University of Southern California Dissertations and Theses

Three essays on comparative economic and clinical outcomes of antipsychotics using retrospective claims data

(USC Thesis Other)

Three essays on comparative economic and clinical outcomes of antipsychotics using retrospective claims data

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THREE ESSAYS ON COMPARATIVE ECONOMIC AND CLINICAL
OUTCOMES OF ANTIPSYCHOTICS USING RETROSPECTIVE CLAIMS
DATA
By
YAWEN JIANG
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALFIORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PHARMACEUTICAL ECONOMICS AND POLICY)
MAY 2016
1
DEDICATION
The dissertation is dedicated to my loving parents.
2
ACKNOWLEDGEMENTS
I wish to express my gratitude to my dissertation advisor, Dr. Jeffrey S. McCombs,
for being patient, encouraging, and supportive in every aspect of my graduate study. His
academic and life experience has always been priceless to me. His optimistic attitude
shall inspire me for many years to come.
I would like to thank Dr. Joel W. Hay for sharing his expertise in econometrics and
helping me to realize my potentials. I would also like to express my appreciation to Dr.
Geoffrey Joyce for his insightful advices which made me realize the importance of
focusing on the very basics of economics, policy, and data. Special thanks to Dr. Michael
B. Nichol and Dr. Steven Fox for their feedback to my dissertation proposal.
My thanks are extended to all my colleagues and friends at USC, who have made my
journey at graduate school enjoyable and delightful.
3
Contents
1. GENERAL TOPIC: THREE ESSAYS ON COMPARATIVE ECONOMIC AND
CLINICAL OUTCOMES OF ANTIPSYCHOTICS USING RETROSPECTIVE
CLAIMS DATA .................................................................................................................11
1.1 INTRODUCTION ..........................................................................................................11
SECTION 1 REFERENCES ................................................................................................. 12
2. COMPARISON OF HEALTH SERVICES USE ASSOCIATED WITH
ZIPRASIDONE AND OLANZAPINE AMONG SCHIZOPHRENIA AND BIPOLAR
DISORDER PATIENTS IN THE USA............................................................................. 15
ABSTRACT ...................................................................................................................... 15
2.1 INTRODUCTION ......................................................................................................... 17
2.2 METHODS ................................................................................................................. 18
2.2.1 Data Sources and Sample Selection .................................................................. 18
2.2.2 Outcomes Measures .......................................................................................... 20
2.2.3 Statistical Analyses ........................................................................................... 21
2.2.4 Sensitivity Analyses .......................................................................................... 22
2.3 RESULTS ................................................................................................................... 23
2.3.1 Baseline Characteristics .................................................................................... 23
2.3.2 Outcomes .......................................................................................................... 25
2.4 DISCUSSION .............................................................................................................. 27
2.5 CONCLUSION ............................................................................................................ 29
CONFLICT OF INTEREST .................................................................................................. 29
SECTION 2 REFERENCES ................................................................................................. 30
SECTION 2 SUPPLEMENTARY MATERIAL ......................................................................... 41
3. IMPROVING MODEL SPECIFICATIONS WHEN ESTIMATING TREATMENT
EFFECTS ACROSS ALTERNATIVE MEDICAL INTERVENTIONS .......................... 44
ABSTRACT ...................................................................................................................... 44
3.1 INTRODUCTION ......................................................................................................... 46
3.2 STATISTICAL CHALLENGES IN OBSERV ATIONAL RESEARCH ...................................... 48
3.3 METHODS ................................................................................................................. 50
3.3.1 Data Sources ..................................................................................................... 50
3.3.2 Definition of the Unit of Analysis ..................................................................... 51
3.3.3 Covariates and Model Sequencing.................................................................... 52
3.4 RESULTS ................................................................................................................... 54
3.5 DISCUSSION .............................................................................................................. 58
SECTION 3 REFERENCES ................................................................................................. 59
4
4. RISK OF ACUTE MAJOR CARDIOV ASCULAR EVENTS AND ACUTE
KIDNEY INJURY ASSOCIATED WITH ANTIPSYCHOTICS ..................................... 68
4.1 INTRODUCTION AND OBJECTIVES ............................................................................. 68
4.2 METHODS ................................................................................................................. 70
4.2.1 Data and Samples .............................................................................................. 70
4.2.2 Statistical Analysis ............................................................................................ 72
4.2.3 Sensitivity Analyses Using Instrumental Variables (IVs) ................................. 73
4.2.4 Other Sensitivity Analyses ................................................................................ 75
4.2.4.1 Extended Pre-index Period ..................................................................... 75
4.2.4.2 Attrition Due to Discontinued Enrollment .............................................. 76
4.2.4.3 Weibull Regressions ................................................................................ 77
4.2.4.4 Less Restrictive Identification of Events ................................................ 77
4.2.4.5 Subsamples of episodes without any overlap ......................................... 77
4.2.4.6 Nonrandom censoring ............................................................................. 78
4.3 RESULTS ................................................................................................................... 78
4.3.1 Descriptive Statistics ......................................................................................... 78
4.3.2 Acute Coronary Syndrome (ACS) and Ischemic Stroke (IS) ........................... 79
4.3.3 Ventricular Arrhythmia ..................................................................................... 80
4.3.4 Acute Kidney Injury .......................................................................................... 80
4.3.5 Hypotension ...................................................................................................... 81
4.3.6 Acute Urinary Retention ................................................................................... 81
4.3.7 Neuroleptic Malignant Syndrome/Rhabdomyolysis ......................................... 81
4.3.8 Pneumonia......................................................................................................... 82
4.3.9 Other Analyses .................................................................................................. 82
4.4 DISCUSSION .............................................................................................................. 83
4.5 CONCLUSIONS .......................................................................................................... 85
SECTION 4 REFERENCES ................................................................................................. 85
APPENDIX TO SECTION 4 .............................................................................................. 100
A4.1. Diagnoses Codes ............................................................................................ 100
A4.2 Covariates .................................................................................................... 101
A4.3 Validity of Instrumental Variables (IVs) ......................................................... 103
A4.3.1 First-stage Statistics .................................................................................. 103
A4.3.2 Exogeneity of IVs ..................................................................................... 104
Appendix A4.3 References ................................................................................... 104
A4.4 Sensitivity Analyses .........................................................................................114
A4.4.1 Endogeneity of choosing between atypical and typical antipsychotics .....114
A4.4.2 Sensitivity analyses using a 360-day pre-index period prior to the first
observed antipsychotic use. ...................................................................................115
A4.4.3 Cox regressions with 2SRI for sample attrition. The number of
months with continuous enrollment prior to the start of the current episode
was used as the IV for sample attrition. .............................................................117
5
A4.4.4 Analyses using Weibull regressions. The results are very similar to
Cox regressions. .................................................................................................119
A4.4.5 Using alternative definition of events. Emergency department (ED)
visits and urgent care visits with target diagnoses were included in addition to
hospitalization with target diagnosis. ................................................................... 121
A4.4.6 Using episodes without overlap. ............................................................... 123
A4.4.7 Using competing risk model to take into consideration of potential
nonrandom censoring. .......................................................................................... 125
6
List of Tables
Table 2. 1 Patient characteristics at baseline. ............................................................... 34
Table 2. 2 Unadjusted differences in outcomes between ziprasidone episodes and
olanzapine episodes ..................................................................................................... 37
Table 2. 3 Adjusted effects of ziprasidone use on costs in relation to olanzapine use . 38
Table 2. 4 Effects of ziprasidone use on number of emergency department
attendances and hospitalizations in post-index period in relation to olanzapine use
evaluated with hurdle models and Poisson models...................................................... 39
Supplementary Table 2. 1 OLS results additionally include interaction between
ziprasidone use and schizophrenia, and OLS results with only first episode of
treatment naïve patients who did not have any poly-therapy episodes (restricted
sample). ........................................................................................................................ 41
Supplementary Table 2. 2 Modified Park Test for GLM regressions of total costs,
medical services costs, and drug costs ......................................................................... 42
Table 3. 1 Impact of atypical antipsychotic use on total cost and duration of
therapy: all episode types (N=731,236) ....................................................................... 61
Table 3. 2 Impact of atypical antipsychotic use on total cost and duration of
therapy: restart episodes (N=445,258) ......................................................................... 62
Table 3. 3 Impact of atypical antipsychotic use on total cost and duration of
therapy: switching episodes (N=71,917) ..................................................................... 63
Table 3. 4 Impact of atypical antipsychotic use on total cost and duration of
therapy: delayed switching episodes (N=97,704) ........................................................ 64
Table 3. 5 Impact of atypical antipsychotic use on total cost and duration of
therapy: augmentation episodes (N=116,357) ............................................................. 65
Table 4. 1 Characteristics of patients at baseline and episode duration across drug
groups ........................................................................................................................... 92
Table 4. 2 Cox regressions: hospitalization with acute coronary syndrome or
ischemic stroke............................................................................................................. 92
7
Table 4. 3 Cox regressions: hospitalization with ventricular arrhythmia .................... 94
Table 4. 4 Cox regressions: hospitalization with acute kidney injury (AKI) ............... 95
Table 4. 5 Cox regressions: hospitalization with Hypotension .................................... 96
Table 4. 6 Cox regressions: hospitalization with acute urinary retention .................... 97
Table 4. 7 Cox regressions: hospitalization with neuroleptic malignant
syndrome/rhabdomyolysis ........................................................................................... 98
Table 4. 8 Cox regressions: hospitalization with pneumonia....................................... 99
Table A4.1. 1 Diagnoses codes used in the study to identify patients and target
outcomes .................................................................................................................... 100
Table A4.2. 1 Covariates used in regressions ............................................................ 101
Table A4.3.1. 1 First-stage multinomial logit regression results ............................... 106
Table A4.3.1. 2 Binary linear probability models of drug choices on IVs ................ 108
Table A4.3.2. 1 Linear regressions of pre-index costs, Charlson Comorbidity
Index (CCI), indicator of hospitalization, and indicator of emergency department
visit on first lag of treatment, second lag of treatment, and specialty indicators ....... 110
Table A4.3.2. 2 Regressions of target events in prior 6-month period on indicators
of prescriber specialties.............................................................................................. 111
Table A4.3.2. 3 Regressions of target events in prior 6-month period on indicators
of first lag of treatment .............................................................................................. 112
Table A4.3.2. 4 Regressions of target events in prior 6-month period on indicators
of second lag of treatment .......................................................................................... 113
Table A4.4.1. 1 Hospitalization with cardiovascular events, acute kidney injury
(AKI), or events that may cause AKI-- Atypical vs. Typical ..................................... 114
Table A4.4.2. 1 Cox regressions: hospitalization with acute major cardiovascular
events using 360-day pre-index period ...................................................................... 115
8
Table A4.4.2. 2 Cox regressions: hospitalization with acute kidney injury (AKI)
or events that may cause AKI using 360-day pre-index period ................................. 116
Table A4.4.3. 1 Cox regressions: hospitalization with acute major cardiovascular
events using 2SRI for sample attrition ....................................................................... 117
Table A4.4.3. 2 Cox regressions: hospitalization with acute kidney injury (AKI)
or events that may cause AKI using 2SRI for sample attrition. ................................. 118
Table A4.4.4. 1 Weibull regressions: hospitalization with acute major
cardiovascular events ................................................................................................. 119
Table A4.4.4. 2 Weibull regressions: hospitalization with acute kidney injury
(AKI) or events that may cause AKI .......................................................................... 120
Table A4.4.5. 1 Cox regressions: hospitalization, ED visits, or urgent care visits
with acute major cardiovascular events ..................................................................... 121
Table A4.4.5. 2 Cox regressions: hospitalization, ED visits, or urgent care visits
with acute kidney injury (AKI) or events that may cause AKI .................................. 122
Table A4.4.6. 1 Cox regressions using episodes without overlap: hospitalization
with acute major cardiovascular events ..................................................................... 123
Table A4.4.6. 2 Cox regressions using episodes without overlap: hospitalization
with acute kidney injury (AKI) or events that may cause AKI .................................. 124
Table A4.4.7. 1 Competing risk model: hospitalization with acute major
cardiovascular events…………………………………………………………………125
Table A4.4.7. 2 Competing risk model: hospitalization with acute kidney injury
(AKI) or events that may cause AKI…………………………………………………126
9
List of Figures
Figure 2. 1 Sample selection process. A total of 15210 episodes representing 9097
patients were selected. 481 patients used both drugs ................................................... 40
Supplementary Figure 2. 1 Diagram of Analysis Period.. . ......................................... 43
Figure 3. 1 Impact of Using Atypical Antipsychotics on Total Cost in First
Post-Treatment Year. .................................................................................................... 66
Figure 3. 2 Impact of Using Atypical Antipsychotics on Duration of Therapy. .......... 67
Figure 4. 1 Sample selection flow chart....................................................................... 91
10
1. General Topic: Three Essays on Comparative Economic and Clinical Outcomes of
Antipsychotics Using Retrospective Claims Data
1.1 Introduction
Antipsychotic medications became the cornerstone of treatments for schizophrenia
and bipolar disorder since their availability in 1950’s (1). Since 1989, several second
generation antipsychotics (SGAs) were introduced into market and soon became first-line
treatment for schizophrenia and bipolar disorder (2). Owing to their reduced
extrapyramidal side effects (EPS) compared to first generation antipsychotics (FGAs),
SGAs dominate the US antipsychotic market today (3). The rapid diffusion of SGAs has
raised controversy in light of their higher costs compared to FGAs (3, 4) and recent
evidence that SGAs may be no more effective than FGAs (4-6). Also, each antipsychotic
drug has a unique safety profile (7), and SGAs are associated with higher risk of
metabolic side effects than FGAs overall (8, 9). Schizophrenia and bipolar disorder
represent two of the most expensive health conditions to private and public healthcare
payers (4, 10, 11). Therefore, it is warranted to obtain reliable evidence on the
comparative effectiveness, safety, and economic impacts of antipsychotics.
The gold standard of drug comparison is randomized clinical trials (RCTs). However,
RCTs fail to provide evidence on many fronts by design. While they have impeccable
internal validity, they exclude many relevant patient groups such as the elderly, those with
11
co-morbidities, and those who have been exposed to other drugs (12). More, they are
conducted in ideal settings to enhance patient adherence which does not necessarily
reflect real-world practice (12). These limitations of RCTs may contribute to discrepancy
between RCT results and daily practice results. Hence, it is important to supplement RCT
results with evidence obtained from real-world data for policy and clinical decision
making.
The proposed dissertation will consist of three separate studies, each comparing costs,
effectiveness or safety across two or more antipsychotics. The first study compares
healthcare services use and costs between two atypical antipsychotics, ziprasidone and
olanzapine using IMS Pharmetrics Database. The second study used Medicaid data to
critique the statistical methods commonly used in observational research and examines
the importance of taking prior treatment history, including episode type into account
when conducting retrospective comparative effectiveness research. The third study
investigates the risks of acute cardiovascular events and acute kidney injury associated
with a wide spectrum of antipsychotics using Humana administrative claims databases.

Section 1 References
1. Alexander GC, Gallagher SA, Mascola A, et al. Increasing off-label use of
antipsychotic medications in the United States, 1995–2008. Pharmacoepidemiology and
drug safety. 2011; 20: 177-84.
2. Leucht S, Kissling W, Davis J. Second-generation antipsychotics for schizophrenia:
12
can we resolve the conflict? Psychol Med. 2009; 39: 1591.
3. Gallini A, Donohue JM, Huskamp HA. Diffusion of Antipsychotics in the US and
French Markets, 1998–2008. Psychiatr Serv. 2013; 64: 680-87.
4. Huskamp HA, O'Malley AJ, Horvitz-Lennon M, et al. How Quickly Do Physicians
Adopt New Drugs? The Case of Second-Generation Antipsychotics. Psychiatr Serv. 2013;
64: 324-30.
5. Swartz M, Stroup TS, McEvoy J, et al. Special Section on Implications of CATIE:
What CATIE Found: Results From the Schizophrenia Trial. Psychiatr Serv. 2008; 59:
500-06.
6. Naber D, Lambert M. The CATIE and CUtLASS Studies in Schizophrenia. CNS
Drugs. 2009; 23: 649-59.
7. Drugs for psychiatric disorders. Treatment guidelines from the Medical Letter. 2013;
11: 53.
8. Meyer JM, Davis VG, Goff DC, et al. Change in metabolic syndrome parameters
with antipsychotic treatment in the CATIE Schizophrenia Trial: prospective data from
phase 1. Schizophr Res. 2008; 101: 273-86.
9. Marder S, Stroup TS. Pharmacotherapy for schizophrenia: Side effect management.
In: Basow DS, ed., Uptodate. Waltham, MA: Uptodate, 2012.
10. Karve SJ, Panish JM, Dirani RG, et al. Health Care Utilization and Costs among
Medicaid-enrolled Patients with Schizophrenia Experiencing Multiple Psychiatric
Relapses. Health Outcomes Research in Medicine. 2012; 3: e183-e94.
13
11. Peele PB, Xu Y , Kupfer DJ. Insurance expenditures on bipolar disorder: clinical and
parity implications. AJ Psychiatry. 2003; 160: 1286-90.
12. Peters DC, van Rensburg A. Drugs: Real World Outcomes. Drugs-Real World
Outcomes. 1-1.
14
2. Comparison of Health Services Use Associated With Ziprasidone and Olanzapine
among Schizophrenia and Bipolar Disorder Patients in the USA
Abstract
Background and Objectives
Ziprasidone is increasingly used for the treatment of schizophrenia and bipolar
disorder. The purpose of this study was to compare healthcare costs and use associated
with ziprasidone and olanzapine.

Methods
Ziprasidone and olanzapine treatment episodes of schizophrenia and bipolar disorder
patients were identified in the 01/2007 –12/2010 IMS LifeLink™ Database. The period
of analysis for each episode has three components: 6 months prior to episode initiation
date (pre-episode period), 1 month immediately following the episode initiation date
(initiation month) and up to 12 months after the end of the initiation month (follow-up).
Ordinary least squares regressions, general linear models, and two-part models were used
to compare various types of costs (2007 US$) associated with the use of ziprasidone and
olanzapine. Logistic regressions, Poisson regressions, and Hurdle models were used to
compare the number of emergency department (ED) visits and hospitalizations associated
with each drug.

15
Results
We identified 7,138 (46.93%) ziprasidone episodes and 8,072 (53.07 %) olanzapine
episodes, and found that patients using ziprasidone were significantly younger (41.50 vs.
45.38 years) and were significantly less likely to be male (29.81 vs. 44.21 %). Regression
analysis showed no significant differences in total costs between the two drugs. However,
ziprasidone was associated with significantly higher medication costs (US$232, p<0.01)
and outpatient costs (US$501, p<0.05), yet lower ED costs (-US$73, p<0.05).
Ziprasidone was also associated with fewer ED visits (0.266, p<0.001) and
hospitalizations (1.117, p<0.001).
Conclusions
Ziprasidone is associated with higher medication costs and outpatient costs than
olanzapine; however it reduces patients’ use of ED and inpatient services.
Key words: ziprasidone, olanzapine, costs, utilization, antipsychotics.
16
2.1 Introduction
Schizophrenia and bipolar disorders are chronic mental illnesses that affect
approximately 1 and 4.5 % of the US population, respectively (1, 2). In 2002, the
economic burden of schizophrenia in the USA was estimated to be US$62.7 billion, 36 %
of which were direct medical costs (3, 4). Bipolar disorder, which is considered one of
the most expensive behavioral healthcare conditions (5), had direct total medical costs of
US$30.7 billion and indirect total costs of US$120.3 billion in the US in 2009 (6). The
first-line treatment for schizophrenia is atypical antipsychotics (7), which have also
become an integral part of the treatment of bipolar disorder (8, 9). Olanzapine is one of
the earliest atypical antipsychotics, and has one of the largest shares of the antipsychotic
market worldwide (10, 11). Ziprasidone is a relatively newer atypical antipsychotic agent,
but has seen rapid increases in use in recent years (12). In general, the two drugs are
comparable in reducing psychotic symptoms (13); however ziprasidone is associated with
lower risks of some adverse effects such as weight gain and diabetes mellitus when
compared with olanzapine (12). Despite its increasing popularity and superior safety
profile to olanzapine, there are few studies that have looked at the healthcare use and
costs associated with ziprasidone. Two previous studies compared the total costs
associated with ziprasidone and olanzapine, and both found no significant difference in
total costs between the two drugs (14, 15). The first study used data from a clinical trial
setting (14), and therefore only included data from patients that met the strict
inclusion/exclusion criteria. The second study was a simulation (15) designed to predict
17
costs associated with atypical antipsychotics, but did not directly evaluate comparative
costs. A third study used claims data and analyzed 1-year post-index average total costs,
medication costs, emergency department (ED) costs, and hospitalization costs across four
atypical antipsychotics (16). The results of this study showed that on average patients
using ziprasidone had lower medication costs, higher ED costs and hospitalization costs,
and similar total costs compared with patients using olanzapine. However, this study had
a relatively small sample size of 1,516. To address gaps in the current literature, we use
multivariate analytic techniques on a large sample of claims data, to compare the costs
and use associated with ziprasidone in relation to olanzapine. Using a retrospective study
design with a large sample size, the aim of the current study was to provide real-world
evidence on the costs and use associated with the two drugs.

2.2 Methods
2.2.1 Data Sources and Sample Selection
Data in this study were obtained from IMS LifeLink™ Health Plan Claims Database,
which comparises medical and pharmaceutical insurance claims for the period of January
2007 to December 2010 from 98 health plans across the USA (17). The database contains
information on inpatient and outpatient diagnoses (in ICD-9-CM format) and procedures
(in Current Procedural Terminology, 4
th
Edition [CPT-4], and Healthcare Common
Procedures Coding System [HCPCS] formats). The data set also includes retail and
mail-order prescription fill information, including the National Drug Code, days’ supply,
18
amounts paid by insurers and beneficiaries, as well as dates of services for all claims.
Demographic data (age, sex, and geographic areas) and enrollment information are also
available. The study received approval from the University of Southern California
Institutional Review Board.
Patients were eligible for the study if they were continuously enrolled beginning 6
months prior to the first observed antipsychotic date through the end of the analysis
period. The date of each patient’s first observed antipsychotic medication treatment
attempt (first observed antipsychotic date), and the initiation date of each patient’s last
treatment attempt with ziprasidone or olanzapine (last observed episode date) were
identified. The date of each patient’s first observed prescription of any antipsychotic was
used instead of the date of first ziprasidone or olanzapine use because this allows a
wash-out period of at least 6 months and maintains information on any other
antipsychotic use the patient may have. In a subset analysis we restricted patient
eligibility to include only treatment-naïve patients, that is, patients whose first observed
antipsychotic date is also their first observed episode date, and found no significant
differences in our results (Supplementary Table 2.1). As such, we proceed with the rest of
our analysis using the first observed antipsychotic date. Patients included in the study
were between the ages of 18 and 100 years, and had to meet the following inclusion
criteria: (1) a diagnosis of schizophrenia (ICD-9-CM codes: 295.xx) or bipolar disorder
(ICD-9-CM codes: 296.4--296.8); (2) at least one prescription for ziprasidone or
olanzapine. Our period of analysis has three temporal components (Supplementary Figure
19
2.1): 6 months prior to the episode initiation date (pre-episode period), 1 month
immediately following the episode initiation date (initiation month) and up to 12 months
after the end of the initiation month (follow-up period). An episode was defined as each
time a patient initiated use of ziprasidone or olanzapine without a gap longer than 15 days
(18, 19). As such, a patient can have one or multiple episodes. If a patient initiated both
medications on the same day, the patient would have two episodes with the same
initiation date, whereby each of the two episodes were designated as polytherapy
episodes. The duration of an episode had to be at least 30 days, to ensure sufficient
exposure (16). To comply with previous studies in literature (18, 20), the combined
period of the pre-episode and initiation month was defined as the baseline period.
Information in the baseline period was used to identify baseline patient characteristics.
2.2.2 Outcomes Measures
We used a retrospective intent-to-treat approach in the current study. The primary
outcome measures in this study were costs in the 12-month post-index period. Total costs,
medical services costs, medication costs, ED costs, outpatient costs, and hospitalization
costs associated with the use of ziprasidone and olanzapine were analyzed. The sum of
all-cause medical costs was defined as total costs. Costs were based on total amount paid
by insurers and patients, and were adjusted to 2007 US$ using the Consumer Price Index
for All Urban Consumers (21). The number of ED visits and the number of
hospitalizations in post-index period were also analyzed.
20

2.2.3 Statistical Analyses
Summary statistics of baseline characteristics included age, sex, healthcare use, type
of schizophrenia (ICD-9-CM codes: 295.0x: simple; 295.1x: disorganized; 295.2x:
catatonic; 295.3x: paranoid; 295.4x: schizophreniform disorder; 295.5x: latent
schizophrenia; 295.6x: residual; 295.7x: schizoaffective disorder; 295.8x: other specified;
295.9x: unspecified), type of bipolar disorder (ICD-9-CM codes: 296.4x: manic; 296.5x:
depression; 296.6x: mixed; 296.7x: unspecified; 296.8x: other), drug use, type of episode
(monotherapy vs. polytherapy), and episode duration. Differences in baseline
characteristics and outcomes associated with the two groups were evaluated using t tests
and χ

tests (or Fisher’s exact tests for categories with at least a number smaller than 10
in any cell). Multivariate linear regressions were used to investigate adjusted effects on
costs. Age, sex, type of episode (monotherapy vs. polytherapy), and total costs in e
baseline period were used as control variables. We also include a vector of utilization and
disease diagnosis indicator variables as additional controls (Table 2.1). These control
variables reflect comorbid conditions often associated with schizophrenia and bipolar
disorder, and have been used in previous studies to control for potential confounding (16,
18). Number of ED visits and number of hospitalizations were modeled using both
Poisson regressions and Hurdle models. Poisson regressions were used to evaluate the
total effects of ziprasidone use on event counts (22). Hurdle models for count data were
conducted to take into account that a proportion of the patients did not have events of
21
interest in the post-index period. In Hurdle models, analyses were carried out in two
processes (22, 23). First, logistic regressions were used to model the probabilities of
having ED visits or hospitalizations. Truncated Poisson regressions were then used to
model positive counts (22). For both logistic and truncated Poisson regressions in Hurdle
models, marginal effects (partial effects) in each of the steps were reported. The
significance level in all analyses was set at 0.05. Analyses were carried out using SAS
(version 9.2; SAS Institute, Cary, NC, USA) and Stata (version 12; Stata Corp, College
Station, TX, USA) statistical software.

2.2.4 Sensitivity Analyses
We carried out three sets of sensitivity analyses of costs. In the first set, general
linear models (GLMs) defined by a logarithmic link function with a gamma distribution
were used for total costs, medical services costs, and medication costs. This specification
is popular in modeling healthcare costs with skewed distributions (23-27). In the second
set, two-part models (TPMs) with logistic regressions in the first steps and GLMs in the
second steps were used for analyses of outpatient costs, ED costs, and hospitalization
costs to account for excessive zero costs (22, 28). TPMs are similar to hurdle models,
only that the truncated Poisson models were replaced with GLMs with a log-link function
and a gamma distribution. The probability of having any cost and the expected cost
conditioning on having any cost were multiplied to estimate the expected cost. The
difference between expected costs of ziprasidone use and olanzapine use for each patient
22
was calculated for the various types of costs. The average of the sample differences in
each type of cost, or sample average incremental effect, was reported (22, 23, 29). The
third set of sensitivity analyses focused on sample inclusion instead of statistical
technique. Specifically, we relax the assumption that episodes should be longer than 30
days, and therefore include all episodes irrespective of length. We perform similar
sensitivity analyses for event count outcomes.

2.3 Results
2.3.1 Baseline Characteristics
A total of 15,210 episodes, representing 9,097 patients (481 patients used both drugs),
met the inclusion criteria. Of the total, 7,138 (46.93 %) of the episodes representing 4,665
patients were ziprasidone episodes, and 8,072 (53.07 %) episodes representing 4,913
patients were olanzapine episodes (Figure 2.1). Descriptive statistics and tests of baseline
characteristics are summarized in Table 2.1. The majority of ziprasidone (98.87 %) and
olanzapine (99.05 %) episodes were monotherapy episodes. Patients using ziprasidone
were significantly younger (41.50 vs. 45.38 years; p<0.001), and were significantly less
likely to be male (29.81 vs. 44.21 %; p<0.001). In the baseline period, patients using
ziprasidone had significantly lower total costs (US$16,930 vs. US$18,150; p<0.001),
medical services costs (US$11,479 vs. US$13,041; p<0.01), ED costs (US$726 vs.
US$803; p<0.05), hospitalization costs (US$5,066 vs. US$6,713; p<0.001), as well as
number of hospitalizations (6.29 vs. 7.76; p<0.001). However, these patients had higher
23
medication costs (US$5,434 vs. US$5,109; p<0.001) and outpatient costs (US$11,864 vs.
US$11,437; p<0.001) than those using olanzapine. In addition, significantly higher
proportions of patients using ziprasidone than olanzapine used following medical
resources: quetiapine (20.76 vs. 13.14 %; p<0.001), risperidone (8.62 vs. 7.12 %;
p<0.001), other antipsychotics (30.76 vs. 24.13 %; p<0.001), antidiabetics (9.88 vs.
7.40 %; p<0.001), anti-obesity drugs (0.28 vs. 0.06 %; p<0.001), ED attendance (41.64
vs. 39.46 %; p<0.01), glucose test (25.90 vs. 22.58 %; p<0.001). Moreover, significantly
higher proportions of patients using ziprasidone than olanzapine had following diagnoses
in baseline period: depression type bipolar disorder (21.34 vs. 13.85 %; p<0.001), mixed
bipolar disorder (17.86 vs. 13.79 %; p<0.001) and other types of bipolar disorder (37.41
vs. 31.67 %; p<0.001). In contrast, significantly lower proportions of patients using
ziprasidone had the following utilization or diagnoses: medications for substance abuse
(9.46 vs. 11.12 %; p<0.001), disorganized schizophrenia (0.13 vs. 0.28 %; p<0.05),
paranoid schizophrenia (3.52 vs. 4.51 %; p<0.001), residual schizophrenia (0.42 vs.
0.68 %; p<0.05), other specified types of schizophrenia (0.36 vs. 0.66 %; p<0.01) and
unspecified types of bipolar disorder (11.67 vs. 12.25 %; p<0.001). Notably, ziprasidone
episodes and olanzapine episodes had similar average durations (85.4 vs. 84.0 days,
p=0.4244).

24
2.3.2 Outcomes
Table 2.2 presents unadjusted differences in outcomes between ziprasidone and
olanzapine episodes. Ziprasidone episodes had significantly lower average: total costs
(US$16,930 vs. US$18,150; p<0.05), medical services costs (US$11,479 vs. US$13,041;
p<0.01), ED costs (US$$726 vs. US$803; p<0.05), and hospitalization costs (US$5,066
vs. US$6,713; p<0.001) in the post-index period. Additionally, the number of
hospitalizations associated with ziprasidone use was significantly lower (7.29 vs. 9.58;
p<0.001). However, ziprasidone episodes were associated with significantly higher
medication costs (US$5,433 vs. US$5,109; p<0.001). The two types of episodes did not
differ significantly in the average post-index outpatient costs or number of ED visits.
Table 2.3 lists the results of baseline regressions and sensitivity analyses. The results
from our ordinary least squares (OLS) regression models suggest that ziprasidone use did
not have significant impact on post-index total costs (US$265, p>0.05) or medical
services costs (US$33, p>0.05) in relation to olanzapine. However, it was significantly
associated with higher medication costs (US$232, p<0.01) and outpatient costs (US$501,
p<0.05), yet lower ED costs (-US$73, p<0.05) and hospitalization costs (-US$236,
p>0.05) when compared with olanzipine. GLM results also suggest that ziprasidone use
did not have any significant effects on total costs (US$160, p>0.05) or medical services
costs (-US$95, p>0.05) when compared with olanzapine, but ziprasidone use
significantly increased medication costs (US$244, p<0.01). The GLM estimate of effect
on medical services costs has a different direction from the OLS estimate; however, both
25
estimates are insignificant. The TPM estimates of effects on outpatient cost, ED costs,
and hospitalization costs are: $482 (p<0.01), -$61 (p<0.01) and -$356 (p<0.05). The TPM
estimate of the cost-saving effect on hospitalization costs is significant, which is different
from the OLS estimate. When all episodes regardless of duration were included in OLS
regressions, results are similar to baseline OLS regressions, except for the slight
difference that the reduced ED costs associated with ziprasidone use was -US$62 and
insignificant (-US$73 in OLS and -US$61 in TPM). Modified Park tests results suggest it
was appropriate to use gamma distribution in GLMs because all the coefficients in the
tests are close to 2 (Supplementary Table 2.2).
Table 2.4 displays adjusted effects of ziprasidone use on the number of post-index
ED attendances and hospitalizations in relation to olanzapine use. Using ziprasidone
insignificantly increased the probability of having ED visits by 0.9 %. However, it
significantly decreased number of visits by 0.266 (p<0.001) among those who had ED
attendances in the post-index period. Overall, it reduced the number of ED visits by 0.076
(p<0.001). Similarly, using ziprasidone insignificantly increased the probability of having
hospitalizations by 0.6%, but it significantly decreased the number of hospitalizations by
4.780 (p<0.001) among those who had hospitalizations in post-index period. Overall, it
decreased the number of hospitalizations by 1.117 (p<0.001). The results produced by
regressions with expanded samples were not significantly different from those in the
baseline models.

26
2.4 Discussion
Ziprasidone and olanzapine are both first-line second-generation antipsychotics used
for the treatment of schizophrenia and bipolar disorder. To our knowledge, the current
study is the first analysis of large-sample claims data to comprehensively assess multiple
dimensions of healthcare costs and use during a 12-month follow-up period associated
with the two medications. Previous studies comparing costs associated with ziprasidone
looked only at total costs (30) and had relatively small sample sizes (16). In our analyses,
the total healthcare costs were not significantly different between ziprasidone users and
olanzapine users, nor were medical services costs or hospitalization costs. Using
ziprasidone increased medication costs and outpatient costs, but reduced ED costs. Our
findings on total healthcare costs are consistent with findings from several previous
studies (14, 30) which did not investigate other aspects of costs. Moreover, the results of
the current study on total costs and hospitalization costs are consistent with findings by
Zhu et. al (16); however ED costs and medication costs contradict the findings in the
study by Zhu et al. (16) in which the investigators reported average 1-year post-index
costs instead of adjusted effects of regressions. The simulation study by Edwards et al.
(15) showed that using olanzapine saved US$900 when compared with ziprasidone,;
however it did not report significance test statistics, making it difficult to compare results
across studies. Our study demonstrates that although using ziprasidone does not impact
the probabilities of having ED visits or hospitalizations after adjusting for patient
characteristics, it does reduce the number of ED visits or hospitalizations among those
27
who have such events. Overall, using ziprasidone is associated with fewer ED visits and
hospitalizations. This result may be a reflection of the recent findings that ziprasidone
decreases risks of adverse effects compared with olanzapine among those who are
vulnerable (31). Further research is needed to examine detailed comparative safety
profiles of ziprasidone and other atypical antipsychotics.
The current study used large sample-size claims data to compare the economic
impact of using ziprasidone in relation to olanzapine. Claims data are well suited for the
analysis of costs and use. We used a variety of statistical techniques to examine the
impacts of interest, and our results are robust across the various techniques. We
conducted a set of sensitivity analyses by relaxing the exposure window used in the
inclusion criteria to further demonstrate the robustness of our results. Using relatively
complex statistical techniques further confirmed the results from baseline analyses. There
are, nevertheless, limitations to the current analysis caused by issues inherent in medical
claims data. Although a comprehensive list of variables was included in regressions to
control for confounding effects, potential unobserved bias may still exist. Such bias may
arise from lack of information on the severity of illness and health behaviors. Future
research that uses relatively thorough data (e.g., electronic medical records plus claims)
should be explored.
Despite the limitations, the current findings demonstrate cost patterns associated with
ziprasidone and olanzapine use. Evidence from the current study indicates that mild
increases in medication costs or outpatient costs associated with ziprasidone use are
28
partially offset by modest decreases in ED costs, resulting in no significant difference in
total costs. Study results on the number of ED visits and hospitalizations further highlight
that the two atypical antipsychotics differ in safety profiles.
2.5 Conclusion
The total annual medical costs associated with ziprasidone and olanzapine for the
treatment of schizophrenia or bipolar disorder are similar. Ziprasidone is not associated
with lower rates of ED visits and hospitalizations in the post-index period; however it is
associated with fewer ED visits and hospitalizations overall. To the extent that
ziprasidone and olanzapine are clinical substitutes, the findings in the current study can
be used to inform physicians when costs associated with the drugs are at question.
Furthermore, the results from this study provide estimates of a variety of costs associated
with ziprasidone and olanzapine, which can be used in future economic studies pertaining
to the two drugs.
Conflict of Interest
All authors have no conflicts of interest to declare. This research received no grant
from any funding agency in the public, commercial, or not-for-profit sectors.
29
Section 2 References
1. Colombo GL, Caruggi M, Di Matteo S, et al. An economic evaluation of aripiprazole
vs olanzapine adapted to the Italian setting using outcomes of metabolic syndrome and
risk for diabetes in patients with schizophrenia. Neuropsychiatric disease and treatment.
2008; 4: 967.
2. Merikangas KR, Akiskal HS, Angst J, et al. Lifetime and 12-month prevalence of
bipolar spectrum disorder in the National Comorbidity Survey replication. Arch Gen
Psychiatry. 2007; 64: 543-52.
3. Wu EQ, Birnbaum HG, Shi L, et al. The economic burden of schizophrenia in the
United States in 2002. J Clin Psychiatry. 2005; 66: 1122-9.
4. McEvoy JP. The costs of schizophrenia. J Clin Psychiatry. 2006; 68: 4-7.
5. Peele PB, Xu Y , Kupfer DJ. Insurance expenditures on bipolar disorder: clinical and
parity implications. AJ Psychiatry. 2003; 160: 1286-90.
6. Dilsaver SC. An estimate of the minimum economic burden of bipolar I and II
disorders in the United States: 2009. J Affect Disord. 2011; 129: 79-83.
7. Mortimer AM. Atypical Antipsychotics as First-Line Treatments for Schizophrenia.
Dis Manag Health Out. 2004; 12: 169-79.
8. Chengappa KR, Suppes T, Berk M. Treatment of bipolar mania with atypical
antipsychotics. 2004.
9. Keck Jr PE. The role of second-generation antipsychotic monotherapy in the rapid
control of acute bipolar mania. J Clin Psychiatry. 2004; 66: 5-11.
30
10. Leonhauser M. Antipsychotics: multiple indications help drive growth. PM360. 2012:
221-23.
11. Gallini A, Donohue JM, Huskamp HA. Diffusion of Antipsychotics in the US and
French Markets, 1998–2008. Psychiatr Serv. 2013; 64: 680-87.
12. Valenstein M, McCarthy JF, Ignacio RV , et al. Patient-and facility-level factors
associated with diffusion of a new antipsychotic in the V A health system. Psychiatr Serv.
2006; 57: 70-76.
13. Swartz M, Stroup TS, McEvoy J, et al. Special Section on Implications of CATIE:
What CATIE Found: Results From the Schizophrenia Trial. Psychiatr Serv. 2008; 59:
500-06.
14. Rosenheck RA. Cost-effectiveness of atypical antipsychotics in the CATIE
schizophrenia trial. Neuropsychopharmacology (New York, NY). 2005; 30: S32-S32.
15. Edwards NC, Pesa J, Meletiche DM, et al. One-year clinical and economic
consequences of oral atypical antipsychotics in the treatment of schizophrenia*. Current
Medical Research and Opinion®. 2008; 24: 3341-55.
16. Zhu B, Kulkarni PM, Stensland MD, et al. Medication patterns and costs associated
with olanzapine and other atypical antipsychotics in the treatment of bipolar disorder*.
Current Medical Research and Opinion®. 2007; 23: 2805-14.
17. IMS. LifeLink™ Health Plan Claims Data User Guide & Data Dictionary. IMS
Health Incorporated, 2009.
18. Chen L, McCombs JS, Park J. The Impact of Atypical Antipsychotic Medications on
31
the Use of Health Care by Patients with Schizophrenia. Value Health. 2008; 11: 34-43.
19. Weiden PJ. Partial Compliance and Risk of Rehospitalization Among California
Medicaid Patients With Schizophrenia. Psychiatric services (Washington, DC). 2004; 55:
886-91.
20. Qiu Y , Fu AZ, Liu GG, et al. Healthcare costs of atypical antipsychotic use for
patients with bipolar disorder in a Medicaid programme. Applied health economics and
health policy. 2010; 8: 167-77.
21. Consumer Price Index - All Urban Consumers. Washington, DC: U.S. Bureau of
Labor Statistics.
22. Greene WH. Econometric Analysis. 7 ed. Upper Saddle River, NJ, USA: Pearson
Education, Inc., 2012.
23. Tarride J-E, Haq M, Taylor VH, et al. Health status, hospitalizations, day procedures,
and physician costs associated with body mass index (BMI) levels in Ontario, Canada.
ClinicoEconomics and outcomes research: CEOR. 2012; 4: 21.
24. Blough DK, Madden CW, Hornbrook MC. Modeling risk using generalized linear
models. J Health Econ. 1999; 18: 153-71.
25. Manning WG, Mullahy J. Estimating log models: to transform or not to transform? J
Health Econ. 2001; 20: 461-94.
26. Manning WG, Basu A, Mullahy J. Generalized modeling approaches to risk
adjustment of skewed outcomes data. J Health Econ. 2005; 24: 465-88.
27. Garrido MM, Deb P, Burgess JF, Jr., et al. Choosing models for health care cost
32
analyses: issues of nonlinearity and endogeneity. Health Serv Res. 2012; 47: 2377-97.
28. Deb P, Manning W, Norton E. Modeling Health Care Costs and Counts. 8th World
Congress on Health Economics. Toronto, Canada, 2011.
29. Press S. Stata Base Reference Manual, Release 10. College Station, TX: Stata Press,
2007.
30. Jing Y , Kim E, You M, et al. Healthcare costs associated with treatment of bipolar
disorder using a mood stabilizer plus adjunctive aripiprazole, quetiapine, risperidone,
olanzapine or ziprasidone. Journal of Medical Economics. 2009; 12: 104-13.
31. Grootens K, Van Veelen N, Peuskens J, et al. Ziprasidone vs olanzapine in
recent-onset schizophrenia and schizoaffective disorder: results of an 8-week
double-blind randomized controlled trial. Schizophr Bull. 2011; 37: 352-61.
33
Table 2. 1 Patient characteristics at baseline.
Baseline
characteristics
Ziprasidone
Episodes
(N=7,138)
Olanzapine
episodes
(N=8,072)
All episodes
(N=15,210)
p-value
Age (years) 41.50 (13.89) 45.38 (16.07) 43.56 (15.21) <0.0001
Male [N, (%)] 2,128 (29.81) 3,569 (44.21) 5,697 (37.46) <0.0001
Monotherapy [N,
(%)]
7,057 (98.87) 7,995 (99.05) 15,052
(98,96)
0.272
Episode duration
(days)
85.4 (112.4) 84.0 (107.2) 77.6 (106.3) 0.4244
Costs in baseline
period
a
(US$)
Total costs 16,930 (28792) 18,150 (33400) 11,778
(23,407)
<0.0001
Total costs IQR 9971 10598 10299
Medical services
costs
11,497
(27,082)
13,041
(32,011)
8,785
(22,345)
0.0014
Medical services
costs IQR
8130 8733 8448
Medication costs 5,434 (6,631) 5,109 (5,427) 2,993 (4,066) 0.0003
Medication costs
IQR
2782 2655 2701
ED costs 726 (2,110) 803 (2,388) 551 (1,593) 0.0382
ED costs IQR 438 423 430
Outpatient costs 11,864
(14,249)
11,437
(16,943)
6,854
(12,412)
<0.0001
Outpatient costs
IQR
5980 6027 6040
Hospitalization
costs
5,066 (20,464) 6,713 (25,478) 4,924
(17,694)
<0.0001
Hospitalization
costs IQR
2688 3177 2945
Number of ED
attendances in
baseline period
0.97 (1.96) 0.96 (2.16) 0.97 (2.07) 0.7751
Number of
hospitalizations in
baseline period
6.29 (16.67) 7.76 (21.91) 7.07 (19.64) <0.0001
Utilization of other
antipsychotics in
baseline period [N,
34
(%)]
Clozapine 36 (0.50) 30 (0.37) 66 (0.43) 0.2141
Quetiapine 1,482 (20.76) 1,061 (13.14) 2,543 (16.72) <0.0001
Risperidone 615 (8.62) 575 (7.12) 1,190 (7.82) 0.0006
Other
antipsychotics
2,196 (30.76) 1,948 (24.13) 4,144 (27.25) <0.0001
Utilization of select
healthcare resources
in baseline period
[N, (%)]

Antidiabetics 705 (9.88) 597 (7.40) 1,302 (8.56) <0.0001
Anti-obesity drugs 20 (0.28) 5 (0.06) 25 (0.16) 0.0010
c

Medications for
substance abuse
675 (9.46) 898 (11.12) 1,573 (10.34) 0.0007
Emergency room
attendance
2,972 (41.64) 3,185 (39.46) 6,157 (40.48) 0.0063
Hospitalization 1,991 (27.89) 2,304 (28.54) 4,295 (28.24) 0.3740
Glucose test 1,849 (25.90) 1,823 (22.58) 3,672 (24.14) <0.0001
Diagnosis of
schizophrenia in
baseline period
b
[N,
(%)]

Simple 24 (0.34) 29 (0.36) 53 (0.35) 0.8098
Disorganized 9 (0.13) 23 (0.28) 32 (0.21) 0.0342
c

Catatonic 14 (0.20) 14 (0.17) 28 (0.18) 0.8503
Paranoid 251 (3.52) 364 (4.51) 615 (4.04) 0.0003
Schizophreniform
disorder
46 (0.64) 54 (0.67) 100 (0.66) 0.8517
Latent 19 (0.27) 20 (0.25) 39 (0.26) 0.8731
Residual 30 (0.42) 55 (0.68) 85 (0.56) 0.0311
Schizoaffective
disorder
585 (8.20) 697 (8.63) 1,282 (8.43) 0.3305
Other specified
types
26 (0.36) 53 (0.66) 79 (0.52) 0.0038
Unspecified 220 (3.08) 374 (4.63) 594 (3.91) <0.0001
Diagnosis of bipolar
disorder in baseline
period
b
[N, (%)]

Manic 918 (12.86) 1,063 (13.17) 1,981 (13.02) 0.5730
Depression 1,523 (21.34) 1,118 (13.85) 2,641 (17.36) <0.0001
Mixed 1,275 (17.86) 1,113 (13.79) 2,388 (15.70) <0.0001
35
Unspecified 833 (11.67) 989 (12.25) 1,822 (11.98) 0.2697
Other 2,670 (37.41) 2,556 (31.67) 5,226 (34.36) <0.0001
Values are expressed as mean (SD) unless specified otherwise. Abbreviations: SD,
standard deviation; ED, emergency department; N, number of episodes; IQR,
interquartile range.
a: The baseline period combines the six months prior to the initiation of current episode
and the month immediately following the initiation.
b: Schizophrenia diagnosis types: 295.0x: simple; 295.1x: disorganized; 295.2x:
catatonic;
295.3x: paranoid; 295.4x: schizophreniform disorder; 295.5x: latent schizophrenia;
295.6x: residual; 295.7x: schizoaffective disorder; 295.8x: other specified; 295.9x:
unspecified. Bipolar disorder diagnosis types: 296.4x: manic; 296.5x: depression;
296.6x: mixed; 296.7x: unspecified; 296.8x: other. Patient may have more than one
diagnosis in the baseline period.
c: Fisher’s exact tests were used in place of χ

test for these specific variables to account
for small number of events.
36
Table 2. 2 Unadjusted differences in outcomes between ziprasidone episodes and
olanzapine episodes
Post-index
utilizations
Ziprasidone
episodes
Olanzapine
episodes
All episodes p-value
Total costs
(US$)
16,930
(28,792)
18,150
(33,400)
17,506
(31,207)
0.0165
Total costs IQR 14519 15097 14841
Medical
services costs
(US$)
11,497
(27,082)
13,041
(32,011)
12,371
(29,732)
0.0014
Medical
services costs
IQR
10335 10756 10510
Medication
costs
5,433 (5,630) 5,109 (5,427) 5,134 (5,423) 0.0003
Medication
costs IQR
5771 5468 5587
ED costs (US$) 726 (2,110) 803 (2,388) 762 (2,236) 0.0382
ED costs IQR 594 587 590
Outpatient
costs (US$)
11,864
(14,249)
11,437
(16,943)
11,520
(15,406)
0.0949
Outpatient
costs IQR
10470 10460 10560
Hospitalization
costs (US$)
5,066 (20,464) 6,713 (25,478) 5,985 (23,473) <0.0001
Hospitalization
costs IQR
1153 1925 1599
Number of ED
attendances
1.25 (2.64) 1.28 (2.90) 1.27 (2.78) 0.5171
Number of
hospitalizations
7.29 (23.22) 9.58 (30.61) 8.53 (27.20) <0.0001
Values are expressed as mean (SD) unless specified otherwise. Abbreviations: SD,
standard deviation; IQR, interquartile range; ED, emergency department.
37
Table 2. 3 Adjusted effects of ziprasidone use on costs in relation to olanzapine use
Costs (US$) OLS GLM TPM OLS with all
episodes
a,b

Total costs 265
(-659, 1189)
160
(-538, 858)
-65
(-941, 812)
Medical
services costs
33
(-853, 920)
-95
(-757, 567)
-306
(-1147, 536)
Medication
costs
232**
(57, 407)
244**
(86, 403)
241**
(78, 404)
Outpatient
costs
a

501*
(29, 974)
482**
(144, 820)
496*
(56, 936)
ED costs
a
-73*
(-145, 0)
-61**
(-115, -6)
-62
(-130, 5)
Hospitalization
costs
a

-236
(-961, 489)
-356*
(-895, -184)
-560
(-1255, 134)
Values are expressed as estimates (95% CI) unless otherwise specified. Abbreviations: CI,
confidence interval; OLS ordinary least squares; GLM, general linear model; TPM,
two-part model.
*: p<0.05, **: p<0.01, ***: p<0.001
a: The number of episodes which met the inclusion criteria of this study is 15210. When
episodes with duration shorter than 30 days were included, there are 16873 episodes. The
numbers of episodes with non-zero costs in TPMs for outpatient costs, ED costs and
hospitalization costs are respectively 14886, 6538 and 3998.
b: Without restriction of episode duration >= 30 days.
38
Table 2. 4 Effects of ziprasidone use on number of emergency department attendances and hospitalizations in post-index period in
relation to olanzapine use evaluated with hurdle models and Poisson models
Logistic regression Truncated Poisson regression Poisson regression
Events parameter partial effect parameter partial effect parameter total effect
N=6,687
a

Number of ED attendances
a
0.371
(-0.032, 0.106)
0.009
(-0.007, 0.024)
-0.102***
(-0.136, -0.069)
-0.266***
(-0.353, -0.179)
-0.060***
(-.090, -0.031)
-0.076***
(-0.114, -0.039)
N=3,998
Number of hospitalizations
b
0.032
(-0.047, 0.111)
0.006
(-0.008, 0.020)
-0.148***
(-0.159, -0.136)
-4.780***
(-5.160, -4.400)
-0.131***
(-0.143, -0.120)
-1.117***
(-1.217, -1.017)
Sensitivity analyses using all episodes (without restriction of episode duration >= 30 days)
N=7,470
Number of ED attendances
c
0.041
(-0.024, 0.107)
0.009
(-0.006, 0.024)
-0.107***
(-0.138, -0.075)
-0.278***
(-0.361, -0.196)
-0.064***
(-0.092, -0.036)
-0.082***
(-0.117, -0.046)
N=4,491
Number of hospitalizations
d
0.021
(-0.053, 0.06)
0.004
(-0.010, 0.017)
-0.170***
(-0.181, -0.159)
-5.440***
(-5.797, -5.083)
-0.160***
(-0.171, -0.149)
-1.366***
(-1.461, -1.271)
Values are expressed as estimates (95% CI). Abbreviations: CI, confidence interval; ED, emergency department.
*: p<0.05, **: p<0.01, ***: p<0.001
a: The sample size is 6687. 149 episodes (0.98% of all eligible episodes) had ED visits in post-index period but did not have ED costs.
b: The sample size is 3998.
c: The sample size is 7470.
d: The sample size is 4491.
39
Figure 2. 1 Sample selection process. A total of 15210 episodes representing 9097
patients were selected. 481 patients used both drugs
40
Section 2 Supplementary Material
Supplementary Table 2. 1 OLS results additionally include interaction between
ziprasidone use and schizophrenia, and OLS results with only first episode of treatment
naïve patients who did not have any poly-therapy episodes (restricted sample).
Costs Including interaction Using restricted sample
OLS Poisson
Regression
OLS Poisson
Regression
Total costs (US$) 331
(-655, 1317)
373
(-1481,
2228)
Medical services
costs (US$)
107
(-834, 1054)
-41
(-2626,
1194)
Medication costs
(US$)
224*
(37, 410)
414**
(128, 701)
Outpatient costs
(US$)
563*
(59, 1067)
611
(-191, 1413)
ED costs (US$) -57
(-134, 20)
-44
(-162, 74)
Hospitalization
costs (US$)
-232
(-1006, 542)
-237
(-1601,
1126)
Number of ED
attendances
- -0.062***
(-0.102, -0.022)
-0.035
(-0.099,
0.029)
Number of
hospitalizations
-1.300***
(-1.409, -1.192)
-0.87
(-1.04, -0.70)
Values are expressed as estimates (95% confidence interval). The interaction term was
not significant in any of the regressions and therefore not shown.
The number of observations in the restricted sample is 4967. Abbreviations: OLS,
ordinary least squares; ED, emergency department.
*: p<0.05, **: p<0.01, ***: p<0.001
41
Supplementary Table 2. 2 Modified Park Test for GLM regressions of total costs, medical
services costs, and drug costs
Total
costs
Medical
services
costs
Medication
costs
Coefficient 1.95 1.83 1.94

tests for distribution
and the corresponding
p-values
χ

tests p-value χ

tests p-value χ

tests p-value
Gamma 0.20 0.6515 12.23 0.0005 1.87 0.1717
Poisson 68.10 0.0000 290.54 0.0000 500.49. 0.0000
Inverse Gaussion or wald 83.82 0.0000 577.86 0.0000 630.28 0.0000
Gaussian 287.50 0.0000 1412.81 0.0000 2126.14 0.0000
These results suggest gamma distribution should be used because coefficients are close to
2. Χ

tests indicate that using gamma distribution produced good fits for total costs and
medication costs. The fit using gamma distribution for medical services costs is better
than any other distribution. Abbreviation: GLM, general linear model.
42
Supplementary Figure 2. 1 Diagram of Analysis Period. Initiation month is the 30 days
immediately following the episode initiation date.

Follow-up period is the period in
which costs and utilization are analyzed, and can be up to 12 months in length. The time
period that encompasses pre-episode, initiation month, and follow-up period is defined as
the “analysis period”.
43
3. Improving Model Specifications When Estimating Treatment Effects across
Alternative Medical Interventions
Abstract
Objective
The purpose of this paper is to critique the list of independent variables commonly
used in observational research and test the impact of variables for prior use and treatment
history on estimates of treatment effects.
Methods
Using data from the California Medicaid program, this study generated a series of
OLS estimates of the effect of atypical antipsychotic medications on costs and duration of
therapy to illustrate the impact of alternative model specifications on treatment effects.
The first sequence of estimates consisted of six model specifications, the last of which
included variables reflecting the type of episode defined according to prior treatment
history and compliance. The second sequences repeats the specification of the first 6
models but were carried out separately by episode type to examine the heterogeneity of
treatment effect. The second sequence of models documented the impact of additional
drug history variables. .
Results
Estimates of the impact of atypical antipsychotic use on total costs and duration on
initial drug were statistically significant in the first 6 models. Estimates changed
significantly when dummy variables indicating prior use of inpatient service and nursing
44
home care were included in the model specification. Estimated effects changed
substantially when prior total cost was included in cost analysis, or when prior treatment
duration was included in duration analysis. Significant variation also existed in estimated
effects across episode types, and it was particularly pronounced before controlling for
prior cost/duration.
Conclusion
It is important to add prior measures of the outcome variable to control for
unobserved bias in retrospective studies. Also, the accuracy and utility of results to
clinicians can be improved significantly if analyses are performed by episode type.
Keywords: treatment, effect, intervention, episode, history
45
3.1 Introduction
Clinicians and policy makers require medical evidence with which to effectively
integrate new technologies into real-world practice. This need is especially acute when
new treatment alternatives are introduced into competition with older, well established
treatments. In the case of new medications, these data come from two sources: the
clinical trials required for FDA and other registry approvals and observational studies that
establish the “essential need” for a new treatment alternative (1). Both sources of medical
evidence are necessary to estimate the cost-effectiveness of a new technology at product
launch.
Efforts to document the essential need for a new technology actually begin very early
in product development. Product innovators assess how well older, competing therapies
are meeting the therapeutic needs of patients treated in the real world. Therapeutic gaps
with older treatments typically arise when patient adherence to current therapies is
sub-optimal or treatment efficacy for compliant patients is limited. Other sources of
essential need are high treatment costs or high indirect cost to the patient and their
caregivers (2). These indirect costs may include the costs of side effects, caregiver time,
reductions in the quality of life and the like. Essential need data are used in a series of
‘go/no go’ decisions that are made as the product is developed and tested.
If the evolving data on essential need are promising and/or the new product is
efficacious, the new technology will move through the required registry trials testing
safety and efficacy. These studies use experimental research designs [RCTs] which
maximize the internal validity through random assignment and other techniques [e.g.,
blinding] (1). However, the generalizability [external validity] of results from randomized
clinical trials is limited:
46
1. RCTs are limited to a small, homogeneous study population due to cost and
patient safety concerns. Data on treatment outcomes for high risk patients may be
missing or, conversely, it may be ethical to only include very high risk patients
who have no remaining treatment options, as in cancer trials.
2. Patient outcomes are measured over a limited time, again due to cost and to
patient burden and risk of dropout. This mis-match between study duration and
time to potential treatment effect is most acute for drug therapies intended to
manage chronic disease such as hypertension or hyperlipidemia.
3. By design, RCTs cannot measure patient adherence to treatment under real-world
conditions. RCTs employ significant effort and resources to insuring patient
adherence to the study protocol.
4. Finally, FDA-registration RCTs may require only placebo-controlled trials or the
list of active comparators may be constrained due to cost concerns.
Conversely, essential need studies based on retrospective data can provide CE evidence
for the full range of treatment alternatives, reflect real-world clinical practice and
real-world adherence. The patients included in an essential need study also include risk
groups not studied in the RCT environment. Finally, retrospective observational studies
can provide evidence on long term outcomes and the [rare] clinical risk associated with
existing therapies.
Drug companies combine data from real-world essential need studies and registry
RCT into an initial computer-based CE model to support product marketing at launch.
These models project the impact of the new technology in clinical use. However, the
accuracy of the initial CE models is limited by the gaps in research on real-world
47
adherence for the new drug, long term patient outcomes using the new drug and outcomes
achieved by patient sub-groups not included in the product’s clinical trials [poor external
validity]. While retrospective essential need studies fill in some of these gaps, the
statistical validity of observational studies can be questionable if not executed well. Of
equal concern, physicians, P&T committees and government program administrators may
not fully understand the complexity and pitfalls of the statistical methods use in
observational research.
The purpose of this paper is to critique the statistical methods commonly used in
observational research by presenting a sequence of analyses which document how
statistical results can change significantly as more care is taken to maximize the use of
available data. Specifically, we will present a sequence of models moving from simple
models to models using explanatory variables that are rarely derived from available
claims data. The paper also documents the impact of alternative estimation strategies.
3.2 Statistical Challenges in Observational Research
Satisfactory internal validity can only be achieved in observational research by
controlling for confounding factors associated with both treatment selection and patient
outcomes. For example, it is challenging to measure the impact of a new medication
relative to competing older drugs if the new medication is reserved for high risk patients,
or if the new medication is used initially to treat patients who failed therapy using the
older alternatives (3). This bias can be reduced using multivariate statistical methods that
adjust statistically for the impact of observable factors on treatment outcomes. However,
treatment selection bias will continue to exist if important factors are missing from the
multivariate statistical models. In the econometrics literature, this is referred to as missing
48
variable bias. In comparative effectiveness research, missing variable bias is referred to as
unobserved treatment selection bias [UTSB].
UTSB is often a function of the data available for analysis. For example, data from a
health insurance program or government program [e.g., Medicare] includes the paid
claim for common laboratory tests but provides no information concerning the laboratory
result itself. Fortunately, the growing availability of electronic medical records [EMR]
data will provide increasing opportunities for reducing the impact of UTSB in
observational studies in medicine.
The first line of defense against UTSB is to use the available data to document all
factors that may impact both treatment selection and patient outcomes. Researchers often
ignore episodes of drug therapy initiated following the first observed treatment episode
which is concerning since patient outcomes can be radically different for the second or
third treatment attempt using the same drug, or for episodes of switching therapies,
episodes of augmentation therapy or episodes involving combination therapy. Moreover,
the later episodes contain more information about the treatment history of the patients,
such as prior compliance behavior, which could significantly impact patient outcomes.
This expanded use of available pharmacy data may be particularly important when newly
approved medications are significantly less likely to be used as first therapy in treatment
naïve patient.
Alternative model specifications make better use of available data and will also be
investigated. Both difference-in- difference models (DD) (4) and fixed effects models (FE)
(4) assume that UTSB is invariant across time periods (e.g. pre-treatment and
post-treatment). For example, genetic factors which affect disease severity or response to
49
drug treatment are invariant across time, and they are usually unobservable to researchers.
Diff-in-diff models are popular in analysis of panel data. By differencing out fixed effects
or controlling for them using dummy variables representing clusters, these models
eliminate the effects of the time-invariant UTSB. But the time-invariant assumption of
UTSB does not necessarily hold in practice. Even though time-invariant effects can be
removed using such techniques, potential bias caused by time-varying confounding
factors is still left unresolved. For instance, some clinical symptoms and health behavior
are not captured by automated data systems, yet they are unlikely to remain exactly the
same across time periods.
3.3 Methods
3.3.1 Data Sources
This study conducts a series of retrospective analyses of the impact of atypical
antipsychotic medications to illustrate the impact of alternative model specifications and
estimation methods on treatment effects. The study uses an existing California Medicaid
(Medi-Cal) data set which was derived for a string of earlier studies (5, 6) from paid
claims data from the fee-for-service portion of Medi-Cal. The data cover the period of
1994-2003 during which Medi-Cal revoked its restriction on the use of typical
antipsychotics to patients who had failed at least two previous treatment attempts using
typical antipsychotics. This formulary restriction was lifted in October 1997, three years
after the introduction of risperidone in 1994 and exactly one year after the approval of
olanzapine in 1996. Quetiapine was approved by the FDA in October 1997 and was
immediately available to Medi-Cal patients without restrictions. This formulary
expansion resulted in an immediate increase in the diffusion of atypical antipsychotics
50
which are now accepted as first line drug therapy for these patients (7).
Initial inclusion criteria required that patients have a paid claim with a recorded
diagnosis of schizophrenia (ICD-9 code=295.xx) or bipolar disorder (ICD-9 codes =
296.4-296.8) and with at least one prescription for an antipsychotic medication.
Additional exclusion criteria were applied once all episodes of care were identified.
3.3.2 Definition of the Unit of Analysis
The “standard of practice” for the unit of analysis in a retrospective CE research
design data mirrors the RCT design: The episode of treatment. In the case of
observational studies, the data of randomization is replaced by an “index date” defined
based on the patient’s first prescription of one of the study drugs. Like most RCTs, the
patient is typically subjected to a “wash-out” period by requiring that the patient has not
filled a prescription of any study drug prior to their initial prescription. Wash-out periods
vary in length and 6 months to a year are common. Most studies then limit their analysis
to these “first episodes” and ignore any subsequent use of related drugs such as
augmentation therapy or the switching to an alternative medication. Limiting the analysis
to first episodes excludes a large majority of treatment episodes. Moreover, new
medications are seldom used as the first drug of choice and are regulated to treating
“treatment failures” or providing augmentation therapy.
The data set used here includes all episodes of psychotropic drug therapy initiated by
patients. An episode of treatment was defined each time a patient started a drug treatment
using an antipsychotic, antidepressant or mood stabilizer not used previously or restarted
an earlier drug treatment after a gap that was at least 15 days. The 15-day gap was
defined in collaboration with the Medi-Cal program and was to comply with earlier
finding by Weiden et al (8), who reported that the risk of hospitalization increased
51
substantially after breaks in therapy as short as 10 days.
The follow up period was the 12 months after the month of initiation. The 12-month
follow up period was specified for the measurement of treatment outcomes which mimics
intent to treat methods implemented in clinical trials. Patient episodes were then screened
for eligibility during the entire pre- and post-treatment period. The amount paid for all
services were inflation adjusted to 2004 using service specific rates of fee inflation from
the Medi-Cal program.
Many patients had more than one treatment episode, which is very common in
schizophrenia and bipolar disorders as patients switched from one antipsychotic to
another or start and stop therapy. While this approach violates the usual assumption of
independence across units of analysis, excluding subsequent episodes initiated by the
patient was judged to generate stronger bias than hypothetical independence of sampling
units (6, 9, 10). Excluding these follow-on episodes severely restricts the utility of the
analysis to clinicians who required data on treatment effects for a wide range of treatment
histories.
3.3.3 Covariates and Model Sequencing
The focus of the proposed study is to examine how the use of an expanded list of
unconventional independent variables impacts estimates of total costs and duration of
therapy using standard ordinary least squares (OLS) regressions. Specifically, the
following sequence of models will be estimated:
Model 1: The basic models include only age [categories with an interval of 10],
gender, county population density [urban/rural/urban-rural-mix] and Medi-Cal aid
categories
52
Model 2: The second set of models adds dichotomous variables based on
non-mental health comorbidities based on ICD-9 diagnoses at baseline.
Model 3: Mental health diagnoses were added to the model specification separately
to test the impact of diagnostic mix data related directly to the disease state under study.
Model 4: The list of independent variables was extended to include two
dichotomous variables indicating whether or not the patient used inpatient hospital
services or nursing home services in the 6 months prior to the episode start date.
Model 5: Pre-treatment measures of the outcome variables [total costs, duration of
therapy] were added in this model. This specification is mathematically equivalent to
difference-in-difference modeling which re-defines the outcome variable by differencing
the value of the outcome measure before and after treatment.
Models 6: This model is the first to used data on the drug history of the patient at the
time of treatment. The initial drug history covariates are dichotomous variable for episode
type. Five types of episodes were defined in this data set:
1. First Observed Episode: The “first” episode was defined based on the patient’s
first psychotropic drug therapy attempt.
2. Restart Episodes: A restart episode was defined if the patient was not on active
psychotropic drug therapy for 15 days or longer and initiated therapy with the
medication used in their most recent episode [intermittent use].
3. Switching Episodes: A switching episode was defined if a patient changed
medication while still on active therapy or within 15 days of terminating a
previous therapy, and discontinued use of all previous medications within 60
days.
53
4. Delayed Switching Episodes: A delayed switching episode was defined if a
patient changed drug therapy after a break in therapy in excess of 15 days
5. Augmentation Episodes: An augmentation episode was defined when a patient
added a second medication while continuing to purchase one or more of their
previous medications beyond 60 days.
This analysis excludes first observed treatment episodes due to the lack of data on patient
treatment history. The following analyses only used restart, delayed switching, switching,
and augmentation episodes. In order to facilitate comparisons to Models 1-6, first
episodes were also excluded from the sample of episodes included in these models.
Models 7-12: The remaining drug treatment history variables are entered sequentially
in Models 7-12: count of the number of prior treatment attempts, monotherapy vs
combination therapy, days off therapy (for restart and delayed switching episodes), and
prior use of related drugs [typical and atypical antipsychotics, mood stabilizers,
antidepressants, depot-formulated drugs]. At this point, the analyses are conducted by
episode type primarily because episode type is a significant predictor of cost and duration
of therapy [Model 6]. It follows that clinicians will require information on the CE of
atypical vs. typical antipsychotics by episode type.
3.4 Results
Results for the first six models for the impact of using atypical antipsychotics that
used all episodes are summarized in Table 3.1. The outcome variables used in these
models are total cost over the first post-treatment year and duration of therapy on the
‘initial’ drug of the episode. For example, in the case of augmentation episode, the initial
drug is the augmenting drug. In addition to the impacts of atypical use, we also include
54
the estimates of the effects of episode type indicators on cost and duration in Model 6
which are also included in Table 3.1.
Estimates of the impact of atypical antipsychotic use on total costs and duration on
initial drug are statistically significant in the first 6 models. In models 1-3, the estimated
impact of using an atypical antipsychotic range from $1230 to $1399, and the estimates
of the impacts on duration range from 90.2 to 95.9 days. Estimates changed significantly
when dummy variables indicating prior inpatient service use and prior nursing home use
were included in the model specification. The effect of atypical use on total cost
decreased to $398 whereas the effect on duration only slightly changed to 89.1 days.
Equally important, the R
2
of the model for total cost increased substantially (0.182 to
0.571).
Difference-in-difference modeling is frequently used in observational research
testing the effect of new treatments or policy changes on patient outcomes. When prior
total cost was included in cost analysis [Model 5], the estimated effect of atypical use
increased from $398 to $615 and the R
2
further increased from 0.571 to 0.710. Similarly,
when prior treatment duration was included in duration analysis, the estimated effect of
atypical use decreased from 89 days to 76 days and the R
2
doubled from 0.063to 0.130.
Model 6 estimates the impact of atypical use controlling for episode type. The results
from this model demonstrate the importance of drug use history when estimating the
impact of atypical antipsychotics on cost and duration of therapy in two ways. First, the
estimated effect of atypical use changed to $751 while the estimated effect on duration
decreased to 55 days. But more importantly, episode type has very significant impacts of
costs and duration. Compared with restart episodes, switching episodes, delayed
55
switching episodes, and augmentation episodes increased total cost by $1221, $1360, and
$2237, respectively. However, the impacts of the episode type on duration were not
uniformly positive. Switching and delayed switching episodes lasted an additional 137
days, 74 days relative to re-start episodes. Conversely, the use of the initial drug
decreased by 76 days in augmentation episodes relative to re-start episodes, possibly
reflecting intended short term use of augmentation therapy.
The results from Model 6 provide an estimate of the average impact of using an
atypical antipsychotic on cost and duration of therapy controlling for how atypical
antipsychotic drugs are used by episode type. However, clinicians need to know how
these new drugs perform by episode type, not on average. This dictates that these
analyses be conducted separately by episode type. Conducting analyses by episode type
also allows researchers to add other treatment history variables to the analyses which can
vary by episode type. Our analyses of use and cost by episode type are displayed in
Tables 3.2-3.5. The results for the average impact of atypical use derived in Model 6
using data for all episode types is also listed in these tables as a reference.
Table 3.2 presents the results using restart episodes starting with the original set of
independent variable used in Model 1. Models 5 and 6 are equivalent when estimated
using only restart episodes. In models 1-3 using restart episodes, the estimated effects of
atypical antipsychotic use on total cost range from $2301 to $2616. Including prior
inpatient services use and prior nursing home use decreased the estimated effect to $1077.
It further decreased to $384 after controlling for prior total cost. The estimated effect
remained stable across models 7-11 ($ 493-567). The R
2
increased significantly at the
stages of model 4 (0.185 to 0.588) and model 5 (0.588 to 0.740). The estimated effects of
56
atypical antipsychotic use on duration is much more stable than estimated for cost across
all models using restart episodes are between 20.3 and 38.2 days. Also, the increase in the
R
2
was modest from model 1 to model 11 (0.018 to 0.062).
Table 3.3 presents the results of analyses using switching episodes. In models 1-3
using switching episodes, the estimated effects of atypical antipsychotic use on total cost
are between $1678 and $1901. Including prior inpatient services use and prior nursing
home use in the model decreased the estimated effect to $1289. Including prior total cost
changed the estimated effect to $1171. The estimated effects on total costs are between
$1122 and $1270 in models 7-11. The R
2
increased by a large amount at the stages of
model 4 (0.220 to 0.571) and model 5 (0.571 to 0.667). The estimated effects of atypical
antipsychotic use on duration in models 1-4 are between 106.1 and 108.0 days. But the
estimated effect of atypical use dropped significantly to 24.6 days after controlling for
prior treatment duration. In models 7-11, the estimated effects are between 25.4 days and
37.2 days. The R
2
increased from 0.071 to 0.453 at the stage of model 5.
Table 3.4 lists the results of analyses using delayed switching episodes. Throughout
the 10 models, the estimated effects of atypical antipsychotic use on total cost range from
$1120 to $1487, and the estimated effects on duration range from 84.0 to 95.9 days. The
R
2
in the cost analysis increased from 0.236 to 0.586 at the stage of model 4 and
increased from 0.586 to 0.667 at the stage of model 5. However, the R
2
in duration
analysis only increased modestly from 0.031 in model 1 to 0.077 in model 11.
Finally, the results of analyses using augmentation episodes are included in Table 3.5.
In models 1-3, the estimated effects of atypical antipsychotic use on total cost are
between -$4936 and -$4390. The estimated effect changed to -$1741 after controlling for
57
prior inpatient services use and prior nursing home use, and further changed to -$289
after controlling for prior total costs. In models 7-11, the estimated effects on total costs
are between -$156 and $74 and are all statistically insignificant. These are the only set of
insignificant estimates in all analyses in the current study. The estimated effects on
duration are between 142.6 days and 170.0 days for all 10 models. The R
2
in the cost
analysis increased from 0.165 to 0.520 at the stage of model 4 and increased from 0.520
to 0.677 at the stage of model 5. Likewise, the R
2
in duration analysis increased from
0.096 to 0.171 at the stage of model 5.
3.5 Discussion
The purpose of this study was to investigate the changes in estimated treatment
effects in response to including a series of explanatory variables, some of which are
rarely derived from claims databases. The results from this series of estimates are
illustrated in Figure 3.1[Total Cost] and Figure 3.2 [Duration of Therapy]. Two statistical
effects are evident. First, controlling for prior total cost/treatment duration led to
significant changes of estimates in most analyses, and the results for the impact of using
atypical antipsychotics [treatment effect] tended to settle down across model
specifications after that stage. This result validates the value of adding prior measures of
the outcome variable which corresponds to the popular difference-in-difference
estimation technique. Second, it is evident that great variation exists in estimated effects
of atypical antipsychotic use across episode types which persists across model
specification and is particularly pronounced before adding prior total cost/treatment
duration to the model specification. But as an added bonus, conducting the analysis of
treatment effects by episode type significantly increases the utility of study results to
58
clinicians who are looking for guidance as to what works best for patients with different
treatment history.
Episode type can significantly impact the estimated treatment effects because episode
type has a major impact on treatment outcomes. Accordingly, comparative effectiveness
research should take into account the differential treatment effects in episode-type
subgroups.
A major limitation of observational result to measure treatment effects stems from
the nature of claims databases. Claims databases do not usually capture important
information such as disease severity and clinical symptoms. Although we controlled for a
long list of variables and used various model specifications in the regressions, potential
bias due to unmeasured covariates could not be ruled out thoroughly. However, the future
of observational research in comparative effectiveness research is bright as data from
electronic medical record [EMR] systems become more available. The internal validity of
estimated differences between alternative treatments will only improve as better clinically
relevant data are available.
Section 3 References
1. Faries DE, Leon AC, Haro JM, et al. Analysis of observational health care data using
SAS. SAS Institute, 2010.
2. Berger ML, Dreyer N, Anderson F, et al. Prospective observational studies to assess
comparative effectiveness: the ISPOR good research practices task force report. Value
Health. 2012; 15: 217-30.
3. Soumerai SB, Zhang F, Ross-Degnan D, et al. Use of atypical antipsychotic drugs for
schizophrenia in Maine Medicaid following a policy change. Health affairs (Millwood,
59
Va). 2008; 27: w185-95.
4. Greene WH. Econometric Analysis. 7 ed. Upper Saddle River, NJ, USA: Pearson
Education, Inc., 2012.
5. Chen L, McCombs JS, Park J. Duration of Antipsychotic Drug Therapy in
Real-World Practice: A Comparison with CATIE Trial Results. Value Health. 2008; 11:
487-96.
6. Chen L, McCombs JS, Park J. The Impact of Atypical Antipsychotic Medications on
the Use of Health Care by Patients with Schizophrenia. Value Health. 2008; 11: 34-43.
7. Narayan S, Sterling KL, McCombs JS. The impact of open access to atypical
antipsychotics on treatment costs for Medi-Cal patients with bipolar disorder. Dis Manag
Health Out. 2006; 14: 287-301.
8. Weiden PJ. Partial Compliance and Risk of Rehospitalization Among California
Medicaid Patients With Schizophrenia. Psychiatric services (Washington, DC). 2004; 55:
886-91.
9. Gianfrancesco FD, Grogg AL, Mahmoud RA, et al. Differential effects of risperidone,
olanzapine, clozapine, and conventional antipsychotics on type 2 diabetes: findings from
a large health plan database. J Clin Psychiatry. 2002; 63: 920-30.
10. Gianfrancesco F, Pesa J, Wang R-H, et al. Assessment of antipsychotic-related risk of
diabetes mellitus in a Medicaid psychosis population: sensitivity to study design. Am J
Health Syst Pharm. 2006; 63: 431-41.
60
Table 3. 1 Impact of atypical antipsychotic use on total cost and duration of therapy: all
episode types (N=731,236)
Total Cost First
Post-Treatment Year
Duration of Drug
Therapy
Model Model Specification OLS
(SE)
R-squared
In OLS
OLS
(SE)
R-squared
In OLS
1 Demographic variables
only
1350
***
(63.7)
0.095 95.9
***
(0.8)
0.037
2 Add: Medical
Diagnostic Mix
1399***
(62.0)
0.160 93.0***
(0.8)
0.046
3 Add: Mental Health
Diagnostic Mix
1230***
(61.6)
0.182 90.2***
(0.8)
0.060
4 Add: Prior Use of Hospital
and Nursing Home Care
398***
(44.6)
0.571 89.1***
(0.8)
0.063
5 Add: Pre-Treatment
Costs and Duration of
Therapy
615***
(36.7)
0.710 76.4***
(0.8)
0.130
6 Add: Episode Type 751***
(37.8)
0.712 55.0***
(0.8)
0.157
Estimated Impact of Episode Type
on Total Cost and Duration of Drug
Therapy in Model 6
(Restart as baseline)
Switching 1221***
(65.3)
136.9***
(1.4)
Delayed switching 1360***
(55.3)
73.8***
(1.2)
Augmentation 2237***
(58.3)
-75.9***
(1.2)
OLS results are presented as estimate (SE). Abbreviations: N, number of episodes; OLS,
ordinary least squares; SE, standard error.
61
Table 3. 2 Impact of atypical antipsychotic use on total cost and duration of therapy:
restart episodes (N=445,258)
Total Cost First
Post-Treatment Year
Duration of Drug
Therapy
Model Model Specification OLS
(SE)
R-squared
In OLS
OLS
(SE)
R-squared
In OLS
1 Demographic variables 2301***
(77.3)
0.098 38.2***
(0.9)
0.018
2 Demo + Medical Diagnosis 2616***
(75.6)
0.166 33.0***
(0.9)
0.026
3 Demo+Medical Dx+MH Dx 2485***
(75.7)
0.185 28.1***
(0.9)
0.039
4 +prior
hospitalization
+prior long term care
1077***
(53.9)
0.588 26.2***
(0.9)
0.043
5 +prior and switch total
costs/prior episode duration
384***
(42.8)
0.740 24.4***
(0.9)
0.056
6 Model 6 Specification Using
Data for All Episodes
751***
(37.8)
0.712 55.0***
(0.8)
0.157
7 Add: prior episode count 500***
(43.3)
0.741 22.2***
(0.9)
0.060
8 Add: mono/poly 493***
(43.4)
0.741 21.7***
(0.9)
0.060
9 Add: prior depot use indicator 497***
(43.5)
0.741 21.7***
(0.9)
0.060
10 Add: time off Rx 567***
(44.2)
0.741 20.3***
(0.9)
0.060
11 Add: prior Rx mix 563***
(111.3)
0.741 24.2***
(2.3)
0.062
OLS results are presented as estimate (SE). Abbreviations: N, number of episodes; OLS,
ordinary least squares; SE, standard error.
62
Table 3. 3 Impact of atypical antipsychotic use on total cost and duration of therapy:
switching episodes (N=71,917)
Total Cost First
Post-Treatment Year
Duration of Drug
Therapy
Model Model Specification OLS
(SE)
R-squared
In OLS
OLS
(SE)
R-squared
In OLS
1 Demographic variables 1678***
(217.4)
0.147 107.0***
(4.0)
0.037
2 Demo + Medical Diagnosis 1796***
(211)
0.203 106.1***
(4.0)
0.053
3 Demo+Medical Dx+MH Dx 1901***
(208.8)
0.220 108.0***
(3.9)
0.068
4 +prior hospitalization
+prior long term care
1289***
(154.9)
0.571 106.6***
(3.9)
0.071
5 +prior and switch total
costs/prior episode duration
1171***
(136.4)
0.667 24.6***
(3.1)
0.453
6 Model 6 Specification Using
Data for All Episodes
751***
(37.8)
0.712 55.0***
(0.8)
0.157
7 Add: prior episode count 1128***
(145.4)
0.670 26.3***
(3.2)
0.453
8 Add: mono/poly 1122***
(145.6)
0.670 25.4***
(3.2)
0.454
9 Add: prior depot use indicator 1270***
(149.1)
0.670 26.4***
(3.2)
0.454
10 Add: time off Rx
11 Add: prior Rx mix 1262***
(152.2)
0.670 37.2***
(3.2)
0.459
OLS results are presented as estimate (SE). Abbreviations: N, number of episodes; OLS,
ordinary least squares; SE, standard error.
63
Table 3. 4 Impact of atypical antipsychotic use on total cost and duration of therapy:
delayed switching episodes (N=97,704)
Total Cost First
Post-Treatment Year
Duration of Drug
Therapy
Model Model Specification OLS
(SE)
R-squared
In OLS
OLS
(SE)
R-squared
In OLS
1 Demographic variables 1140***
(185.1)
0.161 95.9***
(2.8)
0.032
2 Demo + Medical Diagnosis 1487***
(179.7)
0.219 93.0***
(2.8)
0.045
3 Demo+Medical Dx+MH Dx 1454***
(178.2)
0.236 93.4***
(2.8)
0.061
4 +prior hospitalization
+prior long term care
1128***
(131.2)
0.586 92.2***
(2.8)
0.069
5 +prior and switch total
costs/prior episode duration
1179***
(117.6)
0.667 88.6***
(2.85)
0.075
6 Model 6 Specification Using
Data for All Episodes
751***
(37.8)
0.712 55.0***
(0.8)
0.157
7 Add: prior episode count 1217***
(121.9)
0.671 86.6***
(2.9)
0.076
8 Add: mono/poly 1231***
(123.5)
0.671 84.0***
(2.9)
0.076
9 Add: prior depot use indicator 1287***
(126.7)
0.671 84.2***
(2.9)
0.076
10 Add: time off Rx 1283***
(127)
0.671 84.3***
(2.9)
0.076
11 Add: prior Rx mix 1120***
(137.1)
0.672 91.7***
(3.1)
0.077
OLS results are presented as estimate (SE). Abbreviations: N, number of episodes; OLS,
ordinary least squares; SE, standard error.
64
Table 3. 5 Impact of atypical antipsychotic use on total cost and duration of therapy:
augmentation episodes (N=116,357)
Total Cost First
Post-Treatment Year
Duration of Drug
Therapy
Mode
l
Model Specification OLS
(SE)
R-squared
In OLS
OLS
(SE)
R-squared
In OLS
1 Demographic variables -4936***
(185.1)
0.105 170.0***
(2.1)
0.075
2 Demo + Medical Diagnosis -4549***
(180.8)
0.152 168.7***
(2.1)
0.086
3 Demo+Medical Dx+MH Dx -4390***
(179.9)
0.165 168.6***
(2.1)
0.095
4 +prior hospitalization
+prior long term care
-1741***
(136.6)
0.520 167.6***
(2.1)
0.096
5 +prior and switch total
costs/prior episode duration
-289**
(112.2)
0.677 146.7***
(2.1)
0.171
6 Model 6 Specification Using
Data for All Episodes
751***
(37.8)
0.712 55.0***
(0.8)
0.157
7 Add: prior episode count 66
(115.4)
0.676 142.6***
(2.1)
0.173
8 Add: mono/poly 65
(115.6)
0.676 143.3***
(2.1)
0.174
9 Add: prior depot use indicator 74
(118.7)
0.676 143.4***
(2.1)
0.174
10 Add: time off Rx
11 Add: prior Rx mix -156
(122.0)
0.677 155.7***
(2.2)
0.179
OLS results are presented as estimate (SE). Abbreviations: N, number of episodes; OLS,
ordinary least squares; SE, standard error.
65
Figure 3. 1 Impact of Using Atypical Antipsychotics on Total Cost in First
Post-Treatment Year.
66
Figure 3. 2 Impact of Using Atypical Antipsychotics on Duration of Therapy.
67
4. Risk of Acute Major Cardiovascular Events and Acute Kidney Injury Associated With
Antipsychotics
4.1 Introduction and Objectives
In the US, the total antipsychotic use per 1,000 inhabitants increased by 78% from
1998-2008 (1). The major driver was the increased use of atypical antipsychotics (1). In
2008, atypical antipsychotics accounted for 86% of antipsychotic use in the US (1),
owing to the lower risk of extrapyramidal side effects (EPS) compared to typical
antipsychotics (2). However, atypical antipsychotics pose significant risks of their own
(3), and with the exception of clozapine, they may be no more effective than typical
antipsychotics (3). In fact, four atypical antipsychotics have been shown to be associated
with similar all-cause time-to-discontinuation as typical antipsychotics (3), and
medication side effects are a major reason of nonadherence (4). Therefore, evidencing the
comparative safety profiles of antipsychotics is critical to prescribing in clinical practice.
Recent studies raise concerns of cardiovascular risk associated with antipsychotics. A
retrospective cohort study in Italy found that users of phenothiazine, and four atypical
antipsychotics (olanzapine, quetiapine, risperidone, and clozapine) had higher risk of
ischemic stroke compared to non-users of antipsychotics (5). A Canadian retrospective
cohort study examined the risk of ischemic stroke associated with olanzapine, risperidone,
quetiapine, and 16 typical antipsychotics, and did not find significantly different risk
across the drugs (6). Using Tennessee Medicaid data, Ray et al. reported higher risk of
sudden cardiac death among users of four atypical antipsychotics (olanzapine, quetiapine,
risperidone and clozapine) and users of two typical antipsychotics (haloperidol and
tioridazine) compared to non-users of antipsychotics in the general population (7). A
68
retrospective study using Danish data found that the risks of acute coronary syndrome,
ischemic stroke, and cardiovascular mortality among young and middle-age adults were
not significantly different for olanzapine, quetiapine, and risperidone (8).
The potential risk of acute kidney injury (AKI) associated with antipsychotic use was
discovered more recently. A study using Ontario Drug Benefit Data compared the risk of
hospitalization with AKI and six related events among elderly (> 66 years) users of
olanzapine, risperidone, and quetiapine to the risks among non-antipsychotic drug users
(9). The six events are hypotension, acute urinary retention, neuroleptic malignant
syndrome/rhabdomyolysis, pneumonia, acute myocardial infarction, and ventricular
arrhythmia. The study concluded that olanzapine, risperidone, and quetiapine were
associated with higher risks of all events except neuroleptic malignant
syndrome/rhabdomyolysis (9). Currently, there are no other studies which examined risks
of AKI and related events associated with antipsychotics.
All the aforementioned studies included only olanzapine, risperidone, quetiapine, and
in some cases clozapine when studying atypical antipsychotics. Since their marketing in
early 2000’s, ziprasidone and aripiprazole have seen increased use over the years (3).
However, evidence on acute cardiovascular risk profile and acute kidney injury risk
profile associated with the two antipsychotics is still unknown. In addition, the risk of
acute kidney injury associated with typical antipsychotics has not been investigated. In
light of the paucity of related evidence, the aims of the current study are to compare the
risks of acute major cardiovascular events and acute kidney injury associated with a
broad range of atypical and typical antipsychotics. Evidence on whether different
antipsychotics carry different risks will help antipsychotic selection and policy making on
69
monitoring of antipsychotic users (10).
4.2 Methods
4.2.1 Data and Samples
Data in this study were obtained from the January 2007 to June 2013 nationwide
medical and pharmacy claims databases from Humana (~14 million covered lives). The
databases contain information on diagnoses (in ICD-9-CM format) and procedures. They
also include prescription fill information, including days of supply, amounts paid by
insurers and beneficiaries, as well as dates of services for all claims. Demographic
information and enrollment records are also available. The enrollees in the managed care
organization comprised both privately insured and Medicare beneficiaries. Initially,
samples were selected if they had at least one diagnosis of schizophrenia or bipolar
disorder, and had at least one paid claim of any of the study antipsychotics.
The claims were summarized into antipsychotic treatment episodes, which are the
unit of observation of this study. An episode of treatment was defined each time a patient
started a drug treatment using an antipsychotic not used previously or restarted an earlier
drug treatment after a gap that was at least 15 days. This start or restart date was defined
as the index date of each treatment episode observation. The 15-day definition is
compliant with earlier studies in literature (11, 12). The study samples were then
restricted to the episodes during which patients: 1) had at least 180 days of pre-index
enrollment and 360 days of post-index enrollment; 2) aged>18 years. Patients with
Medicare Part D plans and Medicare Supplemental plans were also considered as not
having continuous enrollment. Each 180-day pre-index period was used to measure
baseline characteristics. We then excluded asenapine, chlorpromazine, clozapine,
70
iloperidone, loxapine, molindone, paliperidone, perphenazine, pimozide, thioridazine,
thiothixene, and trifluoperazine from subsequent analyses because their sample sizes
were smaller than 5,000 (2.57% of the total sample). These groups had very few events
(typically < 20 in each group), making estimation of the impact of these specific drugs
unreliable. As a result, the antipsychotics which were included in analysis are aripiprazole,
fluphenazine, haloperidol, olanzapine, quetiapine, risperidone, and ziprasidone.
The acute major cardiovascular events analyzed in this study included: 1) acute
coronary syndrome (ACS) or ischemic stroke (IS) (8, 9); and 2) ventricular arrhythmia
(VCA). AKI and known cause of AKI events included: 1) AKI; 2) hypotension; 3) acute
urinary retention; 4) neuroleptic malignant syndrome/rhabdomyolysis; and 5) pneumonia
(9). Hospitalization with the target diagnoses was required to identify clinically
significant events. The criterion of hospitalization allows reliable identification of events.
The ICD-9-CM codes for diagnoses are listed in Appendix Table A4.1.1.
An event was recorded if it occurred while the patient was on drug therapy or within
30 days after the end of therapy. A 30-day period was added because residual effects of
drugs could last for several days up to several weeks (13-18). The outcomes were
measured as time to event. If an event did not occur, then this episode was defined as a
censored observation (19). The episode was censored either at the the 30
th
day after the
discontinuation of therapy or at the end of the 360-day post-index continuous enrollment
period. Censoring due to discontinuation of therapy is a potential source of nonrandom
censoring which can cause bias (20). A sensitivity analysis relaxing the random censoring
assumption of Cox model was carried out and is described in section 4.2.4.6.
71
4.2.2 Statistical Analysis
T-tests and chi-square tests were used to examine baseline characteristics.
Multivariate Cox proportional hazard regressions were used to estimate adjusted relative
risks across drug groups. Compared to other survival models such as accelerated failure
time (AFT) models or parametric proportional hazard models, Cox models do not require
a fully parametric assumption (21, 22) to produce consistent estimates. AFT models can
take a linear form which only need correct specification of the conditional mean to
produce consistent maximum likelihood estimates (22) when there is no censoring.
However, the non-censoring restriction is impractical in survival analysis because it
requires all observations to encounter the event during the follow up period. Hence, AFT
estimates still need the density to be fully correctly specified in order to be consistent
when there is censoring (22). Therefore, estimates from Cox proportional hazard models
are likely more robust than estimates from AFT models because the underlying data
generating process of observed data is usually unknown. In fact, lack of robustness is
considered the major weakness of parametric methods in survival analysis (22). Clustered
standard errors at patient level were used.
A long list of covariates (Appendix Table A4.2.1) were included in all multivariate
analyses to control for treatment selection bias. The covariates covered demographic
information, mental health diagnoses, non-mental-health diagnoses, prior 6-month total
healthcare costs, use of other healthcare services in the prior 6-month period,
antipsychotic treatment history, and dose. Dose was averaged over days of supply and
categorized according to clinically recommended dosing as above target range, in target
range, and below target range (23) so that it is comparable across drugs.
72
4.2.3 Sensitivity Analyses Using Instrumental Variables (IVs)
Most atypical antipsychotics are recommended as first-line therapy for the treatment
of schizophrenia (24). Atypical antipsychotics are also prescribed more frequently than
typical antipsychotics for the treatment of bipolar disorder (25). Consequently, potential
systematic difference which is not observed to investigators could exist across users of
atypical antipsychotics and users of typical antipsychotics, leading to potential omitted
variable bias. Instrumental variable models can be used to correct omitted variable bias
caused by the unobserved difference.
Instrumental variables are variables that correlate with the treatment but do not
directly affect the outcome (21, 22). We used lags of treatment as well as prescriber
specialties as IVs in this study. Because the analytic file is episode-based, an individual
can have one or more observations. Given the multiple-episode structure of our analytic
data file, it is possible to exploit the panel data nature in IV models. Specifically,
Anderson & Hsiao (22) and Arellano & Bond (22) proposed to use lag values (or
differenced lag values) of dependent variables as IVs in first-difference model of dynamic
panel. In estimating the equation
y

y
,
= y
,
y
,
+ x

x
,

+

,
,
Δ y
,
= y
,
y
,
can be used to instrument
y
,
y
,
if y
,
is uncorrelated with

,
. Kawatkar et al. (26)
extended this theory to use first and second lag values of observed treatment as IVs for
the current treatment in estimating costs associated with disease modifying antirheumatic
drugs. The justification for the correlation between the IVs and the current treatment was
that physician treatment decision was made after observing the patient’s experience and
73
outcomes based on the patient’s past treatment choice (26). We used the lagged
treatment approach in the current study. The first and second lag values of antipsychotic
treatment as IVs for contemporary antipsychotic treatment. Although physicians make
decisions based on prior treatment experience of patients, they likely use information on
treatment response and well-known adverse events such as EPS, metabolic side effects,
and prolactin elevation (27). Variation in treatment across time is less likely to be caused
by uncommon and scantly documented events. Therefore, the lag values of treatment
should explain the current treatment, yet not having direct impact on the target events in
the current episode. As a result of using the lag values of treatment as IVs, the first and
the second episodes of each patient were not included in IV-Cox regressions.
Prescriber specialty indicators were also used as IVs in our analysis. Physicians of
different specialties likely have different practice patterns and therefore have different
preferences for treatment (28), yet physician specialties should have minimum direct
correlation with target side effects in the current study. Specialties were categorized into:
1) psychiatry/mental health; 2) family medicine, primary care, and general internal
medicine; and 3) all other specialties.
In order to compare IV model results to results from Cox regressions, we ran non-IV
Cox regressions using the restricted sample used in the IV models (excluding the first and
the second episodes).
Cox models are nonlinear models. Therefore, conventional IV estimators including
two-stage least squares (2SLS) and general method of moments (GMM) cannot be used
here. Instrumental variable-Cox models in this study used two-stage residual inclusion
(2SRI). Terza et al. showed that 2SRI generates consistent estimates in nonlinear models
74
when IVs satisfy the rank condition and the order condition (29). The first stage in the
current 2SRI analyses was a multinomial logit model (30). In the second stage, residuals
from the multinomial logit model were calculated using observed choice and predicted
probability of each choice and were taken into the Cox regressions as additional
regressors. Standard errors of coefficients were obtained with bootstrapping for
2SRI-Cox models. Testing the joint significance of the residuals in the Cox regressions is
equivalent to Hausman tests of endogeneity in the linear setting (30). Currently, there are
no over-identification tests in 2SRI regressions. Therefore, the exogeneity of IVs was
tested indirectly. Specifically, proxy risk factors including pre-index costs, pre-index
Charlson Comorbidity Index, an indicator for pre-index hospitalization, an indicator for
pre-index emergency department visit, and the occurrence of each target event in the prior
6-month period was regressed on the IVs conditional on other covariates to examine the
correlation between the target events and the IVs (Appendix A4.3.2). It was expected that
the lag values of treatment were not statistically significant predictors of prior 6-month
event occurrence.
4.2.4 Other Sensitivity Analyses
4.2.4.1 Extended Pre-index Period
In the base-case analysis, a period of 180 days prior to the treatment index date was
used. Alternatively, a sensitivity analysis was conducted using the episodes of the patients
with at least 360-day pre-index period.
75
4.2.4.2 Attrition Due to Discontinued Enrollment
Attrition bias, also known as sample selection bias, arises when the sample selected
for analysis is systematically different from the sample excluded from the analysis and
the factors affecting the selection also affects the outcomes (31). In some treatment
episodes in our data, patients did not have continuous enrollment over the full pre- and
post-treatment observation period. Patients in the excluded episodes may not have
continuous enrollment in prior period, post period, or both. The baseline characteristics of
those who did not have prior period could not be measured. Therefore, the characteristics
of patients in the excluded episodes who had continuous enrollment in the prior period
but not the post period were compared to the sample which had continuous enrollment
over the entire observation period. If there were substantial differences, then it is
reasonable to think there was potential nonrandom sample attrition.
Nonrandom attrition could be determined by observable variables or unobserved
variables. 2SRI was used to test whether there is attrition contributed by unobserved
variables. The instrumental variable for this step was days of continuous enrollment prior
to the current treatment episode. Previous enrollment should be related to subsequent
enrollment, yet enrollment should not be a determinant of side effects.
Inverse probability weighted (IPW) regression is used if attrition is caused by
selection on observables (32). The consistency of IPW Cox regression estimates have
been proved in several studies (33-36). However, the consistency of IPW estimators in
Cox regressions relies on either knowing the true probability of sample selection or
correctly specifying the probability estimation equation such that the estimated
probability is consistent (34, 36, 37). To that end, we estimated the selection probability
using a logistic regression which is fully parametric and a local nonlinear least squares
76
(LNLLS) regression which is semi-parametric (38). For each of the method,
cross-validation was carried out by splitting the sample into two halves. The
cross-validation procedure was bootstrapped for 200 repeats and the Brier score in each
repeat was calculated. The Brier scores obtained from bootstrapped logistic regressions
and LNLLS regressions were tested against each other. If significant difference existed,
the method with a significantly smaller mean Brier score was chosen to calculate the
probability of selection-into-sample.
4.2.4.3 Weibull Regressions
Although Cox model are more popular than parametric survival models because of
the robustness of Cox models (39), parametric models are sometimes chosen over Cox
models for efficiency (39). In the current study, we conducted sensitivity analyses using
parametric Weibull models (22).
4.2.4.4 Less Restrictive Identification of Events
In the base-case analysis, only target diagnoses which resulted in hospitalization
were analyzed as outcomes. This method is relatively accurate but very restrictive and
may undercount events, resulting in very low event rates. In a set of sensitivity analyses,
target diagnoses with emergency department (ED) visits and urgent care visits were
added to the event counts.
4.2.4.5 Subsamples of episodes without any overlap
In the base-case analysis, all episodes with or without overlap with other episodes
were included. Events happened during overlap could be attributable to one of the drugs
77
or both drugs. It is not possible to determine whether or which one of the drugs incurred
an event during overlap with certainty. Therefore, carrying out a sensitivity analysis using
episodes without overlap provides a potentially “clean” comparison.
4.2.4.6 Nonrandom censoring
Whether censoring due to discontinuation of therapy is random censoring cannot be
tested (20). However, it is possible to take into consideration of possible nonrandom
censoring by using a competing risk model (19). A competing risk is an event that hinders
the observation of the event of interest (40). The competing risk model in the current
study defined censoring due to discontinuation of therapy as a competing event that
prevented the observation of the outcome of interest.
4.3 Results
4.3.1 Descriptive Statistics
Figure 4.1 shows the sample selection process. 172,313 episodes were selected for
the base-case analytic sample. Table 4.1 lists results of baseline characteristics of patients
in each drug group. Patients had statistically significant differences in each of the
characteristics across drug groups. For example, quetiapine users [mean: $10,509 (SD:
20,959)] and olanzapine users [$10,379 (19,943)] had the highest prior total costs. Also,
quetiapine users [0.46 (1.12)] and risperidone users [0.44 (1.06)] had the highest
Charlson Comorbidity Index. In addition, haloperidol users (22.88%) and quetiapine
users (22.38%) had the highest rates of prior all-cause hospitalization, whereas quetiapine
users (35.75%) and risperidone users (34.53%) had the highest rates of prior all-cause ED
visit. Risperidone users had the best persistence [190.9 days (239.5)] whereas
78
fluphenazine users had the worst persistence [145.5 days (197.8)].
4.3.2 Acute Coronary Syndrome (ACS) and Ischemic Stroke (IS)
Table 4.2 shows results from the base-case Cox regression, the restricted-sample Cox
regression, and the 2SRI-IV Cox regression for the impact of alternative antipsychotics
relative to haloperidol on the likelihood of ACS and IS. The base-case Cox model
estimated using all patient episodes found no statistically significant difference across
alternative antipsychotics. The results from the Cox models estimated over the restricted
sample of episodes used in the IV analysis were consistent with the base-case analysis
with the exception of fluphenazine which was estimated to increase the risk of ACS and
IS [HR (hazard ratio): 1.674, 95% CI: 1.017-2.757] relative to haloperidol.
The last column of Table 4.2 presents the results from the instrumental variable
analysis which attempted to control for treatment selection bias. It is interesting to note
that the results for the IV models are uniformly larger than the estimated effects in the
corresponding Cox analysis. Also, aripiprazole (HR: 2.415, 95% CI: 1.214-4.802),
olanzapine (HR: 2.175, 95% CI: 1.076-4.398), quetiapine (HR: 2.138, 95% CI:
1.095-4.175), risperidone (HR: 2.043, 95% CI: 1.081-3.864), and ziprasidone (HR: 2.231,
95% CI: 1.047-4.754) were associated with significantly higher risks of ACS and IS
compared to haloperidol in 2SRI-IV analyses. Fluphenazine had a similar risk level as
atypical antipsychotics, but its increased risk was not statistically significant compared to
haloperidol. Although first-stage residuals were not statistically significant in the second
stage of the 2SRI-IV regression, we could not conclude that endogeneity of drug choice
was not identified in the analyses of ACS and IS based on the substantial difference
between base-case results and IV results. As such, the results of 2SRI-IV regressions are
79
used as primary results. According to this set of results, the antipsychotics investigated in
the current study were associated with significantly different risks of ACS and IS.
4.3.3 Ventricular Arrhythmia
Table 4.3 displays adjusted relative risks of ventricular arrhythmia associated with
the study drugs. The study antipsychotics were not associated with different risks of
ventricular arrhythmia in the base-case Cox regression, the restricted-sample Cox
regression, and the 2SRI-IV Cox regression. Selection bias was not identified using the
results from IV regression because first-stage residuals were not statistically significant in
the second stage of the 2SRI regression
4.3.4 Acute Kidney Injury
Results for AKI analyses are presented in Table 4.4. The base-case results suggest
that olanzapine (HR: 1.344, 95% CI: 1.057-1.708), quetiapine (HR: 1.350, 95% CI:
1.082-1.685), and ziprasidone (HR: 1.338, 95% CI: 1.035-1.729) were associated with
significantly higher risks of AKI compared to haloperidol. In contrast, the results from the
restricted-sample Cox regression indicate the study antipsychotics did not have
significantly different AKI risks. However, the first-stage residuals were jointly
marginally significant in the 2SRI-IV regression of AKI. Consequently, the 2SRI-IV
results of AKI are used as primary results. According to the 2SRI-IV results, aripiprazole
(HR: 1.893, 95% CI: 1.149-3.119), olanzapine (HR: 1.964, 95% CI: 1.182-3.261),
quetiapine (HR: 2.377, 95% CI: 1.472-3.837), and risperidone (HR: 1.759, 95% CI:
1.119-2.765) were associated with significantly higher risks of AKI compared to
haloperidol, whereas other drugs were not associated with significantly different risks of
80
AKI.
4.3.5 Hypotension
Table 4.5 displays adjusted relative risks of hypotension associated with the study
drugs. According to the base-case Cox regression results, quetiapine was associated with
a significantly higher risk of hypotension (HR: 1.293, 95% CI: 1.009-1.655) compared to
haloperidol while other drugs had similar risks. However, both quetiapine (HR: 1.372, 95%
CI: 1.019-1.848) and ziprasidone (HR: 1.563, 95% CI: 1.124-2.173) were associated with
higher hypotension risk than haloperidol in the restricted-sample Cox regression. The
study antipsychotics were not associated with different hypotension risks in the 2SRI-IV
Cox regression, and the residuals were not statistically significant. Hence, the base-case
results are used as primary results.
4.3.6 Acute Urinary Retention
Results for acute urinary retention are shown in Table 4.6. The Base-case Cox
regression, the restricted-sample Cox regression, and the 2SRI-IV Cox regression
unanimously did not identify different risks of acute urinary retention across drugs. Also,
the first-stage residuals were statistically insignificant in the second stage of 2SRI-IV
Cox regression, indicating a lack of endogeneity. Therefore, the study antipsychotics were
not associated with different risks of acute urinary retention.
4.3.7 Neuroleptic Malignant Syndrome/Rhabdomyolysis
Table 4.7 lists results for neuroleptic malignant syndrome/rhabdomyolysis.
Quetiapine was associated with a significantly higher risk of neuroleptic malignant
81
syndrome/rhabdomyolysis (HR: 1.606, 95% CI: 1.043-2.474) compared to haloperidol in
the base-case regressions, and no significantly different risks were found for other drugs.
Estimates from the restricted-sample Cox regression and the 2SRI-IV Cox regression
were not statistically significant, and the first-stage residuals were not significant in the
IV regression.
4.3.8 Pneumonia
Results from analyses of pneumonia are displayed in Table 4.8. According to the
base-case results, olanzapine was associated with a significantly higher risk of pneumonia
(HR: 1.314, 95% CI: 1.058-1.633) in relation to haloperidol, but other estimates in
base-case Cox regressions of pneumonia were not statistically significant. No
endogeneity was identified because the first-stage residuals were statistically insignificant
in the second stage of 2SRI-IV Cox regression. Therefore, the base-case results are used.
4.3.9 Other Analyses
Choosing between atypical antipsychotics and typical antipsychotics (Appendix
A4.4.1) was endogenous in the analysis of AKI, but not in the analyses of other outcomes.
Atypical antipsychotics were associated with a significantly higher risk of AKI (HR:
1.313, 95% CI: 1.083-1.591) compared to typical antipsychotics in base-case non-IV
estimates, but had similar risks of other outcomes as typical antipsychotics. Using IV
estimates, the AKI risk associated with atypical antipsychotics compared to atypical
antipsychotics is even higher (HR: 2.041, 95% CI: 1.339-3.113). This result is consistent
with drug-specific estimates in both base-case and 2SRI-IV analyses of AKI. If atypical
antipsychotic users use typical antipsychotics instead, the number needed to treat to avoid
82
one AKI hospitalization is 166.
The validity of IVs is discussed in Appendix A4.3 with results of tests. Other
sensitivity analyses results are listed in Appendix A4.4.2-A4.4.7. Results in each of the
sensitivity analyses were very similar to results in base-case analyses. In general, the
results were robust to extending the pre-index period, changing the specification of
distribution assumption, adjusting for sample selection, using an alternative identification
strategy of events, deleting episodes with overlap, and using competing risk model.
4.4 Discussion
In this retrospective cohort study of administrative claims databases, we investigated
differential risks of acute cardiovascular events, AKI, and known causes of AKI
associated with two commonly used typical antipsychotics and five commonly used
atypical antipsychotics.
The findings of this study suggest that antipsychotics have different levels of acute
major cardiovascular event risk and different levels of AKI risk. Specifically, all atypical
antipsychotics in this study are associated with higher acute coronary syndrome and
ischemic stroke risks compared to haloperidol. Also, aripiprazole, olanzapine, quetiapine,
and risperidone are associated with higher AKI risks in relation to haloperidol, and other
antipsychotics have similar AKI risks. With few exceptions, the study drugs also have
very similar risks of ventricular arrhythmia and known causes of AKI. Our findings are in
line with previous evidence that typical and atypical antipsychotics are not significantly
different with respect to cardiovascular risk (6) and that atypical antipsychotics have
similar cardiovascular risks with each other (8). However, our results also suggest that the
cardiovascular risk associated with typical antipsychotics is driven by fluphenazine. In
83
addition, the differential AKI risks associated with antipsychotics were not documented
in current literature and should raise concerns in clinical practice. Use of the
antipsychotics which are associated with relatively high risks of AKI should be avoided
among patients who are vulnerable to kidney disease. Also, when patients develop AKI,
use of the high-risk antipsychotics should be considered as a potential cause.
Prior studies evaluating cardiovascular and renal risks of antipsychotics left out some
very commonly used antipsychotics including haloperidol, fluphenazine, aripiprazole,
and ziprasidone. Especially, the only available study on AKI related to antipsychotics
only included olanzapine, risperidone, and quetiapine (9). Hence, whether the
aforementioned commonly used antipsychotics have differential acute cardiovascular
risks and AKI risks was unknown. Kirkham and Seitz (10) raised the questions of
whether AKI risk related to olanzapine, quetiapine, and risperidone is a class effect and
whether the risk level varies across a broad spectrum of antipsychotic drugs. The findings
of the current study fill the evidence gap in literature by partially answering their
questions. Also, this study investigated AKI risk associated with antipsychotics in a
population of adults with a wide range of age compared to the previous study which only
included elderly patients (9). More, the current study included both “new users” and
“prevalent users” (41). Including only “new users” is supposedly associated with reduced
precision and suboptimal generalizability (41, 42). Even more, the results are robust to
alternative analytic strategies with varying inclusion criteria, distribution assumption, and
definition of events.
The results of the current study should be interpreted with several limitations. The
first limitation is innate to the retrospective cohort nature of the study. Although a long
84

list of variables were controlled for in the regressions and the IV models were used to
identify potential endogeneity, omitted variable bias cannot be ruled out thoroughly. In
addition, our sample did not include Medicaid population. Accordingly, the
generalizability could be limited because Medicaid is the major payer for schizophrenia
patients in the United States. Also, the event rates of several side effects were very low,
and the statistical power of our analyses for such events might not be sufficient.

4.5 Conclusions
Antipsychotics are associated with differential acute major cardiovascular event risks
and acute kidney injury risks. Especially, several atypical antipsychotics are associated
with higher cardiovascular risks and AKI risks than haloperidol. Further research should
be conducted to determine whether the differential risks should be considered in clinical
practice. Also, the question as to whether atypical antipsychotics are safer than typical
antipsychotics needs to be comprehensively re-evaluated.

Section 4 References
1. Gallini A, Donohue JM, Huskamp HA. Diffusion of Antipsychotics in the US and
French Markets, 1998–2008. Psychiatr Serv. 2013; 64: 680-87.
2. Huskamp HA, O'Malley AJ, Horvitz-Lennon M, et al. How Quickly Do Physicians
Adopt New Drugs? The Case of Second-Generation Antipsychotics. Psychiatr Serv. 2013;
64: 324-30.
3. Donohue J, O'Malley AJ, Horvitz-Lennon M, et al. Changes in physician
antipsychotic prescribing preferences, 2002-2007. Psychiatr Serv. 2014; 65: 315-22.
4. Gonzalez JM, Thompson PM, Moore TA. Review of the safety, efficacy, and side
85
effect profile of asenapine in the treatment of bipolar 1 disorder. Patient preference and
adherence. 2011; 5: 333.
5. Sacchetti E, Trifirò G, Caputi A, et al. Risk of stroke with typical and atypical
anti-psychotics: a retrospective cohort study including unexposed subjects. J
Psychopharmacol (Oxf). 2008; 22: 39-46.
6. Gill SS, Rochon PA, Herrmann N, et al. Atypical antipsychotic drugs and risk of
ischaemic stroke: population based retrospective cohort study. BMJ. 2005; 330: 445.
7. Ray WA, Chung CP, Murray KT, et al. Atypical antipsychotic drugs and the risk of
sudden cardiac death. N Engl J Med. 2009; 360: 225-35.
8. Pasternak B, Svanström H, Ranthe M, et al. Atypical Antipsychotics Olanzapine,
Quetiapine, and Risperidone and Risk of Acute Major Cardiovascular Events in Young
and Middle-Aged Adults: A Nationwide Register-Based Cohort Study in Denmark. CNS
Drugs. 2014; 28: 963-73.
9. Hwang YJ, Dixon SN, Reiss JP, et al. Atypical Antipsychotic Drugs and the Risk for
Acute Kidney Injury and Other Adverse Outcomes in Older AdultsA Population-Based
Cohort StudyAtypical Antipsychotic Drugs and the Risk for Acute Kidney Injury. Ann
Intern Med. 2014; 161: 242-48.
10. Kirkham J, Seitz D. New use of antipsychotics associated with a small increased risk
of acute kidney injury in older adults with mental disorders. Evidence Based Mental
Health. 2015; 18: 55.
11. Weiden PJ. Partial Compliance and Risk of Rehospitalization Among California
Medicaid Patients With Schizophrenia. Psychiatric services (Washington, DC). 2004; 55:
886-91.
86
12. Chen L, McCombs JS, Park J. The Impact of Atypical Antipsychotic Medications on
the Use of Health Care by Patients with Schizophrenia. Value Health. 2008; 11: 34-43.
13. Hasnain M, Vieweg WVR. QTc interval prolongation and torsade de pointes
associated with second-generation antipsychotics and antidepressants: a comprehensive
review. CNS Drugs. 2014; 28: 887-920.
14. Gugger JJ. Antipsychotic Pharmacotherapy and Orthostatic Hypotension. CNS Drugs.
2011; 25: 659-71.
15. Chiang C-L, Liu C-C. Micturition difficulty associated with aripiprazole: report of 2
cases. J Clin Psychopharmacol. 2011; 31: 128-29.
16. Parekh R, Fattah Z, Sahota D, et al. Clozapine induced tubulointerstitial nephritis in a
patient with paranoid schizophrenia. BMJ case reports. 2014; 2014: bcr2013203502.
17. Belvederi Murri M, Guaglianone A, Bugliani M, et al. Second-generation
antipsychotics and neuroleptic malignant syndrome: systematic review and case report
analysis. Drugs R D. 2015; 15: 45-62.
18. Nosè M, Recla E, Trifirò G, et al. Antipsychotic drug exposure and risk of
pneumonia: a systematic review and meta‐analysis of observational studies.
Pharmacoepidemiology and drug safety. 2015.
19. Satagopan J, Ben-Porat L, Berwick M, et al. A note on competing risks in survival
data analysis. Br J Cancer. 2004; 91: 1229-35.
20. Jackson D, White IR, Seaman S, et al. Relaxing the independent censoring
assumption in the Cox proportional hazards model using multiple imputation. Stat Med.
2014; 33: 4681-94.
21. Greene WH. Econometric Analysis. 7 ed. Upper Saddle River, NJ, USA: Pearson
87
Education, Inc., 2012.
22. Cameron AC, Trivedi PK. Microeconometrics: Methods and Applications. New York,
NY , USA: Cambridge University Press, 2005.
23. Gardner DM, Murphy AL, O'Donnell H, et al. International consensus study of
antipsychotic dosing. AJ Psychiatry. 2010; 167: 686-93.
24. Mortimer AM. Atypical Antipsychotics as First-Line Treatments for Schizophrenia.
Dis Manag Health Out. 2004; 12: 169-79.
25. Gentile S. Atypical antipsychotics for the treatment of bipolar disorder: more
shadows than lights. CNS Drugs. 2007; 21: 367-87.
26. Kawatkar AA, Hay JW, Stohl W, et al. Incremental Expenditure of Biologic Disease
Modifying Antirheumatic Treatment Using Instrumental Variables in Panel Data. Health
Econ. 2013; 22: 807-23.
27. Drugs for psychiatric disorders. Treatment guidelines from the Medical Letter. 2013;
11: 53.
28. Brookhart MA, Rassen JA, Schneeweiss S. Instrumental variable methods in
comparative safety and effectiveness research. Pharmacoepidemiology and drug safety.
2010; 19: 537-54.
29. Terza JV , Basu A, Rathouz PJ. Two-stage residual inclusion estimation: Addressing
endogeneity in health econometric modeling. J Health Econ. 2008; 27: 531-43.
30. Geraci A, Fabbri D, Monfardini C. Testing exogeneity of multinomial regressors in
count data models: does two stage residual inclusion work? , Quaderni - Working Paper
DSE N° 921, 2014.
31. Miller RB, Wright DW. Detecting and correcting attrition bias in longitudinal family
88
research. Journal of Marriage and the Family. 1995: 921-29.
32. Wooldridge JM. Inverse probability weighted M-estimators for sample selection,
attrition, and stratification. Portuguese Economic Journal. 2002; 1: 117-39.
33. Sasieni P. Maximum weighted partial likelihood estimators for the Cox model.
Journal of the American Statistical Association. 1993; 88: 144-52.
34. Lin D. On fitting Cox's proportional hazards models to survey data. Biometrika. 2000;
87: 37-47.
35. Robins J, Rotnitzky A. Inverse probability weighted estimation in survival analysis.
Encyclopedia of Biostatistics. 2005: 2619-25.
36. Subramanian S, Bandyopadhyay D. An augmented inverse probability weighted
survival function estimator. CAMS Technical Report, Department of Mathematical
Sciences, New Jersey Institute of Technology, 2008.
37. Pan Q, Schaubel DE. Proportional hazards models based on biased samples and
estimated selection probabilities. Canadian Journal of Statistics. 2008; 36: 111-27.
38. Blevins JR, Khan S. Local NLLS estimation of semi-parametric binary choice
models. The Econometrics Journal. 2013; 16: 135-60.
39. Nardi A, Schemper M. Comparing Cox and parametric models in clinical studies.
Stat Med. 2003; 22: 3597-610.
40. Noordzij M, Leffondré K, van Stralen KJ, et al. When do we need competing risks
methods for survival analysis in nephrology? Nephrology Dialysis Transplantation. 2013;
28: 2670-77.
41. Ray WA. Evaluating Medication Effects Outside of Clinical Trials: New-User
Designs. Am J Epidemiol. 2003; 158: 915-20.
89

42. Johnson ES, Bartman BA, Briesacher BA, et al. The incident user design in
comparative effectiveness research. Pharmacoepidemiol Drug Saf. 2013; 22: 1-6.

90
Figure 4. 1 Sample selection flow chart.
14,274,483 distinct
enrollees during 2007/13
14,117,969 persons
w/o schizophrenia or
bipolar disorder
(98.90%)
156,514 persons
with schizophrenia
or bipolar disorder
(1.10%)
64,073 patients did
not use
antipsychotics
(40.94%)
92,441 patients used
antipsychotics
(59.06%)
89,302 patients
age>=18
(96.60%)
49,666 patients
with 194,307
episodes having 18-
month enrollment
(55.62%)
172,313 episodes
after excluding
drugs with < 5,000
episodes
(88.68%)
21,994 episodes
belonging to the
drugs with <5,000
episodes
(11.32%)
39,636 patients
patients w/o 18-
month continuous
enrollment episodes
(44.38%)
3,139 patients
age <18
(3.40%)
91

Table 4. 1 Characteristics of patients at baseline and episode duration across drug groups
Aripiprazole
(N=31,201)
Fluphenazine
(N=5,672)
Haloperidol
(N=13,215)
Olanzapine
(N=18,013)
Quetiapine
(N=54,261)
Risperidone
(N=35,511)
Ziprasdione
(N=14,440)
p-value
Age (years) 48.2 (13.0) 52.2 (11.8) 52.2 (13.3) 54.1 (14.5) 51.5 (13.3) 53.0 (13.8) 48.6 (11.9) <0.0001
Male [N (%)] 10,292
(32.99)
2,859 (50.41) 6,626
(50.14)
8,271
(45.92)
20,778
(38.29)
15,251
(42.95)
4,776
(33.07)
<0.0001

Prior 6-month
total costs ($)
9,507
(17,020)
7,143 (14,794) 9,161
(17,655)
10,379
(19,943)
10,509
(20,959)
8,861
(17,209)
10,080
(16,017)
<0.0001

Prior 6-month
CCI
0.38 (1.00) 0.35 (0.91) 0.43 (1.07) 0.43 (1.10) 0.46 (1.12) 0.44 (1.06) 0.41 (1.04) <0.0001

Had
hospitalization
in prior
6-month
period [N
(%)]
5,722 (18.34) 1,023 (18.04) 3,024
(22.88)
3,973
(22.06)
12,145
(22.38)
7,860
(22.13)
3,126
(21.65)
<0.0001

Had ED visit
in prior
6-month
period [N
(%)]
9,619 (30.83) 1,745 (30.77) 4,520
(34.20)
5,984
(33.22)
19,399
(35.75)
12,262
(34.53)
4,913
(34.02)
<0.0001

Episode
duration
(days)
153.6 (191.4) 145.5 (197.8) 153.3
(208.1)
175.6
(225.8)
174.9
(220.9)
190.9
(239.5)
178.0
(229.1)
<0.0001
Values are presented as mean (SD) unless specified otherwise. Abbreviation: N, number; CCI: Charlson Comorbidity Index.
92
Table 4. 2 Cox regressions: hospitalization with acute coronary syndrome or ischemic
stroke
Acute coronary syndrome / ischemic stroke
Base-case Cox
regressions
Restricted-sample
Cox regressions
2SRI-IV
N sample 172,293 126,599 126,599
N events 1,145 791 791
aripiprazole 1.318 1.426 2.415
*

[0.964,1.801] [0.973,2.090] [1.214,4.802]
fluphenazine 1.323 1.675
*
2.084
[0.840,2.084] [1.017,2.757] [0.967,4.489]
olanzapine 1.225 1.295 2.175
*

[0.888,1.691] [0.870,1.926] [1.076,4.398]
quetiapine 1.222 1.299 2.138
*

[0.908,1.644] [0.903,1.869] [1.095,4.175]
risperidone 1.212 1.307 2.043
*

[0.907,1.620] [0.916,1.865] [1.081,3.864]
ziprasidone 1.164 1.330 2.231
*

[0.818,1.655] [0.874,2.024] [1.047,4.754]
Wald test of
significance of
residuals
a

N.A. N.A. P=0.7086
Likelihood-ratio test of
significance of
residuals
a

N.A. N.A. P=0.6836
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator:
haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
a: H0: each of the residual equals 0. Rejection of this hypothesis means endogeneity of
treatment choice cannot be ignored and instrumental variable estimates should be used.
93

Table 4. 3 Cox regressions: hospitalization with ventricular arrhythmia
Ventricular arrhythmia
Base-case Cox
regressions
Restricted-sample
Cox regressions
2SRI-IV
N sample 172,313 126,612 126,612
N events 198 126 126
aripiprazole 0.579 0.525 1.017
[0.287,1.170] [0.232,1.189] [0.222,4.660]

fluphenazine 0.974 0.531 0.324
[0.312,3.045] [0.114,2.462] [0.0174,6.055]

olanzapine 0.802 0.523 0.474
[0.398,1.615] [0.220,1.245] [0.0899,2.501]

quetiapine 0.727 0.608 0.550
[0.386,1.369] [0.291,1.273] [0.122,2.477]

risperidone 0.955 0.777 0.880
[0.513,1.778] [0.377,1.601] [0.210,3.687]

ziprasidone 0.593 0.602 0.409
[0.254,1.384] [0.231,1.573] [0.0569,2.944]
Wald test of
significance of
residuals
a

N.A. N.A. P=0.3828
Likelihood-ratio test of
significance of
residuals
a

N.A. N.A. P=0.2961
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator:
haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
a: H0: each of the residual equals 0. Rejection of this hypothesis means endogeneity of
treatment choice cannot be ignored and instrumental variable estimates should be used.

94
Table 4. 4 Cox regressions: hospitalization with acute kidney injury (AKI)
AKI
Base-case Cox
regressions
Restricted-sample
Cox regressions
2SRI-IV
N sample 172,283 126,589 126,589
N events 2,011 1,396 1,396
aripiprazole 1.152 1.036 1.893
*

[0.908,1.462] [0.790,1.357] [1.149,3.119]
fluphenazine 0.729 0.644 0.773
[0.483,1.102] [0.410,1.012] [0.402,1.488]
olanzapine 1.344
*
1.140 1.964
**

[1.057,1.708] [0.865,1.501] [1.182,3.261]
quetiapine 1.350
**
1.252 2.377
***

[1.082,1.685] [0.977,1.605] [1.472,3.837]
risperidone 1.147 1.010 1.759
*

[0.923,1.426] [0.791,1.290] [1.119,2.765]
ziprasidone 1.338
*
1.187 1.671
[1.035,1.729] [0.888,1.588] [0.953,2.930]
Wald test of
significance of
residuals
a

N.A. N.A. p=0.0598
Likelihood-ratio test of
significance of
residuals
a

N.A. N.A. P=0.0493
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator:
haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
a: H0: each of the residual equals 0. Rejection of this hypothesis means endogeneity of
treatment choice cannot be ignored and instrumental variable estimates should be used.
95
Table 4. 5 Cox regressions: hospitalization with Hypotension
Hypotension
Base-case Cox
regressions
Restricted-sample
Cox regressions
2SRI-IV
N sample 172,285 126,590 126,590
N events 1,557 1,094 1,094
aripiprazole 0.863 0.973 1.164
[0.657,1.133] [0.702,1.348] [0.661,2.049]
fluphenazine 1.158 1.438 1.189
[0.778,1.724] [0.935,2.213] [0.595,2.375]
olanzapine 1.069 1.199 1.419
[0.813,1.405] [0.862,1.668] [0.795,2.532]
quetiapine 1.293
*
1.372
*
1.548
[1.009,1.655] [1.019,1.848] [0.900,2.662]
risperidone 1.031 1.140 1.420
[0.805,1.320] [0.847,1.535] [0.846,2.384]
ziprasidone 1.287 1.563
**
1.706
[0.970,1.707] [1.124,2.173] [0.931,3.129]
Wald test of
significance of
residuals
a

N.A. N.A. p=0.7876
Likelihood-ratio test of
significance of
residuals
a

N.A. N.A. P=0.7810
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator:
haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
a: H0: each of the residual equals 0. Rejection of this hypothesis means endogeneity of
treatment choice cannot be ignored and instrumental variable estimates should be used.
96
Table 4. 6 Cox regressions: hospitalization with acute urinary retention
Acute urinary retention
Base-case Cox
regressions
Restricted-sample
Cox regressions
2SRI-IV
N sample 172,297 126,602 126,602
N events 736 519 519
aripiprazole 0.848 0.972 1.375
[0.596,1.206] [0.645,1.464] [0.636,2.976]
fluphenazine 0.964 0.851 1.463
[0.569,1.634] [0.462,1.569] [0.614,3.483]
olanzapine 0.777 0.791 0.819
[0.539,1.121] [0.513,1.219] [0.363,1.849]
quetiapine 0.900 0.913 1.174
[0.652,1.243] [0.625,1.334] [0.560,2.464]
risperidone 0.810 0.792 1.260
[0.588,1.117] [0.542,1.157] [0.625,2.539]
ziprasidone 0.952 1.150 1.374
[0.649,1.396] [0.746,1.773] [0.585,3.223]
Wald test of
significance of
residuals
a

N.A. N.A. p=0.3003
Likelihood-ratio test of
significance of
residuals
a

N.A. N.A. P=0.3003
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator:
haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
a: H0: each of the residual equals 0. Rejection of this hypothesis means endogeneity of
treatment choice cannot be ignored and instrumental variable estimates should be used.
97
Table 4. 7 Cox regressions: hospitalization with neuroleptic malignant
syndrome/rhabdomyolysis
Neuroleptic malignant syndrome/rhabdomyolysis
Base-case Cox
regressions
Restricted-sample
Cox regressions
2SRI-IV
N sample 172,307 126,608 126,608
N events 510 363 363
aripiprazole 0.851 0.821 0.715
[0.520,1.392] [0.468,1.439] [0.264,1.933]
fluphenazine 0.888 0.853 1.337
[0.431,1.830] [0.396,1.837] [0.462,3.873]
olanzapine 1.272 1.340 2.364
[0.791,2.046] [0.792,2.266] [0.958,5.835]
quetiapine 1.606
*
1.594 2.033
[1.043,2.474] [0.987,2.575] [0.841,4.914]
risperidone 1.510 1.298 1.764
[0.991,2.301] [0.809,2.082] [0.767,4.053]
ziprasidone 1.299 1.325 1.938
[0.784,2.154] [0.755,2.326] [0.715,5.251]
Wald test of
significance of
residuals
a

N.A. N.A. p=0.2840
Likelihood-ratio test of
significance of
residuals
a

N.A. N.A. P=0.2770
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator:
haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
a: H0: each of the residual equals 0. Rejection of this hypothesis means endogeneity of
treatment choice cannot be ignored and instrumental variable estimates should be used.
98

Table 4. 8 Cox regressions: hospitalization with pneumonia
Pneumonia
Base-case Cox
regressions
Restricted-sample
Cox regressions
2SRI-IV
N sample 172,280 126,590 126,590
N events 2,408 1,716 1,716
aripiprazole 1.077 1.035 1.287
[0.868,1.338] [0.809,1.324] [0.830,1.996]

fluphenazine 0.928 0.996 1.147
[0.659,1.307] [0.696,1.424] [0.681,1.931]

olanzapine 1.314
*
1.247 1.330
[1.058,1.633] [0.973,1.598] [0.849,2.082]

quetiapine 1.184 1.144 1.336
[0.966,1.450] [0.909,1.439] [0.875,2.041]

risperidone 1.057 1.027 1.154
[0.864,1.293] [0.818,1.289] [0.770,1.730]

ziprasidone 1.066 0.959 1.120
[0.840,1.353] [0.730,1.261] [0.686,1.830]
Wald test of
significance of
residuals
a

N.A. N.A. p=0.8750
Likelihood-ratio test of
significance of
residuals
a

N.A. N.A. P=0.8740
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator:
haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
a: H0: each of the residual equals 0. Rejection of this hypothesis means endogeneity of
treatment choice cannot be ignored and instrumental variable estimates should be used.

99
Appendix to Section 4
A4.1. Diagnoses Codes
Table A4.1. 1 Diagnoses codes used in the study to identify patients and target outcomes
Diagnoses ICD-9 Codes
Schizophrenia 295.X
Bipolar disorder 296.4-296.8
Acute coronary syndrome, ischemic
stroke
410.X, 411.1, 433.X, 434.X
Ventricular arrhythmia 427.1, 427.2, 427.4, 427.5
Acute kidney injury
a
584.5-584.9
Hypotension
a
458.X
Acute urinary retention
a
788.2
Neuroleptic malignant syndrome/
rhabdomyolysis
a
333.92, 728.88
Pneumonia
a
480.X-486.X, 770.0
a: Converted from ICD-10 codes in order to be compliant with existing studies.
Reference: http://www.icd10data.com/Convert
100
A4.2 Covariates
A4.2. 1 Covariates used in regressions
Demographic
Age categories (10 year intervals)
Sex
Medicare Advantage vs. Commercial
Diagnoses in prior 6 months
Prior occurrence of target events
Infectious diseases
Neoplasms
Endocrine disorders
Diabetes
Obesity
Diseases of the blood and blood-forming organs
Dementias
Alcohol-induced mental disorders
Drug-induced mental disorders
Transient mental disorders due to conditions classified elsewhere
Persistent mental disorders due to conditions classified elsewhere
Simple type schizophrenia
Paranoid type schizophrenia
Schizophreniform disorder
Schizoaffective disorder
Other specified types of schizophrenia
Major depressive disorder single episode
Bipolar I disorder-single manic episode, Manic disorder recurrent episode
Bipolar I disorder, most recent episode (or current) manic
Bipolar I disorder, most recent episode (or current) depressed
Bipolar I disorder, most recent episode (or current) mixed
Bipolar I disorder, most recent episode (or current) unspecified
Other and unspecified bipolar disorders
Other and unspecified episodic mood disorder
Delusional disorders
Other nonorganic psychoses
Anxiety states
Obsessive-compulsive disorders
Dysthymic disorder
Neurasthenia, Depersonalization disorder, Hypochondriasis, Somatoform disorders,
Unspecified nonpsychotic mental disorder
Personality disorders
Alcohol dependence syndrome
Drug dependence
Other nonpsychotic mental disorders
Diseases of the nervous system and sense organs
Acute myocardial infarction
101

Angina pectoris
Conduction disorders, cardiac dysrhythmias
Heart failure
Other cardiovascular diseases
Diseases of the respiratory system
Diseases of the digestive system
Diseases of the genitourinary system
Diseases of the skin and subcutaneous tissue
Diseases of the musculoskeletal system and connective tissue
Congenital Anomalies
Injury And Poisoning
External Causes Of Injury And Poisoning
Blood sugar test and monitor
Lab tests for blood sugar
Blood sugar monitoring equipment
Use of other classes of drugs in prior 6 months
Hypnotics
Antiparkinson agents
Anxiolytics
Antiepileptic drugs/ anti-seizure medications
Mood stabilizer
Anticonvulsants
Antidepressants
Primary Extrapyramidal symptoms drugs
Secondary Extrapyramidal symptoms drugs
Antidiabetics
Antihyperlipidemic agents
Other medical services
Long-term care facility
Assisted living facility
Inpatient admission
Emergency department visit
Treatment history and current episode characteristics
Whether the patient restarted an antipsychotic used before, switched from another
antipsychotic, or augmented with an additional antipsychotic
The number of treatment episodes a patient had
Used depot antipsychotic
Monotherapy vs poly therapy (any overlap of time with the use of another antipsychotic)
Number of different types of previously used antipsychotics
Dose categories
Within target range (reference)
Below target range
Above target range
102

A4.3 Validity of Instrumental Variables (IVs)
Ideally, IVs should be correlated with the endogenous treatment variable, but not
correlated with the outcomes except through predicting the treatment variable (1) such
that they are orthogonal to the unobserved heterogeneity. Therefore, whether the IVs are
strongly correlated with the treatment choice and whether they are not directly correlated
with the outcomes were tested.

A4.3.1 First-stage Statistics
The first stage regression for 2-stage residual inclusion (2SRI)-IV regressions was a
multinomial logit regression. Therefore, the traditional way of using F-statistic as
evidence of the strength of IVs is not applicable. In fact, consensus on how the strength
of IVs should be determined in nonlinear regressions is still absent in literature. In the
following tables, we first show the results of the multinomial logit regression. Then, we
show the results of a series of binary linear probability models (LPMs) for each of the
drugs and the F-statistics for each of the LPM regressions.
As shown in Table A4.3.1.1, The IVs were statistically significant predictors of the
drug choice in the current episode. Also, the pseudo-R
2
of the multinomial logit
regression was 0.384. McFadden suggested that a pseudo-R
2
between 0.2-0.4 is evidence
of very good fitting (2).
Table A4.3.1.2 lists the results of the binary LPMs. Bootstrapped standard errors with
200 repeats were used for this set of analyses. Traditionally, an F-statistic value greater
than 10 is considered as the threshold of non-weak IVs (3). In the current LPMs, the
F-statistics were substantially greater than 10.

103
A4.3.2 Exogeneity of IVs
Exogeneity of IVs cannot be directly tested in nonlinear regressions. In fact, tests for
this purpose, known as overidentification tests, are hardly gold standards even though
they exist in a linear setting (4). As mentioned in the method section of main text, proxy
risk factors in the prior 6-month period was regressed on the IVs using linear models to
examine the correlation between the target events and the IVs conditional on other
covariates. Bootstrapped standard errors with 200 repeats were used. Instrumental
variables should not be correlated with proxy risk factors conditional on observed
covariates (5). Absence of correlation is indirect evidence of the exogeneity of IVs. The
results are displayed in Table A4.3.2.1, Table A4.3.2.1, Table A4.3.2.3, and Table
A4.3.2.4. Overall, lags of treatment were not statistically significant correlated with prior
occurrence of target events. Indicators of prescriber specialties were statistically
significantly but very weakly correlated with prior occurrence of target events. Therefore,
we did not find plausible evidence that the IVs were associated with the outcomes in
ways other than through affecting the treatment.
Appendix A4.3 References
1. Greene WH. Econometric Analysis. 7 ed. Upper Saddle River, NJ, USA: Pearson
Education, Inc., 2012.
2. McFadden D. Quantitative Methods for Analysing Travel Behaviour of Individuals:
Some Recent Developments',(in) Hensher. DA and Stopher, PR (eds) Behavioural Travel
Modelling, Croom Helm, London. 1978.
3. Stock JH, Yogo M. Testing for weak instruments in linear IV regression.
Identification and inference for econometric models: Essays in honor of Thomas
104
Rothenberg. 2005; 1.
4. Murray MP. Avoiding invalid instruments and coping with weak instruments. The
journal of economic perspectives. 2006; 20: 111-32.
5. Schneeweiss S, Setoguchi S, Brookhart A, et al. Risk of death associated with the use
of conventional versus atypical antipsychotic drugs among elderly patients. CMAJ :
Canadian Medical Association Journal. 2007; 176: 627-32.
105
Table A4.3.1. 1 First-stage multinomial logit regression results
Aripiprazole Fluphenazine Olanzapine Quetiapine Risperidone Ziprasidone
Speciality:
psychiatry
1.187
***
1.017 1.102
**
1.097
**
1.197
***
1.168
***

Specialty: family
medicine, primary
care, general
internal medicine
1.042 0.600
***
1.288
***
1.428
***
1.295
***
0.851
**
lag 1 aripiprazole 72.50
***
12.04
***
7.363
***
11.24
***
8.297
***
9.107
***

lag 1 fluphenazine 11.13
***
286.2
***
11.26
***
9.866
***
10.57
***
9.006
***

lag 1 olanzapine 9.174
***
11.78
***
51.47
***
6.413
***
6.786
***
7.130
***

lag 1 quetiapine 13.57
***
11.03
***
6.810
***
37.72
***
9.422
***
9.929
***

lag 1 risperidone 9.259
***
10.53
***
6.175
***
7.992
***
41.51
***
7.271
***

lag 1 ziprasidone 10.08
***
8.996
***
6.641
***
8.652
***
7.187
***
62.37
***

lag 2 aripiprazole 15.92
***
4.127
***
4.948
***
5.808
***
5.477
***
5.866
***

lag 2 fluphenazine 3.776
***
35.38
***
4.181
***
3.860
***
3.865
***
3.683
***

lag 2 olanzapine 5.452
***
4.435
***
16.73
***
5.574
***
5.051
***
5.310
***

lag 2 quetiapine 5.789
***
4.300
***
5.008
***
12.87
***
5.044
***
5.444
***

lag 2 risperidone 5.235
***
4.561
***
5.114
***
5.328
***
11.80
***
5.091
***

106

lag 2 ziprasidone 6.481
***
4.778
***
5.647
***
5.988
***
5.473
***
20.08
***

N 126,612 126,612 126,612 126,612 126,612 126,612
Exponentiated coefficients. Baseline comparator of specialties: other specialties. Baseline comparators of lags of treatment:
haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001

107
Table A4.3.1. 2 Binary linear probability models of drug choices on IVs
Aripiprazole Fluphenazine Haloperidol Olanzapine Quetiapine Risperidone Ziprasidone
Specialty:
psychiatry
0.00616
***
-0.00186
*
-0.00640
***
-0.000481 -0.00705
**
0.00695
***
0.00268
Specialty: family
medicine,
primary care,
general internal
medicine
-0.0138
***
-0.00944
***
-0.00930
***
0.00647
**
0.0338
***
0.00804
**
-0.0157
***
lag 1 aripiprazole 0.475
***
0.00456
**
-0.452
***
-0.0158
***
0.0137
**
-0.0189
***
-0.00624
*

lag 1
fluphenazine
-0.000644 0.504
***
-0.465
***
-0.00261 -0.0179
*
-0.00853 -0.00952
*

lag 1 olanzapine 0.0112
**
0.00771
***
-0.442
***
0.467
***
-0.0280
***
-0.0120
*
-0.00381
lag 1 quetiapine 0.0228
***
0.00356
*
-0.451
***
-0.0185
***
0.444
***
-0.00160 0.000853
lag 1 risperidone 0.00421 0.00386
*
-0.447
***
-0.0179
***
-0.0129
**
0.477
***
-0.00684
*

lag 1 ziprasidone 0.0133
**
0.00306 -0.442
***
-0.0103
**
0.0113 -0.00830 0.433
***

lag 2 aripiprazole 0.223
***
-0.00377
*
-0.280
***
0.00884
*
0.0196
***
0.0206
***
0.0119
***

lag 2
fluphenazine
-0.0100 0.307
***
-0.272
***
-0.00430 -0.00758 -0.0104 -0.00274
lag 2 olanzapine 0.00773 -0.00336 -0.282
***
0.221
***
0.0336
***
0.0147
**
0.00856
*

108
lag 2 quetiapine 0.00449 -0.00268 -0.280
***
0.0106
**
0.255
***
0.00684 0.00631
*

lag 2 risperidone 0.00913
*
-0.000814 -0.279
***
0.0149
***
0.0268
***
0.220
***
0.00904
**

lag 2 ziprasidone 0.0156
***
-0.00153 -0.285
***
0.0128
**
0.0224
***
0.0152
**
0.221
***

N 126,612 126,612 126,612 126,612 126,612 126,612 126,612
F-statistic 3012.1 1132.0 1694.6 1847.1 5019.1 3677.9 1231.1
Baseline comparator of specialties: other specialties. Baseline comparators of lags of treatment: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
109
Table A4.3.2. 1 Linear regressions of pre-index costs, Charlson Comorbidity Index (CCI), indicator of hospitalization, and indicator of
emergency department visit on first lag of treatment, second lag of treatment, and specialty indicators
Pre-index costs Pre-index CCI Pre-index hospitalization
indicator
Pre-index emergency
department visit indicator
lag 1 aripiprazole 316.0 0.00702 -0.00147
-0.0104
*
lag 1 fluphenazine 348.9 -0.00869 -0.00456 0.00462
lag 1 olanzapine
520.9
*
0.0220 0.00436 -0.000903
lag 1 quetiapine
478.7
*
0.0192 0.00880 0.00485
lag 1 risperidone 360.5 -0.0175 0.00413 0.000742
lag 1 ziprasidone
523.5
*
-0.00251 0.00131 -0.00644
lag 2 aripiprazole 229.8 0.0169 0.00276 -0.00763
lag 2 fluphenazine -16.95 0.00965 -0.00477 -0.00421
lag 2 olanzapine 220.3 0.0248 0.000380 -0.00456
lag 2 quetiapine 158.0 0.0163 0.00350 -0.000454
lag 2 risperidone -135.2 0.00778 0.00379 -0.000296
lag 2 ziprasidone 274.2 0.0130 0.000381 -0.00305
Specialty: psychiatry
-866.1
***
-0.0424
***
-0.0190
***
-0.0245
***
Specialty: family medicine,
primary care, general internal
medicine
-357.9
**
-0.0145 -0.00502
-0.0179
***
N 126,612 126,612 126,612 126,612
110
Table A4.3.2. 2 Regressions of target events in prior 6-month period on indicators of prescriber specialties
(1) (2) (3) (4) (5) (6) (7)
Acute
coronary
syndrome /
ischemic
stroke
Ventricular
arrhythmia
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic
malignant
syndrome /
rhabdomyolysis
Pneumonia
Specialty:
psychiatry
-0.000748 -0.000254 0.000356 0.000165 -0.000349 -0.0000977 0.000277
Specialty:
family
medicine,
primary
care, general
internal
medicine
-0.000671 0.000138 0.00492
***
0.00201 -0.000243 0.000584 0.00412
**

N 126,612 126,612 126,612 126,612 126,612 126,612 126,612
Baseline comparator: other specialties.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
111
Table A4.3.2. 3 Regressions of target events in prior 6-month period on indicators of first lag of treatment
(1) (2) (3) (4) (5) (6) (7)
Acute
coronary
syndrome /
ischemic
stroke
Ventricular
arrhythmia
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic
malignant
syndrome /
rhabdomyolysis
Pneumonia
lag 1
aripiprazole
0.00146 -0.000579 0.000431 0.000750 -0.0000880 0.0000686 0.00280
*

lag 1
fluphenazine
0.00219 -0.000958 -0.00198 0.00158 -0.00162 0.00106 -0.000331
lag 1
olanzapine
-0.0000722 -0.000966 0.00299 0.00441
**
-0.00168 0.00165 0.00223
lag 1
quetiapine
0.000400 -0.000541 0.000688 0.00158 -0.000804 0.00101 -0.000296
lag 1
risperidone
-0.000129 0.000228 0.00193 0.000791 -0.000625 0.000767 0.000663
lag 1
ziprasidone
0.000957 -0.000585 0.00270 0.00187 0.000543 0.00157 -0.000665
N 126,612 126,612 126,612 126,612 126,612 126,612 126,612
Baseline comparators of lags of treatment: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
112
Table A4.3.2. 4 Regressions of target events in prior 6-month period on indicators of second lag of treatment
(1) (2) (3) (4) (5) (6) (7)
Acute
coronary
syndrome /
ischemic
stroke
Ventricular
arrhythmia
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic
malignant
syndrome /
rhabdomyolysis
Pneumonia
lag 2
aripiprazole
0.00219 -0.000213 0.00154 0.000247 -0.000906 0.000162 0.00203
lag 2
fluphenazine
0.00212 --0.000514 -0.00174 0.00174 -0.00170 0.000835 0.00153
lag 2
olanzapine
0.000958 -0.000534 0.00244 0.00292 -0.00118 0.00238
*
0.000683
lag 2
quetiapine
0.00165 -0.000560 0.00135 0.000371 -0.000995 0.000994 -0.000411
lag 2
risperidone
0.000374 0.0000859 0.00313
*
0.00113 -0.000676 0.00109 0.00165
lag 2
ziprasidone
0.00210 -0.000342 0.00154 0.000852 -0.000779 0.000954 -0.000607
N 126,612 126,612 126,612 126,612 126,612 126,612 126,612
Baseline comparators of lags of treatment: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
113
A4.4 Sensitivity Analyses
A4.4.1 Endogeneity of choosing between atypical and typical antipsychotics
According to results shown in Table A4.4.1, endogeneity not only existed across specific drug groups, but also existed between the
larger groups of atypical antipsychotics and typical antipsychotics. Atypical antipsychotics were not associated with a higher risk of
any target events in base-case regressions, but were associated with a higher risk of acute kidney injury (AKI) in 2SRI-IV regressions.
Table A4.4.1. 1 Hospitalization with cardiovascular events, acute kidney injury (AKI), or events that may cause AKI-- Atypical vs.
Typical
(1) (2) (3) (4) (5) (6) (7)
Acute
coronary
syndrome /
ischemic
stroke
Ventricular
arrhythmia
Acute kidney
injury
Hypotension Acute
urinary
retention
Neuroleptic
malignant syndrome
/ rhabdomyolysis
Pneumonia
Base-case Cox regressions
Atypical indicator 1.142 0.775 1.313
**
1.055 0.834 1.425 1.148
[0.893,1.459] [0.454,1.325] [1.083,1.591] [0.858,1.297] [0.636,1.094] [0.988,2.055] [0.966,1.364]
N 172,293 172,313 172,283 172,285 172,297 172,307 172,280
Restricted sample Cox regression
Atypical indicator 1.132 0.718 1.203 1.095 0.912 1.363 1.081
[0.848,1.511] [0.378,1.366] [0.970,1.491] [0.862,1.393] [0.664,1.252] [0.912,2.036] [0.893,1.310]
2SRI-IV regressions
Atypical indicator 1.606 0.995 2.041
***
1.267 1.166 1.442 1.136
[0.927,2.783] [0.251,3.943] [1.339,3.113] [0.806,1.991] [0.630,2.159] [0.711,2.926] [0.796,1.622]
Test of exogeneity P=0.138 P=0.596 P=0.004 P=0.455 P=0.356 P=0.849 P=0.744
N 126,599 126,612 126,589 126,590 126,602 126,608 126,590
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: typical antipsychotics.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
114
A4.4.2 Sensitivity analyses using a 360-day pre-index period prior to the first observed antipsychotic use.
Table A4.4.2. 1 Cox regressions: hospitalization with acute major cardiovascular events using 360-day pre-index period
Acute coronary syndrome / ischemic stroke Ventricular arrhythmia
aripiprazole 1.381 0.542
[0.973,1.961] [0.261,1.126]
fluphenazine 1.365 0.555
[0.824,2.262] [0.122,2.526]
olanzapine 1.270 0.684
[0.885,1.823] [0.328,1.426]
quetiapine 1.291 0.628
[0.926,1.801] [0.324,1.215]
risperidone 1.332 0.750
[0.964,1.842] [0.389,1.446]
1.381 0.542
ziprasidone [0.973,1.961] [0.261,1.126]
N 140,356 140,368
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
115
Table A4.4.2. 2 Cox regressions: hospitalization with acute kidney injury (AKI) or events that may cause AKI using 360-day pre-index
period
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic malignant syndrome
/ rhabdomyolysis
Pneumonia
aripiprazole 1.117 0.890 0.815 0.987 1.071
[0.864,1.444] [0.660,1.199] [0.555,1.198] [0.571,1.707] [0.848,1.353]
fluphenazine 0.724 1.225 1.124 0.980 1.027
[0.465,1.128] [0.798,1.880] [0.655,1.928] [0.451,2.131] [0.719,1.466]
olanzapine 1.330
*
1.130 0.741 1.462 1.333
*

[1.028,1.720] [0.838,1.524] [0.496,1.106] [0.863,2.478] [1.055,1.684]
quetiapine 1.347
*
1.305 0.921 1.866
*
1.182
[1.062,1.708] [0.995,1.713] [0.649,1.308] [1.152,3.024] [0.949,1.471]
risperidone 1.131 1.083 0.780 1.752
*
1.053
[0.896,1.429] [0.826,1.420] [0.550,1.106] [1.096,2.802] [0.848,1.308]
ziprasidone 1.410
*
1.405
*
1.062 1.506 1.021
[1.072,1.855] [1.033,1.912] [0.705,1.600] [0.858,2.645] [0.787,1.325]
N 140,341 140,340 140,352 140,362 140,341
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
116
A4.4.3 Cox regressions with 2SRI for sample attrition. The number of months with continuous enrollment prior to the start of the
current episode was used as the IV for sample attrition.
We found significant evidence of sample selection due to unobserved factors in the analyses of the vast majority of the outcomes.
As a result, the results with 2SRI for sample selection are listed in what follows.
Although the existence of sample selection bias could not be ruled out, the results with correction of sample selection were similar
to base-case analyses.
Table A4.4.3. 1 Cox regressions: hospitalization with acute major cardiovascular events using 2SRI for sample attrition
Acute coronary syndrome / ischemic stroke Ventricular arrhythmia
aripiprazole 1.339 0.586
[0.979,1.831] [0.290,1.184]
fluphenazine 1.352 1.013
[0.858,2.130] [0.325,3.161]
olanzapine 1.224 0.784
[0.887,1.689] [0.389,1.578]
quetiapine 1.239 0.726
[0.921,1.667] [0.386,1.368]
risperidone 1.203 0.922
[0.900,1.608] [0.495,1.717]
ziprasidone 1.184 0.602
[0.832,1.685] [0.258,1.406]
N 172,293 172,313
Walt test of significance of the residual for
sample attrition
P=0.010 P=0.064
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
117
Table A4.4.3. 2 Cox regressions: hospitalization with acute kidney injury (AKI) or events that may cause AKI using 2SRI for sample
attrition. The residual from attrition was significant. The results were similar to results in Table 5 of main text.
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic malignant syndrome
/ rhabdomyolysis
Pneumonia
aripiprazole 1.205 0.889 0.858 0.888 1.097
[0.949,1.530] [0.677,1.168] [0.603,1.221] [0.543,1.453] [0.883,1.362]
fluphenazine 0.770 1.203 0.992 0.934 0.950
[0.509,1.163] [0.808,1.792] [0.585,1.682] [0.453,1.925] [0.675,1.339]
olanzapine 1.334
*
1.070 0.761 1.262 1.310
*

[1.050,1.696] [0.814,1.406] [0.528,1.097] [0.786,2.029] [1.054,1.628]
quetiapine 1.393
**
1.324
*
0.902 1.646
*
1.201
[1.116,1.739] [1.034,1.695] [0.653,1.247] [1.069,2.534] [0.981,1.472]
risperidone 1.129 1.022 0.782 1.476 1.048
[0.908,1.403] [0.798,1.308] [0.567,1.079] [0.969,2.248] [0.857,1.282]
1.414
**
1.335
*
0.978 1.364 1.090
ziprasidone [1.094,1.827] [1.006,1.771] [0.667,1.436] [0.823,2.262] [0.859,1.384]
N 172,283 172,285 172,297 172,307 172,280
Walt test of
significance of the
residual for sample
attrition
P<0.0001 P<0.0001 P<0.0001 P<0.0001 P<0.0001
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
118
A4.4.4 Analyses using Weibull regressions. The results are very similar to Cox regressions.
Table A4.4.4. 1 Weibull regressions: hospitalization with acute major cardiovascular events
Acute coronary syndrome / ischemic stroke Ventricular arrhythmia
aripiprazole 1.358 0.589
[0.994,1.855] [0.292,1.189]
fluphenazine 1.324 0.925
[0.841,2.086] [0.295,2.898]
olanzapine 1.226 0.797
[0.889,1.691] [0.396,1.604]
quetiapine 1.240 0.725
[0.922,1.668] [0.385,1.363]
risperidone 1.221 0.948
[0.914,1.632] [0.509,1.764]
ziprasidone 1.172 0.582
[0.824,1.667] [0.249,1.359]
N 172,293 172,313
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
119
Table A4.4.4. 2 Weibull regressions: hospitalization with acute kidney injury (AKI) or events that may cause AKI
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic malignant syndrome
/ rhabdomyolysis
Pneumonia
aripiprazole 1.180 0.885 0.846 0.866 1.101
[0.930,1.497] [0.674,1.162] [0.595,1.203] [0.530,1.415] [0.887,1.367]
fluphenazine 0.724 1.158 0.964 0.889 0.922
[0.479,1.093] [0.778,1.724] [0.569,1.634] [0.431,1.832] [0.655,1.299]
olanzapine 1.355
*
1.081 0.768 1.278 1.322
*

[1.066,1.722] [0.822,1.420] [0.533,1.107] [0.796,2.052] [1.064,1.643]
quetiapine 1.361
**
1.301
*
0.886 1.602
*
1.189
[1.090,1.697] [1.017,1.666] [0.642,1.224] [1.042,2.464] [0.971,1.455]
risperidone 1.153 1.037 0.793 1.505 1.060
[0.928,1.433] [0.810,1.327] [0.575,1.093] [0.989,2.291] [0.867,1.296]
ziprasidone 1.349
*
1.290 0.942 1.295 1.065
[1.044,1.744] [0.973,1.712] [0.642,1.381] [0.782,2.145] [0.840,1.352]
N 172,283 172,285 172,297 172,307 172,280
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
120

A4.4.5 Using alternative definition of events. Emergency department (ED) visits and urgent care visits with target diagnoses were
included in addition to hospitalization with target diagnosis.

Table A4.4.5. 1 Cox regressions: hospitalization, ED visits, or urgent care visits with acute major cardiovascular events
Acute coronary syndrome / ischemic stroke Ventricular arrhythmia
aripiprazole 1.320
*
0.639
[1.014,1.720] [0.326,1.249]

fluphenazine 1.369 0.904
[0.931,2.013] [0.293,2.789]

olanzapine 1.177 0.812
[0.895,1.548] [0.414,1.594]

quetiapine 1.278 0.762
[0.995,1.641] [0.414,1.403]

risperidone 1.255 0.964
[0.982,1.605] [0.530,1.753]

ziprasidone 1.146 0.595
[0.849,1.546] [0.264,1.341]
N 172,293 172,313
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001

121

Table A4.4.5. 2 Cox regressions: hospitalization, ED visits, or urgent care visits with acute kidney injury (AKI) or events that may
cause AKI
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic malignant syndrome
/ rhabdomyolysis
Pneumonia
aripiprazole 1.185 0.854 0.863 0.823 1.091
[0.936,1.500] [0.667,1.094] [0.615,1.212] [0.514,1.320] [0.893,1.334]

fluphenazine 0.787 1.035 1.062 0.900 0.909
[0.530,1.170] [0.711,1.507] [0.646,1.744] [0.452,1.792] [0.661,1.250]

olanzapine 1.365
*
1.034 0.764 1.133 1.284
*

[1.076,1.731] [0.806,1.327] [0.537,1.088] [0.715,1.795] [1.049,1.571]

quetiapine 1.401
**
1.226 0.900 1.533
*
1.177
[1.125,1.745] [0.979,1.536] [0.658,1.231] [1.014,2.316] [0.975,1.421]

risperidone 1.175 1.000 0.779 1.454 1.048
[0.948,1.457] [0.799,1.252] [0.570,1.065] [0.972,2.174] [0.870,1.263]

ziprasidone 1.350
*
1.165 0.945 1.242 1.028
[1.047,1.741] [0.898,1.513] [0.653,1.368] [0.764,2.018] [0.823,1.283]
N 172,282 172,279 172,296 172,307 172,272
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001

122
A4.4.6 Using episodes without overlap.
Table A4.4.6. 1 Cox regressions using episodes without overlap: hospitalization with acute major cardiovascular events
Acute coronary syndrome / ischemic stroke Ventricular arrhythmia
aripiprazole 1.307 0.456
[0.870,1.963] [0.186,1.121]
fluphenazine 1.077 1.001
[0.573,2.026] [0.207,4.833]
olanzapine 1.060 0.617
[0.693,1.621] [0.251,1.515]
quetiapine 1.152 0.500
[0.780,1.702] [0.222,1.126]
risperidone 1.193 0.798
[0.814,1.748] [0.362,1.758]
ziprasidone 1.050 0.331
[0.653,1.687] [0.0971,1.130]
N 71,585 71,592
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
123
Table A4.4.6. 2 Cox regressions using episodes without overlap: hospitalization with acute kidney injury (AKI) or events that may
cause AKI
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic malignant syndrome
/ rhabdomyolysis
Pneumonia
aripiprazole 1.511
*
0.970 0.887 0.569 1.296
[1.056,2.163] [0.644,1.462] [0.509,1.547] [0.297,1.091] [0.936,1.793]
fluphenazine 0.733 1.018 2.204
*
1.185 0.788
[0.384,1.397] [0.533,1.945] [1.053,4.610] [0.487,2.882] [0.452,1.375]
olanzapine 1.848
***
1.430 0.834 0.800 1.607
**

[1.291,2.647] [0.952,2.148] [0.467,1.490] [0.424,1.509] [1.158,2.229]
quetiapine 1.786
***
1.654
**
1.036 1.275 1.398
*

[1.270,2.512] [1.132,2.417] [0.616,1.745] [0.728,2.233] [1.024,1.909]
risperidone 1.450
*
1.223 0.889 1.132 1.235
[1.035,2.032] [0.836,1.788] [0.526,1.503] [0.652,1.964] [0.906,1.683]
ziprasidone 1.896
**
1.382 1.176 0.846 1.166
[1.293,2.781] [0.892,2.139] [0.647,2.137] [0.419,1.708] [0.812,1.676]
N 71,579 71,579 71,584 71,588 71,579
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
124
A4.4.7 Using competing risk model to take into consideration of potential nonrandom censoring.
Table A4.4.7. 1 Competing risk model: hospitalization with acute major cardiovascular events
Acute coronary syndrome / ischemic stroke Ventricular arrhythmia
aripiprazole 1.292 0.562
[0.945,1.766] [0.278,1.134]
fluphenazine 1.339 1.017
[0.850,2.110] [0.326,3.170]
olanzapine 1.233 0.790
[0.893,1.702] [0.393,1.589]
quetiapine 1.237 0.725
[0.919,1.664] [0.386,1.363]
risperidone 1.225 0.943
[0.917,1.638] [0.507,1.752]
ziprasidone 1.166 0.585
[0.820,1.658] [0.251,1.363]
N 172,293 172,313
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
125
Table A4.4.7. 2 Competing risk model: hospitalization with acute kidney injury (AKI) or events that may cause AKI
Acute kidney
injury
Hypotension Acute urinary
retention
Neuroleptic malignant syndrome
/ rhabdomyolysis
Pneumonia
aripiprazole 1.141 0.849 0.824 0.833 1.069
[0.899,1.448] [0.647,1.115] [0.579,1.173] [0.510,1.363] [0.861,1.327]
fluphenazine 0.739 1.167 0.977 0.896 0.938
[0.489,1.116] [0.784,1.734] [0.576,1.657] [0.435,1.847] [0.665,1.321]
olanzapine 1.366
*
1.077 0.773 1.279 1.331
*

[1.075,1.736] [0.820,1.416] [0.536,1.114] [0.796,2.056] [1.071,1.654]
quetiapine 1.379
**
1.313
*
0.900 1.613
*
1.208
[1.105,1.721] [1.026,1.681] [0.651,1.243] [1.047,2.483] [0.986,1.480]
risperidone 1.173 1.049 0.807 1.532
*
1.079
[0.944,1.457] [0.820,1.343] [0.585,1.113] [1.005,2.334] [0.883,1.320]
ziprasidone 1.343
*
1.288 0.948 1.292 1.069
[1.039,1.735] [0.971,1.709] [0.646,1.390] [0.780,2.140] [0.843,1.357]
N 172,283 172,285 172,297 172,307 172,280
Exponentiated coefficients; 95% confidence intervals in brackets. Baseline comparator: haloperidol.
*
p < 0.05,
**
p < 0.01,
***
p < 0.001
126

Abstract (if available)

Abstract The gold standard of drug comparison is randomized clinical trials (RCTs). However, RCTs fail to provide evidence on many fronts by design. While they have impeccable internal validity, they exclude many relevant patient groups such as the elderly, those with co-morbidities, and those who have been exposed to other drugs. More, they are conducted in ideal settings to enhance patient adherence which does not necessarily reflect real-world practice. These limitations of RCTs may contribute to discrepancy between RCT results and daily practice results. Hence, it is important to supplement RCT results with evidence obtained from real-world data for policy and clinical decision making. ❧ The dissertation consists of three separate studies, each comparing costs, effectiveness or safety across two or more antipsychotics. The first study compares healthcare services use and costs between two atypical antipsychotics, ziprasidone and olanzapine using IMS Pharmetrics Database. The second study used Medicaid data to critique the statistical methods commonly used in observational research and examines the importance of taking prior treatment history, including episode type into account when conducting retrospective comparative effectiveness research. The third study investigates the risks of acute cardiovascular events and acute kidney injury associated with a wide spectrum of antipsychotics using Humana administrative claims databases.

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

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

PDF

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

PDF

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

PDF

Hepatitis C in the post-interferon era: selected essays in health economics

PDF

The development of the prognostic propensity score: an introduction to a method to identify optimal treatment according to individual tailoring variables when heterogeneous treatment effects are ...

PDF

Advances and applications for economic evaluation methods in health technology assessment (HTA)

PDF

Health economics and outcomes research for informed decision making in rapidly evolving therapeutic areas

PDF

US biosimilar market entry and uptake: US commercial payer preferences in access decision making

PDF

Selected papers on methods for evaluating the impact of marijuana use on BMI and other risk factors for metabolic syndrome

PDF

Evaluating approaches to reduce inappropriate antibiotic use in the United States

PDF

An evaluation of risk predictors for disease progression in Veteran Affairs patients with chronic hepatitis C

PDF

Essays on the economics of infectious diseases

PDF

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

PDF

Understanding the value of hemophilia therapies

PDF

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

PDF

The price and clinical impact of specialty drug management in the US healthcare system

PDF

Three essays on emerging issues in hemophilia care

PDF

Impact of pharmacy-based transitional care on healthcare utilization and costs

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 Jiang, Yawen (author)

Core Title Three essays on comparative economic and clinical outcomes of antipsychotics using retrospective claims data

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Pharmaceutical Economics and Policy

Publication Date 04/22/2016

Defense Date 03/10/2016

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

Tag antipsychotic,comparative,economic,effective,inclusion,instrumental,OAI-PMH Harvest,residual,retrospective,stage,Survival,variable

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor McCombs, Jeffrey S. (committee chair), Hay, Joel W. (committee member), Joyce, Geoffrey F. (committee member)

Creator Email jiangyawenac@gmail.com,yawenjia@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-241556

Unique identifier UC11279202

Identifier etd-JiangYawen-4357.pdf (filename),usctheses-c40-241556 (legacy record id)

Legacy Identifier etd-JiangYawen-4357.pdf

Dmrecord 241556

Document Type Dissertation

Format application/pdf (imt)

Rights Jiang, Yawen

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA