Improving decision making at the
point of care: opportunities and
challenges
Christopher Saigal MD MPH
Associate Professor
Department of Urology
Geffen School of Medicine at UCLA
Approaches to decision making
How do we make decisions?
• Based on facts and figures: Apollonian
rationality?
• Gut instincts: Dionysian feeling?
• Both?
One model of decision making:
pure rationality
$900
Rush St
New hot
dog
stand
location?
$450
$500
LaSalle St
$200
“Expected value”
La Salle Street safe strategy:
(.9 x $500) + (.1 x $200)= $470/week
Rush Street risky strategy:
(0.2 x $900) + (0.8 x $450) = $540/week
The rational decider goes for the Rush
Street location
Is this a descriptive theory of
human
decision making?
• NO
• ‘behavioral economics’
• Framing biases/loss aversion
• Bubbles and panics
Intuitive decision making can be
key
• Many decisions are best executed in
response to gut feelings (“blink”)
• See a prairie fire coming towards you: run
to the river
• Without the orbitofrontal cortex, decision
making becomes impossible
Rational decision making can
be key
Some decisions are best made with a
rational framework
Which credit card:
- intro teaser rate of 2.9% for 1 year, then
goes to 16%
- intro rate of 4.9% that goes to 12% at one
year
Best model: useful combination
of both styles of decision
making
• Humans function decide best when
knowing which method to rely on- or when
to combine
Medical decision making
The double-edged sword
• Constant innovation in treatments for
patients
• Treatments can offer trade-offs
• Decisions have multiple moving parts
• Patient preferences and values are key
deciding factors in many situations
Decision choice for a man with moderate risk localized prostate cancer
Robotic
prostatectomy
surgery
If I choose surgery, I may leak
urine…if I choose surveillance, I
may worry about cancer
spreading
radiation
Active
surveillance
Open radical prostatectomy
‘experimental’
options (cryotherapy, primary
hormonal therapy, etc)
External beam radiotherapy
Brachytherapy
“Bounded rationality”
• Complex decision
• Time constraints
• Limits on human computational ability
“ A wealth of information creates a poverty of
attention”
Can software expand these “bounds?”
Simon, Am Economic
Review, 1978
What is the ideal decision in
healthcare?
Patient-centered decision
A patient-centered decision is one which
reflects the needs, values and expressed
preferences of a well-informed patient
Sepucha, Health Affairs 2004
Defining decision quality
A high quality patient decision is one in which the
patient has:
• Leveraged a useful level of decision specific
knowledge
• Expressed his values for the outcomes of interest
for the decision at hand
• Achieved congruence between values and
ultimate treatment choice
Sepucha 2004
Achieving the ideal decision:
Shared Decision Making
• Many definitions
• Shared decision making is the
collaboration between patients and
physicians to come to an agreement about
a healthcare decision
• It is especially useful when there is no
clear "best" treatment option
But…..
• This takes a long time
• Not compensated
• Not all patients prefer this mode of
decision making/feel comfortable with
numbers/ science
Potential solution: decision aids
• Many formats
• Can take advantage of IT to personalize
information, use video, interactivity
• Save time, can be used at home, in
waiting rooms, etc
Challenges addressed by shared decision
making tools
Decision Aids
• Increase patient involvement
• Increase patient knowledge
• Clarify values, increase concordance between
values and choices
• Reduce decisional conflict, regret (? Lawsuits
O’Connor Cochrane Collaboration 2006
Next generation approach:
personalized decision analysis
• “rational model”
• Accounts for all possible outcomes
• Accounts for the probabilities of the
outcomes
• ‘Weighs’ the desirability of the outcomes
Decision analysis for prostate
cancer
Erectile dysfunction 50%
Urinary incontinence 5%
Cancer death 15%
radiation
Erectile dysfunction 20%
Urinary incontinence 3%
Cancer death 30%
Erectile dysfunction 10%
Urinary incontinence 1%
Cancer death 35%
Decision analysis for prostate
cancer
radiation
Erectile dysfunction 50%
Value:40
Urinary incontinence 5%
Value:80
Cancer recurrence 15%
Value: 5
Erectile dysfunction 20%
Urinary incontinence 3%
Cancer recurrence 30%
Erectile dysfunction 10%
Urinary incontinence 1%
Cancer death 35%
How can we measure the strength of
your desire to avoid diapers after
surgery?
Patient preference assessment
What is a ‘utility’value?
• Derived from classical economics
• A health ‘utility’ is a number, ranging from 0.0 to
1.0, which corresponds to a person’s desire for a
health state
• Determined under a conditions of uncertainty
• Expected utility theory is a ‘normative’
description
Von Neumann and Morgenstern 1944
Ways in which we can use
patient preferences
1 year in health state with a utility of 0.85
=
0.85 quality adjusted life years
(QALY)
How do you measure utility?
Traditional ways to quantify preferences:
• Standard Gamble
• Time Trade Off
• Rating Scale
Consumer preference
measurement: conjoint analysis
Phone A
Phone B
Touch screen
Keyboard
2 month wait
No wait
4G network
3G network
Conjoint analysis
• Can more easily incorporate non-clinical
treatment attributes of importance to
patients
• More accurate assessments of
preferences may lead to treatment choices
more congruent with patients’ goals
• More intuitive- leverages emotional
intelligence
Developing a conjoint application
• “Voice of the customer” approach
• Relevance for other patient/stakeholder
engagement efforts?
Methods
“Voice of the Patient” Process
60-90 min.
Interviews:
treatments,
Side effects,
outcomes
Listen
Side
Research Researchers Patients Researchers
Team
effects
Team
Narrow
Group Analyze piles Identifies
Outcomes Identifies From 1,000 Similar Using AHC Conjoint
1,000
15
to 70
Quotes for consensus Attributes
quotes
Themes
quotes
into piles groupings From piles
Parse
Themes
Objective
Select
Subjective
Affinity
Analyze
More Subjective
Translate
Methods
Sample narratives from men treated for prostate cancer
Treatment Issues
Side Effects
Cutting: I don't want to be cut; I don't
want to have surgery.
Sex: If you have an understanding
partner, the ED thing can be ok.
Others' Advice: I only follow doctors’
advice up to a point. Not 100%
Urinary: Changing pads frequently…feels
as if you don't have control of your life.
Caution: I could wait for a while if the
numbers stay stable…
Lifespan: It is more important to stay
alive, regardless of the side effects.
Action: I was just thinking "we have
got to do something"
Bowel: The bowel issue is the biggest deal
because it is socially unacceptable.
Listen
Parse
Themes
Select
Affinity
Analyze
Translate
Methods
• Randomized trial of conjoint analysis versus time
trade off and rating scale methods
• “Voice of the customer” adaptation to identify
attributes of importance to patients
• Development of rating scale and time trade off
applications
• Development of novel form of real-time conjoint
analysis:
Adaptive Best-worst Conjoint (ABC)
Methods
(7) Seven Patient-derived attributes:
1.
2.
3.
4.
5.
6.
7.
Sexual function
Urinary function
Bowel function
Survival
“Active/Cautious”
Requirement for incision
Opinion of significant others
Methods
• Recruited men at the VA urology clinic undergoing
prostate needle biopsy for suspicion of prostate
cancer
• Eligible men:
Negative biopsy, able to read English
• Subjects and task order randomized to:
Rating Scale vs. Adaptive Best-worst Conjoint
Time Tradeoff vs. Adaptive Best-worst Conjoint
Results
Characteristic
Age
Race/ethnicity
White (non-Hispanic)
Black/African American
Hispanic/Latino
Other or mixed race/ethnicity
Partnership status
Living with spouse or partner
Signif. relationship, not living together
Not in a significant relationship
Marital status
Currently married
Not currently married
Employment status
Employed
Not employed
Retired
Educational attainment
High school graduate or less
Some college
College graduate
Household income
Less than $10,000
$10,000 to $30,000
More than $30,000
Mean (% of n=31)
64 ± 4, range 55 to 73
10 (32%)
13 (42%)
5 (16%)
3 (10%)
19 (61%)
2 (6%)
10 (32%)
14 (45%)
17 (55%)
10 (32%)
9 (29%)
12 (39%)
4 (13%)
17 (55%)
10 (32%)
5 (17%)
13 (43%)
12 (40%)
Characteristic
Mean (% of n=31)
Current smoker
Yes
5 (16%)
No
26 (84%)
Medical conditions
Diabetes
7 (23%)
Heart attack
6 (19%)
Stroke
0 (0%)
Amputation
1 (3%)
Circulation problems
7 (23%)
Asthma, emphysema, breathing probs.
4 (13%)
Stomach ulcer or irritable bowel
3 (10%)
Kidney disease
1 (3%)
Major depression
4 (13%)
Seizures
0 (0%)
Alcoholism or alcohol problems
5 (16%)
Drug problems
4 (13%)
Control preferences scale
Mostly doctor making decision
3 (10%)
Doctor and self together
15 (48%)
Mostly self
13 (42%)
Problems in last 4 weeks
Urinary function
11 (35%)
Bowel habits
2 (6%)
Sexual function
11 (35%)
Hot flashes
0 (0%)
Breast tenderness/enlargement
0 (0%)
Depressed
0 (0%)
Lack of energy
1 (3%)
Change in body weight
1 (3%)
Functioning problems were dichotomized as no (no problem or very
small problem) or yes (small, moderate or large problem)
Results
Outcome metrics:
-Compared internal validity of methods
-Comparative ability of stated preference data
to predict preferences for health states that
were not explicitly rated by patient
-Compared patient acceptability in men being
evaluated for prostate cancer
Results: Internal validity
(R2 = % of variance in 16 stimuli scores
explained by utility functions)
P>.05
Mean R2
90%
80%
88%
87%
70%
P=.001
60%
55%
50%
Conjoint
Ratings
Time
Tradeoff
P-values are from paired comparisons (t-tests) with conjoint analysis.
Hit Rate: 1 of 4
Results: Predictive validity for 3
methods
(hit rate:1st choice out of 4 options)
65%
68% 68%
55%
P>.05
P>.05 63%
56%
45%
P>.05
P>.05
47% 47%
35%
1st Choice Hit Rate Conjoint Stimuli
1st Choice Hit Rate Holdout Stimuli
25%
Conjoint
Ratings
Time Tradeoff
P-values are from paired comparisons (McNemar tests) with conjoint analysis.
Results: Patient satisfaction and
Ease-of-Use scores
Preference assessment method ease of use and satisfaction (categories collapsed)
Conjoint
analysis
(N = 31)
Ease of use
Very easy/easy/ somewhat easy
Somewhat/very difficult
Satisfaction
Extremely/somewhat
Neutral/not very/not at all
Time tradeoff
Rating scale
(N = 15)
(N = 16)
18 (58%)
13 (42%)
10 (67%)
5 (33%)
14 (88%)
2 (12%)
26 (84%)
9 (60%)
13 (81%)
5 (16%)
6 (40%)
3 (19%)
Conjoint vs. time
tradeoff
(N = 15)
Conjoint vs. rating scale
(N = 16)
P = .99
P = .03
P = .38
P = .99
P-values obtained by comparing responses within same subjects using the exact version of McNemar’s test of paired proportions.
Rating Scale perceived to be easier than Conjoint…
but Conjoint’s satisfaction ratings are just as good
Conclusions
• Conjoint analysis is a feasible method to
collect real-time, individual level
preferences from patients
• Conjoint analysis is viewed by patients as
a satisfactory way to collect preference
data, though challenging
Additive value of conjoint analysisbased preference assessment over
tradictional SDM aid
Methods
• Men randomized to education and
preference assessment receive a report
detailing their preferences
• Counseling physicians briefed on report
interpretation
• Physicians could use the report during the
counseling session.
Methods
Decision quality measures (pre/post):
•
•
•
•
•
Satisfaction with care
Disease specific knowledge
Decisional Conflict Scale
Shared decision making questionnaire
Yes/No has made a treatment choice
Results
Decisional Conflict
Satisfaction with Care
Results: Prostate Cancer
Knowledge
80
78
76
74
72
70
68
66
64
62
60
Intervention
Control
Conclusions
Conjoint analysis is a feasible method to collect
real-time, individual level preferences from
patients in a busy clinic
Pilot data indicate:
-increased patient satisfaction after formal
preference assessment, reduced decisional
conflict
-perception of physician thoroughness
enhanced
Next frontiers
• Deployment of integrated decision
analysis- preference measurement
application at (UCLA)
• Identify barriers to actual shared decision
making behaviors in men who have
viewed a decision aid and express
readiness to engage in shared decision
making (PCORI)