Transcript EPI-820_Lect6_CDA_Ro..

Medical Decision Making
• Although decision making is a prime
activity of physicians, little time is spent on
this subject in medical school curricula
• This lecture deals with the quantitative
methods that can be used to aid physicians
in decision making under uncertainty
Questions for the Decision Maker
• Efficacy and Effectiveness of Dx Tech
• Evaluation of outcomes of care not process
• New sources of information, written, www,
other high tech
• Rational limits to memory
• Understandability of probability concepts
by Pat and Phy. Understand uncertainty
PRESSURES FOR QUANTITATIVE MEASURES
TO AID HUMAN JUDGMENT
1. Economic
2. Information explosion and limitations of human
ability to process information
3. Advent of Computers
4. Widespread use of statistics in literature
5. Bayes' theorem
6. Expected utility theory
CLASSICAL METHODS OF PHYSICIAN
DECISION MAKING

Pathophysiological-hypothetical-deductive

Pattern Matching

Exhaustive (big banana)
Sources of Uncertainty
 Normal statistical variation. Concept of normal
distribution curve
 Errors in clinical data
a. Memory
b. Inattention
c. Transcription
 Variation in interpretation
 Uncertainty about meaning of clinical data
 Uncertainty about effects of treatment
HUMAN BIASES LIMIT ABILITY
OF PHYSICIANS TO BE RATIONAL

Rationality defined for this course

Expected utility theory (decision trees)

Framing effects (prospect theory)

Representativeness and availability

Anchoring and adjusting

Regression toward the mean

Conditional Probability
Rationality Defined
• Based upon decision maker’s current assets
• Based upon possible consequences of
choice
• When consequences uncertain, evaluated
without violating basic rules of probability
theory
Prospect Theory
(Kahneman and Tversky)
Imagine I will give you $200.00.
I now offer you more in the form of 1 of 2 options:
1. I will give you an additional $100.00
2. I will toss a fair coin. If it lands heads, I will
give you an additional $200.00. If it lands tails I
will give you nothing.
Which offer would you choose?
Prospect Theory (cont.)
(Kahneman and Tversky)
Now imagine I will give you $400.00, but there is a
penalty involved. Choose between these two
options:
1. You must give me back $100.00.
2. Toss a fair coin. If it lands heads, you must give
me back $200.00. If it lands tails, you may keep
the $400.00.
Representativeness Difficulties
•
•
•
•
Prior probability not taken into account
Insensitivity to sample size
Misconstruing randomness
Not taking disconfirmatory evidence into
account
• Misconstruing regression to the mean
Imagine you have the following
scenario
1. If program A is adopted, 200 people will be
saved
2. If program B is adopted, there is a 1/3
probability that 600 people will be saved
and a 2/3 probability that no people will be
saved
Now compare the following
1. If program C is adopted, 400 people will
die
2. If program D is adopted, there is a 1/3
probability that no one will die and a 2/3
probability that 600 people will die
Which Sequence Looks More Like
a Random Sequence
1.
OXOXOXOOXOOXOXXOXOXX
2.
XOOXXXOOXOOOXXOOXXOO
Availability
• When data are recent, familiar or salient
they are more easily retrieved from memory
and therefore seem more real than
naturalistic frequencies
Anchoring and Adjusting
• Original estimates of probability biased
because of availability and
representativeness heuristics
• When asked to adjust estimates based upon
new information unlikely to apply correct
amount
Correct Use of
Representativeness
•
•
•
•
•
Evaluate how good are the cues
Are they independent
What is their prior probability
Is the sample large enough
Evaluate whether the results may be only
regression to the mean
CLINICAL DECISION ANALYSIS
A Systematic Approach to Decision
Making Under Uncertainty
1. EXPLICIT
2. QUANTITATIVE
3. PRESCRIPTIVE
ELEMENTS OF CLINICAL DECISION ANALYSIS
1. Identify and bound problem
2. Structure problem temporally
3. Obtain data:
Probabilities
Utilities
4. Choose preferred action
Decision Tree Concepts
 Time flows from left to right
 All outcomes must be represented
 Three major types of nodes
a. terminal
b. decision
c. chance
 Branches emanating from a decision node must
represent all the options. Each must be nonoverlapping
 Branches emanating from a chance node must
represent all the possible outcomes, must be
non-overlapping, and must sum to 1.0
 Data includes the bounded problem,
probabilities and Utilities
Sample Problem
Mrs. Hull
• 56 year old,white female,160#,5’7” tall
• Smokes 1.5 pack cigarettes daily
• Increasing frequent episodes of typical
angina, relieved by nitroglycerine
• Complete physical exam normal
• Special studies performed
Mrs. Hull (cont1)
• Coronary angiography shows 2 vessel
disease
• Question of therapy and outcomes
• Medical vs. surgical therapy
SURGERY
CABG Medical or Surgical Rx
MEDICAL
[+]
[+]
OpDeath
SURGERY
CABG Medical or Surgical Rx
pOpDeath
Survive
#
MNoSurv3
MEDICAL
pMNoSurv3
MSurv3
#
UDead
[+]
UDead
[+]
OpDeath
SURGERY
UDead
pOpDeath
NoSurv3
Survive
#
pNoSurv3
Surv3
CABG Medical or Surgical Rx
UDead
[+]
#
MNoSurv3
pMNoSurv3
MEDICAL
UDead
MSymL1
MSurv3
#
pMSymL1
MSymL2
pMSymL2
MSymL3
#
USymL1
USymL2
USymL3
OpDeath
UDead
pOpDeath
SURGERY
NoSurv3
pNoSurv3
Survive
UDead
SymL1
#
pSymL1
SymL2
Surv3
CABG M edical or Surgical Rx
#
pSymL2
SymL3
#
M NoSurv3
pM NoSurv3
M EDICAL
UDead
M SymL1
M Surv3
#
pM SymL1
M SymL2
pM SymL2
M SymL3
#
USymL1
USymL2
USymL3
USymL1
USymL2
USymL3
Mrs. Hull (cont2)
• Perioperative mortality 2.5%
• Three year mortality 4.8% after surgery and
7.2% after medical therapy
• Probabilities of 3 levels of symptoms and
death as outcomes
Mrs. Hull Outcomes
Treatment Level 1
Level 2
Level 3
Surgical
P=.41
P=.50
P=.09
Medical
P=.30
P=.60
P=.10
Utilities
1.0
0.8
0.4
Sensitivity Analysis on
Prob dying in periop period
Expected Value
0.820
SURGERY
0.800
M EDICAL
0.780
Threshold Values:
Prob dying in periop period = 0.0386
EV = 0.800
0.760
0.740
0.720
0.700
0.680
0.660
0.0150
0.0705
0.1260
0.1815
Prob dying in periop period
DECISION THEORY
Pros:
Flexible, easily adapted to various problems
Well suited to analyze lack of consensus
Linked to risk/benefit analysis and
cost/effectiveness analysis
Structuring process helps improve decisions
DECISION THEORY
Cons:
Unfamiliar task - Estimating probabilities
Assessing utilities
Forces precision where world is/maybe fuzzy
Normative - Dictates structure of thinking
May emphasize difficulties with human biases