Escolar Documentos
Profissional Documentos
Cultura Documentos
D
Decision Analytic Modeling
P Muenning, Columbia University, New York, NY, USA
2008 Elsevier Inc. All rights reserved.
Introduction
Life is full of uncertainties. You could invest in a mutual
fund or a certificate of deposit account. You could buy a
home or rent. You could go to public health school or
write a novel.
If you knew the risks and benefits of each alternative,
the decision would be a lot easier. Decision analysis provides a mathematical framework for making decisions
under conditions of uncertainty. Lets see how this works
by using an example.
Imagine that you have $100 000 to invest. Youve always
wanted to study public health, but youve also dreamed
about writing a novel ever since you were in college. If
you write a novel, you reason, you can simply invest the
money in a mutual fund and earn interest, making small
withdrawals as needed. Since you are undecided between
the two options, you decide to examine which will be the
better option financially.
Decision analysis is based on a concept called expected
value, in which the value of an uncertain event (such as
making $10 000 in the stock market) is weighed against the
chances that the event will occur. For example, if you know
that the historical average increase in a particular mutual
fund is 10% per year, then the expected value of the return
on your $100 000 investment is 0.1 ! $100 000 $10 000
over one year.
You set your sights five years down the road and assume
that if you go to public health school, all of the $100 000
would be spent on your education after living expenses are
taken into account, but you would be able to earn about
$50 000 per year working in public health. After taxes and
living expenses, you estimate that you would save about
$10 000 over the 3 years after graduation.
If you decide to write a novel, you will have spent your
$100 000, along with the interest on the mutual fund, on
living expenses over the 5-year period. But if you publish
the novel, you could earn an additional $30 000. Your
71
International Encyclopedia of Public Health, First Edition (2008), vol. 2, pp. 71-76
Supportive care
Remains well
Cost = $0
0.9
Becomes ill
Cost = 83
Vaccination
0.1
Remains well
Cost = $10
0.97
Becomes ill
Cost = $38
Cost
Cost
Cost
Cost
0.03
Figure 1 Decision analysis tree comparing supportive care to
vaccination.
International Encyclopedia of Public Health, First Edition (2008), vol. 2, pp. 71-76
Remains well
Cost = Cost + 0
73
Cost
0.9
Hospitalized
Supportive care
Sees doctor
Cost = Cost + 110
0.2
Becomes III
Cost
0.01
Not hospitalized
Cost = Cost + 0
Cost
0.99
Cost = Cost + 12
Hospitalized
0.1
No doctor
Cost = Cost + 0
0.8
Cost
Cost
0.99
Remains well
Cost
Cost = Cost + 0
0.97
Hospitalized
Vaccination
Sees doctor
Cost = Cost + 10
Cost
0.001
Not hospitalized
Cost = Cost + 0
Cost
0.999
Cost = Cost + 12
Hospitalized
0.03
No doctor
Cost = Cost + 0
0.9
Cost
0.001
Not hospitalized
Cost = Cost + 0
Cost
0.999
Figure 2 A more complete decision analysis tree comparing supportive care to vaccination.
International Encyclopedia of Public Health, First Edition (2008), vol. 2, pp. 71-76
Year
Women surviving
1
2
3
4
5
6
Total
10
9
8
7
6
5
Years lived in
intervala
Cost of treating
breast cancer
10
9.5
8.5
7.5
6.5
5.5
48
$100 000
$90 000
$80 000
$70 000
$60 000
$50 000
$450 000
Well
1 QALY
$0
Sick
0.5 QALY
$10 000
Dead
0 QALY
$0
International Encyclopedia of Public Health, First Edition (2008), vol. 2, pp. 71-76
Sensitivity Analysis
The value of decision analysis model inputs can be very
difficult to establish with absolute certainty. For instance,
health-related quality of life scores a key ingredient in
the calculation of QALYs can be obtained from various
instruments, each producing a slightly different number
(Gold and Muennig, 2002). Because these differences
arise due to differences in the way the studies are
designed, they represent a form of nonrandom error.
Most model inputs will be derived from a sample, and
therefore also contain random error.
We can usually guess the range of plausible values
within which the true value might lie. For instance, looking through the literature, we might see that the healthrelated quality of life score for the disease we are studying
varies by roughly 20% when using different instruments.
We might also know the standard error in datasets or
published values.
The inputs we are least likely to be certain about are
the assumptions we have made. In the earlier influenza
example, we might have made an assumption about the
amount of time it takes for a nurse to administer the
influenza vaccine. There are other values that we might
be fairly confident about, such as the cost of the influenza
vaccine itself. (The average wholesale price is usually easy
to get.)
The parameters the researcher is least certain about
should be tested over the widest range of values (because
it is plausible that the values are much higher or much
lower than our baseline estimate). Parameters that the
researcher is somewhat more confident about can be
tested over a narrower range of values. When a particular
strategy remains dominant over the range of plausible
values for the inputs that we are uncertain about, the
model is said to be robust.
There are many different ways of testing variables in a
sensitivity analysis. These include a one-way (univariate)
sensitivity analysis, in which a single variable is tested
over its range of plausible values while all other variables
are held at a constant value; a two-way (bivariate) sensitivity analysis, in which two variables are simultaneously
tested over their range of plausible values while all others
are held constant; a multi-way sensitivity analysis, in
75
$90.00
$80.00
$70.00
$60.00
$50.00
$40.00
$30.00
$20.00
$10.00
$0.00
6.99
Treat
Vaccinate
International Encyclopedia of Public Health, First Edition (2008), vol. 2, pp. 71-76
Summary
Decision analysis is a formalized approach to making optimal choices under conditions of uncertainty. It allows the
user to enter costs, probabilities, and health-related quality
of life values among other inputs of interest, and then
calculates probabilistically weighted means of these outcome measures. In public health, these outcome measures
usually include costs and QALYs. Typically, therefore, decision analysis is the heart of cost-effectiveness analyses in
public health and medicine (Gold et al., 1998).
However, just about any outcome measure can be modeled, including vaccine-preventable illnesses averted,
deaths avoided, and so forth. Therefore, local health departments, pharmaceutical companies, or other agencies can use
decision analysis for internal decision-making processes.
Decision analysis is often used by non-health businesses
interested in deciding whether they should release a product, perform internal restructuring, and so forth.
One great strength of decision analysis modeling is that
it allows for the calculation of a range of possible values
around a given mean. This approach, called sensitivity
analysis, allows the user to better understand the chances
that he or she will make a bad decision if a given strategy
is taken.
Decision analysis, like cost-effectiveness analysis, is
highly dependent on the accuracy and completeness of
model inputs, as well as the assumptions that the analysts
make. Drugs can have unforeseen side effects, or interventions can have long-term costs that may not be apparent to the analysts. Any of these effects can lead to
suboptimal outcomes.
For instance, the optimal treatment strategy for tuberculosis in most instances is a low-cost combination of
medications that can be effectively delivered in developing
countries. By using the most cost-effective medications, it
is possible to maximize the number of lives saved within a
given budget. However, as Farmer points out, these medications will be wasted if delivered to populations with a
high percentage of drug-resistant tuberculosis (Farmer,
2004). Therefore, decision analysis and cost-effectiveness
analysis must be viewed as an adjunct to optimal decision
making rather than the final word in health policy.
Citations
Drummond MF, OBrien BO, Stoddart GL, and Torrance GW (2005)
Methods for the Economic Evaluation of Health Care Programmes,
3rd edn. London: Oxford University Press.
Farmer P (2004) Pathologies of Power: Health, Human Rights, and the
New War on the Poor. Berkeley, CA: University of California Press.
Gold M, Siegel J, Russell L, and Weinstein M (1996) Cost-Effectiveness
in Health and Medicine. New York: Oxford University Press.
Gold MR, Franks P, McCoy KI, and Fryback DG (1998) Toward
consistency in cost-utility analyses: Using national measures to
create condition-specific values. Medical Care 36(6): 778792.
Gold MR and Muennig P (2002) Measure-dependent variation in burden
of disease estimates: Implications for policy. Medical Care 40(3):
260266.
Halpern EF, Weinstein MC, Hunink MG, and Gazelle GS (2000)
Representing both first- and second-order uncertainties by Monte
Carlo simulation for groups of patients. Medical Decision Making
20(3): 314322.
Mandelblatt JS, Schechter CB, Yabroff KR, et al. (2004) Benefits and
costs of interventions to improve breast cancer outcomes in African
American women. Journal of Clinical Oncology 22(13): 25542566.
Muennig P (2007) Cost-Effectiveness Analysis in Health, a Practical
Approach. San Francisco, CA: Jossey-Bass.
Muennig P, Pallin D, Sell RL, and Chan MS (1999) The cost
effectiveness of strategies for the treatment of intestinal parasites in
immigrants. New England Journal of Medicine 340(10): 773779.
Muennig P, Pallin D, Challah C, and Khan K (2004) The costeffectiveness of ivermectin vs. albendazole in the presumptive
treatment of strongyloidiasis in immigrants to the United States.
Epidemiology and Infection 132(6): 10551063.
Muennig P, Franks P, and Gold M (2005) The cost effectiveness of
health insurance. American Journal of Preventive Medicine 28(1):
5964.
Muennig PA and Khan K (2001) Cost-effectiveness of vaccination
versus treatment of influenza in healthy adolescents and adults.
Clinical and Infectious Disease 33(11): 18791885.
Sonnenberg FA and Beck JR (1993) Markov models in medical decision
making: A practical guide. Medical Decision Making 13(4): 322338.
Further Reading
Drummond MF, OBrien BO, Stoddart GL, and Torrance GW (2005)
Methods for the Economic Evaluation of Health Care Programmes,
3rd edn. London: Oxford University Press.
Gold MR, Siegel JE, Russell LB and Weinstein MC (eds.) (1996)
Cost-Effectiveness in Health and Medicine. New York: Oxford
University Press.
Hunink M, Glasziou P, Siegel JE, et al. (2001) Decision Making in Health
and Medicine. Cambridge, UK: Cambridge University Press.
Muennig P (2002) Designing and Conducting Cost-Effectiveness
Analysis in Health and Medicine. San Francisco, CA: Jossey-Bass.
Muennig P (2007) Cost-Effectiveness Analysis in Health, a Practical
Approach. San Francisco, CA: Jossey-Bass.
Sonnenberg FA and Beck JR (1993) Markov models in medical decision
making: A practical guide. Medical Decision Making 13(4): 322.
Weinstein MC, Fineberg HV, Elstein AS, et al. (1980) Clinical Decision
Analysis. Philadelphia, PA: W. B. Saunders.
International Encyclopedia of Public Health, First Edition (2008), vol. 2, pp. 71-76