Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Physician adherence to national hypertension guidelines in an elderly Medicaid population
(USC Thesis Other)
Physician adherence to national hypertension guidelines in an elderly Medicaid population
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI films
the text directly from the original or copy submitted. Thus, some thesis and
dissertation copies are in typewriter face, while others may be from any type of
computer printer.
The quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality illustrations
and photographs, print bleedthrough, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send U M I a complete manuscript
and there are missing pages, these w ill be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and continuing
from left to right in equal sections with small overlaps.
ProQuest Information and Learning
300 North Zeeb Road, Ann Arbor, M l 48106-1346 USA
800-521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
PHYSICIAN ADHERENCE TO NATIONAL HYPERTENSION
GUIDELINES IN AN ELDERLY MEDICAID POPULATION
by
Tripthi Vasant Kamath
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
(PHARM ACEUTICAL ECONOMICS AND POLICY)
December 2001
Copyright 2001 Tripthi Vasant Kamath
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 3065803
Copyright 2001 by
Kamath, Tripthi Vasant
All rights reserved.
UMI
UMI Microform 3065803
Copyright 2002 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Tripthi V. Kamath
Michael B. Nichol
ABSTRACT
PHYSICIAN ADHERENCE TO NATIONAL HYPERTENSION GUIDELINES IN
AN ELDERLY M EDICAID POPULATION
Objective: This research focuses on im pact o f physician adherence to hypertension
guidelines for first line therapy, in elderly patients.
Methodology: Hypertensive patients greater than 65 years o f age were selected. The
JNC IV and the JNC V reports o f the national hypertension guidelines were
specifically studied since they had differing first line therapy recommendations. An
interrupted time series methodology analyzed the prescribing trend o f physicians
over different time periods. Survival analysis techniques analyzed the time to event
(switch in medications, hospitalizations and discontinuation in therapy) and
compared it between patients who got guideline therapy versus not. Finally, a sample
selection model was applied to study differences in health care costs between the two
groups o f patients.
Results: The prescribing behavior o f physicians seemed to be somewhat consistent
with guideline recommendations, although reasons other than guideline
recommendations could be responsible for physician prescribing patterns. The results
also showed that patients getting treated in accordance with guidelines had better
l
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
outcomes (e.g. shorter time to event) and lower health care costs compared to
patients not treated in accordance with guidelines. Specifically, patients treated in
accordance with the JNC IV recommendations had better outcomes and lower health
care costs compared to patients treated in accordance with the JNC V
recommendations.
Conclusion: Not all physicians follow guideline recommendations when prescribing
first line therapy to hypertensive patients. The above results show that there is a
beneficial impact o f adhering to guideline recommendations. It is imperative that in
order to optimally treat the elderly hypertensive patient in the US. health
administrators need to continually target physicians and intervene in ways that will
make physicians more aware o f the repercussions o f inappropriate prescribing in the
elderly hypertensive population.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UNIVERSITY OF SOUTHERN CALIFORNIA
The Graduate School
University Park
LOS ANGELES, CALIFORNIA 90089 1695
This dissertation, w ritten by
-Tripthi V.. Kaaath--------------------------
Under the direction o f h . * r . C . Dissertation
Committee, and approved by all its members,
has been presented to and accepted by The
Graduate School, in partial fulfillment o f
requirements for the degree o f
DOCTOR OF PHILOSOPHY
o f Graduate Studies
l,2-.L?-2nfl.l.
DISSER T A TION COMMITTEE
Chairperson
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
DEDICATION
To My Parents
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ACKNOWLEDGEMENTS
I would like to thank a number o f people who have helped me through my entire
academic career. Without their help and timely support, it would not have been
possible to achieve my academic dreams.
First, I would like to express my gratitude to Dr. Michael B. Nichol for his
invaluable advice and constant encouragement during my dissertation research, for
m aking me realize my potential, and the importance o f staying focused. I am also
greatly indebted to Dr. Kathleen A. Johnson for giving me the opportunity to work
on numerous projects that helped me develop my research and organizational skills. I
also want to thank her for her warmth and support that helped me through all the
difficult times in graduate school.
I would also like to thank Dr. Denise Globe and Dr. Elizabeth Graddy for their
suggestions towards my dissertation. Dr. Jeffrey M cCombs played an important role
by brightening up even the dullest day with his witty humor and I want to thank him
for that.
I will forever cherish the people that came into my life while I was at USC. Joseph
Parker, Maria Barreto, Caroline Bongrand and M ary Gambrell. They
iii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
showed me how to enjoy life to its fullest and to appreciate the wonderful moments
that life has to offer. I thank them for sharing my good times and bad times and I will
always hold them dear to my heart. I would like to acknowledge my friend and
colleague, Usa Chaikledkaew, who constantly stood by me through thick and thin,
each day o f the week, each week o f the year and continues to do so.
I would like to mention a special thank you to my friend Bijal Vyas. whom I have
known since second grade, for always making me see the lighter side o f things and
encouraging me to do the right thing. I especially want to express my gratitude to my
friend Mala Pattanayek, for believing in me, for encouraging me to dream, and for
always standing by me. Special thanks also go to Abhijit Shah, for putting up with
my unpredictable moods in the past year. I want to thank all my other friends who
have helped me in numerous ways and I will always value their friendship.
Most o f ail, I would like to thank my sister, Dr. Amrita V. Kamath, and my parents.
Dr. Vasant K. Kamath and Shanthala V. Kamath, for their love, patience and steady
encouragement through these past years.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF CONTENTS
Dedication ii
Acknowledgements iii
List o f Tables vii
List of Figures ix
Chapter 1: Introduction 1
Background 1
Significance o f this study 6
Preliminary Studies/Related Research 6
Limitations o f previously published literature 11
Specific Aims o f this Research 15
Chapter 2: Research Design and Methods 16
Data Source 16
Sampling Methods 17
Research Question 1 19
The Segmented Regression Model 27
Research Question 2 30
Kaplan M eier Method for Time to Event Analysis 32
Cox Proportional Hazards Model 37
Research Question 3 40
Controlling For Selection Bias 42
Chapter 3: Results 52
Specific Aim 1 56
Specific Aim 2 65
Time to Switch in Medication Class 66
Time to First Hospitalization 70
Time to Discontinuation o f Antihypertensive therapy 74
Time to Augmentation in Antihypertensive therapy 77
Specific Aim 3 78
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4: Discussion 83
Limitations o f the Research 89
Policy Implications o f this Research 91
Future Research 92
References 95
Appendix 1 97
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Tables
1 JNC guidelines for first line antihypertensive therapy 4
2.1 Antihypertensive M edication Class Classification 21
2.2 Number o f patients with 1,2 and 3 year health care costs
by JNC period 42
2.3 Independent variables for first stage Probit model 47
2.4 Independent variables in the outcome equation 49
3.1 General Characteristics o f the Sample (N =2162) 53
3.2 Characteristics o f sample by JNC 4 and JNC 5 periods 55
3.3 Segmented Regression Output 63
3.4 Cox regression Model output
Dependent variable: Time to switch in medication class 69
3.5 Cox regression model output
Dependent variable: Time to first hospitalization 73
3.6 Cox regression model output
Dependent variable: Time to discontinuation o f
antihypertensive therapy 76
vii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Tables (continued)
3.7 Factors associated with 1,2 and 3 year health care costs -
Comparing patients that received first line treatment in
accordance with guidelines
3.8 Adjusted 1,2 and 3 year Health Care Costs
3.9 Adjusted 1.2 and 3 year Health Care Costs
(JNC 4 versus JNC 5)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Figures
3.1 Overall Utilization o f Initial Antihypertensive therapy in the JNC 4
and the JNC 5 periods 56
3.2 Proportion o f new starts o f diuretics per month 58
3.3 Proportion o f new starts o f P-blockers per month 58
3.4 Proportion o f new starts o f ACE Inhibitors per month 59
3.5 Proportion o f new starts o f CCBs per month 59
3.6 Proportion o f diuretics- 3 month moving average 61
3.7 Proportion o f P-blockers - 3 month moving average 61
3.8 Proportion o f ACE Inhibitors - 3 month moving average 62
3.9 Proportion o f CCBs - 3 month moving average 62
3.10 Plot o f Survival Function - Time to switch in medication class
for patients receiving treatment in accordance with guidelines
versus not 67
3.11 Plot o f Survival Function - Time to switch in medication class
Comparing patients treated in accordance with JNC 4 and JNC 5 67
ix
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Figures (continued)
3.12 Plot o f Survival Function - Time to first hospitalization
for patients receiving treatment in accordance with guidelines
versus not
3.13 Plot o f Survival F unctio n -T im e to first hospitalization
comparing patients treated in accordance with JNC 4 and JNC 5
3.14 Plot o f Survival Function - Time to discontinuation in therapy
comparing patients treated in accordance with JNC 4 and JNC 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 1: INTRODUCTION
Background:
Hypertension, the disease known as the “silent killer”, is an asymptomatic disease
and affects about 50 million Americans (Joint National Committee on Detection.
Evaluation and Treatment o f High Blood Pressure, JNC V, 1993). This condition
costs more than $10 billion in direct medical expenditures (Stason. 1991). O ver the
past few decades, awareness o f the complications o f this disease has been growing
and numerous attempts have been made to control and treat hypertension. However,
o f the 50 million Americans suffering from hypertension today, only one in four
Americans, have their blood pressure under control (JNC VI, 1997). This is very
disturbing since hypertension is a risk factor for cerebrovascular diseases and
cardiovascular diseases such as stroke, coronary heart disease, myocardial infarction
and congestive heart failure. Although the mortality rates from stroke and CHD have
slightly declined in the past two decades, the incidence o f end stage renal disease and
prevalence o f heart failure have been on the rise. With appropriate treatment
however, the risk o f cardiovascular morbidity and mortality resulting from
hypertension, can be controlled (Mehta et al., 1999).
In the early years o f anti hypertensive drug therapy (1940s and 1950s). drugs such as
reserpine, hydralazine and ganglionic blocking agents were used to treat
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hypertension, although they caused considerable side effects (Moser M, 1997). There
has been a dramatic change since the mid nineteen hundreds, in the choice o f
antihypertensive agents available for treatment. At least 6 classes o f antihypertensive
agents for treatment o f hypertension have been identified namely diuretics, p-
blockers, ACE inhibitors, CA antagonists, a blockers, central a 2 agonists and
peripheral neuronal blockers. (Frohlich ED. 1998).
Physician's role in management o f hypertension
With the vast number of pharmacological agents that are now available to control
hypertension, the physicians are the ones who have to make the choice o f prescribing
the optimal antihypertensive drug to the patient. When faced with a decision on the
best anihypertensive agent to prescribe, the physician must be aware o f the
coexisting diseases that the patient may have. It is essential that the physicians keep
themselves up to date with results from recent clinical trials in the treatment o f
hypertension and the indications and contraindications for the use o f certain
antihypertensives in some coexisting clinical conditions. The role o f physicians in
the treatment o f hypertension remains predominant today since in the long term
about 80-85% o f all hypertensive patients can be treated successfully with the
available therapy. (Moser et al., 1980). In all the major guidelines that have been
published so far regarding the treatment o f hypertension, there is unanimous
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
agreement on the fact that the physician should assess each patient carefully and
consider the cardiovascular risk profile o f the patient.
Guideline Recommendations o f the Joint National Committee on the Detection.
Evaluation and Treatment o f Hypertension (JNC):
Several anti-hypertensives can be used to treat other illnesses in addition to treating
hypertension. Conversely, some antihypertensive agents may be relatively or
absolutely contraindicated in the treatment o f hypertension when there are specific
co-existing diseases, and alternative agents should be selected in such cases. The
initial drug o f choice in the treatment o f hypertension is o f considerable importance
since each of the drug classes has their side effects and the patient could be suffering
from comorbidities. Since 1977, the Joint National Committee on the Detection.
Evaluation and Treatment o f Hypertension (JNC) has issued six reports on the
detection evaluation and treatment o f hypertension. Each report updates the
information based on the latest scientific research and lays down recommended
guidelines for the treatment and management o f hypertension. Through the years the
guidelines have varied considerably in their choice o f initial antihypertensive
therapy. The 1984 JNC report (JNC III) recommended that either P-blockers or
thiazide type diuretics could be used as first line therapy, unless contraindicated.
Four years later the fourth report o f the Joint National Committee (JNC IV)
recommended that first line antihypertensive therapy could include diuretics.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
P-blockers, ACE inhibitors or calcium channel blockers (CCBs), since there was
clinical evidence that ACE inhibitors and CCBs were effective as antihypertensive
agents. Then, in 1993, the fifth report o f the JNC reinstated the use o f only diuretics
and P-blockers as first line therapy, unless otherwise contraindicated or when there
was no specific indication for the use o f another drug. This was due to the fact that
P-blockers and diuretics were the only drugs that had clinically shown to reduce
morbidity and mortality in long term clinical trials. Table 1 highlights the major
additions/changes in the JNC guidelines over the past years.
Table 1: JNC guidelines for first line antihypertensive therapy
JNC Year
published
Diuretics P-blockers Calcium
Channel
Blockers
ACE
inhibitors
JNC III 1984 Yes Yes No No
JNC IV 1988 Yes Yes Yes Yes
JNC V 1993 Yes Yes No No
JNC VI 1996 Yes Yes No No
Contraindicated Drugs in the JNC Guidelines:
Many o f the elderly patients who suffer from hypertension, also suffer from other
diseases such as diabetes, asthma and other cardiovascular diseases. The JNC
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
guidelines provide a list o f antihypertensive drugs that are recommended and drugs
that are contraindicated for use with certain disease conditions.
According to the JNC publications, P-blockers worsen asthma and peripheral arterial
disease and hence are contraindicated in patients with asthma, COPD, congestive
heart failure and heart block. Calcium antagonists are contraindicated in patients
with second or third degree heart block and ACE inhibitors are contraindicated for
bradycardia. Two studies in 1995 showed that the treatment o f hypertensive patients
with CCBs might increase the risk o f myocardial infarction (Pahor et al., 1995; Psatv
et al., 1995). It has been shown that large doses o f nifedipine given immediately after
an acute MI increases mortality (Furberg et al., 1995). Kaplan et al. 1996 list the
drugs that are indicated and contraindicated for the treatment o f hypertension based
on coexisting conditions. Diuretics are preferred in congestive failure and renal
insufficiency and are contraindicated in dyslipidemia. CCBs are preferred agents in
angina, cerebrovascular diseases and renal insufficiency, but, they are
contraindicated in congestive failures. ACE inhibitors are preferred for post-
myocardial infarction, congestive failure, renal insufficiency and diabetes.
While the Joint National Committee on Detection, Evaluation and Treatment of
Hypertension does not absolutely mandate the application o f these guidelines in
practice, it does recommend these guidelines for the optimal treatment and control of
hypertension in the interest o f individual and public health (Lenfant, 1997).
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Significance o f this study
The impact o f these guidelines on various aspects o f health care such optimal blood
pressure control, quality o f life o f the patients, cardiovascular co-morbidities and
health care costs and utilization is o f considerable interest to the guidelines
committee. Different classes o f antihypertensive drugs drastically differ in their costs
and hence prescribing a more expensive drug class can have a tremendous impact on
US health care expenditures. By analyzing the physician adherence to guideline
recommendations, specifically the two JNC guidelines with differing treatment
protocols involving expensive antihypertensive drugs, we can study the economic
impact o f physician prescribing behaviors. This will give an idea about the impact o f
guideline recommendations since the aim o f guideline developm ent is to minimize
the cost o f managing hypertension, as well as to achieve optimal control
(Johannesson, 1992). The results o f this study will aid health plan administrators to
identify the physician prescribing behaviors and subsequently develop effective
interventions to influence care in a manner that is appropriate and consistent with
national guidelines for the management of hypertension.
Preliminary Studies/ Related Research (See Appendix 1 for summary)
There has been some research performed in the past that has studied the adherence to
the JNC guidelines. Four studies looked at the effect o f JNC guidelines on physician
6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
prescribing patterns (Alderman et al., 1996; Siegel et al., 1997; M ehta et al, 1999.
Kozma et al., 1999) and one article studied the prevalence o f hypertension in diabetic
patients before and after the JNC 5 ( Tam ow et al., 1994).
The general consensus from the above studies was that physicians were not
prescribing antihypertenisve medications in accordance with guideline
recommendations. All previously published literature in this field studied adult
patients who were suffering from hypertension. Two o f the studies that attempted to
study the adherence o f physicians to guideline recommendations (Alderman et al..
Siegel et al.) concluded, from the drug utilization profiles o f the patients in the
period following the publication o f the JNC guideline recommendations, that the
change in the guideline recommendations from JNC 4 to the JNC 5
recommendations, did little to change the prescribing patterns o f the physicians.
These studies looked at the overall utilization o f the various classes o f
antihypertenisve medications. They found that in the period following the JNC 5
guideline publications the utilization o f the antihypertensive prescriptions did not
reflect the recommendations o f the JNC 5 guidelines which recommended the use o f
only P-blockers and diuretics as first line antihypertenisve therapy. This conclusion
was made based on the utilization o f m edications by patients suffering from
hypertension, in the periods following the publications o f the JNC 4 and the JNC 5
guidelines. The possibility that the physicians could have altered their prescribing
7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
practices over a period o f time after the publication o f the guidelines was not
accounted for, since only the aggregate utilization in the entire period was studied.
The only study that looked at the physicians’ perspective on practice patterns in
relation to the guideline recommendations was the study by Mehta et al where they
surveyed physicians regarding their practice patterns for hypertensive patients, and
w hether the physicians considered looking into the JNC guideline recommendations.
The authors concluded that based on the survey responses, most physicians do not
consider the guideline recommendations while prescribing antihypertensive
medictaions, mainly due to lack o f awareness o f the guideline recommendations and
also due to the physician’s individual clincial decision making skills. However, this
does not reflect the real prescribing practices since there was no prescription
utilization data to support the physician responses and also, the survey asked
physician about their prescribing practices eighteen months into the past, hence the
accuracy o f these results can be questioned.
Up until now, previous research that looked at the utilization o f antihypertensive
medications and studied physician prescribing patterns in relation to the JNC
guidelines have been reviewed.
Few researchers studied the im pact o f the guideline recommendations on the patient
health outcomes. A couple o f articles studied the impact o f following guideline
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
recommendations on short-term patient outcomes ranging from only blood pressure
levels to clinic attendance and drop-out rates (Alderman et al., Tam ow et al.). The
results were mixed, where one study showed that the blood pressure levels remained
the sam e in the periods before and after the JNC 5 guideline publication (Alderman
et al.), while the other study showed that blood pressure levels increased in the
period following the publication o f the JNC 5 guidelines (Tamow et al.). In the latter
article however, when making conclusions about the increased blood pressure levels
in patients in the period following the JNC 5 guideline publications, there was no
mention o f whether they had actually checked blood pressure levels o f patients that
had in fact been treated in accordance with the guideline recommendations. Hence,
there is no evidence to conclude that the increase in blood pressure levels in the
period following the publication o f the JNC 5 guidelines, was in fact, due to the the
physicians prescribing antihypertensive medication in accordance with the JNC 5
guideline recommendations.
From an economic point o f view, a few articles studied the impact o f the JNC
guideline recommendations on the health care costs. The results from the studies by
Alderman et al and Siegel et al were consistent in that they found that the
prescription expenditures for patients taking ACE Inhibitors and CCBs increased in
the period following the JNC 5 guideline publication. However, there was no
com parison made to similar costs in the period prior to the JNC 5 publication, which
would allow one to compare costs and come to any conclusion regarding the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
significant change in costs between periods, if any, that was due to the
recommendations o f the JNC 5 guidelines. Also, only prescription costs were taken
into account. Health care costs related to events such as hospitalizations subsequent
to antihypertensive were not accounted for, hence the economic impact of these
guidelines on the total health care costs was not explored. In relation to the economic
impact o f the JNC guidelines, Ramsey et al studied the clincial cost o f managing
hypertension when following the JNC guidelines. This study focussed on the JNC 6
guidelines, whose guideline recommendations for first line antihypertensive therapy
is same as the recommendtaions o f the JNC 5 guidelines. The authors found that
when following the JNC guidelines, from an economic and clincial standpoint,
diuretcis and b-blockers were the preferred antihypertensive drugs.
Although some work has been done in the past with regard to the JNC guidelines,
none o f the articles really studied the change in the physician prescribing patterns for
initial antihypertensive therapy, from the period following the JNC 4 guidelines to
the period following the JNC 5 guidelines. This is especially important since the
recommendations for first line antihypertensive therapy for newly diagnosed
hypertensive patients are different in JNC 4 and the JNC 5. Along with the above
gap in the literature, there are some more limitations that are apparent in the past
studies and which are delineated below.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Limitations o f previously published literature
Although the articles summarized above, presented certain important results, there
are some limitations that need to be addressed. They are listed below:
1. None o f the articles specifically studied the elderly antihypertensive pateints
aged 65 years and more, which constitutes a major proportion o f the hypertensive
population. This is an especially important population since the elderly patients
present challenges to the physicians in terms o f prescribing the appropriate
antihypertensive medications, because they are more likely to suffer from
concom itant disease conditions.
2. None o f the articles mentioned whether there had been sufficient amount o f time
allowed after the publication o f the guidelines, for the dissemination o f the
guideline information to the physicians. It could have been possible that
physicians became aware o f the guidelines immediately after their
publications and made changes in their prescribing patterns immediately
following the publication o f the guideline recommendations, or it is also possible
that physicians took a while, maybe a few weeks or months, to know that newer
guidelines were available for the treatment of hypertension, and hence did not
change their prescribing pattern for a long time after guideline publication.
W hatever the case, the methods used in the previously published literature to
study the physician adherence to the JNC guidelines are very simplistic and do
ll
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
not allow the reseracher to capture the changing nature o f the physician
prescribing behavior over time.
3. There is limited literature on the physician prescribing practices for hypertensive
patients with comorbidities. Both, the JNC 4 and the JNC 5 guidelines mention
certain drugs that should be avoided in antihypertensive patients that have
concom itant diseases. The elderly patients over 65 years o f age constitute a large
percent o f the hypertensive population and these patients usually suffer from one
or more disease conditions. There has been no study so far that has looked at
physician adherence to first line therapy guidelines in the JNC reports, in the
elderly hypertensive population with co-morbidities.
4. None o f the above authors studied the long term impact on patient health
outcomes and economic outcomes, as a result o f receiving treatment in
accordance with the JNC guidelines. Patients starting on one type o f
antihypertensive medication class, may be switched to another medication class
if not responding well to the first class, or there could be an augmentation in
anti hypertensive therapy. Such outcomes have never been studied before in
relation to the appropriateness o f initial antihypertensive therapy.
Hospitalizations associated with hypertension related diagnosis, after the
initiation o f antihypertensive therapy has not been studied in the past. Since the
JNC 4 guidelines and the JNC 5 guidelines differ in thier choice o f first line
antihypertensive therapy, it is assumed that patients receiving these differing
treatments will eventually have different outcomes in the long run. However, no
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
study so far has pursued this aspect o f the analysis, where any differences in the
health outcomes between patients receiving antihypertensive treatment in
accordance with the JNC 4 guidelines versus the JNC 5 guidelines, and also
differences between patients receiving antihypertensive treatment in accordance
with any guidelines in general (JNC 4 or the JNC 5) versus not receiving
appropriate treatment, have been addressed.
How will this study address the above limitations?
This research addresses the gaps in the literature and proposes to conduct a
comprehensive analysis on JNC guideline adherence. This analysis will focus only
on the elderly patients above 65 years o f age. This will be the first study that will
analyse and compares physician guideline adherence in elderly patients, under the
JNC 4 and the JNC 5 guideline recommendations. This research will aim to study the
pattern o f physician prescribing patterns for the elderly hypertensive patients, under
the two differing JNC guideline recommendations, study the long term patient
outcomes and economic outcomes, while simultaneously providing some insight into
the significant differences in these impacts between patients being treated in
accordance with the JNC 4 and the JNC 5 guidelines, and also track signifciant
differences in these impacts between pateints that receive first line antihypertensive
therapy in accordance with guidelines in general (JNC 4 and the JNC 5) versus not.
Some or most o f these patients are bound to have concomitant disease conditions,
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hence treatment o f such patients and long term patient outcomes will be studied and
compared under the two different JNC guideline recommendations. This study will
also account for the time lag that is required for dissemination o f the guideline
information to the physicians via continuing education programs, seminars or journal
articles. Cardiovascular co-morbidities develop over a period o f time, hence this
research will take into account the incidence o f these comorbidities while analyzing
patient health care outcomes. The long term patient health care outcomes will be
compared between the two different JNC guidelines to reach some conclusion as to
which o f the two treatment guidelines have a more beneficial treatment impact.
Different classes o f antihypertensive drugs drastically differ in their costs and hence
prescribing a more expensive drug class can have a tremendous impact on US health
care expenditures. By analysing the physician adherence to the JNC guidelines with
differing treatment protocols involving expensive antihypertensive drugs, we can
study the economic impact o f adhering to these guidelines. This proposed research
will give an idea as to which o f the two treatment protocols (JNV 4 and the JNC 5)
was more beneficial to the patient outcomes and also economic outcomes, since the
aim o f guideline development should be to minimize the cost o f managing
hypertension, as well as to achieve optimal control (Johannesson. 1992).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Specific Aims of this Research:
The main objective o f this study is to examine the adherence o f physicians to the
national guidelines for hypertension, in an elderly hypertensive population and study
the effects o f being adherent versus non-adherent, on patient outcomes. Two specific
reports o f the Joint National Com m ittee on the Detection. Evaluation and Treatment
o f hypertension (JNC) will be targeted, the JNC IV and the JNC V. since they have
differing recommendations for first line antihypertensive therapy. Specifically this
study aims to answer the following questions:
1. Do the physicians adhere to the JNC guidelines when prescribing
antihypertensive medications to the elderly?
2. What is the impact o f these individual guidelines on patient outcomes such as
hospitalizations, switches in medication class, augmentation in therapy and
discontinuation of therapy?
3. What is the economic im pact o f following guidelines versus not following
guidelines, in an elderly M edicaid population?
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 2: RESEARCH DESIGN AND METHODS
Data Source:
The California Medicaid program (MediCal) dataset was used for this retrospective
analysis. MediCal finances a wide range o f health care services for the poor and the
disabled population. The MediCal research database is a random 5% sample o f all
Medicaid beneficiaries. This dataset contains patient level demographic data, type o f
service, date o f service, amount billed, amount paid, and units (days) o f service.
Prescription drug claims identify the specific product dispensed, quantity, strength,
and the date the prescription was filled.
Advantages o f the Medical dataset
The MediCal dataset is beneficial for this type o f analysis since details o f each claim
for each patient is available along with the date o f claim, hence this provides ease o f
following the health care resource utilization o f patients over time. Detailed
information on prescriptions filled by the patients allows for the analysis o f pattern
o f drug utilization. The cost information in each o f the paid claims allows us to
calculate totals costs and also costs for individual health care resources used. This
database provides the opportunity for longitudinal analysis since once an individual
is selected for inclusion, their utilization data are collected for as long as they remain
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
eligible. It is also important to note that elderly beneficiaries rarely lose eligibility.
The MediCal datatset provides us with vast amount o f information that is ideal for
this type o f retrospective database analysis and hence this dataset has been utilized to
answer the research questions in this study. The limitations o f the MediCal dataset
are briefly discussed in Chapter 4.
Selection o f Patients:
In this study, patients are selected based on their prescription profile. Prescription
claims from 1986 to 1996 are investigated. This encompasses the time periods during
which the JNC IV (May 1988) and the JNV V (Jan 1993) were published.
Individuals are selected into the study if they:
1. Receive at least one prescription o f an antihypertensive medication.
2. Maintain Medical eligibility for at least one year (eligibility criteria explained
below)
3. Start a new episode o f antihypertensive therapy (confirmed by the absence o f
any prescription for antihypertensive medications 6 months prior to the first
available antihypertensive medication) in the time period specified above.
Sampling Methods:
The dataset is in an episode level format. An episode is defined as a treatment period,
with antihypertensive medications, starting from the date o f the first
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
anti-hypertensive medication going 3 years forward, plus a 6 m onth washout period
prior to the date o f the first antihypertensive medication. Hence, the length o f an
episode will be 42 months, and a patient can appear more than once in the dataset
depending on the num ber o f treatment episodes that the patient experiences during
the 10 year time frame (1986-1996).
The MediCal paid claims data was used to identify patients with at least one
antihypertensive prescription in the time frame selected above. The date o f the first
antihypertensive prescription is considered to be the start date o f antihypertensive
therapy for the patient and henceforth is called the 'Index D ate’. To circumvent the
possibility that the patient is not a newly diagnosed hypertensive, we select those
patients that do not have an antihypertensive prescription at least 6 months prior to
the keydate (the pre-period).
Medicaid patients may temporarily lose eligibility over a period o f time. To ensure
that the patients are continuously enrolled into the program, in order to reduce
potential bias, patients are removed from the data if they experience three
consecutive months in which no paid claims are processed, in the period following
the keydate. Patients older than 100 years are eliminated due to possible reporting
errors in date o f birth. The follow up period for all patients was the same regardless
o f when they entered the study period.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The MediCal dataset that we started out with had a potential study population o f
97,274 patient episodes that corresponded to 55,829 patients. Applying the above
selection criteria, the final sample included 22.145 patients. Patients who had a
diagnosis o f hypertension on or before the key date were identified using the
International Classification o f Diseases. 9th revision (ICD-9) codes for hypertension
on their claims. The ICD-9 codes used for identifying patients with hypertension
were ICD-9 codes 401.401.0.401.1, 401.9,405, 405.0.405.01, 405.11,405.19.
405.9,405.91 and 405.99.
There were 2.162 patients with a diagnosis o f hypertension on or before the keydate.
For the following analysis, the sample o f 2,162 patients will be used, to affirm that
the patients are receiving antihypertensive treatment for the treatment of
hypertension and not for other disease conditions. The study procedures for the three
main questions for this research are explained in detail in the section below.
Research Questions
Research Question 1: Do physicians follow the guideline recommendations o f the
Joint National Committee for the Detection, Evaluation and Treatment o f
Hypertension (JNC) for initial antihypertensive therapy, while prescribing to an
elderly M edicaid population?
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Specific study procedure
In order to answer this research question, the trend o f antihypertensive drug
utilization was studied using an interrupted time-series analysis procedure, to see if
the physician prescribing trend was consistent with the JNC guidelines. The JNC
guidelines specifically differ in their initial therapy recommendations i.e. the first
drug (or drug class) that should be given to newly diagnosed hypertensives.
Specifically, the four drug classes that are recommended for initial therapy by the
various reports o f the JNC guidelines include diuretics. P-blockers, ACE inhibitors
and calcium channel blockers (CCBs). To study the pattern o f utilization o f these
four classes o f drugs over time, the proportion o f new starts per month, o f each o f the
above drug classes was plotted over time. For each month, the total number o f
prescriptions at keydate was identified. Using this number as a 100% sample for that
particular month, the proportion o f each o f the above four antihypertensive
medications for that month was calculated. A monthly approach is employed to
increase the number o f data points and make the interpretation o f the results more
robust. A monthly analysis also allows for the modeling o f the time lag, which is
imperative for this analysis, since some time needs to be allocated from the
publication o f the guidelines for the dissemination of the information to the
physicians. By calculating proportion o f starts o f each medication class per month,
we can track the changes in the proportions o f the different antihypertensive drug
classes being prescribed, before and after the publication o f the guidelines and also
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
track the time required for the guidelines to take effect, if at all the guidelines have
an impact. Table 2.1 highlights the different drugs that were grouped under a
medication class, hence if a patient received any one o f the drugs as their first
prescription, then they were classified as receiving the particular medication class
that the drug corresponded to.
Table 2.1: Antihypertensive Medication Class Classification
Medication Class Drug
Diuretics Chlorthalidone. Hydrochlorthiazide,
Indapamide. Ethacrynic Acid. Furomeside,
Amiloride HCL, Spironolactone
P-blockers Acebutolol, Atenolol. Betaxolol, Bisoprolol.
Metoprolol, Nadolol, Levatol. Pindolol,
Propranolol. Timolol Maleate
ACE Inhibitors Benazepril, Captopril, Enalapril. Fosinopril,
Lisinopril, Quinapril, Ramipril
Calcium Channel
Blockers (CCBs)
Diltiazem, Verapamil, Vaseretic, Zestoretic,
Dyazide, Aldactazide, Esimil, Apresazide,
Aldoril, Hydropres, Ser-AP-ES Tablet,
Combipres, Hydropres, Demi-Regroton,
Diupres, Minizide, Lopressor HCT, Hydralazine
Complex Tablet
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The dataset included prescription claims from June 1986 to November 1996. The
JNC 4 guidelines were published in May 1988 and the JNC 5 guidelines were
published in January 1993. The 10-year tim e frame was divided into three study
periods as follows:
Period 1 Period 2 Period 3
June
1986
May
1988
Jan
1993
Dec
1996
1. Period 1: Pre JNC 4 period
All patients who received their first antihypertensive medication on or before May
1988 (date o f JNC 4 publication) belonged to period 1. also known as the ore-JNC 4
period.
2. Period 2: JNC 4 period
All patients who received their first antihypertensive medication between June 1988
and January 1993 (date o f publication o f the JNC 5) belonged to period 2, also
known as the JNC 4 period.
3. Period 3: JNC 5 period
All patients who received their first antihypertensive medication after January 1993
belonged to period 3 or the JNC 5 period.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For this specific research question, we were interested only in the very first
antihypertensive prescription that the patient received, hence only the first episode o f
each patient was taken in consideration. The patients in one period were mutually
exclusive from the patients in other periods. An interrupted time series analysis was
applied to this data structure to study the trend in physician prescribing pattern over
time.
Visual Plot o f prescribing patterns
Constructing the variables for monthly proportion of new starts o f antihypertensive
medication classes and the three study periods as described above, a visual graph
was plotted over time. If considerable variation was found in the data points from
month to month, moving average graphs were plotted to smooth the curve and the
data was visually inspected for change in prescribing trends. Three-month moving
average and 6-month moving average graphs were plotted. A single data point for
each month in a 3-month moving average plot is an average o f the values o f that
current month and the two immediately prior months. Hence in the 3-month moving
average plot, the first two months will not have data points since neither o f them
have 2 months o f prior data points, hence the plot starts in the 3 rd month from the
start o f the study period. Similarly in a 6-month moving average plot, each data point
in a month is an average o f that current month and 5 months immediately prior to
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
that month. The moving average graphs allow for smoothing o f the curve when there
is some amount o f variation in the datapoints form month to month.
Interrupted time series analysis
The interrupted tim e series design belong to the general class o f time series models
that characterizes a pattern or a trend o f the dependent variable over a period o f time
(Campbell and Stanley. 1996). The interrupted time series analysis is also known as
the "segmented’ or the "piecewise' regression analysis. This type o f design allows for
evaluation o f any program impact that can occur immediately after the program
comes into effect or gradually over a period o f time. The interruption in the
"interrupted time series analysis’ is the month o f publication o f the JNC 4 and the
JNC 5 guidelines. A visual plot o f the monthly proportion o f each o f the four classes
o f diuretics was plotted over time, in Microsoft Excel 97. The timeline on the X axis
went from June 1986 (start date o f dataset) till November 1996 (end date o f dataset),
while the Y axis was the proportion o f monthly starts. The 3 periods o f study as
defined earlier were encompassed in this timeline. Each period o f study is called a
piece, or a segment, hence the name ‘segmented’ or "piecewise* regression. The
months that divide the timeframe into the individual periods or segments are called
join points as shown below.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Join Point 1 Join Point 2
Period 1 Period 2 Period 3
June May Jan Dec
1986 1988 1993 1996
Four separate graphs were plotted, one for each o f the four antihypertensive
medication drug classes. To be able to detect trends within and between different
segments, trend lines were fitted for each graph. These plots are then analyzed for
statistical interpretation using an interrupted time series or segmented regression
analysis. Segmented regression is similar to the ordinary least squares regression
(OLS), but unlike the OLS procedure, the segmented regression model allows for
detecting changes in slopes between periods. The assumptions for using the
segmented regression model are similar to the assumptions for OLS regression
models i.e. variance homogeneity and independent error terms.
Variance Homogeneity
This is a critical assumption for using the segmented regression model. Variance
homogeneity means that the error variance is constant in the model. When the
monthly proportion o f each antihypertensive drug class in plotted over time, there is
bound to be some fluctuations in the observations which can be attributed to chance.
These fluctuations correspond to the error variances and it is required that these
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
variances be constant every month. When the assumption o f variance homogeneity is
violated the phenomenon called heteroscedasticity exists. When the problem o f
heteroscedasticity exists in the data, the parameter estimates from the model are not
efficient and also the statistical tests used to determine the significance o f the
parameter estimates are not valid (Intrilligator, 1978). Hence if heteroscedasticity
exists, it must be appropriately dealt with prior to fitting the regression model to the
data. One way o f dealing with heteroscedasticity is by performing appropriate
transformations to the data such as a log transformation. In this analysis, the
regression residuals were plotted against the predicted values o f the regression
models to detect the presence o f heteroscedasticity (fan shaped plot implies
heteroscedasticity exists). In the event that the heteroscedasticity exists, appropriate
transformations would be applied to the data.
Independent Error terms
This is another critical assumption for using the segmented regression method. In
order to use the segmented regression, the random errors associated with the monthly
proportion o f new starts o f each drug class should be independent. To check for
independent error terms in this segmented regression model, the Durbin-W atson test
was performed to check for autocorrelation o f error terms. If the null hypothesis that
the autocorrelation parameter equals zero is rejected this implies that the error terms
are not independent, and in this case, the underlying assumption is violated and
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hence the segmented regression model cannot be used. The DW test was performed
and the autocorrelation parameter was tested for significant deviations from zero. If
the Durbin-Watson test statistic is approximately equal to 2, it is likely that the
assumption of independent error term s is not violated.
The Segmented Regression Model
The segmented regression model for the monthly proportion o f new starts o f each
antihypertensive medication class was specified as follows:
E (y) = p0 + (3|X, + p2 (xr 23) x2 + p3 (x,-79) x3
(Equation 1)
Where
y = dependent variable
= monthly proportion o f the antihypertensive medication class
xi = month of first prescription - June 1986
1 if month o f first prescription > May 1988
0 if month <= May 1988
1 if month o f first prescription > Jan 1993
0 if month <= Jan 1993
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Po = intercept for period 1
Pi = slope for period 2 or trend in monthly proportion from June 1986 to May 1988
P: = difference in slopes between period 1 (June 1986 to May 1988) and period 2
(June 1986 to Jan 1993)
P3 = difference in slopes between period 2 and period 3 (Jan 1993 to Nov 1996)
In the regression equation, the composite variables (xi-23).\2 and (.\|- 7 9 )x3 are set up
to reflect the number of months between the start date o f the dataset and join point 1
(i.e. 23 months) and join point 2 (i.e. 79 months) respectively.
The model is fitted like a normal multiple regression model and the coefficients for
the independent terms are estimated using PROC GLM in SAS. The parameters o f
interest are specifically P2 and P3. If period 2 has the same trend as period 1 then
P2=0 . If period 3 has the same trend as period 2 then P3=0 .
Expectations o f the Model
Since the JNC 4 recommended the use o f any o f the four antihypertensive
medication classes for initial therapy and the JNC 5 restricted the first line therapy
recommendations to P-blockers and diuretics, it was expected that there will be a
decrease in the prescribing o f p-blockers and diuretics from period 1 (Pre-JNC 4) to
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
period 2 (JNC 4) and an increase in the prescribing o f p-blockers and diuretics from
period 2 (JNC 4 period) to period 3 (JNC 5 period) in proportion to the prescribing
o f ACE inhibitors and CCBs. Simultaneously an increase in the prescription o f ACE
and CCBs from period 1 to period 2 and a decrease in the prescription o f ACE and
CCBs from period 2 to period 3 was expected.
Sensitivity Analysis
Sensitivity analyses are a must for drawing plausible conclusions from the results. In
this analysis too, a sensitivity analysis was performed by altering the join points i.e.
the segments that divide the time frame into the three study periods. When guidelines
are published, there is usually some time, also refered to as the technology
innovation lag, that is needed for the dissemination o f the guideline information to
physicians. Unless the physician is always up to date with the current literature, there
is usually a time lag that follows guideline publications, before the physicians are
aware o f what information is available in the health care market. To allow for this
time lag in modeling this analysis, the join points that divided the study periods were
moved ahead by 3 months and 6 months, to see if the prescribing patterns
significantly changed from one period to the next.
The advantage o f using an interrupted or segmented time series approach is that it
allows us to infer if the guideline publication had a gradual impact on the proportion
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o f antihypertensive drugs being prescribed. Another advantage o f the time series
analysis is that it allows us to evaluate the maturational trend o f prescribing prior to
the publication o f the guideline.
Research Question 2: What is the impact o f these guidelines on patient outcomes?
Specific Study Procedure
G iven the MediCal dataset that is being used, the following measures are defined for
determ ining the 'im pact' of guideline recommendations on patient outcomes:
1. Tim e to switch in medication class starting from the index date (i.e. once the
patient has started antihypertensive therapy).
2. Tim e to augmentation in antihypertensive therapy from the index date.
3. Tim e to the first hospitalization from the index date.
4. Tim e to discontinuation of therapy starting from the index date.
The above four measure are chosen since they com e as close as possible to
describing the patient health outcomes as a result o f the initial antihypertensive
therapy.
I. Defining a ‘switch’ in medications:
For this analysis, the patient’s first episode was studied. Each episode consists o f a 6
month washout period prior to the keydate plus a 36 month period after the keydate.
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The dataset contains information on the drugs that the patients received starting from
the keydate up to 3 years from the index date or for as long as they were eligible for
that particular episode. At the Keydate, the patient received one or more drugs that
belonged to certain medication drug classes. This analysis is limited to the following
antihypertensive drug classes: diuretics, p-blockers, ACE Inhibitors and CCBs since
these are the four drug classes that are recommended by different JNC reports for the
initial treatment o f hypertension i.e. the first line therapy hypertensive patients.
If the antihypertensive medication class prescribed to the patient at the second
prescription fill date (i.e. the fill data following the keydate), is not the same as the
antihypertensive medication class received at keydate (also referred to as the first
date), then this is considered a switch in medication drug class. The time period
between the keydate and the second date at which a switch was recorded is
documented as the time to switch in medication class. If the medication class is the
same at Keydate and the second prescription date, then the third date o f prescription
fill is checked to see if the medication class o f the prescription filled at the third date
was similar to the medication class filled at keydate and the second date o f fill. If it is
different, a switch is recorded and the time to switch is docum ented as the time
period between the date o f third prescription fill and the keydate. This process is
continued till all the observations have a recorded time to switch. The time to switch
was recorded for each observation as described above. The aim was to study the time
to switch for the study sample and look for significant differences in time to event (in
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
this case, a switch) between patients who were treated in accordance with the
guidelines versus those that were not, and also to study the time to event for patients
that received therapy according to the JNC 4 guidelines versus the JNC 5 guidelines.
Kaplan M eier Method for Time to Event Analysis
A simple method that performs this type o f analysis is the Kaplan Meier method for
survival analysis. In this analysis, survival is recorded as the time between the
keydate and the first switch in medication class. The KM estimate is the most widely
used method for performing survival analysis. Before going to the KM methodology,
a few basic concepts o f survival analysis need to be addressed.
Origin o f time: Each observation should have an origin o f time, at which the data
will begin to be analyzed. In this dataset. The origin of time is the keydate. i.e. the
date at which the patient received the first antihypertensive medication.
Censoring: Each observation is followed over time starting from keydate until the
occurrence o f the event. In this analysis, the event is a switch in medication class.
The dataset contains observations on patients starting from the keydate for a period
o f 3 years or until the patient discontinues the antihypertensive therapy. For patients
that were present in the dataset for the entire three year period, a switch was recorded
as described above. If there was no switch during the entire three year period, the
32
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
observation was right censored at 3 years (number o f days=1092). If the observation
did not make it till the end o f three years and discontinued therapy in between, the
observation was randomly censored at that point in time at which the observ ation
was discontinued. In this analysis, the origin o f tim e is the index date for each
observation, and since the index date for each observation is well recorded, there is
no problem with left censoring.
The KM method was applied to the dataset and time to switch in medication class
(i.e. the survival curves) was analyzed. Different survival curves were studied for
patients receiving first line antihypertensive therapy according to guidelines and
patients not receiving first line therapy according to guidelines. Survival plots were
also compared between patients that received first line therapy according to :he JNC
4 guidelines and patients that received first line therapy according to JNC 5
guidelines.
Criteria for defining a patient as having received treatment in accordance with
the guidelines
Patients that had their index date fall under the JNC 4 period in addition to not
having an ICD 9 diagnosis for a coexisting illness on or before the index were
defined as receiving treatment in accordance with guidelines if they received either
diuretics, p-blockers, ACE inhibitors or CCBs as their initial antihypertensive
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
medication. If the patient had a comorbidity for which either of the above four
medication classes were specifically contraindicated, and if the patient did receive
any o f these medications, then he was defined as not receiving therapy in accordance
with guideline recommendations, even though these four classes were recom mended
for initial therapy under the JNC 4 period.
Similarly, patients with no comorbidity prior to the index date and with an index date
in the JNC 5 period were defined as receiving therapy in accordance with guideline
recommendations if they received either diuretics or b-blockers as initial therapy.
However, if they had a comorbidity for which these medication classes were
contraindicated, and yet they received these medictaion classes, then they were
defined as not receiving therapy in accordance with guideline recommendations.
2. Tim e to augmentation in antihypertensive therapy
For this analysis, the patient's first recorded episode was studied. The guidelines
recommend that if the hypertensive patients are not well controlled after receiving
their first antihypertensive medications, then the physicians should either switch their
medication class or add another medication to the existing regimen. A switch in
medication was recorded as described above. For those patients that received an
addition to their existing therapy, an augmentation in therapy was recorded. An
augmentation in therapy could be an addition o f a drug that belonged to the sam e
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
medication class as the existing drug therapy or it could be a drug from another
medication class. The time to the first augmentation in therapy, starting from the
keydate is recorded and the Kaplan M eier method was applied as above to study
differences in patients receiving treatment according to guidelines versus those that
did not.
3. Tim e to first hospitalization
For this part o f the analysis, data from the hospital paid claims was analyzed. In the
dataset, the costs due to hospitalizations are partitioned in 37 periods around the
keydate: a 6-month prior period and 36 one-month-post-keydate periods. Hence if
the cost for hospitalization in the first month after keydate is not zero then time to
first hospitalization is assumed to be one month. If the first month hospital costs are
zero, we will assume that the patient did not have a hospitalization in the first month.
If subsequently the second month costs are non zero, the time to first hospitalization
will be assumed to be 2 months. This is an approximate measure o f time to
hospitalization. Data in available if the patient was hospitalized with a diagnosis o f
hypertension, any other diagnosis, or no diagnosis at all. The time to hospitalization
analysis will focus on time to hospitalization with a hypertension diagnosis to affirm
that this hospitalization was related to hypertension and not some other disease
condition. If the data is analyzed in this manner, conclusions can be drawn regarding
the impact o f the anytihypertensive therapy for antihypertensive patients.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The time to hospitalization analysis will be conducted in the same manner as the
time to switch in medication class as described above, using the Kaplan Meier
method.
4.Tim e to Discontinuation of antihypertensive therapy
For this analysis, the patients prescription profile was studied. The dataset contains
information on MediCal drug code and date o f service among other variables. The
dataset does not contain information on the number o f days o f prescription supply. In
the absence o f this information and given that the patients in the study were
continuously MediCal eligible (eligibility criteria explained above), patients that
interrupted antihypertensive therapy were identified using the date o f service and
MediCaTs policy o f a limitation o f 30 days supply for each prescription. Allowing
for a 15 day grace period, if the patient did not fill an antihypertensive prescription
within 45 days o f the previous prescription, then that patient was assumed as having
interrupted antihypertensive therapy. If the patient stopped taking prescriptions at a
certain point in time before the end o f the study period (i.e. first episode length of 3
years form keydate), then the patient was defined as having terminated
antihypertensive therapy. Whether the patient interrupted or discontinued
anyhypertensive therapy, the time to the first o f these events starting from keydate
was recorded as time to discontinuation of anytihypertensive therapy. The time to
discontinuation was compared between patients that received initial therapy
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
according to guidelines versus those that did not, also between patients receiving
initial therapy in accordance with the JNC 4 guidelines versus the JNC 5 guidelines.
The analysis method was the Kaplan Meier Method, as described above.
Effect o f Covariates on Time to Event
To study the effects o f covariates on the time to event, the Cox Proportional Hazards
Model is preferred over the KM method. One reason is that the KM method (PROC
LIFETEST in SAS) does not give the coefficient estimates and hence the effect of the
covariate cannot be quantified. The Cox Proportional Hazards model (PROC
PHREG in SAS) quantifies the coefficient estimates and hence this method is
suitable.
Cox Proportional Hazards Model:
The outcome variable o f interest is the time (t) to first event (hospitalization, switch,
discontinuation). The unit o f the time to event variable is days .
The model is specified as follows:
h,(t) = A0(t) exp {/?/.r,/+....... + p k xlk ) (Eq. 1)
An(t) is the baseline hazard function and \\\ to x* are a set o f k fixed covariates.
The hazard function hi(t) is said to be the product o f the two above mentioned
factors.
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Taking the logarithm o f both sides, Eq. 1 becomes
log h,(t) = a (t)+ fiix ,i+ +PkX,k
where aft) - log Aoft)
Using this formula, the Cox proportional hazards model gives the hazard for any
individual as a fixed proportion o f the hazard for any other individual. For eg. the
ratio o f the hazards for two individuals i and j is as follows:
h,(t) = exp {b,(x,,-Xji) + +bk(xlk -xjk)} (Eq. 2)
hj(0
Eq. 2 shows that the ratio o f hazards is constant over time.
Covariates
In the above Cox proportional hazards model, the dependent variables are time to
switch, time to discontinuation and time to first hospitalization. The factors that
could influence these dependent variables are included as covariates in the model.
The covariates defined for this model are age o f patient at keydate, gender, the JNC
period that the keydate fell under, whether the patient received first line therapy
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
according to guideline recommendations and whether they developed and
comorbidities during that time period i.e. time period from keydate to the time o f
event. Demographic variables such as age and gender are included since these
variables are expected to influence the time to a switch, hospitalization or
discontinuation., for eg. older patients may be expected to have shorter time to an
event, because they could be more frail as they age. Whether the patients received
initial antihypertensive therapy in accordance with the guideline recommendations is
a major factor that could influence the subsequent hospitalizations, switches in
medictaion class and discontinuation o f therapy. Hence this variable was included.
Presence o f cardiovascular comorbidities could greatly influence any o f the three
events under consideration and hence these comorbidities have been included as
covariates that could impact the time to these events.
Estimation o f the Cox Proportional Hazards Model:
This model can be estimated by the partial likelihood method. The partial likelihood
estimates obtained are approximately unbiased in large samples and their sampling
distribution is approximately normal. From partial likelihood estimates, the Wald test
or the likelihood ratio test can be performed to test for significant effects of any .v,.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Research Question 3
What is the economic im pact o f following guideline recommendations?
Specific Study Procedure
Estimation o f health care costs:
In the MediCal dataset the patient health care costs were partitioned into monthly
costs for each o f the follow ing types o f services: hospital services, ambulatory care,
prescription drugs, long-term care and other costs. Hence there were 36 months,
starting from the keydate, which had cost data for each o f the above services in each
month and 6 months o f cost data for each o f the months prior to the keydate. There
were some observations that had zero costs in some o f the months. This does not
mean that patients lost eligibility and hence they had zero costs, because as
mentioned earlier in the sample selection criteria, patients selected into the study
sample have already been screened for being continuously eligible. The presence o f
zero costs could reflect the fact that these patients were ‘cross-over’ patients i.e. they
were dually eligible for Medicare and Medicaid. For cross-over patients. M edicare is
the primary payer for all services covered by the Medicare program, hence the dollar
values recorded on the MediCal paid claim will be understated if the health care
service was a Medicare service that was consumed by a cross-over patient. This issue
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
was taker, care o f by utilizing the length o f stay data for hospital and skilled nursing
home (SNF) services that were recorded on the MediCal paid claim every month.
The total num ber o f hospital days in a particular month are multiplied by the average
hospital per diem cost to MediCal to calculate the monthly hospital costs. Similarly,
the number o f days o f stay in the SNF is multiplied by the average SNF cost per
diem to Medical, to calculate monthly SNF costs. The monthly costs were then
totaled to calculate the total health care costs.
Some patients had their index date after 1993 (the time frame o f the dataset was up
to Dec 1996), they did not have an entire three year period for calcu latin g health
care costs. Hence the costs were calculated as one year, two year and three year
costs, in order to compare costs between pateints that had the same follow-up cost
data.
All monthly costs were totaled to calculate the total health care costs for the 3 year
period starting from the keydate. For this analysis, patients that started their episode
on or after January 1, 1994 were not included in the analysis since the cost data for
the entire 3 year post keydate period was not available. Table 2.2 shows the number
o f patients in each period that had one, two, and three year cost data.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2.2: Number o f patients with 1. 2 and 3 year health care costs by JNC period
Period 2
JNC 4
Period 3
JNC 5
Total N um ber o f patients 907 995
Patients with 1 year follow-up costs, N 907 805
Patients with 2 year follow-up costs, N 907 530
Patients with 3 year follow-up costs. N 907 280
C ontrolling for Selection Bias
As in any non-randomized observational study, there is a high potential o f having a
bias in the estimation of treatment effects. This is caused by the failure to control for
the selection bias that occurs due to some unobserved variables in the model that
correlate with the treatment selection and subsequently the treatment outcomes. The
best form o f controlling for potential biases in studies that involve treatment effects,
is to perform a randomized controlled trial (RCT). However in most cases, it is not
possible to conduct such studies and often the researcher has to deal with the
selection biases that arise when using non-randomized observational databases to
study a treatment effect.
Usually, treatment effects are estimated using regression analyses, where the
treatm ent is modeled as a dummy variable that takes on a value o f 1, if the patients
42
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
received the treatment, or a value o f 0, if the patients did not receive the treatment. In
the regression analysis model, observed variables are controlled for when making
inferences about the treatment effects. However, in non-randomized studies, there
always are some variables that are unobservable to the researcher and beyond the
scope o f the dataset. It is precisely these unobservable variables that bring the
potential selection bias into the study. If the patients in the two groups, one receiving
the treatment and the other not receiving treatment, differ in certain unobservable
characteristics that could be related to the outcome under study, then the researcher
will incorrectly attribute the outcome to the treatment received, when in reality the
outcome was due to the initial unobservable differences in characteristics between
the two groups. Hence this leads to erroneous conclusions about the treatment effect,
leading to a serious problem o f selection bias.
In this analysis, the goal is to study the economic impact o f following guidelines
versus not following guidelines in the elderly antihypertensive patient. The outcome
variable is the three-year total health care costs o f the patients, starting from the
keydate. At keydate the patient receives initial antihypertensive therapy from the
physician, the therapy being either in accordance with the guidelines or not in
accordance with the guidelines.
This feature classifies the patients into the two study groups, one group o f patients
that received treatment in accordance with the guidelines and the second group that
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
did not receive treatment in accordance with the guidelines. The question that needs
to be answered is whether the patients that received treatment in accordance w ith the
guidelines had significantly different health care costs compared to the other group.
A multivariate regression analysis that controls for observable variables such as age
and gender will allow us to study the treatment effects (in this case the treatment
refers to the patient getting treated in accordance with the guidelines) on the outcom e
variable. However, as discussed earlier, at the time o f receiving initial therapy, if the
two groups o f patients differed in certain characteristics that are beyond the scope o f
the dataset, these differences need to be controlled for before making any
conclusions o f the treatment effects on the outcomes variable.
The sample selection model used in this analysis, (Heckman J) is a method that
allows us to control the bias in treatment selection due to unobservable variables
which may or may not be associated with the outcome variable o f interest. The
sample selection model for the treatment selection and the outcome o f interest as
described above is simple and is estimated as follows. The procedure for controlling
for bias is a two-stage process. In the first stage, a Probit model is estimated to study
the probability o f receiving treatment in accordance to guidelines or not. From the
estimates o f the Probit model, a ratio called the inverse mills ratio (IMR). also
known as the lambda, is calculated. The IMR is defined as the probability o f
receiving treatment in accordance with the guidelines given that the patient was ‘at
risk’ for receiving such a treatment. Once the IMR is calculated from the Probit
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
procedure, it is then used as an additional independent variable in the second stage of
the sample selection model which is the OLS regression stage, or the stage where
the parameters associated with the outcome o f interest are estimated. If the estimated
coefficient o f the Inverse Mills Ratio is significantly different from zero then this
implies that there was a selection bias to start with, however, due to the inclusion of
the inverse mills ratio, from the first stage Probit model, as a covariate in the
outcom e equation, the selection bias has been accounted for. If the estimated
coefficient o f the inverse mills ratio is not significantly different from zero then there
was no selection bias in the first place and the significance o f the covariates in the
outcom e equation would have been the same even if the classical linear regression
model had been used.
Stage One: The Probit Model (Selection Stage)
The independent variables that are hypothesized to be associated with the probability
o f receiving treatment according to guidelines are outlined in table 2.3. The
dependent variable for this stage is a dumm y variable where a value o f 1 indicates
that the patient received the first antihypertensive treatment in accordance with the
guidelines, and a value o f 0 indicates otherwise.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2.3: Independent variables for the first stage Probit model
Variable Description
Demographics
Age Age o f the patient at the time o f
first antihypertensive
prescription i.e. at keydate
Geographical region
North, South and Central Different counties in CA coded
into north, south and central
CA, to account for variation in
physician practice patterns
Prior diagnosis for a co-morbid condition
Arrythmia (A n), Atherosclerosis
(Athero),Coronary artery disease, (CAD),
Chronic ischemic heart disease (CHD),
Congestive heart failure (CHF), Chronic
obstructive pulmonary disorder (COPD),
Heart block (HB), Thrombosis
Co-morbidity in the 6-month
period prior to keydate,
included to allow for likelihood
o f special prescribing decisions.
This dataset was limited in that there was not much information on patient or
physician characteristics that would have allowed us to better understand the
physician’s prescribing behavior. The only information that was available prior to
index date that could have had a potential impact on physician prescribing pattern
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
was information such as patient demographics, geographic location and
comorbidities prior to the index date. These factors were thus included in step 1 o f
the sample selection model as independent variables with a potential impact on the
selection process, i.e. whether patients got first line antihypertensive therapy in
accordance with guidelines or not.
Table 2.3: Independent variables for the first stage Probit model
Variable Description
Demographics
Age Age o f the patient at the time o f
first antihypertensive
prescription i.e. at keydate
Geographical region
North, South and Central Different counties in CA coded
into north, south and central
CA, to account for variation in
physician practice patterns
Prior diagnosis for a co-morbid condition
Arr, Athero, CAD, CHD, CHF,
COPD.HB, Thrombo
Co-morbidity in the 6-month
period prior to keydate,
included to allow for likelihood
o f special prescribing decisions.
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The IM R for each patient is calculated from the above model, and this represents the
effects o f the unobserved variables that could be associated with receiving treatment
in accordance with the guidelines or not. The IMR is plugged into the outcome
equation in stage 2, as an additional explanatory variable.
Stage 2: The OLS M odel (Outcome equation)
In this stage, the factors associated with health care costs can be identified. The
dependent variable is the total health care costs. The independent variables used in
this second stage need to be different from the independent variables used in the first
stage to avoid the problem o f multicollinearity. Health care costs in the pre-period,
com orbidities that the patient developed developed during the course o f therapy and
whether the patient visited only pharmacy for all his prescriptions were among the
independent variables used in this stage o f the analysis (Table 2.4)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2.4: Independent variables in the outcome equation
Variable Description
Demographics: Gender Fem ale=l, male=0
Precost Prior health care costs in the 6-
month period before the keydate
One Pharmacy Patients received all their
antihypertensive prescriptions from
one pharmacy only.
Post diagnosis for a co-morbid
condition: Arr. Athero. CAD.
CHD, CHF, COPD, HB. Thrombo
Co-morbidity in the 3-year period
following the keydate.
Lambda (IMR) Inverse Mills Ratio
Patient received any guideline
therapy (JNC 4 or JNC 5) versus
none
OR
Patient received JNC 4 guideline
therapy versus JNC 5 guideline
therapy
Dummy variables which make it
possible to compare differences in
health care costs between two
groups o f patients (both variables
not included in same model,
separate models run for each of
these variables, with the same
covariates)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Retransformation of the Logged Dependent Variable
Since the dependent variable, i.e. the total health care cost, was not normally
distributed, a log transformation was applied to achieve normality. After the least
square regression model was fitted and before inferences were made on the
significance and magnitude o f the parameter estimates, a retransformation procedure
was carried out. This is desirable because the consistent and unbiased param eter
estimates on the transformed scale are not necessarily consistent and unbiased on the
untransformed scale. Thus, in order to study the magnitude and significant
differences in the health care costs between the two groups, the logged dependent
variable needs to be retransformed. The regular procedure for retamsformation is to
exponentiate the transformed scale regression coefficents. however this will lead to
unbiased and inconsistent estimates on the untransformed scale due to the problem of
heteroscedasticity in the error term distribution. This problem was dealt with by
using the smearing estimate retransformation procedure (Duan N) for consistent and
unbiased parameter estimates after retransformation o f the regression coefficients.
The smearing estimate procedure was applied to the models with 1, 2 and 3 year
health care costs and the magnitudes and significant differences in health care costs
between patients who were treated in accordance with guideline recommendations
and those that did not, were compared.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Sample Selection Bias in the Cox Proportional Hazards Models in the Time to Event
Analyses (Switches, Hospitalizations and Discontinuation in Therapy)
In the Cox proportional hazards model for the time to event analyses described in
research question 2, there is a possibility o f selection bias. However, in this type o f
analysis, it is a challenge to staistically control for selection given the nature o f the
data. In this data, there is frequently the issue o f random censoring and right
censoring o f data. Such theoretical issues are built into the models dealing with the
Cox proportional hazard for time to event, and the results account for the censored
data and hence we are able to draw conclusions. However, in the sample selection
models described above, there is no such accomodation for censored data and hence,
controlling for selection bias in the time to event analysis is difficult and theortically
challenging. Hence selection bias was not acounted for in the time to event analysis
and this is a limitation for that particular analysis.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 3: RESULTS
The initial MediCal sample consisted o f 55,829 patients. After including only those
patients who were elderly ( 65 years o f age or older), the sample size was cut dow n
to 22,147 patients. Patients who did not have a diganosis o f hypertension on or
before the index date were excluded from the sample bringing the final sample size
down to 2,162 hypertensive elderly patients. Table 3.1 shows the general
characteristics o f the overall sample. The average age o f the population was 74.58
years and about 67% percent o f the population was female.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.1: General Characteristics o f the Sample (N=2162)
Characteristic Number Percent
Age, yrs (mean, std dev) 74.58 (7.4) -
min-65
Max-99.83
Gender (female) 1455 67.3
Initial Drug Therapy
Diuretics 295 13.64
P-blockers 285 13.18
- ACE Inhibitors
472 21.83
- Calcium Channel Blockers (CCBs)
584 27.01
- Others
646 29.88
Use o f one pharmacy for all 938 43.39
antihypertensive prescriptions
Patients receiving initial therapy in :
Pre JNC 4 period 260 12.03
- JNC 4 period 907 41.95
JNC 5 period 995 46.02
Less than 50% o f the patients received all their antihypertensive medications from
one pharmacy only. Majority o f the sample belonged to the JNC 5 period i.e. they
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
were prescribed their first antihypertensive prescription in the period after the
publication o f the JNC 5 guidelines (January 1993).
Table 3.2 shows the univariate analysis o f patient dem ographics and other variables,
by JNC 4 and the JNC 5 periods. The pre-JNC 4 period was not included in the
univariate analysis because the sample size was small and the focus of the research
was specific to the differing guideline recommendations o f the JNC 4 and JNC 5
reports.
The age o f the population and proportion o f female patients were not significantly
different in the two periods. In the JNC 5 period, a significantly lower proportion o f
patients had various coexisting disease conditions com pared to the JNC 4 period. A
higher number o f patients in the JNC 5 period started antihypertensive therapy with
ACE inhibitors compared to the JNC 4 period, while significantly fewer patients
started with Calcium Channel Blockers (CCBs). A significantly higher number o f
patients in the JNC 5 period used one pharmacy for all their anti-hypertensive
medications com pared to the JNC 4 period. There were no significant differences in
the proportion o f patients starting with diuretics or b-blockers between the two
periods.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.2: Characteristics o f the sample by JNC 4 and JNC 5 Periods
Characteristic JNC 4 Period
N=907
JNC 5 Period
N=995
Age 74.96 (7.45) 74.21 (7.34)
Gender (Female) 625 (68.91) 645 (64.82)
Diagnosis
Ischemic Heart Disease (IHD) 66 (7.28) 50 (5.03)*
Chronic Ischemic Heart Disease (CHD) 213 (23.48) 149 (14.97*
Arrythmia (ARR) 180(19.85) 127 (12.76)*
Thrombosis (THROMBO) 22 (2.43) 10(1.01)*
Congestive Heart Failure (CHF) 221 (24.37) 141 (14.17*
Angina (ANG) 171 (18.85) 145 (14.57)*
Coronary Artery Disease (CAD) 133 (14.66) 107(10.75)*
Acute Myocardial Infarction (AMI) 29 (3.2) 22 (2.21)
Bradycardia (BRADY) 63 (6.95) 56 (5.63)
Atherosclerosis (ATHERO) 75 (8.27) 47 (4.72)*
Asthma 99(10.92) 88 (8.84)
Chomic Obstructive Pulmonary
Disease (COPD)
11 (1.21) 21 (2.11)
Initial Drug Therapy
Diuretics 130(14.33) 113(11.36)
b-blockers 128(14.11) 115 (11.56)
ACE Inhibitors 198 (21.83) 261 (26.23)*
Calcium Channel Blockers 264 (29.11) 212(21.31)*
Vasodilators 3 (0.33) 1(0.1)
Anti-adrenergic agents 45 (4.96) 25 (2.51)*
Fixed Combinations 8 (0.88)
1 (0.1)*
Alpha-blockers 178 (19.63) 299 (30.0)*
Alpha-beta blockers 6 (0.66) 1(0.1)*
One Pharmacy for all
antihypertensive prescriptions
341 (37.6) 518 (52.06)*
* p < 0.05
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Specific Aim 1
The first main goal o f this research was to study if physicians were prescribing in
accordance with guideline recommendations for first line antihypertensive therapy.
Figure 3.1 shows the overall initial utilization o f each o f the four classes o f
antihypertensive drugs recommended for initial drug therapy, in the JNC 4 and the
JNC 5 periods.
Figure 3.1: Overall utilization o f initial antihypertensive therapy in JNC IV and JNC
V period
□ JNC 4
■JNC 5
Diuretics b-blockers ACE CCB Others
Inhibitors
The above graph shows that the utilization ACE inhibitors was higher in the JNC 5
period while the utilization o f b-blockers and CCBs was lower in the JNC 5 period
compared to the JNC 4 period. The utilization o f diuretics was similar in both the
periods. Although the above graph shows the overall utilization in each period, the
goal o f this research was to study the trend o f antihypertensive utilization i.e. study
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the pattern o f drug utilization over the different periods and look for significant
differences in this pattern. As described in the Methods section the segmented time
series regression methodology was applied to achieve this goal. Figures 3.2 to 3.5
represent the data on the proportion o f utilization o f each o f the four classes o f
antihypertensive drugs: diuretics, b-blockers, ACE inhibitors and CCBs over the 10
year time frame. The time frame was divided into 3 periods with join points
delineating the periods into pre JNC-4, JNC 4 and the JNC 5 period. Join point 1 was
when the JNC 4 guidelines were published (May 1988) and join point 2 was when
the JNC 5 guidelines were published (January 1993).
Figures 3.2 to 3.5 show the proportion o f new starts o f diuretics, b-blockers, ACE
inhibitors and CCBs per month. It is very apparent that there is a lot o f variation in
the data points and it is not possible to graphically visualize a change in the slopes
form period to period. However after careful observation there seems to be a
decreasing trend in the monthly proportion o f b-blockers towards the second half to
period 2 and an increasing trend in utilization in period 3. In figure 3.4, there is an
increasing trend o f initial starts o f ACE inhibitors from period 2 to period 3, while
figure 3.5 shows a decreasing trend in initial starts o f CCBs from period 2 to period
3.
Due to a large amount o f variation in the monthly readings a 3-month moving
average graph was plotted. The 3 month moving average graph aimed at smoothing
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.2: Proportion o f new starts o f Diuretics per month
t
o
a
o
k .
a
o
s
0
b
3
o
5 0
40
30
20
10
0
— •” — dLzt=— •/. ~ -.“ ~ z i — -7. ~ =5 — -7 . ^ r i — -7 .^5 -
Month
Figure 3.3: Proportion o f new starts o f p-blockers per month
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.4: Proportion o f new starts o f ACE Inhibitors per month
90
80
70
60
50
30
20
10
0
Month
Figure 3.5: Proportion o f new starts o f CCBs per month
100
c
o
t
o
a
o
w
Q .
ID
O
U
Month
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the curve to get a better visual understanding o f the pattern o f the utilization o f each
o f the four antihypertensive agents over time. In the 3 month moving average graph,
the data point for each month is an average o f that particular month and the two
m onths prior to it. Figures 3.6 to 3.9 show the 3 month moving average graphs for
diuretics, P-blockers, ACE inhibitors and CCBs respectively. As with the monthly
graphs plotted earlier, it was not possible to detect any changing trends between
periods for the monthly utilization o f diuretics. The monthly utilization o f new starts
o f p-blockers had an increasing trend in period 3. For ACE inhibitors, there was an
increasing trend from period 2 to period 3 and for CCBs. there was a decreasing
trend from period 2 to period 3, similar to the observations in the previous graphs o f
monthly proportions. Although the visual graphs give us a feel for the pattern o f
utilization, the statistical significance o f the curves with respect to inter-period
differences was tested in the segmented time series regression model. Four separate
models were tested, one for each o f the four classes of initial antihypertensive
medications under consideration, diuretics, P-blockers, ACE inhibitors and CCBs.
Each model was tested for variance homogeneity and independent error terms, the
two assumptions o f the linear regression model. The plot o f residuals versus the
predicted values o f the dependent variables was observed for heteroscedasticity. For
each o f the four models, the scatter plot o f residuals versus time showed a patternless
scatter o f points, thus allowing for the assumption of homoscedasticity. For the four
models tested above, the DW statistic was 2.2,2.1,
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.6: Proportion o f Diuretics- 3 month moving average
45.00
40.00
. 35.00
0 30.00
r 25.00
1 20.00
I 15.00
& 10.00
5.00
0.00
Time
Figure 3.7: Proportion o f P-blockers - 3 month moving average
40.00
35.00
e 30.00
| 25.00
s 20.00
g 1500
a. 1 0 . 0 0
500
0.00
Time
Periodl
P e r i o d 2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.8: Proportion o f ACE Inhibitors - 3 month moving average
i
Time
Figure 3.9: Proportion o f CCBs - 3 month moving average
70.00
60.00
5Q00
4Q 0O
X.00
20.00
10. 00
Q 00
Time
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.08 and 1.6 for diuretics. P-blockers, ACE inhibitors and CCBs respectively. Hence
the assumptions o f the model were not violated and there was no apparent need for
transformation o f the variables. Table 3.3 shows the output o f the regression model
for the segmented regression models o f the four classes o f antihypertensive
medications.
Table 3.3: Segmented Regression Output
Antihypertensive
Medication Class
Parameter Estimate(s.e)
po
PI
P2
P3
Diuretics 22.34 -0.44 -0.05 0.05
(4.95) (0.27) (0.32) (0.15)
P-blockers 2.12 0.73* -0.89* 0.28
(4.3) (0.23) (0.28) (0.13)
ACE Inhibitors -6.29 1.05 (0.29) -0.98 0.02
(5.31) (0.34) (0.16)
CCBs 53.66 -1.018 0.92* -0.12
(5.84) (0.32) (0.38) (0.18)
* p < 0.05
In the above table, the beta estimates are defined as follows:
Po = intercept for period 1
Pi = slope for period 2 or trend in monthly proportion from June 1986 to M ay 1988
63
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
02 = difference in slopes between period 1 (June 1986 to may 1988) and period 2
(June 1986 to Jan 1993)
03 = difference in slopes between period 2 and period 3 (after Jan 1993)
The expectations from the model were that, in an ideal situation, if guidelines should
have been followed, the utilization o f ACE inhibitors and CCBs would increase from
period 1 (Pre-JNC 4 period) to period 2 (JNC 4 period) and decrease from period 2
to period 3 (JNC 5 period) due to the restrictive recommendations o f the JNC 5
guidelines for first line antihypertensive therapy, compared to the JNC 4 guidelines.
Hence the 02 parameter estimate was expected to be positive and the 03 estimate
negative for ACE inhibitors and CCBs. Also, the utilization o f 0-blockers and
diuretics was expected to decrease from period 1 to period 2 and increase from
period 2 to period 3.
From the above results it is apparent that from period 1 to period 2, the utilization o f
b-blockers significantly decreased and that o f CCBs significantly increased, as
expected. The direction o f the 02 estimate for diuretics was in the expected direction
(i.e. negative) but was not significant. From period 2 to period 3, the utilization o f 0-
blockers significantly increased as expected and although the 03 estimate was in the
expected direction for diuretics and CCBs, it was not significant in either case. The
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
parameter estimates for ACE inhibitors were not in the expected direction for either
P2 or 03.
To allow for some time for the guideline information to be disseminated to the
physician population once the guideline recommendations have been published,
sensitivity analyses were performed by shifting the join points that delineate the time
frame into the Pre-JNC 4, JNC 4 and JNC 5 periods. By shifting the join points
ahead by 3, 6 ,9 and 12 months, the slope o f the trend in utilization could potentially
change and this could affect the results in terms o f significantly differing slopes
between periods, and subsequently significantly differing prescribing patterns
between periods.
The segmented regression model was run for each o f the above four models by
shifting the join points ahead by 3 ,6 ,9 and 12 months. The antihypertensive
prescribing pattern for diuretics, 0-blockers, ACE inhibitors and CCBs remianed the
same between periods as in the original model. Hence, incorporating time lag into
the analysis did not change the results.
Specific Aim 2
The second goal o f the research was to study the impact o f physician adherence to
guidelines on patient outcomes such as time to the first hospitalization, switch in
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
medication class, augmentation in drug therapy and time to discontinuation of
therapy, starting from the keydate.
Time to switch in medication class
The Kaplan Meier survival curves ( plot o f time to first switch, as described in the
methods section) were analyzed for difference in time to switch in medication class
in days), between patients that were treated according to guidelines versus those that
were not. A total o f 1971 patients were studied. Of these 1971 patients 808 patients
were censored observations. Figure 3.10 shows the Kaplan M eier survival curve for
differences between the times for patients receiving the first antihypertensive
medication class according to guideline recommendations (either JNC 4 or JNC 5
guideline recommendations), compared to patients that are not treated according to
guideline recommendations. From the visual inspection o f the survival curves there
seems to be no apparent difference in the two curves and infact after about 600 days
(X-axis) the curves are virtually indistinguishable. Results from the PROC
LIFETEST procedure in SAS confirm the conclusions made by visual observation o f
the survival curves. Both, the Wilcoxon and the Log-Rank Statistics are not
significant with p values o f about 0.4 and 0.6 respectively. This confirms that the two
survival curves are homogenous and not significantly different from each other.
Hence the null hypothesis o f no difference in time to switch in medication class
between the two groups is not rejected.
66
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.10: Plot o f Survival Function for the Two Groups:
Dependent Variable = Time to Switch (ttswitch); G uideyes=l: Patients received
treatment in accordance with guidelines (JNC 4 or JNC 5)
% 1.00
u
e
£ 0.75
e
0
1 0.50
£
* 0.25
5 0.00
>
L.
200 400 600 800 1000 1200
ttswitch
ST R A T A --------------- guidyes=0 O O O censored guidyes=0
-------------- guidyes= 1 O O O censored guidyes= 1
Figure 3.11: Plot o f Survival Function for the Two Groups: Dep Variable =
Time to Switch (ttswitch); periodyes=JNC 4: Patients treated in accordance
with JNC 4 guidelines; periodyes=JNC 5: Patients treated in accordance with
JNC 5 guidelines
C
o
< J
c
e
o
3
- C
C 3
>
E
3
V i
1.00
0.75
0.50
0.25
0.00
0 200 400 600 800 1000 1200
ST R A T A periodyes= JNC 4
periodyes= JNC 5
ttswitch
O O O censored periodyes=JNC 4
O O O censored periodyes=JNC 5
67
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A nother model was run that tested the null hypothesis o f no difference in time to
switch in medication class between patients receiving treatment in accordance with
the JNC 4 guidelines versus patients that were treated in accordance with the more
restrictive JNC 5 guidelines. Figure 3.11 shows the Kaplan Meier survival curves for
the tw o groups.
From visual observation, it is apparent that there is a difference in the two curves.
The survival curve for JNC 4 has a higher survival time compared to the JNC 5
curve. The PROC LIFETEST procedure in SAS yielded the following results. Both
the Log-Rank and the Wilcoxon statistics were significant with a p value o f < 0.001.
Thus the null hypothesis o f homogenous curves is rejected and it is concluded that
the survival curve o f patients receiving treatment in consensus with the JNC 4
guidelines declined less rapidly than the survival curve o f patients who received
treatment according to the JNC 5 guidelines.
To study the effects o f covariates on the hazard o f having a switch in medication
class, the Cox Proportional Hazards Model was employed. The dependent variable
was tim e to switch in medication class (table 3.4).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.4: Cox regression model output
Dependent variable: Time to switch in medication class
Variable Parameter
Estimate (S.E)
Hazard
Ratio
P
Value
Age 0.02 (0.01) 0.998 0.75
Gender (female) 0.50 (0.25) 1.60 0.04
Received treatment in accordance with
guidelines (compared to patients not
receiving treatment in accordance with
guidelines)
-0.83 (0.22) 0.43 0.04
Diagnosis of:
ARR 0.007 (0.18) 1.01 0.96
Athero 0.34 (0.16) 1.40 0.04
Hblock -0.05 (0.12) 0.94 0.64
CHD -0.09 (0.24) 0.91 0.69
Thrombo 0.13(0.10) 1.14 0.18
CHF -0.06 (0.14) 0.94 0.66
CAD -0.09 (0.48) 0.90 0.83
From the above table, looking at the hazard ratio estimates and the p values, it is
clear that for patients receiving treatment in accordance guidelines, the hazard o f a
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
switch in medication class is 60% higher for females than for males. Patients
receiving treatment in accordance with guidelines had a 57% lower hazard for a
switch in medication class compared to patients that did not receive antihypertensive
therapy in accordance with guidelines, controlling for other covariates.
Time to First Hospitalization
The time to the first hospitalization (in number o f months), from the keydate was
calculated as described in the methods section. Figure 3.12 shows the Kaplan M eier
survival curve for differences between the times for patients receiving the first
antihypertensive medication class according to guideline recommendations (either
JNC 4 or JNC 5 guideline recommendations), compared to patients that are not
treated according to guideline recommendations. From the survival curves there
seems to be a gap between the two curves, suggesting that he two curves are not
homogenous.
From the results o f the PROC LIFETEST procedure, the significance o f the Log-
Rank and the Wilcoxon statistic confirms that the survival curves o f patients
receiving treatment according to guidelines declined significantly more slowly than
the survival curve o f patients not receiving treatment according to guidelines. This is
the expected result, since patients being treated according to guidelines are assumed
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.12: Plot o f Survival Function for the Two Groups:
D ependent Variable = Time to hospitalization (tthosp); Guidyes=l: Patients
received treatment in accordance with guidelines (JNC 4 or JNC 5)
_o
U
3
3
- C
>
3
C/3
ST R A T A
1.00
0.75
0.50
0.25
0.00
25 20 30 0 10 15 35 40 3
tthosp
guidyes=0 O O O censored guidyes=0
guidyes=l O O O censoredguidyes= 1
Figure 3.13: Plot o f Survival Function for the Two Groups: Dep Variable=Time
to hospitalization (tthosp); periodyes=JNC 4: Patients treated in accordance with
JNC 4 guidelines; periodyes=JNC 5: Patients treated in accordance with JNC 5
o
e
o
3
- O
7 1
s
*3
*E
3
C /3
ST R A T A periodyes= JNC 4 O O O censored periodyes=JNC 4
periodyes= JNC 5 O O O censored periodyes=JNC 5
71
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
to have better outcomes, or in this case, greater number o f patients surviving until the
event (hospitalization) over time compared to patients not being treated in
accordance with guideline recommendations.
Another model was run that tested the null hypothesis o f no difference in time to first
hospitalization, between patients receiving treatment in accordance with the JNC 4
guidelines versus patients that were treated in accordance with the more restrictive
JNC 5 guidelines. Figure 3.13 shows the Kaplan Meier survival curves for the two
groups. From visual observation, it is apparent that up till about 6 months the two
curves are virtually indistinguishable, however after that, there is a gap between the
two curves. The survival curve for JNC 4 has a higher survival time compared to the
JNC 5 curve. The PROC LIFETEST results confirmed that neither the Log-Rank
statistic nor the Wilcoxon statistic was significant, hence the null hypothesis of
homogenous curves is not rejected and it is concluded that the two curves are not
different from each other. Hence the survival curves (time to hospitalization) for
patients treated in accordance with the JNC 4 guidelines and the JNC 5 guidelines,
are similar.
To study the effects o f covariates on the hazard o f having a hospitalization, the Cox
Proportional Hazards Model was employed. The dependent variable here was time to
72
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
first hospitalization (in months) starting from the keydate. Table 3.5 shows the
output o f the Cox Proportional Hazards regression model.
Table 3.5: Cox regression model output
Dependent variable: Time to first hospitalization
Variable Parameter
Estimate (S.E)
Hazard
Ratio
P
Value
Age 0.01 (0.01) 1.02 0.15
Gender (female) 0.61 (0.23) 1.85 0.01
Received treatment in accordance with the
guidelines (compared to not receiving
treatment in accordance with guidelines)
-0.58 (0.22) 0.55 0.008
Diagnosis of:
ARR -0.53 (0.60) 0.58 0.37
Athero 0.50 (0.37) 1.66 0.17
Hblock -0.21 (0.31) 0.80 0.48
CHD -0.52 (0.10) 0.59 0.60
Thrombo 0.22 (0.24) 1.25 0.36
CHF 0.39(0.32) 1.48 0.22
CAD 0.005 (0.49) 1.05 0.90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In the above table, females had a 85% higher hazard o f having a hospitalization than
men. Patients receiving treatment in accordance with guidelines had a 45% lower
hazard o f a discontinuation in therapy compared to patients that did not receive
therapy in accordance with guidelines.
Tim e to discontinuation o f antihypertensive therapy
The variable for time to discontinuation o f antihypertensive therapy starting for the
keydate, is defined in the methods section and was calculated accordingly. The
Kaplan M eier curves shown in figure 3.14 display the differences in survival curves
for time to discontinuation, differentiating between patients that got treated
according to the JNC 4 guideline and patients that got treated according to the JNC 5
guidelines. The figure shows that patients receiving treatment according to the JNC 4
guidelines have a survival function that declines more rapidly until about 180 days
and then the curve crosses over to having a slower decline in survival, compared to
the curve for patients receiving treatment according to the JNC 5 guidelines. The
PROC LIFETEST results showed that neither the Log-Rank statistic nor the
W ilcoxon statistic was significant, hence the null hypothesis of homogenous curves
was not rejected and it was concluded that the two curves were not statistically
different from each other.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.13: Plot o f Survival Function for the Two Groups: Dependent
Variable=Time to discontinuation in therapy (ttdis): periodyes=JNC 4:
Patients treated in accordance with JNC 4 guidelines; periodyes=JNC 5:
Patients treated in accordance with JNC 5 guidelines
I 100
a
^ 0.75
s
o
0.50
0.25
5 o.oo
E
3
C/3
0 50 250 300 100 150 200 350 400
ttdis
S T R A T A ------ periodyes= JNC 4 O O O censored periodyes=JNC 4
periodyes= JNC 5 O O O censored periodyes=JNC 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The effects o f covariates was tested in another model, the Cox proportional hazards
model and the covariates tested in this model were identical to the covariates tested
in the above models. The output for this model is shown is Table 3.6.
Table 3.6: Cox regression model output
Dependent variable: Time to discontinuation o f antihypertensive therapy
Variable Parameter
Estimate (S.E)
Hazard
Ratio
P Value
Age -0.007 (0.009) 0.99 0.45
Gender (female) 0.01 (0.13) 1.02 0.90
Received treatment in accordance
with the guidelines
0.79 (0.17) 2.21 0.16
Diagnosis of:
Arr 0.11 (0.42) 1.12 0.77
Athero -0.20 (0.28) 0.81 0.47
Hblock 0.09 (0.18) 1.09 0.60
CHD 0.16(0.47) 1.18 0.72
Thrombo 0.22 (0.17) 1.25 0.21
CHF 0.003 (0.20) 1.00 0.98
CAD 0.43 (0.30) 1.53 0.18
76
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For patients that received treatment in accordance with guidelines, none o f the
factors were associated with a hazard o f discontinuation. Patients that received
treatment in accordance with guideline recommendations did not have a significantly
different hazard for discontinuation o f therapy compared to patients that did not
receive treatment in accordance with guideline recommendations.
Time to Augmentation in Antihypertensive Therapy
After calculating the time to augmentation variable as described in the methods
section, less than one percent o f the population experienced an augm entation in
therapy, prior to a switch in therapy. Starting from the key date, whichever event
came first, switch in therapy or an augmentation in therapy, the event was
respectively recorded as time to switch or time to augmentation. In the event o f a
switch taking place prior to the augmentation, the possibility o f an augmentation in
therapy after the switch was not ruled out. However, augmentation in therapy after a
switch could be attributed to the patient not responding positively to the switched
medication. In this case the augmentation in therapy cannot be described as a result
o f the initial starting therapy that the patient received, and since this analysis is
focused on the initial therapy and it’s outcomes, the augmentation in therapy
occurring after the switch was not used in calculations.
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Specific Aim 3
The third goal o f this research was to measure the economic im pact o f following
JNC guidelines. The total health care costs including cost o f prescription drugs,
hospital services, ambulatory care, long-term care and other costs were totaled and
used as the dependent variable for this analysis. Total health care costs for ‘cross
over’ patients (dually eligible for Medicare and Medicaid) were calculated as
described in the methods section. The distribution of the cost variable showed that
the data was skewed, hence a transformation on the cost variable was needed to
conform to the normality assumptions. A log transformation o f the cost variable
rendered it normal, hence the log transformed variable was used for further analysis.
The health care costs were calculated and compared as 1. 2 and 3 year costs for
reasons described earlier.
Controlling for Selection Bias
To estimate the economic impact o f following guidelines, a Heckm an Sample
Selection Model was estimated as described in the methods section. In the first stage
a Probit model estimated the probability o f receiving treatment or not receiving
treatment, and the resulting IMR that accounts for the selection bias, was plugged
78
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
into the outcome equation in stage 2 o f the sample selection model as an additional
explanatory variable.
After applying the Heckman sample selection model to the models for 1.2 and 3
year health care costs, it was found that the Inverse Mills Ratio was not significant in
the second step for either o f the above three models, thus it was concluded that there
was no selection bias between patients receiving treatment in accordance with
guideline recommendations and patients not receiving therapy in accordance with
guidelines.
Hence, the classic least squares regression analysis model was fitted to study the
factors associated with 1. 2 and 3 year health care costs. The smearing estimate
procedure was applied to the regression coefficients to obtain the predicted health
care costs in order to compare the difference in the magnitude o f health care costs
between the two patient groups. The results o f the least square regression analysis
arc presented in table 3.7, while the predicted health care costs (after applying the
smearing procedure) are presented in table 3.8.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.7: Factors associated with 1,2 and 3 year health care costs - Comparing
patients that received fist-line treatment in accordance with guidelines
Independent Variables 1 Year Costs
N=1572
Estimate (s.e.)
2 Year Costs
N=1323
Estimate (s.e.)
3 Year Costs
N=1090
Estimate (s.e.)
Intercept 8.01 (0.24) 8.67 (0.26) 9.25 (0.38)
Gender: Female=l -0.19* (0.07) -0.15* (0.07) -0.17* (0.08)
One pharmacy -0.57** (0.07) -0.64** (0.07) -0.74** (0.08)
Health care costs in the
pre-period
0.00004**
(0.000004)
0.00003**
(0.000004)
0.00003**
(0.000004)
Patient got treated in
accordance with
guideline (1 = Yes)
-0.26 **
(0.07)
-0.18* (0.07) -0.12 (0.08)
Co-morbidities following
keydate
Arr 0.48** (0.11) 0.56** (0.11) 0.54** (0.12)
Athero 0.55** (0.17) 0.57** (0.17) 0.63** (0.18)
CAD 0.32* (0.15) 0.38* (0.15) 0.35* (0.15)
CHD 0.36** (0.12) 0.38** (0.12) 0.38** (0.13)
CHF 0.24** (0.12) 0.34** (0.12) 0.32** (0.12)
Thrombo 1.15** (0.29) 0.98** (0.32) 0.88** (0.31)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The results from the above table showed that the factors associated with lower health
care costs were gender and patients receiving all their antihypertensive medications
form one pharmacy. More importantly, for 1 and 2 year health care costs, patients
that received treatm ent in accordance with guideline recommendations had
significantly low er health car costs compared to patients not receiving therapy in
accordance with guideline recommendations. Patients with comorbidities have
significantly higher health care costs. Table 3.8 compares the magnitudes o f the
adjusted (smeared) health care costs between the two groups o f patients.
Table 3.8: Adjusted 1, 2 and 3 year Health Care Costs
1 Year Costs
S
2 Year Costs
S
3 Year Costs
S
Following ANY
guidelines 8,396* 16,925* 25,048
Not following either
guidelines 11,642 21,164 32,847
* p < 0 .0 5
The above table shows that patients receiving first line antihypertensive therapy in
accordance with guideline recommendations had approximately $3,000 lower in 1
year costs and $4,000 lower in 2 year costs com pared to patients who did not receive
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
treatment in accordance w ith guideline recommendations. The3 year costs were also
lower but not significant.
Similar to the above analyses, comparisons in health care costs w ere made between
patients receiving first line antihypertensive therapy in accordance with JNC 4
guideline recommendations and patients receiving therapy under the JNC 5
recommendations.
The adjusted health care costs were compared between the two groups o f patients
and the results are presented in table 3.9
Table 3.9: Adjusted 1, 2 and 3 year Health Care Costs (JNC 4 versus JNC 5)
1 Y ear Costs
S
2 Year Costs
S
3 Y ear Costs
S
Following JNC 4
guidelines 8,488* 16,520* 24,560*
Following JNC 5
guidelines 12,132 23,413 41,122
* p < 0.05
The results from the above table shows that the 1,2 and 3 year health care costs are
significantly lower (significance tested in the regression model) for patients that got
treated in accordance with the JNC 4 guideline recommendations for first line
anti hypertensive therapy com pared to the JNC 5 guideline recommendations.
82
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 4: DISCUSSION
The purpose o f this research was to study the physician adherence to the national
guidelines for hypertension i.e. the guidelines from the Joint National Committee on
Detection, Evaluation and treatment o f Hypertension. The main purpose o f the six
reports that had been published to date, was to provide a better understanding of
hypertension,, and in this process, provide the health care provider, mainly the
physician a tool to optimally treat the hypertensive patients, with special prescribing
guidelines for special populations such as the elderly, women and young children.
The physician is solely responsible for furnishing the antihypertensive patient with
the necessary antihypertensive medications and this is a challenging task with
numerous antihypertensive agents available in the market. Also, considering the
possibility that the patients suffer from concomitant disease conditions, along with
having hypertension, especially the elderly population, it is vital that the physicians
take these co-morbidities into account when prescribing the appropriate
antihypertensive medication for the patients. The JNC guidelines serve as a
prescribing aid for physicians since they lay out specific indications and
contraindications for the use o f certain antihypertensive medication classes.
In the past, researchers have studied the physician adherence to the JNC guidelines
and the consensus on the results is that in general, physicians are not prescribing in
accordance with guidelines. However, none o f the previous research focused
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
specifically on the elderly population and also, nobody looked at the consequences o f
following guidelines versus not following guidelines on patient outcomes. In this
study, a unique methodology called the interrupted time series analysis was used to
study the change in prescribing behavior o f the physicians over a period o f time that
encompassed the publication o f the JNC 4 and the JNC 5 guidelines. The patient
outcomes were also compared between patients that received treatment in
accordance with the guidelines and those that did not.
From the interrupted time series analysis results, it is apparent that the prescribing
behavior o f physicians seem to be somewhat consistent with guideline
recommendations, especially seen in the increase in CCB prescribing and decrease in
the b-blocker prescribing in the JNC 4 period and vice versa in the JNC 5 period.
The JNC 4 and the JNC 5 guidelines had differing recommendations for first line
antihypertensive therapy. While the JNC 4 guidelines published in 1988
recommended the use o f either o f the four classes o f antihypertensive agents for first
line antihypertensive therapy (diuretics, p-blockers, ACE inhibitors and CCBs). the
JNC 5 guidelines recommended the use o f only diuretics and p-blockers as first line
antihypertensive therapy. The increased utilization o f CCBs and the decreased
utilization o f the P-blockers in the JNC 4 period could be due to the fact that in the
late 1980s, newer antihypertensive medications such as CCBs started becoming
popular due to their decreased side effects profile compared to the b-blockers which
had been around for a longer time. In the JNC 4 period, the physicians probably
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
substituted CCBs for b-blockers and hence there as an increased utilization in the
CCBs in the JNC 4 period as witnessed in the interrupted tim e series analysis.
Simultaneously the utilization o f b-blockers decreased in the JNC 4 period and
increased in the JNC 5 period. This pattern suggested that as the CCBs started
getting more popular due to the fewer side effects compared to b-blockers,
physicians their prescribing from b-blockers to CCBs. however, due to lack o f
evidence o f the impact o f CCBs on long term morbidity and mortality, physicians
switched back to prescribing P-blockers.
Although the results from the interrupted tim e series analysis do suggest that
physicians’ prescribing patterns tended to change with the change in guideline
recommendations, it would be an invalid assumption to make that physicians do
follow guideline recommendations in general.
The hypothesis for this research, was that since elderly patients have special
requirements in lieu o f their age and the fact that in general they are at higher risk for
having comorbidities hence utilizing a greater chunk of the health care finances, the
physicians would be m ore cautious in their prescribing behaviors. However,
considering the fact that in the short term, ACE inhibitors and CCBs have shown to
have a more positive im pact on blood pressure control, it seems to be the norm that
physicians prescribe these medications irrespective of the guideline
recommendations o f the JNC 5 report which specifically state that ACE inhibitors
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and CCBs have not shown long term cardiovascular benefits for hypertensive
patients, while there is clinical evidence that diuretics and p-blockers have proven
long term cardiovascular benefits.
There could be many reasons that could explain why physicians do not adhere to
guideline recommendations. It is obvious that when guidelines are published, not all
physicians are aware o f their existence. It is up to their personal need for knowledge
about current literature to keep abreast o f the literature and in part the responsibility
for disseminating information also lies with health care policy m akers that need to
explore ways o f getting the physician to be aware o f practice guidelines and
recommendations. In this analysis, sufficient time lag was modeled into the trend
analysis to see if there was some time that was required for physicians to start
following guidelines, from the date o f publication of these guidelines.
Another reason why physicians don’t follow guidelines is that they continue to
follow their practice patterns that they have developed through their experiences.
Also, although physicians are aware o f guideline recommendations, they do not
agree with the prescribing recommendations and hence do not follow them.
This research went a step further to see if receiving treatment in accordance with the
guidelines recommendations did in fact have any impact on patient outcomes and
total health care costs. Time to switch in medication class from keydate. time to
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hospitalization from keydate and time to discontinuation o f therapy from keydate
were used as key measures o f patient outcomes.
When controlling for covariates. patients receiving treatment in accordance with the
guideline recommendations had a significantly lower hazard o f switch compared to
patients that were not receiving treatment in accordance with the guideline
recommendations. Higher confidence should be placed in multivariate analysis, since
it accounts for other covariates, hence it can be concluded from the above analysis
that following guideline recommendations did have a positive impact on at least one
patient outcome.
Patients that received treatment in accordance with guideline recommendations in
fact had a longer survival time to first hospitalization and discontinuation, compared
to patients that did not receive treatment in accordance with guidelines. This is a
positive finding and when controlling for covariates, it was confirmed that patients
that received treatment in accordance with guideline recommendations did in fact
have a lower hazard for a hospitalization compared to patients that did not receive
treatment in accordance with guideline recommendations. When controlling for other
covariates, it was found that females had a significantly higher hazard o f a
hospitalization compared to males.
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The results from the above analyses repeatedly show that patients receiving
treatment in accordance with the guidelines have more beneficial patient outcomes
compared to patients not receiving treatment in accordance with guideline
recommendations.
The sample selection model indicated that the unobserved variables did not pose a
problem o f selection bias that could have been present when making treatment
selection, It was found that controlling for sample selection bias and other observed
variables, patients that received treatment in accordance with the guidelines had
significantly lower health care costs when compared to patients that did not receive
treatment in accordance with the guidelines. However, caution should be used when
interpreting the results o f the sample selection model. It must be noted that due to the
limited availability o f variables in the dataset, all variables that could possibly impact
the treatment selection in the first stage o f the sample selection model could not be
included in the step I equation. Hence the IMR calculated from step 1 and
introduced in step 2 might not have captured the impact o f the unobserved variables
on the health care costs. Thus, although the results show that there was no sample
selection bias in the model, it must be interpreted with caution. The addition o f more
observable variables in step 1 o f the model might make an enormous impact on
whether the IMR is significant in the second stage. This issue needs to be further
studied in detail.
88
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
From the above results it is also apparent that when the JNC opened its first line
therapy recommendations to include ACE and CCBs in their fourth report, patients
appeared to have better outcomes and significantly lower health care costs. It
appeared that there was some substitution in prescribing medications during the JNC
4 period and the JNC 5 period, and the physicians were changing their prescribing
patterns. This information is important to health care administrators since they can
devise appropriate means o f educating physicians on the beneficial impacts o f
following guidelines versus not, considering the limitations o f the above analyses.
Limitations of the Research
There are certain limitations to this research, as in any other retrospective claims
database analysis. First o f all, clinical information on the patient is unavailable such
as blood pressure readings and other laboratory values, which would have been
beneficial in identifying if the patient’s high blood pressure had actually been under
control had he been treated in accordance with the guidelines. In the absence o f these
clinical measures, conclusions cannot be made regarding how useful guideline
recommendations are in controlling hypertension in the population, however that was
not the goal o f this study.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The results could have been embellished by the availability o f physician
characteristics. Specifically physician age, specialty and practice setting could have
shed more light on the differing prescribing behaviors o f physicians. While making
conclusions about the prescribing patterns from the above results, this research
assumes that physicians follow the same prescribing patterns for all patients in the
population. However, physician prescribing patterns might not be consistent in
patient populations other than the Medicaid population, hence caution should be
taken while generalizing these results to the population.
Another important limitation is the assumption o f newly diagnosed hypertensive
patients. The entire analysis was based upon the initial antihypertensive therapy that
the patient received and whether the physician followed the guideline
recommendations for initial therapy. The patient was assumed as a newly diagnosed
hypertensive if there was no use of an antihypertensive medication in the 6 month
period prior to the keydate.
The inability to control for selection bias in the time to event analysis (time to
switch, hospitalization and discontinuation in therapy) poses another limitation in
this research. As mentioned earlier, censored observations, that are common in
survival analysis data, are not accounted for in models that control for selection bias.
Hence caution m ust be used when interpreting the results o f the proportional hazards
models.
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Policy Implications o f this research
Hypertension is a costly disease to treat with about $US 30 billion spent in 1997 in
treatment (AMA) including pharmaceutical products which cost about $7 billion.
The results o f these research lean towards indicating that following guidelines has a
beneficial impact on patient outcomes and also the total health care costs are
significantly different for patients that did get treated according to guidelines
compared to those that did not. However, there is evidence from clinical trials that
the use o f P-blockers and diuretics have proven long term beneficial cardiovascular
effects as opposed to ACE inhibitors and CCBs that have not (reference). Keeping
this in mind, it is important that physicians are aware o f the impacts o f following
guidelines versus not, and with this information in mind, prescribe appropriate
medications to the hypertensive patient in order to control hypertension and
eventually protect them from potentially avoidable cardiovascular morbidities. Our
results indicate that patients who developed cardiovascular morbidities in the period
following the keydate had higher health care costs compared to those that did not.
This in itself is an important indication that treatment o f cardiovascular
comorbidities is expensive and can lead to big health care expenditures. The sample
size used in this analysis was rather small. There were only about 750 elderly
patients in each arm when comparing patients that received treatment in accordance
with the guidelines to patients that did not. With a larger sample size, an emergence
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o f long term cost savings for patients that received treatment in accordance with
guidelines seems likely. If health care policy makers can design effective
intervention methods to educate physicians on the existence o f guidelines and their
recommendations, it will greatly benefit not only the patients who are the ultimate
beneficiaries o f optimal health care, but also aid in containing health care
expenditures. The aim o f guideline development is to achieve optimal hypertension
control and simultaneously minimize the costs o f managing hypertension.
Hypertension being an insidious disease has far reaching consequences. This study
shows that physicians are not prescribing in accordance with guideline
recommendations, and consequently some patient outcomes are being negatively
impacted. Hence it is imperative that in order to optimally treat the hypertensive
elderly in the US population, health administrators need to continually target
physicians and intervene in ways that will make physicians more aware o f the
repercussions o f inappropriate prescribing in the elderly hypertensive population.
Future Research
Although the results of this study have given us an understanding o f the physician
adherence to the national hypertension guidelines, the JNC guidelines, and the
impact this has on patient and econom ic outcomes, further research into this matter is
required. This study did not include demographics of the physicians prescribing the
antihypertensive medication and in the future, by incorporating these characteristics
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
into the above models, it will help us to understand the physician related factors that
impact their prescribing behavior and in turn their disposition to follow guideline
recommendations. Also, the inclusion o f a sample o f elderly patients and their
physicians from across the country into the study will allow us to understand the
geographical differences if any in the antihypertensive prescribing patterns. This
research looked at physician adherence to guidelines and the impact only on the
elderly patients. Future research might look at other special populations such as
minority races and children and adolescents. This would encompass different types
o f patients and comparisons could be drawn in physician prescribing patterns
between these different samples o f patients.
The JNC published it's sixth report (JNC 6) in November 1997, which w as an
updated version o f the JNC 5, published in 1993. Although the treatm ent
recommendations for initial antihypertensive treatment remained the sam e, it might
be interesting to look at any changes in physician prescribing patterns w ith the
advent o f the newer guidelines.
Conclusions
This research was an ambitious attempt to study the adherence o f physicians to the
JNC guideline recommendations for first line antihypertensive therapy and its impact
on patient outcomes and health care costs. Although there are certain lim itations to
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the dataset and the research design, there appears to be som e change in physician
prescribing practices with the changing guideline recommendations. Patient
outcomes are better and health care costs are lower for patients who are treated in
accordance with guideline recommendations. In the past, researchers have shown
that there has been little or no change in physician prescribing practices after the
publication o f the JNC 5 guidelines. However, this study shows that there are in fact
some changes in prescribing trends over the different time periods. This study has
gone a step further in evaluating the patients outcomes and the economic outcomes
o f following guideline recommendations and has shown that there is a beneficial
impact o f following guideline recommendations. Whether these beneficial impacts
are solely due to the adherence o f physicians to the guidelines or whether other
external factors are also involved in this outcome, needs to be further investigated.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
Alderman MH, Madhavan S, Cohen H. Antihypertensive drug therapy: the effect o f
JNC criteria on prescribing patterns and patient status through the first year. A M J
Hyperten 1996;9:413-418
Burt VL, Cutler JA, Higgins M. Trends in the prevalence, awareness, treatment and
control o f hypertension in the adult US population: data form the Health
Examination surveys, 1960 to 1991. Hypertension. 1995;26:60-69
Chalmers J. Treatment guidelines in hypertension: current limitations and future
solutions. Journal o f Hypertension 1996,14 (suppl4):S3-S8.
Frohlich ED. Treatment guideline in the USA: hypertension in the elderly. B rJ Urol.
1998;81:26-28
Crown WH, obenchain RL, Englehart L, Lair T, Buesching DP, Croghan T. The
application o f sample selection models to outcomes research: The care o f evaluating
the effects o f antidepressant therapy on resource utilization. Stcitis.Med.
1998,17,1943-1958
Furberg CD, Psaty BM, Meyer JV. Nifedipine: Dose related increase in mortality in
patients with coronary heart disease. Circulation 1995;274:620-625
Joint National Committee: the 1988 report o f the joint national committee on
detection, evaluation and treatment o f high blood pressure. Arch Int Med.
1988;148:694-698
Kaplan NM, Gifford RW. Choice o f initial therapy for hypertension. JAMA.
1996;275:1577-1580
Kozma CM , Pannone RD, Clayton G, Hedblom EC. Evaluation o f initial drug
selection for newly medicated hypertensives at the W estinghouse Electrical
Corporation. J Managed Care Pharm 1999:505-509
Lenfant C. JNC guidelines: is the message getting through? JAMA. 1997;278:1778-
1779
Mehta S, W ilcox CS, Schulman KA. Treatment of hypertension in patients with
comorbidities: results from the study of hypertensive prescribing practices (ShyPP).
Am J H ypertens. 1999;12:333-340
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
M oser M. Why are physicians not prescribing diuretics more frequently in the
management o f hypertension. JAMA. 1998;279:1813-1816
Moser M. Evolution o f the treatment o f hypertension from the 1940s to JNC V. Am J
Hypertension. 1997;10:2S-8S
Moser M, Okin P. Grellet K. Long term management o f hypertension-private
practice experience. NYJ 1980;80:1102-1106.
National center for health statistics. Health, United States, 1996-97, an d Injury
Chartbook, Hyattsville. Md: National Center fo r health statistics; 1997
Pahor M, Guralnik JM , Corti M-C. Long term survival and use o f antihypertensive
medications in older subjects. J Am G eriatrS oc 1995,43:1191-1197.
Psaty BM, Heckbert S, Koepsell TD et al. The risk o f myocardial infarction
associated with antihypertensive drug therapy. JAMA 1995;274:620-625
Psaty BM, Smith NL, Siscovick DS et al. Health outcomes associated with
antihypertensive therapies used as first line agents. JAMA. 1997;227:739-745
Ramsey S, Neil N, Sullivan S, Perfetto E. An economic evaluation o f the JNC
hypertension guidelines using data from a randomized controlled trial. J Am Board
Fam Pract. 1999;12:105-114
Siegel D. Lopez J. Trends in antihypertensive drug use in the United States: do the
JNC V recommendations affect prescribing? JAMA. 1997;278:1745-1748
TamowL, Rossing P, Gall MA, Nielsen FS, Parving HH. Prevalence o f arterial
hypertension in diabetic patients before and after the JNC V. D iabetes Care.
1994;17:1247-1251
The Fifth report o f the Joint National Committee on the detection, evaluation and
treatment of high blood pressure. Arch Intern M ed 1993;153:154-183
The Sixth report o f the Joint National Committee on the detection, evaluation and
treatment o f high blood pressure. Washington DC;US Department o f Health and
Human Services, 1997;NIH publication nO. 98-4080
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix 1: Summary o f Previously Published Literature
Study Objective Methods Outcome
measured
Results Conclusion/
comments
Siegel Describe 200 most ff Kx tor fro m 1992 to JNC V has little
et al. anti-htn frequently each 1995, use o f influence on
prescribin prescribed dosage CCBs and ACE prescribing patterns.
1997 g patterns drugs for form inhibs increased
for 1992- 1992-1995 Cost and use o f No information
1995 from estimates B_blockers and whether patient is
35,000 o f meds diuretics newly diagnosed.
retail dispensed decreased.
Estimate pharmacies Costs
impact o f Costs o f CCBs underestimated.
JNC V o n and ACE inhibs didn't include
cost increased physician visits.
clinic visits etc.
Alder Evaluate 500 Iype ot Pattern ot initial Shift in initial therapy
man et impact o f untreated drug used drug therapy choice had little
al JNC IV hypertensiv as initial changed from impact on short term
and JNV es divided therapy P_blockers and patient outcomes
1996 V on into three diuretics to
prescribin periods Changes ACE inhibs and No long-term
g patterns based on in BP, CBS, form outcome studies.
and short tim e o f clinical period I to II economic
term entry I:pre chem istry and evened out implications not
patient JNC IV. 1 1 : measures. across 4 groups studied, small sam ple
outcomes post JNC clinic in period III size to be able to
IV, III: attendance draw reliable
during the and BP response conclusions regarding
developme dropout sim ilar across side effects or lab
nt o f JNC rate the three time use.
V
periods
Dropouts
dim inished over
time
Mehta To Survey Most Physicians not Overall, physicians
et al analyze m ailed to frequently likely to follow did not adhere to JNC
the 500 prescribed guidelines when guidelines when
1999 prescribin primary drug class. prescribing to treating patients with
g practices care Treatm ent blacks, elderly. comorbidities
o f prim ary physicians recommen
care dations for Younger Small sample size.
physicians Average pats with physicians more Survey responses do
for age o f com orbidi likely to comply not respect true
patients physicians ties with guideline prescribing patterns
with =46.2 years recommendatio since no data on
comorbs. ns actual prescriptions.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Study Objective Methods Outcome
measured
Results Conclusion/
comments
Ramsey Determine Randomize Drug and Initial costs to D iuretics and
et al clinical controlled medical achieve and P_blockers are most
costs o f trial data care costs. maintain efficient form clinical
1999 managing from including hypertension and econom ic
hypertensi Treatm ent drug control lowest standpoint in the
on when o f Mild therapy. for treatm ent o f
following Hypertensi complianc chlorthalidone. hypertension.
the JNC V on Study e and cost followed by follow ing JNC
guidelines (TO M H S) o f acebutolol. guidelines
switching amlodipine and
4 agents after enalapril. After
studied: therapeuti hypertension
enalapril. c failures control is
am lodip. achieved costs
acebutololc for managing is
hlorthal- lowest for
idone cholrthalidone
and highest for
acebutolol
lam ow To Iw o Arterial Arterial JNC V criteria led to
et al compare cohorts o f Blood hypertension an increase in arterial
the white pressure increased after hypertension in
1994 prevalenc patients the JNC V. in diabetic patients.
e o f studied both groups of
arterial over a patients.
hypertensi three-year
on in period.(ID
patients DM,
with n=957.
IDDM mean
and age=40;
NIDDM NIDDM ,
before and n=549.
after the mean
JNC V age=60)
Kozma To 1,218 Antihyper Percentage ot JN C V
et al. evaluate newly tensive initially recom m endations had
the initial diagnosed medicatio prescribed not significantly
1999 drug hypertensiv n usage medications are im pacted choice o f
selection es as follows: b- initial therapy by
for newly Prescripti blockers: 20%, 1996.
medicated Paid on diuretics: !(%,
hypertensi m edical/ph expenditur calcium channel The study shows that
ves, after arm acy es blockers: 23%, initial therapy as
JNC V claim s ACE inhibitors: recom m ended by
recommen from 1994- 23%, others: guidelines are used
dations 96 were 15%. only 39% o f the time.
analyzed.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
The construct validity of the Health Utilities Index in patients with chronic respiratory disease in a managed care population
PDF
A methodology to identify high -risk patients with diabetes in the California Medicaid populations (Medi -Cal)
PDF
Effects of a formulary expansion of the use of SSRIs and health care services by depressed patients in the California Medicaid program
PDF
Assessing the cost implications of combined pharmacotherapy in the long term management of asthma: Theory and application of methods to control selection bias
PDF
Compliance study of second-generation antipsychotics on patients with schizophrenia
PDF
Controlling for biases from measurement errors in health outcomes research: A structural equation modeling approach
PDF
A new paradigm to evaluate quality-adjusted life years (QALY) from secondary database: Transforming health status instrument scores to health preference
PDF
Time dependent survival analysis of Kaiser Permanente/USC pharmacists' consultation intervention study
PDF
Assessment of prognostic comorbidity in hospital outcomes research: Is there a role for outpatient pharmacy data?
PDF
Variations in physician practice patterns for eye care under the National Health Insurance of Taiwan
PDF
The influence of drug copay change on drug utilization: The case of small-firm employees in California
PDF
Out -of -pocket health expenditures by older adults in relation to age, race, and insurance
PDF
Physician profiling and clinical pathways: Combining the tools to change physician resource utilization
PDF
Studies on the mechanism and regulation of the transport of 5-fluorouracil (5FU) into tumor cells
PDF
Using network perspective to examine the organization of community -based elder care systems across four communities
PDF
Regionalization of fire protection and emergency medical aid services: A comparative case study analysis of economic and socio-political impacts
PDF
Prescription drug profiles as health risk adjusters in capitated payment systems: An applied econometric analysis
PDF
Essays on organizational forms and performance in California hospitals
PDF
Care management for the uninsured: A force field analysis of the business case
PDF
The political economy of procyclical fiscal policy in Mexico, 1970--1988
Asset Metadata
Creator
Kamath, Tripthi Vasant
(author)
Core Title
Physician adherence to national hypertension guidelines in an elderly Medicaid population
Degree
Doctor of Philosophy
Degree Program
Pharmaceutical Economics and Policy
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
health sciences, health care management,Health Sciences, Pharmacy,OAI-PMH Harvest
Language
English
Contributor
Digitized by ProQuest
(provenance)
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c16-196044
Unique identifier
UC11339094
Identifier
3065803.pdf (filename),usctheses-c16-196044 (legacy record id)
Legacy Identifier
3065803.pdf
Dmrecord
196044
Document Type
Dissertation
Rights
Kamath, Tripthi Vasant
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA
Tags
health sciences, health care management
Health Sciences, Pharmacy