University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

UNDERSTANDING PRIMARY NONADHERENCE TO MEDICATIONS AND ITS ASSOCIATED
HEALTHCARE OUTCOMES: A RETROSPECTIVE ANALYSIS OF ELECTRONIC MEDICAL
RECORDS IN AN INTEGRATED HEALTHCARE SETTING

by

Janet Shin

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

August 2012

Copyright 2012 Janet Shin
ii

TABLE OF CONTENTS

List of Tables iv
List of Figures v
Abstract vii
Chapter 1: Introduction 1
Chapter 2: Statement of research questions 2
Chapter 3: Background and Literature Review
3.1 Significance of primary nonadherence 3
3.2 Extent of primary nonadherence 3
3.3 Reasons for primary nonadherence 4
3.4 Healthcare outcomes of primary nonadherence 6
3.5 Health policy implications 7
3.6 Identified issues in conducting primary nonadherence research
3.6.1 Measuring primary nonadherence 7
3.6.2 Endogeneity of primary nonadherence variable 9
3.6.2.1 Instrumental Variables 9
3.6.2.2 Fixed Effects and Random Effects 10
3.6.2.3 Difference-in-difference Estimator 10
Chapter 4: Factors associated with primary nonadherence to chronic and acute
medications (Phase I)

4.1 Objectives 12
4.2 Methods
4.2.1 Study design, setting, and data source 12
4.2.2 Study population 13
4.2.3 Study variables
4.2.3.1 Study outcomes 15
4.2.3.2 Patient, prescriber, and prescription characteristics 15
4.2.4 Statistical analysis 16
4.3 Results
4.3.1 Rates of primary nonadherence 17
4.3.2 Unadjusted comparison between the primary adherent and
nonadherent
18
4.3.3 Multivariable logistic results 19
4.3.3.1 Patient characteristics 19
4.3.3.2 Prescriber characteristics 21
4.3.3.3 Prescription characteristics 22
4.3.4 Time to first fill analysis 23
4.4 Discussion 24
4.5 Conclusions 27
Chapter 5: Outcomes associated with primary nonadherence to diabetes and
cholesterol medications (Phase II)

5.1 Objectives 28
5.2 Methods
5.2.1 Study design, setting, and data source 28
5.2.2 Study populations 28
iii

5.2.3 Study variables
5.2.3.1 Study covariates 28
5.2.3.2 Study outcomes 30
5.2.4 Statistical analysis
5.2.4.1 Descriptive analysis 31
5.2.4.2 Analysis of Lab Values 32
5.2.4.3 Analysis of event outcomes 32
5.3 Results for cholesterol cohort
5.3.1 Descriptive results 33
5.3.2 Lab values 38
5.3.3 Event outcomes 40
5.4 Results for diabetes cohort
5.4.1 Descriptive results 43
5.4.2 Lab values 49
5.4.3 Event outcomes 51
5.5 Discussion 53
5.6 Conclusions 56
Bibliography 57
iv

LIST OF TABLES

Table 1. Summary of study inclusion/exclusion criteria for Phases I and II

14
Table 2. Results for patient characteristics from multivariable logistic regression
predicting primary nonadherence

20
Table 3. Results for prescriber characteristics from multivariable logistic
regression predicting primary nonadherence

22

Table 4. Results for prescription characteristics from multivariable logistic
regression predicting primary nonadherence

23
Table 5. Healthcare outcomes of Phase II study

31
Table 6. Unadjusted comparison of baseline patient characteristics among
cholesterol adherence groups

36
Table 7. Comparison of LDL testing among cholesterol adherence groups

38
Table 8. Comparison of change in LDL values among cholesterol adherence
groups (n=18,571)

40
Table 9. Comparison of unadjusted rates of event outcomes occurring between 12
and 18 months after index date among cholesterol adherence groups

42
Table 10. Adjusted hazard ratios for event outcomes occurring between 12 and 18
months after index date in cholesterol

43
Table 11. Unadjusted comparison of baseline patient characteristics among
diabetes adherence groups

47
Table 12. Comparison of Ha1c testing among diabetes adherence groups

49
Table 13. Comparison of change in Ha1c (%) values among diabetes adherence
groups

51
Table 14. Comparison of unadjusted rates of event outcomes occurring between
12 and 18 months after index date among diabetes adherence groups

53
Table 15. Adjusted hazard ratios for event outcomes occurring between 12 and 18
months after index date in diabetes

53

v

LIST OF FIGURES

Figure 1. Reasons found in the literature for primary nonadherence

5
Figure 2. Identification and selection of prescriptions in final cohort of Phase I study

17
Figure 3. Primary nonadherence rates by therapeutic drug group

18
Figure 4. Identification of cholesterol study cohort

33
Figure 5. Total days supply during 12-month follow-up period for primary adherent
cholesterol patients

34
Figure 6. Total number of fills dispensed during 12-month follow-up period for primary
adherent cholesterol patients

34
Figure 7. Medication Possession Ratio (MPR) for any cholesterol medication during
12-month follow-up period for primary adherent cholesterol patients

35
Figure 8. Persistence (time to discontinuation) in primary adherent cholesterol
patients

35
Figure 9. Average number of LDL tests per patient by adherence group and months
from index date

39
Figure 10. Average LDL values over time by adherence group (n=18,571)

40
Figure 11. Rate of ER visits per patient over time by adherence group (n=21,291)

41
Figure 12. Rate of hospitalizations per patient over time by adherence group
(n=21,291)

41
Figure 13. Identification of diabetes study cohort

44
Figure 14. Total days supply during 12-month follow-up period for primary adherent
diabetes patients

44
Figure 15. Total number of fills dispensed during 12-month follow-up period for
primary adherent diabetes patients

45
Figure 16. Persistence (time to discontinuation) in primary adherent diabetes patients

46
Figure 17. MPR for any diabetes medication during 12-month follow-up period for
primary adherent diabetes patients

46

vi

Figure 18. Average number of Ha1c tests per patient by adherence group and
months from index date

50
Figure 19. Average Ha1c values over time by adherence group

51
Figure 20. Rate of ER visits per patient over time by adherence group

52
Figure 21. Rate of hospitalizations per patient over time by adherence group

52

vii

ABSTRACT
OBJECTIVES: Phase I objectives were to measure the extent of primary nonadherence (PNA) in
an integrated healthcare system and identify factors significantly associated with PNA to chronic
and acute medications. The objective of phase II was to examine healthcare outcomes
associated with PNA to cholesterol and diabetes medications.
METHODS: This retrospective cohort study was conducted at Kaiser Permanente Southern
California using data from its integrated electronic medical record system. All new prescriptions
for 10 therapeutic drug groups (antiinfectives, analgesics, migraine medications, antidiabetics,
osteoporosis medications, cardiovascular agents, antihyperlipidemics, antiasthmatics,
antidepressants, and anticoagulants) written for patients who satisfied membership and other
criteria were included in phase I. PNA rates were calculated by therapeutic drug group as the
number of prescriptions not filled within 14 days over the total number of prescriptions. A
multivariate logistic regression was used to identify patient, prescriber, and prescription factors
significantly associated with PNA. Patients who filled an antidiabetic or antihyperlipidemic were
eligible for inclusion in Phase II if they were at least 18 years of age and satisfied additional
inclusion and exclusion criteria. A more conservative definition of PNA was applied in phase II as
the failure to fill within 180 days. Descriptive statistics were used to summarize baseline
differences among patient groups and to describe secondary nonadherence and unadjusted rates
of event outcomes. The paired t-test and a multivariate difference-in-difference model were used
to evaluate changes in lab values. A multivariate Cox Proportional Hazards model was used to
measure the effect of PNA and secondary nonadherence on event outcomes of new disease
complications, ER visits, and hospitalization.
RESULTS: In phase I, a total of 569,095 new prescriptions for 398,025 patients satisfied study
criteria. The overall PNA rate was 9.8% but individual PNA rates varied widely by therapeutic
drug group. The effects of patient race, baseline comorbidities, prior fills, and being treatment-
naive on PNA were consistent across drug groups. The strongest predictors were prior fills and
being treatment-naïve. Some factors (mostly prescriber and prescription factors) depended on
viii

whether the treatment was acute or chronic. Phase II included 21,291 cholesterol patients and
10,137 diabetes patients. PNA rates were12.0% and 5.1% for cholesterol and diabetes,
respectively. Secondary nonadherence was relatively worse in cholesterol than diabetes with
about 20% of primary adherent patients only filling the first prescription. Adherent patients were
on average older and sicker than nonadherent patients. Patients who were nonadherent to their
cholesterol or diabetes medication were significantly associated with worse health outcomes in
terms of lab values and ER visits. In cholesterol, the risk for ER visits was significantly higher in
secondary nonadherent patients (HR=1.24, 95% CI: 1.14, 1.34) and the risk of hospitalization
was significantly higher in primary nonadherent patients with prior cholesterol medication use
(HR=1.69, 95% CI: 1.19, 2.41) in comparison to adherent patients. For diabetes, secondary
nonadherent patients were associated with significantly increased risk for ER visits (HR=1.30,
95% CI: 1.11, 1.51).
CONCLUSIONS: The results of this study help clinicians and healthcare decision-makers to
understand the extent of PNA and how PNA varies by therapeutic drug group in an integrated
healthcare setting. Most factors of PNA were consistent across drug groups, but some factors
depended on whether the treatment was acute or chronic. In addition, these results can facilitate
the decision-making process in designing and implementing cost-effective patient interventions to
improve overall adherence to cholesterol and diabetes medications. Future research should
examine the effect of PNA on longer-term health outcomes such as mortality.
OBJECTIVES: Phase I objectives were to measure the extent of primary nonadherence (PNA) in
an integrated healthcare system and identify factors significantly associated with PNA to chronic
and acute medications. The objective of phase II was to examine healthcare outcomes
associated with PNA to cholesterol and diabetes medications.
METHODS: This retrospective cohort study was conducted at Kaiser Permanente Southern
California using data from its integrated electronic medical record system. All new prescriptions
for 10 therapeutic drug groups (antiinfectives, analgesics, migraine medications, antidiabetics,
osteoporosis medications, cardiovascular agents, antihyperlipidemics, antiasthmatics,
ix

antidepressants, and anticoagulants) written for patients who satisfied membership and other
criteria were included in phase I. PNA rates were calculated by therapeutic drug group as the
number of prescriptions not filled within 14 days over the total number of prescriptions. A
multivariate logistic regression was used to identify patient, prescriber, and prescription factors
significantly associated with PNA. Patients who filled an antidiabetic or antihyperlipidemic were
eligible for inclusion in Phase II if they were at least 18 years of age and satisfied additional
inclusion and exclusion criteria. A more conservative definition of PNA was applied in phase II as
the failure to fill within 180 days. Descriptive statistics were used to summarize baseline
differences among patient groups and to describe secondary nonadherence and unadjusted rates
of event outcomes. The paired t-test and a multivariate difference-in-difference model were used
to evaluate changes in lab values. A multivariate Cox Proportional Hazards model was used to
measure the effect of PNA and secondary nonadherence on event outcomes of new disease
complications, ER visits, and hospitalization.
RESULTS: In phase I, a total of 569,095 new prescriptions for 398,025 patients satisfied study
criteria. The overall PNA rate was 9.8% but individual PNA rates varied widely by therapeutic
drug group. The effects of patient race, baseline comorbidities, prior fills, and being treatment-
naive on PNA were consistent across drug groups. The strongest predictors were prior fills and
being treatment-naïve. Some factors (mostly prescriber and prescription factors) depended on
whether the treatment was acute or chronic. Phase II included 21,291 cholesterol patients and
10,137 diabetes patients. PNA rates were12.0% and 5.1% for cholesterol and diabetes,
respectively. Secondary nonadherence was relatively worse in cholesterol than diabetes with
about 20% of primary adherent patients only filling the first prescription. Adherent patients were
on average older and sicker than nonadherent patients. Patients who were nonadherent to their
cholesterol or diabetes medication were significantly associated with worse health outcomes in
terms of lab values and ER visits. In cholesterol, the risk for ER visits was significantly higher in
secondary nonadherent patients (HR=1.24, 95% CI: 1.14, 1.34) and the risk of hospitalization
was significantly higher in primary nonadherent patients with prior cholesterol medication use
x

(HR=1.69, 95% CI: 1.19, 2.41) in comparison to adherent patients. For diabetes, secondary
nonadherent patients were associated with significantly increased risk for ER visits (HR=1.30,
95% CI: 1.11, 1.51).
CONCLUSIONS: The results of this study help clinicians and healthcare decision-makers to
understand the extent of PNA and how PNA varies by therapeutic drug group in an integrated
healthcare setting. Most factors of PNA were consistent across drug groups, but some factors
depended on whether the treatment was acute or chronic. In addition, these results can facilitate
the decision-making process in designing and implementing cost-effective patient interventions to
improve overall adherence to cholesterol and diabetes medications. Future research should
examine the effect of PNA on longer-term health outcomes such as mortality.

1

CHAPTER 1: INTRODUCTION
Healthcare professionals and payers are constantly looking for ways to improve patient
health outcomes while decreasing costs. One possible approach to this challenge is to improve
medication adherence or compliance, which is often defined as the extent to which a patient
takes his medication as prescribed (Haynes et al 1979). Poor medication adherence is common,
with nonadherence rates typically as high as 50% (Haynes et al 1996). Prior research has
suggested that improved medication adherence is associated with better health outcomes and
lower overall health costs for certain chronic diseases such as diabetes, hypertension,
hypercholesterolemia, and congestive heart failure (Sokol et al 2005, Osterberg & Blaschke 2005,
Roebuck et al 2011).
Most adherence research has focused on “secondary nonadherence”, which occurs
when patients do not refill their prescriptions on time or discontinue their medications altogether.
A less-studied form of medication nonadherence is called “primary nonadherence” (PNA), which
occurs when patients fail to pick up a newly prescribed prescription from the pharmacy (Beardon
et al 1993). Incidence, causes, and outcomes of PNA are relatively unknown in comparison to
those of secondary nonadherence. Since most chronic and acute diseases in the U.S. are often
managed by prescription medications, PNA could potentially be a significant factor in determining
healthcare outcomes and costs.

2

CHAPTER 2: STATEMENT OF RESEARCH QUESTIONS
This research aims to fill in the gaps within PNA research by answering the following questions:
1) What are the rates of PNA to chronic and acute prescription medications within an
integrated healthcare system?
2) What patient, prescriber, and prescription characteristics are associated with PNA to
chronic and acute prescription medications?
3) What healthcare outcomes are associated with PNA to medications for chronic
disease (specifically, hyperlipidemia and diabetes type II)?
The research was conducted in two phases. Phase I measured the extent of PNA in an
integrated healthcare system and identified factors significantly associated with filling the first
prescription for chronic and acute medications (Chapter 4). Phase II focused on the cholesterol
and diabetes patient populations from Phase I in order to closely examine healthcare outcomes
associated with PNA to medications for chronic disease (Chapter 5).

3

CHAPTER 3: BACKGROUND AND LITERATURE REVIEW
3.1 Significance of primary nonadherence
PNA is an important healthcare issue because it can potentially contribute to disease
progression, death, and increased health care costs, similar to the effects of secondary
nonadherence (Senst et al 2001; McDonnell and Jacobs 2002; Sokol et al, 2005; Balkrishnan
2005; Osterberg and Blaschke 2005; Cramer et al 2008; Muszbek et al 2008; Roebuck et al
2011). The potential clinical benefit from any prescription medication can only be realized if the
patient is adequately compliant with their medication regimen. Due to the high prevalence in the
U.S. of chronic disease, which is most often managed by prescription medications, the unknown
impact of PNA on healthcare outcomes could potentially be a significant factor in healthcare
outcomes and costs.
There are several potential benefits from improving PNA to medication. First, patients
themselves would benefit from improvements in quality and/or quantity of life gained from the
prevention of disease complications (e.g. amputations, blindness, and cardiovascular events in
diabetes). Second, reductions in the number of outpatient visits, emergency room (ER) visits,
and hospitalizations resulting from improved primary adherence would help to lower overall
healthcare costs (Sokol et al 2005) and ultimately, lower insurance premiums for patients. Last,
overall society would also benefit from increases in worker productivity and in size of the labor
workforce.
In summary, understanding the effects of PNA could benefit patients, 3
rd
party payers,
and overall society. Determining to what extent PNA impacts healthcare outcomes is a
worthwhile issue for U.S. policymakers and insurers, especially during a time when the U.S.
spends 16% of its GDP on healthcare and lags behind other developed countries in terms of life
expectancy and morbidity (OECD Health Data 2008).
3.2 Extent of primary nonadherence
Estimates of PNA rates vary widely in literature from 0.5% to 57.1%. In general, PNA
rates were higher for studies examining patients discharged from the hospital or ER (1.3% to
4

52.4%) compared to those from health care centers (0.5% to 44.4%) or the general population
(1.4% to 29.0%) (Freeman and Guly 1985; Beardon et al 1993; Loong 1999; Ekehdahl and
Mansson 2004; Gadkari and McHorney 2010). Studies in managed care settings tended to have
lower PNA rates ranging from 4.7% to 13% (Lash 1995; Karter et al 2009; Trinacty et al 2009;
Raebel et al 2011). A recent study reported the overall prescription abandonment rate of 3.27%
at a large retail pharmacy chain in the United States (Shrank WH et al 2010).
Estimated PNA rates also vary by disease. In patients with diabetes, PNA rates range
from 4.7% (Karter et al 2009) to 20.4% (Hanko et al 2007). Karter et al (2009) used electronic
medical record (eMR) data for patients in the Kaiser Permanente Northern California Diabetes
Registry (n=27,329) to calculate medication adherence rates. The Hanko et al (2007) study took
place in 14 community pharmacies in Hungary using patient questionnaires related to adherence
and lifestyle (n=142).
Rates of PNA in the literature also vary based on the time allowed for filling prescriptions
before classifying as non-adherent varies greatly among studies, from 2 days to 2 years. PNA
rates were higher on average for studies using time spans of one month or less than studies
using a year or more to define PNA (Gadkari &McHorney, 2010). Karter et al (2009) found that
among their primary adherent diabetic patients, 50% of them filled their prescription on the same
day as it was prescribed, while 90%, 95%, and 99% were dispensed within 6, 12, and 89 days,
respectively.
3.3 Reasons for primary nonadherence
Previous studies have identified reasons why patients do not pick up their initial
prescription medication. Table 1 summarizes commonly reported reasons for PNA according to a
recent narrative systematic review (Gadkari and McHorney 2010). The most common reasons
for PNA are associated with patients’ concerns regarding the medication itself (Pound et al 2005).
This includes patient fears of potential side effects or addiction, reluctance to taking generic
rather than brand name medications, aversion to taking any medications at all, and fear of being
stigmatized with a disease requiring chronic treatment.
5

Figure 1. Reasons found in the literature for primary nonadherence (source: Gadkari &
Mchorney, 2010)

6

Other reasons for PNA are related to patients’ perceived need for the prescription
medication. Patients may feel that taking the medication is unnecessary due to preconceived
notions that the medication will not work or that their condition will improve on its own. Another
major barrier to primary adherence is medication affordability either due to lack of drug benefits or
high copay amounts. Other reasons leading to PNA include transportation problems,
inconvenience, forgetfulness, and lack of patient awareness that a prescription was prescribed for
them (Gadkari and McHorney 2010).
3.4 Healthcare outcomes of primary nonadherence
The outcomes of PNA have been largely ignored in comparison to outcomes associated
with secondary nonadherence. To date, only two studies have been published in the literature
that examine the association between PNA and healthcare outcomes.
Jackevius et al (2008) measured 1-year mortality associated with PNA after acute
myocardial infarction (AMI) in patients identified from an AMI registry in Ontario, Canada.
Unadjusted 1-year mortality rates for patients who filled all (mortality rate of 12.8%), some
(20.5%), or none (30.4%) of their discharge prescriptions suggested an association between PNA
and mortality. Compared to patients who fill all prescriptions, the adjusted 1-year mortality rate
was also higher for both patients who filled some of their medications (OR 1.44; 95% CI 1.15 to
1.79) and none (OR 1.80; 95% CI 1.35 to 2.42).
Ho et al (2010) examined patients who were prescribed clopidogrel upon discharge from
a hospital within 3 large integrated health care systems after drug-eluting stent implantation.
Patients who delayed filling clopidogrel had higher rates of all-cause mortality/MI rates compared
to patients filling the drug on day of discharge (14.2% vs. 7.9%, p<0.001).
Clinical outcomes of patients who are primary nonadherent to chronic medications such
as antidiabetic or antihyperlipidemic drugs are unknown. Patients who never initiate their chronic
prescription may in fact be more problematic and nonadherent to other healthy behaviors (e.g.
diet and exercise) as well. They may have worse outcomes and need more assistance with
managing their health than patients who pick up at least some of their prescriptions. On the other
7

hand, these primary nonadherent patients may be “borderline” sick patients who are relatively
healthier and may decide to use other methods (e.g. diet and exercise) to lower their blood sugar
or cholesterol. A new prescription for diabetes or cholesterol may be the “wake up” call these
patients needed to start taking their health seriously. Understanding which patients are primary
nonadherent and the consequences of PNA will greatly aid clinicians and health policy makers to
decide where resources would be best spent to help patients.
3.5 Health policy implications
The most common reason for PNA is patient concern regarding the medication or the
need for the medication. This likely indicates poor patient-prescriber communication. In addition,
identifying primary nonadherent patients in the clinical setting can be difficult and healthcare
providers overestimate patient adherence and are unlikely to raise the issue of whether the
patient picked up their medications from the pharmacy during routine medical visits (Mushlin and
Appel 1977; Copher et al 2010; Gadkari and McHorney 2010).
Most health policy interventions designed to educate or clinical intervention programs
have been ineffective in improving adherence or patient outcomes (Haynes et al 2002; Haynes et
al 2008). Electronic and in-person interventions either by a pharmacist or at hospital discharge
are more successful at improving adherence (Gatwood and Erickson 2010). Further research is
needed to improve the effectiveness of interventions designed to improve PNA.
Although there are also important health policy implications regarding medication
affordability, little of this research is considered in PNA. Prior research has demonstrated that a
reduction of out-of-pocket costs improves secondary adherence and reduces overall payer costs
(Mahoney 2008; Roebuck et al 2011), but the effect of medication affordability on PNA remains
unclear.
3.6 Identified issues in conducting primary nonadherence research
3.6.1 Measuring primary nonadherence
Until recently, research in PNA has been limited due to limitations associated with typical
claims-based data, i.e. prescription data are only captured once a prescription has been sold and
8

a pharmacy claim has been filed for payment. Examining PNA has become more feasible with
the implementation of electronic prescribing (e-prescribing), which involves the automatic
transmission of prescriptions to pharmacies. However, the use of e-prescribing is not ubiquitous
and accessing this data is usually difficult. Consequently, PNA studies are mostly limited to
select healthcare settings with e-prescribing systems, which may limit the generalizability of study
results.
Self-reported data from mail surveys, telephone surveys, online surveys, or face-to-face
interviews are alternatives to administrative data based on eMR or e-prescribing systems. Data
from self-reports are valuable because they capture data on factors of PNA that are otherwise
unavailable from administrative data, such as the patient’s level of confidence with their
physician. However, patient self-reports are less objective than a retrospective database, subject
to recall bias, and can be time-consuming to collect.
Accuracy in measuring the extent of PNA improves when the risk of outcome
misclassification is minimized. A prescription can be misclassified as nonadherent if it is filled
outside of the network or if patients pay cash or fill their prescription at another pharmacy during a
promotional deal (e.g. $4 generics at Walgreens). These prescriptions will not be submitted for
payment from the insurance company and the prescription will be misclassified as nonadherent
(Karter et al 2009). Even e-prescriptions may become misclassified if the prescription is verbally
changed by the prescriber (e.g. change in drug) after it was ordered and then filled with a different
drug. However, such prescriptions are usually a very small percentage of all e-prescriptions.
Data from an integrated closed-network setting pharmacy are particularly well suited to
minimize sources of outcome misclassification and easily identify outcomes (Karter et al 2009;
Gadkari and McHorney 2010). In a closed-network setting, pharmacy benefits are only honored
at pharmacies within the network, providing a disincentive for patients to fill their prescriptions
outside of the network. The same is true for medical visits and procedures. Thus, the outpatient,
inpatient, and laboratory data from an integrated closed-network setting would likely be more
comprehensive.
9

Lastly, patients who pick up their initial medication are assumed to have consumed the
drug once they get home. In reality, patients may pick up their medications but forget to actually
take the medication or ultimately choose not to take the medication. However, direct measures of
medication adherence (e.g. blood level monitoring or direct observation) are often costly and
burdensome. In addition, laboratory tests for drug levels are not available for most drugs and
may be too invasive for patients (Vermeire et al 2001).
3.6.2 Endogeneity of primary nonadherence variable
Describing the primary nonadherent population and the healthcare outcomes associated
with PNA is of clinical interest and was the focus of this research. Estimation of the incremental
causal effect of PNA on healthcare outcomes is also of interest. However, the PNA variable may
be endogenous due to unobservable variables correlated with PNA that are also correlated with
the outcomes of interest. In particular, patients who are adherent to medications may also be the
sicker. On the other hand, adherent patients may engage in additional healthy behaviors such as
eating well and exercising regularly more so than nonadherent patients (Roebuck et al 2011).
Thus, patients who are adherent are different from patients who are nonadherent, making it
difficult to accurately estimate the effect of primary nonadherence on healthcare outcomes.
Several econometric methods can be applied in order to avoid biased and inconsistent
estimates of the treatment effect in the presence of endogeneity. These include methods such as
the Instrumental Variables (IV) method and Fixed Effects (FE) and Random Effects (RE).
Although the Difference-in-Difference (DID) estimator does not address the issue of endogeneity
and usually assumes the treatment effect is exogenous, it addresses changes in outcomes that
occur over time and rules out alternative explanations of why outcomes differed between
treatment groups (Wooldridge 2002). Each of these methods was considered for this research
and the strengths and limitations of each method are briefly discussed in the sections below.
3.6.2.1 Instrumental Variables
The instrumental variable by definition is an observable variable that is uncorrelated with
the error term and correlated with the endogenous variable, and provides consistent estimates of
10

the treatment effect. However, there are several limitations with the IV approach. First, IV
estimates are still biased. Second, we cannot test the major assumption that the IV is
uncorrelated with the unobservable error term. Furthermore, if the correlation is weak between
the IV and unobservable, then the estimates are worse off than not using the IV in the first place.
Lastly, IV’s are difficult to identify in real-world practice (Wooldridge 2002). As such, an
appropriate IV that was correlated with PNA but uncorrelated with the unobserved variables could
not be identified and so this approach was not used in this study.
3.6.2.2 Fixed Effects and Random Effects
In the case where the outcome of interest is change in laboratory values over time,
multiple observations are available for patients over multiple time periods. The fixed effects panel
data model provides consistent estimates when unobserved effects are correlated with the
explanatory variables in the same time period by “differencing out” the effect of time-constant
unobserved variables. The major limitation of this approach is that the effect of time-constant
variables, such as our variable of interest, i.e. PNA, cannot be estimated since these variables
also get differenced-out (Wooldridge 2002).
Random effects models effectively put the unobserved variables into the error term and
assume that the unobserved variable is independent of all covariates included in the model
(Wooldridge 2002). When examining the effect of PNA on laboratory values over time, it is likely
that the unobserved variables are also correlated with some of the factors that determine
laboratory values, such as patient age and comorbidities. In summary, the FE and Re models
were not appropriate for our study.
3.6.2.3 Difference-in-Difference Estimator
The DID model compares time changes in the mean outcome for the treatment and
control groups, thereby allowing both group-specific and time-specific effects. When comparing
the average lab values over time between primary adherent and nonadherent patients, the DID
model is given by:
y
ipt
= α + γD
p
+ λd
t
+ δ(D
p
*d
t
) + X’
ipt
β + ε
ipt

11

where y
ipt
= average lab value for individual i in group p at time t, D
p
= 1 if patient is primary
adherent or =0 if patient is primary nonadherent, d
t
= 1 if lab was collected after the index date or
=0 if collected before the index date, X
ipt
= observable variables that predict lab values, and ε
ipt
=
error term. The OLS estimate of δ can be interpreted as the change in average lab values among
the primary adherent over time minus the change in average lab values among the primary
nonadherent over time:
δ = (y
p,2
– y
p,1
) – (y
pn,2
– y
pn,1
)
Although the DID model does not address endogeneity issues, it does address the
temporal variation in lab values and the effect of events besides PNA that occurred both before
and after the index date. The crucial assumption in the DID model is that in the absence of
primary adherence, the average outcomes for the primary nonadherent and primary adherent
groups would have followed similar paths over time (Meyer 1995). This assumption can be
tested by comparing trends in lab values between the two patient groups prior to the index date.

12

CHAPTER 4: FACTORS ASSOCIATED WITH PRIMARY NONADHERENCE TO CHRONIC
AND ACUTE MEDICATIONS (PHASE I)
4.1 Objectives
Study objectives of Phase I were to 1) measure PNA rates for ten therapeutic drug
groups, and 2) identify patient, prescriber, and prescription risk factors for PNA to chronic and
acute medications.
4.2 Methods
4.2.1 Study design, setting, and data source
This retrospective cohort study was conducted at Kaiser Permanente Southern California
(KPSC) and was approved by the KPSC Institutional Review Board. Kaiser Permanente is a
large managed care organization providing comprehensive healthcare and is comprised of 8
regions across the nation. KPSC is the largest region with 14 medical centers and an estimated
3.4 million members, representing 15% of the underlying population in Southern California.
Medical centers vary in population size from Kern County serving only 92,745 patients to San
Diego serving 490,154 members. Membership is largely employer-based and only about 5% of
the KPSC population are Medicaid patients.
KPSC members receive the majority of their healthcare and prescriptions at Kaiser
Permanente facilities. Patient information on demographics and healthcare encounters
(diagnoses, procedures, laboratory results, and prescriptions) are captured in an electronic
medical record (eMR) system, which was rolled-out over a period of 4 years by medical center,
beginning in 2004 and ending in 2008. Electronic prescribing (e-prescribing) at KPSC also
became available almost simultaneously with the implementation of the eMR. All prescribers
enter all prescriptions in the eMR system and this information is electronically sent to the Kaiser
pharmacy of the patient’s choice. Once the patient checks in at the pharmacy, the prescription is
released from the electronic queue and filled while the patient waits. Copay information is only
available for sold prescriptions.

13

4.2.2 Study population
Study inclusion and exclusion criteria for Phase I are summarized in Table 2. New
prescriptions for a study drug prescribed between December 1, 2009 and February 28, 2010
were included in the study. Study drugs included a total of ten therapeutic drug groups selected
based on disease prevalence, clinical interest, or the potential impact of PNA on healthcare
outcomes: antiinfectives, analgesics, migraine medications, antidiabetics, osteoporosis
medications, cardiovascular agents, antihyperlipidemics, antiasthmatics, antidepressants, and
anticoagulants. The first 3 drug groups were categorized as acute therapy and the remaining
were categorized as chronic therapy. Each therapeutic drug group consists of several drug
classes. For example, the antidiabetics therapeutic drug group includes drug classes such as
insulin and sulfonylureas. A total of 874 individual drug products were included in the study.
A new prescription was defined as having no prior dispensing of any drug in the same
drug class during the 12 months before the date the prescription was ordered (index date). For
example, a sulfonylurea prescription would be considered as a new prescription if no other
sulfonylureas were dispensed during the 12 months prior to the index date. This minimized the
effect of preexisting drug supplies on current filling and also excluded prescriptions used as
augmentation therapy or episodes of switching drugs within a drug class.
Patients were required to have continuous membership and drug benefits for 12 months
before and after the index date. The post-index drug benefit criterion ensured patients would
have a financial incentive to fill prescriptions at Kaiser and maximized the likelihood that
patients filled their medications at Kaiser rather than at outside pharmacies. The 12-months pre-
index period was used to identify baseline patient characteristics.
Prescriptions that were renewed, transferred from an outside pharmacy, verbally ordered
by the prescriber, printed out in paper form in the doctor’s office, and hard copy prescriptions from
outside providers were excluded from the study. Prescriptions that were switched to a different
drug or later cancelled by the prescriber before it was picked up by the patient were also
excluded. In addition, prescriptions with missing patient demographic information or filled at a
14

Kaiser pharmacy located outside of Southern California were excluded. Lastly, since pregnancy
complicates medication therapy and may result in drug discontinuation, prescriptions written for
female patients who became pregnant (based on gestation date) during the study period were
excluded.
Table 1. Summary of study inclusion/exclusion criteria for Phases I and II
Phase I Phase II
Inclusion criteria
Newly initiated
prescription
No dispensing of any drug in the same drug class as index drug during the
12 months prior to the index date
Eligible index
medication
antiinfectives, analgesics,
antiasthmatics, cardiovascular,
antidepressants,
antihyperlipidemics,
antidiabetics, antimigraine
medications, anti-osteoporosis
medications, warfarin
antihyperlipidemics or antidiabetics
Membership and
drug benefits
Patient has continuous
membership and drug benefits
for 12 months prior and after
index date
Patient has continuous membership
and drug benefits for 12 months prior
and 18 months after index date
Patient age None 18 years or older on index date
Exclusion criteria
Prescriptions not
generated by e-
prescribing
system
Prescriptions transferred from an outside pharmacy, verbally ordered by
the prescriber, renewed or switched to a different drug, printed out in the
doctor’s office, and paper prescriptions from providers outside of network
Cancelled
prescriptions
Prescriptions later cancelled by the prescriber before it was picked up by
the patient
Missing
information
Missing information regarding patient demographics, provider gender, and
drug brand.
Non-KPSC
region
Prescriptions filled at a Kaiser pharmacy located outside of Southern
California.
Pregnancy Prescriptions written for female patients who become pregnant (based on
gestation date) during the study period.
Type I diabetics None Patients identified with an ICD-9 code of
250.01, 250.03, 250.11, 250.13, 250.21,
250.23, 250.31, 250.33, 250.41, 250.43,
250.51, 250.53, 250.61, 250.63, 250.71,
250.73, 250.81, 250.83, 250.91, or
250.93 during the baseline period and
has insulin as their only index drug
during the 3-month enrollment period.

15

4.2.3 Study variables
4.2.3.1 Study outcome
The primary study outcome was PNA, which was defined in Phase I as the failure to fill a
prescription within 14 days of the index date. Previous studies have demonstrated that most
patients fill their medications within the first two weeks of the index date (Karter et al 2009;
Liberman et al 2010). Sensitivity analysis was performed to examine changes in PNA rates when
the definition of PNA varied from 14 days to 30 days and 90 days. The extent of PNA was
reported as a rate calculated as the number of primary nonadherent prescriptions divided by the
total number of new prescriptions.
4.2.3.2 Patient, prescriber, and prescription characteristics
Patient characteristics included patient age at index date and gender as recorded in the
eMR. Patient race was geocoded based on 2010 census-tract data. The patient’s zip code was
linked to U.S. Census 2000 data to assign median household income. The 12-month pre-index
period was used to identify baseline comorbidities based on occurrence of at least one ICD-9
code. Selected disease comorbidites included the 17 standard diseases used to calculate the
Charlson Comorbidity Index (Charlson et al 1987) and 5 additional disease states (alzheimer’s
disease, hyperlipidemia, migraine, depression, osteoporosis) corresponding to the therapeutic
drug groups examined in the study.
The pre-index period was also used to calculate baseline healthcare utilization, which
included the number of prior clinic visits, ER visits or hospitalizations. Patients who did not have
any prior use of prescriptions in the same therapeutic drug group were flagged as treatment-
naïve to disease therapy. Thus, in our previous example, diabetic patients who were treatment-
naïve did not use sulfonylureas or any other classes of antidiabetic medications, such as insulin
or biguanides, during the pre-index period. A flag was created to indicate if the patient had filled
at least one prescription in the pre-index period as a measure of the patient’s compliant
medication behavior. Pharmacy benefits (dual insurance, primary/dependent, plan type:
Medicare, Medicaid, Commercial) were also captured.
16

Prescriber characteristics included prescriber age and gender, as well as prescriber
race/ethnicity, years of experience practicing at KPSC, and specialty. Dummy variables were
used to indicate if the patient and physician were of the same gender or race/ethnicity in order to
assess whether having a prescriber of the same gender or race/ethnicity would influence the
patient-prescriber relationship and result in improved primary adherence.
Characteristics of the index prescription were identified from pharmacy records and
included acute vs. chronic therapy, generic vs. brand, pharmacy regional location, month
prescribed, and weekday vs. weekend prescribing. Since copay information was unavailable for
prescriptions that were not filled, it was not included in the analysis. The total number of
prescriptions written for study drugs on the index date was examined to investigate if patients
who are prescribed multiple medications and likely face an increased pill and copay burden are
more likely to be primary nonadherent. Lastly, dummy variables were used to indicate if the
prescription was used to treat symptomatic disease, and included any medication in the following
drug groups: antimigraine, analgesics, antiinfectives, antiasthmatics, and antidepressants.
4.2.4 Statistical analysis
This analysis was performed at the prescription level so any patient may have had more
than one observation in the data set. PNA rates were calculated overall and by drug group. The
frequency distribution of the time between the index date and fill date was examined. Descriptive
statistics using t-tests and chi-square tests were used to test unadjusted differences in
characteristics of filled prescriptions to those of unfilled prescriptions.
Multivariable logistic regression was used to calculate odds ratios and identify significant
factors associated with PNA to study drugs when adjusting for other study variables. All patient,
prescriber, and prescription characteristics were considered for inclusion in the model.
Interaction terms with the acute vs. chronic variable and patient characteristics were tested. The
final inclusion of variables into the model was based on statistical significance or theoretical
plausibility. A significance level of less than 0.05 was considered statistically significant.
17

We also analyzed significant factors associated with the time to first prescription fill,
which was calculated as the number of days between the index date and fill date, using a Cox
Proportional Hazard model. Covariates included in the model were similar to those in the logistic
model. Patients will be followed until the index prescription was filled or at the end of the 12-
month follow-up period. All statistical analyses were performed using the SAS statistical package
version 9.1 (SAS Institute, Cary, NC).
4.3 Results
4.3.1 Rates of primary nonadherence
A total of 569,095 new prescriptions were written for 398,025 patients during the 3-month
period that satisfied study inclusion and exclusion criteria (Figure 2).
Figure 2. Identification and selection of prescriptions in final cohort of Phase I study

Most prescriptions were written for antiinfectives (43.5%) or analgesics (24.6%), followed
by antiasthmatics (9.7%), cardiovascular (8.6%), antidepressants (4.8%), antihyperlipidemics
(3.9%), antidiabetics (2.5%), antimigraine (1.0%), anti-osteoporosis (0.9%), and anticoagulants
(0.3%). The average (SD) patient copay for sold prescriptions was $9.89 (14.06).
18

The overall PNA rate was 9.8%. PNA varied by therapeutic drug group, ranging from
2.9% for antiinfectives to 22.4% for osteoporosis medications (Figure 3). Varying the definition of
PNA from 14 days to 90 days decreased the overall PNA rate to 8.0%. PNA rates were highest
for osteoporosis medications, antihyperlipidemics, and analgesics. Of the 569,095 prescriptions,
437,940 (77%) prescriptions were filled on the same day as they were written and 525,449 (92%)
prescriptions were eventually filled within 6 months of the index date.
Figure 3. Primary nonadherence rates by therapeutic drug group

4.3.2 Unadjusted comparison between the primary adherent and nonadherent
An unadjusted comparison between primary adherent and nonadherent prescriptions
revealed several small but significant differences in patient, prescriber and prescription
characteristics (tables not shown). PNA was more common for minority race/ethnicities, lower
22.4% 22.3%
21.9%
12.6%
10.8%
7.8% 7.7%
5.6%
4.1%
2.9%
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
19

household incomes, and patients prescribed a greater number of prescriptions on the index date.
A significantly greater proportion of primary nonadherent prescriptions were written by younger
prescribers with <10 years of Kaiser experience. Prescriptions written for brand-name
medications and treatments for asymptomatic diseases were more likely to be primary
nonadherent.
4.3.3 Multivariable logistic results
Results from the preliminary multivariable logistic regressions revealed significant
interaction effects between the acute vs. chronic therapy variable and other important variables in
the model. Thus, the multivariable logistic regression results were stratified by chronic or acute
therapy. The adjusted odds ratios (OR) and 95% confidence intervals (CI) resulting from the
logistic regression are presented separately in Tables 2-4 for patient, prescriber, and prescription
characteristics, respectively. Although presented separately, results from three tables are all
from the same multivariate logistic model that controlled for all study covariates listed in the 3
tables.
4.3.3.1 Patient characteristics
The effects of patient gender and age differed depending on whether the treatment was
acute or chronic. However, when analgesics were excluded as part of a sensitivity analysis, the
trend in effect of patient age was similar between acute and chronic treatments. Patient race had
a consistent effect across medications where blacks [acute: OR=1.30 (1.25, 1.36), chronic:
OR=1.26 (1.18, 1.33)] and hispanics [acute: OR=1.19 (1.16, 1.23), chronic: OR=1.06 (1.01, 1.11)]
were more likely to be primary nonadherent in comparison to whites. Baseline patient
comorbidities such as hyperlipidemia, diabetes without complications, paraplegia, and peripheral
vascular disease increased the risk of PNA. However, cancer and renal disease lowered the risk
of PNA.

20

Table 2. Results for patient characteristics from multivariable logistic regression
predicting primary nonadherence
Acute
Treatments
OR (95% CI)
Chronic
Treatments
OR (95% CI)
Female patient (vs. male) 0.87 (0.85, 0.89) 1.04 (1.01, 1.08)
Patient age group (years)
10 or less 1.35 (1.24, 1.47) 1.16 (0.98, 1.37)
11 to 20 0.92 (0.85, 1.00) 1.55 (1.35, 1.79)
21 to 30 0.75 (0.69, 0.80) 1.66 (1.47, 1.86)
31 to 50 0.76 (0.71, 0.81) 1.44 (1.32, 1.57)
51 to 64 0.9 (0.85, 0.96) 1.27 (1.17, 1.38)
65 or more (reference group) 1.00 1.00
Patient race/ethnicity
American Indian/ Alaskan Native 0.87 (0.63, 1.19) 1.24 (0.84, 1.84)
Asian/pacific islander 1.24 (1.18, 1.29) 0.99 (0.93, 1.05)
Black 1.3 (1.25, 1.36) 1.26 (1.18, 1.33)
Hispanic 1.19 (1.16, 1.23) 1.06 (1.01, 1.11)
Multi-racial 1.37 (1.18, 1.60) 1.24 (0.97, 1.58)
White (reference group) 1.00 1.00
Patient household income
$30,000 or less 1.03 (0.97, 1.09) 1.07 (0.98, 1.16)
$30,000 to 50,000 1.02 (0.98, 1.07) 1.06 (0.99, 1.13)
$50,000 to $70,000 1.00 (0.95, 1.04) 1.05 (0.98, 1.12)
Unknown 1.00 (0.93, 1.07) 1.20 (1.08, 1.33)
$70,000 or more (reference group) 1.00 1.00
Baseline patient comorbidities
Alzheimer's disease 1.07 (0.90, 1.27) 1.53 (1.24, 1.89)
Hyperlipidemia 1.13 (1.09, 1.16) 1.19 (1.15, 1.24)
Migraine 1.05 (0.99, 1.11) 1.01 (0.92, 1.10)
Cancer 0.80 (0.75, 0.86) 0.88 (0.80, 0.97)
Cerebrovascular disease 1.28 (1.20, 1.36) 1.07 (0.98, 1.16)
CHF 1.01 (0.94, 1.08) 0.98 (0.90, 1.06)
Dementia 1.19 (0.98, 1.44) 1.44 (1.12, 1.85)
Depression 0.98 (0.94, 1.01) 1.08 (1.03, 1.13)
Diabetes with complications 0.78 (0.73, 0.82) 0.97 (0.90, 1.03)
Diabetes without complications 1.43 (1.37, 1.49) 1.26 (1.20, 1.32)
HIV 0.82 (0.65, 1.03) 1.07 (0.76, 1.53)
Metastatic cancer 1.00 (0.87, 1.16) 0.97 (0.79, 1.19)
Mild liver disease 0.91 (0.86, 0.98) 0.99 (0.90, 1.07)
Myocardial infarction 1.56 (1.48, 1.66) 0.89 (0.82, 0.97)
Moderate-severe liver disease 0.74 (0.57, 0.96) 0.78 (0.56, 1.10)
Osteoporosis 1.03 (0.98, 1.08) 1.09 (1.02, 1.16)
Paraplegia 1.21 (1.05, 1.40) 1.39 (1.14, 1.70)
Peptic ulcer disease 1.00 (0.89, 1.14) 1.21 (1.02, 1.44)
Pulmonary disease 0.84 (0.81, 0.86) 1.18 (1.13, 1.24)
Peripheral vascular disease 1.22 (1.15, 1.29) 1.10 (1.02, 1.19)
Renal disease 0.90 (0.86, 0.95) 0.90 (0.85, 0.96)
Rheumatic disease 0.96 (0.88, 1.05) 1.00 (0.89, 1.13)
Commercial insurance 0.93 (0.86, 1.02) 1.11 (0.95, 1.31)
Medicaid insurance 0.92 (0.83, 1.03) 0.61 (0.49, 0.78)
21

Table 2 (Continued)
Acute
Treatments
OR (95% CI)
Chronic
Treatments
OR (95% CI)
Medicare insurance 1.04 (0.98, 1.11) 1.02 (0.94, 1.11)
Dual coverage from spouse 0.78 (0.72, 0.83) 0.73 (0.65, 0.82)
Primary subscriber 1.04 (1.01, 1.07) 1.08 (1.03, 1.12)
Filled prescription in prior year 0.06 (0.06, 0.07) 0.11 (0.10, 0.12)
Treatment naïve patient 2.52 (2.36, 2.7) 1.07 (1.03, 1.12)
ER visit in prior year 1.19 (1.16, 1.22) 0.88 (0.85, 0.92)
Hospitalization in prior year 0.96 (0.93, 0.99) 0.96 (0.91, 1.01)

Patients with dual insurance or Medicaid coverage (for chronic drugs only) were less
likely to be primary nonadherent. Patients who filled at least one prescription in the pre-index
period were also less likely to be primary nonadherent [acute: OR=0.06 (0.06, 0.07), chronic:
OR=0.11 (0.10, 0.12)]. Treatment-naïve patients were more likely to be primary nonadherent,
especially to acute medications [acute: OR=2.52 (2.36, 2.70), chronic: 1.07 (1.03, 1.12)].
However, the effect of treatment naïve patients for acute therapy becomes statistically
insignificant when analgesics are excluded from the analysis. Lastly, patients with prior ER visits
had mixed effects [acute: OR=1.19 (1.16, 1.22), chronic: 0.88 (0.85, 0.92)], while prior
hospitalization lowered the risk of PNA to acute medications [OR=0.96 (0.93, 0.99)].
4.3.3.2 Prescriber characteristics
Patients prescribed acute medications were more likely to be primary nonadherent if
seen by providers who were younger or of minority race/ethnicity. Patients primary nonadherent
to chronic medications were more likely to be female or black. Prescriptions from emergency
medicine, OBGYN, pediatrics, or urgent care were more likely to be filled. Patient-provider
gender concordance was not significant, but patients of the same race/ethnicity as the prescriber
were more likely to be primary nonadherent to acute medications.

22

Table 3. Results for prescriber characteristics from multivariable logistic regression
predicting primary nonadherence
Acute
Treatments
OR (95% CI)
Chronic
Treatments
OR (95% CI)
Prescriber female gender (vs. male) 0.98 (0.96, 1.00) 1.11 (1.07, 1.15)
Prescriber age group (years)
≤35 1.18 (1.14, 1.22) 0.87 (0.83, 0.92)
36 to 45 1.18 (1.14, 1.22) 0.97 (0.92, 1.01)
46 to 55 1.07 (1.04, 1.11) 1.07 (1.01, 1.12)
Unknown 0.89 (0.68, 1.17) 1.91 (1.24, 2.96)
>55 (reference group) 1.00 1.00
Prescriber race/ethnicity
American Indian/ Alaskan Native 0.95 (0.81, 1.11) 0.87 (0.69, 1.11)
Asian/pacific islander 1.06 (1.03, 1.09) 0.97 (0.93, 1.02)
Black 1.18 (1.13, 1.24) 1.10 (1.03, 1.19)
Hispanic 1.08 (1.04, 1.12) 0.91 (0.86, 0.97)
Multi-racial 1.13 (0.82, 1.56) 1.53 (0.90, 2.60)
Unknown 0.85 (0.75, 0.96) 0.95 (0.76, 1.18)
White (reference group) 1.00 1.00
Prescriber specialty
Emergency medicine 0.97 (0.92, 1.02) 0.54 (0.45, 0.64)
Internal medicine 1.18 (1.14, 1.21) 1.01 (0.97, 1.05)
OBGYN 0.67 (0.62, 0.72) 0.69 (0.54, 0.88)
Other 1.03 (1.00, 1.07) 1.13 (1.08, 1.18)
Pediatrics 0.39 (0.37, 0.42) 0.86 (0.75, 0.98)
Urgent care 0.63 (0.59, 0.68) 0.47 (0.39, 0.56)
Family Practice (reference group) 1.00 1.00
Patient/prescriber gender match 1.03 (1.00, 1.05) 1.03 (1.00, 1.07)
Patient/prescriber race/ethnicity match 1.05 (1.02, 1.08) 1.04 (0.99, 1.08)

4.3.3.3 Prescription characteristics
Patients prescribed acute medications with brand names [OR=1.49 (1.42, 1.56)] or a
higher number of other medications prescribed on the index date [OR=1.26 (1.25, 1.28)] were
more likely to be primary nonadherent. For patients prescribed chronic medications, having
higher number of medications prescribed on the index date [OR=0.83 (0.81, 0.85)] or
symptomatic disease [OR=0.51 (0.48, 0.53)] decreased the risk of PNA. Prescriptions written on
the weekend were more likely to be primary nonadherent for chronic medications [OR=1.16 (1.09,
1.24)] but less likely for acute medications [OR=0.70 (0.68, 0.73)].

23

Table 4. Results for prescription characteristics from multivariable logistic regression
predicting primary nonadherence
Acute Treatments
OR (95% CI)
Chronic Treatments
OR (95% CI)
Brand name 1.49 (1.42, 1.56) 0.99 (0.94, 1.03)
Number of drugs prescribed on index
date 1.26 (1.25, 1.28) 0.83 (0.81, 0.85)
Symptomatic disease - 0.51 (0.48, 0.53)
Prescribed on weekend 0.70 (0.68, 0.73) 1.16 (1.09, 1.24)
Pharmacy area
Antelope Valley 1.77 (1.65, 1.90) 1.05 (0.94, 1.16)
Baldwin Park 1.02 (0.97, 1.08) 0.98 (0.91, 1.06)
Coachella Valley 1.38 (1.20, 1.60) 1.21 (1.00, 1.46)
Downey 1.20 (1.15, 1.27) 0.86 (0.80, 0.93)
Fontana 1.02 (0.97, 1.07) 0.94 (0.88, 1.00)
Kern County 1.16 (1.07, 1.26) 1.19 (1.07, 1.32)
Los Angeles 1.39 (1.32, 1.46) 0.99 (0.92, 1.07)
Orange County 1.15 (1.10, 1.21) 0.87 (0.82, 0.94)
Panorama City 1.02 (0.96, 1.08) 0.80 (0.74, 0.87)
Riverside 1.39 (1.32, 1.46) 1.02 (0.95, 1.10)
South Bay 1.25 (1.18, 1.32) 1.00 (0.93, 1.08)
Ventura 1.50 (1.33, 1.68) 1.04 (0.89, 1.23)
West Los Angeles 1.26 (1.19, 1.34) 0.98 (0.90, 1.07)
Woodland Hills 1.11 (1.05, 1.18) 0.99 (0.91, 1.08)
Unknown 9.46 (8.31, 10.77) 2.56 (2.30, 2.85)
San Diego (reference group) 1.00 1.00

4.3.4 Time to first fill analysis
Results from the Cox Proportional Hazards model analyzing factors associated with time
to first prescription fill were very consistent with those from the multivariate logistic regression
model in that most statistically significant Hazard Ratios (HR) were in the same direction as the
OR identified from the logistic regression.
The effects of patient gender and age on the time to first fill again varied by acute vs.
chronic treatment. In contrast to previous results from the logistic regression model, patients less
than 10 years of age were significantly less likely to fill their chronic medication rather than their
acute medications. Patient race again was a highly significant factor. Similar to the logistic
results, black patients were less likely to fill than whites for both acute [HR=0.95 (p<0.0001)] and
chronic [HR=0.96 (p<0.0001)] medications. Asian/pacific islanders and hispanics were less likely
to fill than whites for acute medications only [HR=0.97 (p<0.0001) for both races]. The effects of
24

baseline patient comorbidities and insurance type were also similar to results previously found
with the logistic model. As before, the effect of having filled at least one prescription in the pre-
index period was strongly associated with an increased likelihood of filling the first prescription
[acute HR=3.22 (p<0.0001), chronic HR=2.93 (p<0.0001)], and treatment-naïve patients were
more likely to be primary nonadherent to acute medications [HR=0.92 (p<0.0001)]. Prior ER
visits and hospitalizations had similar effects as identified previously.
The effects of prescriber gender and age group were similar as before. Prescriptions
written by OBGYN were no longer significantly associated with an increased likelihood of filling
the first prescription, but prescriptions written by emergency medicine, pediatrics, and urgent care
were still less likely to be primary nonadherent. Unlike before, now both patient-provider
concordance in terms of gender and race/ethnicity were significantly associated with primary
nonadherence to acute medications [HR=0.99 (p<0.05) for both concordance variables].
As before, brand name prescriptions were associated with an increased risk of primary
nonadherence for acute medications [HR=0.92 (p<0.0001)]. However, in addition, the time-to-fill
analysis also found that patients were more likely to fill their chronic medications if they were
brand name prescriptions [HR=1.03 (p<0.01)]. The effects from having a higher number of other
medications prescribed on the index date, symptomatic disease, and being prescribed on the
weekend on primary nonadherence were similar to those previously found in the logistic
regression results.
4.4 Discussion
Although the overall PNA rate across all ten drug groups was only 9.8%, rates for certain
therapeutic drug groups such as anti-osteoporosis medications, antihyperlipidemics, and
analgesics were higher. For these drug groups, about one in five patients did not pick up their
medications after 14 days. This is concerning, especially since medications for osteoporosis and
hyperlipidemia have been shown to decrease morbidity and mortality (Ross et al 2011; MRC/BHF
Heart Protection Study 2002). In particular, statin and bisphosphonate therapy have also been
25

demonstrated as cost-effective (Lazar et al 2011; Pham et al 2011). The availability of OTC
analgesics may contribute to the relatively higher PNA rate associated with analgesics.
The PNA rates found in this study are somewhat higher than those reported in previous
studies in integrated healthcare settings (Karter et al 2009; Raebel et al 2011). These differences
could be attributed to the criteria used to define PNA. Previous studies defined PNA as the failure
to fill within 30 or 60 days. This study used 14 days to define PNA since most prescriptions were
filled within two weeks and it would be unreasonable from a policy perspective for clinicians to
wait 30 days or more to contact patients for intervention. Changing our definition from 14 days to
30 days produced PNA rates closer to those of previous studies.
The multivariate logistic analysis revealed that some factors consistently increased or
decreased the likelihood of PNA across all drug groups. Black patients were more likely to be
nonadherent, which is consistent with prior research (Wroth & Pathman 2006). However, the
effect of race (and income) should be cautiously interpreted since these socioeconomic variables
were geocoded. Patients with baseline comorbidities (except for cancer and renal disease) were
less likely to fill their prescription, which is also consistent with findings from prior studies (Raebel
et al 2011; Kennedy et al 2008; Shah et al 2009). Patients with multiple comorbidities are likely
taking several medications and adding another medication to their current regimen is likely to
increase pill burden, the complexity of medication therapy, and the risk of drug interactions.
However, if the prescription was used to treat symptomatic disease, the patient was more likely to
fill the prescription.
Similar to a recent study by Liberman et al (2010), the logistic regression analysis found
that patients with prior fills were more likely to fill their current prescription. Prior prescription fill
history is a strong indicator of a patient’s compliant behavior and willingness to take medications
in the future. Treatment-naïve patients were more likely to be primary nonadherent to
medications. These patients may be at the early stage of their disease and elect to postpone
filling their prescriptions in order to try alternative methods like diet and exercise or over-the-
counter products. Alternatively, prescriptions written by emergency medicine and urgent care
26

may indicate increased severity and acuteness of the patient’s illness, and thus, were associated
with a decreased likelihood of PNA.
In the logistic regression analysis, the effects of some covariates differed based on
whether the treatment was acute or chronic. Younger patients were more likely to fill acute
medications (mostly antiinfectives or analgesics) and less likely to fill their chronic medications
(mostly includes medications for asthma, cardiovascular and depression) relative to older
patients. The effects of brand name medications and a greater prescription and copay burden on
the index date only affected acute medications, suggesting that patients may be more sensitive to
the cost of the drug when the disease is acute. Patients were less likely to pick up prescriptions
prescribed on a weekend for chronic medications and more likely for acute medications,
suggesting that patients may feel less urgency in treating chronic diseases in comparison to
acute diseases.
Similar to Raebel et al (2011), only a few variables were strongly associated with PNA,
such as prior prescription fill and being treatment-naïve, despite having a comprehensive model
that includes a long list of patient, prescriber and prescription covariates. Such results highlight
the difficulty of quantifying complex human medication-taking behavior using administrative data.
Patient surveys have identified that the most common reasons for PNA are those related to
patient concerns regarding the medication, such as potential side effects and the perceived need
for the medication (Gadkari & McHorney 2010). Capturing these effects can be difficult when
using quantitative administrative data.
Results from the Cox Proportional Hazard model were very similar to those from the
multivariate logistic regression. We found that the majority of primary adherent prescriptions
were filled on the day they were prescribed, so adjusting for time to fill did not change our results
drastically.
This is one of the first studies to compare factors of PNA between chronic and acute
medications. Strengths in study design include the study setting of an integrated healthcare
system and requiring membership and prescription benefits in that patients have a financial
27

incentive to fill their prescriptions at Kaiser. In addition, the use of e-prescribing and eMR data
allowed direct linkage of prescription orders to dispensing information. Use of uniquely rich eMR
data also allowed the inclusion of many more patient, prescriber, and prescription characteristics
in the logistic regression model than those in most other studies. Lastly, the sample size for this
study is very large (n=398,025) in comparison to recent studies (Karter et al 2010; Raebel et al
2011), which improves the robustness of our results. A study limitation is the limited
generalizability of our results to patient populations in non-integrated healthcare systems, as well
as Medicaid or uninsured patients, who were not well-represented in this study.
4.5 Conclusions
Overall PNA was 9.8% but individual PNA rates varied by therapeutic drug group.
Factors of PNA were mostly consistent across drug groups, but some factors such as patient age
and brand name depended on whether the treatment is acute or chronic. The results of this study
help clinicians and healthcare decision-makers to understand the extent of PNA and how PNA
varies by therapeutic drug group in an integrated healthcare setting, and hopefully facilitate the
decision-making process in designing and implementing cost-effective patient interventions to
improve adherence to chronic and acute medications. Future research should focus on
measuring the effect of PNA on patient health outcomes and cost.

28

CHAPTER 5: OUTCOMES ASSOCIATED WITH PRIMARY NONADHERENCE TO
CHOLESTEROL AND DIABETES MEDICATIONS (PHASE II)
5.1 Objectives
The objective of Phase II was to examine healthcare outcomes of change in lab values,
disease complications, ER visits, and hospitalization associated with PNA to cholesterol and
diabetes medications in the adult patient population.
5.2 Methods
5.2.1 Study design, setting, and data source
This retrospective cohort study was conducted at Kaiser Permanente Southern California
(KPSC) and was approved by the KPSC Institutional Review Board. The research setting and
data source has been discussed previously in Section 4.2.1 for Phase I.
5.2.2 Study populations
The inclusion and exclusion criteria used to identify the Phase II study population are
summarized in Table 1. The cholesterol and diabetes populations included all patients in Phase I
who were prescribed either a new antidiabetic or antihyperlipidemic that also satisfied additional
criteria of: 1) age ≥18 years on index date and 2) continuous membership and drug benefits for
12 months prior and 18 months after the index date. For patients who had more than one index
date for a cholesterol or diabetes medication during the 3-month study enrollment period, only the
very first index date was included in the analysis.
All exclusion criteria applied in Phase I were also applied in Phase II. In addition,
patients identified with an ICD-9 code for Diabetes type I during the baseline period and insulin as
their only index drug during the 3-month enrollment period were excluded from the diabetes
population in order to focus the analysis on type II diabetics.
5.2.3 Study variables
5.2.3.1 Study covariates
The covariate of interest in Phase II was adherence to cholesterol or diabetes
medications. Patients were classified into one of three mutually exclusive groups: primary
29

nonadherent, secondary nonadherent, and adherent. The definition of PNA was modified from
Phase I by extending the number of days allowed between the index date and dispensed date
from 14 days to 180 days. This was designed to minimize the risk of misclassifying patients who
were actually primary adherent as primary nonadherent. Based on descriptive results from
Phase I, almost all patients who eventually filled their first prescription filled within 180 days. If a
patient eventually filled their prescription on day 181, then they were still considered as primary
nonadherent.
Once patients were determined as primary adherent, secondary adherence in was
determined by the Medication Possession Ratio (MPR), which was calculated as the total days
supplied for any cholesterol (or diabetes) medication during the 12-month period after the index
date divided by 365 days (or 12 months). MPR was truncated so that it could not be below zero
or above 1. Patients were considered secondary nonadherent if MPR< 80% and adherent if
MPR≥ 80% (Bates et al, 2009; Sokol et al, 2005). As part of the descriptive analysis, persistence
to cholesterol medication was also examined. A gap in therapy of 180 days or more qualified as
discontinuation of therapy. A 180-day definition was used to be consistent with the definition of
primary nonadherence (failure to fill within 180 days of index date).
Patient characteristics examined in Phase II included patient age at index date, gender,
race/ethnicity, median household income, and insurance type (Medicare, Medicaid, Commercial,
primary subscriber, dual coverage). Baseline health status measures were evaluated during a 2-
year period starting from the year prior to the index date until the year after the index date when
secondary adherence was measured. These measures included baseline comorbidities identified
using ICD-9 codes, weighted Charlson comorbiditiy index, prior prescriptions count, prior clinic
visit count, prior ER visit or hospitalization, and most recent lab value prior to the index date.
Additional patient characteristics were included in Phase II. First, patients who filled any
cholesterol or diabetes medication during the year prior to the index date were flagged as an
indicator of patients who were prescribed a new drug in order to augment or switch from previous
therapy. Next, patients who filled cholesterol or diabetes medications other than the index drug
30

subsequent to the index date were also flagged. For diabetes only, an indicator was also created
for patients who filled insulin either during the 12 months prior or after the index date. These
indicators further control for disease severity as patients with prior or subsequent diabetes or
cholesterol medications are likely to differ from patients without these indicators.
Lastly, patients identified with disease complication(s) anytime starting from the year prior
to the index date until the year after the index date were flagged as secondary prevention
patients. For cholesterol, patients with previous diagnoses for myocardial infarction,
percutaneous transluminal coronary angioplasty (PTCA), coronary artery bypass graft (CABG)
surgery, coronary ischemic heart disease, coronary atherosclerosis, or stroke were identified as
secondary prevention patients. For diabetes, patients with any microvascular or macrovascular
complication starting from the year prior to the index date and until the year after the index date
were identified as secondary prevention patients. Patients who have already experienced
complications from diabetes or cholesterol may have more advanced or severe disease.
Prescriber characteristics examined in Phase II included prescriber gender, age at index date,
race/ethnicity, and specialty.
5.2.3.2 Study Outcomes
Study outcomes included lab values (LDL for cholesterol and Hemoglobin A1c (Ha1c) for
diabetes), new complications of diabetes or hyperlipidemia, ER visits, and hospitalizations (Table
5). Change in lab values was only examined in patients with lab values available before and after
the index date, which was a subset of patients from the entire diabetes or cholesterol cohort.
All new disease complications were identified using ICD-9 codes and were only analyzed
in primary prevention patients, defined as patients without disease complications during the year
prior and year after the index date (in total 2-year period). All event outcomes of disease
complications, ER visits, and hospitalizations were identified between 12 and 18 months after the
index date since the first 12 months after the index date were used to determine the patient’s
medication adherence level. Furthermore, beneficial effects of statins have been seen as soon
as after one year of therapy (Ho et al 2006).
31

5.2.4 Statistical analysis
5.2.4.1 Descriptive analysis
Baseline patient characteristics were compared among patient groups using ANOVA for
continuous variables and Chi-square or Fisher’s exact tests for categorical variables. Secondary
nonadherence was examined in several ways by describing the total days supplied dispensed,
total number of prescriptions filled, MPR, and persistence levels in the primary adherent during
the 12-month follow-up period.
Table 5. Healthcare outcomes of Phase II study
Outcome Patient population Definition
Change in
laboratory values
for Hemoglobin A1c
(HA1c) or LDL
Patients with both
baseline and
follow-up labs
available
Change in laboratory value calculated using most
recent values prior to index date and average lab
value during the 18 months after the index date.
Complications of
diabetes
Primary prevention
patients only
Microvascular and macrovascular complications
identified between the 12 and 18 months after
the index date.
Microvascular complications include neurological
disorders, diabetic foot problems, retinopathy,
and nephropathy (ICD-9 codes 250.40, 250.42,
250.50, 250.52, 250.60, 250.62, 250.70, 250.72,
250.80, 250.82, 250.90, 250.92, 785.4, 443.81,
357.2, 362.02, 249.61, 249.80, 249.41, 249.90,
443.81, 337.1, 356.8, 358.1, 713.5, 729.2, 447.1,
599.0, 791.0, 354.xx, 355.xx, 362.xx, 364.0x,
364.4x, 365.xx, 366.xx, 367.xx, 368.xx, 369.xx,
377.xx, 430.xx, 431.xx, 432.xx, 435.xx, 436.xx,
437.xx, 438.xx, 440.xx, 443.xx, 444.xx, 451.xx,
452.xx, 453.xx, 454.xx, 459.xx, 707.xx, 885.xx,
886.xx, 887.xx, 895.xx, 896.xx, 897.xx, 580.xx,
581.xx, 583.xx, 584.xx, 585.xx, 586.xx, 587.xx,
588.xx, 590.xx, 593.xx, 595.xx, 596.xx).
Macrovascular complications include acute MI
(ICD-9 code 410.xx) or stroke and other
occlusive cerebrovascular disease (433.xx,
434.xx).
Complications of
hyperlipidemia
All patients
Acute MI (ICD-9 code 410.xx) or stroke and other
occlusive cerebrovascular disease (433.xx,
434.xx) identified between the 12 and 18 months
after the index date.
Emergency room
visit
All patients
Determined from electronic medical record
outpatient and inpatient data using data between
the 12 and 18 months after the index date.
Hospitalization All patients
All-cause. Determined from electronic medical
record inpatient data using data between the 12
and 18 months after the index date.
32

5.2.4.2 Analysis of Lab Values
Average lab values were graphed over time by adherence group and average rates of lab
testing before and after the index date were compared among patient groups using ANOVA.
Unadjusted differences in average lab values before and after the index date among adherence
groups were compared using ANOVA. The paired t-test was used to test whether the average
lab value change within an individual patient was statistically significant. Lastly, the difference-in-
difference model using a multivariate regression was used to test the significance of the
difference-in-difference of lab values between the primary nonadherent and adherent, as well as
between the primary nonadherent and secondary nonadherent.
5.2.4.3 Analysis of event outcomes
Average number of ER visits per patient and average number of hospitalizations per
patient were graphed over time by adherence group. Unadjusted differences in the proportion of
patients identified with new disease complications, ER visits, and hospitalizations were compared
among patient groups using the chi-square test. The unadjusted average number of ER visits or
hospitalizations during the follow-up period was compared among adherence groups using
ANOVA.
Time to first disease complication, ER visit, or hospitalization was analyzed using
multivariate Cox proportional hazard models. Patients were followed until the event occurred and
censored from the analysis at the end of the follow-up period. An interaction term for adherence
group and prior cholesterol/diabetes medication fill was tested to examine if the effect of
adherence differs between patients newly initiating therapy and patients switching/augmenting
prior therapy. Sensitivity analysis included running all models without the most recent lab value
as a covariate, since including this covariate excluded a considerable number of patients without
baseline lab values available.
For all analyses, a significance level of 0.05 will be considered as statistically significant.
All statistical analyses will be performed using the SAS statistical package version 9.1 (SAS
Institute, Cary, NC).
33

5.3 Results for cholesterol cohort
5.3.1 Descriptive results
A total of 21,291 patients satisfied all Phase II inclusion and exclusion criteria for the
cholesterol analysis (Figure 4). Of the total cholesterol cohort, 91% of patients were identified as
primary prevention patients. Overall, 2,557 patients (12.0%) were primary nonadherent. The
PNA rate for primary prevention patients (12.6%) was almost double the PNA rate in secondary
prevention patients (6.7%). About 85% of patients who were primary adherent filled their new
prescription on the day it was written.
Figure 4. Identification of cholesterol study cohort

The total days supply and total number of fills dispensed during the 12-month period
immediately after the index date indicate only about a quarter of primary adherent patients were
dispensed all 365 days of medication or filled four prescriptions (Figures 5 and 6).

34

Figure 5. Total days supply during 12-month follow-up period for primary adherent
cholesterol patients

Figure 6. Total number of fills dispensed during 12-month follow-up period for primary
adherent cholesterol patients

MPR values revealed that only 27% of primary adherent patients had an MPR around
100% (Figure 7). 7,963 patients were identified with an MPR ≥80% and 10,772 patients with an
MPR <80%. About 23% of primary adherent patients filled 100 days supply or 1 prescription fill,
which was consistent with persistence levels (Figure 8). A significant proportion of patients only
filled the very first prescription without any refills during the 12-month follow-up period.
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500
0
5
10
15
20
25
30
Percent
ttldsup
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
0
5
10
15
20
25
30
Percent
num_fills
35

Figure 7. Medication Possession Ratio (MPR) for any cholesterol medication during 12-
month follow-up period for primary adherent cholesterol patients

Figure 8. Persistence (time to discontinuation) in primary adherent cholesterol patients

An unadjusted comparison of baseline patient characteristics revealed significant
differences among adherence groups (Table 6). Adherent patients were on average the oldest
and sickest in comparison to nonadherent groups. Primary nonadherent patients were
significantly more likely to be of minority race and lower incomes. Adherent patients were more
likely to be secondary prevention patients, have prior cholesterol medication, and switch or
augment therapy after the index date. Furthermore, adherent patients were more likely to have
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
0
5
10
15
20
25
30
Percent
mpr
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 250 260 270 280 290 300 310 320 330 340 350 360 370
0
10
20
30
40
50
P e rce n t
d a ys_ th e ra p y
36

renal disease, paraplegia, CHF, pulmonary disease, depression, rheumatic disease, peripheral
vascular disease, cancer, osteoporosis, and diabetes. Lastly, adherent patients were significantly
more likely to have prior utilization of prescriptions, ER visits, and hospitalizations.
Table 6. Unadjusted comparison of baseline patient characteristics among cholesterol
adherence groups
Characteristic
Primary
nonadherent
(n= 2,557)
Secondary
nonadherent
(n=10,771)
Adherent
(n=7,963)
Male (%) 1,329 (52.0%) 5,494 (51.0%) 4,102 (51.5%)
Mean age (SD)*** 56.2 (12.6) 56.5 (12.8) 61.1 (12.1)
Race ethnicity (%)***
White
1,123 (43.9%) 4,219 (39.2%) 4,487 (56.4%)
Hispanic
854 (33.4%) 4,207 (39.1%) 1,983 (24.9%)
Black
307 (12.0%) 1,306 (12.1%) 618 (7.8%)
Asian/pacific islander
258 (10.1%) 979 (9.1%) 839 (10.5%)
Multi-racial
13 (0.51%) 40 (0.37%) 24 (0.30%)
American Indian/ Alaskan
Native 2 (0.08%) 20 (0.19%) 12 (0.15%)
Patient household income***

$22,000 or less
15 (0.6%) 121 (1.1%) 38 (0.5%)
$22,000 to 45,000
1,201 (47.0%) 5,226 (48.5%) 3,293 (41.4%)
$45,000 to $55,000
519 (20.3%) 2,365 (22.0%) 1,787 (22.4%)
$55,000 or more
743 (29.1%) 2,675 (24.8%) 2,607 (32.7%)
Unknown
79 (3.1%) 384 (3.6%) 238 (3.0%)
Secondary prevention (%)***
189 (7.4%) 1,007 (9.4%) 1,379 (17.3%)
Use of cholesterol medications
prior to index date (%)*** 272 (10.6%) 1,325 (12.3%) 2,301 (28.9%)
Subsequent use of other
cholesterol medications after
index date (%)*** 658 (25.7%) 1,207 (11.2%) 2,196 (27.6%)
Average (SD) Charlson
Comorbidity Index*** 1.14 (1.80) 1.44 (1.96) 1.81 (2.22)
Baseline patient comorbidities

Renal disease***
230 (9.0%) 1,135 (10.5%) 1,293 (16.2%)
Migraine
99 (3.9%) 472 (4.4%) 318 (4.0%)
Alzheimer's disease
13 (0.5%) 49 (0.5%) 55 (0.7%)
Paraplegia*
18 (0.7%) 57 (0.5%) 70 (0.9%)
CHF***
77 (3.0%) 423 (3.9%) 481 (6.0%)
Pulmonary disease***
384 (15.0%) 2,032 (18.9%) 1,616 (20.3%)
Depression**
346 (13.5%) 1,786 (16.6%) 1,416 (17.8%)
Rheumatic disease*
41 (1.6%) 202 (1.9%) 186 (2.3%)
37

Table 6 (Continued)
Characteristic
Primary
nonadherent
(n= 2,557)
Secondary
nonadherent
(n=10,771)
Adherent
(n=7,963)
Peripheral vascular disease***
154 (6.0%) 831 (7.7%) 911 (11.4%)
Peptic ulcer disease
26 (1.0%) 129 (1.2%) 120 (1.5%)
Dementia**
10 (0.4%) 23 (0.2%) 47 (0.6%)
HIV
2 (0.08%) 20 (0.19%) 22 (0.28%)
Cancer***
93 (3.6%) 466 (4.3%) 464 (5.8%)
Metastatic cancer
19 (0.7%) 102 (1.0%) 88 (1.1%)
Mild liver disease**
112 (4.4%) 666 (6.2%) 483 (6.1%)
Moderate to severe liver
disease 1 (0.04%) 26 (0.24%) 19 (0.24%)
Osteoporosis***
166 (6.5%) 939 (8.7%) 843 (10.6%)
Diabetes without
complications*** 612 (23.9%) 3,429 (31.8%) 2,587 (32.5%)
Diabetes with complications***
253 (9.9%) 1,328 (12.3%) 1,146 (14.4%)
Average (SD) prescription
count in prior year*** 5.2 (5.1) 7.4 (5.8) 8.9 (6.3)
Average (SD) clinic visit count
in prior year*** 7.2 (8.2) 8.7 (9.7) 10.1 (10.7)
ER visit in prior year***
467 (18.3%) 2,711 (25.2%) 2,189 (27.5%)
Hospitalization in prior year***
241 (9.4%) 1,228 (11.4%) 1,271 (16.0%)
Medicare insurance***
590 (23.1%) 2,592 (24.1%) 2,924 (36.7%)
Medicaid insurance*
8 (0.3%) 77 (0.7%) 41 (0.5%)
Commercial insurance
2,531 (99.0%) 10,651 (98.9%) 7,888 (99.1%)
Primary subscriber**
1,960 (76.7%) 8,004 (74.3%) 6,099 (76.6%)
Dual coverage (from spouse)
47 (1.8%) 253 (2.4%) 199 (2.5%)
No. patients with baseline LDL
value (%)*** 2,492 (97%) 10,255 (95%) 7,682 (96%)
Patient's average baseline LDL
value‡*** 147.9 (37.8) 148.6 (39.4) 141.0 (41.7)
***p<0.0001, **p<0.01, *p<0.05

In general, secondary nonadherent patients were sicker than primary nonadherent
patients but not as sick as adherent patients. Primary nonadherent patients were the healthiest
and most likely to take cholesterol medications for primary prevention, have the lowest CCI, and
least likely to have renal disease, CHF, pulmonary disease, depression, rheumatic disease,
peripheral vascular disease, cancer, osteoporosis, and diabetes. In addition, primary
nonadherent patients had the least prior utilization of prescriptions, ER visits, and
hospitalizations.
38

5.3.2 Lab values
At least 95% of all patients had at least one LDL test during the baseline period (Table 7).
During the follow-up period, patients in the secondary nonadherent and adherent groups were
more likely to have at least one LDL test than primary nonadherent patients. Adherent patients
were most likely to have LDL tests available both before and after the index date (94%), followed
by secondary nonadherent patients (85%) and primary nonadherent patients (77%) (Figure 9).
LDL testing per patient increased after the index date in all patient groups.
Table 7. Comparison of LDL testing among cholesterol adherence groups

Primary
nonadherent
(n= 2,557)
Secondary
nonadherent
(n=10,771)
Adherent
(n=7,963)
Patients with no LDL tests during study
period 20 (0.8%) 82 (0.8%) 16 (0.2%)
Patients with at least one LDL test
during the baseline period‡ 2492 (97%) 10255 (95%) 7682 (96%)
Patients with at least one LDL test
during the follow-up period
&
2020 (79%) 9576 (89%) 7719 (97%)
Patients with at least one LDL test
during baseline and follow-up periods 1975 (77%) 9142 (85%) 7454 (94%)
Total number of LDL tests during
baseline period 4,386 17,867 14,927
Total number of LDL tests during
follow-up period 4,567 24,314 21,931
Average (SD) number of LDL tests per
patient during baseline period*** 1.76 (0.98) 1.74 (0.98) 1.94 (1.11)
Average (SD) number of LDL tests per
patient during follow-up period*** 2.26 (1.40) 2.54 (1.50) 2.84 (1.56)
***p<0.0001

39

Figure 9. Average number of LDL tests per patient by adherence group and months from
index date

Change in average lab value was examined by adherence group among the 18,571
patients with lab values available both before and after the index date (Figure 10). During the 12-
month baseline period, average (SD) LDL values were lowest in the adherent group (140.7 mg/dL
(41.6)) and significantly higher in the primary nonadherent (144.6 mg/DL (37.7)) and secondary
nonadherent groups (147.6 mg/dL (39.3)) (Table 8). LDL values dropped by an average (SD) of
45.3 (38.2) mg/dL, 28.3 (34.6) mg/dL, and 14.9 (27.7) mg/dL in the adherent, secondary
adherent, and primary nonadherent groups, respectively, during the 18-month period after the
index date. These decreases were significantly different from zero and the differences in change
among groups were statistically significant. The difference in difference in lab values between the
primary nonadherent and adherent was 30.5 mg/dL and statistically significant. The difference in
difference in lab values between the primary nonadherent and secondary nonadherent was 13.4
mg/dL and also statistically significant.

40

Figure 10. Average LDL values over time by adherence group (n=18,571)

Table 8. Comparison of change in LDL values among cholesterol adherence groups
(n=18,571)

Primary
nonadherent
(n=1,975)
Secondary
nonadherent
(n=9,142)
Adherent
(n=7,454)
Difference
PNA-
Adherent
Difference
PNA-SNA
Average (SD)
baseline LDL value
(most recent
value)*** 144.6 (37.7) 147.6 (39.3) 140.7 (41.6) 3.9 -3.0
Average (SD) LDL
value during 18-
month follow-up
period*** 129.7 (35.2) 119.3 (34.7) 95.3 (28.2) 34.4 10.4
Average (SD)
change in LDL
value within a
patient***
-14.9
(27.7)‡***
-28.3
(34.6)‡***
-45.3
(38.2)‡*** 30.5*** 13.4***
***p<0.0001
‡paired t-test tests whether average change within patient was significantly different from zero (p<0.05)
5.3.3 Event outcomes
Event outcomes of new and acute MI or stroke were examined in all 19,284 primary
prevention patients. ER visits and hospitalizations were examined in the entire cholesterol cohort
of 21,291 patients. The rate of ER visits and hospitalizations per patient over time is displayed
over time by adherence group in Figures 11 and 12, respectively.
41

Figure 11. Rate of ER visits per patient over time by adherence group (n=21,291)

Figure 12. Rate of hospitalizations per patient over time by adherence group (n=21,291)

A comparison of unadjusted rates of events revealed that primary nonadherent patients
were least likely to be seen in the ER or be hospitalized (Table 9). The average number of ER
visits per patient was also significantly lower in the primary nonadherent group in comparison to
the secondary nonadherent and adherent patients.

Rate per patient per month
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
0.11
Months from index date
-12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
adh_grp pna sa sna
Rate per patient per month
0.00
0.01
0.02
0.03
0.04
0.05
0.06
Months from index date
-12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
adh_grp pna sa sna
42

Table 9. Comparison of unadjusted rates of event outcomes occurring between 12 and 18
months after index date among cholesterol adherence groups
Outcome
Primary
nonadherent
(n= 2,557)
Secondary
nonadherent
(n=10,771)
Adherent
(n=7,963) TOTAL
No. primary prevention
patients (%) with at least one
MI or stroke 10 (0.4%) 48 (0.5%) 26 (0.4%) 84
No. patients (%) with at least
one ER visit** 287 (11.2%) 1,504 (14.0%) 1,092 (13.7%) 2,883
No. patients (%) with at least
one hospitalization** 169 (6.6%) 757 (7.0%) 642 (8.1%) 1,568
Average no. ER visits per
patient** 0.16 (0.64) 0.21 (0.75) 0.21 (0.65)
Average no. hospitalizations
per patient 0.14 (1.12) 0.10 (0.59) 0.12 (0.60)

After adjusting for all observed patient and prescriber characteristics, the risk of MI or
stoke was not statistically higher in the primary nonadherent or secondary nonadherent group
relative to the adherent reference group (Table 10). However, the risk of ER visit was
significantly higher by 23% in secondary nonadherent patients in comparison to adherent patients
and was statistically significant (HR=1.23, 95% CI: 1.13, 1.33). The risk of hospitalization
depended on whether patients had previously used cholesterol medications during the baseline
period. In patients newly initiating cholesterol therapy, the risk of hospitalization was similar
among adherence groups. However, patients with prior cholesterol medication were associated
with significantly higher risk of hospitalization if they were primary nonadherent rather than
adherent (HR=1.69, 95% CI: 1.19, 2.41).
Results from the sensitivity analysis which excluded baseline LDL value from the
regression model were similar to those from the main analysis except for hospitalization risk.
Excluding baseline LDL values allowed inclusion of all 21,291 patients in the analysis, which
resulted in a significantly higher risk of hospitalization for primary nonadherent (HR=1.69, 95% CI:
1.18, 2.40) and secondary nonadherent patients (HR=1.27, 95% CI: 1.02, 1.58).

43

Table 10. Adjusted hazard ratios for event outcomes occurring between 12 and 18 months
after index date in cholesterol

MI or stroke
n=18,026
ER visit
n=20,429
Hospitalization
n=20,429
Primary nonadherent 1.29 (0.61, 2.75) 1.13 (0.99, 1.30)
Secondary nonadherent,
MPR<80% 1.46 (0.89, 2.41) 1.23 (1.13, 1.33)
Adherent, MPR≥80%
(reference group) 1.00 1.00
Patients without cholesterol medication use prior to index date
Primary nonadherent 1.05 (0.86, 1.29)
Secondary nonadherent,
MPR<80% 1.01 (0.88, 1.15)
Adherent, MPR≥80%
(reference group) 1.00
Patients with cholesterol medication use prior to index date
Primary nonadherent 1.69 (1.19, 2.41)
Secondary nonadherent,
MPR<80% 1.24 (1.00, 1.54)
Adherent, MPR≥80%
(reference group) 1.00

5.4 Results for diabetes cohort
5.4.1 Descriptive results
A total of 10,229 patients satisfied all Phase II inclusion and exclusion criteria for the
diabetes analysis (Figure 13). Only 3,036 patients (30%) were identified as primary prevention
patients. Overall, 528 patients were primary nonadherent, which resulted in an overall PNA rate
of 5.2%. The PNA rate was similar in primary prevention patients (5.2%) and secondary
prevention patients (5.1%). About 60% of patients who were primary adherent filled their new
prescription on the day it was written.
About 65% of primary adherent patients filled more than 365 days supplied during the 12-
month period immediately following the index date (Figure 14) and 19% of primary adherent
patients filled four prescriptions during this period (Figure 15). About 10% of primary adherent
patients discontinued therapy around 100 days (Figure 16). MPR values were also consistent
with these results in that about 60% of patients had an MPR nearly 100% (Figure 17). A total of
44

7,053 patients were adherent and 2,648 patients had an MPR <80% and classified as secondary
nonadherent.
Figure 13. Identification of diabetes study cohort

Figure 14. Total days supply during 12-month follow-up period for primary adherent
diabetes

0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
2.5
5.0
7.5
10.0
12.5
15.0
17.5
20.0
22.5
Percent
ttldsup
45

An unadjusted comparison of baseline patient characteristics revealed several significant
differences among adherence groups (Table 11). Adherent patients were, again, the oldest in
comparison to other adherence groups on average. Primary nonadherent patients were more
likely to be of minority race. As before, adherent patients were on average the sickest in terms of
previous diabetes complications, prior insulin use, and CCI. Specifically, adherent patients had
significantly higher proportions of patients with baseline comorbidities of renal disease,
depression, and cancer. In addition, adherent patients had significantly higher prior utilization of
prescriptions and clinic visits.
Figure 15. Total number of fills dispensed during 12-month follow-up period for primary
adherent diabetes patients

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57
0
2.5
5.0
7.5
10.0
12.5
15.0
17.5
20.0
Percent
num_fills
46

Figure 16. Persistence (time to discontinuation) in primary adherent diabetes patients

Figure 17. Medication Possession Ratio (MPR) for any diabetes medication during 12-
month follow up period for primary adherent diabetes patients

0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340 360 380 400 420 440 460 480 500 520 540
0
1
2
3
4
5
6
7
8
P e rce n t
d a ys_ th e ra p y
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
0
10
20
30
40
50
60
70
Percent
mpr
47

Table 11. Unadjusted comparison of baseline patient characteristics among diabetes
adherence groups
Characteristic
Primary
nonadherent
(n= 521)
Secondary
nonadherent
(n=2,642)
Adherent
(n=6,974)
Male (%)* 261 (50.1%) 1306 (49.4%) 3643 (52.2%)
Mean age (SD)*** 56.5 (12.) 52.7 (14.8) 58.4 (12.7)
Race ethnicity (%)***
White
182 (34.9%) 828 (31.3%) 2876 (41.2%)
Hispanic
204 (39.2%) 1188 (45.0%) 2513 (36.0%)
Black
67 (12.9%) 310 (11.7%) 728 (10.4%)
Asian/pacific islander
65 (12..5%) 301 (11.4%) 824 (11.8%)
Other
3 (0.6%) 15 (0.6%) 33 (0.5%)
Patient household income

$22,000 or less
10 (1.9%) 33 (1.3%) 64 (0.9%)
$22,000 to 45,000
260 (49.9%) 1,315 (49.8%) 3317 (47.6%)
$45,000 to $55,000
106 (20.4%) 583 (22.1%) 1564 (22.4%)
$55,000 or more
128 (24.6%) 623 (23.6%) 1802 (25.8%)
Unknown
17 (3.3%) 88 (3.3%) 227 (3.3%)
Previous complications of diabetes (%)***
425 (81.6%) 1995 (75.5%) 6105 (87.5%)
Use of diabetes medications prior to index
date (%)***
259 (49.7%) 533 (20.2%) 4149 (59.5%)
Subsequent use of other diabetes
medications after index date (%)*** 76 (14.6%) 609 (23.1%) 4955 (71.1%)
Use of insulin prior to index date (%)***
26 (5.0%) 49 (1.9%) 541 (7.8%)
Subsequent use of insulin after index date
(%)*** 48 (9.2%) 131 (5.0%) 1192 (17.1%)
Average (SD) Charlson Comorbidity
Index*** 2.5 (2.3) 2.2 (1.9) 3.0 (2.2)
Baseline patient comorbidities

Renal disease***
90 (17.3%) 301 (11.4%) 1344 (19.3%)
Migraine
19 (3.7%) 100 (3.8%) 203 (2.9%)
Alzheimer's disease
0 11 (0.4%) 32 (0.5%)
Paraplegia
2 (0.4%) 17 (0.6%) 46 (0.7%)
CHF*
27 (5.2%) 94 (3.6%) 340 (4.9%)
Pulmonary disease
101 (19.4%) 511 (19.3%) 1384 (19.9%)
Depression**
67 (12.9%) 417 (15.8%) 1224 (17.6%)
Rheumatic disease
4 (0.8%) 48 (1.8%) 122 (1.8%)
Peripheral vascular disease***
37 (7.1%) 140 (5.3%) 603 (8.7%)
Peptic ulcer disease
5 (1.0%) 28 (1.1%) 68 (1.0%)
Dementia
1 (0.2%) 11 (0.4%) 24 (0.3%)
48

Table 11 (Continued)
Characteristic
Primary
nonadherent
(n= 521)
Secondary
nonadherent
(n=2,642)
Adherent
(n=6,974)
HIV
0 3 (0.1%) 15 (0.2%)
Cancer**
14 (2.7%) 94 (3.6%) 346 (5.0%)
Metastatic cancer
4 (0.8%) 20 (0.8%) 73 (1.1%)
Mild liver disease*
27 (5.2%) 218 (8.3%) 608 (8.7%)
Moderate to severe liver disease
0 10 (0.4%) 26 (0.4%)
Osteoporosis*
28 (5.4%) 141 (5.3%) 462 (6.6%)
Average (SD) prescription count in prior
year*** 7.5 (5.7) 9.0 (6.0) 11.2 (6.2)
Average (SD) clinic visit count in prior
year*** 8.1 (7.3) 9.1 (9.8) 10.3 (10.0)
ER visit in prior year*
103 (19.8%) 682 (25.8%) 1716 (24.6%)
Hospitalization in prior year
57 (10.9%) 296 (11.2%) 850 (12.2%)
Medicare insurance***
124 (23.8%) 513 (19.4%) 2025 (29.0%)
Medicaid insurance
5 (1.0%) 21 (0.8%) 54 (0.8%)
Commercial insurance
514 (98.7%) 2616 (99.0%) 6888 (98.8%)

Primary subscriber**
393 (75.4%) 1918 (72.6%) 5303 (76.0%)
Dual coverage (from spouse)
8 (1.5%) 78 (3.0%) 208 (3.0%)
No. patients with baseline Ha1c value
(%)*** 475 (91.2%) 1976 (74.8%) 6192 (88.8%)
Patient's average baseline HA1c
value***‡ 8.2 (1.5) 8.1 (1.8) 8.5 (1.7)
***p<0.0001, **p<0.01, *p<0.05
‡using most recent lab value prior to index date

Primary and secondary nonadherent patients were healthier than adherent patients.
Primary nonadherent patients had the least prior utilization of prescriptions, clinic visits, and ER
visits. They were also least likely to augment/switch therapy after the index date and significantly
less like to have baseline comorbidities of depression and cancer. Secondary nonadherent
patients were significantly less likely to have had diabetes complications or antidiabetic use or
insulin use prior to the index date in comparison to other groups. They were also least likely to
start insulin therapy after the index date. Secondary nonadherent patients also had the lowest
CCI on average, and least likely to have baseline comorbidities of renal disease, CHF, and
peripheral vascular disease.
49

5.4.2 Lab values
Ha1c testing rates during both baseline and follow-up periods were lower in the
secondary nonadherent group (81% during baseline and 86% during follow-up) in comparison to
primary nonadherent and adherent groups (at least 93% during baseline and follow-up) (Table
12). Again, adherent patients were more likely to have Ha1c tests available both before and after
the index date (91%), now followed by primary nonadherent patients (88%) and lastly, secondary
nonadherent patients (74%). On average, the number of Ha1c tests per patient increased after
the index date for all 3 patient groups. However, the average number of Ha1c tests per patient
remained the highest in the adherent group both before and after the index date (Figure 18).
Table 12. Comparison of Ha1c testing among diabetes adherence groups

Primary
nonadherent
(n= 521)
Secondary
nonadherent
(n=2,642)
Adherent
(n=6,974)
Patients with no lab tests during study
period 10 176 28
Patients with at least one Ha1c test
during baseline period 484 2,130 6,485
Patients with at least one Ha1c lab test
during follow-up period 484 2,284 6,866
Patients with at Ha1c test(s) during both
baseline and follow-up periods 457 (88%) 1948 (74%) 6378 (91%)
Total number of labs during baseline
period 1,121 3,974 15,530
Total number of labs during follow-up
period 1,455 6,797 25,411
Average (SD) number of labs per
patient during baseline period*** 2.3 (1.1) 1.9 (1.1) 2.4 (1.3)
Average (SD) number of lab per patient
during follow-up period*** 3.0 (1.5) 3.0 (1.6) 3.7 (1.6)

50

Figure 18. Average number of Ha1c tests per patient by adherence group and months from
index date

Changes in Ha1c were examined among the 8,783 patients with lab values available both
before and after the index date. Average Ha1c values are displayed over time by adherence
group in Figure 19. During the 12-month baseline period, average (SD) Ha1c values were
highest in the adherent group (8.60% (1.82)) in comparison to primary nonadherent (8.20%
(1.51)) and secondary nonadherent patients (8.21 (1.92)) (Table 13). Ha1c values dropped by an
average (SD) of 0.24% (1.16) in the primary nonadherent, 0.78% (1.58) in the secondary
nonadherent, and 1.05% (1.59) in the adherent. These decreases were significantly different
from zero and the differences in change among groups were statistically significant. The
difference in difference in lab values between the primary nonadherent and adherent was 0.83%
and statistically significant. The difference in difference in lab values between the primary
nonadherent and secondary nonadherent was 0.55% and also statistically significant.

Number of labs per patient
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Months from index date
-12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
adh_grp pna sa sna
51

Figure 19. Average Ha1c values over time by adherence group

Table 13. Comparison of change in Ha1c (%) values among diabetes adherence groups

Primary
nonadherent
(n=457)
Secondary
nonadherent
(n=1,948)
Adherent
(n=6,378)
Difference,
PNA-
Adherent
Difference,
PNA-SNA
Average (SD)
baseline Ha1c
value (most recent
value)*** 8.20 (1.51) 8.21 (1.92)
8.60
(1.82) -0.40 -0.01
Average (SD)
Ha1c value during
follow-up period*** 7.97 (1.52) 7.43 (1.51)
7.54
(1.29) 0.43 0.54
Average (SD)
change in Ha1c
value within a
patient***
-0.24
(1.16)‡**
-0.78
(1.58)‡***
-1.05
(1.59)‡***
Change in
average Ha1c
value -0.23 -0.78 -1.06 0.83*** 0.55***
***p<0.0001, **p<0.01
‡paired t-test tests whether average change within patient was significantly different from zero

5.4.3 Event outcomes
New and acute cases of diabetes complications were examined in a total of 3,036
primary prevention patients, while ER visits and hospitalizations were examined in the entire
diabetes cohort of 10,229 patients. The rate of ER visits and hospitalizations per patient was
Average Ha1c
6.9
7.0
7.1
7.2
7.3
7.4
7.5
7.6
7.7
7.8
7.9
8.0
8.1
8.2
8.3
8.4
8.5
8.6
8.7
8.8
Months from index date
-12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
adh_grp pna sa sna
52

displayed over time by adherence group in Figures 20 and 21, respectively. Comparison of
unadjusted rates of outcome events revealed that adherent patients were significantly most likely
to have diabetes complications (Table 14). However, the average number of hospitalizations per
patient was significantly higher in primary nonadherent patients than adherent patients.
After adjusting for all observed patient and prescriber characteristics, the risk of diabetes
complication and hospitalization was not significantly higher in the primary
Figure 20. Rate of ER visits per patient over time by adherence group

Figure 21. Rate of hospitalizations per patient over time by adherence group

Rate per patient per month
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
0.11
Months from index date
-12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
adh_grp pna sa sna
Rate per patient per month
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
0.020
0.022
0.024
0.026
0.028
0.030
0.032
0.034
0.036
0.038
0.040
0.042
0.044
0.046
Months from index date
-12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
adh_grp pna sa sna
53

nonadherent and secondary nonadherent group relative to the adherent reference group (Table
15). However, the risk of ER visits was significantly higher by 30% in secondary nonadherent
patients relative to adherent patients (HR=1.30, 95% CI: 1.11, 1.51). Results from the sensitivity
analysis which excluded baseline Ha1c value from the regression model were similar to those
from the main analysis except for the outcome of diabetes complications.
Table 14. Comparison of unadjusted rates of event outcomes occurring between 12 and 18
months after index date among diabetes adherence groups
Outcome
Primary
nonadherent
(n= 521)
Secondary
nonadherent
(n=2,642)
Adherent
(n=6,974)
No. primary prevention patients (%) with at
least one diabetes complication event** 22 (22.9%) 92 (14.2%) 176 (20.3%)
No. patients (%) with at least one ER visit 54 (10.4%) 372 (14.1%) 986 (14.1%)
No. patients (%) with at least one
hospitalization* 37 (7.1%) 181 (6.9%) 586 (8.4%)
Average no. ER visits per patient 0.15 (0.51) 0.21 (0.65) 0.22 (0.71)
Average no. hospitalizations per patient* 0.20 (1.92) 0.09 (0.41) 0.13 (0.72)
**p<0.01, *p<0.05

Table 15. Adjusted hazard ratios for event outcomes occurring between 12 and 18 months
after index date in diabetes

Diabetes
complications
n=1,127
ER visit
n=8,643
Hospitalization
n=8,643
Primary nonadherent 1.42 (0.82, 2.45) 1.01 (0.75, 1.35) 1.00 (0.69, 1.44)
Secondary nonadherent 0.91 (0.65, 1.27) 1.30 (1.11, 1.51) 0.83 (0.67, 1.02)
Adherent (reference group) 1.00 1.00 1.00

5.5 Discussion
The overall rate of PNA in cholesterol (12.0%) and diabetes (5.2%) were relatively low in
this patient population. However, secondary nonadherence rates for cholesterol medications
were poor as shown in the descriptive results for MPR levels and as 25% of patients who were
initially primary adherent failed to refill their prescription(s). In comparison, only 10% of diabetes
patients who were initially primary adherent to their first medication did not refill their
prescription(s). About 51% of the cholesterol population was secondary nonadherent and only
37% was adherent. In comparison, only 26% of the diabetes population was secondary
54

nonadherent and 69% was adherent. Thus, cholesterol patients were more likely to be primary
and secondary nonadherent to their medications relative to diabetes patients.
When comparing baseline patient characteristics between the primary nonadherent,
secondary nonadherent, and adherent patients, it was apparent that these patient groups were
different. For both chronic disease states, the adherent patient group was on average sicker than
the other two nonadherent patient groups at baseline. Sicker patients were more likely to be
adherent and this may lead to biased results for health outcomes if differences in observed
baseline health among patient groups are inadequately controlled.
The unadjusted rates of ER visits and hospitalization were lowest in primary nonadherent
patients for cholesterol. In diabetes, the unadjusted rate of diabetes complications was highest in
adherent patients. These results are likely due to sicker patients selecting to be more adherent,
and these results are expected to change once controlling for patients’ baseline disease states.
Although the average number of hospitalizations per patient was significantly higher in primary
nonadherent patients than in adherent patients, this anomaly may be due to the relatively small
sample size of primary nonadherent patients of 521 in the diabetes analysis.
Using a multivariate Cox proportional hazard model, the effect of PNA and secondary
nonadherence on outcomes was estimated while adjusting for many baseline patient and
prescriber characteristics. For cholesterol and diabetes, the risk of ER visits was higher for
secondary nonadherent patients relative to adherent patients after controlling for differences in
baseline characteristics among patient groups. The risk of hospitalization was also significantly
higher in primary nonadherent cholesterol patients with prior cholesterol medications (i.e. patients
who were augmenting or switching therapy).
For both cholesterol and diabetes, lab values decreased by a significant amount after the
index date for all adherence groups, including primary nonadherent patients. Despite not taking
their prescriptions, primary nonadherent patients experienced decreases in lab values that were
statistically significant but small in comparison to secondary nonadherent and adherent patients.
Potentially, lab values in primary nonadherent patients may have decreased after the index date
55

due to lifestyle changes in diet and exercise. Another potential explanation for the small but
significant decrease in lab values in the primary nonadherent could be due to regression to the
mean, where some patients found to have extreme lab levels on the index date but on average
their lab values were lower.
LDL and Ha1c levels decreased the most in adherent patients, followed by secondary
nonadherent patients, and lastly, primary nonadherent patients. Results from the difference-in-
difference analysis revealed that the lab values between primary nonadherent and adherent
patients and between primary nonadherent and secondary nonadherent patients were still
significant after controlling for both group-specific and time-specific effects. This is consistent
with what we would expect since more drug exposure was consistently associated with greater
beneficial effects on lab values. LDL and Ha1c levels decreased most rapidly during the first
three months after the index date, and increased slightly to a steady state level after four months.
This may be due to secondary nonadherent patients stopping the drug after the first prescription,
which was likely for a 100 days supply.
Lab testing rates were highest in adherent patients and lower for nonadherent patients.
The difference seen in testing rates among adherence groups was greater for Ha1c than LDL. It
is likely that patients who do not have lab tests also do not take their medications. The average
lab values may actually be higher than what was reported for primary and secondary
nonadherent patients when accounting for the missing labs. This would make the difference in
lab values between the adherent and nonadherent groups even greater.
The main strength of this study was the use of data from an integrated system, which
provided the opportunity to control for many patient and prescriber characteristics, including lab
values. Any bias in results due to endogeneity was limited since most patient characteristics
were observable and controlled for in the multivariate Cox Proportional Hazard model. This
further confirmed by the results from the multivariate analysis which revealed an increased risk of
event outcomes in primary and secondary nonadherent patients.
56

There are two main limitations in this study. First, as in Phase I, the generalizability of
our results is limited to patient populations in integrated healthcare systems, and may not be
generalizable to Medicaid or uninsured patients. Second, an 18-month follow-up period may be
relatively short when examining the long-term outcomes associated with adherence to
antihyperlipidemics and antidiabetics. However, this study focused on short-term outcomes of
changes in lab values, and found significant differences in ER visits and hospitalizations despite
the relatively short follow-up period.
Overall, patients who were primary or secondary nonadherent to their cholesterol or
diabetes medication were associated with worse health outcomes. In cholesterol, secondary
nonadherent patients were at increased risk for ER visits, and primary nonadherent patients were
at increased risk of hospitalization. Both primary and secondary nonadherent patients should be
targeted by clinicians for interventions addressing hyperlipidemia. For diabetes, only secondary
nonadherent patients were found to have a significantly increased risk for ER visits.
5.6 Conclusions
Primary and secondary nonadherence to cholesterol and diabetes medications were
associated with worse outcomes in terms of lab values, ER visits and hospitalizations. The
results from this study may be helpful to clinicians and policy makers when designing and
implementing interventions to improve medication adherence. The results of this research
advance the study of medication adherence and directly improve clinical care. The results from
this research may help clinical interventionists to target patients who are at risk of PNA and
improve the effectiveness of interventions.
57

BIBLIOGRAPHY
Bates TR, Connaughton VM, Watts GF. Non-adherence to statin therapy: a major challenge for
preventive cardiology. Expert Opin. Pharmacother. (2009) 10(18): 2973-2985.
Bazargan M, Barbre AR, Hamm V. Failure to have prescriptions filled among Black elderly. J
Aging Health. 1993 May;5(2):264-82.
Beardon PH, McGilchrist MM, McKendrick AD, McDevitt DG, MacDonald TM. Primary non-
compliance with prescribed medication in primary care. BMJ. 1993 Oct 2;307(6908):846-8.
Begg D. Do patients cash prescriptions? An audit in one practice. J R Coll Gen Pract. 1984
May;34(262):272-4.
Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic
comorbidity in longitudinal studies: development and validation. J Chron Dis. 1987; 40(5): 373-
383.
Copher R, Buzinec P, Zarotsky V, Kazis L, Iqbal SU, Macarios D. Physician perception of patient
adherence compared to patient adherence of osteoporosis medications from pharmacy claims.
Curr Med Res Opin. 2010 Apr;26(4):777-85.

Cramer JA, Benedict A, Muszbek N, Keskinaslan A, Khan ZM. The significance of compliance
and persistence in the treatment of diabetes, hypertension and dyslipidaemia: a review. Int J Clin
Pract. 2008 Jan;62(1):76-87. Epub 2007 Nov 5.

Ekedahl A, Månsson N. Unclaimed prescriptions after automated prescription transmittals to
pharmacies. Pharm World Sci. 2004 Feb;26(1):26-31.
Fischer MA, Stedman MR, Lii J, Vogeli C, Shrank WH, Brookhart MA, Weissman JS. Primary
medication non-adherence: analysis of 195,930 electronic prescriptions. J Gen Intern Med. 2010
Apr;25(4):284-90. Epub 2010 Feb 4.
Freeman CP and Guly HR. Do accident and emergency patients collect their prescribed
medication? Arch Emer Med 1985 (2): 41-43.
Gadkari AS, McHorney CA. Medication nonfulfillment rates and reasons: narrative systematic
review. Curr Med Res Opin. 2010 Mar;26(3):683-705. Review.

Gardner TL, Dovey SM, Tilyard MW, Gurr E. Differences between prescribed and dispensed
medications. N Z Med J. 1996 Mar 8;109(1017):69-72.
Gatwood J, Erickson SR. Medication adherence: the search for interventions that work. Am J
Manag Care. 2010;16(12):925-6.
Hankó B, Kázmér M, Kumli P, Hrágyel Z, Samu A, Vincze Z, Zelkó R. Self-reported medication
and lifestyle adherence in Hungarian patients with Type 2 diabetes. Pharm World Sci. 2007
Apr;29(2):58-66. Epub 2006 Dec 23.
Haynes RB, Ackloo E, Sahota N, McDonald HP, Yao X. Interventions for enhancing medication
adherence. Cochrane Database Syst Rev. 2008 Apr 16;(2):CD000011.
Haynes RB, McKibbon KA, Kanani R. Systematic review of randomised trials of interventions to
assist patients to follow prescriptions for medications. Lancet. 1996;348(9024):383-386.
58

Haynes RB, Taylor DW, Sackett DL, et al. Compliance in health care. Baltimore: Johns Hopkins
University Press, 1979.
Haynes RB, Yao X, Degani A, Kripalani S, Garg A, McDonald HP. Interventions to enhance
medication adherence. Cochrane Database Syst Rev. 2005 Oct 19;(4):CD000011.
Health Expenditure: Total expenditure on health, % gross domestic product. OECD Health Data
2010.

Ho P, Spertus J, Masoudi F, et al. Impact of medication therapy discontinuation on mortality after
myocardial infarction. Arch Intern Med 2006; 166: 1842-7.

Jackevicius CA, Li P, Tu JV. Prevalence, predictors, and outcomes of primary nonadherence
after acute myocardial infarction. Circulation. 2008 Feb 26;117(8):1028-36.
Janz NK and Becker MH. The Health Belief Model: A Decade Later. Health Educ Behav 1984;
11:1 (1-47).
Jones and Britten. Why do some patients not cash their prescriptions? Brit J Gen Prac: Jan 1998
(903-905).
Karter AJ, Parker MM, Adams AS, Moffet HH, Schmittdiel JA, Ahmed AT, Selby JV. Primary non-
adherence to prescribed medications. J Gen Intern Med. 2010 Aug;25(8):763; author reply 765.
Karter AJ, Parker MM, Moffet HH, Ahmed AT, Schmittdiel JA, Selby JV. New prescription
medication gaps: a comprehensive measure of adherence to new prescriptions. Health Serv Res.
2009 Oct;44(5 Pt 1):1640-61. Epub 2009 Jun 3.
Kennedy J, Tuleu I, Mackay K. Unfilled prescriptions of medicare beneficiaries: prevalence,
reasons, and types of medicines prescribed. JMCP 2008; 14(6):553-60.
Kimball C. Non-cooperation: an examination of factors leading to non-compliance in a
hypertension clinic. Psych J Univ. Ottawa 1980 Dec; 5-243-9.
Kirking MH, Kirking DM. Evaluation of unclaimed prescriptions in an ambulatory care pharmacy.
Hosp Pharm. 1993 Feb;28(2):90-1, 94, 102.
Kirking MH, Zaleon CR, Kirking DM. Unclaimed prescriptions at a university hospital’s ambulatory
care pharmacy. Am J Health Syst Pharm. 1995 Mar 1;52(5):490-5.
Lash S and Harding J. “Abandoned Prescriptions” A Quantitative Assessment of Their Cause. J
Managed Care Pharm 1995: 1: 193-199.
Lazar LD, Pletcher MJ, Coxson PG, Bibbins-Domingo K, Goldman L. Cost-Effectiveness of Statin
Therapy for Primary Prevention in a Low-Cost Statin Era. Circulation. 2011;124(2):146-53.
Liberman JN, Hutchins DS, Popiel RG, Patel MH, Jan SA, Berger JE. Determinants of Primary
Nonadherence in Asthma-Controller and Dyslipidemia Pharmacotherapy. Am J Pharm Benefits.
2010; 2(2):111-118.
Mahoney JJ. Value-based benefit design: using a predictive modeling approach to improve
compliance. J Manag Care Pharm 2008; 14 (6 Suppl. B): 3-8.

59

McDonnell PJ, Jacobs MR. Hospital admissions resulting from preventable adverse drug
reactions. Ann Pharmacother. 2002 Sep;36(9):1331-6.

Meyer B. Natural and Quasi-Experiments in Economics. J Business Econ Stats. 1995 Apr; 13(2):
151-161.

MRC/BHF Heart Protection Study of cholesterol lowering with simvastatin in 20,536 high-risk
individuals: a randomised placebo-controlled trial. Lancet. 2002;360(9326):7-22.

Mushlin A, Appel F. Diagnosing patient noncompliance. Arch Intern Med 1977; 137: 318-21.

Muszbek N, Brixner D, Benedict A, Keskinaslan A, Khan ZM. The economic consequences of
noncompliance in cardiovascular disease and related conditions: a literature review. Int J Clin
Pract. 2008 Feb;62(2):338-51.

Osterberg L, Blaschke T. Adherence to medication. N Engl J Med. 2005 Aug 4;353(5):487-97.

Pham AN, Datta SK, Weber TJ, Walter LC, Colón-Emeric CS. Cost-Effectiveness of Oral
Bisphosphonates for Osteoporosis at Different Ages and Levels of Life Expectancy. Journal of the
American Geriatrics Society. 2011.

Pound P, Britten N, Morgan M, Yardley L, Pope C, Daker-White G, Campbell R. Resisting
medicines: a synthesis of qualitative studies of medicine taking. Soc Sci Med. 2005 Jul;61(1):133-
55. Epub 2005 Jan 26.
Raebel MA, Ellis JL, Carroll NM, et al. Characteristics of Patients with Primary Non-adherence to
Medications for Hypertension, Diabetes, and Lipid Disorders. Journal of general internal
medicine. 2011.
Rashid A. Do patients cash prescriptions? Br Med J (Clin Res Ed). 1982 Jan 2;284(6308):24-6.
Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in
epidemiology. Epidemiology. 2000 Sep;11(5):550-60.
Roebuck MC, Liberman JN, Gemmill-Toyama M and Brennan TA. Medication Adherence Leads
to Lower Health Care Use and Costs Despite Increased Drug Spending. Health Affairs. 2011;
30(1): 91-99.
Rosenbaum PR and Rubin DB. The central role of the propensity score in observational studies
for causal effects. Biometrika 1983; 70 (1): 41-55.
Ross S, Samuels E, Gairy K, et al. A meta-analysis of osteoporotic fracture risk with medication
nonadherence. Value in health. 2011;14(4):571-81.
Senst BL, Achusim LE, Genest RP, Cosentino LA, Ford CC, Little JA, Raybon SJ, Bates DW.
Practical approach to determining costs and frequency of adverse drug events in a health care
network. Am J Health Syst Pharm. 2001 Jun 15;58(12):1126-32.
Shah NR, Hirsch AG, Zacker C, et al. Factors associated with first-fill adherence rates for diabetic
medications: a cohort study. Journal of general internal medicine. 2009;24(2):233-7.
60

Shah NR, Hirsch AG, Zacker C, et al. Predictors of first-fill adherence for patients with
hypertension. American journal of hypertension. 2009;22(4):392-6.

Sokol MC, McGuigan KA, Verbrugge RR, Epstein RS. Impact of medication adherence on
hospitalization risk and healthcare cost. Med Care. 2005 Jun;43(6):521-30.

Wamala S, Merlo J, Bostrom G, Hogstedt C, Agren G. Socioeconomic disadvantage and primary
non-adherence with medication in Sweden. Int J Qual Health Care. 2007 Jun;19(3):134-40. Epub
2007 Apr 19.

Waters WH, Gould NV, Lunn JE. Undispensed prescriptions in a mining general practice. Br Med
J. 1976 May 1;1(6017):1062-3.

Watts RW, McLennan G, Bassham I, el-Saadi O. Do patients with asthma fill their prescriptions?
A primary compliance study. Aust Fam Physician. 1997 Jan;26 Suppl 1:S4-6.

Wooldridge, Jeffrey M. Econometric Analysis of Cross Section and Panel Data. Massachusetts:
The MIT Press, 2002.

Wroth TH, Pathman DE. Primary medication adherence in a rural population: the role of the
patient-physician relationship and satisfaction with care. J Am Board Fam Med. 2006 Sep-
Oct;19(5):478-86.

Yu AP, Yanni FY, Nichol MB. Estimating the Effect of Medication Adherence on Health Outcomes
among Patients with Type 2 Diabetes – An Application of Marginal Structural Models. Value
Health. 2010 Dec 13(8): 1038-1045.

Abstract (if available)

Abstract OBJECTIVES: Phase I objectives were to measure the extent of primary nonadherence (PNA) in an integrated healthcare system and identify factors significantly associated with PNA to chronic and acute medications. The objective of phase II was to examine healthcare outcomes associated with PNA to cholesterol and diabetes medications. ❧ METHODS: This retrospective cohort study was conducted at Kaiser Permanente Southern California using data from its integrated electronic medical record system. All new prescriptions for 10 therapeutic drug groups (antiinfectives, analgesics, migraine medications, antidiabetics, osteoporosis medications, cardiovascular agents, antihyperlipidemics, antiasthmatics, antidepressants, and anticoagulants) written for patients who satisfied membership and other criteria were included in phase I. PNA rates were calculated by therapeutic drug group as the number of prescriptions not filled within 14 days over the total number of prescriptions. A multivariate logistic regression was used to identify patient, prescriber, and prescription factors significantly associated with PNA. Patients who filled an antidiabetic or antihyperlipidemic were eligible for inclusion in Phase II if they were at least 18 years of age and satisfied additional inclusion and exclusion criteria. A more conservative definition of PNA was applied in phase II as the failure to fill within 180 days. Descriptive statistics were used to summarize baseline differences among patient groups and to describe secondary nonadherence and unadjusted rates of event outcomes. The paired t-test and a multivariate difference-in-difference model were used to evaluate changes in lab values. A multivariate Cox Proportional Hazards model was used to measure the effect of PNA and secondary nonadherence on event outcomes of new disease complications, ER visits, and hospitalization. ❧ RESULTS: In phase I, a total of 569,095 new prescriptions for 398,025 patients satisfied study criteria. The overall PNA rate was 9.8% but individual PNA rates varied widely by therapeutic drug group. The effects of patient race, baseline comorbidities, prior fills, and being treatment-naive on PNA were consistent across drug groups. The strongest predictors were prior fills and being treatment-naïve. Some factors (mostly prescriber and prescription factors) depended on whether the treatment was acute or chronic. Phase II included 21,291 cholesterol patients and 10,137 diabetes patients. PNA rates were12.0% and 5.1% for cholesterol and diabetes, respectively. Secondary nonadherence was relatively worse in cholesterol than diabetes with about 20% of primary adherent patients only filling the first prescription. Adherent patients were on average older and sicker than nonadherent patients. Patients who were nonadherent to their cholesterol or diabetes medication were significantly associated with worse health outcomes in terms of lab values and ER visits. In cholesterol, the risk for ER visits was significantly higher in secondary nonadherent patients (HR=1.24, 95% CI: 1.14, 1.34) and the risk of hospitalization was significantly higher in primary nonadherent patients with prior cholesterol medication use (HR=1.69, 95% CI: 1.19, 2.41) in comparison to adherent patients. For diabetes, secondary nonadherent patients were associated with significantly increased risk for ER visits (HR=1.30, 95% CI: 1.11, 1.51). ❧ CONCLUSIONS: The results of this study help clinicians and healthcare decision-makers to understand the extent of PNA and how PNA varies by therapeutic drug group in an integrated healthcare setting. Most factors of PNA were consistent across drug groups, but some factors depended on whether the treatment was acute or chronic. In addition, these results can facilitate the decision-making process in designing and implementing cost-effective patient interventions to improve overall adherence to cholesterol and diabetes medications. Future research should examine the effect of PNA on longer-term health outcomes such as mortality.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

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

PDF

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

PDF

Documenting the impact of under-treatment for chronic disorders

PDF

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

PDF

The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins

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

Three essays in health economics

PDF

Outcomes of antibiotic use among children with acute respiratory tract infections

PDF

The causal-effect of childhood obesity on asthma in young and adolescent children

PDF

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

PDF

Healthcare provider recommendations: a panacea to improving influenza vaccination rates?

PDF

The impact of healthcare interventions using electronic health records: an evaluation within an integrated healthcare system

PDF

Three essays on emerging issues in hemophilia care

PDF

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

PDF

Delivering better care for children with special health care needs: analyses of patient-centered medical home and types of insurance

PDF

A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system

PDF

Factors influencing the decision and timing to undergo breast reconstruction after mastectomy for breast cancer in public hospital vs. private medical center from 2007 to 2013: a retrospective co...

PDF

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

PDF

Depression severity, self-care behaviors, and self-reported diabetes symptoms and daily functioning among low-income patients receiving depression care

Asset Metadata

Creator Shin, Janet (author)

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

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Pharmaceutical Economics and Policy

Publication Date 07/27/2012

Defense Date 06/07/2012

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

Tag electronic medical records,healthcare outcomes,medication adherence,OAI-PMH Harvest

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor McCombs, Jeffrey (committee chair), Joyce, Geoffrey (committee member), Wilber, Kathleen H. (committee member)

Creator Email janetshi@usc.edu,janetshinlee@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-72363

Unique identifier UC11288951

Identifier usctheses-c3-72363 (legacy record id)

Legacy Identifier etd-ShinJanet-1044.pdf

Dmrecord 72363

Document Type Dissertation

Rights Shin, Janet

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