preferences for seasonal influenza vaccine for working adultsin ho chi minhcity a discrete choice experiment

MINISTER OF EDUCATION
UNIVERSITY OF ECONOMICS HO CHI MINH CITY

MINH HAI, PHAN

PREFERENCES FOR SEASONAL INFLUENZA VACCINE FOR
WORKING ADULTS IN HO CHI MINH CITY: A DISCRETE CHOICE
EXPERIMENT

THESIS FOR MASTER OF HEALTH ECONOMIC MANAGEMENT

Ho Chi Minh, City – 2015


MINISTER OF EDUCATION
UNIVERSITY OF ECONOMICS HO CHI MINH CITY

MINH HAI, PHAN

PREFERENCES FOR SEASONAL INFLUENZA VACCINE FOR
WORKING ADULT IN HO CHI MINH CITY: A DISCRETE CHOICE
EXPERIMENT

Faculty: Development Economic
Code: 60310105

THESIS FOR MASTER OF HEALTH ECONOMIC MANAGEMENT

INSTRUCTOR: DR. DANG THUY, TRUONG

Ho Chi Minh, City – 2015

ABSTRACT
Introduction: Seasonal influenza is viral infection which affects 5%–10% in adults and
20%–30% in children, and the influenza vaccine, also known as flu shot, is an
annual vaccination using a vaccine that is specific for a given year. There is increasing
interest in benefit of vaccination coverage in health working adults which reduced work
loss. From 2005, the vaccine manufacturing capacity are increasing globally and especially
in Viet Nam, that would support in price and number of doses in broadening the
vaccination. We conduct the study to assess the influencing factors for vaccination in
working adults and how to increase the vaccination in population via the willingness to pay
model.
Method: We conducted a discrete choice experiment, a quantitative approach which often
used in health economic studies. In which, the questionnaire with socio-economic and
discrete choice experiment questions was collected for 172 respondents. The conditional
logit model was used to estimate the relative importance of influenza vaccine attributes:
effectiveness (50%, 70%, 90%); adverse event (0%, 5%,10%), time of vaccination process
(45 minutes, 60 minutes, 90 minutes); place of vaccination (working place, healthcare
center); cost (150,000 VND 200,000 VND 250,000 VND and 300,000 VND). Based on
the utility functions, the willingness to pay and potential vaccination coverage was
estimated for different vaccine scenarios and price support strategy.
Result: The results indicated the effectiveness and adverse event are two valuable
influence (p value<0.1), and the cost dose not impact significantly to the decision of
vaccination. Meanwhile, the coefficient of place of vaccination is too high which is
suspected as implausible value.
Conclusion: Respondent would choose the high effectiveness and low adverse event
for vaccination, by contrast, it will reduce the preference if high cost, long time for
vaccination, and working place. The availability of more than 1 vaccine can increase the
number of vaccination cases, then the well-fare were also increased. It could be explained

that people will prefer the vaccination if there more options for consideration. Although,
the statistical number could help us to predict the trend of vaccination, but the education to
community is very important.
Keywords: Working adults, influenza vaccination, and discrete choice experiment, utility,
conditional logit model.


ACKNOWLEDGEMENT
I would like to deeply express my gratitude and appreciation to Dr. Truong Dang
Thuy, who provided the intensive, supportive, valuable and vital supervision for thesis
completion. I had a great opportunity to access the econometric method and how to apply
in Viet Nam which has very a few application in health sectors. I was inspired by his
knowledge, enthusiasm and professionalism.
Throughout my thesis process, he was always patient, supportive in the right
directions, encouraged me to pursue my own research ideas and provided a great amount
of valuable suggestions and guidance. I appreciate all his contributions of time, advice to
make me successfully complete the thesis.
I would like to express my appreciation and respect for teachers of master of health
economic management program, who have delivered the useful and practical knowledge
in health economic, which I had perceived at the first time.
I would also like to thank all the faculty, staff and students for their friendships and
collaboration during last 2 years.
I would like to be grateful to my colleges, friend and especial my bother in
supporting me for business, life and activities of data collection.
Thanks for my family always as my support overcomes all the difficulties and
pursue the real goal in my life.
With all appreciation, HCM, Aug 2015

TABLE OF CONTENT

ABSTRACT ........................................................................................................................ 2
ACKNOWLEDGEMENT ................................................................................................... 4
TABLE OF CONTENT....................................................................................................... 5
ABBREVIATION ............................................................................................................... 9
CHAPTER 1: INTRODUCTION AND RESEARCH ISSUE ............................................ 1
1

2

Introduction ............................................................................................................... 1
1.1

Introduction of Influenza .................................................................................... 2

1.2

Burden of influenza ............................................................................................ 3

1.3

Influenza vaccination .......................................................................................... 4

Research objectives ................................................................................................... 4
2.1

Problem issue ...................................................................................................... 4

2.2

Research objective .............................................................................................. 5

2.3

Research methodology and organization of thesis ............................................. 6

CHAPTER 2: LITTERATURE REVIEW .......................................................................... 7
1

Random utility Theory (RUT) ................................................................................... 7

2

Lancaster's New Approach to Consumer .................................................................. 8

3

Data collection method .............................................................................................. 9

4

3.1

Contingent valuation method.............................................................................. 9

3.2

Discrete choice experiment .............................................................................. 10

Applications of discrete choice experiments in health care .................................... 11
Relevant attributes ...................................................................................................... 12
Levels of attribute ....................................................................................................... 13
Discrete choice experiments design............................................................................ 13
Model estimation......................................................................................................... 15
Validity issues ............................................................................................................. 16

5

Discrete choice experiment application in healthcare, Viet Nam ........................... 17

6

Discrete choice experiment application in seasonal influenza vaccine .................. 18

CHAPTER 3: RESEARCH METHODOLOGY ............................................................... 21


1

Discrete choice experiment methodology ............................................................... 21

2

Discrete choice experiment construction ................................................................ 22

3

2.1

Identification of relevant attributes .................................................................. 22

2.2

Selection of attributes ....................................................................................... 26

2.3

Definition of levels ........................................................................................... 28

Discrete choice experiment design .......................................................................... 33
3.1

Discrete choice experiment design ................................................................... 33

3.2

Design evaluation and modification ................................................................. 34

3.3

Questionnaire design ........................................................................................ 36

3.4

Data collection .................................................................................................. 36

CHAPTER 4: RESULT ..................................................................................................... 38
1

2

Demographic and history ........................................................................................ 38
1.1

Socio-economic ................................................................................................ 38

1.2

Influenza and vaccination ................................................................................. 40

Model estimation ..................................................................................................... 40
2.1

Conditional logit model result .......................................................................... 40

2.2

Model interpretation: ........................................................................................ 43

2.3

Willingness to pay for vaccine ......................................................................... 44

2.4

Predicting influenza vaccination uptake rate for existing vaccine ................... 45

2.5

Assess the feasibility of new vaccine ............................................................... 47

2.6

Compensating Variation (CV) .......................................................................... 48

2.7

Vaccination up-take .......................................................................................... 50

2.8

Interaction between “non-vaccination” and socio-economic ........................... 51

CHAPTER 5: DISCUSSION AND CONCLUSION ........................................................ 53

Reference
APPENDIX 1 Short survey for importance of attributes relating to influenza vaccination
APPENDIX 2 (Questionnarie#1: Survey for the factors of influenza vaccination)


LIST OF TABLE
Table 1: Example of choice set of discrete choice experiment design .............................. 11
Table 2: Attributes and levels of influenza vaccine in Japan ............................................ 18
Table 3: Willingness to pay for influenza vaccine in Japan .............................................. 20
Table 4: Primary list of characteristics and attributes influencing choice of influenza
vaccination ......................................................................................................................... 23
Table 5: Secondary list of attributes influencing influenza vaccination up-take .............. 25
Table 6: Result of survey for influencing attributes .......................................................... 27
Table 7: Level of vaccinating time .................................................................................... 32

Table 8: Levels of unacceptable cost ................................................................................. 32
Table 9: T-test of level of unacceptable cost ..................................................................... 33
Table 10: Attributes and levels .......................................................................................... 33
Table 11: Orthogonality design ......................................................................................... 34
Table 12: 16 choice sets of discrete choice experiment design ......................................... 37
Table 13: Socio-economic summary ................................................................................. 39
Table 14: Influenza infection and vaccination history ...................................................... 40
Table 15: Marginal willingness to pay .............................................................................. 44
Table 16: Willingness to pay for vaccine .......................................................................... 45
Table 17: An example of predicting influenza vaccination uptake for existing vaccine .. 46
Table 18: New vaccine versus existing vaccine and non-vaccination .............................. 47
Table 19: CV for price support program versus new vaccine introduction ...................... 49
Table 20: The probability of vaccination for the 16 hypothetical vaccines: ..................... 51
Table 21: interaction with “non-vaccination” ................................................................... 52


LIST OF FIGURE
Figure 1: The percentage of specimens (ILI or SARI) positive for influenza (average:
20.37% for Viet Nam (from week 1-2014 to week 18 – 2015)) ......................................... 3
Figure 2: Average of health-related DCE studies/year ..................................................... 12
Figure 4: Number of attributes .......................................................................................... 26
Figure 5: Example of effectiveness explanation ............................................................... 29


ABBREVIATION
AE: Adverse Event
CV: Compensating Variation
CVM: Contingent Valuation Method
DCE: Discrete Choice Experiments
LIV: Live attenuated influenza vaccine

MNL: The MultiNomial Logit
MWTP: Marginal willingness to pay
RP: Revealed Preference
SP: Stated Preference
TIV: Trivalent inactivated influenza vaccine
WTP: Willingness To Pay


1

CHAPTER 1: INTRODUCTION AND RESEARCH ISSUE
1 Introduction
The finite health care resources (including capital, medical treatment capacity) and
rising costs requests the government to determine which health service/treatment to be
provided and the appropriate level at which to provide them. By providing information of
benefit valuation to assess alternative health care solution to matters, the health economics
has been developing and playing the key role in health policy. The application of health
economics is to obtain the maximum value for money with evaluation techniques including
cost minimization analysis, cost effectiveness analysis, cost utility analysis and cost benefit
analysis. By starting to identify the important health issue, the health economists will
explore the appropriate method to assess the alternatives.
Seasonal influenza is the virus infection, which caused the respiratory disease, with
most mild cases but widely impacted. It is the major economic burden and potential
pandemic. The World Health Organization cites studies from developed countries that
suggest the total annual cost of influenza is between U$1 million to U$6 million per
100,000 population.
Influenza infection can result in increased healthcare costs (direct cost) and
workplace absences and reduced productivity (indirect cost). Studies shows the indirect
cost occupies over 50% of total cost of influenza cases. With 48.4% population in age
group of 25-54 years, the indirect cost of influenza could contribute the high proportion of

health cost in Viet Nam.
Vaccination is the most effective measure at preventing influenza and its severe
outcomes. But the vaccination rate is low for Southeast Asia including Viet Nam, with less
than 1% vaccination in total population. The vaccine manufacturing program is expected
to supply 500 millions of influenza vaccine in 2016, which is enough to cover the
vaccination for all Southeast Asia countries.


2

This thesis is to answer how to increase the vaccination rate in working adults to
provide the policy-makers information to assess the alternatives.
1.1

Introduction of Influenza
Seasonal influenza infection

Influenza is a viral infection (including influenza A and influenza B strains) that
affects mainly the nose, throat, bronchi and, occasionally, lungs. Infection usually lasts for
about a week, and is characterized by sudden onset of high fever, aching muscles, headache
and severe malaise, non-productive cough, sore throat and rhinitis.
The virus is transmitted easily from person to person via droplets and small particles
produced when infected people cough or sneeze. Influenza tends to spread rapidly in
seasonal epidemics.
Most infected people recover within one to two weeks without requiring medical
treatment. But, the high-risk persons (age below 2 or above 65 years old, pregnant women
and people of any age with certain medical conditions) are vulnerable to the complications
of influenza that cause the hospitalizations, deaths and high healthcare costs. Meanwhile,
in health working adult group, the major impact of influenza is related to work absenteeism,
impaired working performance and daily activities.

Influenza prevalence (WHO, 2013)
The Global Influenza Surveillance and Response System (GISRS) consists of over
140 National Influenza Centers around the world that collect and test clinical specimens,
submitting a sample of these to WHO Collaborating Centers and Essential Regulatory
Laboratories. The Influenza-Like Illness or Severe Acute Respiratory Infection cases could
be caused by a variety of microbial agents other than influenza viruses.
ILI case definition: an acute respiratory infection with measured fever of >=38oC,
and cough, with onset within the last 10 days.


3

SARI case definition: an acute respiratory infection with history of fever >= 38oC,
and cough, with onset within the last 10 days, and requires hospitalization.
As the primary goal of influenza surveillance is to recognize trends, describe
patterns of risk, and estimate impact, it is not necessary to identify every cases. So the case
definition of ILI and SARI was used for doing the influenza surveillance and integrating
with virological laboratory test (the most common method is: Real-time reverse
transcription polymerase chain reaction) for data.
The patients who attended outpatient department or admitted a sentinel hospital,
will be collected for specimen if meeting the clinical case definition for SAR or ILI, and
the onset of symptoms falls within 10 days of sample collection
The data of Viet Nam from 1st week of 2014 to 18th week of 2015 was extracted
from GISRS to show the average of 20.37% positive influenza among ILI/ SARI.

% POSITIVE INFLU
60.00%
50.00%
40.00%
30.00%

20.00%
10.00%

2015-17

2015-14

2015-11

2015-08

2015-05

2015-02

2014-52

2014-49

2014-46

2014-43

2014-40

2014-37

2014-34

2014-31

2014-28

2014-25

2014-22

2014-19

2014-16

2014-13

2014-10

2014-07

2014-04

2014-01

0.00%

Figure 1: The percentage of specimens (ILI or SARI) positive for influenza (average:
20.37% for Viet Nam (from week 1-2014 to week 18 – 2015))
1.2

Burden of influenza

According to WHO, influenza occurs globally with an annual attack rate estimated

at 5%–10% in adults and 20%–30% in children result, with about 3 to 5 million cases of
severe illness, and about 250 000 to 500 000 deaths. In tropic coutries, influenza-related


4

deaths are estimated to range from 4 to 20 deaths per 100,000 persons (Sophia Ng and
Aubree Gordon, 2015 Apr). A national surveillance from 2006-2010 in 15 hospital
sentinels of Viet Nam, the 19% of hospitalization with ILI is positive with influenza
viruses. In another literature review for South-East Aisa from 1980 to 2006, 11-26 % of
outpatient febrile illness and 6-14% of hospitalized pneumonia cases had laboratory
confirmed influenza infection (James M. Simmerman and Timothy M. Uyeki, 2008).
A systemtic review study (includes 140 articles, before 2010) estimated for the per
capita cost (indirect and direct cost) of case of influenza illness ranged from 27-52 USD
(or 0.04-0.13%GDP) in European countries, 45-63 USD (or 0.14%GDP) in US, 3 USD (or
0.01%GDP) in Hong Kong, and 01 USD (or 0.02%GDP) in Thailand. The percentage of
total cost for productivity losses (indirect cost) was more than 80% in European countries
(except Spain), 38% in US, 12.5% in Hong Kong and 55.3% in Thailand.
1.3

Influenza vaccination

The influenza vaccine, also known as flu shot, is an annual vaccination using a
vaccine that is specific for a given year to protect against the highly variable influenza virus
(Couch, 2008).
Generally, the influenza vaccines included the common trivalent strains (mixture of
Influenza A (H1N1; H3N2) and influenza B strains) were produced into two kinds of form:
the trivalent inactivated influenza vaccine (TIV) and the live attenuated influenza vaccine
(LAIV). In Viet Nam, the TIV is the most popular and being circulated in market, so the
data of TIV was used for model.

The antigenic properties of influenza viruses are highly variable, so the annual
vaccination is required to have the effective immunogenicity.
2 Research objectives
2.1

Problem issue

The cost-effectiveness of influenza vaccination in risk group was widely accepted
and high recommended by WHO. However, there is increasing interest in benefit of


5

vaccination coverage in health working adults which reduced work loss. In cost-saving
annalysis, when the indirect costs included, the vaccination in adults under age 65 is still
highly cost effective. With over fifty millions of employed persons, the benefit of
vaccination in working adults should be considered for Viet Nam.
Moreover, the anaylysis of genome sequences by Nelson et al. found evidence
compatible with either northern-to-southern hemisphere migration or migration from
tropical regions, including Southeast Asia (Russell CA et al., 2008). For this point of view,
a relatively high vaccination coverage which creates the strong community immunity to be
achieved, that would prevent the transmission and mutation (Anon., 2015). The study in
USA, France and Australia suggests that interrupting transmission of seasonal influenza
would require a relatively high vaccination coverage (>60%) in healthy individuals who
respond well to vaccine, in addition to periodic re-vaccination due to evolving viral
antigens and warning population immunity. (G. CHOWELL1, 2008)
The vaccine sale during 2010-1011 is less than 1000 per 100,000 population in
South-East countries, it implies that vaccination is very low (Gupta V et al., 2012). It could
be caused by vaccine capacity or the policy of influenza vaccine programme in that
countries.

From 2005, the vaccine manufacturing capacity are increasing globally and
especially in Viet Nam, that would support in price and number of doses in broadening the
vaccination (Anon., 2013). Once the vaccine capacity addressed, the question on how to
increase the vaccination in population will be an issue to be addressed, especially for
majority of working adults in Viet Nam.
2.2

Research objective

Our primary objective is to assess the influencing factors for vaccination in working
adults.
Secondary objectives:
-

The willingness to pay for vaccination


6

-

The vaccination strategy for achieving the community immunity.
2.3

Research methodology and organization of thesis

This thesis is based on the random utility theory, and the stated preference choice
experiment is used to collect the choice from respondents for influenza vaccination.
Discrete choice experiment was chosen as the approach of stated preference which is
advantage in evaluating the different features of vaccination.

Chapter 2 provide the overview of customer theory, random utility maximization
(RUM) and the literature review.
Chapter 3 describes choice discrete choice experiment and the design of survey.
Chapter 4 presents the empirical estimation results of vaccination choice.
Chapter 5 provides conclusions and discussions on the limitation of study as well as the
suggestions for future research.


7

CHAPTER 2: LITTERATURE REVIEW
1 Random utility Theory (RUT)
Random utility models, a subset of the class of probabilistic choice models, were
firstly developed by psychologists in the attempt to characterize observed inconsistencies
in pattern of individual behavior. In 1927, Thurstone identified the comparison process
(known as Thursone’s Law) with below assumptions:
 Assumption one is that choice is a discrete event. Customer cannot leave market with
0.3432 cans of Coke and 0.6568 cans of Pepsi, but need to leave with 1 cans of their
chosen brand, and 0 cans of their not chosen brand.
 Assumption two is that the attraction or utility towards a brand varies across individuals
as a random variable.
 The last assumption is that the consumer chooses the brand with the highest utility. This
makes our consumer an economically rational being.
The utility of an alternative i in a choice set Cn (as perceived by individual n) is
considered to be decomposable into two additively separable parts: (1) a systematic
(explainable) component specified as a function of (i.e. caused by) the attributes of the
alternatives

; and (2) a random (unexplainable) component

representing

unmeasured variation in preferences. This random variation may be due to unobserved
attributes affecting choice, inter-individual differences in utilities depending upon the
heterogeneity in tastes, measurement errors and/or functional specification.
(Eq. 1)
Alternative is chosen if and only if:
(Eq. 2.1)
(Eq. 2.1)

Or:
In DCE, the utility function

was assumed as:


8

With

is an alternative-specific constant captures the mean effect of the

unobserved factors in the error term for each of the alternative.
The individual choice probability is given by: (Ben-Akiva, M. and Lerman, S.,
1985)
(Eq. 3)
Most commonly, welfare analysis refers to the estimation of WTP for policy
changes. And if there is no income effect (i.e. no change in purchasing power), the
compensated demand curve can be considered to be equivalent to customer surplus.
Customer surplus can be defined as the maximum utility, in monetary terms, an individual

receives by choosing the alternative in a choice situation. A general formula to estimate
mean aggregate Willingness To Pay (WTP) (compensating variation or CV) for a
determined change is (Small, K.A. and Rosen, H.S., 1981) (Williams, H.W.C.L, 1977):
(Eq. 4)
2 Lancaster's New Approach to Consumer
The traditional approach sates that goods are the direct objects of utility and goods
are consumed not because they have intrinsic value but because they are goods. The new
approach supposes that it is the properties or characteristics of the goods from which
utility is derived.
It assumes that consumption, singly or in combination, are inputs and the output
that we get is a combination of characteristic. A product does not have only one
characteristic but may have numerous characteristics which may other products may
share. A product does not have to be a close substitute to share the same properties but
may vary to the extent of a diamond to bread.
The essence of the new approach can be summarized in the following three points:
1. The goods, per se, do not give utility to the consumer. It possesses
characteristics which gives rise to utility.


9

2. In general, a good may possess more than one characteristic, and many
characteristics will be shared my more than one good.
3. Goods in combination may possess characteristics different from those
pertaining to the goods separately.
3 Data collection method
There are two fundamental pathways in data collection for monetary benefit
valuation including “stated preference” (SP), and “revealed preference” (RP). SP relies on
what customers say they will do, and RP relies on what they actually do. RP is market
data, used for product/service which had been launched.

The monetary terms in health care is a remarkable issue to policy decision-maker
including government and pharmaceutical companies, and the health services/products are
usually not traded or characters can be changed (such as efficacy, safety, consequence, or
administration). There are a number of compelling reasons why health economists should
be interested in SP data. The most importance in the health sector is that it may not be
possible to infer consumer preferences or values from RP. The two best-known SP
approaches for providing estimates of monetary valuation are the contingent valuation
method (CVM) and discrete choice experiments (DCE).
The choice of SP method depends, in part, on how much detail is required on the
characteristics of the health care intervention being valued. A CVM is appropriate for
answering questions only about the good or service as a whole (e.g. what is the monetary
value placed on a screening test). In other contexts, what matters is the importance of
different characteristics of the programme being valued. In these cases, DCEs are more
useful. To the extent that DCEs also allow estimating total values, they provide more
information than a single (CVM) experiment (Ryan. M et al., 2008)
3.1

Contingent valuation method

Contingent valuation method is a choice-based approach to ask the individual how
much they are willing to pay for specific good with certain set of attributes and levels. In


10

some cases, they are also asked the amount they will accept in compensation to give up a
specific one. It is called “Contingent” valuation, because people are asked to state their
willingness to pay, contingent on a particular hypothetical scenario and description of the
commodity being valued. The CVM approach estimates the good for a whole.
The survey was created with the hypothetical good constructed by the relevant

attributes and levels, and this was presented with different prices. The range of price was
chosen with assumptions that most people agree to pay at the lowest price and that most
everyone would reject the highest price.
The survey will be started with average price to be offered to respondent, if it is was
rejected, then the lower price will be offered until respondent agree with that price or until
the lowest price was rejected. Otherwise, the higher price will be offered until the highest
price or respondents reject the offer.
The CVM has been applied with varying degrees of success in health care both for
benefit valuation and for elicitation of public view. As monetary benefit valuation is
increasingly advocated in health care and many methodological issues become better
understood, the use of CVM for valuing the multiple-dimension of health care benefits can
be expected to grow. (Ryan. M et al., 2008)
3.2

Discrete choice experiment

The DCE is an attribute-based survey method for measuring benefits (utility) which
respondents are presented with the samples of hypothetical choice sets. The choice sets
comprise two or more alternatives which contain the combination of attributes of
goods/service with various levels. Respondents will make the trade-offs between attributes
levels by choosing the alternative with high utility in the choice sets of the survey.
The DCE assumes that individuals derive utility from the underlying attributes of
the commodity under valuation (rather than the commodity per se) and that individuals’
preferences (as summarized by their utility function) are revealed through their choices.


11

The results from the experiment are used to model preferences within a random utility
maximization (RUM) framework (McFadden, 1974).

As with consumer theory, the respondents in DCE are assumed that they are rational
decision makers and they seek to maximize innate, stable preferences. However, in DCE
there are there important extensions:
 Attributes of goods/services determines the utility
 Participants deal with a choice among a set of finite and mutually exclusive
alternatives (choose one and only one alternative from this choice set).
 Individual choice behavior is intrinsically probabilistic, hence random
Table 1: Example of choice set of discrete choice experiment design
Attributes

Option 1: No

Option 2: Influenza

vaccination

vaccination

0

90%

Adverse event

0%

5%

Time of vaccination

0 minute

60 minutes

Effectiveness of influenza
vaccination
Influenza-infected
Non influenza-infected

Cost of vaccination (by
pocket)

0VND

300,000 VND

Which option you would
choose?

4 Applications of discrete choice experiments in health care
Discrete choice experiments (DCEs) are increasingly used in health economics to
address a wide range of health policy-related concerns in figure 2 (Michael D., 2014).


12

Within 3 years from 2009 – 2012, there are 179 studies, nearly five times more than
duration of 10 years from 1990-2000.

Figure 2: Average of health-related DCE studies/year

DCE in healthcare mostly conducted in UK at the beginning, then it was extended
to USA, AUS, and other countries.
Relevant attributes
It is acceptable reality that researchers cannot observer all the factors affecting
individual preference, and the more factored included the more complexity required. The
correct specification of relevant attributes along with their levels are vital role for
successful elicitation.
Attributes can be quantitative (e.g. cost, number of injection) or qualitative (e.g.
type of healthcare, healthcare service grade), and be generally derived from published
literature, textbook, regulatory documents. Or new attributes can be collected throughout
a discussion with expert or pilot testing with targeted subject pool.
The main domain of attributes in health economic includes money, time, risk, health
care, health status. For the vaccination preference, the attributes of vaccination derived
from literature are: effectiveness, adverse event, cost, number of injection, time, allocation,
physician’s recommendation.


13

There is no rule for the number of attributes to be used, but the multiplying increase
of choice sets by number of attributes requires capture the main attribute for majority of
respondents. Almost studies recruit 4-6 attributes for DEC design.
In case the number of possible attributes exceeds which could lead to the potential
large design, then rating or ranking exercise could be help for maintaining the important
attributes.
Levels of attribute
Once attributes was decided, the levels for each attribute must be identified. They
can be categorical (e.g. central healthcare or district healthcare), continuous (e.g. 1USD or
2 USD cost, 1 or 2 injection) or probability (e.g. 50% efficacy, 5% adverse event…). The
range to define attributes (e.g. cost from 5-10 USD) should be avoid because this requires

interpretation and result in ambiguous.
A sufficiently difference between level should be used to avoid respondents
ignoring attributes because the little difference in levels, especially for the choice set
designed with two comparing alternatives. Unless special required, it does not need span
the full spectrum of levels. For example, we may not use the longest time of waiting in
describing the level for attribute of waiting time. Instead, the interquartile range or at plus
and minus one standard deviation from the mean can help. Number of level is usually
limited to three or four per attribute and the same pace between levels is not required.
Discrete choice experiments design
The experimental design is the combination of the attributes with levels used to
construct the alternatives included in the choice sets.
A full factorial design includes all possible combinations of the levels of the
attributes and allows for estimation of main effects and interaction effects.
A main effect refers to the direct effect of each independent variable (the difference
in attribute in attribute levels) on the dependent variable (choice variable). An interaction


14

effect is the effect of the interaction between two or more independent variables (by varying
two or more attribute levels together) on the dependent variable.
For an experiment with k attributes and each attribute q (q = 1, 2 ... k) defined by lq
levels, the total number of possible combinations, L, is given by the product of the number
of levels for each attribute

. For example, with five attributes (k = 5), two at four levels

and three at five levels (often denoted 4253), there are 4*4*5*5*5 = 2,000 combinations
in the full factorial. A fractional design was chosen to reduce the complexity of full factorial
design which includes the main effects and sometime plus the significant interaction

effects.
For the construction of choice sets, if a binary choice DCE is used (e.g. would you
use this service, yes/no) then the scenarios derived from the full factorial or fractional
factorial design are the choice. If two or more alternatives are employed the scenarios must
be properly placed into choice sets.
And to minimize the number of choice sets required for each respondent, the choice
sets could be blocked into parts and randomly assigned when implementing.
The main objectives of design are identification and efficiency, while the
identification is priority before constructing the design. The identification is referred to the
effects included in the indirect utility function, which requires the chosen relevant attributes
and levels, with the model of main effects and/or interactive effects. The identification
cannot be changed once a design is constructed, while the efficiency could be improved by
increasing the sample size.
Efficiency is related to the precision with which the effects are estimated. In this
respect, proposed desirable design criteria that are orthogonally (attribute level appear with
equal frequency with each level of each other attribute in all the included alternatives),
level balance (the levels of each attribute in appear with equal frequency in all the included
alternatives), minimal overlap (there are as few overlaps of levels as possible for each
attribute in each choice set) and utility balance (the options in each choice set should have


15

similar probabilities of being chosen). The orthogonality is most important character for
design.
The variance-covariance matrix (i.e. det(C-1)) was used to compare different design,
in which the optimal design will have the smallest value of det(C-1) or largest value of
det(C), we called it as det(Copt). With the comparing design of Cd, the D-efficiency of design
is given by:
D-efficiency=

with p is parameters to be estimated

The researcher could use the orthogonal main arrays, or software package (such as:
SPSS, SPEED, SAS) to get the design with D-efficiency score.
Model estimation
If the choice set includes binary choice (yes/no) or only two alternatives, then binary
probit or logit model are appropriate. Both models, binary probit and logit, lead to
equivalent parameter estimates up to scale.
If the study collects multinomial rather than binary choice data, the alternative
model estimation was developed with start using the McFacdden’s multinomial logit
(MNL). The MNL has four important assumptions: (i) identically distributed errors (i.e.,
constant error variance or homoscedasticity); (ii) independent errors (i.e., independence of
irrelevant alternatives (IIA); thus assuming that all options are equal substitutes); (iii) no
panel data (i.e., no correlation allowed for within responses); and (iv) no taste variation
(i.e. homogenous preferences across respondents).
Three alternative families of models which were developed to relax the restrictions of the
McFadden’s MNL model (Ryan, M. et al., 2008): (i) the heteroscedastic model, which
relax the assumption of identically distributed errors; (ii) the generalized extreme value
(GEV) models, which relax the assumptions of independent errors and allow for random
taste variation (MNL, GEV, and heteroscedastic models allow for heterogeneous
preferences using sub-group analysis; the key contribution of the flexible models is that