Decision Analysis

1
Decision Analysis
2
Objectives

• By the end of this session you will
• Be aware of/reminded about the concept of
  decision analysis
• Have seen examples of its use
• Be better enabled to consider in wider terms how
  and why decisions in medicine are made

3
Definition of decision analysis

• The application of explicit quantitative methods
  to analyse decisions made under conditions of
  uncertainty. A decision analysis model must
  compare at least two decision options. The
  process involves identifying all the available
  management options, and the potential outcomes of
  each, in a series of decisions that have to be
  made about patient care. The range of choices are
  plotted on a decision tree.

4
Definition of decision tree

• Illustrates all the potential choices and
  subsequent outcomes in diagrammatic form. The
  decisions and outcomes are presented in the order
  in which they are likely to occur, hence it is
  hierarchical in structure

5
How does one start?

• Commonly accepted format is a tree diagram.

Treat
UTI?
Dont Treat
Decision Node
6
Definition of decision node

• A point in a decision tree where a decision has
  to be made. Generally illustrated by a square.
  The lines emanating from a decision node
  represent the clinical strategies being compared.

7
Decision data
Chance Node
better
0.9
Treat
Not better
0.1
UTI?
better
0.5
Dont Treat
Not better
0.5
8
Definition of a chance node

• Chance events that may occur following a
  decision. Generally illustrated by a circle.

9
Definition of an outcome node

• The final outcome of a decision path. Generally
  illustrated by a rectangle or triangle.

10
Probability

• The chance of the event occurring. The
  probabilities resulting from a chance node must
  add up to 1.0

11
Result 0.9 is better than 0.5
Chance Node
better
0.9
Treat
Not better
0.1
UTI?
better
0.5
Dont Treat
Not better
0.5
12
Urinary Tract Infection

• Patient presents with symptoms
• What would happen if you opted for one path in
  preference to another?

13
How do we develop a detailed decision analysis
tree for UTI?

• Need to know all the baseline data for UTI (see
  Fenwick BJGP 2000).
• What proportion of patients with typical symptoms
  have UTI?
• What is the sensitivity and specificity of
  dipstix urine analysis (different sticks).
• What is the success rate of antibiotic treatment?

14
Add reality to probability scores

• Cost of tests
• Cost of treatment
• Days lost from work
• Cost of re attending clinician
• Multiply the probabilities by the costs

15
Cost Data 2 for antibiotic, 20 per consultation
Costs
220
better
0.9
Treat
2 2020
Not better
0.1
UTI?
better
20
0.5
Dont Treat
Not better
2020
0.5
16
Rollback Costs
22
(22x0.9) (42x0.1)24
better
0.9
Treat
42
Not better
0.1
UTI?
better
20
0.5
Dont Treat
Not better
40
0.5
(20x0.5) (40x0.5)30
17
What does the patient think?

• Utilities (The preference or desirability of a
  particular outcome)..these may vary between
  individuals
• Consider how would you feel regarding whether or
  not to have antibiotics for a suspected UTI?

18
Utility

• The preference or desirability of a particular
  outcome

19
Utilities
Utilities
1
better
0.9
Treat
0.7?
Not better
0.1
UTI?
better
1
0.2
Dont Treat
Not better
0.7?
0.8
20
Rollback
1
(0.9x1) (0.7x0.1)0.97
better
0.9
Treat
0.7
Not better
0.1
UTI?
better
1
0.5
Dont Treat
Not better
0.7
0.5
(0.5x1)(0.5x0.7) 0.85
21
Results

• More people get better (90 vs 50)
• It is cheaper (24 vs 30)
• The utilities are better (0.97 vs 0.85)
• Probably should treat??
• We have not considered the negative aspects of
  prescribing antibiotics

22
UTI What are the options?

• Treat on symptoms alone?
• Treat after urine analysis in surgery?
• Treat after msu results available?
• Treat but send off msu anyway?

23
Can I apply the results to my patient?

• Do the probability estimates fit my patients'
  clinical features?
• Do the utilities reflect how my patients would
  value the outcomes of the decision?

24
Reflections

• The power of decision analysis is the ability to
  change the utilities and probabilities
• You can watch how this affects the decision node
• Thus it should be seen as a dynamic tool