Transcript section of decision sciences and clinical systems modeling

Conference on Evidence-based Public Policy
Ottawa, Canada October 24th-25th, 2005
Risk Preferences, Perceptions and Medical
Decision Making
Mark S. Roberts, MD, MPP
[email protected]
Associate Professor of Medicine,
Health Policy and Management and Industrial Engineering
Chief, Section of Decision Sciences and Clinical Systems Modeling
University of Pittsburgh
SDS
CSM
SECTION OF
DECISION SCIENCES
AND
CLINICAL SYSTEMS
MODELING
SDS-CSM
University of Pittsburgh
Outline of Talk
• Brief review of the use of clinical decision
analysis
• Example of a major health care disparity
• Description of work to evaluate whether risk
perceptions and preferences can help
explain/understand some disparities that are
observed
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Types of Decision Analysis
• Descriptive
– analyses and explanations of how decision makers
actually make decisions
• Involves decision psychology
• Ability to make predictions of future decisions is key element
• Prescriptive
– analyses of how decision makers should make
decisions (which options are optimal given that an
outcome criteria has been chosen)
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Context of Medical Decision Making
OUTCOME 1
VALUE 1
p1
CHOICE 1
p2
Specific
situation with
specified
choices and
outcomes
OUTCOME 2
OUTCOME 3
Patients have
preferences
for the
various
outcomes
VALUE 2
VALUE 3
p3
CHOICE 2
p4
OUTCOME 4
VALUE 4
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Decision Analysis: Analytic mechanics
OUTCOME 1
Expected value
of choice 1
CHOICE 1
VALUE 1
p1
p2
OUTCOME 2
Specific
situation with
specified
choices and
outcomes
OUTCOME 3
VALUE 2
VALUE 3
p3
CHOICE 2
Expected value
of choice 2
p4
OUTCOME 4
VALUE 4
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Decision Analytic Foundations
• Fundamental assumption is that people maximize
utility
– Choose that branch which maximizes the expected
value of the outcome
– Or, at least, make decisions as if that is what they are
doing
• Standard versions of these analyses assume risk
neutrality
• So, why do we kneed to understand (and be able
to measure) this?
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Health Care Disparities
• Multiple studies have found that utilization of
health care procedures varies by characteristics
(race, gender, etc)
– Bypass surgery rates are lower in African Americans
than whites
– Women have fewer catheterizations during acute heart
attacks than men
• The reasons for this are multiple:
– Bias/racism
– Access
– Difference in outcome preferences
– Differences in risk preferences
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Knee Replacement
• Knee replacement is a proven therapy to decease
pain and increase mobility in patients with
osteoarthritis
– Most common surgical procedure in the US
• Dramatic differences in procedure utilization rates
have been found across gender and ethnicity
– African Americas, Hispanics use less than whites
– Women use more than men
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Raw rates by hospital referral region
Skinner et al in NEJM 2003;349:1350-1359.
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Disparities in Rates of Knee Arthroplasty
Rate/1000 patients
7
5.97
6
5
4
p <.001
5.37
4.82
4.84
p <.001
3.46
3
1.84
2
1
0
Men
Women
White
Hispanic
Black
Adapted from Skinner et al in NEJM 2003;349:1350-1359.
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Decision to have surgery: individual variation
• Two 78 year old women with
essentially identical knees by
x-ray, other comorbidities
– Woman A: “There is no way
I’m having an operation - I’m
too old!”
– Woman B: “I can’t bear the
pain - when can I have
surgery?”
• We would never call that
difference a disparity
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Structure of Knee replacement decision
Outcomes are uncertain,
preferences for risk matter
Die
Dead
p1
Knee
Replacement
Values for outcome
vary between people
p2
78 Year old
female with
degenerative
arthritis of
knee
No Pain
Success
p3
Live
p4
Pain
No Better
No Knee Replacement
Pain
This decision involves both risks and preferences
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Disparity or Differences in Preferences?
• We assume that there is a distribution of
preferences in the population
– For the outcome state
– For risk (risk averse vs risk seeking)
• To fully understand the presence of disparities, we
need to understand whether these preferences
systematically vary with race, gender, etc.
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Estimates of Outcome Preferences
• There are several methods for assessing
preferences for outcomes (all pretty poor!)
• Visual Analog Scale (easy, reproducible)
0
100
• Standard Gamble (closest to von Neumann-Morgenstern utilities, tradeoff
and risk)
• Time Tradeoff (tradeoff, but no risk)
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Estimates of outcome preferences
Death
“Gamble”
Standard gamble
p(death)
Perfect Health
Choose:
Sure Thing
HEALTH STATE
Particular Health State
Time trade off
Perfect Health
0
Death
8
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
10 Years
SDS-CSM
University of Pittsburgh
Empiric Estimates of Risk Preferences
• Remainder of the work has primarily been
conducted by Carol Stockman, PhD under a grant
from the NHLBI
• Examining disparities in acceptance of invasive
cardiovascular procedures
– Assessing risk preferences and perceptions
– Assessing values of the outcome state
– Assessing belief in luck
• Evaluate whether preferences and risk differ
systematically by ethnicity and gender
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Methods
• Patients undergoing nuclear stress test to evaluate
cardiovascular problems
• 2 risk preferences instruments
– Health: Decisions over drugs to treat angina or
shortness of breath
– Monetary: Decisions over $5 lottery tickets
• Belief in luck scale
• Demographic questionnaire
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Methods, continued
• Health Risk Preferences Instrument
– For some of the subjects, the choices presented in the
health risk preferences instrument were purely
hypothetical
– For other subjects, who had either angina or shortness
of breath, the health scenarios instrument was tailored
to their symptom and its frequency
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Methods, continued
New Drug
p1
p2
Patient with
angina or
asthma
No improvement
Expected value of risky drug is
always the effectiveness of the
safe drug
Improvement of X%
Standard Drug
Improvement of p2*X%
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Risk Preferences over Health Outcomes
40
35
30
25
% 20
15
10
5
0
0
1
2
3
4
5
6
# Times Subject Chose Risky Drug A
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Risk Preferences over Health Outcomes
25
Mean = 0.346
Std. Dev. = 0.32332
N = 79
20
15
10
5
0
0.00
0.50
1.00
Percentage Risky Drug Choice
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Effect of hypothetical nature of question
Percentage of Subjects Choosing Risky Drug A, by
Probability Drug A works
70%
60%
50%
40%
30%
20%
10%
0%
10% 20% 30% 40% 45% 50% 55% 60% 70% 75% 80% 90%
Hypothetical
Actual Symptom-Based
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Risk Preferences over Money
• 12 lotteries
– All tickets paid $5.00 if you won
– You could sell your ticket for a price if there was a
buyer
– Certainty equivalents elicited using Becker-DeGrootMarschak mechanism
• Subjects are actually paid their earnings from the
sale or winnings from the lottery
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Risk Seeking
5.0
Certainty Equivalent
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0
0.5 1.0
1.5 2.0
2.5 3.0
3.5 4.0
4.5 5.0
Expected Value of Lottery
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Risk Averse
5.0
Certainty Equivalent
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0
0.5 1.0
1.5 2.0
2.5 3.0
3.5 4.0
4.5 5.0
Expected Value of Lottery
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Average over all patients
5.0
Certainty Equivalent
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0
0.5 1.0
1.5 2.0
2.5 3.0
3.5 4.0
4.5 5.0
Expected Value of Lottery
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Correlation between risk preferences
% Monetary Lotteries with CE> EV
1.00
0.80
ρ= -.039,
p=.74
0.60
0.40
0.20
0.00
0.00
0.20
0.40
0.60
0.80
1.00
% Health Choices where Risky Drug Chosen
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Belief in Luck Survey
• 28 questions measuring 5 concepts related to
belief in luck
–
–
–
–
–
Belief in personal good luck
Belief in personal bad luck
Belief in Others’ luck
General Belief in luck
Influence of Luck on Decision Making
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Belief in good and bad luck
20
20
15
15
10
10
5
5
0
0
0.00
1.00
2.00
3.00
Personal Bad Luck
4.00
5.00
0.00
1.00
2.00
3.00
4.00
5.00
Personal Good Luck
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Ethnic Differences in Belief in Luck
African
Caucasian
American
I consider myself to be an unlucky person.
The things in life I can’t control tend not to
go my way because I’m unlucky.
I mind leaving things to chance because I’m
an unlucky person.
Some people are consistently lucky and
others are unlucky.
Luck never works in my favor.
Luck works in my favor.
p value
30%
18%
.08
17%
8%
.09
17%
6%
.02
63%
46%
.03
21%
48%
10%
63%
.05
.05
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Effects on decision making
• We have not yet finished the acquisition of
patients in the CV study
• We have found some differences in these measures
across:
– Gender
– Race
– Patient type
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING
SDS-CSM
University of Pittsburgh
Conclusions
• People’s risk preferences are different over
–
–
–
–
Monetary outcomes
Health outcomes
Risk preferences vary over level of risk
Risk preferences very with respect to knowledge of
state
• Risk and luck appear different to many people
– Risk is an attribute of the world
– Luck is an attribute of the person
SECTION OF DECISION SCIENCES AND CLINICAL SYSTEMS MODELING