advancing and applying stated- preference methods among ...€¦ · patients and to update the...
TRANSCRIPT
Johns Hopkins Bloomberg School of Public Health
Advancing and applying stated-preference methods among
patients with type 2 diabetes John FP Bridges
Associate Professor Department of Health Policy and Management
1
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Acknowledgements • This work is supported by the Patient-Centered
Outcomes Research Institute (PCORI) Methods Program Award (ME-1303-5946).
• John Bridges is supported by the FDA-Johns Hopkins Center for Excellence in Regulatory Science and Innovation (CERSI)
• Co-investigators: • Albert Wu, Daniel Longo, Lee Bone, Karen
Bandeen-Roche, Jodi Segal, Tanjala Purnell, Karen Edwards, Ellen Janssen, Allison Oakes, Mo Zhou
2
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Overview This presentation will 1. Give a brief overview of stated-preference methods
and instrument development of two choice experiments to measure the preferences of people with type 2 diabetes
2. Discuss results of a national survey comparing the use of discrete-choice experiment and best-worst scaling to measure preferences for diabetes medications
3. Discuss preference heterogeneity for diabetes medications using different analytical techniques
3
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Objectives of the PCORI study 1. Demonstrate and disseminate good practices for
patient and community involvement in patient centered outcomes research projects by applying principles of community-based participatory research
2. Address several key methodological questions pertaining to the use of stated-preference methods
3. Demonstrate and disseminate good practices for the application of stated-preference methods in patient centered outcomes research
4
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Aims of the PCORI study 1. Compare two survey methods for assessing the
priorities of patients with type 2 diabetes (rating/ranking vs. best-worst scaling)
2. Compare two survey methods for measuring the preferences of patients with type 2 diabetes (choice based conjoint/discrete choice experiment vs. best-worst scaling)
3. Compare stratification and segmentation methods for analyzing preference heterogeneity
4. Assess patients’ and stakeholders beliefs about the relevance of our methods and results
5
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Diabetes action board (DAB) The diabetes action board (DAB) is a group of local and national stakeholders that has played and will continue to play a role in:
• Developing this study to measure the preference of patients in type 2 diabetes
• Assisting in the broad dissemination of the research findings and in leverage further applications and action in type 2 diabetes
• Building personal and professional relationships to enrich our work
6
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Study Overview - Progress
PCORI Diabetes Preferences Study Progress
FirstDABmee2ng
Systema2cReview
Whitepaper
Focusgroups(n=25)
SecondDABmee2ng
Report:focusgroups
Pretest(n=25)
Pilottest(n=50)
ThirdDABmee2ng
Na2onal(n=1000)
Report:aggregatefindings
FourthDABmee2ng
Report:heterogeneity
Followupsurvey(n=600)
FiMhDABmee2ng
Report:followupfindings
FinalDABmee2ng
Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov
20162014 2015
7
1. Instrument development
8
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Stated-preference methods • Stated-preference methods are survey techniques
aimed at documenting the priorities and preferences of respondents
• Preferences relate to an a priori assessment of possible alternatives (e.g. health states, treatment options etc). Implicit in the assessment of preferences is a tradeoff between the perceived benefits and harms
9
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Preferences in Decision Making • Regulatory benefit-risk assessments
– Patient Focused Drug Development – Medical Device Innovation Consortium – FDA recently approved a weight loss device based
on patients’ risk tolerance – FDA released Patient Preference Information draft
guidance
10
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Preferences in Diabetes • Chronic condition that requires lifestyle intervention
and has a range of available treatment options • Three systematic reviews explored the literature on
treatment preferences in diabetes (Joy et al. 2013, Purnell et al. 2014, VonArx et al., 2014)
• 67 unique title reviewed • Most studies were industry funded, had small
sample sizes, and were of low to moderate quality
11
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Instrument development Evidence synthesis
Expert consultation
Stakeholder engagement
Pretest interviews
Pilot testing
12
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Evidence synthesis • Objective: To gather existing evidence on preferences
and utilize the existing literature to develop the instrument
• Combined articles from three literature reviews on treatment preferences of adults with diabetes
• Extracted attributes from discrete choice experiments (DCE) and conjoint analyses (CA)
• Within each study, relative attribute importance (RAI) for each attribute was calculated
13
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Preferences in diabetes literature • 10aUributeswithrela2veaUributeimportanceextractedfrom12publishedDCEsondiabetespreferences
14
0 1 2 3 4 5 6 7 8 9 10
CVDrisk(6)
Glucosemonitoring(3)
TreatmentBurden(23)
Nausea(12)
Hypoglycemia(17)
Sideeffects(10)
Weight(14)
Glucosemeasures(19)
QualityofLife(3)
Cost(11)
StandardizedRela-veA0ributeImportance
A0ributes
MinMedianMax
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Expert consultation • Objective: To ensure clinical accuracy and relevance of
attributes and decision framework • Decision Framework:
“Suppose that your doctor says that your current diabetes medicine is not working to keep your blood sugar controlled. Your doctor recommends that you add another diabetes medicine to lower your A1c.”
15
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Pretest interviews (n=25) • Objective: To ensure acceptability of the instrument to
patients and to update the instrument based on participants’ feedback.
• In-person semi-structured cognitive interviews (20-60 min) among patients with type 2 diabetes in Baltimore.
• Participants were presented with a paper version of the instrument and verbalized their thoughts.
16
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Pilot Testing (n=50) • Objective: To quantitatively test the instrument and to
estimate attribute priors for the experimental design of a large-scale national survey.
• Administered through a national online panel (GfK KnowledgeNetwork)
• Collected demographic information • Randomized design was utilized:
– DCE (n=27) v BWS (n=23) to measure preferences
17
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Final survey - Attributes
Attributes Highest benefit/
Lowest risk
Medium benefit and
risk
Lowest benefit/
Highest risk A1c levels go
down 1% 0.5% 0%
Stable blood glucose
6 days per week
4 days per week
2 days per week
Low blood glucose None During the day
only During the day
and/or night
Nausea None 30 minutes per day
90 minutes per day
Additional medicine 1 pill per day 2 pills per day
1 pill and 1 injection per
day Additional out-of-pocket costs $10 per month $30 per month $50 per month
18
2. Comparing two preference-elicitation methods
19
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
DCE • A repeated discrete-
choice response indicating preference between two or more profiles according to objective/subjective criteria
• Strengths: most frequently used and studied stated-preference method
• Limitations: complicated design and analysis
20
Attributes Medicine A
Medicine B
A1c levels go down 1% 0.5%
Stable blood sugar 2 days/wk 4 days/wk
Low blood glucose
During the day None
Nausea None 90 min/day
Additional medicine 2 pills/day 1 pill/day
Additional out-of-pocket costs $50/mo $30/mo
Which medicine would you choose?
Medicine A ☐
Medicine B ☐
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
BWS A repeated discrete-choice response assessing the best/worst aspect of a profile according to objective/subjective criteria Strengths: simple design and analysis Limitations: possible floor and ceiling effects
21
Attributes Medicine A Best Worst
A1c levels go down 1% ☐ ☐
Stable blood sugar 4 days/wk ☐ ☐
Low blood glucose
During the day ☐ ☐
Nausea None ☐ ☐ Additional medicine 2 pills/day ☐ ☐ Additional
out-of-pocket costs
$50/mo ☐ ☐
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Experimental design DCE • 6 attributes at 3 levels each • Bayesian efficient design:
• 48 profile pairs divided into 4 blocks • Added 2 holdout tasks to each block • 18 profile pairs per participant
• Prompt: Consider the following two diabetes medicines. Which medicine would you prefer?
BWS • 6 attributes at 3 levels each • Orthogonal design:
• 18 profiles per participant • Prompt: Which of this medicine’s characteristics is the
best and which is the worst? 22
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
National Survey - Overview • 1103 participants with self-reported type 2 diabetes. • Survey was administered through GfK Knowledge Panel,
a nationally representative online panel. • The survey was in the field for 16 days from October 10
to October 25, 2015 • Collected preference data as well as self-reported
demographic, personality, and clinical information
23
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Research Question • Are treatment preferences estimated using BWS Case 2
the same as treatment preferences estimated using DCE?
• Determine: • Respondent burden of the methods • Correlation between methods • Equivalence of the methods
• Do different subgroups display different preferences? • Stratification • Segmentation
24
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Demographic characteristics BWS 2 DCE P-value Total (N, prop) 551 (0.50) 552 (0.50) Age (mean, range) 63 (25, 89) 61 (24, 91) .082 Gender
Male (N, prop) 274 (0.49) 279 (0.51) .787 Race .985
White (N, prop) 286 (0.51) 289 (0.52) Black (N, prop) 128 (0.23) 126 (0.23) Hispanic (N, prop) 117 (0.21) 119 (0.22) Other (N, prop) 20 (0.04) 18 (0.03)
Education .393 No HS degree (N, prop) 39 (0.07) 43 (0.08) HS degree (N, prop) 174 (0.32) 188 (0.34) Some college (N, prop) 182 (0.33) 156 (0.28) Bachelor’s or higher (N, prop) 156 (0.28) 165 (0.30)
25
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Diabetes Related Characteristics BWS 2 DCE P-value
Years of diagnosis (mean, range) 13.2 (11.9, 14.5)
12.6 (11.4, 13.7) .645
Hypoglycemia At least once in past 6 mo (N, prop) 273 (0.50) 259 (0.47) .820
A1c level .169 ≥ 8.0% (N, prop) 83 (0.15) 80 (0.15) ≥ 7.0%, but < 8.0% (N, prop) 144 (0.27) 153 (0.28) < 7.0% (N, prop) 232 (0.43) 228 (0.41) Don’t know (N, prop) 84 (0.15) 89 (0.16)
Diabetes medicine .049 No medicine (N, prop) 62 (0.11) 37 (0.07) Only pills (N, prop) 321 (0.58) 345 (0.63) Only insulin/injection (N, prop) 48 (0.09) 42 (0.08) Pills and injections (N, prop) 119 (0.22) 127 (0.23)
26
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Time spend per Section (minutes)
Task N Median (minutes)
Min (minutes)
Q1 (minutes)
Q3 (minutes)
Max (minutes)
DCE 552 10.1 0.9 8.8 16.6 191.6 BWS 2 551 12 1.3 7.4 14.8 146.7
27
025
50
75100
125
150
175
200
BWS2DCE
Q1
Min
Median
Max
Q3
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Evaluation of preference tasks
28
0.0000
0.2000
0.4000
0.6000
0.8000
1.0000
1.2000
Ifounditeasytounderstandthe
ques2ons
Ifounditeasytocompletetheques2ons
Iansweredinawayconsistentwithmy
preferences
Standardizedscoreonascalefromstronglydisagree(-2)tostronglyagree(+2)
DCE
BWS2
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
DCE vs. BWS Case 2 (rho = 0.93)
29
Analyzedusingmixedlogitandeffectscoding
-1.6-1.4-1.2-1.0-0.8-0.6-0.4-0.20.00.20.40.60.81.01.21.41.6
1%
0.50%
0%
6days/w
eek
4days/w
eek
2days/w
eek
Non
eDa
yDa
yand/ornight
Non
e30m
inutes
90m
inutes
1pill
2pills
1pilland1injec2on
$10
$30
$50
DCE BWS
A1cdecrease
Stablebloodglucose
Lowbloodglucose
Nausea
Treatmentburden
Out-of-pocketcost
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
DCE vs. BWS Case 2 - scaled
30
Analyzedusingmixedlogitandeffectscoding
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
0.3
0.5
0.71%
0.50%
0%
6days/w
eek
4days/w
eek
2days/w
eek
Non
eDa
yDa
yand/ornight
Non
e30m
inutes
90m
inutes
1pill
2pills
1pilland1injec2on
$10
$30
$50
DCE BWS
A1cdecrease
Stablebloodglucose
Lowbloodglucose
Nausea
Treatmentburden
Out-of-pocketcost
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Standardized attribute importance
31
0
2
4
6
8
10
12
DCE
BWS
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Comparing results with the literature
32
0 1 2 3 4 5 6 7 8 9 10
CVDrisk(6)
Glucosemonitoring(3)
TreatmentBurden(23)
Nausea(12)
Hypoglycemia(17)
Sideeffects(10)
Weight(14)
Glucosemeasures(19)
QualityofLife(3)
Cost(11)
StandardizedRela-veA0ributeImportance
A0ributes
MinMedianMaxDCEBWS
3. Preference heterogeneity
33
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Preference heterogeneity • Different individuals or individuals of different
subgroups might have different preferences. • Preference heterogeneity can be accounted for using:
• Stratification • Mixed logit models • Finite mixture models (segmentation)
34
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Preferences by Gender
Analyzedusingcondi4onallogitandeffectscoding
35
-0.7
-0.5
-0.3
-0.1
0.1
0.3
0.5
0.7 1%
0.50
%
0%
6 da
ys/w
eek
4 da
ys/w
eek
2 da
ys/w
eek
Non
e
Day
Day
and
/or n
ight
Non
e
30 m
inut
es
90 m
inut
es
1 pi
ll
2 pi
lls
1 pi
ll an
d 1
inje
ctio
n
$10
$30
$50
Male Female
A1cdecrease
Stablebloodglucose
Lowbloodglucose
Nausea
Treatmentburden
Out-of-pocketcost
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Preferences by A1c levels
36
Analyzedusingcondi4onallogitandeffectscoding
-1.4
-1.2
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1%
0.50
%
0%
6 da
ys/w
eek
4 da
ys/w
eek
2 da
ys/w
eek
Non
e
Day
Day
and
/or n
ight
Non
e
30 m
inut
es
90 m
inut
es
1 pi
ll
2 pi
lls
1 pi
ll an
d 1
inje
ctio
n
$10
$30
$50
<7% >8%
A1cdecrease
Stablebloodglucose
Lowbloodglucose
Nausea
Treatmentburden
Out-of-pocketcost
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
DCE – Mixed Logit
37
-2.3
-1.9
-1.5
-1.1
-0.7
-0.3
0.1
0.5
0.9
1.3
1.7
2.1 1%
0.50
%
0%
6 da
ys/w
eek
4 da
ys/w
eek
2 da
ys/w
eek
Non
e
Day
Day
and
/or n
ight
Non
e
30 m
inut
es
90 m
inut
es
1 pi
ll
2 pi
lls
1 pi
ll an
d 1
inje
ctio
n
$10
$30
$50
A1c decrease
Stable blood glucose
Low blood glucose
Nausea Treatment burden
Out-of-pocket cost
Mea
n pr
efer
ence
wei
ghts
with
95%
Con
fiden
ce In
terv
al
and
95%
CI o
f pre
fere
nce
dist
ribut
ion
©2015, Johns Hopkins University. All rights reserved. ©2015, Johns Hopkins University. All rights reserved. © 2014, Johns Hopkins University. All rights reserved.
Predicted individual preferences 0
.2.4
.6.8
-2 -1 0 1 2 3
1% A1c decrease0
12
34
-.5 0 .5 1
0.5% A1c decrease
0.5
11.
52
-1 -.5 0 .5 1 1.5
Stable 6 dy/wk
01
23
-1 -.5 0 .5 1
Stable 4 dy/wk
01
23
-.5 0 .5 1
No hypoglycemia
0.5
11.
52
-1 -.5 0 .5 1
Daytime hypoglycemia
0.2
.4.6
-2 0 2 4
No nausea
01
23
4
-.5 0 .5 1
30 minutes Nausea
0.2
.4.6
.81
-2 -1 0 1 2
1 pill
0.5
11.
5
-1 0 1 2
2 pills
Density y
38
0.2
.4.6
.8
-2 -1 0 1 2 3
1% A1c decrease
01
23
4
-.5 0 .5 1
0.5% A1c decrease
0.5
11.
52
-1 -.5 0 .5 1 1.5
Stable 6 dy/wk
01
23
-1 -.5 0 .5 1
Stable 4 dy/wk0
12
3
-.5 0 .5 1
No hypoglycemia
0.5
11.
52
-1 -.5 0 .5 1
Daytime hypoglycemia
0.2
.4.6
-2 0 2 4
No nausea
01
23
4
-.5 0 .5 1
30 minutes Nausea
0.2
.4.6
.81
-2 -1 0 1 2
1 pill
0.5
11.
5
-1 0 1 2
2 pills
Density y
0.2
.4.6
.8
-2 -1 0 1 2 3
1% A1c decrease0
12
34
-.5 0 .5 1
0.5% A1c decrease
0.5
11.
52
-1 -.5 0 .5 1 1.5
Stable 6 dy/wk
01
23
-1 -.5 0 .5 1
Stable 4 dy/wk
01
23
-.5 0 .5 1
No hypoglycemia
0.5
11.
52
-1 -.5 0 .5 1
Daytime hypoglycemia
0.2
.4.6
-2 0 2 4
No nausea
01
23
4
-.5 0 .5 1
30 minutes Nausea
0.2
.4.6
.81
-2 -1 0 1 2
1 pill
0.5
11.
5
-1 0 1 2
2 pills
Density yPredictedindividual
coefficientsEs2mateddistribu2on
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Latent Class Analysis - DCE
-2.3 -2.1 -1.9 -1.7 -1.5 -1.3 -1.1 -0.9 -0.7 -0.5 -0.3 -0.1 0.1 0.3 0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9
1%
0.50
%
0%
6 da
ys/w
eek
4 da
ys/w
eek
2 da
ys/w
eek
Non
e
Day
Day
and
/or
nigh
t
Non
e
30 m
inut
es
90 m
inut
es
1 pi
ll
2 pi
lls
1 pi
ll an
d 1
inje
ctio
n
$10
$30
$50
Class 1 Class 2
A1cdecrease
Stablebloodglucose
Lowbloodglucose
Nausea
Treatmentburden
Out-of-pocketcost
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Latent Class Analysis (cont.)
Control17%
Stable15%
Hypo11%
Nausea10%
Dose21%
Cost26%
Class1(65.8%)
Control12%
Stable10%
Hypo10%
Nausea45%
Dose19%
Cost4%
Class2(34.2%)
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Latent Class Analysis (cont.) OddsRaCo(SE)
Age 1.033**(0.01)
Female 0.979(0.21)
Race/Ethnicity
Black 1.149(0.30)
Hispanic 0.623(0.18)
Other 0.855(0.50)
Yearsofdiabetes 1.031*(0.01)
Self-reportedHealthStatus
Verygood 1.719(0.82)
Good 1.502(0.72)
Fair 1.512(0.79)
Poor 1.655(1.12)
HaveotherchroniccondiCons 1.703*(0.45)
*Othercovariatesincludeeduca4on,income,employmentstatus,andhavingcomplica4ons.
© 2014, Johns Hopkins University. All rights reserved.
© 2014, Johns Hopkins University. All rights reserved.
©2015, Johns Hopkins University. All rights reserved.
Conclusion • Participants did not express a strong preference
towards BWS or DCE. • Preference weights obtained from BWS or DCE had
high correlation, but were on a different scale. • Significant preference heterogeneity was observed in
mixed logit models and latent class models.
42
Protecting Health, Saving Lives—
Millions at a Time
43