Medical Decision Making: Getting Started Primer On Medical Decision Analysis: Part 1

Medical Decision Making
http://mdm.sagepub.com/
Primer on Medical Decision Analysis: Part 1Getting Started
Allan S. Detsky, Gary Naglie, Murray D. Krahn, David Naimark and Donald A. Redelmeier
Med Decis Making 1997 17: 123
DOI: 10.1177/0272989X9701700201
The online version of this article can be found at:
http://mdm.sagepub.com/content/17/2/123
Published by:
http://www.sagepublications.com
On behalf of:
Society for Medical Decision Making
Additional services and information for Medical Decision Making can be found at:
Email Alerts: http://mdm.sagepub.com/cgi/alerts
Subscriptions: http://mdm.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://mdm.sagepub.com/content/17/2/123.refs.html
>> Version of Record - Apr 1, 1997

What is This?
Downloaded from mdm.sagepub.com at HAM-TMC LIBRARY on November 13, 2013
Tutorial:
How to
Primer on Medical Decision

Part
Perform
Decision
Analysis
Analysis:
1—Getting Started
ALLAN S. DETSKY, MD, PhD, GARY NAGLIE, MD,

MURRAY D. KRAHN, MD, MSc, DAVID NAIMARK, MD,
DONALD A. REDELMEIER, MD, MS(HSR)
This paper is Part 1 of a five-part series covering practical issues in the performance
of decision analysis. The intended audience is individuals who are learning how to
perform decision analyses, not just read them. The series assumes familiarity with the
basic concepts of decision analysis. It imparts many of the recommendations the authors have learned in teaching a one-semester course in decision analysis to graduate
students. Part 1 introduces the topic and covers questions such as choosing an appropriate question, determining the tradeoff between accuracy and simplicity, and deciding on a time frame. Key words: decision analysis; expected value; utility; sensitivity
analysis; decision trees; probability. (Med Decis Making 1997;17:123-125)
learned has
from teaching our students how

model.
The intended audience for this series is individuals who are learning how to perform decision analyses. The series assumes that the reader is already
familiar with the concepts of decision analysis and
has read and comprehended several decision analyses before trying his or her hand at one. Before
embarking on this series, the reader should read a
two-part users guide aimed at consumers of decision analyses 15 16 to achieve a grounding in basic
concepts such as decision nodes, probability nodes,
expected value, tree representation, sensitivity, and
threshold analyses, and is also referred to two textbooks.1718 The goal of the present series is to give
practical suggestions for performing decision anal-
Over the past ten years, our group at the University

of Toronto has taught a one-semester course on decision analysis. Unlike educational initiatives that
solely focus on readings about and discussions of
to
decision analysis, our course has required that each

student perform a decision analysis. This requires
building a decision tree on paper, modeling that tree
on a computer, obtaining probabilities and utilities
by direct measurement or literature review, evaluating the tree, performing sensitivity analyses, presenting the study to peers, and writing up the analysis. Several of the students have gone on to publish
their work.1-s To pass the course, each student is
required to develop a model that &dquo;works.&dquo;
In this series9-&dquo; and the accompanying note on
presentation, 14 we attempt to impart what we, as
teachers, have learned about the practical issues of
performing decision analysis. Much of what we have
develop
come
working
ysis.
When Is Decision
Received December 27, 1995, from the University of Toronto

Programme in Clinical Epidemiology and Health Care Research
(The Toronto Hospital and The Sunnybrook Health Science Centre Units) and the Departments of Health Administration, Medicine, and Clinical Biochemistry, Toronto, Ontario, Canada. Revision accepted for publication December 12, 1996. Drs Detsky
and Redelmeier are partially supported by Career Awards from
the National Health Research and Development Programme and
the Ontario Ministry of Health, respectively. Drs. Naglie and
Krahn are partially supported by Arthur Bond Fellowships from
the Physicians Services Incorporated Foundation.
Address correspondence and reprint requests to Dr. Detsky.
EN G-246, General Division, The Toronto Hospital, 200 Elizabeth
Street, Toronto, ON M5G 2C4, Canada e-mail. (detsky@utstat.
Analysis Appropriate?
One of the first steps in a decision analysis is to

decide whether the technique is appropriate for the
given question. Two issues merit special attention.
First, there should be some uncertainty about the
appropriate clinical strategy for patients with a given
health state. There are some circumstances in clinical medicine when published primary clinical evidence already clearly identifies the optimal clinical
strategy. For example, patients who have carotid artery stenosis of more than 70% after suffering transient ischemic attacks clearly benefit from carotid
endarterectomy, as demonstrated by the NASCET
toronto.edu).
123
124
Trial.9 One need

this
not
perform
decision
analysis of
most clinical decisions
question. However,
clearly supported by
are
direct evidence.
Sometimes randomized trials have been performed but the optimal decision remains uncertain
because several outcomes are involved. The randomized trials of anticoagulation in atrial fibrillation, for example, show a reduction in the risk of
stroke but an increased risk of bleeding and, in
some trials, a considerable dropout because patients
do not wish to be on coumadin for indefinite periods of time. A decision analysis20 can incorporate
these three outcomes as well as possible different
levels of risk depending on patient characteristics
(e.g., age, other existing cardiac disease). Such a
model can help the decision maker understand the
not as
risk-benefit tradeoff more explicitly.

The second main issue in assessing the potential
value of a decision analysis is to ensure that there is
a meaningful tradeoff in the problem. Decision analyses always require comparison of at least two clinical strategies. One strategy ought to contain advantages and countervailing disadvantages. If one
strategy clearly dominates the other because it
re-
sults in lower rates of adverse disease outcomes, less

risk of treatment side effects and better utilities, then
a decision analysis is not necessary. An example of
this situation might be a noninvasive diagnostic
technique (such
as
echocardiography) compared
with an equally accurate but more invasive diagnostic technique (such as angiography) for assessing
myocardial wall motion or thickness. Of course, decision-analytic models can also be used to estimate
the cost-effectiveness of interventions involving
dual outcomes such as costs and utilities. In this
circumstance, one strategy may clearly be better
than the other in the clinical sense but result in
higher expected costs. In this series, however, we
focus on models that have only one clinical outcome
measure.
A second
major consideration concerns availabilsome circumstances, analysts will wish
ity of data. In
models to better fit the

problem. However, data for such a
such
as a Markov model (described
tree,
complex
in Part 5 of this series)13 that uses transition probabilities from one state to another over successive
short time periods, are often unavailable. In these
circumstances, it may be wise to develop a simpler
model. Like others,21 we have often found that the
insights derived from a simple model are similar to
those derived from a complex formulation. We encourage beginners to develop the simplest models.
How many strategic alternatives should be included ? Some clinical questions require consideration of multiple combinations of several interventions. For example, in suspected pulmonary
embolism, a simple model would consider empirically anticoagulating the patient, or performing tests
such as a ventilation-perfusion lung scan, venous
to
develop
complex
more
nature of the
doppler ultrasonography, impedence plethysmography, and angiography. The various combinations

would result in dozens of possible strategies. The
best approach for modeling such a circumstance is
to
limit the number of
options
to
those that
are
clearly different from each other and cover the spectrum of the problem. Thus, one might consider only
three options in the pulmonary embolism example:
doing nothing further, anticoagulating without further tests, and performing a ventilation-perfusion
lung scan and going on to angiography only if its
result is neither clearly positive (high-probability
scan) nor clearly negative (normal scan). One can
always
add other
options
later
on.
Time Frame
In comparing outcomes of alternative strategies,
the analyst must determine an appropriate time
frame. When
considering alternative strategies for

treating chronic illnesses, the time
frame should be long. When considering therapeutic strategies to avoid short-term complications (e.g.,
preventing
Accuracy versus Simplicity

Decision-analytic models must be sufficiently
complex to incorporate the important events and
values, yet sufficiently simple to be understandable.
A decision tree is not a complete representation of
&dquo;the real world&dquo; but rather a simplified and highly
stylized model of the most important components.
To determine the appropriate level of complexity,
the analyst must consider whether the model captures the key issues necessary to fully describe the
risk-benefit tradeoff. If a key element of the riskbenefit tradeoff is missing, the model will not
achieve its goal of helping the reader understand the
tradeoff.
or
perioperative management issues),
the time frame
be brief. Thus, the time frame will

the nature of the clinical problem.
can
In
choosing
an
appropriate
time
depend upon
frame,
one con-
fronts the same tradeoff as noted in the above section on accuracy versus simplicity, namely desire for
completeness versus availability of data. Most clinical studies including randomized trials have relatively limited time frames. Even for trials involving
chronic diseases such as coronary artery disease,
the period of observation is usually between one and
three years and rarely as long as five years. Occasionally one gets reports of long-term follow-up, but
125
the observation periods in these reports rarely exceed ten years.22 Some cohort studies, such as the
Framingham Study, have much longer periods of
observation. One, therefore, must balance the desire
for complete long-term follow-up with the availability of valid and precise data. Most analysts will have
to extend the period of observation of a randomized
trial for their analysis and therefore will have to extrapolate the data. The performance of extensive
sensitivity analyses around these extrapolations is
References
HJ, Detsky AS, Armstrong PW Management of asymp-
tomatic chronic aortic
function:
regurgitation with left ventricular dysanalysis. J Gen Intern Med. 1990;5:394-
decision
401.
2. Krahn M,
Basinski AS, Detsky AS Comparison of

a less aggressive (Canadian) policy
for cholesterol screening and treatment. Ann Intern Med.
an
Naylor CD,
aggressive (U.S ) and
1991;115:248-55.
UD, Guerriere M, Read SE, Detsky AS. Percutaneous
injuries among health care workers: the real value of HIV
testing of "donor" blood. Arch Intern Med. 1991;151:2033-
3. Allen
Krahn MD, Naimark D, Redelmeier, Detsky AS. Primedical decision analysis: Part 3—estimating probabilities and utilities. Med Decis Making. 1997;17.136-41.
12. Krahn MD, Naglie G, Naimark D, Redelmeier D, Detsky AS
Primer on medical decision analysis: Part 4—analyzing the
model and interpreting the results. Med Decis Making. 1997;
11
17:142-51.
13
esophagogastrectomy
with
without
or
drainage?
18.
analysis.
J Rheumatol.
pneumonia
decision
analysis
Am Rev
Respir
Dis
Endarterectomy Trial
Engl J Med. 1991;325:445-53.

Naglie IG, Detsky AS. Treatment of chronic nonvalvular atrial
fibrillation in the elderly: a decision analysis. Med Decis
Making. 1992;12:239-49.
Moskowitz AJ, Dunn VH, Lau J, Pauker SG. Can "hypersimplified" decision trees be used instead of Markov models?
[abstr]. Med Decis Making. 1984;4:530
stenosis. N
20
Yusuf S, Zucker D, Peduzzi P, et al. Effect of coronary artery

bypass surgery on survival overview of 10-year results from
randomized trials by the Coronary Artery Bypass Graft Surgery Trialists Collaboration. Lancet 1994,344:563-70.
1993;148:370-7
Barrett BJ, Parfrey PS, Foley RN, Detsky AS. An economic
analysis
of
strategies
for the
use
of contrast media for
nostic cardiac catheterization. Med Decis

325-35.
Making.
diag-
1994,14
Krahn MD, Naimark D, Redelmeier D.

Primer on medical decision analysis: Part 1—getting started
Med Decis Making. 1997,17.123-5.
10. Detsky AS, Naglie G, Krahn MD, Redelmeier D, Naimark D.
9.
Carotid
terectomy in symptomatic patients with high-grade carotid
22.
JV, Biem HJ, Detsky AS Bronchoscopy vs empirical therapy in HIV-infected patients with presumptive Pneumocystis
Symptomatic
(NASCET) Collaborators. Beneficial effect of carotid endar-
1992,19.
7. Tu
Phil-
PA: W. B.
19. North American
Management of suspected giant
1220-8.
Analysis.
Saunders, 1980.
Sox H, Blatt MA, Higgins MC, Marton KI. Medical Decision
Making. London, England: Butterworth & Co., 1988.
adelphia,
21
9-13
6. Buchbinder R, Detsky AS.
cell arteritis. a decision
JAMA 1995,273.1610-3.
MC, Fineberg HV. Clinical Decision
cability
17 Weinstein
Ann
Thorac Surg. 1992;53:493-7.

5. Warde P, Detsky AS. Is steroid receptor data useful in patients with metastatic breast cancer. Am J Med Sci. 1992;304
Naimark D, Krahn MD, Naglie G, Redelmeier D, Detsky AS.

Primer on medical decision analysis: Part 5—working with
Markov processes. Med Decis Making. 1997;17:152-9.
Redelmeier DA, Detsky AS, Krahn MD, Naimark D, Naglie G.

Guidelines for verbal presentations of medical decision analyses. Med Decis Making. 1997;17.228-30.
15 Richardson WS, Detsky AS, and the Evidence-Based Medicine
Working Group. Users guides to the medical literature. VII
How to use a clinical decision analysis. A. Are the results of
the study valid? JAMA. 1995;273:1292-5.
16. Richardson WS, Detsky AS, and the Evidence-Based Medicine
Working Group. Users guides to the medical literature VII.
How to use a clinical decision analysis. B Results and appli-
Olak J, Detsky AS Surgical decision analysis: esophagectomy/
carimi
Naglie G,
mer on
40.
4.
2—building a tree.
14.
important.
1. Biem
Primer on medical decision analysis: Part

Med Decis Making. 1997;17:126-35.
Detsky AS, Naglie G,
Glossary
Markov Model: A decision-analytic model that involves a Markov process, a modeling technique derived from matrix alge-
bra, which describes the

among
tervals
or
transitions
number of health states
cycles.
cohort of patients make

a series of short in-
during

Medical Decision Making: Getting Started Primer On Medical Decision Analysis: Part 1

Uploaded by

Copyright:

Available Formats

Medical Decision Making: Getting Started Primer On Medical Decision Analysis: Part 1

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Medical Decision Making: Getting Started Primer On Medical Decision Analysis: Part 1

Uploaded by

Copyright:

Available Formats

Medical Decision Making

Primer on Medical Decision Analysis: Part 1Getting Started

Society for Medical Decision Making

>> Version of Record - Apr 1, 1997

Downloaded from mdm.sagepub.com at HAM-TMC LIBRARY on November 13, 2013

Primer on Medical Decision

ALLAN S. DETSKY, MD, PhD, GARY NAGLIE, MD,

from teaching our students how

Over the past ten years, our group at the University

decision analysis, our course has required that each

Received December 27, 1995, from the University of Toronto

One of the first steps in a decision analysis is to

Downloaded from mdm.sagepub.com at HAM-TMC LIBRARY on November 13, 2013

Trial.9 One need

most clinical decisions

risk-benefit tradeoff more explicitly.

strategy clearly dominates the other because it

sults in lower rates of adverse disease outcomes, less

major consideration concerns availabilsome circumstances, analysts will wish

models to better fit the

doppler ultrasonography, impedence plethysmography, and angiography. The various combinations

limit the number of

considering alternative strategies for

Accuracy versus Simplicity

perioperative management issues),

the time frame

be brief. Thus, the time frame will

Downloaded from mdm.sagepub.com at HAM-TMC LIBRARY on November 13, 2013

tomatic chronic aortic

regurgitation with left ventricular dysanalysis. J Gen Intern Med. 1990;5:394-

Basinski AS, Detsky AS Comparison of

aggressive (U.S ) and

Engl J Med. 1991;325:445-53.

Yusuf S, Zucker D, Peduzzi P, et al. Effect of coronary artery

of contrast media for

nostic cardiac catheterization. Med Decis

Krahn MD, Naimark D, Redelmeier D.

terectomy in symptomatic patients with high-grade carotid

(NASCET) Collaborators. Beneficial effect of carotid endar-

19. North American

Management of suspected giant

Thorac Surg. 1992;53:493-7.

Naimark D, Krahn MD, Naglie G, Redelmeier D, Detsky AS.

Redelmeier DA, Detsky AS, Krahn MD, Naimark D, Naglie G.

Olak J, Detsky AS Surgical decision analysis: esophagectomy/

Primer on medical decision analysis: Part

Detsky AS, Naglie G,

bra, which describes the

number of health states

Downloaded from mdm.sagepub.com at HAM-TMC LIBRARY on November 13, 2013

cohort of patients make

You might also like