Pharmacoeconomics and Outcomes Research: Evaluating the Studies
Adjusting Costs for
Differential in Timing-Discounting7
Why discount?
treatment. Although difficult to measure, the cost of pain and
suffering is assessed in the courts daily. However, in health care,
these costs are difficult to measure and quantify, and thus usu-
.
ally omitted from pharmacoeconomics evaluations.
Drummond recommendations
Measuring direct costs is essentially counting units-number of
physician visits, number of days missed from work, number of
doses, etc., and multiplying by.the dollar value of that unit. To
Time preference for receiving money and time
preference for receiving public goods-social discount rate
.
Present costs in their undiscounted form, so oth-
ers can investigate the implications of using differ-
ent discount rates.
.
Undertake a base case analysis using the rates of
3% and 5%, currently recommended by analysts
from the U.S. Public Health Service Panel on CostEffectiveness in Health and Medicine.
.
Undertake a sensitivity analysis, making sure. to
include 0%, 3%, and 5%.
.
In situations where the discount rate has a substantial impact on study results, alert decision makers to the importance of selecting a discount rate.
Measuring Units
measur~ productivity losses, patient self-report data can be
used, *hough this technique is somewhat controversial due to
concerns about recall accuracy, among other things. For example, in themigraine study rrt'êntjoned
earlier, patient self-reports
were used to measure hours: of work missed and to assess how
producb'vepatients felt at J~rk:whensuffering from a migraine.
'I
But were the missed days reþorJ:edreally due to migraine headaches? ,Such questions pose. credibility issues for researchers
when using recall data.
Valuing Costs
Once resources have been identified and measured, they must
be valued or assigned a unit cost. Opportunity cost is an important concept in valuing resources and is defined as the amount
an input could earn in its best alternative use. Opportunity
costs are difficult to assess, so the market price of a good or service is typically used. This method assumes a competitive mar-
Identifying and Measuring Costs
Another important step in conducting pharmacoeconomics
evaluations is identifying and measuring costs. Customarily,
pharmacoeconomics evaluations measure four basic types of
nonmedical costs, indirect
ket exists and uses the market price as a proxy for valuing medical resources. Because the health care market falls short of a
competitive market in many ways, charges or reimbursement
rates often serve as proxies in valuing pharmaceuticals and
costs: direct medical costs, direct
other medical resources. However, this method poses certain
costs and intangible costs (see Table 3, p.7).
Direct medical costs are resources spent on medical ser-
problems.
vices or products, including physician visits, emergency depart-
ment visits, medication, hospitalizations, and nursing or pharmacy staff time. Direct nonmedical costs can include transportation for medical treatment, boarding for family during
treatment, and other expenses related to the provision of medical care but incurred outside the medical sector.
Although controversial, indirect costs are included in many
studies. Loss of productivity due to temporary illness or disability or to permanent loss of employment are examples of
indirect costs. These costs are often measured as the wage rate
of an individual multiplied by hours of work lost. In studies
focusing on migraine and other chronic conditions, indirect
costs are an important consideration, particularly from the
employer's perspective. One study found that patients lost an
average of 2.2 days of work per month when suffering from
migraine. The costs associated with these lost days can be significant to employers~6
Intangible costs are those costs associated with real or perceived pain and suffering caused by the illness or resulting from
8
Supplement to Journal of Managed Care Pharmacy
jMCP
Once costs have been valued, they then should be adjusted
for diffFences in timing by d.i,~cpuntingfuture costs to a present
value. the importance of dis~(jünting relates to our time preference for receiving money We would rather receive benefits
today than a year from now. Conversely, we would prefer to
incur costs later rather than in the immediate future. The same
time preference for receiving money exists in industries such as
health care. Drummond has established criteria for which discount rates should be used to adjust costs for differences in timing (see Table 4).7
The importance of discounting is highlighted by the following example (see Table 5).7 In health care, particularly when
treating a patient with a chronic disease, up-front costs may lead
to benefits that are not observed until five or 10 years in the
future. With benefits of $95,000 and costs of $90,000, a $5,000
net benefit is observed when no discounting is conducted.
Given a positive net benefit, a decision maker would probably
decide to implement such a program. However, if the future
benefits are discounted-even at a low rate of 3%-the positive
net benefit is no longer observed.s
JanuarylFebruary 2000
Vol. 6,
No.1
Pharmacoeconomics and Outcomes Research: Evaluating the Studies
:'::'1/,');
i:::<':<"
20,000'
.'. ;,,;..;
,',
"
"
;
','.,
',;'í:',,""'"
.,
,.,.,
I.
"".'.
"
'
*Assume that no costs or benefits are realized in any year other than those presented.
Reprinted with permission from Warner et al.'
".'
_Study Designs
Four major types of study designs are used in pharmacoeconomics evaluations: clinical trials, naturalistic (or large simple)
trials, models, and retrospective analyses. No study design is
inherently superior to another for conducting pharmacoeconomics research. The approaches are complementary, and each
has distinct advantages and disadvantages that should be considered.
Clinical Trials
The most familiar type of study to clinicians is probably the
clinical trial. It typically has strong internal validity but can be
of questionable relevance in the real world. Clinical trials are
characterized by random assignment of subjects to study
groups, a control group (often a placebo), and blinding of subjects and investigators. Clinical trials are the gold standard for
demonstrating efficacy of drug therapies.
coeconomics study often can be piggybacked onto an existing
clinical trial. Second, subjects are randomized to treatments,
which enhances the internal validity of clinical trials (the confidence that the observed effect is due to the treatment and not
some other factor). A major disadvantage is that the results are
often of questionable relevance to routine practice because such
controlled conditions are rarely found in clinical practice.
More specifically, clinical trials may lack an appropriate
comparator. In addition, surrogate outcome markers (e.g., blood
pressure) often are used to make results readily available in a
timely manner. This is particularly problematic if the surrogate
marker and the outcome of interest are not closely linked, which
may be the case with ulcer symptoms and endoscopy-based
ulcer rates.
Another limitation inherent in clinical trials is that patient
compliance typically is higher than in routine care. Populations
.
a
outside
a
clinical trial are often described as "professional patients."
In addition, because these patients are monitored closely, hj\ve
frequent contact with providers, and receive prompting from
Vol. 6. No. I
clinical trial.
Furthermore, because patient retention is an unstated goal
of clinical trials, adverse effects are managed intensively, reducing the likelihood that patients will discontinue taking medica-
tion. An example of how compliance in clinical trials can be
misleading appeared in recent literature. Investigators included
only clinical trial data in a meta-analysis of efficacy rates of tricyclic versus selective serotonin reuptake inhibitor (SSRI) antidepressants and found no significant difference. Since then, a
number of studies have demonstrated that in more routine clinical settings, substantial differences exist between the compliance rates with tricyclics and with SSRIs, particularly in the first
two months of therapy, and that this compliance can affect success rates.
Two major advantages exist in conducting a pharmacoeconomics study using a clinical trial approach. First, the pharma-
in
investigators to take their medication, patients may be more
compliant with the medication regimen than they would be
9
Naturalistic Trials
The alternative to the clinical trial is the naturalistic, or large
simple, trial. Its goal is to eliminate most of the disadvantages
clinical trial while retaining randomization and a control
LV
group. Typically, naturalistic trials are conducted in the Phase
Postmarketing stage. Settings and providers are more represenof
a
tative of routine clinical practice than those in a clinical trial.
and providers are not blinded to the treatment
Patients
received. Inclusion apd exclusion criteria are less extensive. In
addition, the drug being evaluated is compared with standard
a placebo. The antidepressant trial discussed
previously is an example of a naturalistic trial9
The primary advantage of naturalistic trials is that they more
accurately ref1ect routine clinical care. However, naturalistic trials typically can be even more expensive than clinical trials and,
in many instances, are not a practical alternative. Furthermore,
care rather than
naturalistic trial may not be as representative as the investigators claim. Decision makers should read the methodology carefully and decide whether the trial accurately mirrors c1i,nical
a
practice.
january/February 2000
jMCP
Supplement to Journal of Managed Care Pharmacy
9
Pharmacoeconomics and Outcomes Research: Evaluating the Studies
a chance node or a probability node. At this juncture, a
probability that the Helicobacter pylori bacteria will be eradicated can be assigned to one branch and a probability that it will
not to the other. The assigned probabilities are based on the lit-
called
Models
When to use?
erature and should total 100%. The two branches at the next
chance node represent the probability that the ulcer will recur
.
Clinical trial does not capture economic data
.
Measures clinical intermediaries only
and the probability that it will not. Even if the ulcer is eradicated, the probability exists that the patient will have a recurrence;
this probability is much lower than if the ulcer had not been
.
Compares treatment with placebo
.
Want to translate clinical trial data into that which
eradicated, but it exists nevertheless. Beyond those chance
nodes-"ulcer" and "no ulcer recurrence"-a chance exists that
better reflects routine clinical care
Models
Models are commonly used in pharmacoeconomics research
and comprise an increasing portion of published studies. A
model is defined as a systematic approach to decision making
under conditions of uncertainty A model
is used
when direct
observation is not available and prospective data cannot be collected. A model does not banish uncertainty but makes the uncertainty explicit by including junctures at which a decision
must be made or the likelihood of an outcome predicted. In
pharmacoeconomics, investigators use two major types of models
-decision-analytical and Markov. The latter are sophisticated
and represent a minority of the models in the literature; decision-analytical models are used more often.
Figure 5 is an example of a decision-analytical model of
ulcer pharmacotherapy. Each square box is called a decision
node, where a choice is made about which treatment alternative
to use. Two branches are shown here, but other branches could
be added at this decision node, such as the dual-based or triplebased proton pump inhibitor regimens.
Once a triple therapy has been selected, the clinical course
can take more than one direction. The circle on the model is
Example of
a
an outpatient visit will occur, as well as a chance that the patient
will be hospitalized.
The reader should then critically analyze a model to determine whether the model has been constructed as simply as possible, yet stil1 represents clinical reality The model shown here
is reasonable because it captures several important elements in
the course of treating a patient with an ulcer.
Table 6 details the situations in which models are useful. For
example, if no pharmacoeconomics data were collected during
a chnical trial but the investigators want to perform a pharmacoeconomics analysis, a model can be developed that incorporates the efficacy data from the clinical trial and cost data from
other sources. If the clinical study has measured only intermediary or surrogate measures, the investigators can use a model
to extrapolate those results to a lifetime.
Assume one clinical trial has compared drug X with placebo
and another compared drug Y with placebo. The investigator
later decides to conduct a head-to-head comparison of the two
drugs. The results from the two clinical trials may be used to
model the economic impact of selecting drug X versus drug Y.
In addition, if the investigator has clinical trial data and wants
to construct a model that more closely mirrors routine clinical
care, data from medical claims databases and the hterature can
be used.
For example, to compare the cost of outpatient treatment of
depression with tricyclics versus SSRIs, the investigator knows
Decision-Analytic Model
,/
Ulcer
H2 antagonist
.~
Not eradicated
I
.?l}
No ulcer
I
Outpatient
~i
Hospitalization
recurrence
Triple therapy
atient
Fl
Ulcer
Eradicated
No ulcer
1.;{
recurrence
10
Supplement to Journal of Managed Care Pharmacy
jMCP
JanuarylFebruary 2000
Vol. 6,
No.1
Hospitalization