vs.
0.12, p = 0.04). There was no significant difference in terms of number of
hospitalization and the days of hospital care between the two groups.

Table 3 Baseline Health Care Utilization by Drug Type

HealthCare
Utilization
Second
Generation
(n = 1845 )
Typical
(n =926 )
p All
(n = 2761)
Number of
Hospitalizations
0.68
(SD =1.08)
0.63
(SD=1.08)
0.28 0.66 (SD=1.08)
Days of Hospital
Stay
2.55
(SD=13.64)
3.69
(SD = 22.42)
0.19 2.89
(SD= 17.06)
Days of Long-term
Care stays
12.83
(SD=62.61)
23.66
(SD=83.90)
0.0002 16.41
(SD = 70.54)
Number of Suicide
Attempts
0.12
(SD=0.42)
0.09
(SD=0.38)
0.04 0.11
(SD=0.41)

Factors Affecting Treatment Decision
Table 4 shows the results from the logistic regression with the class of
antipsychotic medication as the dependent variable. Females were less likely to be
prescribed for a second-generation antipsychotic medication than males (OR: 0.72;
95% CI: 0.60 – 0,86). African-American patients (OR: 0.78; 95% CI: 0.61-0.99) were
significantly less likely to be treated with second-generation antipsychotics than
Caucasian patients. Patients with other ethnic backgrounds including Hispanics and
57
Asians were also less likely to use second-generation antipsychotic medications than
Caucasians. However, the results were not statistically significant (OR: 0.86; 95% CI:
0.71-1.05). Patients living in urban areas were significantly more likely to be on
second-generation antipsychotics compared to their peers in a rural area (OR: 1.54;
95% CI: 1.17-2.02). Generally, the impact of age was not significant in any of the age
groups compared to the patients younger than 35 years.
Patients who had a diagnosis of bipolar disorder, depressed (ICD-9-code:
296.5) or bipolar disorder, atypical (ICD-9-code: 296.8) in the previous year were
significantly more likely to be prescribed with a second-generation antipsychotic
medication compared to the patients who did not have these diagnosis (OR: 1.78; 95%
CI: 1.33-2.37 and OR: 1.63; 95% CI: 1.20-2.20, respectively). Having one or more
previous diagnosis of major depressive disorder (MDD, ICD-9-code 296.2-296.3) was
also associated with an increased likelihood of second-generation antipsychotic use
(OR: 1.54; 95% CI: 1.14-2.07) while existence of other mental disorders including
schizophrenia, mania, and phobia, anxiety, and personality disorders did not have
significant impact on treatment selection. Existence of other co-morbid conditions
measured as the Charlson co-morbidity index was a significant factor. Patients with
the higher Charlson co-morbidity index (i.e. having more co-morbid conditions) were
significantly less likley to be prescribed a second-gereration antipsychotics (OR: 0.93;
95% CI: 0.88-0.98). Patients who had long-term care use in the baseline period were
significantly less likely to initiate second-gerenration antipsychotic treatment (OR:
0.69; 95% CI: 0.50 - 0.97).

Table 4 Logistic regression model for the selection of antipsychotic medications

Effect Odds Ratio
95% Confidence
Interval
Female 0.721 0.602 - 0.862

urban 1.540
1.009
1.173 - 2.021
0.746 - 1.364
58

mixed

35 ≤ age < 45
45 ≤ age < 55
55 ≤ age < 65
0.869
0.898
0.835
0.882
0.680 - 1.112
0.701 - 1.152
0.625 - 1.116
0.588 - 1.325

65 ≤ age
African-American 0.775 0.606 - 0.990

Others 0.862 0.709 - 1.047

Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
1.160
1.778
1.091
1.026
0.916 - 1.471
1.333 - 2.371
0.839 - 1.419
0.795 - 1.324

Bipolar Disorder - Others 1.625 1.200 - 2.201

OAA
Disabled
AFDC
1.805
1.755
2.270
0.495 - 6.586
0.537 - 5.735
0.677 -7.612

Others 1.969 0.573 -6.764
Charlson Index
Major Depressive Disorder
Schizophrenia
Long-term care use
Hospitalization
Substance abuse
Other mental disorders
Manic
0.931
1.535
1.163
0.692
0.828
1.252
1.046
1.492
0.961
0.883 - 0.982
1.137 - 2.071
0.972 - 1.391
0.497 - 0.965
0.671 - 1.020
0.981 - 1.599
0.839 - 1.304

0.876 - 2.539
Prescription drug use 0.918 - 1.007

59
Duration of therapy
The mean durations of the initial continuous therapy for typical antipsychotic
medication and second-generation antipsychotic medications were 62.9 (SD=153.4)
and 115.0 (SD=216.01), respectively (p < 0.0001)(Table 5). The total days a patient
had the initial antipsychotic available was calculated by summing days of
uninterrupted therapy episodes during the one year period after the initiation of the
antipsychotic therapy and capped at 365 days. The average days covered by the initial
first-generation antipsychotics for the first year after the initiation of therapy was
102.5(SD=113.7). The average days covered was 177.1 (SD=124.9) for the second-
generation antipsychotics (p < 0.0001). The long-term antipsychotic therapy was
defined as continuous therapy with the initial antipsychotic medicine ≥ 180 days. The
proportion of patients who had a long-term antipsychotic therapy was significantly
higher in the second-generation group than in the typical group (37.6% vs. 9.2%, p <
0.0001).

Table 5 Duration of Therapy and Rate of Switch/Augmentation by Drug Type

Second Generation Typical P
Initial Duration of
Therapy
115.0
(SD=216.01)
62.9 (SD=153.4) < 0.0001
Total Duration of
Therapy (per year)
177.1
(SD=124.9)
102.5
(SD=113.7)
< 0.0001
Proportion with initial
duration of therapy >
6months
37.6% 9.2% < 0.0001
Proportion Swt/Aug
within 30 days
4.4% 5.9% 0.07

60

Switching to and/or Augmenting with Other Antipsychotics
Table 5 also shows the proportion of patients who switched to or augmented
with a different antipsychotic medication within 30 days after the initiation of
antipsychotic therapy. More patients on typical antipsychotics switched to or
augmented with a second antipsychotic medication within 30 days of initiation of
therapy than patients on second-generation antipsychotic medication (5.9% vs. 4.4%,
p=0.07). Logistic regression models were fitted to evaluate the association between
use of second-generation antipsychotic medications and risk of switching to or
augmenting with another antipsychotic medication within 30 days of initiation of
therapy in comparison to typical antipsychotic medication use (Table 6). Other
independent variables controlled for in the model include patient age, gender, race, co-
morbidities, type of bipolar diagnosis at baseline, and Medicaid eligibility category.
Compared to patients on typical antipsychotics, patients used second-generation
antipsychotic medications were significantly less likely to switch to or augment with
another antipsychotic medication (OR: 0.67; 95% CI: 0.46 - 0.97).

61
Table 6 Likelihood of Switch/Augmentation: Logistic Regression

Odds Ratio 95% Confidence Limits
Second-generation use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Major Depression
Schizophrenia
Previous long term care
Previous hospitalization
Substance abuse
Other mental disorders
0.67
1.17
1.02
0.93
0.60
0.54
0.69
1.04
0.82
0.82
1.04
0.86
0.82
1.50
1.21
0.94
0.51
2.37
0.97
1.46
1.57
0.94
0.46 - 0.97
0.80 - 1.71
0.57 - 1.81
0.49 - 1.79
0.37 - 0.99
0.33 - 0.90
0.38 - 1.23
0.49 - 2.21
0.48 - 1.40
0.53 - 1.28
0.63 - 1.70
0.46 - 1.62
0.45 - 1.51
0.92 - 2.44
0.68 - 2.15
0.83 - 1.07
0.23 - 1.11
1.63 - 3.43
0.48 - 1.98
0.96 - 2.20
0.99- 2.51
0.59 - 1.50

Comparison of Healthcare Costs
Descriptive Statistics
Table 7 shows descriptive statistics of the healthcare costs in year 2 (12 months
following the initiation of antipsychotic therapy). While the typical antipsychotic
group had a significantly higher total healthcare costs in the baseline period, the
second-year total healthcare costs were not statistically different between the two
groups ($20,585 vs. $18,801, typical and second-generation antipsychotics,
respectively, p=0.121). Conversely, pharmacy costs were significantly higher in the
62
second-generation group than in the typical group in the post period ($5690 vs. $4834,
p=0.0013) while the difference was not statistically significant in the baseline ($3539
vs. $3489, p = 0.86). Patients in the typical antipsychotic group still had significantly
higher long-term care costs ($7339 vs. $4562, p=0.0004), outpatient costs ($2793 vs.
$2391, p=0.042), and other healthcare costs ($298 vs. $403, p = 0.05) in the second
year.

Table 7 Health Care Costs by Drug Type – 1 Year Post

Health Care
Costs
Second
Generation
Typical p*

All
Long Term Care

$4562
(SD=18556)
$ 7339
(SD=24222)
0.0004 $ 5494
(SD=20668)
Hospitalization $ 2574
(SD=16077)
$2242
(SD=9667)
0.500 $2464
(SD=14279)
Rehabilitation $ 9
(SD=139)
$ 12
(SD=165)
0.715 $ 10
(SD=148)
Psychologist $ 4
(SD =57)
$ 2
(SD=31)
0.273 $ 3.6
(SD=50)
Prescription
Costs
$ 5690
(SD=7261)
$ 4834
(SD=6264)
0.0013 $5402
(SD=6953)
Community
Mental Health
Center
$ 3253
(SD=6594)
$ 2988
(SD=7533)
0.363 $ 3164
(SD=6923)
Other Costs $ 298
(SD=1197)
$ 403
(SD=1360)
0.047 $333
(SD=1255)
Outpatients $ 2391
(SD=5071)
$ 2793
(SD=4806)
0.042 $ 2526
(SD=4986)
Total Health Care
Costs
$ 18801
(SD=27918)
$ 20585
(SD=28787)
0.121 $19399
(SD=28220)

Panel Data Fixed-Effect Model for Medical Costs
Tables 8-15 show the results of the panel fixed-effect models for the
comparative economic treatment effects of second generation antipsychotics and
typical antipsychotics controlling for various confounding factors and unobservables.
63
Covariates included in the models include: index of comorbidities using Charlson
Index, diagnosis of different bipolar subtypes, diagnosis of other affective mental
disorders such as schizophrenia, alcohol and drug dependency and use of other bipolar
medications such as mood stabilizers and antidepressants. The results indicate that
second generation antipsychotic use was associated with significantly higher total
health care costs compared to typical antipsychotic use (Table 8).

Table 8 Panel Data First Difference Model: Total Healthcare Costs

Total Healthcare Costs
Parameter
Estimates P
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
2366.0
2878.6
1900.4
1315.8
1768.8
4267.2
2029.1
1414.5
1112.9
2745.5
2184.7
1726.0
0.028
0.002
0.092
0.254
0.165
0.009
0.000
0.205
0.353
0.000
0.027
0.038

Specifically, second-generation antipsychotic use was associated with increase of total
healthcare costs by $2366 in the second year compared to typical antipsychotic
medication use (p = 0.03). Compared to patients in the typical antipsychotic
medication group, patients with second-generation antipsychotic had a significant
increase in prescription drug costs ($770.9, p < 0.0001) (Table 9). Medical costs
(defined as the total costs – pharmacy costs) were higher in the second-generation
64
group compared to the typical group ($1687.0). However, the difference was not
statistically significant (p = 0.09)(Table 10).

Table 9 Panel Data First Difference Model: Prescription Costs

Prescription Costs
Parameter
Estimates p
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
770.9
501.5
-638.1
-339.9
-212.8
209.2
410.1
216.5
-425.2
742.7
734.9
235.3
0.000
0.103
0.151
0.398
0.498
0.594
0.006
0.447
0.049
0.000
0.001
0.257

Table 10 Panel Data First Difference Model: Total Medical Costs

Total Medical Costs
Parameter
Estimates P
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
1687.0
2457.4
2302.3
1614.2
1803.1
3995.5
1402.0
959.9
1266.2
1947.8
1408.2
1451.2
0.090
0.019
0.022
0.191
0.080
0.004
0.001
0.319
0.239
0.005
0.172
0.147

65

Comparisons of itemized healthcare expenditures are displayed on Table 11 -
15. Total hospitalization costs increased more in the second-generation group but the
difference was not statistically significant (p=0.07)(Table 11). The comparative
impact of second-generation antipsychotic medication use on other types of health
care costs including long-term care costs (252.4, p = 0.63), outpatients services costs
(part B costs, -175.2, p =0.36), and community mental healthcare facility costs ( -6.4,
p = 0.98) was not statistically and/or clinically significant (Table 12 – Table15).

Table 11 Panel Data First Difference Model: Hospitalization Costs

Hospitalization Costs
Parameter
Estimates p
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
1501.3
878.8
653.6
2.5
680.7
3124.4
618.4
274.6
651.3
730.7
-73.9
-4.6
0.067
0.134
0.295
0.997
0.364
0.033
0.015
0.661
0.399
0.122
0.929
0.993

66

Table 12 Panel Data First Difference Model: Outpatient Costs (Part B)

Part B Costs
Parameter
Estimates p
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
-175.2
131.3
503.2
115.5
506.0
225.1
587.9
453.4
659.4
73.4
140.2
247.1
0.356
0.463
0.284
0.609
0.002
0.355
0.000
0.021
0.001
0.759
0.481
0.238

Table 13 Panel Data First Difference Model: Long-term Care Costs

Long-term Care Costs
Parameter
Estimates P
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
252.4
243.1
588.7
263.0
221.9
-459.5
450.1
-485.8
59.5
517.8
944.4
260.4
0.629
0.678
0.276
0.734
0.725
0.382
0.051
0.240
0.886
0.128
0.045
0.672

67
Table 14 Panel Data First Difference Model: Community Mental Healthcare
Facility Costs

Community Mental Healthcare
Facility
Parameter
Estimates p
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
-6.4
1125.7
769.6
1251.8
584.9
1127.3
-102.6
969.6
252.5
635.1
501.4
981.1
0.979
0.001
0.011
0.002
0.047
0.001
0.032
0.003
0.462
0.007
0.053
0.000

Table 15 Panel Data First Difference Model: Other Costs

Other costs
Parameter
Estimates p
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
34.2
-0.7
24.8
1.6
6.0
10.6
63.0
3.7
-93.8
16.5
-46.2
9.1
0.493
0.990
0.705
0.977
0.866
0.862
0.000
0.923
0.068
0.755
0.388
0.877

To see whether the impact of different antipsychotic medication use was
associated with the duration of therapy, days of therapy was included in the model
(Table 16). The duration of therapy of second-generation antipsychotic medication
use was associated with significant increase in total healthcare costs (8.9 per additional
68
day of therapy, p=0.0005) while the duration of typical antipsychotic therapy was not
statistically significant reduction in healthcare costs (- 4.5 per additional day of
therapy, p = 0.34).

Table 16 Panel Data First Difference Model: Impact of the Duration of Therapy
on Total Healthcare Costs

Total Healthcare Costs
Parameter
Estimates p
Duration of therapy (SG)
Duration of therapy (typical)
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder V
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
8.9
-4.8
2941.2
1881.0
1435.7
1876.3
4390.7
2025.6
1254.2
898.0
2411.4
2438.6
1787.4
0.005
0.346
0.009
0.114
0.236
0.052
0.000
0.000
0.208
0.314
0.007
0.004
0.071

To investigate whether the negative economic outcomes of second-generation
antipsychotic medications was due to the extended therapy observed in this
population, the panel data model for the total healthcare costs was fitted using patients
did not have a long-term antipsychotic therapy (i.e., antipsychotic therapy ≤ 6 months)
(Table 17). When only the “short-term” users were compared, total healthcare costs of
the patients who were treated with second-generation antipsychotic medications were
not statistically different from the patients who were treated with typical
antipsychotics (1444, p = 0.15).

69
Table 17 Panel Data First Difference Model: Total Healthcare Costs excluding
patients with extended antipsychotic therapy

Total Healthcare Costs
Parameter
Estimates p
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder V
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
1444
3310
1194
1706
1527
5119
2016
744
327
2942
2215
2358
0.148
0.018
0.425
0.253
0.209
0.001
0.000
0.544
0.758
0.009
0.034
0.064

Comparison of Healthcare Utilizations
Table 18 shows the changes in healthcare utilization patterns after the initiation
of antipsychotic therapy stratified by antipsychotic class. The mean number of total
hospitalizations decreased by 15 % in the second-generation antipsychotic medication
group while the same decreased by 10% in the typical group (0.68 vs. 0.58 and 0.63
vs. 0.57, respectively). Average length of stay decreased by 5% in the second-
generation group while the same decreased by %41 in the typical antipsychotic group
(2.56 vs. 2.44 and 3.59 vs. 2.13, respectively). Average length of stays at long-term
care facilities increased in both groups; from 12.8 to 17.2 in the second-generation
group (+34%) and 23.7 to 26.8 in the typical group (+13%). Mean number of suicide
attempts also increased in both groups: 25% in the second-generation group (0.12 vs.
0.15) and 11% in the typical group (0.09 vs. 0.1).

70

Table 18 Descriptive Statistics of Medical Utilization Patterns

Second-generation

Typical

Pre Post Pre Post
Number of
hospitalization
0.68
(SD=1.1)
0.58
(SD=1.0)
-15% 0.63
(SD=1.0)
0.57
(SD=1.0)
-10%
Days of
hospital stays
2.56
(SD=13.6)
2.44
(SD=15.3)
-5% 3.59
(SD=22.4)
2.13
(SD=9.2)
-41%
Number of
suicide
attempts
0.12
(SD=0.43)
0.15
(SD=0.53)
25% 0.09
(SD=0.38)
0.1
(SD=0.4)
11%
Days of LTC
stays
12.8
(SD=62.6)
17.2
(SD=69.4)
34% 23.7
(SD=83.9)
26.8
(SD=88.9)
13%

Panel Data Fixed-effect Poisson Regression Models
Table 19 - 22 show the results of the fixed effect Negative Binomial regression
models for the comparison of healthcare utilization patterns between the two groups.
There was no significant difference in the rate of hospitalization, days of hospital
stays, length of long-term care facility stays or rate of suicide attempts between the
two groups.

71

Table 19 Fixed-Effect Poisson Model – Hospitalization rate

Number of Hospitalization RR* 95% CI
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
0.94
1.09
1.47
1.14
1.30
1.33
1.39
1.11
1.62
0.99
1.17
1.06
0.79 - 1.11
0.95 - 1.25
1.21 - 1.79
0.95 - 1.36
1.09 - 1.54
1.13 - 1.56
1.15 - 1.69
1.07 - 1.15
1.38 - 1.90
0.84 - 1.17
0.98 - 1.40
0.90 - 1.24
* Relative Risk

Table 20 Fixed-Effect Poisson Model –Length of Hospital Stays

Days of Hospital Stay RR* 95% CI.
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
1.13
1.11
1.01
1.33
0.86
1.01
1.73
0.96
1.40
1.49
0.98
0.89
0.66 - 1.94
0.73 - 1.68
0.67 - 1.53
0.73 - 2.42
0.48 - 1.54
0.54 - 1.89
1.05 - 2.85
0.83 - 1.12
0.93 - 2.11
0.91 - 2.43
0.68 - 1.42
0.61 - 1.30
* Relative Risk

72
Table 21 Fixed-Effect Poisson Model –Length of LTC Stays

Length of LTC Stays RR* 95% CI
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
1.02
1.18
0.80
1.64
1.35
0.97
0.85
1.01
1.11
1.16
0.60
1.05
0.84 - 1.24
0.94 - 1.47
0.56 - 1.16
0.80 - 3.37
0.72 - 2.51
0.65 - 1.43
0.56 - 1.31
0.94 - 1.08
0.85 - 1.46
0.92 - 1.46
0.40 - 0.90
0.80 - 1.37
* Relative Risk

Table 22 Fixed-Effect Poisson Model – Number of Suicide Attempts

Number of Suicide Attempts RR* 95% CI
Second-generation Use
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
1.05
1.11
1.18
1.01
0.73
1.40
1.29
1.05
1.16
0.97
1.33
1.92
0.72 - 1.53
0.82 - 1.49
0.81 - 1.73
0.69 - 1.47
0.41 - 1.29
1.01 - 1.96
0.84 - 2.00
0.95 - 1.17
0.79 - 1.69
0.70 - 1.32
0.90 - 1.97
1.41 - 2.61
* Relative Risk

Comparisons with Cross Sectional Analyses
Table 23 - 30 show the results of the cross sectional analysis controlling for the
baseline characteristics including patient demographic characteristics, co-morbidities
and utilization. The results from the cross sectional OLS analyses were consistent
73
with those from the panel data fixed effects linear models. Total healthcare costs were
significantly higher in the second-generation antipsychotic group than in the typical
antipsychotic medication group during the 1 year period following the initiation of
therapy controlling the baseline patient characteristics (Table 23, p = 0.01).

Table 23 Cross Sectional Analysis – Total Healthcare Costs

Independent Variables
Parameter
Estimates P
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous hospitalization rate
1985.2
-284.8
-2030.8
-1717.3
-133.5
1209.9
2661.8
13188.0
-1498.5
-232.9
-768.4
-510.6
-1562.3
224.8
-1202.5
1493.1
2711.7
1197.2
-1096.1
174.7
-409.2
0.7
0.010
0.711
0.092
0.204
0.898
0.247
0.032
<.0001
0.167
0.786
0.454
0.661
0.171
0.838
0.332
<.0001
0.001
0.244
0.248
0.819
0.592
<.0001
R
2
= 0.56

74
Patients who were treated with second-generation antipsychotic medications
had significantly higher prescription costs than patients who were treated with typical
antipsychotics (p < 0.0001, Table 24).

Table 24 Cross Sectional Analysis – Pharmacy Costs

Independent Variables
Parameter
Estimate P
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous medical costs
944.8
-241.3
-823.0
-863.8
474.5
174.7
600.6
413.6
-732.5
66.0
-263.9
411.7
-496.4
314.0
-146.0
432.0
375.2
-396.8
-27.4
297.4
221.0
0.7
<.0001
0.19
0.00
0.01
0.06
0.48
0.04
0.25
0.00
0.75
0.28
0.14
0.07
0.23
0.62
<.0001
0.04
0.10
0.90
0.10
0.23
<.0001
R
2
= 0.60

However, there was no significant difference in total medical costs (total healthcare
costs – prescription costs) between the groups (p=0.17, Table 25).

75

Table 25 Cross Sectional Analysis – Medical Costs

Independent Variables
Parameter
Estimate P
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous medical costs
1007.9
-42.4
-1220.0
-876.6
-634.3
1037.9
2092.2
13053.0
-739.5
-322.4
-478.6
-927.4
-1083.4
-100.0
-1016.8
1091.1
2386.2
1638.2
-1064.3
-90.6
-640.8
0.7
0.17
0.95
0.29
0.50
0.52
0.30
0.08
<.0001
0.47
0.69
0.62
0.40
0.32
0.92
0.39
<.0001
0.00
0.09
0.24
0.90
0.38
<.0001
R
2
= 0.56

All other itemized healthcare costs were not significantly different between the two
groups (Table 26 – 30).

76
Table 26 Cross Sectional Analysis – Outpatient costs

Independent Variables
Parameter
Estimate P
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous hospitalization rate
-186.9
299.7
-403.2
-237.6
43.4
129.3
207.2
175.4
6.8
217.3
-230.2
207.4
29.1
264.4
42.2
380.0
93.1
303.1
316.8
-248.4
406.7
0.5
0.27
0.07
0.12
0.42
0.85
0.57
0.44
0.60
0.98
0.25
0.30
0.41
0.91
0.27
0.88
<.0001
0.58
0.18
0.13
0.14
0.01
<.0001
R
2
= 0.34

77

Table 27 Cross Sectional Analysis – Hospitalization Costs

Independent Variables
Parameter
Estimates P
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous hospitalization costs
905.8
-153.8
-466.9
-671.5
-1288.3
-1139.4
-1167.9
59.3
366.7
142.1
-387.7
107.9
-53.5
-826.9
280.2
926.7
520.8
2703.6
-395.3
12.1
227.4
0.4
0.07
0.75
0.54
0.43
0.05
0.09
0.14
0.95
0.59
0.79
0.55
0.88
0.94
0.24
0.72
<.0001
0.29
<.0001
0.51
0.98
0.64
<.0001
R
2
= 0.32

78

Table 28 Cross Sectional Analysis – CMHC costs

Independent Variables
Parameter
Estimate P
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous CMHC costs
189.4
13.7
-710.1
-230.1
-79.5
-124.8
-717.4
-741.0
48.6
4.1
390.1
-261.8
-411.4
-85.2
-207.4
-40.1
987.0
7.9
259.9
-58.6
-125.4
0.7
0.42
0.95
0.05
0.58
0.80
0.70
0.06
0.11
0.88
0.99
0.22
0.46
0.24
0.80
0.58
0.56
<.0001
0.98
0.37
0.80
0.59
<.0001
R
2
= 0.33

79

Table 29 Cross Sectional Analysis – Long Term Care Costs

Independent Variables
Parameter
Estimate p
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous medical costs
320.3
-16.4
-279.8
-67.0
210.2
1313.9
2638.6
6715.6
-198.7
-501.0
-11.6
-822.5
-364.9
126.0
-228.0
208.2
677.6
-182.4
-711.8
-55.8
-1150.0
0.9
0.47
0.97
0.69
0.93
0.73
0.03
0.00
<.0001
0.75
0.31
0.98
0.22
0.58
0.84
0.75
0.12
0.13
0.76
0.19
0.90
0.01
<.0001
R
2
=0.73

80
Table 30 Cross Sectional Analysis – Other Costs

Independent Variables
Parameter
Estimate P
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Other costs
-3.6
-36.5
51.8
64.4
59.3
105.7
274.3
429.6
39.1
-31.6
-28.6
-0.4
-12.0
61.9
-4.1
43.6
-12.4
40.5
36.0
24.0
-40.7
0.6
0.93
0.37
0.42
0.37
0.29
0.06
<.0001
<.0001
0.50
0.49
0.60
0.99
0.84
0.29
0.95
0.00
0.76
0.46
0.48
0.56
0.32
<.0001
R
2
= 0.37

The results from cross sectional Poisson regression models were also
consistent with the results from panel data fixed effects Poisson models. As in the
panel data analyses, second-generation antipsychotic use did not have significant
impact on subsequent healthcare utilizations – number of hospitalization (RR: 1.00;
95% CI: 0.87 – 1.14, Table 31), days of hospital stays (RR: 0.94; 95% CI: 0.64 – 1.38,
Table 32), days of long-term care stays (RR: 1.01; 95% CI: 0.83 – 1.23, Table 33) and
number of suicide attempts (RR: 1.35; 95% CI: 0.99 – 1.84, Table 34).
81

Table 31 Cross Sectional Analysis – Number of Hospitalization

Number of Hospitalizations RR* p 95% CI
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous hospitalization rate
1.00
0.93
0.98
0.92
0.88
0.99
1.04
1.45
1.20
0.93
0.85
1.30
1.05
0.94
0.85
1.05
1.25
1.19
1.25
1.04
0.96
1.34
0.943
0.160
0.859
0.530
0.159
0.959
0.708
0.000
0.041
0.397
0.066
0.005
0.636
0.554
0.161
0.012
0.001
0.043
0.003
0.587
0.652
0.000
0.87 - 1.14
0.84 - 1.03
0.80 - 1.20
0.71 - 1.19
0.74 - 1.05
0.82 - 1.21
0.84 - 1.30
1.18 - 1.77
1.01 - 1.43
0.80 - 1.09
0.72 - 1.01
1.08 - 1.56
0.85 - 1.30
0.76 - 1.16
0.68 - 1.07
1.01 - 1.09
1.09 - 1.43
1.01 - 1.41
1.08 - 1.46
0.90 - 1.21
0.82 - 1.13
1.29 - 1.40
*Relative risk

82

Table 32 Cross Sectional Analysis– Days of Hospital stays

RR* p 95% CI
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous hospital stays
0.94
1.06
0.47
0.53
0.45
0.62
0.93
0.93
1.32
0.85
1.48
1.15
1.18
0.58
2.05
1.22
1.92
1.59
1.16
0.76
0.82
1.01
0.742
0.766
0.000
0.006
0.000
0.036
0.737
0.910
0.286
0.501
0.013
0.542
0.462
0.012
0.010
0.000
0.000
0.013
0.319
0.086
0.139
0.000
0.64 - 1.38
0.73 - 1.53
0.31 - 0.71
0.34 - 0.83
0.32 - 0.64
0.40 - 0.97
0.61 - 1.42
0.29 - 3.03
0.79 - 2.20
0.54 - 1.35
1.09 - 2.02
0.74 - 1.78
0.76 - 1.85
0.38 - 0.89
1.19 - 3.54
1.14 - 1.30
1.38 - 2.66
1.10 - 2.30
0.86 - 1.57
0.55 - 1.04
0.63 - 1.07
1.01 - 1.01
*Relative risk

83

Table 33 Cross Sectional Analysis – Days of Long-term Care stays

Long-term Care Days RR* p 95% CI
Second-generation Use
Female
Urban
Mixed
35 ≤ age < 45
45 ≤ age < 55
55 ≤ age < 65
65 ≤ age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Anti Depressant Use
Mood Stabilizer Use
1.01
0.94
1.01
0.90
1.31
3.13
4.83
5.46
0.87
0.86
1.00
0.63
0.91
1.13
0.98
1.07
1.22
0.78
0.77
0.82
1.00
1.01
0.912
0.580
0.977
0.616
0.575
0.006
0.000
0.000
0.349
0.181
0.983
0.075
0.673
0.593
0.916
0.051
0.165
0.371
0.209
0.048
0.988
0.000
0.83 - 1.23
0.77 - 1.16
0.66 - 1.53
0.60 - 1.35
0.51 - 3.36
1.39 - 7.04
2.15 - 10.84
2.30 - 12.96
0.65 - 1.17
0.69 - 1.07
0.75 - 1.35
0.38 - 1.05
0.58 - 1.43
0.72 - 1.76
0.64 - 1.48
1.00 - 1.14
0.92 - 1.62
0.45 - 1.35
0.51 - 1.16
0.67 - 1.00
0.82 - 1.22
1.01 - 1.01
*Relative risk
84

Table 34 Cross Sectional Analysis – Number of Suicide Attempts

Number of Suicide Attempts RR* P 95% CI
Second-generation Use
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance abuse
Other mental disorders
Anti depressant use
Mood stabilizer use
Previous suicide attempts
1.35
1.16
0.93
0.88
0.58
0.64
0.61
0.19
0.87
0.81
0.87
1.12
1.12
1.14
1.00
1.02
1.62
1.56
1.29
1.14
0.79
1.89
0.060
0.393
0.745
0.658
0.003
0.001
0.060
0.018
0.508
0.310
0.444
0.557
0.621
0.466
0.989
0.608
0.000
0.005
0.140
0.472
0.191
0.000
0.99 - 1.84
0.83 - 1.62
0.61 - 1.42
0.51 - 1.53
0.40 - 0.83
0.49 - 0.83
0.36 - 1.02
0.05 - 0.75
0.58 - 1.31
0.55 - 1.21
0.60 - 1.25
0.77 - 1.64
0.72 - 1.74
0.80 - 1.62
0.67 - 1.49
0.94 - 1.11
1.26 - 2.08
1.14 - 2.14
0.92 - 1.80
0.80 - 1.61
0.55 - 1.13
1.69 - 2.12
*Relative risk

85

Table 35 summarizes the parameter estimates from the cross sectional linear
and nonlinear models as well as the panel data models.

Table 35 Summary of the Panel data and Cross Sectional Analyses

Outcome
Measures
Panel Data Cross sectional
Parameter
Estimates
p Parameter
Estimates
p
Total Healthcare
costs
2366.0

0.028

1985.2

0.010

Pharmacy Costs 770.9

<0.0001

944.8

< 0.0001
Medical Costs 1687.0

0.090 1007.9

0.17
Hospital Costs 1501.3

0.067 905.8 0.07
Outpatient Costs -175.2 0.356 -186.9 0.27

Long-term Care
Costs
252.4

0.629 320.3 0.47
CMHC -6.4 0.979 189.4

0.42

Other Costs 34.2 0.439 -3.6 0.93

RR* 95% CI RR 95% CI
Number of
hospitalization
0.94 0.79 - 1.11

1.00 0.87 - 1.14

Days of hospital
stays
1.13 0.66 - 1.94

0.94 0.64 - 1.38

Days of LTC stays 1.02

0.84 - 1.24

1.01

0.83 - 1.23

Number of Suicide
Attempts
1.05 0.72 - 1.53

1.35

0.99 - 1.84

* Relative Risk

86

SECOND-GENERATION HEAD TO HEAD COMPARISON
Descriptive Statistics of Study Population
Table 36 shows the demographic characteristics of the patients who initiated
second-generation antipsychotic medications by drug type. The three groups were not
statistically different for any of the observable demographic characteristics.

Table 36 Demographic Characteristics of Study Sample – Head to Head
Comparisons

Characteristic Olanzapine
(n = 865 )
Risperidone
(n = 598 )
Quetiapine
(n = 318)
p
Age
Mean

18-24
25-34
35-44
45-54
55-64
Over 65

45.3
(SD= 14.0)
64 (7.4%)
135(15.1%)
226 (26.1%)
253 (29.3%)
117 (13.5%)
70 (8.10%)

44.8
(SD= 14.0)
53 (8.9%)
91 (15.2%)
165 (27.6%)
155 (25.9%)
91 (15.2%)
43 (7.2%)

44.8
(SD = 13.2)
23 (7.2%)
41 (12.9%)
102 (32.1%)
96 (29.9%)
39 (12.3%)
18 (5.7%)

0.72

0.53
0.50
0.13
0.30
0.43
0.36
Gender
Male
Female

345 (40.0%)
520 (60.0%)

252 (42.1%)
346 (57.9%)

122 (38.4%)
196 (61.6%)

0.50
Residence
Urban
Rural
Mixed

615 (71.1%)
83 (9.6%)
167 (19.3%)

416 (69.6%)
57 (9.5%)
125 (20.9%)

207 (65.1%)
31 (9.8%)
80 (25.2%)

0.14
0.99
0.09
Race
White
Black
Hispanic
Asian
Others

517 (59.7%)
119 (13.8%)
38 (4.4%)
20 (2.3%)
171 (19.8%)

350 (58.5%)
92 (15.4%)
36 (6.0%)
10 (1.7%)
110 (18.4%)

211 (66.4%)
37 (11.6%)
12 (3.8%)
4 (1.3%)
54 (17.0%)

0.06
0.29
0.23
0.44
0.53
Eligibility
AFDC
1
OAA
2
Blind
Disabled
Other aid

147 (17.0%)
22 (2.5%)
2 (0.2%)
627 (72.5%)
67 (7.8%)

105 (17.6%)
12 (2.0%)
2 (0.3%)
435 (72.7%)
44 (7.4%)

50 (15.7%)
5 (1.6%)
1 (0.3%)
244 (76.7%)
18 (5.7%)

0.78
0.56
0.93
0.32
0.47

87

Baseline Healthcare Utilizations
Baseline healthcare utilizations and healthcare costs of the three groups are
shown on Table 37. The three groups were not statistically different in their total or
itemized expenditure except for the prescription drug costs in year 1. Patients in the
risperidone group had significantly higher prescription drug costs in the baseline
period compared to the patients in quetiapine group ($4501.4 vs. $3044.1, p < 0.05).
Difference between olanzapine and risperidone was not statistically significant.

Table 37 Baseline Healthcare costs by drug

Olanzapine Risperidone Quetiapine p
Prescription Costs 3600.4
(SD= 9468.7)
3044.1
(SD=4432.2)
4501.4
(SD=10984.8)
0.0007
Hospitalization
Costs
2574.7
(SD=12526.6)
2935.4
(SD=18155.6)
1880.8
(SD=7234.1)
0.48
Long-term care
costs
2965.9
(SD=15891.6)
3611.6
(SD=17024.9)
4277.6
(SD=18473.4)
0.16
Rehabilitation 13.2
(SD=145.7)
20.1
(SD=307.5)
8.0
(SD=112.3)
0.88
Psychologist
services
1.6
(SD =22.9)
5.3
(SD=47.4)
3.5
(SD=51.6)
0.14
Community mental
healthcare facility
2305.8
(SD=4789.0)
2631.2
(SD=5171.7)
3016.6
(SD=5869.9)
0.49
Other Costs 239.2
(SD=1054.9)
215.1
(SD=971.0)
256.6
(SD=1045.1)
0.62
Part B Costs 1877.9
(SD=3346.9)
2193.9
(SD=6133.8)
2366.5
(SD=5696.4)
0.23
Medical Costs 9978.3
(SD=21305.0)
11612.6
(SD=21590.6)
11809.6
(SD=26670.8)
0.13
Total Healthcare
costs
13578.6
(SD=23308.6)
14656.6
(SD=27720.3)
16311.0
(SD=24691.7)
0.02

Baseline healthcare utilization patterns are displayed on Table 38. There was
no statistically significant difference in the medical utilization patterns including the
88
proportion of patients with at least one hospitalization, mean number of
hospitalizations, mean length of hospital stays, long-term care use, length of stays at a
long-term care facility, and mean number of suicide attempts.

Table 38 Baseline Medical Utilization by drug

Olanzapine Quetiapine Risperidone p
Number of Suicide
attempts
0.1 (0.4) 0.1(0.4) 0.1 (0.4)
0.88
Number of hospitalization 0.7 (1.1) 0.7 (1.1) 0.6 (1.1) 0.35

Hospital LOS 2.5 (11.9) 1.8 (6.9) 2.8 (17.2) 0.48
LTC LOS
10.6 (57.1) 15.31 (68.09) 15.3 (65.5) 0.07
Hospital Use
23.0% 20.1% 21.2% 0.51
Long Term Care Use
5.1% 7.6% 7.2% 0.14

Factors Affecting Treatment Decision
Table 39 shows the results from the multinomial logistic model for the
selection of second-generation antipsychotic medications setting olanzapine as the
reference group. Patients who had at least one claim for diabetes medication in the
baseline were significantly more likely to be treated with risperidone than with
olanzapine (RR: 1.71; 95% CI: 1.14 – 2.57) while diabetes medication use did not
have significant impact on the choice between quetiapine and olanzapine (OR: 1.06;
95% CI: 0.62 – 1.82). Patients who had long-term care use in the baseline were
significantly more likely to be treated with risperidone (OR:1.74; 95% CI: 1.06 – 2.86)
than with olanzapine.
For the initiation of quetiapine therapy, patients with following were
significantly more likely to initiate quetiapine than olanzapine: long-term care use
89
(RR: 2.27; 95% CI: 1.25 – 4.13), and previous anti-depression medication therapy
(RR: 1.04; 95% CI: 1.04– 2.29).

Table 39 Multinomial Logit Model for Selection of Second-Generation
Antipsychotic Medication

Risperidone Quetiapine
OR 95% CI OR 95% CI
Female 0.88 0.70 - 1.11 1.13 0.85 - 1.51
Urban
Mixed
0.92
1.06
0.64 - 1.34
0.70 - 1.62
0.83
1.22
0.52- 1.31
0.73 - 2.02
35 ≤ age < 45
45 ≤ age < 55
55 ≤ age < 65
65 ≤ age
1.01
0.85
1.03
0.71
0.75 - 1.38
0.62 - 1.16
0.71 - 1.49
0.41 - 1.23
1.33
1.06
0.87
0.59
0.90 - 1.95
0.72 - 1.57
0.53 - 1.42
0.28 - 1.22
Black
Other race
1.18
1.02
0.86 - 1.62
0.80 - 1.31
0.79
0.78
0.52 - 1.21
0.57 - 1.08
Bipolar - Manic
Bipolar - Depressed
Bipolar - Mixed
Bipolar - NOS
Bipolar - Others
0.84
1.21
0.89
0.80
0.81
0.62 - 1.14
0.88 - 1.66
0.64 1.25
0.58 - 1.11
0.57 - 1.16
0.79
1.19
1.25
1.10
1.08
0.54 - 1.15
0.81 - 1.74
0.85 - 1.85
0.76 - 1.60
0.72 - 1.62
OAA
Disabled
AFDC
Other aids
0.54
0.76
0.80
0.72
0.06 - 4.72
0.10 - 5.58
0.11 - 5.98
0.09 - 5.45
0.51
0.74
0.63
0.52
0.03 - 7.51
0.07 - 8.33
0.05 - 7.17
0.04 - 6.26
Charlson Index
Major Depression
Schizophrenia
Long-term care use
Hospitalization
Substance abuse
Other mental disorder
Manic
Diabetes medication
use
0.97
0.88
1.06
1.74
0.92
0.78
1.04
0.55
1.71
0.90 - 1.05
0.61 - 1.27
0.84 - 1.33
1.06 - 2.86
0.70 - 1.21
0.58 - 1.06
0.78 - 1.37
0.29 - 1.06
1.14 - 2.57
1.00
1.54
1.32
2.27
0.80
0.81
1.08
0.67
1.06
0.92 - 1.10
1.04 - 2.29
1.00 - 1.74
1.25 - 4.13
0.57 - 1.13
0.56 - 1.17
0.76 - 1.52
0.32 - 1.44
0.62 - 1.82

Switching to and/or Augmentation with Other Antipsychotics
Table 40 presents the duration of therapy and the proportion of patients who
switched to or augmented with other antipsychotic medications within 30 days of the
initial therapy. Mean duration of initial uninterrupted therapy was 106.7 days for
90
olanzapine patients, 127.2 for quetiapine patients, and 120.7 days for risperidone
patients. The difference was not statistically significant. Mean days covered per year
was 174.0 for olanzapine, 194.5 for quetiapine, and 172.0 for risperidone (p = 0.02).
Mean days covered was significantly higher for patients on quetiapine compared to
both patients on olanzapine and risperidone.

Table 40 Duration of Therapy and Rate of Switch or Augmentation by Drug

Olanzapine
(n = 865)
Quetiapine
(n = 318)
Risperidone
(n = 598)
p
Duration of initial
therapy
106.7
(SD=200.2 )
127.2
(SD=242.1)
120.7
(SD=225.1)
0.81
Duration of therapy 174.0
(SD=123.4)
194.5
(SD=124.3)
172.0
(SD=127.4)
0.01
Switched within 30
Days
3.1 % 4.7 % 5.5 % 0.07

3.1 % of the patients on olanzapine switched to or augmented with another
antipsychotic medication within 30 days of the initiation of therapy while 5.5% and
4.7% of patients on risperidone and quetiapine respectively did (p = 0.07). Results
from the logistics regression model predicting a switch to or an augmentation with
other antipsychotic medication within 30 days of therapy are presented on Table 41.
Patients on risperidone was significantly more likely to switch to or augment with
another antipsychotic medications compared to patients on olanzapine (OR: 1.53; 95%
CI: 1.09 – 2.16) while the likelihood of a switch or an augmentation for patients on
quetiapine were not statistically different from patients on olanzapine (OR: 1.35; 95%
CI: 0.88 – 2.06).
91
Table 41 Logistic Regression Model – Switch/Augmentation

Odds Ratio 95% CI
Female
Risperidone
Quetiapine
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Major Depression
Schizophrenia
Previous long term care
Previous hospitalization
Substance abuse
Other mental disorders
1.04
1.99
1.67
0.96
0.56
0.42
0.47
0.59
1.15
0.52
1.05
0.80
0.65
0.77
1.91
0.99
0.99
0.30
1.91
0.95
1.50
1.61
0.91
0.64 - 1.71
1.17 - 3.39
0.86 - 3.23
0.45 - 2.04
0.22 - 1.39
0.21 - 0.83
0.25 - 0.91
0.27 - 1.29
0.45 - 2.95
0.23 - 1.21
0.60 - 1.82
0.39 - 1.61
0.29 - 1.47
0.34 - 1.73
1.04 - 3.53
0.45 - 2.14
0.84 - 1.17
0.09 - 0.99
1.17 - 3.12
0.35 - 2.54
0.85 - 2.63
0.88 - 2.94
0.49 - 1.70

Descriptive Statistics of Post-Intervention Healthcare Costs and Utilizations
Table 42 shows the descriptive statistics of healthcare costs and medical
utilizations in the post-intervention period by drug. While there was a significant
difference in prescription costs between the three groups at baseline, the differences
were no longer statistically significant in the post-antipsychotic therapy (p=0.12).
Total healthcare costs were still significantly different between the three groups in
year 2 ($17837, $19126, and $20732: olanzapine, quetiapine and risperidone groups,
respectively, p=0.02). All other types of costs were not significantly different between
the groups.

92

Table 42 Descriptive Statistics of Healthcare Costs by Drug – Year 2

Olanzapine Quetiapine Risperidone p
Pharmacy Costs 5886.7 (SD=8761.3) 5195.9 (SD=4951.7) 6049.5 (SD=6668.1) 0.14

Hospitalization Costs
Long-term Care costs
Rehab costs
Psychologist costs
CMHC costs
Other costs
Outpatient costs
Medical Costs
Total costs
2504.1 (SD=12359.4)
3932.3 (SD=17270.2)
1.5 (SD=30.5)
3.4 (SD=37.4)
2913.3 (SD=5746.7)
330.1 (SD=1259.8)
2266.2 (SD=4775.8)
11950.8 (SD=22899.8)
17837.6 (SD=25102.1)
3053.4 (SD=23094.3)
4489.3 (SD=18647.8)
20.1 (SD=226.8)
6.0 (SD=75.8)
3489.7 (SD=7647.2)
281.5 (SD=1116.9)
2590.9 (SD=6006.9)
14683.2 (SD=23518.8)
19126.8 (SD=33235.3)
2192.1 (SD=9067.1)
6040.2 (SD=21094.5)
8.0 (SD=86.9)
4.2 (SD=67.7)
3698.4 (SD=6711.0)
274.0 (SD=1267.0)
2466.4 (SD=4185.5)
13930.8 (SD=31785.9)
20732.7 (SD=24907.0)
0.23
0.17
0.08
0.70
0.40
0.52
0.73
0.12
0.02

93
Medical utilization patterns in the second year period are presented on Table
43. As in the baseline period, there was no significant difference in terms of number
of hospitalizations, length of hospital stays, and number of suicide attempts between
the three groups in post-antipsychotic therapy. Also similarly to baseline, mean length
of stays at a long-term care facility was marginally higher for patients on quetiapine
but the difference was not statistically significant.

Table 43 Descriptive Statistics of Healthcare Utilization by Drug – Year 2

Olanzapine Quetiapine Risperidone p
Number of Hospitalization
Length of Hospital Stays
Number of suicide attempts
Length of LTC stays
0.6 (SD=1.0)
2.4 (SD=11.7)
0.2 (SD=0.6)
12.7 (SD=60.5)
0.5 (SD=0.9)
2.9 (SD=21.9)
0.1 (SD=0.4)
19.9 (SD=73.7)
0.5 (SD=0.9)
2.1 (SD=8.6)
0.1 (SD=0.4)
14.8 (SD=64.9)
0.53
0.23
0.19
0.06

Panel Data First Difference Models for Healthcare Costs
Table 44 – Table 51 show the results of the panel data first difference models
for the impact of second-generation antipsychotic medication use on healthcare costs.
There was no significant difference in changes in total healthcare costs between year 1
and year 2 between the three comparison groups (+$86 for risperidone and + $957 for
quetiapine). Other factors that were positively related with an increase in total health
care costs include: having a diagnosis of bipolar disorder in year 2 (except bipolar
subtype II), use of other types of bipolar medications including mood stabilizers and
antidepressants, and higher comorbidity index (Table 44).

94
Table 44 Head to Head Comparison Panel Data First Difference Model: Total
Healthcare Costs

Total Healthcare Costs Parameter Estimates p
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
86
957
3500
362
2889
2047
3633
1387
2870
852
2531
2529
834
0.920
0.366
0.001
0.762
0.008
0.092
0.002
0.000
0.000
0.464
0.008
0.002
0.379

Olanzapine use was generally associated with increased pharmacy costs
compared to risperidone and quetiapine; however, the results were not statistically
different (olanzapine vs. risperidone:-$166, p = 0.46; olanzapine vs. quetiapine (-$567,
p = 0.14) (Table 45).

Table 45 Head to Head Comparison Panel Data First Difference Model:
Prescription Costs

Prescription Costs Parameter Estimates p
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Schizophrenia
-111.6
-646.6
659.8
-801.8
-198.6
-237.5
-71.2
258.9
464.3
-486.2
582.6
0.67
0.05
0.04
0.02
0.58
0.40
0.84
0.00
0.12
0.06
0.04

95

The change of medical costs (defined as total costs – pharmacy costs) in the
risperidone group was not statistically different from the olanzapine group ($366, p =
0.61). The medical costs increased marginally greater in the quetiapine group in
comparison to olanzapine use; however the difference was not statistically significant
($1729, p = 0.10) (Table 46).

Table 46 Head to Head Comparison Panel Data First Difference Model: Medical
Costs

Medical Costs Parameter Estimates p
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
366
1727
2799
1116
3029
2197
3617
940
2318
1203
1449
1829
244
0.613
0.098
0.001
0.159
0.004
0.052
0.009
0.001
0.024
0.207
0.159
0.021
0.805

Compared to olanzapine, quetiapine use was associated with increased long-term care
costs. Again, the difference was not statistically significant (+987, p=0.14) (Table
47).

96
Table 47 Head to Head Comparison Panel Data First Difference Model: Long-
term Care Costs

Long-term Care Costs Parameter Estimates p
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
-98
987
325
566
1026
-285
144
445
144
27
318
698
-243
0.85
0.14
0.62
0.43
0.16
0.62
0.84
0.01
0.81
0.96
0.53
0.17
0.67

Analyses of other itemized health care costs show that there was no significant
difference between the three groups in changes in the itemized healthcare costs
including outpatient costs, hospitalization costs, and community mental healthcare
facility costs among the three groups (Table 48 – Table 51).

Table 48 Head to Head Comparison Panel Data First Difference Model:
Hospitalization Costs

Hospitalization Costs Parameter Estimates p
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
169
484
988
-78
449
1527
1729
222
780
44
255
378
-576
0.762
0.433
0.078
0.857
0.359
0.020
0.016
0.069
0.068
0.936
0.644
0.474
0.139

97
Table 49 Head to Head Comparison Panel Data First Difference Model:
Outpatient Services Costs (Part B costs)

Outpatient Services Costs Parameter Estimates p
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
27
-95
307
218
408
562
408
534
460
712
73
173
331
0.884
0.745
0.218
0.599
0.075
0.022
0.058
0.003
0.068
0.002
0.788
0.332
0.105

Table 50 Head to Head Comparison Panel Data First Difference Model:
Community Mental Health Facility Costs

Community Mental Health
Center Costs Parameter Estimates p
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
165
196
1182
545
1254
580
1433
-99
1133
632
829
674
907
0.661
0.617
0.021
0.166
0.013
0.102
0.001
0.160
0.013
0.146
0.002
0.032
0.000

98

Table 51 Head to Head Comparison Panel Data First Difference Model: Other
Costs

Other costs Parameter Estimates p
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
20
-60
76
30
34
-13
-54
49
13
-39
50
6
-51
0.684
0.452
0.047
0.747
0.625
0.828
0.367
0.015
0.845
0.360
0.438
0.901
0.354

Panel Data Fixed Effect Poisson Regression Models
The impact of different second-generation antipsychotic medications on
medical utilization is presented on Table 52 – Table 55. Use of different second-
generation antipsychotic medications did not have significant impact on the rate of
hospitalization in the second year while having a diagnosis of and mixed and manic
subtypes of bipolar disorder and schizophrenia, and a higher co morbidity index was
associated with significant increase in the rate of hospitalization (Table 52).

99
Table 52 Head to Head Comparison Panel Data Fixed Effect Poisson Model:
Number of Hospitalizations

Number of Hospitalization RR* p 95% CI
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
0.691
0.590
0.001
0.449
0.182
0.000
0.001
0.004
0.017
0.912
0.362
0.923
0.000
0.691
0.590
0.001
0.449
0.182
0.000
0.001
0.004
0.017
0.912
0.362
0.923
0.000
0.76 - 1.20
0.73 - 1.20
1.16 - 1.81
0.86 - 1.40
0.92 - 1.53
1.20 - 1.79
1.14 - 1.66
1.03 - 1.15
1.04 - 1.45
0.84 - 1.21
0.91 - 1.29
0.83 - 1.22
1.23 - 1.87
*Relative risk
Similarly, days of hospital stays, days of stay at a long-term care facility, or the
rate of suicide attempts were not significantly different among the three groups (Table
53 – Table 55).

Table 53 Head to Head Comparison Panel Data Fixed Effect Poisson Model:
Days of Hospital Stays

Days of Hospital Stays RR* p 95% CI
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
1.03
1.08
1.52
1.12
0.93
1.44
2.12
1.04
1.27
0.75
1.20
1.01
1.23
0.919
0.789
0.160
0.695
0.866
0.282
0.011
0.580
0.205
0.119
0.470
0.968
0.415
0.54 - 2.00
0.61 - 1.90
0.85 - 2.71
0.64 - 1.97
0.39 - 2.21
0.74 - 2.80
1.19 - 3.79
0.91 - 1.19
0.88 - 1.82
0.52 - 1.08
0.74 - 1.94
0.62 - 1.64
0.75 - 2.00
*Relative risk

100

Table 54 Head to Head Comparison Panel Data Fixed Effect Poisson Model:
Days of Long-term Care Facility Stays

Days of Long-term Care Stays RR* p 95% CI
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
0.89
1.29
0.85
1.73
1.49
0.90
1.00
1.01
0.74
1.21
1.28
1.10
1.17
0.430
0.250
0.526
0.055
0.498
0.658
0.992
0.821
0.511
0.318
0.187
0.345
0.622
0.66 - 1.19
0.84 - 1.98
0.51 - 1.41
0.99 - 3.04
0.47 - 4.66
0.56 - 1.44
0.59 - 1.69
0.93 - 1.10
0.31 - 1.80
0.83 - 1.78
0.89 - 1.83
0.90 - 1.36
0.62 - 2.20
*Relative risk

Table 55 Head to Head Comparison Panel Data Fixed Effect Poisson Model:
Number of Suicide Attempts

Number of Suicide Attempts RR* p 95% CI
Risperidone
Quetiapine
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Schizophrenia
0.84
0.65
1.53
1.10
0.62
1.57
1.40
0.99
1.42
2.16
1.08
1.04
1.20
0.423
0.064
0.049
0.659
0.128
0.038
0.144
0.862
0.161
0.000
0.674
0.866
0.511
0.54 - 1.29
0.41 - 1.02
1.00 - 2.34
0.72 - 1.70
0.34 - 1.15
1.03 - 2.42
0.89 - 2.21
0.88 - 1.11
0.87 - 2.34
1.43 - 3.28
0.75 - 1.57
0.66 - 1.63
0.70 - 2.07
*Relative risk

101

Comparison with Cross Sectional Analyses
Similar to the comparisons between different classes, the results from the head-
to-head cross sectional analyses were consistent with the results from the panel data
analyses (Table 56 – 59). Compared to olanzapine, quetiapine use was associated with
a marginally significant reduction in pharmacy costs (- $523, p = 0.06) (Table 56).

Table 56 Cross Sectional Head to Head Comparisons – Pharmacy Costs

Independent Variables
Parameter
Estimates p
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Previous pharmacy costs
-339.3
-523.0
-474.4
-354.4
-350.0
654.9
402.2
742.3
616.3
-675.5
-3.8
-270.5
362.7
-634.9
372.6
132.6
405.0
404.6
-503.3
218.2
0.7
0.13
0.06
0.02
0.30
0.37
0.02
0.16
0.03
0.15
0.03
0.99
0.33
0.22
0.04
0.21
0.68
<.0001
0.06
0.07
0.40
<.0001
R
2
= 0.68

102
However, the medical costs increased greater, although not significant, in the
quetiapine group ($1458.4, p=0.17) (Table 57) offsetting the reduction in pharmacy
costs in the total healthcare costs ($831, p=0.46) (Table 58). In particular, patients in
the quetiapine group had a marginally greater increase in long-term care costs
compared to their peers in the olanzapine ($1219.7, p = 0.06) (Table 59). The
comparisons between the olanzapine and the risperidone groups were not statistically
significant.

Table 57 Cross Sectional Head to Head Comparisons – Medical Costs

Independent Variables
Parameter
Estimates p
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous medical costs
662.1
1458.4
-167.2
-2316.7
-1714.2
40.9
1498.3
955.6
8920.1
-811.4
-867.5
-390.6
-780.2
-1644.7
-33.5
-701.2
624.6
845.7
904.6
-1583.1
-37.1
-1310.7
0.9
0.44
0.17
0.84
0.08
0.25
0.97
0.17
0.47
<.0001
0.49
0.34
0.72
0.50
0.17
0.98
0.57
0.02
0.30
0.40
0.11
0.96
0.11
<.0001
R
2
=0.63

103

Table 58 Cross Sectional Head to Head Comparisons - Total Healthcare Costs

Independent Variables
Parameter
Estimates p
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous healthcare costs
429.3
831.0
-375.6
-2730.1
-2109.2
487.7
1850.7
1768.5
9816.3
-1287.8
-822.7
-553.6
-324.8
-2220.3
396.8
-375.4
746.1
1334.2
477.0
-1249.4
-212.2
-1323.4
0.8
0.64
0.46
0.66
0.05
0.18
0.67
0.11
0.21
<.0001
0.30
0.39
0.63
0.79
0.08
0.74
0.78
0.01
0.12
0.67
0.24
0.80
0.13
<.0001
R
2
=0.64

104

Table 59 Cross Sectional Head to Head Comparisons – Long Term Care Costs

Independent Variables
Parameter
Estimates p
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous LTC costs
71.2
1219.7
306.0
-1105.8
-1246.3
-328.6
1318.5
994.9
7438.2
-97.1
-339.0
-335.0
-960.8
-411.9
916.8
-823.2
85.6
602.1
-350.5
-804.5
58.2
-857.9
0.9
0.89
0.06
0.53
0.17
0.17
0.62
0.05
0.22
<.0001
0.89
0.54
0.61
0.17
0.57
0.19
0.28
0.59
0.23
0.59
0.19
0.91
0.09
<.0001
R
2
=0.72

All other comparisons of other types of costs were not statistically significant between
the three groups (Table 60 – 63). Similarly, the results from the cross sectional
Poisson regression for healthcare utilization patterns were consistent with those from
the panel data fixed effect models (Table 64 – 67). Table 68 summarizes the
comparison of the results from the panel data approach and cross sectional models.

105
Table 60 Cross Sectional Head to Head Comparisons – Part B costs

Independent Variables
Parameter
Estimates p
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous Part B costs
125.9
-158.6
-25.2
-420.1
-408.4
148.4
218.0
283.0
91.4
-98.0
-99.2
-414.7
128.2
37.8
76.5
-6.0
323.5
166.6
34.2
202.0
-286.5
349.1
0.6
0.57
0.56
0.90
0.21
0.28
0.59
0.43
0.41
0.83
0.74
0.67
0.13
0.66
0.90
0.79
0.99
<.0001
0.43
0.90
0.43
0.17
0.09
<.0001
R
2
=0.37

106

Table 61 Cross Sectional Head to Head Comparisons – CMHC Costs

Independent Variables
Parameter
Estimates P
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous CMHC costs
348.9
294.6
-8.4
-750.5
-38.2
158.9
42.4
-613.2
-570.6
355.5
-17.4
785.5
-29.1
-381.3
218.2
-12.5
-67.3
875.5
-98.1
38.1
-157.8
-321.1
0.6
0.25
0.43
0.98
0.11
0.94
0.68
0.91
0.19
0.33
0.39
0.96
0.04
0.94
0.36
0.59
0.98
0.46
0.00
0.79
0.91
0.58
0.26
<.0001
R
2
=0.28

107

Table 62 Cross Sectional Head to Head Comparisons – Hospitalization Costs

Independent Variables
Parameter
Estimates P
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous hospitalization costs
276.9
342.3
-319.3
-82.2
-9.9
179.1
-84.5
-51.3
1136.5
-840.1
-339.0
-164.1
447.3
-830.2
-941.7
349.5
368.1
-218.9
1610.1
-609.6
300.1
-185.5
0.9
0.62
0.62
0.54
0.92
0.99
0.80
0.90
0.95
0.29
0.27
0.57
0.81
0.55
0.28
0.20
0.66
0.03
0.68
0.02
0.34
0.57
0.72
<.0001
R
2
= 0.60

108

Table 63 Cross Sectional Head to Head Comparisons – Other Costs

Independent Variables
Parameter
Estimates p
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous other costs
-28.6
-48.9
-42.1
-3.4
-31.9
-16.4
100.5
232.5
342.0
-47.4
21.0
-38.5
63.3
-18.5
-14.5
54.2
43.6
12.3
43.0
48.1
-27.2
-69.3
0.7
0.57
0.43
0.37
0.97
0.71
0.80
0.12
0.00
0.00
0.49
0.69
0.54
0.35
0.79
0.83
0.46
0.00
0.80
0.49
0.41
0.57
0.15
<.0001
R
2
=0.41

109

Table 64 Cross Sectional Head to Head Comparison – Number of Hospitalization

Independent Variables RR* 95% CI
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous hospitalization rate
0.88
0.88
0.85
0.90
0.91
0.92
1.05
1.02
1.48
1.26
0.93
0.87
1.39
0.99
0.84
0.88
1.02
1.28
1.17
1.21
1.05
1.00
1.37
0.75 - 1.02
0.71 - 1.10
0.72 - 0.99
0.65 - 1.23
0.63 - 1.31
0.76 - 1.12
0.86 - 1.28
0.78 - 1.32
1.14 - 1.91
1.02 - 1.56
0.77 - 1.12
0.70 - 1.09
1.16 - 1.67
0.80 - 1.22
0.69 - 1.02
0.70 - 1.12
0.95 - 1.09
1.10 - 1.50
0.96 - 1.44
0.98 - 1.50
0.89 - 1.24
0.86 - 1.16
1.27 - 1.47
*Relative risk

110

Table 65 Cross Sectional Head to Head Comparison – Days of Hospital Stays

Independent Variables RR* 95% CI
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous days of inpatient
0.95
0.80
0.79
0.48
0.72
0.44
0.60
0.97
1.56
1.35
0.73
1.37
1.41
1.15
0.41
1.88
1.25
2.06
1.52
1.48
0.87
0.80
1.01
0.62 - 1.46
0.52 - 1.25
0.54 - 1.16
0.28 - 0.83
0.40 - 1.30
0.28 - 0.68
0.37 - 0.99
0.58 - 1.63
0.42 - 5.78
0.83 - 2.20
0.50 - 1.06
0.97 - 1.93
0.92 - 2.17
0.72 - 1.84
0.23 - 0.74
1.03 - 3.44
1.15 - 1.36
1.44 - 2.95
1.02 - 2.26
1.00 - 2.18
0.59 - 1.28
0.58 - 1.09
1.00 - 1.02
*Relative risk

111

Table 66 Cross Sectional Head to Head Comparison – Days of Long-term Care
Stays

Independent Variables RR* 95% CI
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous days of long-term care stays
1.18
1.67
0.99
0.81
0.72
0.60
2.49
3.17
4.99
1.10
0.89
1.04
0.41
0.86
1.36
0.86
1.05
1.11
0.77
0.65
0.85
1.05
1.01
0.81 - 1.72
0.97 - 2.87
0.72 - 1.35
0.38 - 1.75
0.34 - 1.51
0.17 - 2.10
0.94 - 6.62
1.15 - 8.74
1.90 - 13.14
0.70 - 1.72
0.64 - 1.25
0.52 - 2.07
0.20 - 0.82
0.50 - 1.48
0.68 - 2.73
0.47 - 1.57
0.95 - 1.16
0.75 - 1.64
0.43 - 1.38
0.41 - 1.03
0.59 - 1.22
0.81 - 1.36
1.01 - 1.01
*Relative risk

112

Table 67 Cross Sectional Head to Head Comparison – Number of Suicide
Attempts

Independent Variables RR* 95% CI
Risperidone
Quetiapine
Female
Urban
Mixed
35 ≤ Age < 45
45 ≤ Age < 55
55 ≤ Age < 65
65 ≤ Age
Black
Other Races
Bipolar Disorder - Manic
Bipolar Disorder - Depressed
Bipolar Disorder - Mixed
Bipolar Disorder - NOS
Bipolar Disorder - Others
Charlson Index
Schizophrenia
Substance Abuse
Other mental disorders
Mood Stabilizer Use
Anti Depressant Use
Previous number of suicide attempts
0.80
0.69
1.14
1.04
1.04
0.59
0.63
0.43
0.13
0.98
0.85
0.75
0.85
1.11
0.94
0.92
1.06
1.25
1.54
1.18
1.16
0.79
1.97
0.56 - 1.15
0.45 - 1.05
0.82 - 1.57
0.64 - 1.70
0.52 - 2.09
0.39 - 0.87
0.46 - 0.86
0.24 - 0.80
0.00 - 224.84
0.66 - 1.45
0.60 - 1.22
0.48 - 1.19
0.54 - 1.36
0.72 - 1.72
0.60 - 1.47
0.54 - 1.57
0.97 - 1.15
0.91 - 1.71
1.06 - 2.24
0.80 - 1.73
0.79 - 1.69
0.55 - 1.14
1.53 - 2.54
*Relative risk

113
Table 68 Summary of the Panel Data and Cross Sectional Models

Outcome
Measures
Panel Data Cross sectional
Risperidone Quetiapine Risperidone Quetiapine
Est. p Est. P Est. p Est. p
Total
Healthcare
Costs
86

0.920 957 0.366 429.3

0.64 831.0

0.46
Pharmacy -111.6

-646.6

0.67 0.05 -339.3

0.13 -523.0

0.06
Medical
Costs
366 0.613 1727

0.098 662.1

0.44 1458.4

0.17
LTC costs -98 0.85 987 0.14 71.2

0.89

1219.7

0.06

Inpatient 169 0.762 484 0.433 276.9

0.62 342.3

0.62
Outpatient 27

-0.884 -95 0.745 125.9

0.57

-158.6

0.56

CMHC 165

0.661 196

0.617 348.9

0.25 294.6

0.43

Number of
Hospitalizati
on
0.691

0.76 - 1.20

0.590

0.73 - 1.20

0.88 0.75 - 1.02

0.88

0.71 - 1.10

Days of
inpatient
stays
1.03

0.54 - 2.00

1.08

0.61 - 1.90

0.95

0.62 - 1.46

0.80 0.52 - 1.25

Days of LTC
Stays
0.89

0.66 - 1.19

1.29

0.84 - 1.98

1.18

0.81 - 1.72

1.67

0.97 - 2.87

Number of
suicide
attempts
0.84

0.54 - 1.29

0.65 0.41 - 1.02

0.80

0.56 - 1.15

0.69

0.45 - 1.05

114
CHAPTER 4 DISCUSSION AND CONCLUSION
This study evaluated the comparative treatment effects of second generation
antipsychotic medications over the conventional first generation antipsychotics in
treatment of bipolar disorders using the panel data approach to address the unobserved
heterogeneity common in retrospective claims data analysis. To our knowledge, this
is one of the first studies that implemented the panel data models in evaluating the
treatment effects of pharmacotherapy. Panel data approach provides a powerful tool
for controlling for individual-specific time-invariant factors for treatment decisions.
The comparative treatment effects of three most commonly used second-generation
antipsychotic medications were also evaluated.
At baseline, patients in the first generation group had significantly higher total
healthcare costs compared to their peers in the second-generation group. Patients in
the first generation group had significantly higher medical costs (total costs –
pharmacy costs), long-term care costs and outpatients costs than the second generation
group while the two groups were not statically different in terms of other itemized
costs including pharmacy costs and outpatient mental healthcare facility.
It appears African American patients are significantly less likely than white
patients to be treated with second-generation antipsychotic medications controlling for
other factors. Similar but statistically nonsignificant trend was observed for other
minorities. Other ethnic minority groups including Hispanics and patients with other
ethnicities were less likely to be treated with newer generation antipsychotics than
white patients. Disparities in drug treatment patterns have been reported in other

115
disease conditions and drug categories including SSRIs for major depression and
antipsychotics for schizophrenia (Kuno 2002, Opolka 2004, Blazer 2000). The results
from the current study are consistent with the reports from these previous studies and
show that the disparities in treatment patterns expands to antipsychotic therapy with
bipolar disorders enrolled in Medi-Cal. Also disparities among rural populations were
observed in this study. Future studies investigating the multifaceted treatment
decision including clinical symptoms as well as other patient characteristics should be
conducted to understand the persistent racial disparities in pharmacotherapy for mental
disorders. Further research on the factors the implications of ethnic disparities in
treatment patterns for quality of care and patient outcomes is also warranted.
A previous diagnosis of major depression and/or bipolar disorder, depressed
subtype (ICD-9-Code: 296.5) had significant implications for second-generation
antipsychotic use while a previous schizophrenia diagnosis or a previous diagnosis of
other bipolar subtype did not. Generally speaking, it appears that the patients on
typical antipsychotics generally have poorer health than their counterpart.
Specifically, patients who were prescribed for typical antipsychotics ; 1) had more co-
morbid conditions, 2) had higher baseline healthcare costs, and 3) were more likely to
have used long-term care facilities than their peers who used the second generation
antipsychotics.
For the head-to-head comparisons among the second-generation medications,
patients on risperidone had significantly higher pharmacy benefit costs and therefore
total healthcare costs than patients on either olanzapine or quetiapine. However, they

116
were not different in the baseline medical costs. Unlike the comparison between the
first- versus second-generation antipsychotic medications, patient ethnicity was not
significantly associated with significant differences in treatment decision among the
three second-generation medications compared. Patients who used long-term care
facilities were significantly more likely to be given risperidone or quetiapine than
olanzapine. In addition, patients who had a claim for at least on diabetes medication
were significantly less likely to receive olanzapine than risperidone. Olanzapine has
been associated with potential metabolic side effects such as weight gain,
hyperglycemia and/or diabetes and the results may suggest that physicians consider
the potential side effects into consideration when they make treatment decision
between the second-generation antipsychotic medications.
It appears that the second-generation antipsychotic medications were
associated with improved drug utilization patterns. With a well-documented favorable
adverse effect profile, patients on second-generation antipsychotics stayed on their
medications significantly longer than the patients on typical antipsychotics. Similarly,
patients received a second-generation antipsychotic medication stayed on the
antipsychotic therapy continuously significantly longer than the patients who received
a typical antipsychotic. Total duration of antipsychotic therapy over a year period
after the initiation was also significantly higher in the second-generation group than in
the typical antipsychotic group. In addition, patients who received a second-
generation were significantly less likely to switch to or augment with another
antipsychotic medication than patients who received typical antipsychotics.

117
However, the results of panel data analyses suggest that the improvement in
utilization pattern due to improved tolerability and comparable efficacy did not
translate into savings in healthcare costs. Second-generation antipsychotics were not
superior to typical antipsychotics in terms of economic outcomes assessed and second-
generation antipsychotic use resulted in significant increase in total healthcare costs to
Medi-Cal compared to typical antipsychotic use. In particular, the second-generation
group had a substantially greater increase in their pharmacy costs. However, second-
generation antipsychotic medication use did not result in significant improvement in
medical utilization patterns and therefore the increase in the pharmacy costs was not
offset by a reduction in medical costs. Specifically, there was no significant difference
in changes in outpatient costs, community mental healthcare center costs, long-term
care facility costs, or other costs. The second-generation antipsychotic group had a
greater, albeit statistically non-significant, increase in hospitalization costs after the
initiation of antipsychotic therapy compare to the typical antipsychotic group. The
results form the analyses of other medical utilization patterns were consistent with the
economic analyses. Second-generation antipsychotic use did not have significantly
positive impact on the utilization of medical care compared to typical antipsychotic.
The rate of hospitalizations, length of hospital stays or length of long-term care stays
did not differ significantly between the two study groups. Also, the rate of suicide
attempts did not change significantly differently after the antipsychotic therapy
between the two groups.

118
When the duration of therapy was compared, the longer the patient stays on
second-generation antipsychotics, the higher the total healthcare costs became while
typical antipsychotics did not have significant association between the duration of
therapy and total healthcare costs. It can be hypothesized that while the typical
antipsychotic medications are discontinued when manic symptoms are controlled, i.e.
used solely as an anti-manic agent, patients on second-generation antipsychotic
medications stay on their medication much longer often extended to be used as a mood
stabilizer itself or as an adjunct therapy to a mood stabilizer. This argument is further
supported by the fact that significantly higher proportion of patients in the second-
generation group remained on the antipsychotic therapy six months after the initiation.
However, the results of the current study do not seem to warrantee merits of the
extended antipsychotic therapy. When patients who had a shorter antipsychotic
therapy were selected and compared across the groups, the second-generation group
was no longer had significantly higher healthcare costs implying that the negative
economic impact of the second-generation may be resulting from the significantly
longer duration of therapy observed in the patients treated with second-generation
antipsychotics. While evidences of benefits of long-term therapy with other types of
medications used for bipolar disorder such as lithium or valproic acids have been well
documented, few studies have focused on the evaluation of long-term therapy of
antipsychotic medication in a real world setting. Furthermore, even though second-
generation antipsychotic medications have been approved as a maintenance therapy,
the comparative effectiveness of second-generation antipsychotics as a mood stabilizer

119
over other commonly used mood stabilizers have not been well established. The
results of the analysis imply that while second-generation antipsychotics may provide
clinical benefits to the bipolar patients without significant additional financial burden
to Medi-Cal when used as an anti-manic agent, the extended use of second-generation
observed in this population may have resulted in the substantial increase in healthcare
costs without clear improvements either in economic or clinical outcomes. Future
research evaluating the extended use of second-generation antipsychotic medication
and its implication in patient outcome is warranted.
Olanzapine has been associated with weight gain and potential risk of diabetes
and it appears that the metabolic side effects of olanzapine was a significant factor in
making the treatment decisions for risperidone over olanzapine for the patients who
had previously been treated with anti-diabetic medication(s). Quetiapine was more
likely to be prescribed for patients who had other psychiatric comorbidities than
olanzapine. In addition, the patients who were treated with quetiapine had
significantly higher total healthcare costs than the patients treated with olanzapine.
Quetiapine was a relatively newer medication used for the treatment of bipolar
disorder compared to the other second-generation medications included in this study
and patients who had a history of failure of treatment with other second-generation
antipsychotics might have been prescribed for quetiapine. However, the results imply
that quetiapine was not associated with significantly improved outcomes compared to
other second-generation antipsychotic medications.

120
Compared to olanzapine, quetiapine use was associated with decreased
pharmacy costs and increased medical costs and the differences were approaching
statistical significance. Based on average whole sale price, quetiapine was
significantly less expensive than olanzapine in 2002 and the lower pharmacy costs in
the quetiapine group probably reflect the price difference between olanzapine and
quetiapine. However, the quetiapine group had greater increases in utilizations of
other medical services compared to the olanzapine group, which offset the savings in
pharmacy costs. The difference in economic and clinical effectiveness between
olanzapine and risperidone were negligible.
The results from the conventional cross sectional analyses were largely
consistent with the results from the panel data analyses. The consistency in results
may suggest that the omitted variables problem due to unobservable time-invariant
heterogeneity that may have affected the treatment decision was negligible when a set
of well-defined explanatory variables such as patient demographic characteristics,
patients’ previous medical and pharmacy utilization patterns and comorbidities are
included in the cross sectional OLS models. In other words, the treatment decision
was not correlated with the error term, i.e. exogenous. It could also imply that even
when unobservable patient heterogeneity was not negligible, the availability of proxy
variables in the data could eliminate or mitigate the problem. For example, Charlson
Comorbidity index was included as a proxy for patient’s general health status and
health risk, which would have significant impact on all treatment outcomes.

121
For the current study, it appears that implementation of panel data approach
did not provide additional benefits in estimating the treatment effect of different
antipsychotic medications compared to cross sectional approach. However, this
cannot be directly generalized to research based on different settings or data and the
decision to implement panel data models should be made in consideration of the
objectives of the study and the limitations of the data to be used. It should be noted
that multiple factors could have contributed to the consistency of the results from the
two different approaches. The current study used Medi-Cal claims data with a
multiple years of follow-up period that enabled researchers to control for various
patient characteristics even within the limited cross sectional framework. Also, unlike
administrative claims from commercially insured populations, Medi-Cal claims data
provide additional information on patient characteristics such as ethnicity and urbanity
of the residence of the patients that could have contributed to omitted variable
problems. Within the panel data framework, omitted variable problems from this type
of time-invariant patient characteristics can be efficiently dealt with.
The results of this research should be interpreted within the context of its
limitations. First, as with other study with a sample population, the study population
does not represent the general population and the results of this study may have
limited generalizability. This study evaluates the impact of second-generation
antipsychotic use in bipolar patients enrolled in Medi-Cal and the study results may
not be directly applied to other populations or other disease conditions. Patients
enrolled in a commercial plan, for example, often have significantly different patient

122
characteristics and treatment patterns and the results from the current study may
provide limited insight into the treatment effect of second-generation antipsychotics in
a commercially insured population. Also, the results of the study may not be
transferable to other mental disorders that may require antipsychotic therapy such as
schizophrenia. The pharmacotherapy of bipolar disorder typically involves a complex
regimen of different classes of medications including mood stabilizers and
antidepressants as well as antipsychotics and therefore, the implications for the
antipsychotic therapy in bipolar disorder may be different from that in schizophrenia
where antipsychotic medication is often the single medication recommended.
Second, this study only evaluated the comparative treatment effect of second-
generation antipsychotic medications on utilization of healthcare resources from a
payer’s perspective and does not include other facets of treatment outcomes such as
patient health-related quality of life. It can be suggested that with a substantially
improved side effect profiles, second-generation antipsychotics treatment would
positive impact on health related quality of life compared to the typical antipsychotics.
However, the current evidences for superiority of second-generation antipsychotics
over typical antipsychotics in improving patient health-related quality of life are mixed
(Jones 2006).
Limitations of the data used for the study should also be discussed. Most
notably, the diagnosis code for each visit or medical utilization was not available. The
unavailability of the diagnosis code for each visit limited the researcher’s ability to
identify whether the index antipsychotic treatment was due to acute manic symptoms

123
and therefore allowing heterogeneity of the study population. To resolve this issue,
patients were required to have a gap in antipsychotic therapy to be entered into the
study population. Also, due to the unavailability of the diagnosis information, this
study did not differentiate the bipolar disorder – specific healthcare costs from the
total healthcare costs. However, the current study is to evaluate the impact of different
antipsychotic use on Medi-Cal and therefore, the total healthcare costs should be more
relevant. Also, given the characteristics of the study population - patients with severe
mental illnesses - it can be reasonably assumed that the disease-specific healthcare
costs would follow the similar pattern. Similarly, due to the unavailability of the
diagnosis codes, physician specialty and/or detailed place of service codes, the
utilization patterns investigated in this study do not include the nature of the visits
such as ‘severity’ or ‘necessity’. For example, ER visits were not differentiated from
regular hospitalizations in the current dataset and therefore, it was not feasible to
evaluate the impact of second-generation on ER visits separate from hospitalizations.
Although beyond the scope of the current study, this type of analysis may provide
implications for the impact of different antipsychotics on the appropriate management
of bipolar disorder.
Finally, while panel data approach may allow researchers to estimate treatment
effects without imposing the assumption of ignorability of treatment (e.g. propensity
scoring method) or without an instrument variable available, its validity hinges on the
assumption that the treatment decision is uncorrelated with time-varying
unobservables that also affects the outcomes of the treatment. In the current study, it

124
has been assumed that unobserved heterogeneity that may affect treatment decisions is
time-invariant. Although reasonable given the treatment patterns observed in patients
with bipolar disorders and other chronic mental conditions, this assumption may be
tested by incorporating time-varying heterogeneity into the model.
Limitations of the current study provide a lead to potential areas of future
research. This study tested the applicability of panel data fixed-effect approach to
estimate treatment effects of different antipsychotic medications. Much research
remains to be done to evaluate the applicability of other econometric techniques that
can effectively address self-selection/omitted variables problems such as instrumental
variable method or propensity scoring methods and compare the advantages of
different approaches. Potential instruments for the treatment selection may include
but not limited to providers’ prescribing habits or the ‘diffusion’ rate of second-
generation antipsychotic use in the area. As previously mentioned, time-varying
unobserved heterogeneity can also be incorporated in the current panel data model
when the treatment decisions are suspected to be correlated with time-varying
unobservables.
This study can be further expanded to evaluate the impact of different
antipsychotics on management of bipolar disorder. For example, the nature of ER
visits can be categorized into different severity based on evidence of subsequent
hospitalizations. Patients whose bipolar symptoms are not properly managed through
medications probably are more likely to use ER as their regular care creating
avoidable financial burden. Evaluating the impact of second-generation particularly

125
on low severity ER visits that does not require subsequent hospitalization may provide
valuable insight into whether the newer antipsychotics had positive impact on proper
management of the illness.
Another area that needs further investigation may include adherence to
antipsychotic medication and its outcomes compared with those of other bipolar
medications. Consistent with existing evidences, patients treated with second-
generation antipsychotics stayed on their antipsychotic medications significantly
longer than compared to the patients treated with typical antipsychotics. However,
the results of this may suggest that this extended antipsychotic therapy might not be
necessarily associated with improved economic outcomes. Few studies have found
the benefits of long-term antipsychotic therapy while a substantial body of literature
exists for the benefits of long-term therapy with conventional mood stabilizers such as
lithium or divalproex in a real world setting. Given the increasingly prevalent use of
second-generation antipsychotics as a mood stabilizer which requires longer-term
therapy than an anti-manic agent, much research remains to be conducted on the
impact of extended therapy with antipsychotic medications on patient outcomes.

126

References

Altamura AC, Slavadori D, Madaro D, Santini A, Mundo E. Efficacy and
tolerability of quetiapine in the treatment of bipolar disorder: preliminary evidence
from a 12-month open-label study. Journal of Affective Disorders 2003;76:267-
271.

American Psychiatric Association. American Psychiatric Association: Diagnostic
and Statistical Manual of Mental Disorders, Fourth Edition. American Psychiatric
Association. Washington, DC 2000.

American Psychiatric Association. Practice guideline for the treatment of Patients
with bipolar disorder (Revision). American Journal of Psychiatry 2002;159:1-50.

Angrist JD, Krueger AB. Instrumental variables and the search for identification:
from supply and demand to natural experiments. Working paper 8456. National
Bureau of Economic Research. Cambridge MA 2001.

Angst J, Selloro R. Historical perspectives and natural history of bipolar disorder.
Biological Psychiatry 2000;48:445-57.

Arellano M, Honore B. Panel Data Models: Some Recent Development. In:
Heckman JJ, Leamer E. (Eds) Handbook of Econometrics. Volume 5. Elsevier
Science B.V. Amsterdam, the Netherlands. 2001.

Baldessarini RJ, Tondo L, Hennen J. Treating the suicidal patient with bipolar
disorder: reducing suicide risk with lithium. Annals of the New York Academy of
Science 2002;932:24-38.

Bartels SJ, Forester B, Miles KM, Joyce T. Mental health service use by elderly
patients with bipolar disorder and unipolar major depression. American Journal of
Geriatric Psychiatry 2000;8(2):160-166.

Berk M, Berk L. Mood stabilizers and treatment adherence in bipolar disorder:
addressing adverse events. Annals of Clinical Psychiatry 2003;15:217-224.

Blanco C, Laje G, Olfson M, Marcus SC, Pincus HA. Trends in the treatment of
bipolar disorder by outpatient psychiatrists. American Journal of Psychiatry
2002;159:1005-1010.

127
Blazer DG, Hybels CF, Simonsick EM, et al. Marked differences in antidepressant
use by race in an elderly community sample: 1986 -1996. American Journal of
Psychiatry 2000;157(7):1089-1094.

Bostwick JM, Pankratz VS. Affective Disorders and Suicide Risk: a
Reexamination. American Journal of Psychiatry 2000;1257:1925-1932.

Bowden CL. Predictors of response to divalproex and lithium. Journal of Clinical
Psychiatry 1995;56(suppl 3):25-30.

Bowden CL, Brugger AM, Swann AC, et al. Efficacy of divalproex vs. lithium and
placebo in the treatment of mania. Journal of American Medical Association
1994;271:918-24.

Bowden CL, Calabrese JR, McElroy SL, et al. A randomized, placebo-controlled
12-month trial of divalproex and lithium in treatment of outpatients with bipolar
disorder. Archives of General Psychiatry 2000;57:481-489.

Brown ES, Suppes T, Adinoff B, Thomas NR. Drug abuse and bipolar disorder:
comorbidity or misdiagnosis? Journal of Affective Disorders 2001;65:105-115.

Calabrese JR, Fatemia SH, Kujawas M, et al. Predictors of response to mood
stabilizers. Journal of Clinical Psychopharmacology 1996;16 (2 Suppl 1):24S-31S.

Calabrese JR, Woyshiville MJ. Lithium therapy: limitations and alternatives in the
treatment in bipolar disorders. Annals of Clinical Psychiatry 1995;72:103-112.

California Department of Health Services. California’s Medical Assistance
Program: Annual Statistical Report, Calendar Year 2003. Sacramento, California
Department of Health Services, Medical Care Statistics Section, 2003.

Chamberlain G. Heterogeneity, omitted variable bias, and duration dependent. In:
Heckman, JJ, Singer, B. (Eds), Longitudinal analysis of labor market data.
Cambridge University Press, Cambridge, United Kingdom 1985.

Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying
prognostic comorbidity in longitudinal studies: Development and validation.
Journal of Chronic Disease 1987;40:373-383.

Dion GL, Tohen M, Anthony WA et al. Symptoms and functiningof patients with
bipolar disorder six months after hospitalization. Hospital & Community
Psychiatry 1988;39:652-657.

128
Duggan MG. Does Medicaid pay too much for prescription drugs? A case study of
atypical anti-psychotics. NBER Working Paper Series. National Bureau of
Economics Research. Cambridge, MA 02138. 2003.

Garfinkel PE, Stancer HC, Persad E. A comparison of haloperidol, lithium
carbonate and their combination in the treatment of mania. Journal of Affective
Disorders 1980;2(4):279-88.

Gianfrancesco FD, Grogg AL, Nahmoud RA, Wang R, Nasrallah HA. Differential
effects of risperidone, olanzapine, clozapine, and conventional antipsychotics on
type 2 diabetes: findings from a large health plan database. Journal of Clinical
Psychiatry 2002;63:920-930.

Gianfrancesco FD, Sajatovic M, Rajagopalan K, Wang R. Antipsychotic treatment
adherence and associated mental health care use among individuals with bipolar
disorder. Clinical Therapeutics 2008;30(7):1358-1374.

Gibson PJ, Damler R, Jackson EA, Wilder T, Ramsey JL. The impact of
olanzapine, risperidone, or haloperidol on the cost of schizophrenia care in a
Medicaid population. Value in Health 2004;7(1):22-35.

Goodwin FK, Fireman B, Simon GE, Hunkeler EM, Lee H, Revicki D. Suicide
risk in bipolar disorder during treatment with lithium and divalproex. Journal of
American Medical Association 2003;290:1467-1473.

Guille C, Sachs GS, Ghaemi SN. A naturalistic comparison of clozapine,
risperidone, and olanzapine in the treatment of bipolar disorder. Journal of Clinical
Psychiatry 2000;61:638-642.

Hausman J, Hall BH, Griliches Z. Econometric models for count data with an
application to the patents-R&D relationship. Econometrica 1984;52(4):909-938.

Henderson DC, Cagliero E, Gray C, Nasrallah RA, Hayden DL, Schoenfeld DA,
Goff DC. Clozapine, diabetes mellitus, weight gain, and lipid abnormalities: a
five-year naturalistic study. American Journal of Psychiatry 2000;157:975-981.

Hsiao C. Analysis of Panel Data. Cambridge University Press, Cambridge, United
Kingdom. 2003.

Judd L, Akiskal HS. The prevalence and diability of bipolar spectrum disorders in
the US population: re-analysis of the ECA database taking into account
subthreshold cases. Journal of Affective Disorders. 2003;73:123-131.

129
Kane JM. The role of neuroleptics in manic-depressive illness. Journal of Clinical
Psychiatry 1988;49(suppl,11):12-13.

Kane JM, Smith JM. Tardive dyskinesia: prevalence and risk factors. Archives of
General Psychiatry 1982;39:473-481.

Keck PE, McElroy SL, Strakowski SM. Compliance with maintenance treatment
in bipolar disorder. Psychopharmacology Bulletin 1997;33:87-91.

Keck PE, McElory SL, Strakowski SM, et al. Factors associated with maintenance
antipsychotic treatment of patients with bipolar disorder Journal of Clinical
Psychiatry 1996:57:147-151.

Keck PE, Nabulsi AA, Taylor JL, Henke CJ, Chmiel JJ, Stanton SP, Bennett JA. A
pharmacoeconomic model of divalproex vs. lithium in the acute and prophylactic
treatment of bipolar I disorder. Journal Clinical Psychiatry 1996;57:213-222.

Keck PE, Versiani M, Potkin S, West SA, Giller E, Ice K. Ziprazidone in the
treatment of acute bipolar mania: a three-week, placebo-controlled, double-blind,
randomized trial. American Journal of Psychiatry 2003;160:741-748.

Kessler RC, McConagle KA, Shao S, et al. Lifetime and 12-month prevalence of
DSM-III-R psychiatric disorders in the United States. Archives of General
Psychiatry 1994;51:8-19.

Koro CE, Fedder Do, L’Italien GJ, et al. Assessment of independent effect of
olanzapine and risperidone on risk of diabetes among patients with schizophrenia:
population based nested case-control study. British Medical Journal 2002;325:243-
245.

Kuno E, Rothbard AB. Racial disparities in antipsychotic prescription patterns for
patients with schizophrenia. American Journal Psychiatry 2002; 159(4): 567-572.

Kusumaker V. Antidepressants and antipsychotics in the long-ter treatment of
bipolar disorder. Journal of Clinical Psychiatry 2002;63(suppl10):23-28.

Lancaster T. The incidental parameter problem since 1948. Journal of
econometrics 2000;95:391-413.

Lewin Group. Access and Utilization of New Antidepressant and Antipsychotic
Medications. Available at:
http://www.aspe.hhs.gov/health/reports/psychmedaccess/chap06.htm. Accessed
January 2005

130
Li J, McCombs JS, Stimmel GL. Cost of treating bipolar disorder in the California
Medicaid (Medi-Cal) program. Journal of Affective Disorders 2002:71;131-139.

Lish JD, Dime-Meenan S, Whybrow PC, et al. The national depressive and manic-
depressive association (DMDA) survey of bipolar members. Journal of Affective
Disorders 1994;31:281-294.

Maj M, Pirozzi R, Magliano L, Bartoli L. Long-term outcomes of lithium
prophylaxis in bipolar disorder: a 5-year prospective study of 402 patients at a
lithium clinic. American Journal of Psychiatry 1998;155(1):30-35.

Manning JS, Hackal RF, Connor PD, et al. One the nature of depressive and
anxious states in a family practice setting: the high prevalence of bipolar II and
related disorders ina cohort followed longitudinally. Comprehensive Psychiatry
1997;38:102-108.

McCombs JS, Mulani P, Gibson PJ. Open access to innovative drugs: treatment
substitutions or treatment expansion? Health Care Financing Review
2004;25(3):35-53.

McCombs JS, Nichol MB, Johnstone BM, Stimmel GL, Shi J, Smith R,
Antipsychotic Drug Use Patterns and the Cost of Treating Schizophrenia.
Psychiatry Services. 2000;51(4):525-527.

McCombs JS, Nichol MB, Stimmel GL, Shi J, Smith RR. Use patterns for
antipsychotic Medications in Medicaid Patients with Schizophrenia. Journal of
Clinical Psychiatry 1999;60 (suppl 19):5-11.

Meyer JM. A retrospective comparison of weight, lipid, and glucose changes
between risperidone- and olanzpine-treated inpatients: metabolic outcomes after 1
year. Journal of Clinical Psychiatry 2002;63:425-433.

Mitchell PB, Malhi GS. The expanding pharmacopoeia for bipolar disorder
Annual Review of Medicine 2002:53;173-188.

Mukherjee S, Rose AM, Caracci G, et al. Persistent tardive dyskinesia in bipolar
patients. Archive of General Psychiatry 1986;43:342-346.

Muller-Oerlinghausen B, Berghofer A, Bauer M. Bipolar Disorder. Lancet
2002;359:241-247.

Murray CJL, Lopez AD. The global burden of disease. World Health
Organization. Harvard University Press. Cambridge, MA 1996.

131
Narayan S, Sterling KL, McCombs JS. The impact of open access to atypical
antipsychotics on treatmente cost for Medi-Cal patients with bipolar disorder.
Disease Management and Health Outcomes 2006;14(5):287-301.

Narrow WE, Rae DS, Robins LN. Revised prevalence estimates of mental
disorders in the United States: using a clinical significance criterion to reconcile 2
surveys’ estimates. Archives of General Psychiatry 2002;59:115-123.

Nemeroff CB. An ever-increasing pharmacopoeia for the management of patients
with bipolar disorder. Journal of Clinical Psychiatry 2000;61(suppl 13):19-25.

Opolka JL, Rascati KL, Brown CM, Gibson PJ. Ethnicity and prescription patterns
for haloperidol, risperidone,and olanzapine. Psychiatric services 2004; 55(2):151-
156.

Peele PB, Xu Y, Kupfer DJ. Insurance expenditures on bipolar disorder: clinical
and parity implications. American Journal of Psychiatry 2003;160(7):1286-1290.

Perkins DO. Adherence to Antipsychotic Medication. Journal of Clinical
Psychiatry 1999;60(suppl 21): 25-30.

Perugi G, Akiskal HS, Lattanzi L, et al. The high prevalence of “soft” bipolar (II)
features in atypical depression. Comprehensive Psychiatry 1998;39:63-71.

Peter B, Jones, Thomas RE. Barnes, Linda Davies, et al. Archive of General
Psychiatry. 2006;63:1079-1087.

Regenold WT, Thapar RK, Marano C Gavirneni S, Kondapavuluru PV. Increased
prevalence of type 2 diabetes mellitus among psychotic inpatients with bipolar I
affective ad schizoaffective disorders independent of psychotropic drug use.
Journal of Affective Disorders 2002;70:19-26.

Regier DA, Farmaer ME, Rae DS, et al. Comorbidity of mental disorders with
alcohol and other drug abuse. Journal of American Medical Association
1990;264:2511-2518.

Ridder G and Tunali I, Stratified partial likelihood estimation. Journal of
Econometrics 92, 193-232.

Romano PS, Roos LL, Jollis JG. Adapting a clinical comorbidity index for use
with ICD-9-CM administrative data: Differing perspectives. Journal of Clinical
Epidemiology 1993;46:1075-1079.

132
Rosenheck R, Perlick D, Bingham S et al. Effectiveness and costs of olanzapine
and haloperidol in the treatment of schizophrenia. A randomized controlled trial.
Journal of American Medical Association 2003;290(20):2693-2702.

Tohen M, Baker RW, Altshuler LL, et al. Olanzapine versus divalproex in the
treatment of acute mania. The American Journal of Psychiatry 2002;159(6):1011-
1017.

Tohen M, Chengappa RK, Suppers T, et al. Efficacy of olanzapine in combination
with valproate or lithium in the treatment of mania in patients partially
nonresponsive to valproate or lithium monotherapy. Archives of general
psychiatry 2002:59(1):62-69.

Tohen M, Jacobs TG, Grundy SL, et al. Efficacy of olanzapine for acute bipolar
mania: a double-blind, placebo-controlled study. Archive of general psychiatry
2000;57(9):841-849.

Tohen M, Zhang F, Keck PE, et al. Olanzapine versus haloperidol in
schizoaffective disorder, bipolar type. Journal of Affective Disorders
2001;67:133-140.

Sachs GS, Grossman F, Ghaemi NS, Okamoto A, Bowden CL. Combination of a
mood stabilizer with risperidone or haloperidol for treatment of acute mania: a
couble-blind, placebo-controlled comparison of efficacy and safety. American
Journal of Psychiatry 2002;159:1146-1154.

Sachs GS, Printz DJ, Kahn DA, et al. The expert consensus guidelines series:
medication treatment of bipolar disorder. New York, McGraw-Hill, April 2000.

Sanger TM, Grundy SL, Gibson PJ, et al. Long-term olanzapine therapy in the
treatment of bipolar I disorder: an open-label continuation phase study. Journal of
Clinical Psychiatry 2001; 62:273-281.

Scott J and Pope M. Nonadherence with mood stabilizers: Prevalence and
predictors. Journal of Clinical Psychiatry 2002;63:384-390.

Sernyak MJ, Griffin RA, Johnson RM, et al. Neuroleptic exposure following
inpatient treatment of acute mania with lithium and neuroleptic. American Journal
of Psychiatry 1994;151:133-135.

Sernyak MJ, Leslie DL, Alarcon RD, Losonczy MF, Rosenheck R. Association of
diabetes mellitus with use of atypical neuroleptics in the treatment of
schizophrenia. American Journal of Psychiatry 2002;159:561-566.

133
Shi L, McCombs JS, Thiebaud P. The impact of unrecognized bipolar disorders for
patients treated for depression with antidepressants in the fee-for-services
California Medicaid (Medi-Cal) program. Journal of Affective Disorder
2004;82(3):373-83.

Simon GE, Unutzer J. Health care utilization and costs among patients treated for
bipolar disorder in an insured population. Psychiatric Services 1999:50(10):1303-
1308.

Stender M, Bryant-Comstock L, Philips S. Medical resource use among patients
treated for bipolar disorder: a retrospective, cross-sectional, descriptive analysis.
Clinical Therapeutics 2002;24(10):1668-1676.

Swann AC, Bowden CL, Calabrese JR, et al. Differential effect of number of
previous episodes of affective disorder response to lithium or divalproex in acute
mania. American Journal of Psychiatry 1999;156:1264-1266.

Vella F. Estimating models with sample selection bias: a survey. Journal of human
resources 1998:33(1):127-169.

Vieta, E, Goikolea JM, Corbella B, Benabarre A, Reinares M, Martinez G,
Fernandez A, Colom F, Martinex-Aran A, Torrent C. Risperidone safety and
efficacy in the treatment of bipolar and schizoaffective disorders: results from a 6-
month, multicenter, open study. Journal of Clinical Psychiatry 2001;62(10): 818-
825.

Vieta E, Reinares M, Corbella B, Bernabarre A, GIlaberte I, Colom F, Martinez-
Aran A, Gasto C, Tohen M. Olanzapine as long-term adjunctive therapy in
treatment-resistant bipolar disorder. Journal of Clinical Psychopharmacology
2001;21(5):469-473.

Weiss RD, Greenfeild SF, Najavits LM, et al. Medication compliance among
patients with bipolar disorder and substance use disorder. Journal of Clinical
Psychiatry 1998;59:172-174.

Woods SW. The economic burden of bipolar disease. Journal of Clinical
Psychiatry 2000;61(Suppl 13): 38-41.

Woodridge JM. Econometric Analysis of Cross Section and Panel Data. The MIT
Press. Cambridge MA 2001

Wyatte RJ, Henter I. An economic evaluation of manic-depressive illness. Social
Psychiatry & Psychiatric Epidemiology 1995:30(5):213-219.

Abstract (if available)

Abstract While the comparative efficacy of second-generation antipsychotics (SGAs) vs. first generation antipsychotics (FGAs) has been well documented, the treatment effect of SGAs in bipolar disorder has not been directly evaluated in a real-world setting. Panel data fixed effects models provide a simple yet powerful tool in measuring treatment effects. The objectives of this study were to investigate the treatment effect of SGAs in bipolar patients enrolled in the fee-for-service (FFS) Medi-Cal program using panel data fixed effect models. The retrospective paid claims files from the Medi-Cal program was used. Patients were included if they had at least 1) one claim(s) for any bipolar medications and 2) one medical claim(s) with ICD-9 codes 296.40-296.99. The population was further defined as: 1) patients who initiated a new antipsychotic therapy between 2000 and 2001 (index date)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Effects of a formulary expansion on the use of atypical antipsychotics and health care services by patients with schizophrenia in the California Medicaid Program

PDF

Estimation of heterogeneous average treatment effect-panel data correlated random coefficients model with polychotomous endogenous treatments

PDF

Understanding primary nonadherence to medications and its associated healthcare outcomes: a retrospective analysis of electronic medical records in an integrated healthcare setting

PDF

The impact of Patient-Centered Medical Home on a managed Medicaid plan

PDF

Selected papers on the evaluation of healthcare costs of prematurity and necrotizing enterocolitis using large retrospective databases

PDF

Crohn’s disease: health outcomes and resource utilization in the biologic era

PDF

The impact of treatment decisions and adherence on outcomes in small hereditary disease populations

PDF

Impact of medication persistence on clinical and health care cost outcomes in patients with major depressive disorder using retrospective claims data

Asset Metadata

Creator Park, Jinhee (author)

Core Title Assessment of the impact of second-generation antipscyhotics in Medi-Cal patients with bipolar disorder using panel data fixed effect models

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Pharmaceutical Economics

Publication Date 07/15/2009

Defense Date 04/07/2009

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag antipsychotics,bipolar disorder,healthcare utilization,Medi-Cal,OAI-PMH Harvest,panel data fixed-effect model

Place Name California (states)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor McCombs, Jeffrey S. (committee chair), Ahn, Jeonghoon (committee member), Hsiao, Cheng (committee member)

Creator Email jinhee.x.park@gsk.com,park4940@yahoo.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m2363

Unique identifier UC1192408

Identifier etd-Park-2939 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-562590 (legacy record id),usctheses-m2363 (legacy record id)

Legacy Identifier etd-Park-2939.pdf

Dmrecord 562590

Document Type Dissertation

Rights Park, Jinhee

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu