Deriving distributional weights for
Quality Adjusted Life Years (QALYs)
through discrete choice experiments
Emily Lancsar
Associate Professor
Centre for Health Economics
Monash University
New Directions in Welfare
7 July 2011
Acknowledgements
• ESRC/MRC/NIHR Fellowship in Economics of
Health
• Joint work with John Wildman, Cam
Donaldson, Mandy Ryan and Rachel Baker
QALYs
• Scarcity – need to prioritize what is funded/covered 
economic evaluation
• Quality adjusted life year (QALY) dominant measure of benefit
assessment in economic evaluation in the health sector
• QALY is LE x QOL; 1 QALY = 1 year in full health
• Wide spread use internationally in HTA – e.g.
– NICE in the UK
– PBAC in Australia
– CADTH in Canada
In England…
• When this study was undertaken….
• NICE appraises new & existing health technologies, makes
recommendations to rest of NHS
• Requires judgments about value of QALY gains
• The mythical £30,000 per QALY!!
• Is it in the right ballpark?
• What determines whether something above the ‘threshold’
might be accepted?
• Even without money values, what weights might be attached
to QALY gains in different situations?
• These questions led to the Social Value of a QALY (SVQ)
project
SVQ
• Funded by Dept of Health’s NCCRM, 2004-2007
• Team: Ian Bateman1, Cam Donaldson2, Michael JonesLee2, Emily Lancsar2, Graham Loomes1, Helen Mason2,
Jose Luis Pinto Prades1 Angela Robinson1, Mandy
Ryan3, Phil Shackley2, Richard Smith1, Kerry Sproston4,
Robert Sugden1, Heather Wardle4, John Wildman2
1.
2.
3.
4.
University of East Anglia
Newcastle University
University of Aberdeen
National Centre for Social Research (NatCen)
SVQ: Key research questions
1. What is the social value of a QALY (to inform
consideration of cost effectiveness thresholds)?
2. Does this value differ depending on the
characteristics of the recipients of such health
gains?
Distributional Weights
• Default normative framework: all QALYs are equal (“a
QALY is a QALY is a QALY”); maximise total
• Alternative: Would members of the public prefer
policy makers to trade off health gains for other
characteristics when allocating scarce health care
resources?
• If so, which characteristics of recipients of QALYs
warrant differential treatment?
• What weight should they receive?
DCEs – what are they?
•
Three components:
1. Surveys used to collect choice data using an
experimental design
2. Discrete choice analysis used to model
preferences from the generated data in a
random utility framework
3. Model of preferences (IUF) used to derive
welfare measures & other policy analysis
Key methodological challenges of this study
1. Identifying attributes over which weights should be
derived
2. Designing & presenting questions so that
respondents can understand & make complex
choices
3. Eliciting quantitative preference data from
members of the general public to allow the
estimation of distributional weights
4. Developing a way to derive weights from the model
of preferences
Qualitative Research
• One year of qualitative work to:
– establish attributes
– develop & pilot the questionnaire
• 17 focus groups
• 1 on 1 cognitive interviews
• Q sort
Attributes, Levels & Experimental Design
Description of attribute
Age at onset (years)
Age at death if untreated (yrs)
Severity: Qol if untreated (%)
Gain in Qol with treatment (%)
Gain in life expectancy (yrs)
Levels
1 10 20 40 60 70
1 10 20 40 60 70 80
0 30 60 90
0 10 20 40 70 100
0 1 5 10 20 40 60 79
• ED – combination of attributes and levels to present as choice
options
• Implausible combinations → constraints → compromised
statistical efficiency
• However, efficiency is relative to an optimal design without
implausible scenarios; optimal design for this study is not known
• Increased realism and “respondent efficiency”
Choice context
• Choice of who to treat out of two types of patients A & B
• For each patient type, described what would happen first
without treatment & then with treatment
• Context: two options cost the same & treat the same number
of patients, but fixed amount of resources making choice
necessary - invoking consideration of a government budget
constraint
• ‘Citizen’ perspective - lives of others
Presenting Choices
• For each option A & B:
• First depict scenario without treatment: full health from birth
until illness at some age depicted as reduction in either QOL
or LE axis or both (light blue)
• Next depict gain with treatment: QOL gain, LE gain or
combination of both (dark blue)
• Two stage presentation – untreated situation useful:
– starting point to demonstrate potential health gain
– to show counterfactual of what happens to group not
chosen
Data collection
• National Centre for Social Research
• Face to face in home using CAPI
• Questionnaire: background, DCE, PTO, attitudinal
questions & socio-demographics
• 587 members of adult population in England
– 243 (41%) were male (compared to 49% in the general
population)
– mean age: 52 years (47 in the general adult population)
Discrete choice model
• V=f(AO, AD, QL, QALY)
• Create a QALY variable: gain LE x QOL
• Cannot be additive (to avoid positive utility even when
QALYs=0)
• Assumed multiplicative underlying model – log-linear model &
accounted for higher order non linearities
 log( V )  1 ( log( AO))   2 ( log( AO)) 2   3 ( log( AO)) 3
  4 ( log( AD))   5 ( log( AD)) 2   6 ( log( AD)) 3
  7 ( log( QL ))   8 ( log( QL )) 2   9 ( log( QL )) 3
 10 ( log( QALY ))  11 ( log( QALY ) 2  12 ( log( QALY )) 3
Choice Model Results
• AO not significant
• AD significant, increasing at decreasing rate
• Severity significant only at 10% in the linear term,
preference for treating individuals in less severe
states
• Increasing QALYs always preferred
Main result: QALYs dominate the model in terms of
size of coefficient & statistical significance
Two types of distributional weights
1. Weights for individual characteristics (AO, AD,
severity)
– Calculated by changing one characteristic at a time,
holding all else constant
2. Weights for whole scenarios, or combinations of
characteristics, describing beneficiaries of QALY
gains
– Derived for 16 types of beneficiaries each receiving 1, 2, 4
& 10 QALYs with treatment (64 cases)
Deriving Distributional Weights
• Hicksian compensating variation from a discrete
choice random utility model:
J
1  J V j0
V 1j 
CV  ln  e  ln  e 
  j 1
j 1

• Traditionally used to calculate monetary value
• Instead, value the change of interest using marginal
utility of a QALY (for a 1 QALY gain) rather than MUy
Deriving distributional weights
• CV values used to calculate weights:
CV
Weight  1 
QALYbase
• CV: number of QALYs required to equalise expected utility in the reference
& alternative case
• QALYbase: number of QALYs gained in the reference case
• Weight greater (less) than one indicates the alternative case is valued
more (less) highly than the reference
Weights for individual characteristics
• AO: No weight is given based on age of onset of illness
• AD: Preference for giving more weight (although weight very
small) to those who die at 10, 70 or 80
– may reflect caring about the very young & old in our society
• Severity: less weight given to the most severe groups relative
to less severe
– In line with Dolan and Tsuchiya (2005)
– May be giving 4 QALYs to those already experiencing very poor QOL
does not seem as useful as giving QALYs to individuals who are more
likely to be returned closer to full health (capacity to benefit?)
Weights for beneficiary types (constant QALYs)
• Very little preference for weighting QALYs
• Where weights significant, small (range 0.89 to 1.14)
• Significant weight associated with giving QALYs to 1
year olds who without treatment die at age 10 &
either face a QOL loss of 0.1 or 0.7
• Preference for 10 year olds over infants – possibly
related to argument that the latter have not really
yet engaged with life whereas the former have?
“Golden age”?
Weights for beneficiary types (varying QALYs)
• Trade off number of QALYs against other
characteristics
• In aggregate, all weights are relatively small &
number of QALYs gained is driving the results
• Exceptions involve 10 year olds
• Suggests a desire to maximise health & a reluctance
to trade off health gain for other characteristics as
the health gain increased
End of Life Weights
• Policy issue: should more weight be given to treating
those people who are at the end of their life at any
age & in particular when this might be considered
premature?
• Our results suggest individuals facing instant &
premature death should not be given a higher
weighting
• In fact, given a lower weighting relative to the
reference case
Conclusions & Contributions
• Generally both sets of weights suggest little evidence on
which to weight QALYs based on the characteristics of the
recipients; where significant, much smaller than PTO lit
• New analytical approach: DCE & novel application of Hicksian
compensating variation
• Methodological challenges
• Derivation of weights for QALYs, not just for life or life years
saved
• Investigation of the impact of the size of the health gain by
allowing the gain to be traded against other characteristics
• Additional, albeit imperfect, evidence to inform the debate on
the need or otherwise to weight QALYs
Thank you & Questions
• Contact:
Emily.Lancsar@monash.edu
• Lancsar et al (2011) ‘Deriving Distributional Weights
for QALYs through Discrete Choice Experiments’,
Journal of Health Economics; 30: 466-478.
• Questions?