Estimating the Impact of Adherence to and Persistence

with Atypical Antipsychotic Therapy on Health Care

Costs and Risk of Hospitalization
Yawen Jiang,* and Weiyi Ni
Department of Clinical Pharmacy, Pharmaceutical Economics and Policy, Leonard D. Schaeffer Center for Health
Policy & Economics, School of Pharmacy, University of Southern California, USC Schaeffer Center, Verna &
Peter Dauterive Hall (VPD), Los Angeles, California

STUDY OBJECTIVE To estimate the impact of adherence to and persistence with atypical antipsychotics
on health care costs and risk of hospitalization by controlling potential sources of endogeneity.
DESIGN Retrospective cohort study using medical and pharmacy claims data.
DATA SOURCE Humana health care insurance database.
PATIENTS A total of 32,374 patients with a diagnosis of schizophrenia or bipolar disorder and who had
a prescription for noninjectable atypical antipsychotics (aripiprazole, asenapine, clozapine, iloperi-
done, lurasidone, olanzapine, paliperidone, quetiapine, risperidone, or ziprasidone), after a washout
period of at least 180 days during which there was no use of any atypical antipsychotics, between
January 2007 and June 2013.
MEASUREMENTS AND MAIN RESULTS The effects of adherence (proportion of days covered by all atypical
antipsychotic prescription fills) to and persistence (time from initiation to discontinuation of
therapy) with atypical antipsychotics on outcomes (all-cause total health care costs, medication
costs, medical services costs, and inpatient admissions) were examined. To exclude potential bias
due to mutual causality between drug use patterns and health care utilization, the effects of adher-
ence and persistence measured in the first year on outcomes measured in the second year were
investigated. Instrumental variable regressions using reimbursement rate and mail order as instru-
mental variables were conducted to correct potential endogeneity due to omitted variable bias. Being
adherent decreased total costs by $19,497 (p<0.05), increased medication costs by $8194 (p<0.001),
decreased medical services costs by $27,664 (p<0.001), and reduced hospitalization risk by 27%
(p<0.001). Being persistent decreased individual total costs by $23,927 (p<0.05), increased medica-
tion costs by $10,278 (p<0.001), and decreased medical services costs by $34,178 (p<0.001). We
could not identify a significant association between persistence and the risk of hospitalization.
CONCLUSION Good adherence to and persistence with atypical antipsychotics led to lower total costs
than poor adherence and persistence. Thus efforts should be made to improve adherence and persis-
tence in patients taking atypical antipsychotics.
KEY WORDS adherence, persistence, antipsychotic, utilization.
(Pharmacotherapy 2015;35(9):813–822) doi: 10.1002/phar.1634

*Address for correspondence: Yawen Jiang, Department

of Clinical Pharmacy, Pharmaceutical Economics and Pol-
icy, Leonard D. Schaeffer Center for Health Policy & Eco-
nomics, School of Pharmacy, University of Southern
California, USC Schaeffer Center, Verna & Peter Dauterive
Hall (VPD), 635 Downey Way, Los Angeles, CA 90089-
3333; e-mail: yawenjia@usc.edu.
Ó 2015 Pharmacotherapy Publications, Inc.
Increasing prescription drug spending is a
leading factor contributing to the growth of
health care costs.1 With the Affordable Care Act
coming into effect, prescription drug spending is
projected to rise even faster in the near future
because of expanded coverage.2 Accordingly,
814 PHARMACOTHERAPY Volume 35, Number 9, 2015
health systems are experiencing increasing bud-
get concerns, and some of them are applying
higher cost sharing to constrain the increase in
prescription costs. For medication recipients,
however, excess cost sharing may lead to under-
use of medications and poor compliance, and
eventually exacerbate health conditions. Hence
excess cost sharing may actually cause increased
medical costs in other aspects such as inpatient
admissions, outpatient services, and emergency
department visits. Therefore, it is important for
health care decision makers to know whether
the overall economic value of good adherence
and persistence is worth the increasing medica-
tions costs.
Endogeneity stands for a correlation existing
between the variable of interest and the error
term, biasing the estimates of associations in
regressions.3 When evaluating the effects of
medication adherence and persistence on health
care utilization, two potential sources of endo-
geneity need to be considered. First, drug use
patterns and outcomes measured in the same
period are likely mutually causal. For example,
patients may have undesirable outcomes because
of poor adherence, thereby incurring a hospital-
ization. Patients may also cease to use a drug
after a hospitalization because they believe the
drug is not effective. Both causal directions are
equally plausible. Several studies assessing the
effects of adherence and persistence on eco-
nomic or clinical outcomes addressed this issue
by evaluating the effects of adherence and per-
sistence in one period on outcomes in a later
period.4–6 Second, compliant and noncompliant
persons are likely unbalanced on unobserved
characteristics. For example, adherent medica-
tion users may have either relatively severe ill-
ness or good health behavior. Both severity of
illness and good health behavior can contribute
to higher health care utilization, yet they are
unobservable to researchers from claims data. As
a result, the estimate of the effect of adherence
on health care utilization will capture the effects
of disease severity or other health behaviors in
addition to adherence. In this case, the unob-
served factors that lead to endogeneity are
known as omitted variable bias. Introducing
instrumental variables (IVs) in econometric anal-
ysis is a standard way to deal with the bias
caused by endogeneity. One study that evaluated
the economic impact of adherence to antidia-
betic medications tried to solve the endogeneity
issues using an IV method.6 However, there is
still an absence of studies on adherence and
persistence to atypical antipsychotics that have
tried to address the endogeneity issues.
Schizophrenia and bipolar disorder are two
distinct psychiatric diagnoses with shared
genetic and pathologic factors,7, 8 both of which
are imposing a great economic burden on health
systems.9–11 Indeed, formulations of mental
illness, including schizophrenia and bipolar dis-
order, based on the Diagnostic and Statistical
Manual of Mental Disorders, Revised Third Edition
(DSM-III-R) define diagnoses by a set of symp-
tom criteria.12 However, emerging conceptual-
izations such as the National Institute of Mental
Health’s Research Domain Criteria Project sug-
gest a spectrum of mental disorders instead of
several categories of mental disorders with clear
cutoffs.13 In particular, the overlap of genetic
variation between schizophrenia and bipolar dis-
order was the highest among major mental dis-
orders.14 As such, both schizophrenia and
bipolar disorder population were included in the
current analyses. Atypical antipsychotics have
gained substantial popularity for the treatment
for schizophrenia and bipolar disorder since the
1990s. One of the major goals of treatment for
schizophrenia and bipolar disorder is to achieve
continuous relief from symptoms.15–17 Accord-
ingly, adherence and persistence are critical to
assess the effectiveness of atypical antipsychotic
treatment.15, 18 Several studies have attempted
to quantify the impact of noncompliance with
antipsychotics on costs and health care utiliza-
tion. Although some studies produced results
that were favorable with good adherence, other
studies produced neutral or disadvantageous evi-
dence. One study19 found that partial compli-
ance with antipsychotics was associated with an
increased risk of hospitalization compared with
full compliance but was not associated with sig-
nificantly different costs. Another report20
showed that early nonadherence (time to discon-
tinuation less than 90 days) to antipsychotics
increased hospitalizations and costs. Two other
studies21, 22 also found that nonadherence
increased total costs. One group of authors23
specifically examined atypical antipsychotics and
reported a positive association between noncom-
pliance and risk of hospitalization. In contrast,
another study24 reported a significant decrease
in total costs associated with nonadherence to
antipsychotics compared with good adherence.
Even more disparate, yet another report25 did
not find any significant association between non-
adherence and rate of hospitalization. However,
none of the previous studies tried to factor out
 ECONOMIC IMPACT OF ADHERENCE TO ATYPICAL ANTIPSYCHOTICS Jiang and Ni
potential bias due to either of the two endogene-
ity issues described earlier. The lack of correc-
tion of endogeneity is perhaps the reason for
inconsistent implications across studies.
The purpose of this study was to examine the
impact of adherence and persistence on health
care costs and risk of hospitalization by taking
the two sources of endogeneity into considera-
tion. We expected that individuals with
schizophrenia or bipolar disorder who were
adherent to and persistent with atypical antipsy-
chotic treatment would have lower total direct
medical costs.
Methods
Data Source and Sample Selection
In this retrospective cohort study, data were
obtained from the medical and pharmacy
claims databases of the Humana health care
insurance plan (~14 million covered lives)
between January 2007 and June 2013. The
Humana databases contain information on diag-
noses in the International Classification of
Diseases, Ninth Revision, Clinical Modification
(ICD-9-CM) format and procedures. They also
include prescription fill information, covering
the National Drug Codes, days of supply, pay-
ment amounts by insurers and beneficiaries, as
well as dates of services for all claims. Demo-
graphic information (age, sex, race/ethnicity,
three-digit ZIP code, and geographic areas) and
enrollment records are also available. The
enrollees in the managed care organization
comprised both privately insured and Medicare
beneficiaries. This study was approved by the
University of Southern California institutional
review board.
Figure 1. Schematic of the observation period for patients included in the study. The observation period consisted of a pre-
index washout period of 180 days, during which there was no use of any atypical antipsychotics, and a post-index period of
720 days in the health care insurance plan databases, during which adherence and persistence were measured in the first
360 days, and outcomes were measured in the second 360 days.
815
Patients in the database with a diagnosis of
schizophrenia (ICD-9-CM code 295.xx) or bipo-
lar disorder (ICD-9-CM code 296.4x–296.8x) at
any time during the period covered by the data
and who had a prescription for noninjectable
atypical antipsychotics—aripiprazole, asenapine,
clozapine, iloperidone, lurasidone, olanzapine,
paliperidone, quetiapine, risperidone, or ziprasi-
done—were identified and selected. The index
date was defined as the date on which the
patient first filled one of these agents. In addi-
tion, the cohort was restricted to patients who
had a pre-index washout period of at least
180 days during which there was no use of any
atypical antipsychotics and had a post-index
period of at least 720 days in the databases.
Only individuals with continuous enrollment
during the observation period were included.
The study period for each patient consisted of a
180-day pre-index period for washing out any
previous atypical antipsychotics, a first 360-day
post-index period (first year), and a second 360-
day post-index period (second year) (Figure 1).
Adherence and persistence were measured in the
first year, and outcomes were measured in the
second year.
Measures
To exclude potential bias due to mutual causal-
ity between drug use patterns and health care
utilization, we investigated the effects of adher-
ence and persistence measured in the first year on
outcomes measured in the second year. The
outcomes examined were all-cause total health
care costs, medication costs, medical services
costs, and inpatient admissions. Costs were
adjusted to 2013 U.S. dollars using the medical
care component of the Consumer Price Index.26
816 PHARMACOTHERAPY Volume 35, Number 9, 2015
The measure of adherence was operationalized
by the proportion of days covered (PDC), with
the PDC numerator the number of days covered
by all atypical antipsychotic prescription fills
during the first year27, 28 and the denominator
360 days. A property of PDC is that it avoids
double counting of overlap days between two
fills. Persistence was measured using time from
initiation to discontinuation of the index drug
therapy. It was defined as the number of days
since the index date until a 15-day or longer gap
between two fills (the end of supply of a fill and
the fill date of the next fill) for the index
drug.29–32 The index drug was the first observed
atypical antipsychotic in the study period for
each patient. This measure was right censored to
be as long as the 360-day measurement period.
A patient was defined as adherent if the PDC
was 80% or higher23, 24 and was considered per-
sistent if the time from initiation to discontinua-
tion was at least 360 days. Adherence
emphasizes the percentage of doses taken,
whereas persistence stresses the duration of ther-
apy.33 The effects of adherence and persistence
were analyzed separately.
Baseline characteristics that were compared
between adherent and nonadherent patients and
between persistent and nonpersistent patients
included the following: demographic variables;
information in the first year such as health care
utilization, diagnoses, and index atypical
antipsychotic drug; and schizophrenia and bipo-
lar disorder diagnoses. Demographic variables
consisted of age, sex, race (white vs nonwhite),
and benefit type (commercial vs Medicare).
Health care utilization consisted of total costs,
medications costs, medical services costs, and
hospitalizations. Diagnoses included major
depression, substance use, and the Charlson
Comorbidity Index (CCI).34 Schizophrenia and
bipolar disorder diagnoses were not limited to
the first year. Instead, these variables were based
on diagnoses at any time in the databases.
A long list of covariates in addition to the
adherent or persistent dummy variable was con-
trolled in multivariate analyses. These covariates
included demographic variables; type of the
index atypical antipsychotic; health care utiliza-
tion in the first year—total costs and indicators
for emergency department visit and hospitaliza-
tion; diagnoses in the first year—CCI, asthma,
epilepsy, hyperlipidemia, major depression, sub-
stance use, and suicide attempt; use of other
drugs in the first year—antidiabetic agents,
diuretics, hypnotics, antihypertensive agents,
ophthalmic agents, cardiotonic agents, cardiovas-
cular agents, antidepressants, antihyperlipidemic
agents, antiparkinson agents, psychotherapeutic
and neurologic agents, antimanic agents,
antianginal agents, anxiety medications, typical
antipsychotics, and use of other atypical antipsy-
chotic (due to switch or augmentation); a
dummy variable for bipolar disorder diagnosis;
and dummy variables for time and state fixed
effects. The diagnoses and use of other drugs
might be associated with both adherence or per-
sistence and health care utilization35, 36 and
were therefore included. Asenapine, clozapine,
iloperidone, and lurasidone were grouped as
“other atypicals” due to their small sample sizes.
Data Analysis
All analyses were conducted by using SAS,
v.9.2 (SAS Institute Inc., Cary, NC) and Stata,
v.12 (Stata Corp., College Station, TX) statistical
software. The significance level threshold was
set at p<0.05. Descriptive statistics were
obtained for the baseline characteristics, and all
outcomes, including frequency for categorical
variables and means and SDs for continuous
variables. Analysis of variance and v2 tests were
used to compare adherent and nonadherent
patients as well as persistent and nonpersistent
patients. In addition, costs were described by
using medians and interquartile ranges, and
were compared by using Kruskal-Wallis tests.
To reduce potential bias due to the second
source of endogeneity, omitted variable bias, IVs
were used to estimate the effects of being adher-
ent and being persistent on costs and hospital-
izations.3, 37 IVs are variables that correlate with
the treatment of interest but do not directly
affect the outcome.3, 37 Naive regressions, or
regressions without using IVs to correct endo-
geneity, were also performed. Two-stage least
squares were used to analyze costs and recursive
bivariate probit (BVP) models were used to ana-
lyze the risk of hospitalization. Numerous
econometric studies have shown that BVP mod-
els can generate consistent estimates using valid
IVs when both the endogenous regressor and the
regressand are dichotomous.38–41 Ordinary least
squares (OLS) and probit models were used as
naive regressions for costs and hospitalizations.
For BVP and probit models, sample average
marginal effects are reported, and standard
errors were obtained by using the delta method.
Two IVs were used in all IV regressions in the
study. The IVs were the reimbursement rate of
 ECONOMIC IMPACT OF ADHERENCE TO ATYPICAL ANTIPSYCHOTICS Jiang and Ni
atypical antipsychotic fills and an indicator for
whether the first observed atypical antipsychotic
fill was a mail order. The reimbursement rate is
a proxy of cost sharing and has been shown to
affect compliance.42, 43 Because the reimburse-
ment rate of a certain drug in a given insurance
policy was determined exogenously, it should be
independent of individual conditions or other
clinical characteristics. Thus it should not affect
the outcomes except through affecting medica-
tion use. In this study, the mean reimbursement
rate was calculated at the aggregate level of
3-digit ZIP code, benefit type, plan type, year, and
generic versus brand name. Mail-order prescrip-
tion filling has been shown to increase adherence
in relation to conventional filling,44, 45 yet it
should not directly affect utilization in ways other
than affecting drug use. The IVs were subject to
tests of validity (Appendix S1).
Results
Baseline Patient Characteristics
The sample selection process identified 32,374
patients (Appendix S1, Figure A1); of these,
11,642 (36.0%) were adherent and 17,772
(54.9%) were persistent. Table 1 lists the demo-
graphic characteristics and first-year measures of
health care use and health conditions. Overall,
baseline characteristics of adherent patients ver-
sus nonadherent patients, as well as persistent
patients versus nonpersistent patients, were
unbalanced (significantly different).
Unadjusted Analysis
Table 2 presents the results from the unad-
justed analyses. During the second year, adher-
ent patients had significantly higher mean
medication costs ($9045 [SD $7616] vs $3971
[SD $5255], p<0.001) and nonsignificantly
lower medical services costs ($9780 [SD
$23,853] vs $10,170 [SD $26,214], p=0.185),
resulting in significantly higher total costs
($18,825 [SD $25,894] vs $14,141 [SD
$27,388], p<0.001) than nonadherent patients.
Adherent patients had a nonsignificantly lower
rate of hospitalization (52.6% vs 53.6%,
p=0.096). Persistent patients had significantly
higher medication costs ($7519 [SD $7026] vs
$3698 [SD $5523], p<0.001) and lower medical
services costs ($9600 [SD $22,267] vs $10,554
[SD $28,731], p<0.001) than nonpersistent
patients in the second year. Total costs ($17,119
817
[SD $24,202] vs $14,252 [SD $19,986], p<0.001)
were significantly higher among persistent
patients. Rates of hospitalization (53.4% vs
53.1%, p=0.503) in the second year did not sig-
nificantly differ between the persistent and non-
persistent groups.
Adjusted Analysis
Table 3 shows the naive regression results and
IV results. OLS estimates showed that being
adherent was related to increased total costs by
$3071 (95% confidence interval [CI] 2228–
3914), increased medication costs by $4312
(95% CI 4159–4465), and decreased medical
services costs by $1240 (95% CI 2072 to
409). Probit model showed that being adher-
ent did not impact the risk of hospitalization
(95% CI 1 to 1). IV estimates indicated that
being adherent was associated with decreased
total costs by $19,497 (95% CI 35,172 to
3822), increased medication costs by $8194
(95% CI 4654–11,734), and decreased medical
services costs by $27,664 (95% CI 43,348 to
11,981). The directions of the effect estimates
of being adherent on total costs were different
across OLS and IV models. Also, the BVP model
estimated that being adherent was associated
with a reduced second-year hospitalization risk
by 27% (95% CI 33 to 22).
According to the OLS results, being persistent
was related to increased total costs by $3384
(95% CI 2602–4164), increased medication
costs by $3705 (95% CI 3567–3843), and only
nonsignificantly decreased medical services
costs by $319 (95% CI 1087 to 448). The
probit model indicated that being persistent
was associated with a significantly increased
risk of hospitalization by 2% (95% CI 1–3%).
According to IV results, being persistent was
associated with decreased total costs by $23,927
(95% CI 45,057 to 2798), increased medica-
tion costs by $10,278 (95% CI 5444–15,311),
and decreased medical services costs by
$34,178 (95% CI 55,936 to 12,420). The
BVP model demonstrated that being persistent
was associated with a nonsignificantly decreased
risk of hospitalization by 5% (95% CI 18
to 8).
Results of the tests of the validity of IVs and
endogeneity of adherent and persistent dummy
variables are shown in Appendix S1, Table A1.
Overall, the tests demonstrate that endogeneity
existed in most naive regressions, IVs were not
weak IVs, and IVs were exogenous.
 818

Table 1. Descriptive Statistics of Demographic and Baseline Characteristicsa

Adherent group Nonadherent group Persistent group Nonpersistent group
Characteristic (n=11,642) (n=20,732) p value (n=17,772) (n=14,602) p value
Age, yrs, mean  SD 51.5  15.1 50.7  15.6 < 0.001 51.8  14.8 50.0  16.3 < 0.001
Male 4660 (40.0) 8314 (40.1) 0.895 7141 (40.2) 5833 (40.0) 0.668
White 7608 (65.4) 12,838 (61.9) < 0.001 11,671 (65.7) 8775 (60.1) < 0.001
Medicare beneficiaries 9839 (84.5) 16,673 (80.4) < 0.001 15,261 (85.9) 11,251 (77.1) < 0.001
Costs, $, mean  SD
First-year total costs 19,805  23,209 15,730  26,666 < 0.001 17,201  23,401 17,187  27,947 0.961
First-year medication costs 10,352  7379 4636  4537 < 0.001 8128  6764 4942  5298 < 0.001
First-year medical services costs 9453  21,259 11,094  25,610 < 0.001 9074  21,734 12,245  26,693 < 0.001
Costs, $, median (IQR)
First-year total costs 13,148 (15,006) 7893 (12,932) < 0.001 10,598 (13,773) 8925 (15,192) < 0.001
First-year medication costs 8767 (7814) 3631 (3640) < 0.001 6274 (6882) 3595 (4182) < 0.001
First-year medical services costs 2267 (8916) 3208 (10,434) < 0.001 2108 (8060) 3692 (12,149) < 0.001
First-year CCI score 0.50 (1.19) 0.55 (1.26) < 0.001 0.49 (1.16) 0.59 (1.31) < 0.001
First-year hospitalization 6146 (52.8) 11,783 (56.8) < 0.001 9303 (52.4) 8626 (59.1) < 0.001
First-year major depression 2625 (22.6) 5757 (27.8) < 0.001 3955 (22.3) 4427 (30.3) < 0.001
First-year substance use 2408 (20.7) 5485 (26.5) < 0.001 3770 (21.2) 4123 (28.2) < 0.001
Schizophrenia diagnosis at any timeb 5654 (48.6) 7421 (35.8) < 0.001 8305 (46.7) 4770 (32.7) < 0.001
Bipolar disorder diagnosis at any timeb 8934 (76.7) 17,365 (83.8) < 0.001 13,832 (77.8) 12,467 (85.4) < 0.001
Index drugc < 0.001 < 0.001
Aripiprazole 1760 (15.1) 4478 (21.6) < 0.001 2821 (15.9) 3417 (23.4) < 0.001
Olanzapine 1508 (13.0) 2531 (12.2) 0.052 2236 (12.6) 1803 (12.4) 0.526
Paliperidone 83 (0.7) 244 (1.2) < 0.001 98 (0.6) 229 (1.6) < 0.001
Quetiapine 4081 (35.1) 7277 (35.1) 0.933 6623 (37.3) 4735 (32.4) < 0.001
Risperidone 2776 (23.8) 4290 (20.7) < 0.001 4208 (23.7) 2858 (19.68) < 0.001
Ziprasidone 1143 (9.8) 1751 (8.5) < 0.001 1470 (8.3) 1424 (9.8) < 0.001
Other atypical antipsychoticsd 291 (2.5) 161 (0.9) < 0.001 316 (1.8) 136 (0.9) < 0.001
PHARMACOTHERAPY Volume 35, Number 9, 2015

Data are no. (%) of patients unless otherwise specified.

CCI
a
= Charlson Comorbidity Index; IQR = interquartile range; SD = standard deviation.
First 360-day follow-up period characteristics are included in regressions as control variables along with demographic characteristics; therefore, they are evaluated as baseline characteristics in
this table.
b

c
Schizophrenia diagnosis and bipolar disorder diagnosis at any time between January 2007 and June 2013 were used.
d
Refers to the drug of the first observed atypical antipsychotic use; the first fill date of this drug is the index date.
Includes asenapine, clozapine, iloperidone, and lurasidone.
 ECONOMIC IMPACT OF ADHERENCE TO ATYPICAL ANTIPSYCHOTICS Jiang and Ni
Table 2. Unadjusted Comparison of Second-Year Outcomes Between the Adherent and Nonadherent Groups, and Between
the Persistent and Nonpersistent Group
Nonadherent
Outcome Adherent group group
Costs, $, mean  SD
Second-year total costs 18,825  25,894 14,141  27,388
Second-year 9045  7616 3971  5255
medication costs
Second-year medical 9780  23,853 10,170  26,214
services costs
Costs, $, median (IQR)
Second-year total costs 11,750 (15,396) 6557 (12,721)
Second-year medication 7375 (8310) 2712 (4194)
costs
Second-year medical 2133 (8511) 2443 (9100)
services costs
Second-year hospitalization 6128 (52.6) 11,112 (53.6)
(no. [%])
IQR = interquartile range; SD = standard deviation.
Discussion
This retrospective cohort study investigated
the effects of adherence to and persistence with
atypical antipsychotics in the first year after ini-
tiating any type of atypical antipsychotic therapy
on health care costs and risk of hospitalization
in the second year. Although several studies
explored this issue with respect to antipsychotics,
or even specifically atypical antipsychotics, none
of the studies addressed potential bias due to
mutual causality and omitted variable bias.
Implications from previous studies on this topic
were not completely in line with each other,
perhaps because of lacking endogeneity correc-
tion. Among those studies, one19 did not find
significant impacts of adherence on total costs
and consequently pointed out that it was possi-
ble that the effects of compliance were altered
substantially by patient severity (effect modifica-
tion), which could not be measured directly. To
our knowledge, our study is the first to obtain
estimates of the impacts of adherence to and
persistence with atypical antipsychotics on
health care costs and utilization by addressing
endogeneity issues.
According to our findings, being adherent and
being persistent were both expected to be
related to increased drug costs and decreased
medical services costs. At the margin, the
decreased medical services costs offset the
increased drug costs and resulted in a net sav-
ings on total health care costs. This is consistent
with previous evidence that medication expendi-
ture leads to savings on medical services expen-
diture.6, 46 These results suggest that poor
819
Nonpersistent
p value Persistent group group p value
< 0.001 17,119  24,102 14,252  29,986 < 0.001
< 0.001 7519  7026 3698  5523 < 0.001
0.185 9600  22,267 10,554  28,731 < 0.001
< 0.001 9986 (14,673) 6330 (13,537) < 0.001
< 0.001 5559 (7113) 2139 (4026) < 0.001
< 0.001 2114 (8399) 2601 (9460) < 0.001
0.096 9494 (53.4) 7746 (53.1) 0.503
adherence or persistence possibly leads to condi-
tions or events that could probably be more
expensive than medication costs. Clinical prac-
tice should incorporate interventions to improve
adherence and persistence to better manage
schizophrenia or bipolar disorder. Future studies
can focus on examining the cost-effectiveness of
adherence improvement programs. In addition,
policymakers and third-party payers should be
cautious about using cost sharing as a measure
to restrain costs because such a policy may actu-
ally lead to suboptimal adherence or persis-
tence,47 thereby causing undesirable clinical
outcomes and high overall costs. Medication
users may choose an adherence level by evaluat-
ing the trade-off between health improvement
and out-of-pocket costs. Hence they may choose
an imperfect adherence level if a marginal unit
of drug use fails to compensate additional costs.
However, this process may result in increased
total costs, which is a negative externality to the
health system. Therefore, it is important that
payers and providers use an appropriate cost-
sharing level to mitigate medication users’ feel-
ings of loss due to out-of-pocket costs.
The diagnostic statistics rejected the hypothesis
that the IVs were weak. Even so, the IV estimates
turned out to be less efficient than the naive esti-
mates. As such, the point estimates of the effects
may not be accurate. However, the lower (or
upper) bounds of the CIs can provide conserva-
tive estimates at the minimum. Adherent patients
and persistent patients had at least $3822 and
$2798 lower total costs, respectively. Programs
aiming at improving adherence and persistence,
820 PHARMACOTHERAPY Volume 35, Number 9, 2015

such as adjunctive psychotherapy,48 can use these

( 55,936 to 12,420)
( 45,057 to 2798)
amounts as conservative thresholds of spending.
Using IVs is critical to this study. Although

( 0.18 to 0.08)
(5444–15,311)
we analyzed first-year medication use and sec-
IV methodb
ond-year outcomes to eliminate the effect of
mutual causality, we still obtained contrary esti-
Table 3. Effects of Being Adherent and Persistent on Costs and Hospitalizations Estimated with Naive Regressions and Instrumental Variable Methods

mates from naive and IV estimates. By checking

naive regressions results only, investigators
*

10,278***
34,178***
0.05
23,927

would have implied that being adherent and

Persistent

being persistent are both associated with higher

total costs, and being persistent is related to a
higher risk of hospitalization. Such results were
contradictory to IV estimates and intuition.
( 1087 to 448)

However, it is noteworthy that the IVs appeared

(2603–4164)
(3567–3843)
Naive regressiona

(0.01–0.03)

to be only marginally exogenous (p=0.048) in

estimating the effect of persistence on medical
services costs. This result may undermine the
validity of the estimated effect of persistence on
***

3705***

0.02***
319

medical services costs. Better IVs in future stud-

3384

ies are necessary.

This study has several limitations. First, as
mentioned earlier, the efficiency of the IV esti-
( 43,348 to 11,981
( 35,172 to 3822)

mates was not optimal. Second, the measures of

( 0.33 to 0.22)

adherence and persistence were based on records

(4654–11,734)

of fills. The existence of a fill does not equal the

IV methods refers to two-stage least squares for costs models and bivariate probit model for hospitalization.
IV methodb

consumption of a fill. Therefore, the measures

Naive regression refers to ordinary least squares for costs models and probit model for hospitalization.

almost certainly overestimate the actual values,

as do all measures of medication use using retro-
Estimates of change in probability of hospitalization in the second follow-up year are reported.

spective databases. Furthermore, the data source

*

8194***
27,664***
0.27***
19,497

for our study did not contain Medicaid benefi-

Adherent

ciaries. This may restrict the generalizability of

the implications from the study. This is an
important issue because Medicaid is the largest
payer for schizophrenia in the United States,49
( 2072 to 409)
( 0.01 to 0.01)

with very different benefit designs.50 Medicaid

(2228–3914)
(4159–4465)
Naive regressiona

beneficiaries with psychiatric disorders may have

different adherence, persistence, and cost pat-
terns. Therefore, the associations between medi-
cation use and costs may also be different.
***

4312***

0.00
1240**

Future studies should examine such associations

3071

among Medicaid beneficiaries.

Data are estimates (95% confidence interval).

Conclusion
medical services costs, $

Both adherence to and persistence with atypi-

medication costs, $

cal antipsychotics are associated with reduced

*p<0.05, **p<0.01, ***p<0.001.
hospitalizationc

medical services costs, which offset increased

total costs, $

medication costs to generate a net savings on

= instrumental variable.

total health care costs. Therefore, policies using

excessive cost sharing of prescription costs to
control health costs may lead to suboptimal
Second-year
Second-year
Second-year
Second-year

adherence and persistence, and consequently

Outcome

increase total costs. Efforts should be made to

control costs and enhance outcomes by improv-
ing adherence and persistence among individuals
IV
b
a

c
 ECONOMIC IMPACT OF ADHERENCE TO ATYPICAL ANTIPSYCHOTICS Jiang and Ni
with schizophrenia or bipolar disorder who are
taking atypical antipsychotics.
References
1. Hartman M, Martin AB, Benson J, Catlin A. National health
spending in 2011: overall growth remains low, but some pay-
ers and services show signs of acceleration. Health Aff (Mill-
wood) 2013;1:87–99.
2. Keehan SP, Sisko AM, Truffer CJ, et al. National health
spending projections through 2020: economic recovery and
reform drive faster spending growth. Health Aff (Millwood)
2011;8:1594–605.
3. Greene WH. Econometric analysis. Upper Saddle River, NJ:
Pearson Education; 2012.
4. Ho PM, Rumsfeld JS, Masoudi FA, et al. Effect of medica-
tion nonadherence on hospitalization and mortality among
patients with diabetes mellitus. Arch Intern Med
2006;17:1836–41.
5. Subotnik KL, Ventura J, Gretchen-Doorly D, et al. The impact
of second-generation antipsychotic adherence on positive and
negative symptoms in recent-onset schizophrenia. Schizophr
Res 2014;1:95–100.
6. Encinosa WE, Bernard D, Dor A. Does prescription drug
adherence reduce hospitalizations and costs? The case of dia-
betes. Adv Health Econ Health Serv Res 2010;22:151–73.
7. Konopaske GT, Lange N, Coyle JT, Benes FM. Prefrontal cor-
tical dendritic spine pathology in schizophrenia and bipolar
disorder. JAMA Psychiatry 2014;12:1323–31.
8. Bramon E, Sham P. The common genetic liability between
schizophrenia and bipolar disorder: a review. Curr Psychiatry
Rep 2001;4:332–7.
9. Dilsaver SC. An estimate of the minimum economic burden of
bipolar I and II disorders in the United States: 2009. J Affect
Disord 2011;1:79–83.
10. Wu EQ, Birnbaum HG, Shi L, et al. The economic burden of
schizophrenia in the United States in 2002. J Clin Psychiatry
2005;9:1122–9.
11. McEvoy JP. The costs of schizophrenia. J Clin Psychiatry
2007; 68 Suppl (14):4–7.
12. Cooper AM, Michels R. Diagnostic and Statistical Manual of
Mental Disorders, revised (DSM-III-R). American Journal of Psy-
chiatry 1988;10:1300–1.
13. Etkin A, Cuthbert B. Beyond the DSM: development of a
transdiagnostic psychiatric neuroscience course. Acad Psychia-
try 2014;2:145–50.
14. Lee SH, Ripke S, Neale BM, et al. Genetic relationship
between five psychiatric disorders estimated from genome-
wide SNPs. Nat Genet 2013;9:984–94.
15. Rascati KL, Richards KM, Ott CA, et al. Adherence, persis-
tence of use, and costs associated with second-generation
antipsychotics for bipolar disorder. Psychiatr Serv
2011;9:1032–40.
16. Dubovsky SL, Dubovsky AN. Maintenance treatment of bipo-
lar disorder with ziprasidone in adjunctive use with lithium or
valproate. Clin Med Insights Ther 2012;4:1–8.
17. Ascher-Svanum H, Zhu B, Faries D, Landbloom R, Swartz M,
Swanson J. Time to discontinuation of atypical versus typical
antipsychotics in the naturalistic treatment of schizophrenia.
BMC Psychiatry 2006;1:8.
18. Lehman AF, Lieberman JA, Dixon LB, et al. Practice guideline
for the treatment of patients with schizophrenia. Am J Psychia-
try 2004;2(Suppl):1–56.
19. Eaddy M, Grogg A, Locklear J. Assessment of compliance with
antipsychotic treatmentand resource utilization in a Medicaid
population. Clin Ther 2005;2:263–72.
20. Offord S, Lin J, Mirski D, Wong B. Impact of early nonadher-
ence to oral antipsychotics on clinical and economic outcomes
among patients with schizophrenia. Adv Ther 2013;3:286–97.
21. Knapp M, King D, Pugner K, Lapuerta P. Non-adherence to
antipsychotic medication regimens: associations with resource
use and costs. Br J Psychiatry 2004;6:509–16.
821
22. Becker M, Young MS, Ochshorn E, Diamond R. The relation-
ship of antipsychotic medication class and adherence with
treatment outcomes and costs for Florida Medicaid beneficia-
ries with schizophrenia. Adm Policy Ment Health Ment Health
Serv Res 2007;3:307–14.
23. Ward A, Ishak K, Proskorovsky I, Caro J. Compliance with
refilling prescriptions for atypical antipsychotic agents and its
association with the risks for hospitalization, suicide, and
death in patients with schizophrenia in Quebec and Saskatche-
wan: a retrospective database study. Clin Ther 2006;11:1912–
21.
24. Gilmer TP, Dolder CR, Lacro JP, et al. Adherence to treat-
ment with antipsychotic medication and health care costs
among Medicaid beneficiaries with schizophrenia. Am J Psy-
chiatry 2004;4:692–9.
25. Sajatovic M, Valenstein M, Blow FC, Ganoczy D, Ignacio RV.
Treatment adherence with antipsychotic medications in bipo-
lar disorder. Bipolar Disord 2006;3:232–41.
26. Bureau of Labor Statistics. Consumer Price Index—All Urban
Consumers. Available from http://data.bls.gov/cgi-bin/survey-
most. Accessed May 16, 2014.
27. Hess LM, Raebel MA, Conner DA, Malone DC. Measurement
of adherence in pharmacy administrative databases: a proposal
for standard definitions and preferred measures. Ann Pharma-
cother 2006;7–8:1280–8.
28. Nau DP. Proportion of days covered (PDC) as a preferred
method of measuring medication adherence. Springfield, VA:
Pharmacy Quality Alliance, 2012.
29. Chen L, McCombs JS, Park J. The impact of atypical antipsy-
chotic medications on the use of health care by patients with
schizophrenia. Value Health 2008;1:34–43.
30. Weiden PJ. Partial compliance and risk of rehospitalization
among California Medicaid patients with schizophrenia. Psy-
chiatr Serv 2004;8:886–91.
31. Horvitz-Lennon M, Donohue JM, Lave JR, Alegria M, Nor-
mand SL. The effect of race-ethnicity on the comparative
effectiveness of clozapine among Medicaid beneficiaries. Psy-
chiatr Serv 2013;3:230–7.
32. Jiang Y, Ni W, McGinnis J. Comparison of health services use
associated with ziprasidone and olanzapine among schizophre-
nia and bipolar disorder patients in the USA. Clin Drug Inves-
tig 2014;7:491–9.
33. Cramer JA, Roy A, Burrell A, et al. Medication compliance
and persistence: terminology and definitions. Value Health
2008;1:44–7.
34. Quan H, Sundararajan V, Halfon P, et al. Coding algorithms
for defining comorbidities in ICD-9-CM and ICD-10 adminis-
trative data. Med Care 2005;11:1130–9.
35. Gianfrancesco F, Pesa J, Wang R-H, Nasrallah H. Assessment
of antipsychotic-related risk of diabetes mellitus in a Medicaid
psychosis population: sensitivity to study design. Am J Health
Syst Pharm 2006;5:431–41.
36. Qiu Y, Fu AZ, Liu GG, Christensen DB. Healthcare costs of
atypical antipsychotic use for patients with bipolar disorder in
a Medicaid programme. Appl Health Econ Health Policy
2010;3:167–77.
37. Cameron AC, Trivedi PK. Microeconometrics: methods
and applications. New York, NY: Cambridge University Press;
2005.
38. Chiburis RC, Das J, Lokshin M. A practical comparison of the
bivariate probit and linear IV estimators. Econ Lett
2012;3:762–6.
39. Bhattacharya J, Goldman D, McCaffrey D. Estimating probit
models with self-selected treatments. Stat Med 2006;3:389–
413.
40. Clarke PS, Windmeijer F. Instrumental variable estimators for
binary outcomes. J Am Stat Assoc 2012;500:1638–52.
41. Arendt JN, Holm A. Probit Models with Binary Endogenous
Regressors, June 2006. Available from http://www.econ.ku.dk/
cam/wp0910/wp0406/2006-06.pdf. Accessed Nov 18, 2014.
42. Eaddy MT, Cook CL, O’Day K, Burch SP, Cantrell CR. How
patient cost-sharing trends affect adherence and outcomes: a
literature review. Pharm Ther 2012;1:45.
822 PHARMACOTHERAPY Volume 35, Number 9, 2015
43. Dor A, Encinosa W. How does cost-sharing affect drug pur-
chases? Insurance regimes in the private market for prescrip-
tion drugs. J Econ Manag Strategy 2010;3:545–74.
44. Duru OK, Schmittdiel JA, Dyer WT, et al. Mail-order phar-
macy use and adherence to diabetes-related medications. Am J
Manag Care 2010;1:33.
45. Zhang L, Zakharyan A, Stockl KM, Harada AS, Curtis BS,
Solow BK. Mail-order pharmacy use and medication adher-
ence among Medicare Part D beneficiaries with diabetes. J
Med Econ 2011;5:562–7.
46. Shang B, Goldman DP. Prescription drug coverage and elderly
Medicare spending (NBER Working Paper Series No. 13358).
Cambridge, MA, USA: National Bureau of Economic Research,
2007. Available from http://www.nber.org/papers/w13358.
Accessed Nov 18, 2014.
47. Reed M, Brand R, Newhouse JP, Selby JV, Hsu J. Coping with
prescription drug cost sharing: knowledge, adherence, and
financial burden. Health Serv Res 2008;2:785–97.
48. Wang C-C, Maciejewski M, Rao J, Sheitman B, Sleath B, Far-
ley J. Examining the relationship between adjunctive psy-
chotherapy use and antipsychotic persistence and
hospitalization. Adm Policy Ment Health Ment Health Serv Res
2014;5:598–607.
49. Karve SJ, Panish JM, Dirani RG, Candrilli SD. Health care
utilization and costs among Medicaid-enrolled patients with
schizophrenia experiencing multiple psychiatric relapses.
Health Outcomes Res Med 2012;4:e183–94.
50. West JC, Wilk JE, Rae DS, et al. Medicaid prescription drug
policies and medication access and continuity: findings from
ten states. Psychiatr Serv 2009;5:601.
Supporting Information
The following supporting information is available in the online
version of this paper:
Appendix S1. Validity of Instrumental Variables (IVs).
Copyright of Pharmacotherapy is the property of Wiley-Blackwell and its content may not be
copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for
individual use.