Decision Modeling Techniques
HINF 371 - Medical Methodologies
Session 3
Objective

To review decision modeling
techniques and discuss their use in
healthcare decision making
Reading

Roberts M S and Sonnenberg F A (2000) Chapter 2:
Decision Modeling Techniques, in Chapman G B and
Sonnenberg F A (eds) Decision Making In Health
Care: Theory, Psychology and Applications,
Cambridge University Press, USA,
Evidence Preparation
Engine where data
is translated into
information
Why do we need them?
To create a quantitative
representation of clinical choices
 To compare alternatives and results of
choices
 To integrate data from various
sources to describe a clinical situation
 To simulate trial results to the whole
population

Requirements for a Decision
Model
Perspective: identification of whose
perspective has been used to develop
the model
 Context: who is involved, what
conditions, what interventions
 Complexity (or granularity): what
should be the level of detail
 Time horizon

Simple Decision Tree
Chance Node
Outcome 1
Value 1 (U1)
p1
Decision Node
Choice 1
p2
Outcome 2
Outcome 3
Value 2 (U2)
Total = 1
Value 3 (U3)
p3
Choice 2
p4 Outcome 4
Value 4 (U4)
Terminology
Sensitivity
HIV+
Test +
p1
p2
p1 HIV+
HIVHIV+
Test -
LERx True Positive LERx
Test +
p3
p4
HIV-
LEToxFalse PositiveLETox
False Negative
LE LateRx
LE LateRx
LE True Negative LE
Specificity
Test -
p2
Test +
p3 HIVTest -
p4
Example
3.5444
HIV+
0.9988
p1
0.4856
p2
Screen 39.2050
LERx
3.5 QALYs
LETox
39.4 QALYs
LE LateRx
2.75 QALYs
Test +
HIV-
0.0012
HIV+
p3
0.5144
p4
Test -
21.89
21.5250
21.53
HIV-
HIV+
p5
HIV-
40.3 QALYs
0.5
LE Late Rx
2.75 QALYs
0.5
LE
40.3 QALYs
No Screenp
6
LE
Influence Diagrams
Test
Result
HIV
Status
Screen for
HIV
Yes/No
Treat for
HIV
Yes/No
Life
Expec
Sensitivity Analysis
Test +
LERx
HIV+ p1
p2
Test Test +
HIV- p3
p4
Test -
LETox
LE LateRx
LE
Markov Processes
Iterative in time, can be repeated
until everybody in the absorbing state
 Based on the probabilities of change
in status
 Three states

Recurrent state
 Transient state
 Absorbing state

Markov Processes
HIV+
Asymptomatic
HIV+
p3
p2
p4
p1
p1
AIDS
HIV+
p2
AIDS
p3
DEAD
p4
AIDS
p5
DEAD
AIDS
DEAD
p5
p1
P6
DEAD
P6
DEAD
AIDS
HIV+
Test +
LERx
HIV+
p1
p2
DEAD
HIV-
Screen
LETox
HIV-
DEAD
AIDS
HIV+
Test -
p3
p4
HIV+
LE LateRx
HIV+
DEAD
HIV-
LE
HIV-
LE Late Rx
HIV+
DEAD
AIDS
p5
DEAD
No Screenp
6
LE
HIV-
HIV-
DEAD
Alternatives to Markov
Processes





Markov Processes has no memory and
based on discrete snapshots in time
Semi Markov Processes – time is
continuous, one does not move to the next
another stage in the next term and
measures holding times
Individual Simulations as a solution:
simulates individuals’ travel
Dynamic influence diagrams creates a new
influence diagram for the next cycle
Discrete event simulation: what is possible
to do