Introduction to Prompt Engineering (Focusing on ChatGPT)
A New Approach to Presenting Health States in Stated Preference Valuation Studies
1. A New Approach to Presenting Health States
in Stated Preference Valuation Studies
Cole, A.1, Shah, K.1, Mulhern, B.2, Feng, Y.1 & Devlin, N.1
In health state valuation studies, respondents are asked to
consider a ‘health state’ and express preferences about it,
e.g.
• How much time in healthy life would you be willing to trade to avoid it?
• Which do you prefer between this health state and this alternative health state?
Health states are generally presented as a list of statements,
each describing an aspect of health and its severity level.
However, presented on their own, respondents may have
trouble understanding how to interpret them.
Acknowledgements
This study was funded by the EuroQol Research Foundation. However,
the views expressed do not necessarily reflect the views of the EuroQol
Research Foundation.
3. METHODS
• Each respondent completed 16 DCE tasks and feedback
questions. Differences across arms were assessed using
regression analyses.
4.RESULTS
• Our sample was representative of the UK general population,
and respondent characteristics were similar across arms
• Respondents assigned to the context arm completed the survey
slightly quicker (mean: 13.18 minutes vs. 12.34 minutes for
context arm respondents) (p<0.001)
• Through regression analyses, coefficients were ordered as
expected for both arms (apart from the ordering of MO2 and
MO3 for the control arm, but this was not statistically
significant)
5. DISCUSSION
• Displaying all levels of the various health dimensions helps
respondents understand the ordering of labels, particularly
of ‘severe’ and ‘extreme’
• Web-based DCE valuation studies can produce well-ordered,
statistically significant coefficients and are efficient and
relatively cheap compared with interviewer-assisted tasks
• The ‘context’ display does not work well on smaller devices
• Our results could have implications for other valuation tasks
such as time trade-off (TTO), and for the valuation of other
patient-reported outcomes (PRO) instruments.
6. CONCLUSION
• Presenting health states ‘in context’ of the whole descriptive
system is feasible, and means that valuation exercises are
matched more closely with how patients self-report their
own health
• Displaying health states in context has an impact on
valuation data, and reduces logical inconsistencies.
Contact: acole@ohe.org 1Office of Health Economics, UK 2University of Technology Sydney, Australia
Which is worse,
‘severe’ or
‘extreme’?
• severe problems in walking about
• no problems washing or dressing myself
• slight problems doing my usual activities
• moderate pain or discomfort
• extremely anxious or depressed
HEALTH STATE DESCRIPTION (EQ-5D)
By contrast, when patients self-complete health
questionnaires, they typically see how the health state fits
within the entire descriptive system and select that which
best describes how they feel.
What does ‘slight’ mean
– how does this
compare with other
types of problems?
1. BACKGROUND
• A two-arm discrete choice experiment
(DCE) conducted online in the UK (n=993)
2. AIM
To test the feasibility of presenting EQ-5D-5L health
states in the context of the entire descriptive
system, and explore how this affects valuation data.
Control
arm
(standard
presentation)
‘Context’
arm
MO2
MO3
MO4
MO5
SC2
SC3
SC4
SC5
UA2
UA3
UA4
UA5
PD2
PD3
PD4
PD5
AD2
AD3
AD4
AD5
-1.4
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
-1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0
ContextArm
Control Arm
• The sizes of decrements between levels were larger and
more significant in the context arm, particularly between
levels 4 and 5 across dimensions
This is especially the case for anxiety/depression
(difference between levels 4 and 5: 0.132 [control]
versus 0.429 [context])
• Preferences differ significantly between arms (likelihood ratio
statistic of restricted pooled model: 42.00)
• Respondents allocated to the context arm made fewer logically
inconsistent responses, tested through the inclusion of two
fixed tasks (testing preference for ‘extreme’ over ‘severe’ (33344
vs 33355) and ‘moderate’ over ‘slight’ (22211 vs 33311)
Control arm: 22.1% of selected options were
‘dominated’
Context arm: 16% of selected options were ‘dominated’
(χ²=24.13, p<0.001)
Most of this difference was accounted for by improvements in
the distinction between ‘extreme’ and ‘severe’ (χ²=46.02, p<0.001)
• Respondents in the control arm were more likely to agree that
they found it “difficult to tell the difference between the
descriptions” (23% versus 14%, (χ²=12.13, p<0.001))MO: mobility
SC: self-care
UA: usual activities
PD: pain/discomfort
AD: anxiety/depression
Dotted line: control=context
Fig. 1 Graphic representation of coefficients for Control and Context arms