University of Southern California Dissertations and Theses

Medication compliance to statin therapy and its impact on disease outcomes in type 2 diabetes

(USC Thesis Other)

Medication compliance to statin therapy and its impact on disease outcomes in type 2 diabetes

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content MEDICATION COMPLIANCE TO STATIN THERAPY AND ITS
IMPACT ON DISEASE OUTCOMES IN TYPE 2 DIABETES
by

Lihua Zhang
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PHARMACEUTICAL ECONOMICS AND POLICY)
May 2007
Copyright 2007 Lihua Zhang
ii
Table of Contents
List of Tables iii
List of Figures iv
Abstract v
Chapter 1: Introduction 1
Chapter 2: Literature review 5
1. Diabetes - Background information 5
2. Literature review on medication compliance 10
3. Statin treatment in type 2 diabetes 17
4. Summary of current studies on statin therapy in diabetes 22

Chapter 3: Study design and methods 24
1. Data source and study population 24
2. Construction and measurement of study variables 28
3. Estimating procedure 32
Chapter 4: Study results 47
1. Statin therapy in the Medi-Cal type 2 diabetic patients 47
2. Compliance to statin therapy in type 2 diabetic patients 58
3. Compare medication compliance to antidiabetics
and statin therapy 73
4. The impact of medication compliance to statin therapy
on disease outcomes 76
Chapter 5: Discussion and conclusion 81
1. Summary of study findings 82
2. Clinical and policy implications 84
3. Contribution of the study to the literature 88
4. Study limitations 89
5. Future research 91
6. Conclusions 91
Bibliography 93
iii
List of Tables
Table 3.1 Identification of CVD events 29
Table 4.1: Frequency distribution of patient characteristics 48
Table 4.2: Number of patients being treated with statin by index statin year 50
Table 4.3: Average time between the diabetes diagnosis and the initial
statin therapy 50
Table 4.4: Index statin by the index statin year 52
Table 4.5: Factors affecting time to statin therapy 56
Table 4.6: Characteristics of the compliance study population 59
Table 4.7: Healthcare utilization and costs during the first two years
of statin therapy 61
Table 4.8: Medication compliance to statin therapy during the first
quarter of treatment 63
Table 4.9: Factors influencing patient compliance to statin therapy in
type 2 diabetes 71
Table 4.10: The impact of patient compliance to statin therapy on
cardiovascular risk 78
Table 4.11: The impact of patient compliance to statin therapy
on hospitalization 79
Table 4.12: The impact of patient compliance to statin therapy on
prescription drug costs 80
iv
Lists of Figures
Figure 2.1: A conceptual framework to study medication compliance 16
Figure 3.1: Flowchart of sample selection for the compliance study 27
Figure 3.2: Causal graph to demonstrate time-depending confounders 35
Figure 4.1: Average time (months) between diabetes diagnosis and
the initial statin therapy by index diagnosis year 51
Figure 4.2: Time between diabetes diagnosis and the initial statin
therapy by index diagnosis year 51
Figure 4.3: Time to statin therapy in different ethnic groups 54
Figure 4.4: Time to statin therapy in different age groups 54
Figure 4.5: Time to statin therapy in patients with baseline
cardiovascular event(s) 57
Figure 4.6: Charlson comorbidity index at baseline year 60
Figure 4.7: Medication compliance to statin over time 64
Figure 4.8: Proportion of patient compliant to statin therapy over time 65
Figure 4.9: Proportion of patients compliant with statin therapy by
index statin year 67
Figure 4.10: Proportion of patients compliant to statin by ethnicity 68
Figure 4.11: Proportion of patients compliant to statin by age group 68
Figure 4.12: Proportion of patients compliant to antidiabetics and statin 74
Figure 4.13: Medication compliance to both statin and antidiabetics 75
v
Abstract
Objective: To examine long-term patient compliance to statin therapy in type 2
diabetes and its impact on various disease outcomes.
Methods: The California Medicaid program (Medi-Cal) claims data collected during
the period of January 1995 to December 2004 was analyzed. Medication compliance
was measured using Proportion of Days Covered (PDC) in each quarter of the study
period. Descriptive analyses were performed to: 1) understand the trend of statin
therapy in type 2 diabetes; 2) examine patient compliance to statin therapy and its
change over time; and 3) compare long-term medication compliance to statin and to
antidiabetic agents. A marginal structural model was employed to examine the impact
of medication compliance to statin therapy on cardiovascular events. The effects of
compliance on hospitalization and total medical costs were estimated by generalized
estimating equations (GEE).
Results: Statin therapy was initiated earlier in patients with type 2 diabetes. Patient
compliance to statin therapy was high (PDC=0.83, SD=0.234) during the first quarter
of statin treatment. However, patient compliance dropped sharply from the second
quarter (PDC=0.63, SD=0.382) and the medication compliance at the end of the first
year of the treatment was decreased to 0.56. Over the 30 quarters, patients were less
likely to comply with statin therapy than with antidiabetic treatment (approximately
35% difference). Hispanic patients and patients with complicated medication regimens
were more likely to exhibit poor compliance to statin therapy. Compliance to statin
vi
therapy was significantly associated with decreased cardiovascular risk events (hazard
ratio = 0.24; p<0.01) and less hospitalization in patients with type 2 diabetes (p<0.01).
Conclusion: Prescribing of statins in diabetic populations has improved in recent
years, but long-term medication compliance to statin therapy and antidiabetic
medications in the California Medicaid population remains problematic.
1
CHAPTER 1: INTRODUCTION
Diabetes is a chronic, progressive metabolic disorder. It is an extremely common and
costly chronic condition, affecting nearly 16 million (5.9%) of the U.S. population
(American Diabetes Association, 1998). The estimated total healthcare expenditure
of people with diabetes was $132 billion in 2002 (Hogan P, 2003).
With disease progression, diabetic patients succumb to both microvascular
(retinopathy, nephropathy, or neuropathy) and macrovascular (coronary heart
disease, cerebrovascular, or peripheral vascular disease) complications. In type 2
diabetes, up to 80% of patients will develop or die of macrovascular disease (Vijan
S, 2004). It has been showed that diabetes related complications are associated with
significantly higher medical costs in managing diabetes and the costs associated with
macrovascular disease are much greater than those for microvascular disease
(Gordois A, 2004; Brandle M, 2003; Bate KL, 2003; ADA, 1998).
Given the epidemiology of diabetic complications, management of cardiovascular
risk has been emphasized in the clinical management of diabetes. Recently, HMG-
CoA reductase inhibitors (statins), a new class of lipid-lowering drugs, have been
shown to be effective in both primary and secondary prevention of cardiovascular
diseases. However, its effects in the management of diabetes are still not fully
investigated.
2
Current evidence supporting the therapeutic benefits of statin in diabetes are mainly
from randomized clinical trials, with very few of them focusing on patients with
diabetes. Furthermore, the effects of statin therapy in diabetic patients have been
rarely evaluated in observational settings. Therefore, it is not clear whether the
therapeutic effects of statins can be realized in real practice.
In an observational setting, many factors can affect the effectiveness of
pharmaceutical therapy. Noncompliance to medication has been found to be a
possible reason for the observed gap between potential therapeutic effect and effect
actually achieved by patients in real practice. As for lipid-lowering therapy, patient
compliance has been emphasized as critical to achieving the effectiveness of primary
and secondary prevention (The National Cholesterol Education Program Adults
Treatment Panel III, JAMA, 2001). However, while there has been considerable
research on statin compliance, little has been conducted to assess patient compliance
to statin therapy and its impact on disease outcomes in diabetes. As improving
medication compliance may lead to improved disease outcomes and even cost
savings, it is of great importance to understand the issue of compliance to statin
therapy in diabetic patients.
In addition, current statin studies have been generally limited to the assessment of
clinical benefits while the economic outcomes of statin therapy in diabetic patients
3
have been rarely examined. Therefore, it remains unclear whether the reduced
cardiovascular risk through statin therapy can translate into decreased healthcare
utilization and medical costs in the management of diabetes.
The purpose of this study was three-fold. First, we examined the trend of statin
therapy in type 2 diabetes. Second, we investigated long-term medication
compliance with statin therapy in type 2 diabetes patients. Third, we determined the
impact of patient compliance with statin therapy on clinical and economic outcomes.
We hypothesized that:
1) Time between diabetes diagnosis and the initiation of statin therapy was shortened
in recent years; 2) Patients are expected to be less compliant with statin therapy than
with antidiabetic agents; 3) Medication compliance with statin will decrease over
time; 4) Poor compliance with statin treatment is associated with increased
cardiovascular risk; and, 5) Noncompliance to statin therapy will result in increased
hospitalization and higher treatment costs.
This empirical study will have important implications in the management of type 2
diabetes. First, the results from this study will provide a better understanding about
the current clinical practice regarding the lipid-lowering therapy in diabetic patients.
Second, the findings from this study can help diabetes disease management programs
4
to understand the impact of patient’s compliance on healthcare utilization and costs
in type 2 diabetes, thus helping them to decide whether interventions should be
designed in order to improve medication compliance in diabetic patients.
5
CHAPTER 2: LITERATURE REVIEW
I. Diabetes - Background information
Epidemiology and economic impact of diabetes
Diabetes is a common chronic metabolic disorder. It is a leading cause of morbidity
and mortality (the fifth leading cause of death) in the United States (Anderson RN,
2002). Currently there are 20 million persons with diabetes in the United States, of
whom more than 5 million remain undiagnosed (Winer N, 2004). Among people
with diabetes, about 15% have type 1 (known as insulin-dependent diabetes), while
about 85% have type 2 (known as non-insulin-dependent diabetes). Type 2 diabetes
occurs more frequently in older persons and in certain racial groups, such as
Hispanics and African Americans (Winer N, 2004). Due to increasing obesity,
sedentary lifestyle, and population aging, the global incidence of diabetes,
particularly the type 2 diabetes, is increasing dramatically.
Diabetes exerts a substantial cost burden to society. The estimated total healthcare
expenditure of people with diabetes in 2002 was $132 billion, approximately 12.5%
of total healthcare expenditure in the U.S. (Hogan P, 2003). The direct medical
expenditures attributable to diabetes in 2002 were estimated at $91.8 billion, with a
significant proportion resulting from chronic complications of diabetes (Hogan P,
2003). Hospitalization accounts for 82% of the total direct cost of treating the disease
6
(Bhattacharyya SK, 1998). The estimated average annual cost of care of a diabetic
person in 1997 was $10, 071, compared to $2,699 for a person without diabetes
(Winer N, 2004).
Disease progression of type 2 diabetes
Diabetes often goes undiagnosed because many of its symptoms seem harmless.
Before people develop type 2 diabetes, they almost always have "pre-diabetes" --
blood glucose levels that are higher than normal but not yet high enough to be
diagnosed as diabetes. Recent research has shown that some long-term damage to the
body, especially the heart and circulatory system, may already be occurring during
the pre-diabetes stage (ADA website).With disease progression, diabetic patients
progress to both microvascular (retinopathy, nephropathy, or neuropathy) and
macrovascular (coronary artery disease, cerebrovascular, or peripheral vascular
disease) complications.
About 10% to 40% of patients with type 2 diabetes will develop severe kidney
disease and end stage renal disease. Diabetic retinopathy is the leading cause of new
cases of blindness among adults aged 20 to 74 years. About 60% to 70% of people
with diabetes have mild to severe forms of nervous system damage. Studies have
shown that more than 60% of nontraumatic lower-limb amputations occur among
diabetic patients with severe nerve disease (National diabetes information
clearinghouse).
7
Compared to diabetic microvascular complications, macrovascular complications are
more common and severe. It is the leading cause of diabetes-related deaths. Up to
80% of patients with type 2 diabetes will develop or die of macrovascular disease
(Vijan S, 2004). Compared to nondiabetic individuals younger than 45 years of age,
those with diabetes are more than tenfold as likely to have cardiovascular diseases
(Winer N, 2004). The cardiovascular mortality rate is more than twice as high in
diabetic men and more than fourfold greater in diabetic women, compared with their
nondiabetic counterparts (Kannel WB, 1979).
Management of diabetic complications
As a chronic disease usually accompanied with multiple complications, diabetes
requires continuing medical care and patient self-management education to prevent
acute complications and to reduce the risk of long-term complications. Current
pharmacological management of diabetes has been focused on the following three
aspects.
Control of blood glucose
Glycemic control is fundamental to the management of diabetes. Clear targets for
glycemic control have been established. The American Diabetes Association
recommends glycosylated hemoglobin (HbA
1c
) as the primary target for glycemic
control and the A1c goal for patients in general is set at <7%. The A1c goal for the
8
individual patients is an A1c as close to normal (<6%) as possible without significant
hypoglycemia (ADA, 2007).

Over the past decade, results from clinical trials have demonstrated that
aggressive management of hyperglycaemia significantly reduces the
microvascular risk of developing diabetic nephropathy, retinopathy and
peripheral neuropathy. However, its effect on reducing macrovascular disease is
still uncertain. The United Kingdom Prospective Diabetes Study, to date the largest
and longest prospective randomized trial in people with type 2 diabetes, showed that
intensive blood glucose control with sulfonylurea or insulin failed to significantly
reduce macrovascular complications such as myocardial infarction or stroke
(UKPDS, 1998).
Control of hypertension
About half of the population with type 2 diabetes has hypertension. As blood
pressure levels increase there is a parallel increase in cardiovascular disease in
addition to diabetic retinopathy and nephropathy (Chobanian AV, 2003).
Randomized controlled trials have demonstrated clear benefits of lowering blood
pressure. In the UKPDS trial, tight blood pressure control with angiotensin-
converting enzyme (ACE) inhibitors or-blockers significantly reduced diabetes-
9
related events and diabetes-related deaths. In addition, lower blood pressure targets
further reduce macrovascular complications (Hansson L, 1998).
Current recommended blood pressure target for individuals with diabetes is 130/80
mmHg (ADA, 2007). ACE inhibitors and angiotensin II receptor antagonists are
generally considered first-line agents for treating hypertension in patients with
diabetes, because of their roles in preventing and treating diabetic nephropathy
(ADA, 2007).
Control of blood lipids
Patients with type 2 diabetes have an increased prevalence of lipid abnormalities that
contributes to higher rate of cardiovascular diseases. Traditionally, clinical
management of type 2 diabetes has been focused on the treatment of hyperglycemia
while overlooking the management of dyslipidemia.
Recently, evidence from clinical trials have demonstrated that lipid improved control
of cholesterol or blood lipids can reduce the incidence of macrovascular diseases and
mortality in patients with type 2 diabetes, particularly those who have had prior
cardiovascular events. Following the introduction of several national guidelines, the
importance of concurrently treating hyperglycemia and dyslipidemia has been
10
recognized when developing a management strategy for type 2 diabetes patients
(Haffner SM, 2003; Verges B, 2005).
II. Literature review on medication compliance
The therapeutic and economic benefits of drug treatment are often demonstrated in
the controlled settings of clinical trials. However, these benefits may not be realized
in day-to-day practice, especially for patients who are only partially compliant to
their prescribed therapy. Recent studies suggest that noncompliance to medication
might be a possible reason for the observed gap between potential therapeutic effect
and effect actually achieved by patients in a practical setting (Sokol MC, 2005). As
former surgeon general C. Everett Koop remarked, “Drugs don’t work in patients
who don’t take it.” (Osterberg L, 2005).
Definition and measurement of compliance
Although medication compliance has been studied for a long time, there is no
unanimous definition of compliance and standard method to measure compliance
(Cleemput I, 2002). Different definitions and measurement methods with different
degrees of accuracy have been used in literature.
Compliance with (or adherence to) a medication regimen is generally defined as the
extent to which patients take medications as prescribed by their health care providers
11
(Osterberg L, 2005). Some researchers prefer the term “adherence” over
“compliance” because the latter one suggests a more passive patient-physician
relationship, whereas “adherence” connotes a more active role of patients in
decisions concerning their own treatment more collaborative patient-physician
partnership in developing and implementing treatment plans (Tabor PA, 2004).
Regardless of which term is preferred, it is clear that the full benefit of medications
will be achieved only if patients follow prescribed treatment regimens reasonably
closely.
Measurement of compliance
The lack of a generally accepted definition of compliance makes attempts to measure
the concept difficult (Kyngas H, 2000). Of the numerous measures of compliance
none appears to be sufficiently reliable and valid. As noticed by Kyngas etal, the
selection of an appropriate measure of compliance is complicated by a number of
considerations. First, chronic conditions with multimodal treatment plans may
require different measures from acute illnesses with simple medication regimens.
Second, compliance with one element of the treatment plan may not suffice if its
success depends on compliance with other elements. Thus, an all-encompassing
assessment strategy is needed.
Currently there is no gold standard exists for measuring medication compliance
(Murray MD, 2004). Methods used to measure compliance can be divided into
12
indirect and direct methods. Direct methods such as drug assays of blood, urine and
saliva to determine drug concentration levels, are considered to be objective, reliable
and easily quantifiable. They have major disadvantages, however, including
relatively high costs and resource limitations in medical settings (Kyngas H, 2000).
Some indirect methods are patient report, pill count, refill adherence, and electronic
monitoring using microchip technologies. Each of these methods has advantages and
disadvantages. For example, Pill counts provide an overall estimate of patients'
medication consumption over time but give no information about the timing of
doses. This method is also subject to pill dumping immediately before the day of the
pill count. Technological medication monitoring devices, such as the Medication
Event Monitor System (MEMS) represent clear improvements in terms of reliability
and objectivity compared with the traditional method of pill counts. However, it is
still not known whether the medication that is removed from the container is actually
taken by the patient. Also, it is not known who is opening and closing the container.
To date this measurement method is still fairly expensive, which is another drawback
for regular use.
Refill adherence is commonly used in retail pharmacy and pharmacy benefit
management to track the patterns of patient refills using computerized prescription
records. Access to refill data permits easy tracking of chronic treatments, but the data
are widely dissociated from day-to-day medication consumption. Refill adherence
13
has been an area of increasing interest in the retrospective study of medication
compliance. The proportion of days covered (PDC) is a common measure of refill
adherence using pharmacy refill records (Chapman RH, 2005). It is calculated as the
number of days with drug coverage divided by the number of days in the specified
time interval. The PDC method avoids double-counting covered days. Thus, it is
always a value between the range of 0 and 1. In contrast to the PDC method, the
medication possession ratio (MPR) method counts the sum of the “days supplied” by
all prescriptions filled during the period, and therefore might generate values that are
greater than 1. Although compliance assessed using prescription refill data provides
insight into the availability of medication, it does not provide information on the
timeliness and consistency of refilling.
The impact of compliance on disease outcomes
Notwithstanding differences in the measures used to estimate noncompliance, the
phenomenon of noncompliance is considered a serious problem in health care (Paes
AH, 1998). A recent meta-analysis revealed that the percentage of patients who do
not adhere to their instructions vary in different studies from 20% to 82% (Dimatto
MR, 2004). The medication compliance rate is especially lower (<50%) in patients
with chronic disorders (including asthma, cardiac disease, diabetes, hypertension,
psychiatric disorders, rheumatoid arthritis) (Dunbar-Jacob J, 2000).
14
The economic costs of non-adherence are substantial. When patients are not
compliant to the prescribed therapy, disease symptoms and complications may
worsen, leading to increased use of hospital and ER services, and higher medical
costs (Sokol MC, 2005). The healthcare costs associated with inadequate adherence
(e.g., more diagnostic tests and treatment, unnecessary therapies, increased dosages,
more frequent clinic visits and monitoring, hospital admissions, nursing home care)
and lost productivity is estimated to be as much as US$100 billion annually (Robiner
WN, 2005). As many as 1.94 million (about 5.5% of) hospital admissions can be
attributed to poor adherence with drug therapy (Sullivan SD, 1990).
Conceptual model for medication compliance research
Current research suggests that medication compliance is a self-care process affected
by many social and behavioral factors such as patient’s perception of the severity of
the illness, medication regimens, patient-provider relationship, etc. Noncompliance
to prescribed medications might be the result of a complex interaction among the
social environment, the patient, and the health providers (Kidd KE, 2000).
The behavioral health care utilization model of Phillips et al (1998) relates
environmental and population characteristics to health behavior, including personal
health choices (eg, medication adherence). Based on this model, Murray et al
15
developed a conceptual model of medication compliance (Murray MD, 2004). The
following compliance model for studying the effect of medication compliance in
diabetic patients was constructed in light of Murray’s model (Figure 2.1).
This model shows a complex relationship between patient characteristics,
environmental factors, and medication compliance. It also reflects the possible
interactions between medication compliance and disease outcomes. The
environmental factors include characteristics of the healthcare delivery system,
external environment (such as economic climate, level of stress and violence,
prevailing norms of the society), and community (eg, the availability of providers in
the community) (Phillips KA, 1998). Patient characteristics including three
components: 1) Predisposing characteristics: These are preexisting factors (such as
age, education level, cognitive disability, etc) that predict medication compliance. 2)
Enabling resources: Despite the patient's predisposition to comply with their
medications, access to certain supportive resources (eg, money or transportation to
the pharmacy, insurance, patient-physician relationship) may also affect patient’s
compliance to medication. 3) Need: Patient’s perceived illness also play an important
role in medication compliance. Even when patients are predisposed to comply with
their medication regimen and they have access to enabling resources, they must
perceive a need to adhere. Patients are less likely to adhere when they do not
perceive the severity of their illness or expect the medication to help.
16
Figure 2.1: A conceptual framework to study medication compliance.
Healthcare system ------- External environment
Predisposing
characteristics
Enabling
Resources
Income
Insurance
Relationship with providers
Distance to health services
Social support
Supervision
Need Perceived and evaluated illness
Demographics
Knowledge, beliefs, expectations
Cognitive resource
Medical-disability-related
Medication
Compliance
Outcomes:
Diabetic
complications,
Healthcare use
and cost; etc
Environment
Patient
Characteristics
17
III. Statin treatment in type 2 diabetes
Several pharmacological agents are available for the treatment of diabetic
dyslipidemia, including resins, niacin, fibrates, and the statins. The current ADA
guidelines recommend that statins as the first-line choice for the primary prevention
against macrovascular complications in patients with type 2 diabetes and other
cardiovascular risk factors (ADA, 2004).
Therapeutic effects of statins in type 2 diabetes
Results from several large-scaled randomized clinical trials (eg., 4S (Scandinavian
simvastatin survival study), CARE (cholesterol and recurrent events), WOSCOPS
(West of Scotland coronary prevention study), LIPID (long term intervention with
pravastatin in ischaemic disease)) have demonstrated that statins are effective for
both primary and secondary prevention of coronary artery disease (Pyorala K, 1997;
Goldberg RB, 1998; Shepherd J, 1995; The LIPID study group, 1998). Current
evidence supporting the use of statin in diabetes is mainly from subgroup analyses
from these clinical trials. Beyond the subgroup analyses, only limited data is
available about the lipid modifying efficacy of statins in type 2 diabetic populations
(Haffner SM, 2003; Colhoun HM, 2004).
18
In a background paper prepared for the American College of Physicians regarding
the pharmacologic lipid-lowering therapy in type 2 diabetes mellitus, Vijan et al
(2004) systemically reviewed the pharmacological lipid-lowering therapy in type 2
diabetes. They identified 14 randomized controlled trials of lipid-lowering therapy
for the primary and secondary prevention. None of these trials were conducted solely
in patients with diabetes. Sample sizes of diabetic patients in these trails are often
small. In several trials, diabetes was an exclusion criterion or very few diabetic
patients were included.
Because of small diabetes subgroups in most trials, not all individual trials found
benefit from treatment with statins. For example, the Prospective Study of
Pravastatin in the Elderly at Risk (PROSPER) failed to demonstrate the benefit of
Pravastatin in both primary and secondary prevention of cardiovascular disease in
diabetic patients. Neither did the LIPID study show any significant risk reduction of
Pravastatin for the secondary prevention. However, the results from the meta-
analysis of the 14 trials suggest that treatment with statins reduces cardiovascular
risks in type 2 diabetes. Actually, for primary and secondary prevention in diabetes,
most of the patients came from the Heart Protection Study (Collins R, 2003). This
study provides direct evidence that cholesterol-lowering therapy is beneficial for
people with diabetes even if they do not already have the manifestation of coronary
disease or high cholesterol concentrations. Statin therapy should now be considered
19
routinely for all diabetic patients at sufficiently high risk of major vascular events,
regardless of their initial cholesterol concentrations.
Despite the increased risk of macrovascular complications in people with diabetes,
very few randomized clinical trials have been focused on patients with type 2
diabetes. The Collaborative AtoRvastatin Diabetes Study (CARDS) is the first large
primary prevention study conducted solely on patients with type 2 diabetes (Colhoun
HM, 2004) that was not included in the above meta-analysis. This study
demonstrated a 37% relative risk reduction in the primary prevention of
cardiovascular morbidity in patients with diabetes. Currently, there are also several
other ongoing studies that will continue to assess the beneficial effects of statins in
the management of diabetes.
Compliance to statin treatment in type 2 diabetes
As mentioned earlier, noncompliance with medications is common among patients
with diabetes. A recent meta-analysis has showed that the average adherence in
patients with diabetes is 67.5%, which is lower than that found among many
conditions (DiMatteo MR, 2004). Findings from a specific systematic review on
adherence to medications for diabetes showed that average adherence to oral
hypoglycemic agents ranged from 36 to 93% (Cramer JA, 2004).
20
Although patient compliance to lipid-lowering therapy has been emphasized as
critical to achieving the effectiveness of primary and secondary prevention (The
National Cholesterol Education Program Adults Treatment Panel III, 2001), studies
have revealed poor compliance to statin therapy (Benner JS, 2002; Jackevicius CA,
2002, Wei L, 2002; Perreault S, 2005). A recent study in elderly patients showed that
compliance to statin therapy decline more than 25% in the first 6 months after the
original prescription, with further declines over time (only 26% of those who
initiated statin treatment maintained a high level of use five years later) (Benner JS,
2002). Noncompliance to statin therapy has been associated with increased
cardiovascular morbidity (Blackburn DF, 2005) and higher risk for recurrent MI
(Wei L, 2002). In patients with type 2 diabetes, noncompliance to lipid-lowering
therapy could impose an extreme danger due to their high risk of developing
cardiovascular complications. Up to date, few studies have assessed compliance to
statin therapy in patients with diabetes. In a study including 677 diabetic patients,
Pladevall et al (2004) found that the prevalence of noncompliance was 36% for
statins. Poor compliance to statins was associated with an increase of cholesterol
levels (Parris ES, 2005).
Many factors appear to affect medication compliance in patient with type 2 diabetes.
Regimen complexity is one of them. As a chronic disease condition, patients
suffering from diabetes are usually on several medications. Therefore, the burden of
medication compliance is relatively high. Actually, recent studies have showed that
21
fixed-dose combination therapy (FDCT) of several antidiabetics (rosiglitazone-
metformin or glyburide-metformin) can improve medication compliance due to
overall reductions in regimen complexity allowed by FDCT (Vanderpoel DR, 2004;
Melikian C, 2002). In addition, research has also showed that other factors, including
patient’s comprehension of the treatment regimen and its benefits, adverse effects,
medication costs, as well as the patient’s emotional well-being, also affect patient’s
compliance with medications (Rubin RR, 2005).
By adding lipid-lowering drugs to the existing hypoglycemic medication regime, the
burden of medication compliance is further increased. In this situation, medication
compliance might involve a much more complex decision making process (require
more cognitive resources, belief of the therapeutic effects of statins, and perceived
disease severity, etc.). Patients’ beliefs and understanding about cholesterol and the
role of cholesterol modifying strategies might play an important role in the long-term
compliance with statin therapy (Yilmaz MB, 2004).
Previous research has shown that patients are generally less motivated to comply
long term with preventive treatments or those targeting asymptomatic conditions
with no immediate apparent benefits (Howell N, 2004). Without foreseeing any
immediate benefit of statin’s risk reduction effects, compliance with statin therapy in
diabetic patients is expected to be poor. Therefore, comprehensive information at
initial prescription might result in better compliance rates. A careful exploration of
22
patient’s beliefs and understanding might provide useful insights in subsequent
medication use and expose potential barriers to compliance (Yilmaz MB, 2004). If
clinicians can identify these barriers, this might improve medical management.
IV. Summary of current studies on statin therapy in diabetes
Although several important clinical trials have demonstrated that statin therapy can
reduce cardiovascular risk, its effects in the management of diabetes has not been
fully investigated (Green ML, 2003). Several issues associated with current studies
on statin therapy in diabetes are noteworthy.
First, current evidence supporting statin therapy in patients with diabetes is mostly
from clinical trials, with very few of them specifically designed for diabetic patients.
Furthermore, observational studies on the therapeutic benefits of statin therapy in
diabetic patients are rare. Therefore, little is known about the effect of statin
treatment in real practice settings. Whether the potential therapeutic effects can be
realized in real practice remains unclear.
Second, medication compliance to statin therapy in patients with type 2 diabetes,
especially over the long term, is poorly understood. As diabetic patients generally
have complicated regimen with several different medications being used
simultaneously to control blood glucose level, hypertension, and dyslipidemia; it
23
may be a challenge for them to comply with the statin treatment. However, the
effectiveness of statin therapy in reducing cardiovascular risk may not be realized if
patient compliance is poor in type 2 diabetes.
Third, although previous research has shown that medication compliance to lipid-
lowering therapy is critical in achieving therapeutic goals, this issue has not been
well understood in patients with diabetes. Little is known about the impact of patient
compliance with statin treatment on the development of complications in diabetes,
healthcare utilization and costs. However, for the purpose of disease management, it
is important to understand how disease outcomes are affected by patient compliance
to statin therapy. If this relationship is significant, then intervention programs can be
designed to improve medication compliance in high-risk patients.
24
CHAPTER 3: STUDY DESIGN AND METHODS
I. Data source and study population
Data source
The data used for this retrospective study was from a publicly available 20% sample
of California Medicaid program (Medi-Cal) fee-for-service claims data. This claims
data includes patient eligibility information, medical service claims, and pharmacy
claims for all patients between January 1995 and December 2004. These claims
represent fee-for-service beneficiaries as all Medi-Cal managed care plans were
excluded from this data set. For dually eligible patients, Medicare claims were
captured by the Medi-Cal crossover system.
Selection of the study population
The proposed research focused on patients with newly diagnosed type 2 diabetes.
The identification of subjects with diabetes was based on ICD-9 code “250.xx” and
the use of antidiabetic agents. The fifth digit ICD-9 codes “250.01” and “250.03”
was used to eliminate patients with type 1 diabetes (Kalsekar ID, 2006). As this
study was restricted to patients with type 2 diabetes, we also excluded patients from
the analysis if they started antidiabetic treatment with insulin.
25
To answer our research questions, two study populations were identified. The first
population was selected to examine the clinical trend of statin therapy in type 2
diabetes. This included patients who met the following selection criteria: 1) Age is
no less than 18 years old; 2) Diabetic patients newly started on antidiabetic
treatment, defined by at least 12 months washout period prior to the index diagnosis
date; 3) Patients must start with statin therapy after the initiation of oral antidiabetic
agents; and, 4) No use of any lipid-lowering drugs for up to 24 months before the
first statin prescription. The last exclusion was for the purpose of identifying diabetic
patients newly treated with statin and to exclude any residual effect from previous
lipid-lowering treatment.
The second population was selected to examine medication compliance to statin
therapy in type diabetic patients. These patients were selected from the first study
population by excluding patients who: 1) were on insulin treatment before the
initiation of the statin therapy; 2) were not continuously eligible through the study
period; and, 3) had only one prescription fill of statin and those who used other lipid-
lowering drugs (LLD) after the initiation of statin therapy (Figure 3.1). Patients with
only one fill of statin were excluded in the compliance study because this patient
population might be different from those with multiple fills. They may quit the
treatment after one fill because: 1) their lipid level may not be high enough to require
lipid-lowering treatment; or 2) they did not respond to statin very well; or 3) they
may quit due to the adverse effect from the treatment. Since our purpose was to
26
examine the risk reduction effect of statin, we also excluded patients who used other
lipid-lowering drugs after the initiation of statin as these patients might have
different clinical characteristics.
The index diagnosis date was defined as the first prescription date for oral
antidiabetic agents. Index statin date was defined as the date of the first statin
prescription from the pharmacy claims. Patient eligibility profile was generated to
ensure the patients were continuously eligible during the study period.
27
Figure 3.1: Flowchart of sample selection for the compliance study
7,015 statin users left after applying the criteria of 24 month washout period
Select 7,404 pts started statin after the diagnosis of type 2
Exclude 1,148 patients that used other LLDs after the index statin date
Exclude 602 patients starting insulin before statin
Exclude 865 patients not continuously eligible
Exclude 580 patients with one fill of statin
22,189 patients with the diagnosis of type
2 diabetes, and age over 18 years old
Finally, 4,422 patients were included in the compliance study
28
II. Construction and measurement of study variables
To examine medication compliance and its impact on disease outcomes, the
following study variables were generated.
Medication compliance to statin
Patient compliance to statin therapy was assessed based on their prescription refill
records and was expressed as proportion of days covered (PDC). It was measured
quarterly throughout the study period by dividing the number of days with statin
drug coverage after treatment initiation by the number of 90. PDC was treated either
as a continuous measure or a dichotomous variable in the statistical analyses. In the
later case, patients were classified into two categories based on the commonly
adopted 80% cutoff point of PDC (Avorn J, 1998). Patients having PDC greater or
equal than 0.8 was defined as compliant while those with PDC less than 0.8 were
treated as noncompliant.
Five statins were included in the study of medication compliance to statin therapy:
atorvastatin, pravastatin, simvastatin, lovastatin, and fluvastatin. The newest statin,
rosuvastatin, was not available during the study period and hence is not considered in
this study.
29
Outcome variables
The primary clinical outcome variable was the time to the first cardiovascular event.
Diagnoses of cardiovascular event were based on ICD-9 and CPT codes.
Cardiovascular events included in the study were angina, myocardial infarct, stroke,
transient ischemic attack, and revascularization procedures such as CABG and
PTCA (Herman WH, 1999; Snow V, 2004), as showed in Table 3.1. The study end
point was the first cardiovascular event occurring after patients started with statin
therapy. Patients who did not have any cardiovascular event were censored at the
date when they lost eligibility or the end of the monitoring period.
Table 3.1 Identification of CVD events
CVD events ICD-9 or CPT code
Myocardial Infarct 410.xx
Angina 413.xx
Stroke 433.x1, 434.xx, 436
TIA 435.xx
PTCA 92982, 92984,92995,92996
CABG 33510 - 33536
The outcome variables for the economic analysis include hospitalization and total
medical costs. Since the dataset used was a claims data, medical costs were measured
30
using payments as a proxy, so this study was conducted from the payer’s perspective.
Total medical costs were separated into: a) inpatient costs; b) ambulatory and
outpatient costs (including Medicare Part B costs); c) nursing facility/intermediate
care facility costs; and d) prescription drug costs. Quarterly costs for these four
categories were derived separately. Total costs within a quarter were the sum of the
costs of the above categories.
Costs of service and prescriptions were credited based on the quarter of service.
Inpatient costs and long-term care costs were estimated by multiplying the number of
service days by the per diem rates for MediCal fee-for-service patient (Watkins J,
2005). Medicare Part B costs were estimated by multiplying the Medicare co-
insurance by four, adding the Medi-Cal paid amount and the paid amount of any
other parties listed on the claims (Nichol MB, 2004). For non-crossover claims,
outpatient costs were the sum of the amount paid by Medi-Cal and other parties.
Prescription costs were the sum of payment made by Medi-Cal and other parties.
Other study covariates
1) Comorbidities
As comorbidities are important predictor of disease outcomes and may affect
medication compliance to statin, baseline comorbidities during the one year period
before the index statin date were measured using the Charlson Comorbidity Index
(CCI). The CCI score was calculated based on ICD-9 codes in the medical claims
31
during the analysis period and was computed using a Deyo-adapted Charlson scale
(Deyo RA, 1992). We also developed separate indicators for depression, and
cardiovascular conditions.
2) Healthcare utilization variables
Hospitalization and emergency room visits were assessed both at baseline and during
the study period.
3) Medications for diabetes treatment
The total number of distinct medications was measured to reflect patient’s
medication burden. Dichotomous variables were also created respectively to indicate
whether patient was on antihypertensive and antiplatelet agents, which are related to
the future cardiovascular risk.
4) Year variable
To account for the possible effects of treatment guideline change and market
expansion of statins during the study period, a calendar year variable were created
based on the index year of the statin therapy.
5) Other study variables
Demographic variables such as age, gender, race and ethnicity, and dual eligibility
were identified based on records from the Medi-Cal data.
32
III. Estimating procedure
All statistical analyses were conducted using SAS for Windows, version 9.1 (SAS
Institute Inc., Cary, NC).
Descriptive analyses were performed to understand: 1) the clinical trend of statin
therapy in patients with type 2 diabetes; and 2) patient compliance to statin therapy
and its change over time. In this study, we also compared long-term medication
compliance to statin and to antidiabetic agents in type 2 diabetic patients as this has
been rarely studied before.
Different statistical models were employed to answer the following questions.
Question 1: How does medication compliance to statin therapy affect the
cardiovascular risk in type 2 diabetes?
To answer this question, we started with addressing the statistical challenge related
to the compliance study faced by healthcare researchers.
Currently, many researchers have studied the relationship between medication
compliance and disease outcomes. It has been showed that medication compliance is
significantly associated with disease outcomes in different disease states (Blackburn
33
DF, 2005). However, most of these studies are cross-sectional, with compliance and
outcomes observed during the same time period. Hence, the causal relationship
between medication compliance and outcomes is not established.
To determine this causal relationship, both compliance and outcome variables need
to be measured in a temporal order. Hence, a longitudinal study design is desirable
for this purpose. However, analytical issues arise when compliance is treated as a
time-varying covariate in observational studies. One of the issues is the confounding
effect from time-dependent variables that are the predictors of medication
compliance and also affected by the medication compliance from the previous
period. A conventional statistical approach does not account for confounding
covariates may provide biased estimates of the causal effect of compliance on
disease outcomes. Recently, Robins and colleagues have proposed a new class of
statistical models, called the marginal structural models (MSMs), to account for this
statistical issue.
Marginal structural model
Marginal structural models are a class of causal models for the estimation, from
observational data, of the causal effect of a time-dependent exposure (eg.
compliance) in the presence of time-dependent covariates that may be
simultaneously confounders and intermediate variables (Robins, 2000). The
parameters of MSM can be consistently estimated using a new class of estimators
34
called the inverse-probability-of treatment weighted (IPTW) estimators. Basically,
IPTW calculates the probability of an individual receiving the treatment they actually
received conditional on their observed covariates. Each subject in the sample is then
weighted by the inverse of that probability. The details of the theory and form of the
marginal structural model have been illustrated in their original papers (Robins JM,
2000; Hernan MA, 2000). Here we provide a brief review of the model.
To understand the marginal structural model, let’s start with a simple example with
two time points (time 0 and time 1). As shown in Fig 3.2, let E
k
be the time-
dependent exposure or treatment at time k (k=0, 1). Let Y be the dichotomous
outcome variable representing the outcome after the second treatment. Z
k
indicate
the time-varying covariates. As showed in the following causal graph, in which each
arrow represents a casual relationship (or path) between the two variables it
connects, Z
1
is affected by treatment E
0
, but it also confounds the treatment effect of
E
1
on Y. Therefore, Z(k) is a time-dependent confounder, as it is a time-dependent
covariate that is both a risk factor for survival and also predicts subsequent exposure.
35
Time-dependent confounders are affected by treatment and relevant to the outcome
of interest. In the presence of such time-dependent confounders, Robins argues that
using standard methods (such as Cox or Logistic regression) to estimate the causal
effect of a time-varying exposure (treatment) may produce biased parameter
estimates. Instead, they proposed the marginal structural model as a tool to deal with
this time-dependent confounding problem. To adjust for the association between the
time-depending confounders and the treatment, they proposed a weighted analysis
procedure for this association model, which gives unbiased estimates of causal
parameter.
The model assumes that there are no unmeasured confounders given data on
measured confounders Z. Under this assumption, the bias caused by time-dependent
confounders can be reduced or eliminated by fitting the time-dependent Cox model
E
0
E
1
Z
1
Y
Z
0
Figure 3.2: Causal graph to demonstrate time-depending confounders
36
with the contribution of a subject i to a risk-set calculation performed at time t
weighted by the stabilized weights (sw
i
(t)), which is derived through the following
formula:
int(t) pr[E(k)=e
i
(k) | (k-1) =
i
(k-1), W=w
i
)]
sw
i
(t) = pr[E(k)=e
i
(k) | (k-1) =
i
(k-1), Ž (k)= ž
i
(k)]
k=0
(1)
In this formula, we use hat (˘) to represent a covariate history. For instance, E(k)
indicates the exposure or treatment at time k; while (k) represents the exposure
history up to time k. We let W represent for time-independent covariates at baseline;
Z(k) is a random vector representing time dependent covariates at time k.
Each factor in the denominator of sw
i
(t) is informally the probability that the subject
received his observed treatment at time t=k, given his past treatment and prognostic
factor history. Each factor in the numerator is, the probability that the subject
received his observed treatment conditional on his past treatment history and
baseline covariates, but not adjusting for his past time-dependent prognostic factor
history.
37
If all relevant time-dependent confounders are measured and included in Z(k), then
weighting by sw
i
(t) creates a pseudo-population for a risk set at time t, in which Ž
(k) no longer predicts the initiation of the treatment at time t (i.e, Ž (k) is not a
confounder), and the causal association between the exposure and the outcome is the
same as in the original study population. Therefore, the effect of weighting is to
create a pseudopopulation which is not confounded by the measured time-varying
covariates so estimates of the causal effect of the treatment are unbiased.
Censoring
In the field of healthcare research, many data are censored. Right censoring occurs
when an event is not observed during the follow-up period or the subject dropped out
of the study. Under the assumption of no informative censoring, a similar procedure
is used to estimate inverse-probability-of-censoring weights cw
i
(t), which is the ratio
of a subject’s probability of remaining uncensored up to time t.
t pr[C(k)=0 | (k-1) =0, (k-1) =
i
(k-1), W=w
i
)]
cw
i
( (t) = pr[C(k)=0 | (k-1) =0, (k-1) =
i
(k-1), Ž (k-1)= ž
i
(k-1)]
k=0
(2)
38
These weights are then multiplied by the stabilized weights sw
i
(t) from formula (1)
to create an overall weight for each subject in each time period, SW
i
*
( (t) = sw
i
(t) x
cw
i
(t). Weighting by SW
i
*
( (t) produces a consistent estimate of the causal parameter

1
under the assumption that the measured covariates are sufficient to adjust for both
confounding and the selection bias due to loss to follow-up.
Estimation of the weights
To derive the total weight for each subject at different time points, we need to
estimate consistently the denominator and the numerator of sw
i
(t) and cw
i
(t). To
estimate sw
i
(t), a pooled logistic regression model is fitted:
Logit pr[E(k)| (k-1), Ž (k)] =
0
(k) +
1
Z(k) +
2
W
(3)
Now we obtain â=( â
0
(k), â
1
, â
2
) for the unknown parameters. The estimated
predicted values p
i
(k)=expit(â
0
(k) + â
1
Z(k) + â
2
W
i
), where expit(x)=e
x
/(1+e
x
).
The denominator of sw
i
(k) can then be obtained. The numerator of sw
i
(k) can be
estimated by fitting the above logistic regression model with Z(k) removed from the
model.
Logit pr[E(k) | (k-1), Ž (k)] =
0
(k) +
2
W
(4)
To correct for censoring, cw
i
(k) can be estimated in a manner analogous to the
estimation of sw
i
(k) except with E(k) replaced by C(k) as the outcome variable, with
E(k-1) added as an additional regressor, and conditioning on (k-1) =0.
39
Once the total weights are derived, we can fit the pooled weighted logistic regression
model to get the parameter estimates for the causal effect of the treatment. The above
analyses can be performed in SAS using the programs developed by Robins et al
(Robins, 2000).
Application of the MSMs in healthcare research
Marginal structural models (MSM) provide a powerful tool for estimating the causal
effect of a treatment. They are particularly useful in the context of longitudinal data
structures, in which each subject's treatment and covariate history are measured over
time, and an outcome is recorded at a final time point.
MSMs are useful tools for assessing causality with complicated, longitudinal
datasets, but they have not been widely used in the healthcare research field.
Recently, the applications of this model in some empirical studies have been seen. In
their study, Bodnar et al (2004) illustrated the application of MSMs in the estimation
of the causal effect of iron supplementation during pregnancy on odds of anemia at
delivery. In their study, they also demonstrate the robustness of the causal model
even when non-confounding covariates are included, as these did not meaningfully
change the results from the MSMs.
MSM has also been employed by Barron et al (2004) in their research to estimate the
effect of discontinuing antiretroviral therapy on survival of women initiated on
40
highly active antiretroviral therapy. Their results showed that the marginal structural
model generated an increased risk of mortality for those who discontinued all
antiretroviral therapy. This risk was greatly attenuated when estimated by the
standard Cox proportional hazard model. In addition to the application with
observational datasets, MSMs have also been used in randomized control studies. In
a recent study by Cook et al (Cook NR, 2002), they used a marginal structural model
to determine the underlying causal effect of aspirin on cardiovascular mortality in the
physicians’ Health Study, which is a randomized clinical trial of aspirin in primary
prevention of cardiovascular disease among physicians. The marginal structural
model yielded a larger effect of aspirin in reducing cardiovascular mortality,
comparing to the intent-to-treat analyses which found no effect of aspirin. In a recent
study, Bryan et al (Bryan J, 2004) employed the methodology to estimate the causal
effect of exercise on mortality in seniors in a longitudinal observational study, in
which they demonstrated that the efficiency gain of the IPTW estimator in the
presence, or even in the absence, of time-dependent confounding factors.
Application of marginal structural model in this research
In our study, we chose to use the MSM model to estimate the causal effect of
medication compliance to statin therapy on the cardiovascular risk in type 2 diabetes.
This is because the causal effect of patient compliance to statin therapy might be
confounded by time-varying covariates (such as hospitalization) that are potentially
associated with both medication compliance and cardiovascular risk. Since marginal
41
structural models can provide consistent estimates in the presence of time-dependent
confounding factors, it is the appropriate statistical method for the estimation of the
causal effect of medication compliance to statins.
The primary outcome variable for the marginal structural model was time to the first
CVD event occurred after the initiation of statin therapy. The principle predictive
variable of interest was medication compliance to statin therapy. The model was
adjusted for study covariates including: 1) patient demographic characteristics such
as age, gender, and race; 2) utilization of healthcare service at baseline, including
hospitalization, ER and outpatient services; 3) baseline comorbidities (CCI), baseline
cardiovascular and cerebrovascular diseases (based on the diagnosis of MI, angina,
stroke, and cardiovascular procedures such as PTCA, etc), congestive heart failure,
and number of distinct medications; 4) index statin year; 5) time-varying variables:
total number of medications, hospitalization, ER visit, number of outpatient visits,
use of antihypertensives and antiplatelet medications.
Question 2: How does compliance to statin therapy affect hospitalization and total
medical costs in type 2 diabetes?
Longitudinal data analyses were conducted to answer this question. Since data are
collected on the same units across successive points in time, these repeated
observations are correlated over time. If this correlation is not taken into account
42
then the standard errors of the parameter estimates will not be valid. To overcome
this problem, the Generalized Estimation Equations (GEE) will be employed for the
statistical analysis (Michaud K, 2003).
Generalized Estimation Equations
GEE was introduced by Liang and Zeger (1986) as a method of estimation of
regression model parameters when dealing with correlated data. It is being
increasingly used to analyze longitudinal and other correlated data. Because it is an
extension of generalized linear models, GEE can be applied to many different types
of outcome variables (categorical, count, and continuous outcomes).
GEE belongs to the marginal effect models. In contrast to random-effects model, it
supposes that the relationship between the outcome Y and the covariate X is the
same for all the subjects (Diggle PJ, 2002). Marginal models give an average
response for observations sharing the same covariates as a function of the covariates.
i.e., for every one-unit change in a covariate across the population, GEE tells how
much the average response would change (Zorn CJW, 2001). GEEs estimate
regression coefficients and standard errors with sampling distributions that are
asymptotically normal (Liang KY, 1986), can be applied to test main effects and
interactions, and can be used to evaluate categorical or continuous independent
43
variables. GEE estimates are the same as those produced by OLS regression when
the dependent variable is normally distributed and no correlation within response is
assumed.
Model specification
GEE model can be modeled as following:
g(µ
ij
)=X
ij

where
Y
ij
denotes the outcome response for subject i measured at time j.
µ
ij
=E(Y
ij
) is the marginal response
X
ij
is a p x 1 vector of study covariates
is a p x 1 vector of unknown regression coefficients
g(.) is the link function
Fitting a GEE model requires to specify: (a) the link function, (b) the distribution of
the dependent variable, and (c) the correlation structure of the dependent variable.
1) Choose an appropriate link function
The link function g(.) is a transformation function that allows the dependent variable
to be expressed as a vector of parameter estimates (y =
0
+
1
X
1
+
2
X
2
+
3
X
3
. . .)
in the form of an additive model. In theory, the link function g(.), can be an
monotonic, differentiable function. However, in practice, only a small set of link
44
functions are actually utilized. The basic link function is the identity link function,
which is used for normally distributed data. The logit link is the standard linking
function for binary dependent variables. For counted data being modeled with
Poisson regression, the log link function is the most appropriate.
2) Distribution of the response variable
GEEs permit specification of distributions from the exponential family of
distributions, which includes normal, inverse normal, binomial, Poisson, negative
binomial, and Gamma distributions. Generally, researchers will have some prior
knowledge of the distribution of the response variable. As a rule, the binomial
distribution is preferred for the binary response data. In cases in which the responses
are counted, a Poisson distribution should be first selected.
3) Specification of the correlation within response variable
A third step involves the specification of the ‘‘working’’ correlation structure of the
repeated measures within subjects. It is this working correlation matrix that allows
GEEs to estimate models that account for the correlation of the responses (Liang KY,
1986). This working correlation matrix R
i
() is a n x n “working” correlation matrix
of Y
i
indexed by a vector of parameters. It is assumed that the correlation matrix R
i
depends on a vector of association parameters denoted . These parameters are
assumed to be the same for all subjects. They represent the average dependence
among the repeated observations across subjects.
45
It is generally recommended that choice of R should be consistent with the observed
correlations. However, the model is robust to errors in the specification of correlation
structure because estimates of the regression parameters remain consistent. Therefore,
the efficiency gains from exact specification of the structure are usually slight (Liang
KY, 1986). This is a very attractive property of GEE model. Efficiency is reduced if
the choice of R is incorrect, however the loss of efficiency is lessened as the number
of subjects gets large.
Once the correlation matrix is specified, it can be combined with the variance
function Var(Y
ij
) to form a model for the covariance of correlated observations Y
i
=
(y
i1
, . . . , y
it
).
It is assumed that the variance of Y
ij
is a known function of the mean:
Var(Y
ij
)=V(µ
ij
)
where
V(µ
ij
) is a known variance function
is a scale parameter which may need to be estimated
Var(.) is the variance function
46
The covariance matrix of Y
i
for a given subject i is:
Vi =A
i
1/2
R
i
() A
i
½
where
A
i
is a n x n diagonal matrix of variance functions v(uij)
R
i
() is the n x n “working” correlation matrix of Y
i
indexed by a vector of
parameters.
The GEE estimator of is the solution of
N
D
i
’
[V(â)]
-1
(Y
i
-µ
i
) = 0;
i=1
where â is a consistent estimate of and D
i
=µ
i
/.
Because the above equation depends only on the mean and the variance of Y, these
are quasi-likelihood estimates. Therefore, GEE avoids the need for joint distributions
of dependent variables by only assuming a functional form for the marginal
distribution at each time point. Estimation of the parameters involves iterating
between the quasi-likelihood solution for estimating and a robust method for
estimating as a function of.
47
CHAPTER 4: STUDY RESULTS
I. Statin therapy in the Medi-Cal type 2 diabetic patients
Using the claims data from the California Medicaid program during the period from
January, 1995 to December, 2004, we identified 28,065 diabetic patients who were
over 18 years old, and eligible for at least 12 months before and after the diagnosis of
type 2 diabetes. A total of 22,189 patients were selected after excluding patients
who were possibly type 1 diabetes. Among the 22,189 patients with type 2 diabetes,
61.7% of them received statin therapy at some time point during the study period.
About 87.7% of the statin therapy was initiated after the diagnosis of diabetes. After
excluding patients who had used lipid-lowering drugs during the two years period
before the initial statin therapy, we derived the first study population of 7,015
diabetic patients who were statin users. Based on this patient population, we
examined the clinical trend of statin therapy in the Medi-Cal type 2 diabetic patients.
1. Patient characteristics
The characteristics of the 7,015 type 2 diabetic patients were displayed in Table 4.1.
Most of these patients (62.55%) were female. The majority (90%) of the patients
were over 40 years old. Patients over 65 years old accounted for 48.3% of the study
48
population. About 33% of patients were Caucasian, Hispanics and African American
each comprised of 10% of the study population.
Table 4.1: Frequency distribution of patient characteristics (N =7,015)
Characteristics Frequency Percent (%)
Sex
F 4388 62.55
M 2627 37.45
Age Group
18-39 633 9.02
40-49 993 14.16
50-64 2001 28.52
65+ 3388 48.3
Ethnicity
White 2321 33.09
Hispanic 735 10.48
African
American
673 9.59
Others 526 7.5
Unknown 2760 39.34
Dual eligibility
0 2141 30.52
1 4874 69.48
Calendar year
of diabetes
diagnosis
1996 1287 18.35
1997 978 13.94
1998 858 12.23
1999 774 11.03
2000 887 12.64
2001 882 12.57
2002 762 10.86
2003 587 8.37
49
2. Time between diabetes diagnosis and the initial statin presciption
In the year 1996, only 190 patients newly diagnosed with type 2 diabetes were
treated with statin. The number of patients who were treated with statin increased
dramatically during the study period (Table 4.2). Among the 7,015 patients who
received statin therapy, about 23% of them started the treatment within 3 months
after the diagnosis of diabetes. In most patients (57.18%), statin therapy was initiated
a year after them being diagnosed with type 2 diabetes. Over the period from the
year 1996 to 2003, the trend of starting statin therapy earlier in the diabetes disease
course was observed (Table 4.3 and Figure 4.1). Among statin users who were
diagnosed with type 2 diabetes in 1996, only about 23% of them were treated within
12 months after the diagnosis of diabetes. This number increased to 42% in the year
of 2000. For statin users diagnosed with type 2 diabetes in year 2003, actually more
than 87% of them were treated within a year after the diagnosis, with 57% of these
patients starting statin in less than 3 months after the diagnosis (Figure 4.2). The
index statin was presented in Table 4.4 by the year in which the statin therapy was
initiated.
50
Table 4.2: Number of patients being treated with statin by index statin year
Index statin year Frequency Percent (%)
1996 190 2.71
1997 385 5.49
1998 505 7.2
1999 624 8.9
2000 753 10.73
2001 1108 15.79
2002 1178 16.79
2003 1268 18.08
2004 1004 14.31
Total 7015 100
Table 4.3: Average time between diabetes diagnosis and the initial statin therapy by
index diagnosis year (Total number of patients = 7, 015)
Time (months)
Year N (%)
Mean Std Dev
1996 1287 (18.35) 40.86 29.95
1997 978 (13.94) 36.23 27.61
1998 858 (12.23) 28.75 23.66
1999 774 (11.03) 25.73 19.93
2000 887 (12.64) 19.77 16.51
2001 882 (12.57) 13.87 12.99
2002 762 (10.86) 9.58 9.56
2003 587 (8.37) 4.72 5.68
51
Figure 4.1: Average time (months) between diabetes diagnosis and the initial statin
therapy by index diagnosis year
Figure 4.2: Time between diabetes diagnosis and the initial statin therapy by the
index diagnosis year
Time between the diabetes diagnosis and the first statin
prescription
0
1 0
20
30
40
50
60
70
80
90
1 996 1 997 1 998 1 999 2000 2001 2002 2003
P roportion of patients(%)
less than 3 months between 3 and 6 months
between 6 and 1 2 months more than 1 2 months
52
Table 4.4: Index statin by the index statin year
Index statin year Proportion(%)
Index statin
1996 1997 1998 1999 2000 2001 2002 2003 2004
Atorvastatin 0 0.26 14.26 51.76 54.58 54.06 57.13 55.05 62.25
Fluvastatin 19.47 17.66 14.26 9.29 7.57 4.24 3.57 3.71 2.99
Lovastatin 38.95 17.14 10.69 0 0 0.09 0.25 1.34 1.39
Niacin/Lovastatin 0 0 0 0 0 0 0.17 0.47 1.29
Pravastatin 32.63 42.6 38.42 24.52 21.25 19.49 18.34 17.19 11.65
Rosuvastatin 0 0 0 0 0 0 0 0 5.98
Simavastatin 8.95 22.34 22.38 14.42 16.6 22.11 20.54 22.24 14.44
53
3. Time to statin therapy by age and ethnicity
Figure 4.3 shows the time to statin therapy in different ethnic groups. In the first few
years of the study period, comparing to other ethnic groups, it generally took longer
for African American patients to start statin treatment after being diagnosed with
type 2 diabetes. In the year 1996, the time between the diagnosis of diabetes and
initial statin treatment averages 50 months for African Americans, while it was only
37.86 months for Caucasian patients. However, this situation has been improved in
recent years. In the year 2003, treatment differences among different ethnic groups
almost diminished.
Through the study period, time to statin therapy after the diagnosis of type 2 diabetes
decreased consistently across all age groups (Figure 4.4). However, for patients
younger than 40, time to initial statin treatment experienced more dramatic changes.
It dropped sharply by 15 months from year 1997 to 1998 and 13 months from year
2000 to 2001. Since then, there was no difference in time to statin treatment among
different age groups.
54
Figure 4.3: Time to statin therapy in different ethnic groups
0
10
20
30
40
50
60
1996 1997 1998 1999 2000 2001 2002 2003
Year of diagnosis
Time (in months)
Asian Black Hispanic White
Figure 4.4: Time to statin therapy in different age groups
Time to statin therapy after diabetes diagnosis
0
10
20
30
40
50
1996 1997 1998 1999 2000 2001 2002 2003
year of diagnosis
Time(in months)
Age < 40 Age between 40 and 50
Age between 50 and 65 Age over 65
55
4. Factors affecting time to statin therapy in the primary prevention
To identify the factors that possibly affect the time to statin therapy in the primary
prevention of cardiovascular complications in type 2 diabetic patients, a generalized
linear regression model was fitted. Dependent variable was the number of months
between the diagnosis of the type 2 diabetes and the initial statin prescription.
Independent variables included the index diagnosis year of the diabetes, age, gender,
dual eligibility, and ethnicity. The result from the regression model was presented in
Table 4.5. It showed that the index year of diabetes diagnosis was a significant
predictor of how soon patients receive statin therapy after they were diagnosed with
type 2 diabetes (p<0.01). Time between diabetes diagnosis and statin treatment was
significantly shortened in recent years. Ethnicity also affected how soon patients
received statin therapy after being diagnosed of diabetes. Comparing to Caucasians,
statin treatment in African American patients was significantly delayed. Statin
therapy was initiated earlier in older patients, but gender is not a significant predictor
of time to statin therapy in type 2 diabetes. Whether patients were dually eligible did
not affect how soon patients got statin therapy after the diagnosis of type 2 diabetes.
5. Statin therapy in the secondary prevention
Among the 7,015 patients who received statin therapy, 29.4% of them had
cardiovascular event(s) before statin therapy was initiated. The time to statin therapy
since the occurrence of the cardiovascular event deceased steadily over the study
56
period (Figure 4.5). In 1996, only 13% of patients received statin therapy within 3
months after the cardiovascular event happened. However, for most of patients who
had cardiovascular event(s), initiation of statin therapy was delayed, with more than
73% of them being treated 12 months after the event happened. This situation has
been continuously improved through the study period. In year 2004, statin therapy
was initiated in more than 87% of patients within 3 months after the cardiovascular
event.
Table 4.5: Factors affecting time to statin therapy in the primary prevention
Parameter Estimate
Standard
Error P-value
Year of diabetes
diagnosis
-4.786 0.131 <.0001**
Age -0.084 0.026 0.0011**
Female 0.135 0.616 0.826
Ethnicity
African American 3.684 1.113 0.0009**
Hispanic 1.758 1.033 0.0888
Other -1.678 1.186 0.1574
Unknown 0.843 0.712 0.2366
White . . .
Dual eligible 1.894 0.765 0.0133
** P<0.01, R-square=0.25
57
Figure 4.5: Time to statin therapy in patients with baseline cardiovascular event(s)
Time between CVD and initial statin prescription
0
10
20
30
40
50
60
70
80
90
100
1996 1997 1998 1999 2000 2001 2002 2003 2004
Calendar Year
Proportion of patients (%)
starting statin w ithin 3 months after CVD
starting statin one year after CVD
58
II. Compliance to statin therapy in type 2 diabetic patients
1. Selection of the study population
As mentioned in the study design section, the compliance study was based on the
second study population, which was generated according to the sample selection
criteria described in Figure 3.1. Totally, 4,422 patients were included in the study of
medication compliance to statin therapy in type 2 diabetic patients.
2. Characteristics of the study population
The characteristics of the study population were displayed in Table 4.6. Female
patients accounted for 63.5% of the study population. More than half (51.6%) of the
patients were over 65 years old. The study population comprised of 32.7%
Caucasians, 9.2% Hispanics, and 9.54% African Americans. The study included
similar number of patients being diagnosed with type 2 diabetes in each calendar
year.
59
Table 4.6: Characteristics of the compliance study population (N=4,422)
Characteristics Frequency Percent (%)
Sex
Female 2,807 63.48
Male 1,615 36.52
Age Group
18-39 341 7.71
40-49 552 12.48
50-64 1,246 28.18
65+ 2,283 51.63
Ethnicity
White 1,445 32.68
Hispanic 406 9.18
African American 422 9.54
Others 347 7.85
Unknown 1,802 40.75
Dual eligibility
3,141 71.03
Index diagnosis year
1996 699 15.81
1997 566 12.8
1998 545 12.32
1999 503 11.37
2000 580 13.12
2001 616 13.93
2002 498 11.26
2003 415 9.38
Index statin year
1996 116 2.62
1997 249 5.63
1998 310 7.01
1999 412 9.32
2000 481 10.88
2001 738 16.69
2002 772 17.46
2003 850 19.22
2004 494 11.17
60
3. Comorbidities and healthcare service utilization among the study population
Among the 4,422 patients being selected for the compliance study, 1,227(27.75%) of
them had cardiovascular event(s) during the baseline year. The Charlson comorbidity
index was less than 3 in 52.4% of patients (Figure 4.6). At the baseline year, 950
patients (21.48%) were hospitalized at least once for all causes; 775 patients
(17.53%) had at least one visit to the emergency room. During the first quarter after
the initiation of the statin therapy, 6.56 % of patients had CVD event(s); 7.08% of
patients were hospitalized; and 6.45% of patients had ER visit. Table 4.7 showed the
Healthcare utilization and costs over the first 2 years after statin therapy was
initiated.
Figure 4.6: Charlson comorbidity index at baseline year
Baseline CCI of the study population
0
5
10
15
20
25
30
35
12 3 456 789 10 11 12 13 14 15 16 17 18
CCI
Proportion of patients (%)
61
Table 4.7: Healthcare utilization and costs during the first two years of statin therapy
Quarter
Variable (frequency) 1 2 3 4 5 6 7 8
Any Hospitalization
313 258 249 300 251 248 209 214
Any ER visit
285 245 232 215 184 173 168 139
Any CVD event
290 216 197 184 181 145 148 152
Average number of outpatient visits
per patient 12.7 11.5 10.8 10.4 9.9 9.1 8.2 7.4
Average number of distinct
medications per patient 7.4 6.6 6.3 6.1 5.7 5.4 5.1 4.7
Average medical costs($) per patient
12,363 12,423 11,809 12,407 13,464 13,465 11,297 10,647
62
4. Compliance to statin therapy in type 2 diabetic patients
Medication compliance to statin therapy in patients with type 2 diabetes was
assessed quarterly. As discussed before, the Proportion of drug days covered (PDC)
was used to measure patient compliance to the treatment. The average PDC during
the first quarter since the index statin date was 0.83 (SD=0.234). Based on the 80%
cutoff point, about 65% of patients were classified as compliant. Medication
compliance in the first quarter after the initiation of statin therapy was presented in
Table 4.8. It showed that the proportion of compliers varied among different
ethnicities (p=0.0034), but there was no statistical difference among different gender
and age groups. Also no difference was found between patients with and without
baseline cardiovascular events.
Medication compliance to statin therapy over time
Figure 4.7 and Figure 4.8 showed patient compliance and the proportion of patients
who were compliant to statin therapy in each quarter during the study period.
Medication compliance to statin therapy was high in the beginning quarter. However,
patient compliance dropped sharply during the first year of the treatment. At the end
of the second quarter, the average PDC decreased from 0.83 (SD=0.23) to 0.63
(SD=0.38) while the proportion of compliers decreased from 65% to 51%. By the
end of the first year, the average PDC further decreased to 0.56 and only 43%
patients were compliant to the treatment. Starting from the second year, medication
compliance to statin therapy decreased very slowly and then it stayed relatively
63
stable over the study period. By the end of 20th and 30th quarters, the average PDC
was around 0.4 and only 35% of patients were compliers.
Table 4.8: Medication compliance to statin therapy during the first quarter of
treatment
Characteristics
Proportion of
compliers P-value
Age Group
18-39 66.86%
40-49 57.98%
50-64 66.37%
65+ 64.87%
0.0443
Gender
Female 64.52%
Male 65.33%
0.5881
Ethnicity
White 67.96%
African American 59.11%
Hispanic 60.43%
Other 63.98% 0.0034**
Baseline CVD
No 64.13%
Yes 66.59% 0.1261
** Chi-square test, P<0.01. Complier was defined as patients with PDC
0.8.
64
Figure 4.7: Medication compliance to statin over time
Compliance to statin over time
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
13 57 9 11 13 15 17 19 21 23 25 27 29
Quarter since index statin date
Compliance (PDC)
65
Figure 4.8: Proportion of patient compliant to statin therapy over time
Proportion of patients compliant to statin therapy over time
0
10
20
30
40
50
60
70
135 79 11 13 15 17 19 21 23 25 27 29
Quarter since index statin date
Proportion of patients (%)
Note: Complier was defined as PDC
0.8; Non-complier: PDC<0.8.
Medication compliance to statin therapy by the index statin year
We compared patient’s compliance to statin therapy during the first 12 quarters by
index statin year and the result was presented in Figure 4.9. It showed that patients
starting with statin in recent years tend to comply better with the treatment over time.
Comparing to those starting statin therapy in 1997, patients started in year 2001
showed significant improvement in medication compliance to the treatment. They
demonstrated higher PDC in the first quarter of the therapy and compliance
decreased at a much slower rate over time. The proportion of compliers among
patients starting statin in year 1997 was 63.86% in the first quarter (vs. 69.51%
66
among patients starting in year 2001). At the end of first year treatment, only 35.4%
of patients were compliers among those starting the therapy in year 1997 as opposed
to 45.6% patients among those starting in year 2001.
Medication compliance to statin therapy in different ethnic and age groups
As showed in Figure 4.10, among the three major ethnic groups, Caucasian patients
started with much higher proportion of compliers than Hispanic and African
American patients. Over the study period, Caucasian patients continued to
demonstrate better compliance to statin treatment, especially compared to Hispanics.
During the first year of the treatment, compliance rate among Hispanic patients fell
sharply from 59% in the first quarter to 31% by the end of the fourth quarter. Similar
compliance rate was observed in diabetic patients over 50 years old and those with
younger age (Figure 4.11).
67
Figure 4.9: Proportion of patients compliant with statin therapy by index statin year
Proportion of compliers over time by index statin year
0
10
20
30
40
50
60
70
80
12 3 4 56 7 8 9 10 11 12
Quarter since index statin date
Proportion of patients (%)
Year 1997 Year 1999 Year 2001
Note: Complier was defined as PDC
0.8; Non-complier: PDC<0.8.
68
Figure 4.10: Proportion of patients compliant to statin by ethnicity
Proportion of compliers by ethnicity
0
10
20
30
40
50
60
70
80
13 5 7 9 11 13 15 17 19 21 23 25
Quarter since index statin date
Proportion of patients (%)
Black Hispanic White
Figure 4.11: Proportion of patients compliant to statin by age group
Proportion of compliers by age
0
10
20
30
40
50
60
70
80
90
1 35 79 11 13 15 17 19 21 23 25 27 29
Quarter since index statin date
Proportion of patients (%)
Age betw een 19-39 Age betw een 40-49
Age betw een 50-64 Age >=65
Note: Complier was defined as PDC
0.8; Non-complier: PDC<0.8.
69
Identify factors affecting patient’s compliance to statin
In order to identify the factors that that affect medication compliance to statin
therapy in type 2 diabetes, we performed a longitudinal data analysis using the
generalized estimating equations model (GEE). As medication compliance over time
is a dynamic process, we fitted two separate models in order to determine the effects
of time-varying variables on patient compliance with statin therapy. The dependent
variable was medication compliance defined as 80% or greater PDC in each quarter.
The covariates included in both model 1 and model 2 were: 1) patient demographic
characteristics such as age, gender, and ethnicity; 2) any hospitalization or ER visit
during the baseline year; 3) baseline comorbidities (CCI), cardiovascular and
cerebrovascular conditions at baseline, congestive heart failure (CHF), depression,
and the total number of distinct medications; 4) the index statin year, and the time to
statin therapy after the diagnosis of type 2 diabetes. In addition to theses variables,
model 2 also included the following time-varying variables: utilization of health
services; number of distinct medications; cardiovascular event and CHF developed
in each quarter after the initiation of statin therapy.
Model fitness was assessed using the likelihood ratio test. The results revealed that
model 2 fitted the data better than model 1 (Chi-square test: P<0.001). Results from
model 2 showed that medication compliance to statin in type 2 diabetes was
significantly related to patient’s characteristics (Table 4.9). Compare to Caucasians,
70
Hispanics were less likely to be compliant with the treatment (P<0.01). Patients
using more medications at baseline were less compliant. Patients who had
hospitalization or ER visit during the baseline year tended to be more compliant with
the treatment. But hospitalization happened after the initiation of the statin therapy
actually had negative effect on patient’s compliance. Patients started the statin
therapy in recent years were more likely to be compliant. Baseline comorbidities
and cardiovascular complications developed after the initiation of statin therapy did
not influence patient’s compliance to the treatment.
71
Table 4.9: Factors influencing the medication compliance to statin in patients with
type 2 diabetes
Parameter Estimate
Standard
Error
95% Confidence
Limits P-value
Age group
18-39 0.0134 0.1039 -0.1902 0.2171 0.8971
40-49 0.0642 0.0856 -0.1036 0.2321 0.4533
50-64 0.0588 0.0605 -0.0598 0.1774 0.331
Female -0.0301 0.0475 -0.1233 0.0631 0.5266
Ethnicity
Hispanic -0.4202 0.0873 -0.5913 -0.2491 <.0001**
African American -0.1679 0.0856 -0.3357 -0.0002 0.0498
Other -0.0267 0.091 -0.2052 0.1517 0.769
Unknown -0.0292 0.0542 -0.1355 0.0771 0.5906
Dual eligible 0.1878 0.0651 0.0601 0.3155 0.0039**
Baseline
variables
CCI -0.0201 0.0141 -0.0477 0.0074 0.1525
cardiovascular
and
cerebrovascular
conditions
0.0047 0.0569 -0.1068 0.1163 0.9339
Congestive heart
failure
0.0363 0.0715 -0.1039 0.1765 0.6116
Depression -0.0002 0.0705 -0.1384 0.138 0.9975
Number of
distinct
medications
-0.0596 0.0048 -0.069 -0.0503 <.0001**
Hospitalization 0.2584 0.0611 0.1387 0.3782 <.0001**
ER 0.1931 0.0621 0.0715 0.3148 0.0019**
Time to statin
therapy after
diabetes
diagnosis
-0.0013 0.0012 -0.0037 0.0011 0.2874
Index statin year 0.1242 0.012 0.1008 0.1477 <.0001**
Time-varying
variables
Number of
distinct
medications
0.1675 0.0053 0.1571 0.178 <.0001**
72
Table 4.9, continued
Hospitalization -0.6589 0.0468 -0.7506 -0.5671 <.0001**
ER 0.0418 0.0521 -0.0604 0.144 0.4223
Number of
outpatient visits
-0.0055 0.0012 -0.0077 -0.0032 <.0001**
Cardiovascular
and
cerebrovascular
complications
0.0144 0.071 -0.1248 0.1535 0.8397
Congestive heart
failure
-0.245 0.0983 -0.4377 -0.0524 0.0127
** P<0.01. Dependent variable: Medication compliance, defined as PDC
80% in
each quarter.
73
III. Compare medication compliance to antidiabetics and statin therapy in type
2 diabetes
To compare patient compliance to antidiabetics and statin therapy over the study
period, medication compliance to antidiabetic agents and statin was both assessed
from the first statin prescription date (index statin date). By the time they started
statin therapy, these patients had been on antidiabetic treatment for an average of 21
months. During the first quarter after the initiation of statin therapy, patients
demonstrated a slightly higher level of compliance to statin than to antidiabetics
(65% vs 61%, P<0.01) (Figure 4.12). However, over time, the proportion of patients
compliant with statin therapy dropped much more rapidly than that of antidiabetics.
Starting from the second quarter after the initiation of statin treatment, patient
compliance rate to statin appeared to be significantly lower than that of antidiabetics
(P<0.01). By the end of the third year of statin therapy, 56% of patients were
compliant with antidiabetics while only 38% of patients were complaint to statin
therapy. Since then, a relatively stable gap (about 20%) between the compliance rate
to antidiabetics and statin was maintained through the study period. As showed in
Figure 4.13, during the first quarter after the initiation of statin therapy, the majority
of patients (46.6%) were compliant to both antidiabetic and statin treatment. But this
number dropped down to 37% by the end of second quarter. Although in the first
quarter, the number of patients that were only compliant to statin therapy (19%) was
slightly more than that of hypoglycemics (14%), over the following study quarters,
74
the latter continuously outnumbered the compliers to statin therapy. At the same
time, the number of patients that were compliant to neither statin nor hypoglycemics
kept increasing.
Figure 4.12: Proportion of patients compliant to antidiabetics and statin
Proportion of compliers to statin vs. antidiabetics
0
10
20
30
40
50
60
70
13 579 11 13 15 17 19 21 23 25 27 29
Quarter since index statin
Proportion of patients (%)
antidiabetics Statin
Note: Complier was defined as PDC
0.8; Non-complier: PDC<0.8.
75
Figure 4.13: Medication compliance to both statin and antidiabetics
Medication compliance to both statin and antidiabetics
0
5
10
15
20
25
30
35
40
45
50
13579 11 13 15 17 19 21 23 25 27 29
Quarter since index statin date
Proportion of patients (%)
Series1 Series2
Series3 Series4
Note: Series 1: Compliance with both statin and antidiabetics; Series 2: Compliance with
antidiabetics but not statin; Series 3: Compliance with statin but not antidiabetics; Series 4:
Compliance with neither of them. Compliance was defined as PDC
80% in each quarter.
76
IV. The impact of medication compliance to statin therapy on disease outcomes
in patients with type 2 diabetes
1. Impact of compliance to statin therapy on cardiovascular risks
A marginal structural model was employed in this study to determine the impact of
medication compliance to statin therapy on the cardiovascular risk in patients with
type 2 diabetes. We also fitted a traditional Cox proportional hazard model with the
same study variables, but without adjusting for the confounding effect from time-
varying variables. Result from the marginal structural model was presented in Table
4.10. It showed that after adjusting for the study covariates, patient’s compliance to
statin therapy was significantly related to decreased cardiovascular risk in patients
with type 2 diabetes (hazard ratio =0.24; 95% CI : 0.093 - 0.615; P<0.01).
Cardiovascular risk increased in older patients (hazard ratio = 1.16) and patients with
more medications. In contrast, results from the Cox model failed to demonstrate the
benefit of medication compliance to statin therapy (hazard ratio =1.09; P=0.0003).
2. The impact of medication compliance to statin therapy on hospitalization and
medical costs
GEE model was used to examine the impact of medication compliance to statin
therapy on hospitalization and total medical costs. The binomial distribution and the
logit link function were chosen to model the effect of medication compliance on
77
hospitalization. Log transformation was applied to account for the skewness of the
cost distribution; the normal distribution and the identity link function were chosen
in the analysis of the impact of patient compliance on treatment costs.
Both models were adjusted for study covariates including patient’s demographic
characteristic variables, baseline comorbidities (CCI), utilization of health services at
baseline. For the cost analysis, the model is also adjusted for the total medical costs
from the previous period.
The results from the GEE model revealed that in patients with type 2 diabetes, better
medication compliance to statin therapy is significantly related with decreased
hospitalization (odds ratio =0.74; P<0.001) (Table 4.11). However, after adjusting
for the study covariates, medication compliance to statin therapy did not lead to
lower medical costs for the California Medicaid program. A separate analysis
showed that better medication compliance to the treatment was significantly related
with higher prescription drug costs (Table 4.12).
78
Table 4.10: The impact of patient compliance to statin therapy on CVD events
Parameter Estimate
Standard
Error
95% Confidence
Limits P-value
Age
0.1513 0.0451 0.0629 0.2398 0.0008**
Female 1.0528 1.1479 -1.1970 3.3026 0.3591
Ethnicity
Hispanic 3.8819 1.8268 0.3014 7.4623 0.0336
African American 2.4237 1.6393 -0.7893 5.6367 0.1393
Other 4.7236 1.8361 1.125 8.3222 0.0101
Unknown 2.2528 1.2675 -0.2315 4.7372 0.0755
Dual eligible
-3.3757 1.4414 -6.2009 -0.5506 0.0192
Baseline variables
CCI
-0.2078 0.2535 -0.7046 0.289 0.4124
cardiovascular and
cerebrovascular
conditions
1.3432 0.9513 -0.5214 3.2078 0.158
Congestive heart
failure
0.731 0.9375 -1.1066 2.5686 0.4356
Number of distinct
medications
0.253 0.0505 0.1541 0.3519 <.0001**
Hospitalization
-2.1128 1.3758 -4.8094 0.5837 0.1246
ER
-1.7958 0.7523 -3.2703 -0.3213 0.017
Index statin year
-0.4197 0.1974 -0.8065 -0.0329 0.0335
Time-varying
variables
Medication
compliance
-1.4328 0.483 -2.3795 -0.486 0.003**
Antihypertensive
med
-1.1597 1.2298 -3.5699 1.2506 0.3457
Antiplatelet med
2.7888 0.8707 1.0821 4.4954 0.0014**
Number of distinct
medications
-0.0535 0.1204 -0.2895 0.1825 0.657
Hospitalization
0.6918 0.15029 0.3973 0.9864 <.0001**
ER
0.0131 2.9873 -5.8418 5.8681 0.9965
Number of outpatient
visits
0.021 0.0128 -0.0041 0.0461 0.1014
** P<0.01. Medication compliance, defined as PDC
80% in each quarter.
79
Table 4.11: The impact of patient compliance to statin therapy on hospitalization
Parameter Estimate
Standard
Error
95% Confidence
Limits P-value
Age
0.01 0.0027 0.0047 0.0153 0.0002**
Female -0.0206 0.0517 -0.1219 0.0807 0.6905
Ethnicity
Hispanic 0.0215 0.1051 -0.1844 0.2275 0.8377
African American 0.0633 0.0918 -0.1166 0.2433 0.4902
Other -0.3461 0.1144 -0.5704 -0.1218 0.0025**
Unknown -0.0862 0.0568 -0.1975 0.0252 0.1293
Dual eligible 0.23 0.0798 0.0735 0.3864 0.004**
Baseline
variables
CCI 0.0768 0.0128 0.0518 0.1019 <.0001**
cardiovascular
and
cerebrovascular
conditions
0.1796 0.0572 0.0674 0.2918 0.0017**
Number of
distinct
medications
0.0258 0.0045 0.0171 0.0346 <.0001**
Hospitalization 0.7212 0.0592 0.6051 0.8374 <.0001**
ER 0.0595 0.064 -0.0659 0.1848 0.3524
Index statin year -0.0383 0.013 -0.0638 -0.0129 0.0032**
Time-varying
variable
Medication
compliance
-0.3056 0.042 -0.3879 -0.2233 <.0001**
** P<0.01. Medication compliance, defined as PDC
80% in each quarter.
80
Table 4.12: The impact of patient compliance to statin therapy on prescription drug
costs
Parameter Estimate
Standard
Error
95% Confidence
Limits P-value
Age
-0.0004 0.001 -0.0023 0.0015 0.6977
Female -0.018 0.0203 -0.0578 0.0218 0.3743
Ethnicity
Hispanic 0.0852 0.041 0.0047 0.1656 0.038
African American 0.0656 0.0404 -0.0136 0.1447 0.1046
Other 0.0171 0.0396 -0.0606 0.0947 0.6665
Unknown 0.0546 0.0228 0.0099 0.0993 0.0166
Dual eligible 0.0574 0.0284 0.0018 0.113 0.0431
Baseline variable
CCI -0.0361 0.0061 -0.048 -0.0242 <.0001**
Cardiovascular and
cerebrovascular event
-0.0496 0.0243 -0.0972 -0.002 0.0413
Number of distinct
medications
0.1148 0.0044 0.1062 0.1234 <.0001**
Hospitalization -0.1966 0.0334 -0.262 -0.1312 <.0001**
ER 0.1436 0.0297 0.0853 0.2019 <.0001**
Number of outpatient visits
-0.0021 0.0006 -0.0032 -0.0009 0.0008**
Index statin year -0.0616 0.0051 -0.0715 -0.0517 <.0001**
Time-varying variable
Total costs from the
previous quarter
0.7596 0.007 0.7459 0.7734 <.0001**
Medication compliance
0.7215 0.0194 0.6835 0.7595 <.0001**
** P<0.01. Medication compliance, defined as PDC
80% in each quarter.
81
CHAPTER 5: DISCUSSION AND CONCLUSION
In patients with type 2 diabetes, control of dyslipidemia with lipid-lowering agents is
important in lowering their risk for macrovascular complications. The benefit of
statin therapy in reducing the cardiovascular risk has been well demonstrated in
clinical trials, though only few of them were conducted exclusively in patients with
type 2 diabetes. Current treatment guidelines suggest statin as the first-line drug for
the management of dyslipidemia in diabetes (ADA, 2004). However, few studies
have examined the issue of medication compliance to statin therapy in patients with
type 2 diabetes in the real-world settings, and surprisingly little is known about the
relationship between non-compliance to statin therapy and disease outcomes in
diabetic patients.
In this study, we examined the statin therapy in type 2 diabetic patients enrolled in
the California Medicaid program from the period of January 1995 to December
2004. To our knowledge, this is the first longitudinal study that specifically
examined the issue of medication compliance to statin therapy and its impact on
various disease outcomes in patients with type 2 diabetes. To overcome the time-
varying confounding effects, we took the longitudinal study design and applied state-
the marginal structural model to study the impact of patient compliance on
82
cardiovascular risk in diabetic patients. The approach of generalized estimating
equations was employed to examine its effect on economic outcomes. In the
following section, we summarized the findings from the study and discussed their
clinical and policy implications in the management of type 2 diabetes. The study
limitations were also briefly discussed.
Summary of study findings
Our study found that in patients who received the treatment after the diagnosis of
type 2 diabetes, statin therapy was initiated much earlier in recent years. Among
statin users who were diagnosed with type 2 diabetes in 1996, only about 23% of
them were treated within 12 months after the diagnosis. For those being diagnosed in
year 2003, actually 87% of them were treated within a year after being diagnosed of
type 2 diabetes, with 57% of these patients starting statin in less than 3 months after
the diagnosis. Comparing to Caucasian patients, the initiation of statin treatment in
African American patients were significantly delayed.
Compliance to statin therapy is especially important for patients with type 2 diabetes
given the associated high cardiovascular risk with the disease. However, findings
from this study suggest poor compliance to the treatment in the Medi-Cal diabetic
population. Although patients generally started with high compliance rate at the
beginning of the treatment, patient compliance continuously dropped during the
study period, with the biggest decrease being observed within the first year after the
83
treatment was initiated. The medication compliance rate found in this research is
much lower than what has been reported from the clinical trials (Collins R, 2003). In
the literature, however, similar pattern of medication compliance to statin therapy in
real practice has been reported previously in other patient population (Benner JS,
2002). Our study also revealed that patients with type 2 diabetes were more likely to
be compliant with antidiabetic agents than to statin therapy. According to our
knowledge, this is the first longitudinal study that specifically compares patient
compliance to hypoglycemics and lipid-lowering agents in type 2 diabetes over long-
term period.
In comparison with Caucasians, Hispanic patients demonstrated poorer compliance
to statin therapy. This is of particular concern because Hispanics have a higher
prevalence of diabetes and are at higher risk for CHD than whites (Winer N, 2004;
American Heart Association, 2006). Previous findings from a recent retrospective
study (Yood MU, 2006) have reported that African American patients were less
likely to be compliant with statin therapy. In this research, we also found that
patients who experienced hospitalization during the baseline year were more likely to
be compliant to statin therapy, but hospitalization occurred after the initiation of the
therapy has a negative impact on patient’s compliance.
Recent studies found that poor compliance to statins was associated with an
increased cholesterol levels and cardiovascular morbidity (Parris ES, 2005;
84
Blackburn DF, 2005; Mozaffarian D, 2004). Although generally, the importance of
medication compliance to statin therapy has now been realized by the healthcare
professionals, the actual relationship between medication compliance to statin
therapy and disease outcomes still remains largely unknown in many disease states.
To understand the impact of patient compliance to statin therapy in type 2 diabetes is
especially urgent because of the important clinical and economic consequences
related to cardiovascular complications in these patients. In this research, we found
that poor compliance to statin therapy was significantly associated with worse
disease outcomes in type 2 diabetes. Increased cardiovascular risk and healthcare
service utilization were found in patients who were non-compliant to statin therapy,
though we did not find that better medication compliance to statin therapy
necessarily lead to lower total medical costs in the management of type 2 diabetes.
Clinical and policy implications
Our study findings have important clinical implications for clinicians, diabetes
disease management programs, and the healthcare system. Type 2 diabetes is a
chronic disease that is very costly to treat. Reducing the risk of cardiovascular
complication with lipid-lowering therapy is very important in the management of
type 2 diabetes. Although lipid-lowering therapy should be life-long in type 2
diabetic patients, and current treatment guideline have emphasized the importance of
lipid-lowering therapy in the management of type 2 diabetes, our results suggest that
in real practice, medication compliance to statin therapy in these patients is far from
85
desirable and it is especially difficult for patients with type 2 diabetes to maintain
medication compliance to statin therapy over long-term period. Hence, the benefits
of statin therapy, which has been well demonstrated in clinical trials, might not be
realized if patients do not comply well with the treatment in real practice.
Many factors can influence patient’s compliance to medication regimen. Certain
patient characteristics, such as age and race, are related to medication compliance to
pharmacotherapy in type 2 diabetes (Hertz RP, 2005; Benner JS, 2002). As found in
our research and other studies, medication compliance is also negatively related to
the complexity of patient’s medication regimen (Perreault S, 2005). As a chronic
disease, patients with type 2 diabetes are generally on several different medications.
Even if patients are aware of the potential benefits of statin therapy, yet in the long
term, it might be a challenge for diabetic patients to be compliant with all of their
medications given their medication burden. Therefore, physicians should try to
minimize patient’s medication burden whenever possible by offering a simpler
treatment regimen with fewer medications or less frequent dosing.
Researchers have found that medication compliance can be affected by other factors
including patient’s understanding of the medication regimen, their perception of the
potential benefits of the treatment, medication costs, whether physicians have
provides appropriate information regarding the treatment benefit (Rubin RR, 2005).
However, findings from recent studies suggest that patients on statin therapy have
86
very limited understanding about the treatment. In their study, Yilmaz and colleagues
(Yilmaz MB, 2004) surveyed patients who were on statin therapy for at least three
months and found that 16.5% of them did not understand the reason for their statin
therapy. More than half of the study participants (58%) stated that they did not know
how long they would have to keep using their statins, while only 21% replied that
they would use them continuously. One fifth of participants had some idea about side
effects.
In our study, we found that patients with type 2 diabetes were more compliant to
antidiabetic treatment than to statin therapy. This may be not only due to their
medication burden, but also due to their lack of understanding that in addition to
control blood glucose, control of lipid level is also a very important aspect in the
management of type 2 diabetes. However, unlike non-compliance to statin (which
may not have impact on patients health status in short term), non-compliance to
antidiabetics can result in immediate consequences (such as coma resulted from
diabetic hyperosmolar syndrome and ketoacidosis). This might partially explain the
better compliance with antidiabetics found in the study.
Given the limited understanding in patients who received statin therapy, patient
education through the healthcare system and disease management programs may be a
useful tool to improve their medication compliance to the treatment. For physicians,
they should discuss with patients about their medication regimen by clarifying the
87
potential benefits of statin therapy in type 2 diabetes and emphasizing the importance
of long-term compliance to the treatment. As patients might quit the therapy because
of concerns about the long-term side effect of the drug, it is important for physicians
to discuss potential adverse effects with their patients and reassure their safety given
the highly favorable benefit-risk profile of statins (Sacks FM, 2002). Recent studies
suggest that initiation of statins during hospitalization may increase patient’s
awareness of the treatment benefits and is an effective strategy for long-term statin
use with associated improvement in compliance and long-term clinical benefits.
(Fonarow GC, 2003).
Healthcare system and diabetes disease management program should consider
interventions to improve patient compliance. Interventions should be designed to
ensure that physicians follow the lipid treatment guidelines for treating type 2
diabetic patients. Other possible intervention including nurse led interventions, home
aids, diabetes education, pharmacy led interventions could also be helpful (Vermeire
E, 2005). Diabetes education programs can help patients to understand the
seriousness of the disease and educate them about how they can make changes that
will reduce their risk of diabetes complications and cardiovascular disease. As has
been showed in some studies, pharmacist could play an active role in the
improvement of medication compliance (Lindenmeyer A, 2006). They can provide
patient education on pharmacologic therapy, and reinforce the importance and
purpose of therapy during medication refills, emphasize the need for adherence,
88
identify and resolve barriers to adherence, and provide long-term monitoring of drug
response and feedback to the patient between visits to the primary care provider.
This allows pharmacists to give the patient feedback on their progress and reinforce
the steps to achieving treatment goals.
Although the cost of implementing these interventions might be a concern for
diabetes disease management programs and healthcare system, they might be
actually cost-effective as better compliance with statin therapy can lead to better
disease outcomes in type 2 diabetes.
Contribution of the study to the literature
This study contributes to the literature in the following aspects.
First, this study examined the long-term medication compliance to statin therapy
specifically in patients with type 2 diabetes, who are at high risk for cardiovascular
complications. The findings from this research contribute significantly to the existing
body of research on the relationship between medication compliance to statin therapy
and disease outcomes, and have very important implications in the management of
dyslipidemia in patients with type diabetes.
Second, this study also makes significant methodological contribution to the
literature. Although the MSM model has been around for a while, its application in
89
the healthcare research field is very limited. This research provided an example of
using MSM in handling the time-confounding issue in longitudinal studies.
Given the limited number of observational studies about statin treatment in diabetic
patients, we believe that this study will provide better understanding about the role of
statin treatment in the management of diabetic patients. Findings from this study
will be able to provide more information to help healthcare organizations and disease
management programs to make their decisions about the intervention.
Study limitations
In this study, we examine the impact of medication compliance of statin therapy in
type 2 diabetes in an observational setting. Several limitations of the study need to be
acknowledged.
In this study, we used the claims data from the California Medicaid program;
therefore, it carries the common limitations of claims data. One of the issues is that
the disease diagnosis may not be accurately recorded and some important clinical
variables are not collected. The Medi-Cal data also do not include death data. So
even though some patients may actually die from a cardiovascular event, they might
be treated as censored in the marginal structural model as this was not recorded in
the claims data. As this study is based on the diabetic patients enrolled in the
90
California Medicaid program, the study results might not be generalized to other
patient populations.
The second limitation of the study is about the measurement of medication
compliance. We used pharmacy claims record to assess medication compliance in
the study population. Pharmacy refill analysis addresses the question of drug
availability and may not reflect the actual consumption of medications. However, it
provides crucial insights into patient willingness to comply, and so far, such records
may be the most accurate estimate of compliance for large cohorts available today
(Jackevicius CA, 2002).
Last but not the least, is the methodological limitation. Although the marginal
structural model used in this study has improved in handling the time-confounding
issue in longitudinal studies, like any other statistical models, it still has its own
limitation. Therefore, the results should be interpreted within the contexts of the
model assumptions: no unmeasured confounding and no informative censoring. As
some variables such as laboratory value of lipid level and whether a patient is a
smoker or not are not available in the claims data, this might affect the magnitude of
the effect of statin compliance to some extent. However, we don’t expect this will
change the direction of its impact since: 1) diabetes itself is a high risk factor for
cardiovascular disease; and 2) from a clinical perspective, if a patient was treated
with statin, usually he (or she) might very likely have elevated blood lipid level. As
91
most researchers assumed censoring at random when they perform survival analysis
using claims data, we think the second assumption should not impact our study
results.
Future research
This study provides insights into the issue of medication compliance to statin therapy
in patients with type 2 diabetes. Although our research has linked poor compliance to
statin therapy with increased cardiovascular risk and hospitalization in type 2
diabetic patients, we didn’t find it had significant impact on medical costs to the
California Medicaid Program, which might be due to the limitation of the data.
Therefore, future research is needed in order to clarify this question. Another
important area worth further exploring is how to monitor patient compliance. If
physician or pharmacist can identify the non-compliers at early stage during the
treatment, then certain interventions can be targeted at these patients.
Conclusions
Patient compliance to statin therapy has improved in recent years, but long-term
medication compliance to statin therapy is still far from desirable in the Medi-Cal
type 2 diabetes patients population. Although the benefits of statin therapy in
lowering cardiovascular risk has been well demonstrated from clinical trials, with the
poor long-term compliance to statin therapy in real world-settings, it is doubtful that
92
patients with type 2 diabetes can achieve the full benefits of the treatment. Our study
results should send an alarming signal to both clinicians and decision makers as
given the low long-term compliance rate to statin therapy in real practice, benefits
from the treatment might be overestimated based on the evidence from the clinical
trials. To improve medication compliance to statin therapy in patients with type 2
diabetes, more efforts should be spent on identifying the barriers to medication
compliance and certain interventions might be considered. Physicians, patients, and
healthcare system should work together in order to achieve better compliance and
thus better disease outcomes.
93
BIBLIOGRAPHY
1. American Diabetes Association (ADA). Economic consequences of diabetes
mellitus in the U.S. in 1997. Diabetes Care; 1998; 21: 296-309.
2. American Diabetes Association (ADA). Standards of medical care in
diabetes. Diabetes Care. 2007; 30(suppl 1): S4–S41.
3. American Diabetes Association (ADA). Dyslipidemia management in adults
with diabetes. Diabetes Care. 2004; 27 (Suppl. 1):S68–S71.
4. American Diabetes Association website: http://www.diabetes.org/
5. Anderson RM, Funnell MM. Compliance and adherence are dysfunctional
concepts in diabetes care. Diabetes Educ. 2000 Jul-Aug;26(4):597-604.
Review.
6. Avorn J, Monette J, Lacour A, et al. Persistence of use of lipid-lowing
medications. A cross-national study. JAMA 1998;279:1458–62.
7. Barron Y, Cole SR, Greenblatt RM, Cohen MH, Anastos K, DeHovitz JA,
Delapenha R, Gange SJ. Effect of discontinuing antiretroviral therapy on
survival of women initiated on highly active antiretroviral therapy. AIDS.
2004 Jul 23;18(11):1579-84.
8. Bate KL, Jerums G. Preventing complications of diabetes. Med J Aust. 2003
Nov 3; 179(9):498-503.
9. Benner JS, Glynn RJ, Mogun H, Neumann PJ, Weinstein MC, Avorn J.
Long-term persistence in use of statin therapy in elderly patients. JAMA.
2002; 288:455–461.
10. Bhattacharyya SK. Predicting hospitalisation of patients with diabetes
mellitus. An application of the Bayesian discriminant analysis.
Pharmacoeconomics. 1998;13(5):519-29.
11. Blackburn DF, Dobson RT, Blackburn JL, Wilson TW. Cardiovascular
morbidity associated with nonadherence to statin therapy. Pharmacotherapy.
2005 Aug;25(8):1035-43.
94
12. Bodnar LM, Davidian M, Siega-Riz AM, Tsiatis AA. Marginal structural
models for analyzing causal effects of time-dependent treatments: an
application in perinatal epidemiology. Am J Epidemiol. 2004 May
15;159(10):926-34.
13. Brandle M. et al. Cost effectiveness of statin therapy for the primary
prevention of major coronary events in individuals with type 2 diabetes.
Diabetes Care. 2003; 26:1796–1801.
14. Bryan J, Yu Z, Van Der Laan MJ. Analysis of longitudinal marginal
structural models. Biostatistics. 2004 Jul;5(3):361-80.
15. Chapman RH, Benner JS, Petrilla AA, et al. Predictors of adherence with
antihypertensive and lipid-lowering therapy. Arch Intern Med
2005;165:1147–52.
16. Chobanian, A.V., et al., Seventh report of the Joint National Committee on
Prevention, Detection, Evaluation, and Treatment of High Blood Pressure.
Hypertension, 2003. 42(6): 1206-52.
17. Cleemput I, Kesteloot K, DeGeest S. A review of the literature on the
economics of noncompliance. Room for methodological improvement.
Health Policy. 2002 Jan; 59(1):65-94.
18. Clemens A, Siegel E, Gallwitz B. Global risk management in type 2 diabetes:
blood glucose, blood pressure, and lipids--update on the background of the
current guidelines. Exp Clin Endocrinol Diabetes. 2004;112(9):493-503.
Review.
19. Colhoun HM, Betteridge DJ, Durrington PN, Hitman GA, Neil HA,
Livingstone SJ, Thomason MJ, Mackness MI, Charlton-Menys V, Fuller JH;
CARDS investigators. Primary prevention of cardiovascular disease with
atorvastatin in type 2 diabetes in the Collaborative Atorvastatin Diabetes
Study (CARDS): multicentre randomised placebo-controlled trial. Lancet.
2004;364(9435):685-96.
20. Collins R, Armitage J, Parish S, Sleigh P, Peto R; Heart Protection Study
Collaborative Group. MRC/BHF Heart Protection Study of cholesterol-
lowering with simvastatin in 5963 people with diabetes: a randomised
placebo-controlled trial. Lancet. 2003 Jun 14;361(9374):2005-16.
95
21. Cook NR, Cole SR, Hennekens CH. Use of a marginal structural model to
determine the effect of aspirin on cardiovascular mortality in the Physicians'
Health Study. Am J Epidemiol. 2002 Jun 1;155(11):1045-53.
22. Cramer JA. A systematic review of adherence with medications for diabetes.
Diabetes Care. 2004 May;27(5):1218-24. Review.
23. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for
use with ICD-9-CM administrative databases. J Clin Epidemiol.
1992;45:613– 619.
24. Diggle PJ, Heagerty PJ, Liang KY, Zeger SL. Analysis of longitudinal data.
Oxford University Press, 2002.
25. DiMatteo MR. Variations in patients' adherence to medical
recommendations: a quantitative review of 50 years of research. Med Care.
2004 Mar;42(3):200-9.
26. Dunbar-Jacob J, Erlen JA, Schlenk EA, Ryan CM, Sereika SM, Doswell
WM. Adherence in chronic disease. Annu Rev Nurs Res. 2000;18:48-90.
Review.
27. Fonarow GC, Watson KE. Effective strategies for long-term statin use. Am J
Cardiol. 2003 Jul 3;92(1A):27i-34i. Review.
28. Goldberg RB, Mellies MJ, Sacks FM, et al, for the CARE Investigators.
Cardiovascular events and their reduction with pravastatin in diabetic and
glucose-intolerant myocardial infarction survivors with average cholesterol
levels: subgroup analyses in the Cholesterol and Recurrent Events (CARE)
trial. Circulation 1998; 98: 2513–19.
29. Gordois A, Scuffham P, Shearer A, Oglesby A. The health care costs of
diabetic nephropathy in the United States and the United Kingdom. J
Diabetes Complications. 2004 Jan-Feb;18(1):18-26.
30. Grant RW, O'Leary KM, Weilburg JB, Singer DE, Meigs JB. Impact of
concurrent medication use on statin adherence and refill persistence. Arch
Intern Med. 2004 Nov 22;164(21):2343-8.
31. Green ML. Management of dyslipidemias in the age of statins.
Prim Care. 2003 Dec;30(4):641-69. Review.
96
32. Haffner SM. Statin therapy for the treatment of diabetic dyslipidemia.
Diabetes Metab Res Rev. 2003;19(4):280-7. Review.
33. Haffner SM. Dyslipidemia management in adults with diabetes. Diabetes
Care. 2004;27(Suppl 1):S68-71.
34. Hansson L, Zanchetti A, Westerling S, et al. Effects of intensive blood-
pressure lowering and low-dose aspirin in patients with hypertension:
principal results of the Hypertension Optimal Treatment (HOT) randomised
trial. HOT Study Group. Lance. 1998; 351: 1755-1762.
35. Herman WH, Alexander CM, Cook JR, Boccuzzi SJ, Musliner TA, Pedersen
TR, Kjekshus J, Pyorala K. Effect of simvastatin treatment on cardiovascular
resource utilization in impaired fasting glucose and diabetes. Findings from
the Scandinavian Simvastatin Survival Study. Diabetes Care.
1999;22(11):1771-8.
36. Hernan MA, Brumback BA, Robins JM. Estimating the causal effect of
zidovudine on CD4 count with a marginal structural model for repeated
measures. Stat Med. 2002;21(12):1689-709.
37. Hertz RP, Unger AN, Lustik MB. Adherence with pharmacotherapy for type
2 diabetes: a retrospective cohort study of adults with employer-sponsored
health insurance. Clin Ther. 2005 Jul;27(7):1064-73.
38. Hogan P, Dall T, Nikolov P; American Diabetes Association. Economic costs
of diabetes in the US in 2002.Diabetes Care. 2003 Mar;26(3):917-32.
39. Howell N, Trotter R, Mottram DR, Rowe PH. Compliance with statins in
primary care. The Pharm J. 2004;272(7281):23-26.
40. Jackevicius CA, Mamdani M, Tu JV. Adherence with statin therapy in
elderly patients with and without acute coronary syndromes. JAMA. 2002 Jul
24-31;288(4):462-7.
41. Kalsekar ID. Depression in Patients with Type 2 Diabetes: Impact on
Adherence to Oral Hypoglycemic Agents. The Annals of Pharmacotherapy.
2006; 40(4): 605-611.
42. Kannel WB, McGee DL. Diabetes and glucose tolerance as risk factors for
cardiovascular disease: the Framingham Study. Diabetes Care 1979; 2: 120-
126.
97
43. Kidd KE, Altman DG. Adherence in social context. Control Clin Trials.
2000; 21:184 –187.
44. Kyngas H, Duffy ME, Kroll T. Conceptual analysis of compliance. J Clin
Nurs. 2000 Jan;9(1):5-12. Review.
45. Liang KY, Zeger SL. Longitudinal data analysis using generalized linear
models. Biometrika. 1986; 73: 13-22.
46. Lindenmeyer A, Hearnshaw H, Vermeire E, Van Royen P, Wens J, Biot Y.
Interventions to improve adherence to medication in people with type 2
diabetes mellitus: a review of the literature on the role of pharmacists. J Clin
Pharm Ther. 2006;31(5):409-19.
47. Melikian C, White TJ, Vanderplas A, Dezii CM, Chang E. Adherence to oral
antidiabetic therapy in a managed care organization: a comparison of
monotherapy, combination therapy, and fixed-dose combination therapy. Clin
Ther. 2002;24(3):460-7.
48. Michaud K, Messer J, Choi HK, Wolfe F. Direct medical costs and their
predictors in patients with rheumatoid arthritis: a three-year study of 7,527
patients. Arthritis Rheum. 2003;48(10):2750-62.
49. Mozaffarian D, Nye R, Levy WC. Statin therapy is associated with lower
mortality among patients with severe heart failure. Am J Cardiol. 2004 May
1;93(9):1124-9.
50. MRC/BHF Heart Protection Study of cholesterol lowering with simvastatin
in 20,536 high-risk individuals: a randomised placebo-controlled trial.
Lancet. 2002; 360: 7-22.
51. MRC/BHF Heart Protection Study of cholesterol-lowering with simvastatin
in 5963 people with diabetes: a randomised placebo-controlled trial. Lancet.
2003; 361:2005–2016.
52. Murray MD, Morrow DG, Weiner M, Clark DO, Tu W, Deer MM, Brater
DC, Weinberger M. A conceptual framework to study medication adherence
in older adults. Am J Geriatr Pharmacother. 2004 Mar;2(1):36-43. Review.
53. National Diabetes Information Clearinghouse. http://diabetes.niddk.nih.gov/
98
54. Nichol MB, Shi S, Knight TK, Wallace DJ, Weisman MH. Eligibility,
utilization, and costs in a California Medicaid lupus population. Arthritis
Rheum. 2004 Dec 15;51(6):996-1003.
55. O'Brien JA, Shomphe LA, Kavanagh PL, Raggio G, Caro JJ. Direct medical
costs of complications resulting from type 2 diabetes in the U.S. Diabetes
Care. 1998; 21(7):1122-8.
56. Osterberg L, Blaschke T. Adherence to Medication. N Engl J Med
2005;353:487-97.
57. Paes AH, Bakker A, Soe-Agnie CJ. Measurement of patient compliance.
Pharm World Sci. 1998 Apr;20(2):73-7.
58. Parris ES, Lawrence DB, Mohn LA, Long LB. Adherence to statin therapy
and LDL cholesterol goal attainment by patients with diabetes and
dyslipidemia. Diabetes Care. 2005;28(3):595-9.
59. Perreault S, Blais L, Dragomir A, Bouchard MH, Lalonde L, Laurier C,
Collin J. Persistence and determinants of statin therapy among middle-aged
patients free of cardiovascular disease. Eur J Clin Pharmacol.
2005;61(9):667-74.
60. Phillips KA, Morrison KR, Anderson R., Aday LA. Understanding the
context of healthcare utilization: Assessing environmental and provider-
related variables in the behavioral model of utilization. Health Serv Res.
1998;33:571-596.
61. Pladevall M, Williams LK, Potts LA, Divine G, Xi H, Lafata JE. Clinical
outcomes and adherence to medications measured by claims data in patients
with diabetes. Diabetes Care. 2004 Dec;27(12):2800-5.
62. Pyorala K, Pedersen TR, Kjekshus J, Faergeman O, Olsson AG, Thorgeirsson
G. Cholesterol lowering with simvastatin improves prognosis of diabetic
patients with coronary heart disease. A subgroup analysis of the Scandinavian
Simvastatin Survival Study (4S). Diabetes Care. 1997 Apr;20(4):614-20.
63. Robiner WN. Enhancing adherence in clinical research. Contemp Clin Trials.
2005 Feb;26(1):59-77. Epub 2005 Jan 27. Review.
64. Robins JM, Hernan MA, Brumback B. Marginal structural models and causal
inference in epidemiology. Epidemiology. 2000 Sep;11(5):550-60.
99
65. Rubin RR. Adherence to pharmacologic therapy in patients with type 2
diabetes mellitus. Am J Med. 2005 May;118 Suppl 5A:27S-34S. Review.
66. Shepherd J, Cobbe SM, Ford I, Isles CG, Lorimer AR, MacFarlane PW,
McKillop JH, Packard CJ. Prevention of coronary heart disease with
pravastatin in men with hypercholesterolemia. West of Scotland Coronary
Prevention Study Group. N Engl J Med. 1995 Nov 16;333(20):1301-7.
67. 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.
68. Snow V, Aronson MD, Hornbake ER, Mottur-Pilson C, Weiss KB; Clinical
Efficacy Assessment Subcommittee of the American College of Physicians.
Lipid control in the management of type 2 diabetes mellitus: a clinical
practice guideline from the American College of Physicians. Ann Intern Med.
2004;140(8):644-9.
69. Sullivan SD, Kreling DH, Hazlet TK. Noncompliance with medication
regimens and subsequent hospitalizations: a literature analysis and cost of
hospitalization estimate. J Res Pharm Econ 1990;2:19–33.
70. Tabor PA, Lopez DA. Comply With Us: Improving Medication Adherence.
J. Pharmacy Prac. 2004; 17-167-181.
71. The Long-term Intervention with Pravastatin in Ischaemic Disease (LIPID)
Study Group. Prevention of cardiovascular events and death with pravastatin
in patients with coronary heart disease and a broad range of initial cholesterol
levels. N Engl J Med. 1998; 339: 1349–1357.
72. The National Cholesterol Education Program Adults Treatment Panel III.
Expert Panel on Detection, Evaluation, and Treatment of High Blood
Cholesterol in Adults. Executive summary of the Third Report of the
National Cholesterol Education Program (NCEP) Expert Panel on Detection,
Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult
Treatment panel III). JAMA. 2001; 285:2486-2497.
73. The Scandinavian Simvastatin Survival Study (4S). Randomised trial of
cholesterol lowering in 4444 patients with coronary heart disease. Lancet
1994; 344: 1383-1389.
100
74. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose
control with sulphonylureas or insulin compared with conventional treatment
and risk of complications in patients with type 2 diabetes (UKPDS 33).
Lancet. 1998; 352: 837-853.
75. UK Prospective Diabetes Study (UKPDS) Group. Tight blood pressure
control and risk of macrovascular and microvascular complications in type 2
diabetes: UKPDS 38. BMJ. 1998;317(7160):703-13.
76. Vanderpoel DR, Hussein MA, Watson-Heidari T, Perry A. Adherence to a
fixed-dose combination of rosiglitazone maleate/metformin hydrochloride in
subjects with type 2 diabetes mellitus: a retrospective database analysis. Clin
Ther. 2004 Dec;26(12):2066-75.
77. Verges B. Diabetic dyslipidaemia: insights for optimizing patient
management. Curr Med Res Opin. 2005;21 Suppl 1:S29-40. Review.
78. Vermeire E, Wens J, Van Royen P, Biot Y, Hearnshaw H, Lindenmeyer A.
Interventions for improving adherence to treatment recommendations in
people with type 2 diabetes mellitus. Cochrane Database of Systematic
Reviews 2005, Issue 2. Art.No.: CD003638. DOI:
10.1002/14651858.CD003638.pub2.
79. Vijan S, Hayward RA. American College of Physicians. Pharmacologic lipid-
lowering therapy in type 2 diabetes mellitus: background paper for the
American College of Physicians. Ann Intern Med. 2004;140(8):650-8.
Review.
80. Watkins J, Cline M, 2005. California's Medical Assitance Program, Annual
Statistical Report, Calendar year 2003. California Department of Health
Services, Sacramento.
81. Wei L, Wang J, Thompson P, Wong S, Struthers AD, MacDonald TM.
Adherence to statin treatment and readmission of patients after myocardial
infarction: a six year follow up study. Heart. 2002 Sep;88(3):229-33.
82. Winer N, Sowers JR. Epidemiology of diabetes. J Clin Pharmacol.
2004;44(4):397-405.
83. Yilmaz MB, Biyikoglu SF, Guray Y, Karabal O, Caldir V, Cay S, Sahin O,
Sasmaz H, Korkmaz S. Level of awareness of on-treatment patients about
prescribed statins. Cardiovasc Drugs Ther. 2004 Sep;18(5):399-404.
101
84. Yood MU, McCarthy BD, Kempf J, Kucera GP, Wells K, Oliveria S, Stang
P. Racial differences in reaching target low-density lipoprotein goal among
individuals treated with prescription statin therapy. Am Heart J. 2006
Oct;152(4):777-84.
85. Zeger SL, Liang KY, Albert PS. Models for longitudinal data: a generalized
estimating equation approach. Biometrics. 1988;44(4):1049-60.
86. Zorn CJW. Generalized estimating equation models for correlated data: A
review with applications. American Journal of Political Science. 2001;45:
470-490.

Abstract (if available)

Abstract Objective: To examine long-term patient compliance to statin therapy in type 2 diabetes and its impact on various disease outcomes.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The effect of using fibrates in conjunction with statins for the management of dyslipidemia in persons with Type II diabetes mellitus

PDF

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

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

Burden of illness in hemophilia A: taking the patient’s perspective

PDF

Angiotensin (1-7): a novel treatment for diabetes-induced kidney and heart dysfunction

PDF

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

PDF

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

PDF

Documenting the impact of under-treatment for chronic disorders

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

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

PDF

The peptide angiotensin-(1-7) as a novel treatment for complications induced by type 2 diabetes mellitus

PDF

New approaches using probabilistic graphical models in health economics and outcomes research

PDF

Discriminating changes in health using patient-reported outcomes

PDF

Predictive performance of asthma quality of care measures – a comparison under skewed outcome and cost distributions

PDF

Application of concepts from prospect theory and response shift when estimating the minimally important difference for health utility measures

PDF

Dentists' and dental hygienists' knowledge, opinions and practice behaviors related to patients with uncontrolled or undiagnosed type 2 diabetes mellitus

PDF

Examining the relationship between common genetic variation, type 2 diabetes and prostate cancer risk in the multiethnic cohort

PDF

Characterization of health outcomes in patients with hemophilia A and B: Findings from psychometric and health economic analyses

PDF

Three essays on emerging issues in hemophilia care

PDF

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

Asset Metadata

Creator Zhang, Lihua (author)

Core Title Medication compliance to statin therapy and its impact on disease outcomes in type 2 diabetes

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Pharmaceutical Economics

Publication Date 04/24/2007

Defense Date 03/30/2007

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

Tag cardiovascular risk,costs,hospitalization,medication compliance,OAI-PMH Harvest,statin,type 2 diabetes

Place Name California (states), USA (countries)

Language English

Advisor Nichol, Michael B. (committee chair), Ahn, Jeonghoon (committee member), Buchanan, Thomas A. (committee member)

Creator Email lihuazha@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m458

Unique identifier UC1183187

Identifier etd-Zhang-20070424 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-484442 (legacy record id),usctheses-m458 (legacy record id)

Legacy Identifier etd-Zhang-20070424.pdf

Dmrecord 484442

Document Type Dissertation

Rights Zhang, Lihua

Type texts

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

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu