DEVELOPING A COST-EFFECTIVENESS ANALYSIS OF
ANASTROZOLE AND MEGESTROL ACETATE FOR THE
PALLIATIVE TREATMENT OF BREAST CANCER
by
Gordon Harper
B.Sc.(Pharm.), University of British Columbia, 1991

THESIS SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE IN COMMUNITY HEALTH

©

Gordon Harper, 1998

THE UNIVERSITY OF NORTHERN BRITISH COLUMBIA
March 1998
1

_ __ _ _ _ __Al1 rights reserved. This work may not be

•

reproduced in whole or in part, by photocopy or
. other means, without the permission of the author .

Cost-Effectiveness Analysis

3

ABSTRACT

This thesis demonstrates the applicability of pharmacoeconomic analysis for two
drug therapies used in the palliative treatment of breast cancer. The
pharmacoeconomic evaluation explores the techniques that would be used in
performing a cost-effectiveness analysis of anastrozole and megestrol acetate
therapies. The thesis discusses the methods for collecting, interpreting, and
extrapolating clinical and economic data. The analytical techniques include:
measuring the effectiveness of breast cancer therapies, applying clinical decision
analysis methods to produce a decision tree, reviewing clinical trial data and
examining the clinical findings with the numbers needed to treat approach,
applying survival data to a pharmacoeconomic evaluation, the valuation of
resource items, constructing and using a costs and outcomes table, and calculating
the cost-effectiveness ratio and the incremental cost-effectiveness ratio.

Cost-Effectiveness Analysis

4

TABLE OF CONTENTS

APPROVAL
ABSTRACT

2
3

ACKNOWLEDGMENTS

6
7
8

Chapter 1
Introduction
Health and economics.
Pharmacoeconomics of cancer therapies.
Breast cancer therapies.

9
9
10
14
18

Chapter 2
Pharmacoeconomic Analysis
Developing the cost-effectiveness analysis.
Components of the cost-effectiveness ratio.
Modeling and decision trees.

25
25
25
28
30

Chapter 3
Clinical Data
Pharmacoeconomics and the randomized clinical trial.
An overview of anastrozole and megestrol acetate randomized clinical
data.

37
37
37

Chapter 4
Developing the Pharmacoeconomic Analysis
Study population.
Method of evaluation.
Time horizon.

47
47
47
48
49

Chapter 5
Quality of Life and Survival
Survival analysis.
Constructing survival curves.
Modeled data.
Evaluating survival curves.

51
51
52
55
60
61

LIST OF FIGURES
LIST OF TABLES

38

Cost-Effectiveness Analysis

5

Chapter 6
Organizing and Presenting Clinical and Economic Data
Decision model.
Determining the cost of therapy.
Costs and outcomes table.

65
65
65
68
72

Chapter 7
Summary
Limitations.
Future research.

77
77
82
84

References

86

Cost-Effectiveness Analysis

6

LIST OF FIGURES

·~
··•

Figure 1.

Treatment of advanced breast cancer .. .... ............ .... .............. .. .. ......... 21

Figure 2.

A decision tree for the treatment of acute abdominal pain .. ...... .... .... .32

Figure 3.

A recursive decision tree of anticoagulant therapy ............................ 34

Figure 4.

A three state Markov model.. ................................................ .. ...... .. ... 35

Figure 5.

Time lines for twelve fictitious study participants .... .........................52

Figure 6.

Survival curves for anastrozole and megestrol acetate ....................... 55

Figure 7.

Survival curve for twelve fictitious study participants .... ................... 59

Figure 8.

Comparing three sets of survival curves .................. .. ........................ 60

Figure 9.

Main branch of the recursive decision tree for anastrozole 1-mg od
and megestrol acetate 40-mg qid therapies ................ .......... .. ............ 65

Figure 10. Five year recursive decision tree for anastrozole and megestrol
acetate ....... .................. ................... ......... ................ ........... .... ......... ... .66

Cost-Effectiveness Analysis

7

LIST OF TABLES

Table 1.

The incidence of adverse effects for anastrozole and megestrol
acetate ................................................................................................. 39

Table 2.

Summary of survival information for anastrozole and megestrol
acetate for clinical trials 0004 and 0005 ............................................ .41

Table 3.

Analysis of clinical trials 0004 and 0005 mortality data for anastrozole
and megestrol acetate ......................................................................... 44

Table 4.

Life table of 12 fictitious participants ................................................ 56

Table 5.

Resource item list and sources of cost information ............................ 68

Table 6.

Cost valuation of continuous therapy resource items .........................70

Table 7.

Costs and outcomes table for anastrozole 1-mg od ............................ 73

Table 8.

Costs and outcomes table for megestrol acetate 40-mg qid ............... 74

Cost-Effectiveness Analysis

8

ACKNOWLEDGMENTS

I wish to express my sincerest thanks to Dr. David Fish for his advice and
guidance in the preparation of this thesis. I would also like to thank Dr. Zafar
Kayani, Dr. Bruno Zumbo, and Professor Geoffrey Weller for their technical
assistance.
Mr. Tuhin Rahman, ofthe B.C. Pharmacoeconomics Initiative, deserves a
special mention as he spent many hours sharing his expertise and knowledge with
me and his patience and understanding assisted me with the development of the
methodologies.
In addition I would like to thank Jacqui for giving me the encouragement
and support to complete this project, and my mother, Sylvia Harper, for her love
and inspiring me to write this thesis.

t
Cost-Effectiveness Analysis

9

Chapter 1
Introduction
This thesis will illustrate the methodologies that might be used in a
pharmacoeconomic evaluation of two drugs for the palliative treatment of breast
cancer. The technical aspects of the pharmacoeconomic evaluation will follow the
Canadian Coordinating Office of Health Technology Assessment, Guidelines for
Economic Evaluation of Pharmaceuticals: Canada (CCOHTA, 1997). The
technical areas will include: the perspective of the evaluation, the measurement of
costs and outcomes, decision analysis, the analytical time horizon, data sources,
uncertainty of data, survival analysis, discounting of costs and outcomes,
constructing a cost and outcomes cohort table, and the application of sensitivity
analyses (CCOHTA, 1997).
The purpose of this thesis is to develop the analytical techniques needed to
perform a cost-effectiveness analysis of anastrozole and megestrol acetate
therapies. Hopefully, by developing the analytical techniques, more
pharmacoeconomic evaluations on palliative breast cancer therapies will be
performed. High quality pharmacoeconomic evaluations will play an important
role in encouraging effective, efficient, and equitable drug therapy selection.
Chapter 1 will provide a short discussion on health and economics,
pharmacoeconomics, and palliative breast cancer treatments. Chapter 2 will

Cost-Effectiveness Analysis

10

review the components of the cost-effectiveness ratio, the incremental costeffectiveness ratio, and the structuring of clinical data into a decision model.
Chapter 3 will describe the problems with using randomized clinical trial data in a
pharmacoeconomic evaluation. Chapter 4 will review the handling of survival
information. Chapter 5 will describe the steps involved in collecting, tabulating,
and presenting cost data. A costs and outcomes table for combining clinical and
economic data will be provided. Chapter 6 will summarize the analytical
techniques and the limitations of these techniques, in performing a costeffectiveness analysis of anastrozole and megestrol acetate therapies.
Health and economics.
In western societies, much of the consumption of health resources is that
which relies on new medical technologies. Economists are advocating the practice
of evidence based medicine to ensure that the provision of health care and the use
of new medical technologies is equitable, effective, and efficient. It is imperative
that effective and economical therapies are promoted in order to ensure the users
of our health care system receive the best possible medical care for the resources
consumed.
Allocative decision making explores ways in which one can analyze the
costs and the benefits of new medical technologies in order to best govern the
distribution of scarce resources. Decisions concerning which medical technologies

Cost-Effectiveness Analysis

11

should be supported by our health care system can be evaluated by using
analytical methods taken from the disciplines of epidemiology, social sciences,
medicine, economics, and pharmacy.
The method used to evaluate the clinical and economic value of drugs is
pharmacoeconomics. Pharmacoeconomics involves the systematic and
comprehensive analysis of costs and benefits of drug therapies in order to
understand their impact on the health care system (Bootman, Townsend &
McGhan, 1996).
Federal and provincial agencies, hospitals, pharmaceutical manufacturers,
and universities perform pharmacoeconomic evaluations. A federal agency, the
Patented Medicine Prices Review Board (PMPRB), uses clinical and economic
information in order to determine whether drugs coming to market offer
innovative therapies and whether the cost of those therapies is reasonable. The
PMPRB uses pharmacoeconomic information to guide the establishment of
Canada wide drug pricing policies. Drug pricing policies are established so that
drugs are priced at what is believed to be a fair market value (PMPRB, 1996).
Provincial Ministries of Health use pharmacoeconomics to determine what
medications should be available on provincial drug formularies and what
medications should be available for government reimbursement. In British
Columbia (B.C.), the Therapeutic Initiative and Pharmacoeconomic Initiative

Cost-Effectiveness Analysis

12

evaluates drugs based on clinical and economic information. The Therapeutic and
Pharmacoeconomic Initiatives membership is composed of physicians,
economists, health policy experts, and pharmacists. The recommendations from
these initiatives are used by the B.C. Ministry of Health to manage the provincial
drug formulary and establish a reference based drug pricing program. The
reference based pricing program sets a maximum reimbursable drug price for
medications that are considered therapeutically equivalent.
While the Therapeutic and Pharmacoeconomic Initiatives evaluate drugs
for the B.C. provincial drug formulary, the B.C. Cancer Agency and B.C.
hospitals do not utilize such initiatives in their formulary review process. The
B.C. Cancer Agency and B.C. hospitals perform few detailed pharmacoeconomic
studies as they do not have the resources, or the expertise, that the Therapeutic
and Pharmacoeconomic Initiatives have.
Canadian hospitals and cancer agencies also perform pharmacoeconomic
evaluations. Hospitals and cancer agencies use pharmacoeconomics to measure
the cost-effectiveness of drug therapies. Hospitals and provincial cancer agencies
try to support drug therapies which offer clinically significant outcomes at a cost
that the institutions can afford. Drug therapies that do not offer clinical outcomes
that are not significantly better or where the costs are not less than other equally
effective treatments, would not be considered.

Cost-Effectiveness Analysis

13

Although there are numerous groups performing pharrnacoeconomic
evaluations, it is only recently that methodological guidelines have been
developed. Analysts and users of pharrnacoeconomic studies need to have a clear
understanding of the methodological principles behind these evaluations. A clear
understanding of the methodological principles will ensure that the
pharrnacoeconomic studies produce reliable and valid information.
In order to facilitate high quality pharrnacoeconomic studies the Canadian
Coordinating Office for Health Technology Assessment (CCOHTA) has
developed pharrnacoeconomic guidelines (CCOHTA, 1997). CCOHTA
guidelines divide the pharrnacoeconomic evaluation process into: a
pharmaceutical review of therapy, defining the perspective of the analysis,
applying a methodology (e.g. cost-effectiveness analysis), identification and
presentation of costs and consequences, measuring costs and consequences,
discounting future costs and consequences, performing sensitivity analysis, and
presenting the analysis.
Even with high quality data, a pharrnacoeconomic evaluation should only
be used as an aid in the decision making process. The primary purpose of
pharrnacoeconomics is to provide the decision maker with clinical and economic
evidence that describes the outcomes of a particular intervention and how it will
affect the distribution of resources. Pharrnacoeconomics should not "replace hard

Cost-Effectiveness Analysis

14

thinking, careful consideration, good judgment and common sense" (CCOHT A,
1994, p. 1).
Pharroacoeconomics of cancer therapies.
The screening and the treatment of cancer are under scrutiny by health
administrators and government bodies because of the large amounts of resources
that are being consumed. It is estimated that, in industrialized countries,
expenditures on cancer treatments may account for as much as six percent of the
gross national product (Jonsson, Clausen & Hansen, 1995). With the increasing
numbers of elderly, with the growing exposure of the public to carcinogenic
agents, and with improvements in the diagnostic testing of cancerous tumors, the
prevalence and the costs of cancer related illnesses will continue to rise (Desch,
Hillner, Smith & Retchin, 1993).
Recent attention has focused on how the best possible care can be
provided to cancer patients while controlling rising costs. Hospitals, governments
and cancer agencies are each trying to find ways to operate as efficiently as
possible. Health professionals are being forced to find ways of reducing the
consumption of scarce resources while still providing quality care.
While ensuring efficiencies in cancer treatments is paramount, researchers
are concerned that there are apparent inequities in funding between different types
of cancer treatments. Jonsson et al. (1995) believe that noncurative cancer

Cost-Effectiveness Analysis

15

treatments are not receiving their fair share of resources. They believe that society
has an obligation not to consider incurable cancers as a failure and limit its
funding because of prognosis. They go on to state:
.. .we would like to point out that economic allocations and the distribution
of resources in the treatment of cancer must be determined by the possible
outcome of treatment; however, outcome and cure cannot be equated. We
would also like to emphasize the conceptual errors that may be introduced
by interpreting cure as 'success' and lack of cure as ' failure ' ,
interpretations that are too narrow. Stabilisation of incurable cancer for a
period of time, together with the provision of a good quality of life, can
itself be considered a satisfactory goal until curative therapy becomes
available (p. 280).
Examples of incurable cancers that could receive improved therapies
include testicular cancer, some hematological cancers, and metastatic breast
cancer (Jonsson et al., 1995). The treatment ofthese cancers should be evaluated
in the same way as curable cancers and the allocation of funds should be based on
sound therapeutic and economic analysis. Pharmacoeconomic evaluations of the
treatment of incurable cancers that could provide effective cytoreduction therapies
need to be promoted.
Given this dilemma of equitable resource allocation and the fact that, when
working with finite resources, one must concede some health benefits over others,
determining how to choose where funds should be made available is a difficult
task. Pharmacoeconomic analysis of cancer therapies should consider measuring
outcomes that are meaningful to the specific disease states that are being studied.

Cost-Effectiveness Analysis

16

For example, event free or disease free survival in curable cancers maybe a more
appropriate outcome measure than quality of life. Event free or disease free
survival is important when cytotoxic treatments in curable cancers are relatively
short in duration compared to the remaining life years and the possibility of
success (e.g. , cure) outweighs the negative health effects received from the toxic
chemotherapy (Jonsson et al., 1995).
For incurable cancers, quality of life and life-years gained are the best
outcome measures for most pharmacoeconomic evaluations. Quality of life is
important for two reasons. First, research has shown that cancer patients find that
severe pain and confinement to a bed is regarded as worse than death (Hall &
Tattersall, 1995). Second, when death is inevitable the toxic effects from the
treatment may not be worth the gain in survival. Life-years gained would be an
appropriate outcome measure for incurable cancer drug therapies where patients
feel the extra time attributed to the treatment outweighs the adverse effects
(Jonsson et al., 1995).
Although certain outcome measures are more applicable to specific types
of cancers, deciding which measurement to use may not be easy. For example, the
clinical data needed for the evaluation may not be collected, or the data used in a
comparative analysis is not sensitive enough to detect differences between the
treatment states being studied (Drummond, Stoddart & Torrance, 1993). In many

Cost-Effectiveness Analysis

17

pharmacoeconomic evaluations, the outcome measure is quality of life and is
considered a cost-utility analysis (CUA). A CUA allows the analyst to apply a
utility score according to the quality of life that the patient experiences. In CUA,
the outcome measure is expressed as a product of utility and length of survival. In
its final form, a CUA is defined by the cost per quality adjusted life years (or
QAL Y) (Bootman et al. , 1996).
In the present thesis, the outcome measure of life-years gained will be
used to examine which drug therapy, anastrozole or megestrol acetate, is the more
cost-effective in treating advanced hormonally responsive breast cancer.
In cost-effective analysis (CEA), the outcome measure could be any
clinically significant quantifiable effect that is common to the treatments being
studied (Bootman et al., 1996). The costs and the effectiveness components are
evaluated separately and then expressed as a cost-to-effectiveness ratio (e.g. , cost
per life-years gained). A cost-effectiveness ratio is determined for each treatment
group and the difference between the two cost-effectiveness ratios is expressed as
an incremental ratio. The components ofthe cost-effectiveness ratio and the
specific calculations to arrive at the ratio will be provided later in the paper.
The survival outcome measure for each treatment group can be determined
by utilizing clinical data from drug trials and/or from using epidemiological data
from cancer registries. The basic strategies underlying the analysis include:

Cost-Effectiveness Analysis

18

estimating the survival for each treatment group at specific points in time,
determining the significance of the survival values, applying the survival rates for
a hypothetical cohort of patients in each treatment group, running the cohort for a
specified time, and performing a final count of patients for each group (Lee,
1980).
Unfortunately, it is only in the last few years that quality
pharmacoeconomic studies of cancer therapies have been performed. Many of the
methods used, such as measuring quality of life and the counting of cancer
treatment related costs, need further development. Also, while clinical trials of
cancer therapies provide valuable information regarding tumor cell response or
time to treatment failure, researchers still have difficulty ascertaining the true
value of such information (Rubens, 1996). Hopefully, pharmacoeconomic
analysis of cancer treatments will promote the use of effective therapies and
ensure that funding priorities are equitable.
Breast cancer therapies.
Breast cancer is the most common cancer found in women in Europe and
North America. The highest incidence rates in the world are in Hawaii, California,
and British Columbia (Veronesi, Goldhirsch & Yarnold, 1995). In British
Columbia (B.C.) the estimated age standardized incidence ofbreast cancer is 121
cases per 100,000 women (National Cancer Institute of Canada, 1996), while in

Cost-Effectiveness Analysis

19

Japan the estimated age standardized incidence is approximately 12 cases per
100,000 women (Veronsi et al., 1995). Women with breast cancer usually live
longer than those with other common types of malignancies, such as lung and
colorectal cancers. In Canada one in nine women will be diagnosed with breast
cancer and one in twenty-five will die from the disease (National Cancer Institute
of Canada, 1996).
Breast cancer is rare in women less than 30 years of age but, as age
increases, the incidence rate also increases. Breast cancer usually appears as a
slow growing painless mass until detected by physical examination or
mammography. The final diagnosis is made by microscopic examination of the
breast tissue. Treatment is determined by the extent of the disease and the
woman' s age. Treatment factors which are to be considered include: "the extent,
pattern and aggressiveness of the disease, indices of likely hormone sensitivity,
such as steroid receptor status, and menopausal status" (Rubens, 1996, p. 2).
When there is no sign of the cancer involving peripheral sites the most common
treatment is a lumpectomy, modified radical mastectomy, or a total mastectomy.
After surgical resection further cytoreduction, such as radiation and/or
chemotherapy, may be performed.
When the cancer has advanced and spread to the lymph nodes and other
sites, the disease is said to be metastatic and treatment is considered palliative.

Cost-Effectiveness Analysis

20

Veronesi et al. (1995) recommend that the treatment objective for metastatic
breast cancer "should be to increase the total duration of time with no or few
disease related symptoms using the therapy associated with the lowest cost in
terms of side-effects" (p. 1274). Figure 1 illustrates a schematic for the treatment
of advanced breast cancer in women.

Cost-Effectiveness Analysis

I Aggressive visceral disease I

21

Other metastatic

I

Potentially hormone
Yes

No

I

Premenopausal

Postmenopausal

Ovarian ablation
LH-RH agonist
Tamoxifen

Tamoxifen

Response?
I

No

I

Yes

I
Progestogens
Aromatase
inhibitors

I

Cytotoxic chemotherapy

1 Treatment of advanced breast cancer. From "Key Issues
in the Treatment of Advanced Breast Cancer: Expectations and
Outcomes" by R. Rubens, 1996, Pharmacoeconomics. 9(Suppl. 2),
p. 3.
~

I

Cost-Effectiveness Analysis

22

Women with aggressive metastatic visceral breast cancer require
chemotherapy to reduce the rapid progression of the disease. For less aggressive
metastatic breast cancers, treatment is dependent upon whether the tumor' s
mitogenic activity is sensitive to the presence of estrogen. Women whose breast
cancer is sensitive to estrogen are considered to have estrogen-positive tumors.
These women make up the majority of breast cancer patients and have been
shown to have an improved prognosis compared to women with estrogen-negative
status (Veronesi et al. , 1995).
For women with hormone responsive breast cancer, the primary treatment
is altering the hormonal environment of the tumor. The course of therapy depends
on menopausal status. Premenopausal women may undergo ovarian ablation or
drug treatments with luteinizing hormone-releasing hormone (LH-RH) or
tamoxifen. Postmenopausal women who are estrogen-positive will likely benefit
from tamoxifen therapy. Tamoxifen competitively forms an estrogen receptor
complex which blocks the growth stimulatory mechanisms controlled by
endogenous estrogen. However, for some women, receptor mediated blockade
will eventually regress and the tumor(s) will grow (Ruben, 1996).
Women who have responded to tamoxifen therapy and then show signs of
failure, may experience secondary regressions from additional hormonal
manipulation. It is postulated that tumor regression is a result of cellular

Cost-Effectiveness Analysis

23

adaptation by the development of estrogen receptor mutations (Santen, 1996). The
tumor cell may then be able to receive hormonal stimulation by mutating the
receptor so that treatment with estrogen receptor antagonists (e.g. , tamoxifen) are
no longer effective. This may help explain why changing from an estrogen
receptor antagonist, when regression develops, to a medication with a different
mechanism of action may help delay further disease progression.
Third generation aromatase inhibitors may help to prolong survival in
postmenopausal women by inhibiting the biosynthesis of estrogen (Zeneca,
1997a). Although circulating estrogen concentrations are low in postmenopausal
women, tumor cells can receive estrogen from the peripheral aromatization from
fat and muscle tissue and from local aromatization from the tumor (Brodie, 1996).
Therefore, inhibiting estrogen biosynthesis at the peripheral and local level may
help to reduce further tumor proliferation. The clinical trials of these third
generation aromatase inhibitors suggest that they may be useful in first and
second-line therapy (Smith & Henderson, 1996).
Due to the high costs associated with treating breast cancer, and the
clinical uncertainty surrounding the benefits of the new pharmaceutical
technologies, economic constraints are forcing provincial cancer agencies to limit
access to third generation aromatase inhibitors. For example, the British Columbia
Cancer Agency (BCCA) has restricted the use of the new aromatase inhibitor

Cost-Effectiveness Analysis

24

anastrozole to "palliative treatment of hormonally sensitive metastatic breast
cancer in patients who have progressed after tamoxifen and megestrol acetate
treatment and have unacceptable side effects from aminoglutethimide" (BCCA,
1997, p. 1). Pharmacoeconomic analysis of third generation aromatase inhibitors
will help to explore the costs and the effectiveness of this type of new therapy and
assist provincial cancer agencies and Ministries of Health to decide how these
new drugs fit into their current cancer treatment policies.

Cost-Effectiveness Analysis

25

Chapter 2
Pharmacoeconomic Analysis
This chapter will describe the technical aspects involved in performing a
pharmacoeconomic analysis. The chapter will discuss the cost-effectiveness ratio
and the incremental cost-effectiveness ratio. Components that make up these
ratios will be highlighted. The chapter will describe three clinical decision
modeling methods for the organization of pharmaceutical data.
~ the cost-effectiveness analysis.

The technical aspects involved in the development of a cost-effectiveness
analysis can be divided into three categories. These are: the development and
structure of the pharmacoeconomic problem; the assessment of methodological
assumptions; and the evaluation of costs (Weinstein, 1981 ).
In the development of the research problem, the pharmacoeconomist needs
to define the population of interest, the treatment options for the specific disease
states, the risks and benefits associated with each treatment, and to assess the
availability of data. The pharmacoeconomist must identify which intervention will
be the comparator and determine what outcomes will be used to measure the
drug' s effectiveness. During the structuring phase of the evaluation a decision tree
model serves as a useful method of structuring and describing the various
treatment options and their outcomes.

Cost-Effectiveness Analysis

26

The second part of a cost-effectiveness analysis is understanding how
assumptions made during the early stages of the evaluation can affect the results.
Economists recognize that a comprehensive evaluation of costs and health effects
often leads to a large number of assumptions being introduced into the study
(Weinstein, 1981 ). These assumptions create uncertainties that impact on the
significance ofthe final results. Testing the significance ofthe assumptions is
carried by sensitivity analysis.
.

·~

.

.

Boatman et al. ( 1996) state that "sensitivity analysis is a method of
determining whether the conclusion of an economic evaluation changes when the

•••

/

value of one variable is varied as all other variables are held constant" (p. 70).
Sensitivity analysis is accomplished by first determining which variables contain
uncertainty. Those variables with uncertainty are then assessed for the magnitude
of the uncertainty and the study is re-run with the revised values (Siegel,
Torrance, Russell, Luce & Weinstein, 1997). The results ofthe re-run study are
compared to the original to determine whether the uncertainty had any impact on
the findings.
The third part of the cost-effectiveness analysis is the measuring and
reporting of all relevant costs and comparing those costs with the effectiveness
portion of the analysis. Costs are determined according to the perspective used.
The perspective is the point of view from which the analyst conducts the

II

Cost-Effectiveness Analysis

27

evaluation. Pharmacoeconomic perspectives include the societal, the institutional,
and the patient's. An analysis which uses the societal perspective would consider
the consumption of resources for all members of society. Institutional and patient
perspectives are more narrow in the evaluation of costs since they encompass only
those costs which affect the institution or individual patient.
Costs can be broken down into direct, indirect and intangible categories.
Direct costs are those that involve the transfer of money. For example, direct costs
include capital costs, drug costs, laboratory costs, labor costs, and patient "out of
pocket" expenses (Drummond, Stoddart & Torrance, 1993). Indirect costs are
those that do not involve the exchange of money but that affect the use of other
resources. For example, indirect costs include lost leisure time and time from
work. Intangible costs are costs where no money is exchanged and the effect on
the consumption of resources is either difficult to measure or, in dollar terms, is
unquantifiable. For example, intangible costs include psychological loss and pain
or suffering.
In pharmacoeconomic analysis, a comprehensive list of costs is gathered
for each treatment. Final costs are then tallied and compared to the effectiveness
of each therapy. The costs and the effectiveness for each treatment are represented
as a ratio of costs to effectiveness. The cost-effectiveness ratio for this study is
expressed as costs (in Canadian dollars) versus the effectiveness outcome (e.g. ,

Cost-Effectiveness Analysis

28

life years gained). Further economic evaluations may then be performed such as
the calculation of an incremental cost-effectiveness ratio.
Components of the cost-effectiveness ratio.
The components ofthe cost-effectiveness ratio can be expressed as follows
(Weinstein, 1977, 1980):

(1)

where: in the

~

c R x is the direct costs of treatment (e.g., drug costs,

drug administration costs, costs associated with laboratory tests), ~ c s£ is costs
associated with the drug' s side effects (e.g., nausea, vaginal bleeding),
and !J. C Mo rh is the savings associated with the prevention of morbid events (e.g.,
brain metastases). In the denominator !J. Y is the change in effectiveness (e.g.,
change in life years), !J. YsE is the adjustment for side effects, !J. Y Mo rh is the
adjustment for a reduction in morbidity, and !J. Ysymp is the adjustment for the
relief of symptoms (Weinstein, 1980).
In Equation 1 the net cost calculated in a cost-effectiveness analysis is
represented by the numerator and the net effectiveness is determined by the
denominator. Adjustments in the effectiveness component are determined by the

Cost-Effectiveness Analysis

29

perspective and the methodologies used in the analysis. For example, modifying
the effectiveness component of the ratio by applying adjustments for side effects,
morbidity and the relief of symptoms is usually considered a cost-utility analysis.
Cost-utility analysis allows the researcher to account for less tangible aspects of a
person' s well being. A person's well being is measured as a utility value. Utility
values are applied to the effectiveness portion of Equation 1. Some methods used
to determine utility values are the standard gamble, time trade off, healthy years
equivalent, and willingness to pay techniques (Bonneterre, Schraub, Lecomte &
Mercier, 1996).
CCOHTA (1994) recommends that, once the cost-effectiveness ratio has
been calculated, the ratios for each drug therapy should be compared and
expressed in incremental terms. The incremental cost-effectiveness ratio is the
ratio of the difference between the net costs in the numerator and the difference
between the net effectiveness in the denominator (Detsky, 1990). The incremental
cost-effectiveness ratio is expressed in Equation 2.

(2)

Cost-Effectiveness Analysis

30

where: in the numerator C 1 is the net cost for treatment 1 and C 2 is the net cost
for treatment 2. In the denominator E 1 is the net effectiveness for treatment 1 and
E 2 is the net effectiveness for treatment 2.

The incremental cost-effectiveness ratio provides valuable information as
it "reveals the cost per unit of benefit of switching from one treatment strategy
(usually already in operation) to a new strategy" (Detsky, 1990, p. 151). For
example, the cost-effectiveness ratio for drug A is $1 ,800 per 2.2 years of life, and
for drug B is $1 ,500 per 1. 9 years of life. The incremental ratio is calculated as
follows:

c $1,800-$1,500
Incremental-=
= $1 ,000 per life year gained (3)
2.2years- 1.9years
E

Since drug A costs $1,000 for each life-year gained over drug B, Drug B is more
cost-effective.
Modelin2 and decision trees.
The most difficult aspect of carrying out a pharmacoeconomic analysis is
first determining what information is required and second deciding how to
organize the clinical and economic data. Pharmacoeconomic information will
need to be structured into an analytical framework that will break the evaluation

Cost-Effectiveness Analysis

31

into manageable units. The manageable units will first need to be defined,
organized, and then presented in a comprehensive format.
Decision models are useful analytical tools for structuring a
pharmacoeconomic evaluation. Decision models organize clinical and economic
information into "its component parts so that they can be analyzed individually
and then recombined in a systematic way" (Weinstein, Fineberg et al., 1980, p. 4).
Decision trees, recursive decision trees, and Markov models are types of decision
models used in structuring clinical and economic information. Decision models
are used not only for systematically organizing pertinent data into a structured
framework, but they can also assist in determining the likelihood of various
treatment events occurring.
The decision tree is an analytic tool that tracks the options and the
outcomes for each treatment group at a particular point in time. The decision tree
contains decision nodes, chance nodes and treatment paths. The tree is structured
from left to right starting with a decision node. The timing of the nodes and the
paths which connect the nodes, correspond to the clinical course of the event
(Weinstein, Fineberg et al., 1980).
Figure 2 illustrates a branch of a decision tree for the treatment of a patient
with acute abdominal pain. Decision nodes are points where alternative actions
are selected and are represented as squares. The first square in Figure 2 is a point

Cost-Effectiveness Analysis

32

where a physician must decide to intervene immediately or wait six hours and
monitor the patient's abdominal pain. The second square is a point where the
physician must decide whether or not to operate. The lines connecting the first
two decision nodes are treatment paths.

Perforated

Operate

Inflamed

NSAP

Survive Survival
Die

Death

Survive Survival
Die

Death

Survive Survival
Die

Death

Decide Now
Perforated
Do Not Operate

Inflamed

Survive Survival
Die
Survive
Die
Survive

'Wait Six Hours

NSAP

Die

Death
Survival
Death

Death

Figure 2. A decision tree branch for the treatment of a patient with acute
abdominal pain. NSAP refers to nonspecific abdominal pain. From Clinical
Decision Analysis (p. 15) by M. C. Weinstein, Fineberg et al. , 1980, Toronto: W.
B. Saunders Company.
Chance nodes are points where "one of several possible events beyond the
control of the decision maker may take place. It is represented in a decision tree as
a small circle" (Weinstein, Fineberg et al, 1980, p. 14). As shown in Figure 2,
following the decision of whether or not to operate, the possible events are either

Cost-Effectiveness Analysis

33

a perforated appendix, an inflamed appendix, or nonspecific abdominal pain
(NSAP). Consequences of the three chance nodes branch out to two other chance
nodes. Those chance nodes are survival and death.
A recursive decision tree is similar to a standard decision tree in that it
outlines the events and outcomes in a systematic diagram. However, a recursive
decision tree also provides a means of identifying events that are repeated
throughout the clinical course. By including repeated events in the recursive tree
the clinical problem can structured according to the time intervals (e.g., a month,
or a year) at which these events occur. The cycling of the repeated events in a
recursive decision tree helps to track their reoccurrence at various points in time.
Figure 3 shows a recursive decision tree. The chance events from
anticoagulant therapy are represented by the first three nodes as bleed, embolus
and no event. Bleed and embolus events can be fatal or non-fatal. Since bleed,
embolus and no event may occur more than once, the tree structure repeats itself.
Period 1 and Period 2 contain the same events in the same order but occur in a
different time frame. However, complex analytical problems that are structured
into a decision tree can become too large and impractical for modeling purposes.
For example, Figure 3 would become too confusing if it extended past period 2.

~

·~

Cost-Effectiveness Analysis

34

r-'F:..:."T
"'-A;;;,.
C - ; O["-:.AD
BLEED

SLEEO

....UN -F A "TAL

~

A.NTfCOAG

FATAl.
E M60LUS J.
TNON-FATAL

1

DEAD

FATAL
ANnCOAG

OEAO
. -ANT"ICOAG

....rroco-.o

NO EVENT
FAT.A&...

IANTlC OAG

BLEED

~

NON-FAT"'- .ANTICOAG

Of!·AD

T NON -FATAL
FA.T AL

DEAD

E MBOLUS.J..

1

ANTICOA O

~

I'NTICOAG

NO EVENT

B EEO

.....
~

ANTICOAG

- - f OEAO

r""'""""'-TO NON- FATAL

A.NTICOAG

' - - - - - - - - _ _ _ __.A..___ _ _-. _ _ _ _ _,/

v

PERIOD 1

v

PERIOD 2

Figure 3. A recursive decision tree of anticoagulant therapy. Period 1 and period 2
is comprised of a main branch starting with bleed, embolus, and no event. From
"Markov Models in Medical Decision Making: A Practical Guide," by F. A.
Sonnenberg and J. R. Beck, 1994, Medical Decision
~ 13, p. 324.
Another way of structuring repeated health events is by using a Markov
model (Briggs & Sculpher, 1997). The Markov model is similar to a recursive
decision tree in that it structures events in chronological order and provides a
means of documenting the cycling between repetitive events. The Markov model,
however, limits the number of health states to only three or four events. More than
four events would make the model cluttered and confusing.

~

~

Cost-Effectiveness Analysis

35

Pauker and Kassirer (1987) describe a Markov model as a small set of
health states with transitions between the states. The likelihood of changing from
one health state to another is called transition probability. Figure 4 illustrates a
"three-state Markov Model" where patients reside in any one ofthe three states
and change to other states at different points in time (Beck & Pauker, 1983, p.
421).

Figure 4. A three state Markov model illustrating transitional probabilities from
time ito i+ 1 for the health states well, ill, and dead. From "The Markov Process in
Medical Prognosis" by, J. R. Beck and S. G. Pauker, 1983, Medical Decision
Makin2. 3(4), p. 421.
The three health states are well, ill, and dead. The model allows the
patients to be in any one of the three states depending on whether their clinical
problem requires them to be in that state (Beck & Pauker, 1983). The patients
distribute to other health states according to the transition probabilities after a
fixed amount of time. The transition probabilities are shown adjacent to the lines

Cost-Effectiveness Analysis

36

flowing from one state to another. For example, the transition probability from the
well state at time i, to the health state ill at time i+ 1, is designated as Pwi.
Whatever decision model that is used, the most effective method will
incorporate all significant clinical events into a structured tool. Choosing the
correct model, and the information to include in the model, should be based on a
thorough understanding of the events and outcomes in each treatment group.
Knowing how those events and outcomes occur over time, and which events and
outcomes are clinically and economically significant, will provide the researcher
with an understanding as to which model would be best suited to the
pharmacoeconomic evaluation.
Once event or transitional probabilities are determined, a cohort
simulation of a defined number of patients can be performed on the decision
model. If the model is time dependent, such as with a recursive decision tree or
Markov model, the patients would be distributed among the various events or
health states after a set period of time (Sonnenberg & Beck, 1993). The simulation
would be run until the time horizon is completed. The results of a simulation of a
cohort can be tabulated and compared to the cost data.

Cost-Effectiveness Analysis

37

Chapter 3
Clinical Data
There are many possible sources of clinical data for use in a
pharmacoeconomic evaluation. Clinical data comes from randomized clinical
trials, retrospective drug information from various databases, epidemiological data
from the literature or databases, or from professional organizations or agencies.
The primary source of clinical data for the anastrozole/megestrol acetate example
used in this thesis could come from randomized clinical trials. This chapter will
review the anastrozole/megestrol acetate randomized clinical trial data and how it
would be used in a pharmacoeconomic evaluation. Although actual clinical data
will be utilized in this chapter, hypothetical values will be used in later chapters
for describing pharmacoeconomic calculations and tables.
Pharmacoeconomics and the randomized clinical trial.
A pharmacoeconomic evaluation relies on interpreting and extrapolating
results from randomized clinical trials (RCTs). A RCT is required to demonstrate
the safety and efficacy of new drugs. A RCT also provides a means of comparing
new drug treatments to existing therapies. Drug therapies which have
demonstrated at least an incremental advantage over existing therapies "usually
require randomized comparison trials to demonstrate convincingly statistically
significant improvement[s]" (Kaufman, 1993, p. 2801). The steps involved in

Cost-Effectiveness Analysis

38

performing a RCT are: assignment of participants into treatment groups,
assessment for defined endpoints or outcomes, analysis of the results, and
interpretation of the findings (Riegelman & Hirsch, 1989). A RCT that has been
carefully designed and has demonstrated a significant improvement in drug
therapy is the basis for undertaking a pharmacoeconomic analysis since it is the
source of data for the effectiveness data that is required.
Many of the recent advances in the treatment of progressive breast cancer
have come from RCT data which have helped to identify safe and effective
therapies (Kaufman, 1993). While RCTs have been "capable of identifying
effective new therapies and eliminating ineffective, unnecessary, or harmful
therapies," pharmacoeconomic evaluations have not kept pace with recent RCT
information (Kaufman, 1993, p. 2801).
An overview of anastrozole and me2estrol acetate randomized clinical

In order to determine whether a pharmacoeconomic analysis is justified, a
close evaluation of the clinical findings is necessary. The assessment of the RCT
data should consider the strengths and weaknesses of the findings and should
consider that clinical trial data may not reflect real-life drug utilization patterns
(Clemens et al. , 1995). Clemens et al. comment on the problems of applying RCT
data to pharmacoeconomic evaluations.

Cost-Effectiveness Analysis

39

Clinical trials are the primary source for efficacy data at approval. Clinical
trials will generally be powered based on the primary clinical endpoint(s).
Such trials may be under-powered for secondary end-points including
resource use and cost data (p. 172).
Table 1 shows drug tolerability results from the RCT data for anastrozole
and megestrol acetate (Zeneca, 1997b).

Table 1
The Incidence of Adverse Effects for Anastrozole and Me2estrol Acetate
Adverse Event Group
Gastrointestinal Disturbance
(nausea, vomiting, diarrhea)
Hot Flushes
Edema
Thromboembolic Disease
Vaginal Dryness
Weight Gain

Anastrozole
1-mg od (n=262)
%
n
(29.4)
77
33
19
9
5
4

(12.6)
(7.3)
(3.4)
(1.9)
(1.5)

Anastrozole
I 0-mg od (n=246)
%
n
(32.9)
81
29
28
4
3

10

( 11.8)
(11.4)
(1.6)
(1.2)
(4.1)

Megestrol acetate
40-mg qid (n=253)
n
%
54
(21.3)
35
35
12
2
30

(13 .8)
(13.8)
(4.7)
(0.8)
(11.9)

~ From "Arimidex: a Significant Advantage in the Only Reliable Parameter -

Survival," by Zeneca, 1997, Product Information Leaflet, Zeneca Pharma,
Ontario.

The table shows that, overall, the number and types of adverse effects are
similar for both drugs. There are, however, a few significant differences. In the
anastrozole 10-mg group, women experienced greater gastrointestinal effects than
the megestrol acetate group. However, in the megesterol acetate 40-mg qid group,
women experienced greater weight gain than both of the anastrozole groups

Cost-Effectiveness Analysis

40

(Budzar et al, 1996). Zeneca (1997a) reported that the megestrol acetate group had
a greater number of participants who withdrew from the study for intolerable
adverse effects than either of the anastrozole groups. The withdrawal rates due to
adverse effects for the megestrol acetate 40-mg qid group was 4.0 percent, in the
anastrozole 1-mg od group 2.7 percent, and in the anastrozole 10-mg od group 3.3
percent. However, there were no statistically significant differences in withdrawal
rates between the three treatments (Budzar et al., 1996).
Critical appraisal and further analysis of the clinical data should be
considered before proceeding with a pharmacoeconomic evaluation. It would be
disheartening to find that the work of performing a pharmacoeconomic evaluation
was wasted because the results reported were based on invalid RCT data or the
new treatment offered no additional clinical benefit. For example, in the
anastrozole 1-mg and 10-mg daily (od) and megestrol acetate 40-mg four times
daily (qid) comparisons the phase III survival data indicates that the anastrozole
10-mg od treatment is less effective than the anastrozole 1-mg od treatment and
also exhibits a greater number of adverse effects. A pharmacoeconomic evaluation
of anastrozole 10-mg od is therefore not worth pursuing.
The RCT data presented by Zeneca suggested that anastrozole "is well
tolerated and as effective as megestrol acetate for the treatment of postmenopausal
women with advanced breast cancer" (Budzar et al., 1996, p. 2000). But before

Cost-Effectiveness Analysis

proceeding with the pharmacoeconomic analysis a review of the findings is
needed.
Table 2 highlights the survival differences for the anastrozole and
megestrol acetate treatments after the median 31 month follow-up period.
Table 2
Summary of Survival Information for Anastrozole and
Clinical Trials 0004 and 0005
Phase III Trial Number
0004
Number of patients who died(%)
2-year survival rate
Median time to death (months)
0005
Number of patients who died(%)
2-year survival rate
Median time to death (months)
0004 & 0005 Combined
Number of patients who died(%)
2-year survival rate
Median time to death (months)

~

Acetate from

Anastrozole
1-mg od

Anastrozole
10-mg od

Megestrol acetate
40-mg qid

66 of 128
(51.6)
62.0%
29.6

81 of 130
(62.3)
58.0%
25 .7

79 of 128
(61. 7)
53 .1%
26.7

85 of 135
(63.0)
50.5%
24.3

70 of 118
(59.3)
50.8%
24.8

92 of 125
(73.6)
39.1%
19.8

151 of263
(57.4)
56.1%
26.7

151 of248
(60.9)
54.6%
25.5

171 of253
(67.6)
46.3%
22.5

Note. Phase Ill refers to a stage of clinical research where large numbers of
human subjects are given new medications in order to evaluate their safety and
efficacy. From "Arimidex: a Significant Advantage in the Only Reliable
Parameter- Survival," by Zeneca, 1997, Product Information Leaflet, Zeneca
Pharma, Ontario.

41

Cost-Effectiveness Analysis

42

A total of 764 women were randomized into the study, 386 women in the
North American trial (Canada and USA, trial number 0004), and 378 in the
European trial (Europe, Australia and South Africa, trial number 0005) (Zeneca,
1997a). Women were randomized into one ofthe three treatment groups:
anastrozole 1-mg daily, anastrozole 10-mg od, or megestrol acetate 40-mg four
times daily. Women were routinely assessed by physical examination, by bone
scans, and by radiographic examinations. Women were withdrawn from the trial if
they experienced serious adverse effects, unwilling or noncompliant with
procedures, or were found to have significant cancer progression (Budzar et al. ,
1996). Screening, drug tolerability and efficacy assessments were performed on a
routine basis. Efficacy assessments included: time to treatment failure, tumor
response, response duration-and time to death (Zeneca, 1997a, Budzar et al. ,
1996).
The median time to death for anastrozole 1-mg od, anastrozole 10-mg od
and megestrol acetate 40-mg qid was reported to be 26.7, 25.5 and 22.5 months
respectively. The two year survival rate for anastrozole 1-mg od, anastrozole 10mg od and megestrol acetate 40-mg qid was reported to be 56.4, 54.6 and 46.3
percent respectively. The results indicate that the anastrozole 1-mg od group had a
median survival advantage of 4.2 months and 3.0 months over the anastrozole 10mg od and megestrol acetate 40-mg qid groups respectively. The anastrozole 1-

Cost-Effectiveness Analysis

43

mg od group was found to have a two year survival advantage of9.8 and 1.5
percent over the anastrozole 10-mg od and the megestrol acetate 40-mg qid
groups respectively.
Since RCT findings are reported in numerous ways, researchers are
recommending that clinical trial information should be presented in a less
confusing and a more standardized format (Laupacis, Naylor & Sackett, 1992,
Therapeutics Initiative, 1996). Laupacis et al. (1992) suggest that "for clinical
trials, a complementary and simple way to represent the difference between 2
groups emphasizes the clinical effect of the treatment being studied" (p. 12). They
state that "the number needed to treat is a useful method of expressing the efficacy
of a therapy because it incorporates the baseline risk in untreated patients, is easily
calculated (the inverse ofthe absolute risk reduction), and allows an estimate of
the effort and cost associated with the therapy" (pp. 13-14). Clinicians who use
RCT findings that have been reported in terms of relative risk reductions have
been described as being less critical about the results than if they were reported as
absolute risk reductions or number needed to treat (Therapeutics Initiative, 1996).
In the anastrozole versus megestrol acetate case, the absolute risk reduction
is calculated by taking the percent mortality rate for the megestrol acetate 40-mg
qid group minus the percent mortality rate for the anastrozole 1-mg od group. The
absolute risk reduction for women taking anastrozole 1-mg od for a median time

Cost-Effectiveness Analysis

44

the 31 months was found to be 10.2 percent. This means that 10.2 percent fewer
deaths were found to occur with the anastrozole 1-mg od group over the 31 month
treatment period than with the megestrol acetate 40-mg qid group. The number
needed to treat was found to be 10, which indicates that if 10 women were treated
with anastrozole 1-mg od for a median duration of 31 months, instead of with
megestrol acetate 40-mg qid, one death would be prevented. Table 3 outlines these
findings .

Table 3
Analysis of Clinical Trials 0004 and 0005 Mortality Data for Anastrozole and
~
Acetate
Megestrol acetate 40-mg qid
# of patients
Death
Total
253
171
(67.6%)

Anastrozole 1-mg od
# of patients
Total
Death
263
151
(57.4%)

Absolute Risk
Reduction

Number Needed
to Treat

67.6%- 57.4% =
10.2%

100/ 10.2 = 10

The phase III efficacy data, and the numbers needed to treat results, indicate
that a 31 month median treatment with anastrozole 1-mg od demonstrates a survival
advantage over that of the megestrol therapy. Although the RCT efficacy data
"refers to the performance of a drug under highly controlled circumstances," a costeffectiveness analysis ofthese two drug therapies is worth considering (CCOHTA,

Cost-Effectiveness Analysis

1997, p. 20). However, it is important to recognize that the findings cannot be
reliably extrapolated beyond the 31 month median follow-up period.

45

Cost-Effectiveness Analysis

46

Chapter 4
Developin2 the Pharmacoeconomic Analysis
This chapter will discuss the study population and the time horizon for the
pharmacoeconomic evaluation. Since CCOHTA recommends that a societal
perspective should be used whenever an economic analysis is performed, the
anastrozole/megesterol acetate example will be based on that perspective. Using
the societal perspective ensures that all clinical and economic outcomes that have
an impact on the study will be accounted for. Methods for determining the costs to
the individual patient, the family, the hospital or agency caring for the patient, and
to the government will be reviewed. A hypothetical list of direct and indirect
costs, for developing the anastrozole/megestrol acetate evaluation, will also be
discussed. Actual costs will not be gathered and so hypothetical values will be
used for calculation purposes.
Study population.
In this example, the study population will be defined as comprising
women with postmenopausal estrogen receptor positive breast cancer, whose
disease has progressed while on tamoxifen and have switched therapy to either
anastrozole 1-mg od or megestrol acetate 40-mg qid.

Cost-Effectiveness Analysis

47

Method of evaluation.
The method of evaluation will be a cost-effective analysis. The
effectiveness portion of the evaluation will consider the differences in survival
between the two treatment groups. A cost-effectiveness ratio will be calculated
from hypothetical cost and survival data and will be expressed as costs versus life
years (time from start of therapy to death). An incremental cost-effectiveness ratio
will also be described by using the cost-effectiveness information from each
cohort. The incremental cost-effectiveness ratio will describe the cost per life-year
gained for one cohort relative to another.
The multicentered, randomized anastrozole and megestrol acetate phase III
trials, and the follow-up period information, will be used as the primary data
source (Budzar et al. , 1996, Zeneca, 1997a). Although Zeneca' s clinical trial data
contains time to disease progression, best tumor response, duration of response
and duration of stable disease, only survival (i.e., time to death) will be included
in the model. Extracting the retrospective data into the evaluation can be
problematic and so limitations in the anastrozole and megestrol acetate data will
be highlighted. Since phase III RCT data measures the efficacy of the treatments
and not effectiveness, all assumptions made should "be explicitly and thoroughly
tested with sensitivity analysis" (CCOHTA, 1997, p. 20).

Cost-Effectiveness Analysis

48

Other sources of data could include: drug prescribing rates, clinic and drug
costs, and epidemiological information from the British Columbia Cancer
Agency; cost information from the Ministry of Health; drug prescribing and cost
information from British Columbia Pharmacare; clinical and drug cost
information from local hospitals; epidemiological information from Vital
Statistics; and information from literature sources. The specific requirements for
the types of data will become more apparent as the analysis progresses.
Time horizon.
When analyzing survival data, the most appropriate time horizon is to run
the study until all the participants have died (Lee, 1980). In the 31 month median
period approximately sixty-two percent of the women participants died. Most
pharmaceutical company sponsored RCTs are run for the shortest possible
duration while still being able to demonstrate statistically significant outcomes. In
order to meet CCOHTA recommendations the time horizon should be extended
"far enough into the future to capture the major clinical and economic outcomes
related to the treatment under study" (CCOHTA, 1997, p. 18).
Since the ideal time horizon should be greater, modeling techniques are
required to make up the missing data. It is important that modeled data be
considered since the survival benefit for either anastrozole and megestrol acetate
at greater than 31 months is unknown. If the clinical trial cannot be run until all

Cost-Effectiveness Analysis

49

patients are absorbed then there should be good clinical evidence to support a
shorter period.
In order to determine the time horizon for the modeled data, a review of
other clinical data for hormonal therapy of progressive breast cancer is needed.
The long term data from the National Surgical Adjuvant Breast and Bowel Project
have shown that pre and post-menopausal women, with or without estrogen
receptor positive breast cancer, receive significant benefit from taking tamoxifen
for up to five years (Frankel, 1995). Treatment with tamoxifen greater than five
years has be shown to provide no additional efficacy (Frankel, 1995). Applying a
similar five year time horizon for this economic evaluation seems appropriate
since treatment with the first line therapy has been demonstrated to be effective
for a maximum of five years.

Cost-Effectiveness Analysis

50

Chapter 5
Quality of Life and Survival
Most clinicians consider quality of life, or survival, as the most important
criteria for evaluating cancer treatments. Although many cost-effectiveness
analyses study both quality of life and survival together, in this evaluation only
survival will be reviewed. Determining whether to include quality of life in a costeffectiveness analysis of cancer therapies should be based on how much impact
the treatment affects the patient's quality of life and whether the data is available.
Quality of life was not considered in this evaluation for two reasons. First,
since treatment with hormonal therapy for progressive breast cancer results in
fewer side effects than with cytotoxic chemotherapy, the effect on quality of life
would be significantly less than other more toxic therapies. Most evaluations that
have included quality of life have been for surgical or cytotoxic therapies where
the intervention have a significant impact on a person' s well being. Second,
although quality of life data was gathered during the phase III trials, the
information was not published by Zeneca due to inconsistencies in data collection.
In order to develop a better understanding of how anastrozole 1-mg od and
megestrol acetate 40-mg qid impacts on survival, this chapter will examine
concepts and methods used in survival analysis. This chapter will also review how
participants of clinical trials are followed, what the identifiable starting point of

Cost-Effectiveness Analysis

51

treatment is, how losses to follow-up affect the findings, and what methods are
used to organize and interpret the survival results.
Survival analysis.
In Zeneca's phase III trials, women were entered into the study based on
specific criteria: menopausal status, current medical status (e.g., other illnesses),
type of breast cancer, and their current medical treatments. Once women met the
selection criteria, they were then randomized into one of the three groups (i.e.
anastrozole 1-mg od, anastrozole 10-mg od, and megestrol acetate 40-mg qid).
The date of randomization was used as the start date for treatment and was
the point from which the duration of therapy was measured. The participants were
followed until the end of the study or until loss to follow-up. The participants of
the study may die before the end of the study, some may withdraw early, and
some may be alive at the end of the trial (Lee, 1980).
The timing of the various events is central to the analysis of survival.
Figure 5 illustrates how timing of crucial events and the censoring of data can
occur in trials where survival is evaluated. Twelve fictitious subjects, identified as
A through L, in a 36 month study are shown on they-axis of Figure 5. Subjects
who relapsed or died, are labeled with a R at the time of relapse. Subjects labeled
withaL were lost to follow-up during the study. Subjects labeled with C were

Cost-Effectiveness Analysis

52

censored. Censored subjects are those who where still alive at the time the study
ended.
A

B

c

R

-----------------------------------R
R

---------------------------------4C

0

L

E
F

R:

G

~

H

R
: L

R

J
K

~

~

L

L

0

6

12

18

24

30

36

Calendar Month
~
5. Time lines for twelve fictitious study participants. Participant's time
lines ending with R died at that point in their treatment. Participant's time lines
ending with L where lost during the study and those ending with C where alive at
the end ofthe 36 months. From "'Stayin Alive': An Introduction to Survival
Analysis" by, D. L. Streiner, 1995, Can. J. Psychiatry. 40, p. 439-444.

Figure 5 illustrates the problems with presenting the data as median
survival time or relapse rate. Based on the figure, median survival time is
determined from the study population of only those individuals who died during
the trial period. For example, only the patients A, B, C, F, Hand J would be used
in calculating the median survival time (Streiner, 1995). Censored patients would

Cost-Effectiveness Analysis

53

not be included. By excluding censored data the analyst would miss important
information of who did not die and who dropped out of the trial.
Without examining the reason for assigning a patient to the censored
category, the pharmacoeconomist may not be aware that the loss to follow-up
could be related to the treatment the participants received. If the loss of follow-up
was related to the treatment then the study would be underestimating the risk of
therapy (Streiner, 1995). By not including censored participants, the median
survival time calculation underestimates the benefit accruing to those who are
most successful in their treatment (Luke, 1993).
Unfortunately, Zeneca's clinical trial data does not comment on patients
who were lost to follow-up or who lived beyond the follow-up period (Budzar et
al., 1996; Zeneca, 1997a). This is a concern since forty percent of the population
were still alive at the end of the study. Since the number of participants lost to
follow-up is also unknown the exact survival time cannot be determined. Unless
there is more information regarding the censoring of data the pharmacoeconomist
cannot be certain about the reliability of the results. Analysts need to ensure that
survival probabilities are an accurate representation of all of the clinical data.
Using relapse rate as a measure of the efficacy can also be problematic. As
with using median survival times, relapse rates do not take into account
participants who were lost to follow-up or who were still alive at the end of the

Cost-Effectiveness Analysis

54

study period. In order to accurately evaluate the survival data the
pharmacoeconomist needs access to the "raw" RCT data. Raw data should contain
basic information for each participant. The information should include the date of
randomization, the participants' randomized treatment group, the date of death,
and the date of loss to follow-up if it occurred. Information as to the reasons for
loss to follow-up should also be obtained. Once this information has been
gathered the analyst can then interpret the data by running it through one of the
many computer programs that analyze survival information, or the analyst may
organize the data into a statistical table and plot a survival curve. Unfortunately,
the raw survival data was not available from Zeneca and therefore a thorough
analysis of survival could not be performed.
Constructin2 survival curves.
There are two methods for constructing survival curves. The first method
is called the product-limit method and is also known as the Kaplan Meier method.
The product limit method is useful when the study population is less than 100 and
the exact date ofthe event (e.g., death) being monitored is known (Lee, 1980).
The analyst plots the exact point in time each participant died (Kaplan & Meier,
1958). The resulting curve "is simply the proportion surviving at various points in
time" (Coldman & Elwood, 1979; p. 1065).

Cost-Effectiveness Analysis

55

Figure 6 shows the Kaplan Meier survival curves for the anastrozole 1-mg
od and megestrol acetate 40-mg qid therapies (Pritchard, 1997). The curves
represent the overall survival data from the two clinical trials number 0004 and
0005 . Time zero represents the point at which the women where randomized. As
time increases along the x-axis the percentage of women surviving decreases. The
curves provide a graphical representation of the survival data and help to depict
the differences in survival between the two types of treatments.

%Surviving
100

80

60
40
20

Megestrol acetate 40 mg qid

0

1

2
Number of Years

3

6. Survival curves from combined trials 0004 and 0005 . From "Arimidex:
A Significant Advantage in the Only Reliable Parameter - Survival" by, Zeneca,
1997, Product Information Leaflet, Zeneca Pharma, Ontario.
~

Another method of evaluating raw survival data is by using a life table.
The life table method is useful when the time of an event (e.g. , death) cannot be

Cost-Effectiveness Analysis

56

accurately determined. The life table method is also useful when the study
population is large (e.g., greater than 100) or when modeled data is required.
Since the time horizon for Zeneca's randomized clinical trial should be extended,
a life table method would be useful in producing the modeled data (Lee, 1980).
A sample of a life table is shown in Table 4.

Table 4
Life

~ Qf 12 FictitiQys Parti!;;ipants

Interval
(months in
the study)
0-6
6-12
12-18
18-24
24-30
30-36

Number
at risk

Number
died

Number lost
to follow-up

Hazard

Proportion
surviving

12
11
9
6
3
2

I
0
2
2
0

0
2

0.0833
0.0000
0.2353
0.3636
0.0000
0.6667

0.9167
1.0000
0.7647
0.6364
1.0000
0.3333

Cumulative
proportion
surviving
0.9167
0.9167
0.7010
0.4461
0.4461
0.1487

Probability
density
function
0.0764
0.0000
0.1649
0.1622
0.0000
0.0991

NQte. From "'Stayin Alive': An Introduction to Survival Analysis" by, D. L.
Streiner, 1995, Can. J. Psychiatry. 40, p. 441.
Table 4 represents the data from the fictitious study of 12 people
previously shown in Figure 5 (Streiner, 1995). The life table contains the number
of participants at risk, the number who died, the number lost to follow-up, the
hazard (the proportion of participants at risk of dying during a specific interval),
the proportion of participants surviving, the cumulative proportion of participants

Cost-Effectiveness Analysis

57

surviving, and the probability density function (probability of a participant dying
in a specific time interval).
The hazard ratio, shown in the fifth column of the table, is a useful statistic
for comparing the survival experiences of two or more treatment groups. Unlike
survival rate or median survival time, the hazard ratio incorporates all of the
participant data whether censored or not (Mathews & Farewell, 1985). A hazard
ratio is calculated from data over time and, unlike other descriptive survival
statistics, reflects the whole treatment period. A hazard ratio of 0.5 indicates that
the risk of death is one-half the hazard ratio of 1.0. A hazard ratio of 1.0 indicates
there is neither an increased nor decreased risk of death. A hazard ratio of 1.5
suggests a 50 percent increase in risk (Pritchard, 1997).The hazard ratio is
calculated according to equation (4) as follows:
Hazard= #ofpeople who died during the interval
# at risk _ # lost during
~ U_E_
2

(4)

For example, the hazard ratio for the 18 to 24 month interval would be calculated
as:
2
Hazard= - - : - = 0.3636
1
6 2

(5)

Cost-Effectiveness Analysis

58

The proportion of participants at risk of dying for the 18 to 24 month interval is
0.3636, or 36 percent. The only interval with a greater hazard ratio is the 30 to 36
month interval (i.e., hazard ratio is 0.6667).
The hazard ratio accounts for participants who were lost during the followup by assuming that the participants lost were at risk for one-half the interval.
This is why in equations 4 and 5 the number lost to follow-up is divided by two
(Streiner, 1995; Lee, 1980). The proportion of participants surviving is equal to
one minus the hazard ratio. The cumulative proportion of participants surviving in
the first interval is the proportion of participants surviving in the first interval
multiplied by one. The cumulative proportion of participants surviving in the
second interval is the proportion of participants surviving in the second interval
multiplied by the cumulative proportion of participants surviving in the first
interval, and so on for the remainder of the table (Streiner, 1995). The probability
density function is the hazard ratio multiplied by the cumulative proportion of
participants surviving for each interval.
The cumulative proportion of participants surviving is also known as the
survival function and when plotted against time is known as the survival curve
(see Figure 7 for the survival curve of the data represented in Table 4). Plots of
the hazard ratios and the probability density functions can also be graphed against
time. These two types of graphs help the analyst understand how the study

Cost-Effectiveness Analysis

59

population's risk of dying during a particular treatment varies over time. For most
treatments, where survival is the outcome being measured, the risk of death will
increase and decrease over the treatment period (Lee, 1980). Understanding how
the risk of death changes over time provides the analyst with additional
information as to the treatment's effectiveness.
1 .0 ICII>

·~

·;;
......
::0

- - - - - - :

0.8

en

0.8

::c

0.4

=
-•

0

·--------------.'
''
'
'
'

f--------------.

''

e o.2

~

0-

0.0 +--.....--,---.....-...,--.....--,---.....--,---.....-....,-----.---,
24
30
18
36
12
0
6

Number of Months

7. Survival curve for 12 fictitious participants. From '"Stayin Alive': An
Introduction to Survival Analysis" by, D. L. Streiner, 1995, Can. J. Psychiatry.
40, p. 442.
~

I

il

Modeled data.
Extending the survival curve past the available data can be accomplished
by using analytical modeling procedures (CCOHTA, 1997). Estimating future
data points on the survival curve can be carried out by using the survivorship,
probability density, and hazard function information. This information provides
the pharmacoeconomist with an understanding of which analytical model would

Cost-Effectiveness Analysis

60

best estimate the survival distribution. The analyst would choose the model
according to a detailed understanding of the data or by fitting the model to the
survival, hazard or density curves (Gehan, 1975). The validity ofthe model is
checked by testing the goodness of fit between the model and the curves.
Choosing and testing an analytical model is beyond the scope of this paper. The
reader will find Lee (1980), Gehan (1975), and Buyse et al (1984) are key
references for creating and handling modeled data.
Evaluatin2 survival curves.
The analyst should also be aware of the shapes of survival curves and the
information they provide. Figure 8 shows three different curves.

(a)

100

Years

Fi2ure 8. Comparing three sets of survival curves for treatment A and B. From
"Cancer clinical trials: methods and practice," by M. E. Buyse, M. J. Staquet, and
R. J. Sylvester, 1984, Toronto, Oxford University Press, p. 383 .

Cost-Effectiveness Analysis

61

The survival curves of anastrozole 1-mg od and megestrol acetate 40-mg
qid therapies shown in Figure 6 are similar in appearance to curve (a) as the
anastrozole 1-mg od curve lies above the megesterol acetate curve and separates
as time increases. The anastrozole and the megestrol acetate curves are different to
curve (a) as they have less separation and that the percentage of participants
surviving does not reach zero. Curve (a) indicates that treatment A is uniformly
superior to treatment B over the entire life of the participant. In curve (b) death
occurs more quickly than with treatment B, but the long term use of either
treatment results in the same proportion of people dying. Since the proportion of
participants surviving greater than three years is unknown in the
anastrozole/megestrol acetate case, the pharmacoeconomist cannot be certain
whether the anastrozole curve will not follow the same shape as curve (b) or curve
(c). Curve (c) indicates that treatment A is initially superior to treatment B, but
treatment B is more superior over the long term than treatment A. Being aware of
the potential long term consequences of these types of drug therapies is important
to consider when evaluating their effectiveness.
Not only do survival curves provide a useful graphical representation of
the data but they can also be used to estimate the proportion of participants
surviving. Knowing the proportion of people surviving is valuable information.

Cost-Effectiveness Analysis

62

Survival probabilities at various points in time can be estimated and used in the
decision model. The probability information would be used to describe specific
survival outcomes for the treatments under investigation.
Estimating the proportion of participants surviving can be extrapolated
directly from the survival curve. Determining survival probabilities from
extrapolating the data from a curve brings with it considerable error or
uncertainty. Buyse et al. estimate that, when using survival curves to determine
the proportion of participants surviving, the "range in error of the curve at a given
time is roughly± 1/..J (N) , where N'is the number of participants who have either
already died or are followed up to that time" (Buyse et al., 1984, p. 368). A two
year survival rate for megestrol acetate extrapolated from the survival curve
would be approximately 46%. Using the above equation:± li..J253 = ± 0.06, the
true survival rate would be somewhere between 40 and 52%.
A z;-test allows the analyst to compare the probability of survival for two
treatments at a single point in time. The analyst would be able to use the survival
data already gathered in the life table to determine whether there is a significant
difference in survival between the two therapies. The z-test equation is as follows
(Streiner, 1995, p. 442):

Cost-Effectiveness Analysis

z

=

63

(6)

The cumulative proportion surviving at a specific point in time for the two
treatments are P 1 and P 2 (this data is already collected in the life table). The
standard error (SE) for treatment number one is (Streiner, 1995, p. 442):

=

I

~
PI
R

(7)

I

The value of R 1 is the number of the study population who are at risk at a specific
point in time for treatment number one. The standard error for the second
treatment group would be calculated the same way as the first treatment. If the
value for z is found to be 1.96 or greater, then there is strong evidence (at the 0.05
significance level) that the probability of surviving between these two types of
therapies are different (Streiner, 1995). If the value for z is less than 1. 96 then the
probability of survival for the two treatments at the point in time does not differ at
the 0.05 level of significance.

Cost-Effectiveness Analysis

64

Chapter 6
~

~ and

~ Clinical and Economic Data

This chapter will discuss the three final steps in the pharmacoeconomic
evaluation of anastrozole and megestrol acetate. It will first describe the
construction of an anastrozole/megestrol acetate decision model, following which
it will review the collection, tabulation, and presentation of cost data. A costs and
outcomes table will be used to organize the clinical and economic information and
provide the basis for the calculation of cost-effectiveness ratios.
Decision model.
The anastrozole and megestrol acetate randomized clinical trial
information will be used to construct a decision model. The decision model will
quantify the life years gained for a hypothetical cohort of patients receiving
anastrozole or megestrol acetate therapies. The model will include clinical events
that track and quantify the amount of life patients receive over the five year time
horizon. The clinical events will include the loss of patients to death and the loss
of patients to drop-out. Patient drop-out may result from disease progression
and/or drug induced adverse effects.
A recursive decision tree will be used to track the probability of fatal and
non-fatal events occurring throughout the five year time horizon using one year
periods. The probability of fatal events (i.e., death) for each drug therapy, will be

Cost-Effectiveness Analysis

65

estimated from survival data. The probability of non-fatal events (e.g., patient
drop-out) will be estimated from drug utilization data obtained from provincial
cancer agencies, by using epidemiologic information, or by informed guesswork
(Weinstein, Fineberg et al., 1980).
Figure 9 illustrates the main branch of the recursive decision tree.

I

Non-Fatal

jAnastrozole
I mgod

L

Megestrol Acetate
40 mg qid

Death

1

Drop-Out

I

Fatal
Continue Therapy

Fatal

Non-Fatal I

Death

I

Drop-Out

I

I

Continue Therapy

Fifi:ure 9. Main branch ofthe recursive decision tree for anastrozole 1-mg od and
megestrol acetate 40-mg qid therapies.
The main branch starts with a decision node that is represented as a square and is
the point at which the anastrozole 1-mg od or megestrol acetate 40-mg qid therapy
is selected. Down the treatment path from the decision node is a chance node. The
chance node is depicted as a circle and branches out to a fatal event, a non-fatal
event, and a "continue therapy" path. The fatal event ends with the outcome

Cost-Effectiveness Analysis

66

"death". The non-fatal event ends with the outcome "drop-out". The "continue
therapy" path represents all patients who would move on to another period of drug
therapy. The outcome of the non-fatal treatment path is labeled drop-out as
patients who are considered as using other therapies are still alive and are no
longer tracked in the model. Patients who drop-out include those who have
experienced ineffective or intolerable drug therapy.
Figure 10 shows a five year recursive decision tree for anastrozole and
megestrol acetate.
Anastrozole
1-mgod

Continue Therapy

Megestrol
40-mgqid

Continue Therap
Continue Therapy

Continue Therapy

Continue Therapy

Continue Therapy
Continue Therapy

Continue Therapy

Continue Therapy

Year I

~

Year 2

Year 3

Year4

Year 5

10. Five year recursive decision tree for anastrozole and megestrol acetate.

Cost-Effectiveness Analysis

67

In the five year tree the main branch represents a one year period and is
repeated five times. Patients who continue therapy from the first year enter the
second year and distribute between the fatal and non-fatal events and the
"continue therapy" path. The remaining patients continue to distribute between
the two events and the one path until the end of the fifth year.
~ the cost of therapy.

Determining costs in a pharmacoeconomic evaluation involves three main
steps: (1) identifying the types of resources consumed, (2) measuring resource
consumption, and (3) placing a dollar value on resource consumption.
Identifying the types of resources consumed can be determined from a
detailed examination of the events found on the decision tree. The costs attached
to fatal events, nonfatal events, and the continue therapy path can be separated
into medical, non-medical and indirect expenses. Medical costs would be divided
into costs for therapy and costs for treating adverse effects. Non-medical costs
would include travel expenses, accommodations, food, and telephone. Indirect
costs would include the loss of earned income by the patient or their families.
Measuring the costs associated with caring for the patient's additional years of
life that result from the treatment would not be included. CCOHTA believes that
"costs associated with persons living longer and consuming health care resources are
subject to debate" and should not be included (CCOHT A, 1996, p. 12).

Cost-Effectiveness Analysis

68

A detailed review of consumable resource items affecting medical, nonmedical and indirect expense groups will need to be performed. A list of consumable
resource items and the sources of cost information are provided in Table 5.

Table 5
Resource Item List and Sources of Cost Information
Resource Item

Sources of Cost Information

Comments

Physician fees or
salaries

-Provincial fee schedules.
-Canadian Institute for Health Information
(CIHI) National Physician Database.
-Victoria Order ofNurses (VON) offices.
-VON and fee schedules from provinces.
-National health expenditure survey.
-Could allocate costs for heating, lighting,
housekeeping, etc. based on size of clinic.
-Fee schedules are available from the
provinces.
-Provincial fee schedules available.
-Hospital or cancer clinic pharmacies.
-Hospital drug contracts.
-Patented Medications Prices Review Board
and provincial price information.
-Individual pharmacies.
-Retail or manufacturer pricing.
-Employer salary costs including benefits.
-Collected by questionnaire Encompass
items paid out by patients and/or family .

-In some instances, physicians
are paid on salary.
-Cost per service.
- Use cost per intervention.
-No classification system for
fee schedules in place.
-Could use hourly rate.

Nursing services
Clinic services or
Home care
serv1ces
Laboratory and
diagnostic tests
Medical imaging
Drugs (hospital,
cancer agencies)
Drugs (Pharmacy
services)
Medical supplies
Lost wages
Out-of-pocket
expenses

-Cost/test should be used.
-Private and public costs differ.
-Use cost/exam.
-Cost should be at invoice price
plus overhead charges.
-Co-payment must be
accounted for according to the
perspective of the analysis.
-Use cost/item.
-Use hourly rates.
-These costs may include nonmedical costs.

"A Guidance Document for the Costing Process" by, CCOHTA,
1996, Ottawa: National Library of Canada. pp. 6-11.

~

Once resource items have been identified, the next step in the evaluation is
to determine the dollar values. There are different methods that can be used to

Cost-Effectiveness Analysis

69

determine the value of resource items. Whatever the method chosen, each
technique "entails a certain amount of complexity, time and effort and yields a
certain precision" (CCOHTA, 1996, p. 5). CCOHTA (1996) describes the
challenge of valuing resource items as "strik[ing] the appropriate balance between
the need for precision and the avoidance of bias and the effort needed to provide
the increased precision" (p. 5).
When performing the cost analysis the pharmacoeconomist should
consider pricing some resource items according to their opportunity cost. The
opportunity cost is "the value of the foregone benefits because the resource is not
available for its best alternative use" (Drummond et al. , 1993, p. 41). For
example, the opportunity cost for a family member to provide home care to an ill
patient would be the family member' s lost wages, plus any other costs or loss of
revenue that the family member might incur while providing care.
The cost per unit of output is useful for evaluating the consumption of
resources for many items since it assigns a monetary value to the consumption of
specific resource units. Examples of cost per unit of resource output include cost
per laboratory test, cost per chest x-ray, cost per 1-mg dose of anastrozole, and
cost per physician service/intervention.
Other resource items, like home care or clinic services, can be determined
from per diem rates. Per diem rates represent a daily average cost per patient for

Cost-Effectiveness Analysis

70

providing the service. These can be multiplied by the average length of stay to
estimate the total costs (CCOHTA, 1996).
When collecting resource item expenses the pharmacoeconomist should
provide a range of costs that can be used in the sensitivity analysis. A range of
costs can be based on the accuracy of the resource item estimates.
Table 6 shows the range of resource unit costs for each year of drug
therapy for a hypothetical case of participants receiving anastrozole 1-mg therapy
(CCOHTA, 1996). Dollar values are for illustrative purposes only and do not
represent actual costs (Walker, 1997; Prince George Regional Hospital, 1998).

Table 6
Cost Valuation of Continuous Therapy Resource Items
Resource Unit
Physician fees
Nursing services
Clinic services
Laboratory
Medical imaging
Anastrozole 1-mg od
Medical supplies
Lost wages
Out-of-pocket
Total

Year I
$285-$305
$135-$150
$360-$440
$30-$35
$170-$190
$1740$1790
$0
$75-$150
$20-$60
$2815$3120

Year2
$190-$220
$90-$100
$240-$305
$20-$25
$0
$1740$1790
$0
$50-$100
$10-$30
$2340$2570

Year 3
$190-$220
$90-$100
$240-$305
$20-$25
$0
$1740$1790
$0
$50-$100
$10-$30
$2340$2570

Year4
$190-$220
$90-$100
$240-$305
$20-$25
$0
$1740$1790
$0
$50-$100
$10-$30
$2340$2570

Year 5
$190-$220
$90-$100
$240-$305
$20-$25
$0
$1740$1790
$0
$50-$100
$10-$30
$2340$2570

Note: Dollar values are for illustrative purposes only and do not represent actual
values. From "A Guidance Document for the Costing Process" by, CCOHTA,
1996, Ottawa: National Library of Canada. p.14.

Cost-Effectiveness Analysis

71

In the case of anastrozole/megestrol acetate, there would be a cost
valuation table for fatal and non-fatal events and the continuous therapy path, for
both anastrozole and megestrol acetate therapies. Resource item costs would
depend on the treatments and the events that are being studied. For example, outof-pocket costs would be greater for patients and their families, in fatal and nonfatal events than in the continuous therapy path. The costs of fatal and non-fatal
events would be greater because these events would most likely result in more lost
time from work, greater travel, greater accommodation requirements, higher
phone costs, and more lost wages. Non-fatal medical costs associated with disease
progression would be larger than continuous therapy medical costs as these
patients would require more medical interventions (e.g. laboratory work,
physician monitoring, nursing support).
Costs and outcomes table.
Once resource item costs have been determined, the next step in the
analysis is to calculate the total costs and the total effectiveness for anastrozole
and megestrol acetate therapies. Total costs and total effectiveness are calculated
on the costs and outcomes table by performing a cohort simulation of patients to
which survival and cost information can be applied (Walker, 1997). The costs and
outcomes table uses mortality, drop-out, and continuing therapy rates to predict
the number of patients surviving, the number of patients who died, the number of

Cost-Effectiveness Analysis

72

patients who dropped out, and the number of patients who continued with therapy.
Once these numbers of patients have been determined, the total survival time and
the total cost of therapy, for each year of treatment, is then calculated. Survival
time can be calculated by summing the years that patients are alive. Patients
receive one-half of a year of survival time for fatal and non-fatal events and a full
year for each year in the continuous therapy path.
Table 7 shows the hypothetical costs and outcomes for anastrozole 1-mg
od therapy while Table 8 shows the hypothetical costs and outcomes for
megestrol acetate therapy. While the values used in the costs and outcomes table
do not represent actual values, the table illustrates how total costs and total
effectiveness, how the cost-effectiveness ratio, and how the incremental costeffectiveness ratio can be calculated.

Cost-Effectiveness Analysis

73

Table 7
Hypothetical Costs and Outcomes Table for Anastrozole 1-mg od

year I

year 2

year 3

year 4

year 5

mortality rate

0.23

0.26

0.39

0.31

0.25

drop-out rate

0.06

0.06

0.09

0.1

0.1

continuing therapy rate

0.71

0.68

0.52

0.59

0.65

costs for death ($/pt)

$2,523

$2,285

$2,285

$2,285

$2,285

costs for drop-out ($/pt)

$2,055

$1 ,817

$1 ,817

$1 ,817

$1 ,817

costs for continuing therapy ($/pt)

$2,815

$2,340

$2,340

$2,340

$2,340

# pts surviving

770

570

348

240

180

# ofpts died

230

200

222

108

60

# of pts dropped out

60

46

51

35

24

# pts continuing therapy

710

524

297

205

156

total survival time in cohort (life years)

855

647

434

276

198

total cost in cohort (Canadian$)

$2,702,095 $1 ,766,626

$1 ,295,937

$789,228 $545,359

discount factor (at a 5% discount rate)
discounted survival time in cohort
discounted costs in cohort

0.9524
0.907
814
587
$2,573,475 $1 ,602,330

0.8638
375
$1 , 119,430

0.8227
0.7835
227
155
$649,298 $427,289

Costs and Outcomes

total survival time for five year treatment (after discounting)
total cost for five year treatment (after discounting)

2158 life years
$6,371 ,822

Cost-Effectiveness Ratio

$2,953/Iife year

Note: A hypothetical cohort of 1000 patients at day 1. Distribution of patients into
death and drop-out events, and continue therapy path determined at the end of
each year of treatment according to mortality, drop-out, and continue therapy
rates.

Cost-Effectiveness Analysis

74

Table 8
Hypothetical Costs and Outcomes Table of

Acetate 40-mg qid

~

Costs and Outcomes

year I

year 2

year 3

year 4

year 5

mortality rate

0.28

0.36

0.39

0.21

0.18

drop-out rate

0.05

0.06

0.09

0.09

0.1

continuing therapy rate

0.67

0.58

0.52

0.7

0.72

costs for death ($/pt)

$2,427

$2,285

$2,285

$2,285

$2,285

costs for drop-out ($/pt)

$1 ,959

$1 ,817

$1 ,817

$1 ,817

$1 ,817

costs for continuing therapy ($/pt)

$2,624

$2,149

$2,100

$2,050

$2,000

# pts surviving

720

461

281

222

182

# ofpts died

280

259

180

59

40

# of pts dropped out

50

43

41

25

22

# pts continuing therapy

670

418

297

197

160

total survival time in cohort (life years)

835

569

408

239

191

total cost in cohort (Canadian $)

$2,535,590 $1 ,568,189 $1 ,109,697

$584,208

$451 ,447

discount factor (at a 5% discount rate)
discounted survival time in cohort
discounted costs in cohort

0.9524
0.907
795
516
$2,414,896 $1 ,422,347

0.8227
197
$480,628

0.7835
150
$353,709

0.8638
352
$958,556

total survival time for five year treatment (after discounting)
total cost for five year treatment (after discounting)

2009 life years
$5 ,630,135

Cost-Effectiveness Ratio

$2,802/life year

Incremental Cost Effectiveness Ratio (ICER = C 1-CiE 1-E 2)

$4978/life year gained

Hypothetical cohort of 1000 patients at day 1. Distribution of patients into
death and drop-out events, and continue therapy path determined at the end of
each year of treatment according to mortality, drop-out, and continue therapy
rates.

~

Cost-Effectiveness Analysis

75

Once total survival time and total cost for the cohort is determined, a
discount factor is applied to survival and cost data (CCOHT A, 1997). The
discount factor is multiplied by the cost and survival time for each year of
treatment (Drummond et al., 1993). The total survival time and the total cost for
the five year treatment of the hypothetical cohort of patients are then calculated.
The cost-effectiveness ratio is determined by dividing the total cost by the
total survival time for the five year treatment after discounting. The incremental
cost-effectiveness ratio can be calculated after the cost-effectiveness ratios for
both drug treatments have been determined.
The costs and outcomes table also provides a simple way of adjusting
variables for performing sensitivity analysis. Walker (1997) recommends
generating the costs and outcomes table on a computer spreadsheet (e.g. Microsoft
Excel®) and programming the simple mathematical formulas so that, when
performing sensitivity analysis, the variables can be easily adjusted and the
calculations can be performed automatically.

Cost-Effectiveness Analysis

76

Chapter 7
Summary
This thesis has outlined the methodological techniques for performing a
cost-effectiveness analysis of anastrozole and megestrol acetate in the treatment of
progressive breast cancer. The Canadian Coordinating Office for Health
Technology Assessment (CCOHTA) document "Guidelines for Economic
Evaluation of Pharmaceuticals" has been used to provide a comprehensive and
structured approach to the methodological techniques needed in a
pharmacoeconomic analysis (CCOHTA, 1997). Hopefully, researchers carrying
out similar pharmacoeconomic studies will find this review useful and will
encourage effective, efficient, and equitable use of pharmaceutical therapies.
The principal pharmacoeconomic methodology used to evaluate
anastrozole versus megestrol acetate was a cost-effectiveness analysis (CEA). The
CEA of anastrozole/megestrol acetate was selected for two reasons. First,
researchers have suggested that inequitable resource allocation has occurred in the
area of palliative cancer therapies and that close evaluation of these therapies
needs to be performed (Jonsson et al. , 1995). Second, the anastrozole/megestrol
acetate case has provided an opportunity to describe methodological techniques ·

Cost-Effectiveness Analysis

77

that would be useful in measuring the effectiveness of palliative breast cancer
drug therapies.
While investigators in the anastrozole/megestrol acetate phase III trials
reported the findings as two year survival rates and median time to death, the data
was reevaluated by methods considered by some researchers as being less biased.
The anastrozole and megestrol acetate survival information was expressed as
absolute risk reduction and numbers needed to treat.
The absolute risk reduction for women taking anastrozole 1-mg od for a
median time of 31 months was found to be 10.2 percent. This indicated that
women being treated with anastrozole 1-mg od for 31 months had 10.2 percent
fewer deaths than women treated with megestrol acetate 40-mg four times daily
(qid) for the same duration.
The number needed to treat was calculated to be 10, indicating that if 10
women were treated with anastrozole 1-mg od for 31 months, instead of megestrol
acetate 40-mg qid, one death would be prevented. This demonstrated that
anastrozole 1-mg od offered a survival advantage over megestrol acetate 40-mg
qid and that a pharmacoeconomic evaluation of these therapies would be
worthwhile. However, because the randomized clinical trial data may not reflect
real-life drug utilization patterns and because the phase III drug trial findings

Cost-Effectiveness Analysis

78

cannot be reliably extrapolated beyond the 31 month median follow-up period,
these results may not be generalizable.
Before methodological techniques were reviewed the pharmacoeconomic
study population was defined, the perspective of the analysis was stated, and the
time horizon for the analysis was determined. The pharmacoeconomic population
was defined as women with postmenopausal estrogen receptor positive breast
cancer whose disease has progressed while on tamoxifen and who have switched
therapy to either anastrozole 1-mg od or megestrol acetate 40-mg qid. The
analysis followed the societal perspective as recommended by CCOHTA. The
time horizon for the analysis was determined to be five years based on efficacy
data from other hormonal breast cancer therapies.
In chapter 2 decision modeling methods were discussed. Decision
modeling was used to systematically organize clinical and economic information
into a structured framework of events. A standard decision tree, a recursive
decision tree, and a Markov model were described. A recursive decision tree was
selected for the analysis of the anastrozole/megestrol acetate case. The recursive
decision tree comprised a main branch of two events and one treatment path. The
events selected were "death" and "drop-out", and the path was "continue therapy".
These events were selected to provide a means of quantifying the survival time
obtained from each therapy.

Cost-Effectiveness Analysis

79

The main branch of the recursive decision tree represented a one year
period of therapy. The main branch was repeated five times and formed the five
year recursive tree. Structuring the tree in this way provided a method of tracking
the distribution of patients through the five year therapy. At the end of the five
year horizon the amount of survival time was estimated by summing up the
amount of life-years for a hypothetical cohort of patients.
In reviewing the phase III trial data it was found that the
anastrozole/megestrol acetate 31 month median treatment period was too short by
CCOHTA' s standards. According to CCOHTA, the ideal time horizon for a
pharmacoeconomic analysis of survival times should be until all participants in
the study have died. If this information is not available, then modeled data would
need to be used. It was determined that modeled data would need to extend the
phase III trial data from 31 months to at least 60 months according to efficacy
data from other hormonal therapies.
In order to determine the likelihood of patients surviving after each year of
drug therapy, the probability of death had to be determined. In order to calculate
the probability of death, raw survival data from the phase III trials was needed.
Raw survival data would provide the analyst with the opportunity to calculate the
mortality rate, the hazard ratio, the probability density function, and predict
survivorship beyond the 31 month median time frame. Since raw survival data

Cost-Effectiveness Analysis

80

was not available another method of estimating mortality/survival rates was
discussed. Survival rates could be estimated by extrapolating the values from the
survival curves but, while extrapolating survival rates is easy to perform
mathematically, the validity and the reliability of the findings in the present case
is too imprecise to justify using this method.
Once the recursive decision tree was developed, the next step in the
evaluation was to determine the costs for anastrozole and megestrol acetate
therapies. The evaluation of costs for therapy encompasses three main steps:
identifying the types of resources, measuring resource consumption, and
determining the dollar value of resource consumption. Resource items were
identified from the events found on the recursive decision tree and were based on
the societal perspective. A list of resource items, the sources of cost information,
and specific points to

~

when measuring those resource items were also

discussed. Even though costing of resource items was not performed, one method
for calculating the value of resource items could be determined from cost per unit
of resource output and per diem rates. When valuing resource items a range of
costs would be gathered and used in a sensitivity analysis.
The final stage of the pharmacoeconomic evaluation involved combining
clinical and economic data into a single table. A sample table of costs and
outcomes for anastrozole and for megestrol acetate therapies were provided (see

Cost-Effectiveness Analysis

81

Tables 7 and 8). The tables were used to demonstrate how survival time,
discounting, and cost-effectiveness ratios could be calculated from a hypothetical
cohort of patients. The data used in the tables did not represent actual values but
were provided for illustrative purposes only.
The costs and outcomes table combined mortality, drop-out, and
continuing therapy rates so as to determine the number of patients in each of the
three events in years one through five. Costs per patient, determined from the
consumption of resource items, were then applied to the number of patients in
each event. Survival times were determined from the number of patients in the
death and drop-out events, and the continue therapy path. Patients who dropped
out, or died, were given one-half of a year of survival time. Patients who
continued with therapy received one full year of survival time. Costs and survival
times were totaled and a discount rate of five percent would be applied to both
costs and survival times. The cost-effectiveness ratios, after discounting, was then
determined. The incremental cost-effectiveness ratio for anastrozole 1-mg od and
megestrol acetate 40-mg qid was determined from the cost-effectiveness ratios.
Limitations.
Performing a retrospective cost-effectiveness analysis from clinical trial
data poses several problems. The randomized clinical trial efficacy data for
anastrozole and megestrol acetate was designed to demonstrate a particular

Cost-Effectiveness Analysis

82

primary endpoint and was not intended for secondary pharrnacoeconomic
analysis.
The anastrozole and megestrol acetate clinical phase III data were found to
use a dosing regimen for megestrol acetate that is not normally being prescribed
in Canada. The clinical trials used a four times daily dosing regimen for ·megestrol
acetate instead of the more popular daily regimen. Noncompliance with the qid
dosing regimen is more likely to occur than with the od regimen. If a significant
portion of patients in the phase III trials was noncompliant then the survival
results could be compromised. For example, the improved survival times for
anastrozole may have been a result of better patient compliance than of superior
pharmacotherapeutic properties of this drug ..
Further limitations include extending the time horizon of the analysis.
Extending the time horizon to five years requires modeling procedures that can
lead to unreliable data. Modeled data for determining the mortality rates for years
three, four, and five, would have to be thoroughly tested by sensitivity analysis.
Sensitivity analysis would also need to be applied to drop-out and continue
therapy rates. Sensitivity analysis would need to be applied to cost valuation of
resource items and discount rates. The sensitivity analysis would be performed by
adjusting variables in the costs and outcomes table.

Cost-Effectiveness Analysis

83

While the recursive decision tree provided a method for tracking the
survival of patients during anastrozole and megestrol acetate therapies, the model
may be too simplistic for actual use. A more detailed recursive decision tree with
a greater number clinical events and paths may help to improve the predictability
of survival for hormonal therapy in palliative breast cancer.
Future research.
Although the practice ofpharmacoeconomics is considered in its infancy,
guidelines for performing and evaluating pharmacoeconomic studies are
becoming commonplace. Guidelines, such as those developed by CCOHTA, are
helping to ensure that the methods of economic and clinical evaluation of health
care programs are based on accepted analytical techniques. Future research will be
required to illustrate how these guidelines can be used in everyday health care
practice. For example, health care managers, responsible for the allocation of
resources, will need to understand how pharmacoeconomics can be used to assist
in drug policy decision making.
Since the accuracy and validity of pharmacoeconomic studies is reliant on
quality data, future pharmacoeconomic research should promote methods of
obtaining accurate and reliable information. Randomized clinical trials are
important sources of information but can be of limited value. Randomized clinical
trials are designed to study drug safety and efficacy under controlled situations,

Cost-Effectiveness Analysis

84

but in pharmacoeconomic evaluations, it is how the drug works in the real world
that is of importance. Future research should find ways to overcome the pitfalls of
using randomized clinical trial data in pharmacoeconomic evaluations.
It would be hoped that the B.C. Cancer Agency and other institutions in

British Columbia will, in the future, use criteria similar to those of the Therapeutic
Initiative and the Pharmacoeconomics Iniative in the selection of drugs for
patients under their care.

Cost-Effectiveness Analysis

85

References
Beck J. R., & Pauker S. G. (1983). The Markov process in medical
prognosis. Medical Decision

~

3(4), 419-458.

Bonneterre J., Schraub S., Lecomte S., & Mercier M. (1996). Quality of life
as an outcome in breast cancer: clinical application. Pharmacoeconomics.
2(Suppl. 2), 23-29.
Bootman J. L. , Townsend R. J. , & McGhan W. F. (1996). Principles of
pharmacoeconomics. Cincinnati: Harvey Whitney.
Briggs A. , & Sculpher M. (1997). Commentary; Markov models of medical
prognosis. British Medical Journal. 314, 354-355.
British Columbia Cancer Agency. (1997, February). Class II drug
registration form. Vancouver: Author.
Brodie A. M. H. (1996). Tumor biological aspects of aromatase. In K.
Pritchard (Chair), Toronto Symposium for the Canadian

~

Society:

Aromatase Inhibitors: Current data and future perspectives for breast cancer.
Mississauga, Ontario: Zeneca Pharma Inc.
Budzar A., Jonat W., Howell A., Jones S. E. , Blomqvist C. , Vogel C. L.,
Eiermann W., Wolter J. M., Azab M., Webster A., & Plourde P. V. (1996).
Anastrozole, a potent and selective aromatase inhibitor, versus megesterol acetate

Cost-Effectiveness Analysis

86

in postmenopausal women with advanced breast cancer: results of overview
analysis of two phase III trials. Journal of Clinical

~

14(7), 2000-2011.

Buyse M. E., Staquet M. J., & Sylvester R. J. (1984). Cancer clinical trials:
methods and practice. Toronto: Oxford University Press.
Canadian Coordinating Office for Health Technology Assessment. (1994).
Guidelines for economic evaluation of pharmaceuticals: Canada (1st ed.). Ottawa:
National Library of Canada.
Canadian Coordinating Office for Health Technology Assessment. (1996).
A guidance document for the costin2 process. Ottawa: National Library of
Canada.
Canadian Coordinating Office for Health Technology Assessment. (1997).
Guidelines for economic evaluation of pharmaceuticals: Canada (2nd ed.).
Ottawa: National Library of Canada.
Clemens K., Townsend R., Luscombe F., Mauskopf J., Osterhaus J., &
Bobula J. (1995). Methodological and conduct principles for pharmacoeconomic
research. Pharmacoeconomics. 8(2), 169-174.
Coldman A. J., & Elwood J. M. (1979). Examining survival data. Canadian
Medical Association Journal. 121. 1065-1 071.
Desch C. E., Hillner B. E., Smith T. J. , & Retchin S.M. (1993). Should the
elderly receive chemotherapy for node-negative breast cancer? A cost-

Cost-Effectiveness Analysis

87

effectiveness analysis examining total and active life-expectancy outcomes.
Journal of Clinical Oncolo2y. 11(4), 777-782.
Detsky A. S., & Naglie I. G. (1990). A clinician' s guide to costeffectiveness analysis. Annuals oflntemal Medicine. 113, 147-54.
Drummond M. F., Stoddart G. L., & Torrence G. W. (1993). Methods for
the economic evaluation ofhealth care pro2rammes. New York: Oxford
University Press.
Frankel D. H. (1995). US cancer institute halts tamoxifen study. Lancet.
.HQ, 1552.

Gehan E. A. (1975). Statistical methods for survival time studies. In
Staquet M. J. (Ed.), Cancer therapy: pro2nostic factors and criteria of response
(pp. 7-35). New York: Raven Press.
Hall J., & Tattersall M. (1995). Economic considerations in cancer care. In
Veronesi U. , Goldhirsch A., & Yamold J. (Eds.), Oxford's Clinical Oncolofi!y.
(pp. 2456-2463). New York: Oxford University Press.
Hennekens C. H. , & Buring J. E. (1987). Epidemiolo&y in medicine.
Toronto: Little, Brown and Company.
Hillner B. E. ( 1996). Role of decision analysis in relation to clinical trials
and a US perspective of the Batelle model. Pharmacoeconomics. 9(Suppl. 2), 3036.

Cost-Effectiveness Analysis

88

Hillner B. E. , & Smith T. J. (1991). Efficacy and cost-effectiveness of
adjuvant chemotherapy in women with node-negative breast cancer. New

~

Journal of Medicine. 324(3), 160-68.
Jonsson V., Clausen S. R., & Hansen M. M. (1995). Pharmacoeconomic
aspects in the treatment of curable and incurable cancer. Pharmacoeconomics .
.8.(4), 275-281.
Kaplan E. L., & Meier P. (1958). Nonparametric estimation from
incomplete observations. Journal of American Statistical Association. 53. 457481.
Kaufman D. (1993). Cancer therapy and the randomized clinical trial: Good
medicine? Cancer. 72(Suppl. 9), 2081-2084.
Laupacis A., Naylor C. D., & Sackett D. L. (1992). How should the results
of clinical trials be presented to clinicians [editorial]? American Clinical
~

Journal Club. May/June. A12-A14.

Lee E. T. (1980). Statistical methods for survival data analysis. California:
Lifetime Learning Publications.
Luke D. A. (1993). Charting the progress of change: a primer on survival
analysis. American Journal of Community
Mathews D. E., & Farewell V. (1985).
statistics. New York: Karger

~
~ and

21 (2), 203-246.
~ medical

Cost-Effectiveness Analysis

89

National Cancer Institute of Canada. (1996). Canadian Cancer Statistics.
Toronto: Author.
Pauker S. G. , & Kassirer J. P. (1987). Medical progress: decision analysis.
New

~

Journal of Medicine. 316(5), 250-258.

Patented Medicine Prices Review Board. (1996). Year-over-year
the patented medicine price index 1988-1995 :

~

~

annual report. Ottawa:

Author.
Prince George Regional Hospital. (1998). [Cost valuation of resource
items]. Unpublished raw data.
Pritchard K. L. (1997). Significantly improved survival with Arimidex®
(anastrozole) vs. megestrol acetate (MA) in postmenopausal advanced breast
cancer; updated results from two randomized trials: Questions and answers.
Denver. Colorado Symposium for the American Society of Clinical

~

Mississauga, Ontario: Zeneca Pharma Inc.
Riegelman R. K. , & Hirsch R. P. (1989).

~ a study and

~ a

test: How to read the medical literature. Toronto: Little Brown and Company.
Rubens R. (1996). Key issues in the treatment of advanced breast cancer:
expectations and outcomes. Pharroacoeconomics. 9(Suppl. 2), 1-7.

in

Cost-Effectiveness Analysis

90

Santen R. J. ( 1996). Clinical endocrinology of aromatase inhibitors. In K.
Pritchard (Chair), Toronto Symposium for the Canadian

~

Society:

Aromatase Inhibitors: Current data and future perspectives for breast cancer.
Mississauga, Ontario: Zeneca Pharma Inc.
Siegel J. E., Torrance G. W., Russell L. B., Luce B. R., Weinstein M. C.
( 1997). Guidelines for pharmacoeconomic studies: recommendations from the
panel on cost effectiveness in health and medicine. Pharmacoeconomics. 11(2),
159-168.
Smith G. , Henderson C. I. (1996). New treatments for breast cancer.
Seminars in

~

23(4), 506-528.

Sonnenberg F. A., Beck J. R. (1993). Markov models in medical decision
making: a practical guide. Medical Decision

~

13. 322-338.

Streiner D. L. (1995). 'Stayin alive:' an introduction to survival analysis.
Canadian Journal of Psychiatry. 40. 439-444.
Therapeutics Initiative. (1996). Evidence based

~ thercwy. What do the

numbers mean? (Issue 15). Vancouver, BC: Author.
Torrance G. W. , Blaker D., Detsky A., Kennedy W. , Schubert F., Menon
D., Tugwell P., Konchak R. , Hubbard E., Firestone I. (1996). Canadian
guidelines for economic evaluation of pharmaceuticals. Phannacoeconomics.
2(6), 535-559.

Cost-Effectiveness Analysis

91

Walker, H. (1997, March). Cancer costs and outcomes. Lecture and
documentation presented at a Health Economics lecture at the University of
Northern British Columbia, Prince George, British Columbia.
Weinstein M. C. (1980). Estrogen use in postmenopausal women: costs,
risks and benefits. New En2land Journal of Medicine. 303(6), 308-16.
Weinstein M. C. ( 1981 ). Economic assessments of medical practices and
technologies. Medical Decision Makin2. 1(4), 309-329.
Weinstein M. C. , Fineberg H. V., Elstein A. S., Frazier H. S., Neuhauser
D., Neutra R. R., McNeil B. J. (1980). Clinical decision analysis. Toronto: W.B.
Saunders Company.
Weinstein M. C., Stason W. B. (1977). Foundations of cost effectiveness
analysis for health and medical practices. New En2land Journal of Medicine,
22.2(13), 716-21.
Veronesi U, Goldhirsch A, Yarnold J (1995). Breast cancer. Oxford's
Clinical Oncolo2y. New York: Oxford University Press.
Zeneca (1997a). Arimidex: a si2nificant advanta2e in the only reliable
parameter- survival [Product Information Brochure]. Mississauga, Ontario:
Author.
Zeneca (1997b).

~ [Monograph]. Mississauga, Ontario: Author.