health-related quality of life measure (eq-5d-5l ... · studies, mahidol univ. thesis / iv...
TRANSCRIPT
HEALTH-RELATED QUALITY OF LIFE MEASURE (EQ-5D-5L): MEASUREMENT PROPERTY TESTING
AND ITS PREFERENCE-BASED SCORE IN THAI POPULATION
JUNTANA PATTANAPHESAJ
A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR
THE DEGREE OF DOCTOR OF PHILOSOPHY (PHARMACY ADMINISTRATION)
FACULTY OF GRADUATE STUDIES MAHIDOL UNIVERSITY
2014
COPYRIGHT OF MAHIDOL UNIVERSITY
iii
ACKNOWLEDGEMENTS
This project is supported by the Burden of Diseases Project, Thailand; and
the EuroQol foundation, The Netherlands. The Health Intervention and Technology
Assessment Program (HITAP) is supported by the Thailand Research Fund under the
Senior Research Scholar on Health Technology Assessment (RTA5580010) and
ThaiHealth Global Link Initiative Program (TGLIP), supported by ThaiHealth
Promotion Foundation. The findings and opinions in this report have not been endorsed
by the above funding agencies and do not reflect the policy stance of these organizations.
I would like to express my gratitude to all those who gave me the possibility
to complete this thesis. I would like to thank the leader of Health Intervention and
Technology Assessment Program (HITAP), Dr.Yot Teerawattananon for giving me
permission to commence this thesis in the first instance.
I am deeply indebted to Assist Professor Dr. Montarat Thavorncharoensap
(major advisor), Dr.Sirinart Tongsiri (co-advisor) for their guidance, supervision, kindly
suggestions and continual encouragement. Special thanks also to Dr.Lily Ingsrisawang,
Miss Wantanee Kulpeng, Miss Pritaporn Kingkaew for their suggestion on data
analysis.
I am also grateful to Dr.Thunyarat Anothaisintawee, six interviewers, all
field coordinators and all respondents for their kindness and facilitation of the data
collection. Grateful acknowledgement is extended to my class participants of pharmacy
administration program, Mahidol University; Health Intervention and Technology
Assessment Program (HITAP) for their encouragement.
Juntana Pattanaphesaj
Fac. of Grad. Studies, Mahidol Univ. Thesis / iv
HEALTH-RELATED QUALITY OF LIFE MEASURE (EQ-5D-5L): MEASUREMENT
PROPERTY TESTING AND ITS PREFERENCE-BASED SCORE IN THAI POPULATION
JUNTANA PATTANAPHESAJ 5337502 PYPA/D
Ph.D. (PHARMACY ADMINISTRATION)
THESIS ADVISORY COMMITTEE: MONTARAT THAVORNCHAROENSAP, Ph.D.,
YOT TEERAWATTANANON, Ph.D., SIRINART TONGSIRI, Ph.D.
ABSTRACT
This research aimed to develop the population-based preference scores of the EQ-
5D-5L (the 5L), Thai version in an effort to compare the measurement properties of the 5L
with those of the EQ-5D-3L (the 3L), and to compare the results of an economic evaluation
from a 2014 study that used the 5L compared with the 3L.
To elicit population-based preference score, face-to-face interviews using the EQ-VT
protocol was undertaken in 12 provinces across Thailand. A representative sample consisting
of 1,207 recruited individuals was used in a stratified stage sampling and quota sampling of
age and gender. Regarding TTO valuation, 86 health states were grouped into 10 blocks. Each
block contained 10 health conditions. For the Discrete Choice Experiment (DCE) valuation,
196 health states grouped into 28 blocks of 7 pairs of health states were used. For each
participant, the block used for TTO valuation and DCE valuation were randomly selected
through the use of the EQ-VT software application. Regarding the comparison of the
measurement properties, a total of 117 diabetes patients treated with insulin completed a
questionnaire including the 3L, the 5L, and SF-36. Measurement properties were then assessed
in terms of distribution, ceiling effect, convergent validity, discriminative power, test-retest
reliability, and patient preference. The result of economic evaluation using the utility derived
from the 5L was compared with those from the 3L in term of incremental cost-effectiveness
ratio (ICER) and cost-effectiveness acceptability curve (CEAC).
The result of the interview showed no inconsistency among 3,125 possible health
states for the 5L. Random effect model with only primary effects was selected. Mobility had
the greatest impact on preference score. The second best score was 0.968 for state 11112 and
the worst score was -0.283 for worst state (55555). In terms of measurement properties,
evidence supported the convergent validity of both 3L and 5L. However, the 5L showed a
trend towards a slightly lower ceiling effect compared with the 3L (33% vs 29%). It also
showed more promise when compared to the 3L in terms of more discriminatory power, more
reliable index score, and more preferable by respondents. In addition, it was found that the
preference scores derived from the 5L yielded lower ICER and produced less uncertainty than
those derived from the 3L. Thus, the 5L could be recommended as a preferred health-related
quality of life measure in Thailand.
KEY WORDS: HEALTH-RELATED QUALITY OF LIFE / EQ-5D / MEASUREMENT
PROPERTY TESTING / PREFERENCE-BASED SCORE / TARIFF / VALUE SETS
180 pages
Fac. of Grad. Studies, Mahidol Univ. Thesis / v
���������������� ��������� EQ-5D-5L: ก��� !�"� �!�"#��ก���# $%�&��'(�)�#ก���*�+,��-.����ก�/�, HEALTH-RELATED QUALITY OF LIFE MEASURE (EQ-5D-5L): MEASUREMENT PROPERTY TESTING AND ITS PREFERENCE-BASED SCORE IN THAI POPULATION
B#���� �#C���!#� 5337502 PYPA/D
�. . ("��)����!#�ก�B)
��ก���ก������JกK����,�����L- : ����#��- *����B��M��#�,-, Ph.D., ,P ����#C�����-, Ph.D., P�����* ��P���, Ph.D.
"��# ,&� ก��PJกK���'���#�* �!��-������#C���$�����*�+,��-!(�)�#"$""!�"*�� EQ-5D-5L (5L)
Q"#"��K�/�, ���,"���,"$""!�"*�� EQ-5D-3L (3L) $% 5L .� R��� �!�"#��ก���# $%S%ก������������ R��&�����P�KTP�!��-
.�ก���#C���$�����*�+,��-B�ก���ก� (�����ก��+ ,!#���K�-ก% &��#��,&��B(���� 1,207 ��,$""�#��&��#�.� 12 B#�)�# �#�����P+ ,.�R+�+���% EQ-VT ! &��#��,&�� R�,$S�ก��! &��#��,&��)%�,W#'�$%�%��ก)�&�,�#��,&��$""+���R����!# !&��W����, $%��P�����.)Rก% &��#��,&���X��#�$��W�����ก�/�, ก�������� ��&�W��!*��! W��� R�,��L�ก��$%ก�%��,���%� (time trade-off, TTO) (�����ก��+ ,.�R!*��! W���B(���� 86 !*�����$"&��X� 10 ก% &� $�&%ก% &��ก�" R�,!*��! W��� 10 !*�� ก�������� ��&�W��!*��! W��� R�,��L�ก�� discrete choice experiment (DCE) (�����ก��+ ,.�R!*��! W���B(���� 196 !*�����$"&��X� 28 ก% &� $�&%ก% &��ก�" R�,!*��! W��� 7 �j& ก% &��#��,&��$�&%��,������ ��&�W��!*��! W��� R�,��L� TTO $% DCE + ,.�R!*��! W������,�ก% &�� �,�kJ��! &�+ ,+�$ก�� EQ-VT !(�)�#"ก�����,"���,"� �!�"#��ก���# (�����ก��+ ,!#���K�-ก% &��#��,&������X�+���"�)���$%�#กK� R�,���kj%��B(���� 117 ��, R�,$""!�"*�� 3L 5L $% SF-36 ������ �!�"#��ก���# B�กก��ก�B�,W���(���" ���L��%�� �� ������������)���� �(���Bก��B(�$�ก ����������#��$""!�"k'(� $%�����J���.BW��SjR��"$""!�"*�� !&��ก�����,"���,"S%ก������������ R��&�����P�KTP�!��-B�กก��.�R�&����*�+,��-���/ RB�ก$""!�"*�� 5L $% 3L �(�ก����B����B�ก�#���!&���R�� ��!��L�S%!&������� (incremental cost-effectiveness ratio, ICER) $%ก��n� #"����� R��&����,���#"/ R (cost-effectiveness acceptability curve, CEAC)
S%ก��PJกK��"�&��$�����*�+,��-W��!*��! W����#'�)� ����X�// RW�� 5L B(���� 3,125 !*�� ������!� �%R��ก#� $""B(�%�����.�R�(���,�$�����*�+,��-!(�)�#"���ก�/�,���$""B(�%�����L��%$""! &� (random effect model) ����ก�" R�,���L��%)%#ก���,��,&��� �,� S%ก��PJกK��"�&�����ก����%����/)���S%�&��$�����*�+,��-��ก���! !*��! W�������$����� ����! �X�%(� #"���!����� 11112 ���$�� 0.968 !&��!*��! W�������$���R�,���! ��� 55555 ���$�� -0.283 S%ก��PJกK�� �!�"#��ก���# �"�&� 5L �����L��%�� ���R�,ก�&� 3L (33% $% 29%) ��กB�ก��',#��"�&� 5L ��� �!�"#����� �ก�&� 3L .� R���(���Bก��B(�$�ก ����������#��$""!�"k'(�W���$�� #��� $%�����J���.BW��SjR��"$""!�"*�� !&��������������)����/�&$�ก�&��ก#��)�&�� 5L $% 3L .� R��ก������������ R��&�����P�KTP�!��-�"�&�ก��.�R�$�����*�+,��-����(����B�ก 5L .)R�&� ICER ��(�ก�&� 3L $%�&�,% ����/�&$�&���W��S%ก��PJกK� #��#'� 5L ���/ R�#"ก��$��(�.)R�X�$""!�"*��� ��������� R��! W���!(�)�#"���P/�,
180 )�R�
vi
CONTENTS
Page
ACKNOWLEDGEMENTS iii
ABSTRACT (ENGLISH) iv
ABSTRACT (THAI) v
LIST OF TABLES x
LIST OF FIGURES xii
LIST OF ABBREVIATIONS xiv
CHAPTER I INTRODUCTION 1
1.1 Background and rationale 1
1.2 Objectives 3
1.3 Expected outcomes and benefits 4
1.4 Definition of terms 4
CHAPTER II LITERATURE REVIEW 7
2.1 Quality of life 7
2.2 Measurement properties of health status questionnaires 14
2.3 SF-36 questionnaires 19
2.4 EQ-5D questionnaires 20
2.5 Multilevel analysis 30
2.6 Discrete choice experiment and logistic regression 35
CHAPTER III METHODOLOGY 37
Session 3.1 : Development of the Thai population-based preference
scores for the 5L Thai version 37
3.1.1 Study design 37
3.1.2 Study location 37
3.1.3 Study population 38
3.1.4 Selection criteria 38
3.1.5 Data collection method 38
vii
CONTENTS (cont.)
Page
3.1.6 Study procedure 43
3.1.7 Data analysis for TTO valuation 43
3.1.8 Data analysis for DCE valuation 53
Session 3.2: Testing the measurement properties of the Thai version of
the 5L compared to the 3L 55
3.2.1 Study design 55
3.2.2 Study location 55
3.2.3 Study population 55
3.2.4 Selection criteria 55
3.2.5 Data collection method 56
3.2.6 Data analysis 56
Session 3.3: Comparison of economic evaluation results using
preference score derived from the 3L and the 5L 60
CHAPTER IV RESULTS 62
Session 4.1 : Development of the Thai population-based preference
scores for the 5L Thai version 62
4.1.1 Respondent’s characteristics 62
4.1.2 Valuation by TTO 64
4.1.3 Data diagnostic tests 66
4.1.4 Data selection to input in the regression model 68
4.1.5 Testing for the functional form 74
4.1.6 The Thai algorithm and the preference scores 80
4.1.7 Comparing Thai preference score with the interim value
sets from mapping technique 82
4.1.8 Valuation by DCE 83
4.1.9 Country-specific data 85
4.1.10 Qualitative data 86
viii
CONTENTS (cont.)
Page
Session 4.2 : Testing the measurement properties of the Thai version of
the 5L compared to the 3L 90
4.2.1 Characteristics of respondents 90
4.2.2 Distribution and ceiling effect 92
4.2.3 Redistribution 93
4.2.4 Convergent validity 94
4.2.5 Discriminative power 95
4.2.6 Test-retest reliability 96
4.2.7 Coefficient of variation 96
4.2.8 Patient preferences 97
Session 4.3 : Comparison of economic evaluation results using
preference score derived from the 3L and the 5L 97
CHAPTER V DISCUSSIONS 103
CHAPTER VI CONCLUSIONS 114
REFERENCES 115
APPENDICES 125
Appendix A Certificate of ethical consideration 126
Appendix B The example of EQ-VT screen 127
Appendix C EQ-5D-3L Thai version 130
Appendix D EQ-5D-5L Thai version 132
Appendix E Background questions 135
Appendix F TTO Health states included in the EQ-VT 136
Appendix G TTO feedback questions 137
Appendix H DCE pairs included in the EQ-VT 138
Appendix I DCE feedback questions 140
Appendix J Country-specific questions 141
Appendix K Qualitative questions 142
ix
CONTENTS (cont.)
Page
Appendix L Questionnaire for testing measurement property 144
Appendix M SF-36v2 Thai version 150
Appendix N Thai preference score for EQ-5D-5L health states 155
BIOGRAPHY 180
x
LIST OF TABLES
Table Page
2.1 Dimension of health profile measures 10
2.2 Dimension of utilities measures 10
3.1 Magnitude of inconsistent responses of 1 respondent 47
3.2 Variables in the TTO model 48
3.3 Functional forms 49
3.4 Example of more severe health state was preferred 54
3.5 Size of (in) consistent response 57
4.1 Demographic characteristic of respondents 62
4.2 The number of respondents by age, gender, and residential area 63
4.3 Health status of respondents by level of severity 64
4.4 Observed mean values by health state’s profiles 65
4.5 TTO feedback 66
4.6 Number of respondents that met the criteria for low quality data for
TTO valuation 69
4.7 Subgroup classification by low quality data and magnitude of
inconsistency 70
4.8 Mean TTO scores by subgroup 71
4.9 Parameter estimates and the fit statistics by subgroup 73
4.10 Coefficients and fit statistics generated from subgroup 2 by functional
form 76
4.11 Coefficients for main effects of the Thai model 80
4.12 Examples for calculating the Thai preference score for the EQ-5D-5L 81
4.13 Comparing EQ-5D-5L value sets’ parameter between Thai’s and interim
scoring 83
4.14 Number of respondents that met the criteria for low DCE data quality 83
4.15 Coefficients and fit statistics of DCE model 84
xi
LIST OF TABLES (cont.)
Table Page
4.16 DCE feedback 84
4.17 Opinions of the respondents towards philosophy of life 85
4.18 Opinions of the respondents on the most and the least important for each
dimension 89
4.19 Demographic characteristic of respondents 91
4.20 Redistribution pattern of response from the 3L to the 5L 93
4.21 Correlation coefficients between EQ-5D and SF-36v2 dimensions 95
4.22 Shannon index ( ) and Shannon’s Evenness index ( ) of the 3L and the
5L 96 4.23 Test-retest reliability of the 3L and the 5L 96
4.24 Economic evaluation results generated from 3 different value sets 98
xii
LIST OF FIGURES
Figure Page
2.1 Concept of quality of life and health-related quality of life 8
2.2 Visual Analog Scale 11
2.3 Standard Gamble 12
2.4 Time Trade-off method 12
2.5 The lead time TTO 13
2.6 The difference between absolute and relative informativity in a 3-level
system and 5-level system 18
2.7 Scatterplot without median trace, and with median trace 31
3.1 TTO screen for state better than dead 40
3.2 Lead time TTO screen for state worse than dead 40
3.3 DCE screen 41
3.4 Example of utility calculation for TTO valuation 44
3.5 Six criteria used to detect low quality of the data for each respondent 46
3.6 Example of the method used to detect logical inconsistency among 3,125
health states 52
4.1 Scatter plot between TTO value (Y) and level of severity (X) of all
respondents 66
4.2 Spaghetti plot between fitted value (Y) and level of severity (X) of 10
selected respondents 67
4.3 Kernel density estimate of residuals 68
4.4 Probability-probability (P-P) plot of residuals 68
4.5 Magnitude of inconsistency among 1,181 respondents 69
4.6 Mean TTO scores by subgroup 73
4.7 Comparison between actual mean score and predicted score of the model
1-8 using data from subgroup 2 78
4.8 Bland-Altman plots of model 1-8 79
xiii
LIST OF FIGURES (cont.)
Figure Page
4.9 Comparing EQ-5D-5L utility score obtaining from surveying and
mapping 82
4.10 Distribution across severity level of the 3L and 5L dimension 92
4.11 Mean, standard deviation, and coefficient variation of preference score 97
4.12 Cost-effectiveness plane comparing SMBG group to no SMBG group for
DM type 1 & 2 99
4.13 Preference score for comparable health states of the 3L and the 5L 100
4.14 Cost-effectiveness acceptability curve for DM type 1 group 101
4.15 Cost-effectiveness acceptability curve for DM type 2 group 102
xiv
LIST OF ABBREVIATIONS
3L EQ-5D-3L
5L EQ-5D-5L
AD Anxiety/depression
AIC Akaike information criterion
CCC Concordance correlation coefficient
DALY Disability Adjusted Life Year
DCE Discrete choice experiment
DM Diabetes
EA Enumeration areas
EQ-5D EuroQol-5dimensions
EQ-VT EuroQol Group’s Valuation Technology
GDP Gross domestic product
H’ Shannon index
HR-QoL Health-related quality of life
HTA Health Technology Assessment
HUI Health Utilities Index
ICC Intraclass correlation coefficient
ICER Incremental cost-effectiveness ratio
IHRP Institute for the Development of Human Research Protections
J’ Shannon’s Evenness index
MO Mobility
MU-IRB Mahidol University Institutional Review Board
MVH Measurement and valuation in Health
NHP Nottingham Health Profile
NICE National Institution for Clinical Excellence
NLEM National List of Essential Medicine
NSO National Statistical Office
xv
LIST OF ABBREVIATIONS (cont.)
OLS Ordinary least square
PD Pain/discomfort
PPS Probability proportional to size
QALY Quality adjusted life year
QoL Quality of life
QWB Quality of Well-Being
RMSE Root mean square error
SC Self-care
SF-12 The SF-12 Health Survey
SF-36 The Medical Outcomes Study 36-item Short-Form
SG Standard Gamble
SIP Sickness Impact Profile
SRM Standardized response mean
TTO Time trade-off
UA Usual activities
UK The United Kingdom
US The United State of America
USD United States Dollar
VAS Visual analog scale
WHO World Health Organization
WTD Worse than dead
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 1
CHAPTER I
INTRODUCTION
1.1 Background and rationale
Due to scarcity of health resource and increasing high cost of available
health technology, demand for health technology assessment (HTA) evidences is
increasing. Recently, HTA has been a significant tool for evidence-based policy
decision making in Thailand (1, 2). According to the Thai national guidelines of HTA
(3, 4), a cost-utility analysis was recommended as a preferred method for assessing the
cost-effectiveness of health technology. For cost-utility analysis, outcome of
intervention or health technology is measured in terms of quality-adjusted life year
(QALY), which enable comparison across different types of health technology. QALY
is suitable to use as measure of health outcomes as it is the most comprehensive and
encompassing both the quantity and quality of life aspects (5). In other words, QALY
is calculated by the amount of life expectancy multiply by the utility score, which is
varied by each individual's preferences of his/her health status. Utility score can be
ranged from 0 (the worse health state or dead) to 1 (the perfect health). At present,
cost-effectiveness threshold endorsed by the Thai National List of Essential Medicine
(NLEM) Committee is 1.2 Gross National Income (GNI) per QALY or approximately
160,000 Baht (6). Based on the given threshold, it can be interpreted that any
pharmaceutical or technology that produces 1 QALY gained with its costs less or
equal to 160,000 Baht or 1.2 GNI will be considered as cost-effective medicine.
At present, guidelines published by many HTA organizations including the
National Institution for Clinical Excellence (NICE) (7), the US panel on Cost-
effectiveness in Health and Medicine (8), and the Thai national guideline of HTA (9)
has recommended EQ-5D as the preferred instrument for assessing the utility for HTA
studies. The EQ-5D, a widely used general health questionnaire for describing and
valuing health outcome, has been developed since 1980s (10). The first version has
Juntana Pattanaphesaj Introduction / 2
now been translated into more than 150 languages and is used worldwide. This is
because it is easy to response, taking a few minutes to complete, so the respondent can
answer by themselves. It is also suited for use in clinical research, and face-to-face
interviews. The first part of questionnaire contains five dimensions i.e. mobility, self-
care, usual activities, pain/discomfort and anxiety/depression. Each dimension of the
EQ-5D-3L (after this “the 3L”) comprises three levels of impairment namely no
problems (level 1), some/moderate problems (level 2), and extreme problems (level 3),
which generates 243 possible health states. The second part values health status using
visual analog scale (VAS), which is a utility measurement using direct method. It is a
20 centimeter vertical scale designed for self-rated. The scale ranges from 0 to 100,
where 0 means dead or the worst health you can imagine and 100 mean the best health
you can imagine. Nevertheless, it should be noted that utilities score should be
calculated from the first part using country-specific preference score or value sets.
In Thailand, the preference scores for the 3L health states were established
since 2009 by Tongsiri et al (11, 12). In estimating the preference-based score for the
3L, a total of 1,409 Thai respondents were randomly selected for interviewing. A total
of 86 health states, grouped into 12 blocks, were employed in the interview. The utility
was directly elicited by time trade-off (TTO) method using Measurement and
valuation in Health (MVH) protocol. Health state 11112 was the second best health
(preference score = 0.766) while the worst health was 33333 (score = -0.454).
As the 3L is limited to three levels of response categories, a substantial
ceiling effect was observed (13-18). In other word, the respondents who are near
highest possible score can’t show any health improvement. In addition, it has
limitations in measuring small changes, especially in mild conditions (19-22).
Previous studies also found that the 3L appeared to be less sensitive when compared to
the SF-12 or SF-36 (13, 14).
In response to the problems previously mentioned, the 5-level of EQ-5D
(EQ-5D-5L, after this “the 5L”) was developed by a task force of the EuroQol group
in 2005 (19, 20). This version includes five levels of impairment for the existing five
dimensions of the EQ-5D. At present, the 5L has now been translated into more than
113 languages including Thai (23). Several studies (21, 22, 24-30) examining the
measurement properties of the 5L have found that it is a reliable and valid measure.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 3
When comparing the 5L with the 3L, it was found that the 5L had a lower ceiling
effect (22, 24-27, 29, 30) and greater discriminative power with the potential to better
detect the differences between groups (21, 22, 24, 26, 27, 30). In addition, it showed
better face validity (19, 21, 31) and test-retest reliability (24, 27, 29).
To our knowledge, the Thai population-based preference scores for the 5L
have not been developed. In the meantime, an interim value sets generated by the
mapping method between the 3L and 5L index scores has been reported by the
EuroQol group (32). The crosswalk project consisted of 3,691 respondents from 6
countries: Denmark, Scotland, England, Poland, Italy, and the Netherlands. In the
project, the respondents had to complete both the 3L and 5L. The statistical
relationship between these two measures is established through regression methods
and presented as an algorithm or a formula. Then, transitional probabilities from 5L to
3L health state are generated. The 5L value sets are obtained by multiplying the 3L
tariffs with their 243 transition probabilities. However, artificial floor effect is
observed. In addition, when mapping the 5L to the 3L value sets, the mapping method
does not allow the value of worst health state (55555) to be lower than that of 33333
because the range of 5L score was limited within the range of the 3L. In addition,
since transitional probabilities were combined from many countries where the
translation process for the 5L was different, cultural difference may occur thus limited
the validity of the value set estimated. At present, the mapping value sets for the 5L
are available for Thailand as well as the following countries: Denmark, Spain, UK,
US, France, Germany, Japan, the Netherlands, and Zimbabwe (33, 34).
Based on the limitation of the 3L mentioned above and the unavailability
of the Thai population-based preference scores for the 5L, there is a clear need to
develop the population-based preference scores for the 5L Thai version for use in
health technology assessment and also to assess the measurement properties of the 5L
in comparison with the 3L.
1.2 Objectives
The objectives of this study are;
1. to develop the Thai population-based preference scores for the 5L,
Juntana Pattanaphesaj Introduction / 4
2. to test measurement properties of the 5L compared to those of the 3L,
and
3. to compare the results of economic evaluation using utilities derived
from the Thai population-based preferences scores for the5L, which were developed
from this study, with those of the 3L
1.3 Expected outcomes
This study is the first study aims to examine the measurement properties of
the 5L (Thai versions). The measurement properties established and documented from
this study will provide vital evidences to justify the use of such instrument to measure
quality of life among Thai population in the future. More importantly, the population-
based preference score for the 5L generated from this study will allow clinicians and
researcher to calculate utility value, which is crucial information for evaluating effect
of health intervention, monitoring treatment process, as well as conducting HTA
research to support health policy decision making. In addition, the results can be used
for international comparison in order to understand similarities and differences of
health preference across population.
1.4 Definition of terms
Cost-effectiveness acceptability curve
Cost-effectiveness acceptability curve (CEAC) was used to summarize the
information on uncertainty in cost-effectiveness analysis (35). It was drawn by plotting
the proportion of the costs and QALY pairs that were cost-effective for the maximum
acceptable ceiling ratio. The CEAC shows the probability that the intervention is
worth for a given value of the maximum acceptable ceiling ratio.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 5
Cost-effectiveness analysis
Type of economic evaluation that measures consequences or health
outcome in terms of physical units (36).
Cost-utility analysis
Type of economic evaluation that measures consequences or health
outcome in terms of utility unit (e.g. QALYs) rather than in physical units (36).
Economic evaluation
A comparative analysis of alternative technology in terms of their costs
and consequences (36).
Health-related quality of life
The value assigned to duration of life as modified by the impairments,
functional states, perceptions, and social opportunities that are influenced by disease,
injury, treatment, or policy (37).
Health states
The differentiated stages of a lifetime or disease progression which can be
temporary or permanent (38).
Incremental cost-effectiveness ratio (ICER)
The ratio between the increase in cost and the increase in mean
effectiveness which is commonly used to compare interventions (39). Lower ICER
indicates better value for money compared with alternative intervention.
Quality of life
The satisfaction of an individual’s value, goals and needs through the
actualization of their abilities or lifestyle (40).
Reliability
The degree of consistency between two measures of the same thing (41).
Juntana Pattanaphesaj Introduction / 6
Preference-based measure
The preference-based measure is an instrument that define an individual’s
health conditions for using in an economic evaluation. Each possible health state is
associated with an estimate of the value (preference or utility weight) that a surveyed
sample of the general population has attributed to these health states. These
preference-based measures are used for estimation of QALYs (38).
Preference
The umbrella term that describes the overall concept of utility and
preference. For this study, the term ‘preference’ is used to denote a latent tendency to
consider desirable or undesirable toward health state (42). The term preference and
utility were used interchangeably in this study.
Utility
The utility is a cardinal measure of the preference for, or desirability of, a
specific level of health status or specific health outcome. Its scale is an interval or ratio
scale, defined by 2 anchor states or outcomes and their scores, on which utilities are
measured. Often defined by full health = 1.0 and death = 0.0 (43).
Validity
Validity is the degree to which certain inferences can be made from test
scores. Since a single test may have different purposes, there is no single validity
index for a test (41).
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 7
CHAPTER II
LITERATURE REVIEW
The chapter consists of 4 parts as follows:
2.1 Quality of life
2.1.1 Concept and definition
2.1.2 HR-QoL measurement
2.2 Measurement properties of health status questionnaires
2.2.1 Reliability
2.2.2 Validity
2.2.3 Discriminatory power
2.3 SF-36 questionnaires
2.4 EQ-5D questionnaires
2.4.1 The Thai version of EQ-5D-3L
2.4.2 Measurement properties of the EQ-5D-3L
2.4.3 Measurement properties of the EQ-5D-5L
2.4.4 Comparing the measurement property between the 3L and 5L
2.4.5 Interim method for mapping EQ-5D-5L to EQ-5D-3L value sets
2.4.6 EQ-VT protocol and health state design for the 5L valuation
2.5 Multilevel analysis
2.6 Discrete choice experiment and logistic regression
2.1 Quality of life
2.1.1 Concept and definition
The consequence of health intervention is usually assessed in term of
clinical outcomes, which mainly provide information to physicians however it does
not interested to patients. Hence, the term ‘quality of life’ is aimed to integrate both
Juntana Pattanaphesaj Literature Review / 8
subjective and objective indicators, a broad range of life domains, and individual
values. Current evidences consistently agreed that quality of life is multidimensional.
It is defined as an overall general well-being that comprises both subjective and
objective assessment of physical, emotional well-being, social, and material together
with the extent of personal development and purposeful activity, all weighted by a
personal value sets (40).
WHO defines quality of life as “the individual’s perception of their position
in life in the context of the culture and value systems in which they live and in relation
to their goals, expectations, standards and concerns”. It is a broad ranging concept
affected in a complex way by the person's physical health, psychological state, level of
independence, social relationships, personal beliefs and their relationship to salient
features of their environment (44). It includes positive well-being, rather than lack of
disease. Regarding “health-related quality of life” (HR-QoL), it is used to specify
“health” and exclude other aspects of life that are not generally considered as “health”
such as income, freedom, and environment (45). Although they may adversely affect
health, these problems are often distant from a health or medical concern. Health-related
quality of life is defined as the value assigned to duration of life as modified by the
impairments, functional states, perceptions, and social opportunities that are influenced
by disease, injury, treatment, or policy (37), as shown in Figure 2.1.
Figure 2.1 Concept of quality of life and health-related quality of life
(adapted from Wilson et al, 1995(46))
HR-QoL Overall QoL
Non medical factors
Biologic / Physiologic
variables
Symptom status
Functional status
General health
perception
Characteristic of environment
Psychological, social and economic support
Characteristic of individual
Motivation, values preferences, belief
e.g. material possession, work life, education, etc
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 9
2.1.2 HR-QoL measurement
HR-QoL provides useful information to patients, clinicians, health care
payer, and healthcare administrators. HR-QoL construction includes physical health,
mental health, social health and general health from the individual’s perspective (45).
There are 2 basic approaches of HR-QoL measurement: specific and generic
instruments. The specific instrument focuses on characteristics of health conditions
which are specific to the area of interest. This kind of instrument has more
responsiveness than generic instruments as it includes only important characteristics of
HR-QoL that are applicable to their patients. The instrument may be specific to the
diseases, certain problem, or population studied also.
The generic instruments include health profiles, and preference-based
instrument. Health profiles provide a set of scores of individual domains. The
summary score is derived by averaging across scales or domains. The examples of
health profile instrument are SF-36, Sickness Impact Profile (SIP), and Nottingham
Health Profile (NHP) (47). The dimensions of these instruments are demonstrated in
Table 2.1.
The other type of generic instrument, preference-based measures, is
developed from economic and decision theory. It reflects the strength of preferences
that individual has for treatment process or outcome. The term ‘preference’ is an
umbrella term that describes the overall concept of utility and preference. It is used to
indicate a hidden tendency to consider undesirable or desirable toward health state
(26). Utility refers to the preferences of the individual over process and outcome (37).
In the context of health, utility could be defined as “a cardinal measure of the
preference for a specific health conditions”.
Preference-based instrument offer a summary score known as utility, but it
does not show the domains in which improvement. The key elements of preference-
based instrument is that it includes both preference measurements and relate health
conditions to death. Thus, it can be employed in cost-utility analysis (5, 45). The
examples of utility measure are EQ-5D, Health Utilities Index (HUI) Mark 3, and
Quality of Well-Being (QWB). Dimensions of HR-QoL utility measures are
demonstrated in Table 2.2.
Juntana Pattanaphesaj Literature Review / 10
Table 2.1 Dimension of health profile measures (47)
NHP SF-36 SIP
1. Energy level
2. Emotional reactions
3. Physical mobility
4. Pain
5. Social isolation
6. Sleep
1. Physical functioning
2. Role limitations due to
physical problems
3. Bodily pain
4. General health perceptions
5. Vitality
6. Social functioning
7. Role limitations due to
emotional problems
8. Mental health
Physical dimension
1. ambulation
2. mobility
3. body care and movement
Psychosocial dimension
4. communication
5. alertness behaviour
6. emotional behaviour
7. social interaction
Independent categories
8. sleep and rest
9. eating
10. work
11. home management
12. recreation and pastimes
Table 2.2 Dimension of utilities measures (47)
QWB EQ-5D HUI Mark 3
1. Mobility
2. Physical activity
3. Social activity
4. Symptoms/problems
1. Mobility
2. Self-care
3. Usual activity
4. Pain/discomfort
5. Anxiety/depression
1. Vision
2. Hearing
3. Speech
4. Ambulation
5. Dexterity
6. Emotion
7. Cognition
8. Pain
There are 2 approaches to measure utility.
1) Direct measurement
For direct methods, respondents are requested to rate the desirability of
different health conditions. They order their preferences, and making trade-offs
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 11
between health conditions and alternatives. The examples of direct measures are
Visual Analogue Scale (VAS), Standard Gamble (SG), and Time Trade-off (TTO).
Visual Analogue Scale (VAS) or another word ‘rating scale’ is a simple
line with defined end points on which respondents are able to indicate their
preferences. The line may be vertical or horizontal, may have interval marked out with
numbers as shown in Figure 2.2. VAS does not offer a choice and has no basis in
economic theory. However, it is still widely used to elicit preferences as it takes less
time and easy administration (48, 49).
Figure 2.2 Visual Analog Scale (10)
10
80
100
90
70
60
50
75
95
85
65
55
45
40
35
30
25
20
15
Full health
5
0
Dead
Juntana Pattanaphesaj Literature Review / 12
In Standard Gamble (SG), the respondents were requested to choose an
option between alternative health outcomes in the uncertainty situation. Given that the
individual was in a health state ‘i’. Each individual then has to express his/her
preference by choosing between two alternatives (Figure 2.3). Alternative one is
receiving treatment with 2 possible outcomes which are recover to perfect health with
a probability ‘p’ or dead with probability ‘1-p’. Alternative 2 is to remain in health
state i for certain period. The probability ‘p’ then be changed until the individual
cannot tell the different between these two alternatives. This probability ‘p’ will be
equal to the utility of health state ‘i’ (5).
Figure 2.3 Standard Gamble
Time trade-off (TTO) method was established from SG, and it was also
designed to decrease the difficulties of explaining probabilities to respondents. The
TTO method requests respondent to select between 2 options. For example, the
number of years (x) in full health and the number of years (t) in the valued health state
(state i), as shown in Figure 2.4. Then the respondent will be asked to reduce the
number of year in full health (x), until they are indifferent between 2 alternatives. The
health state’s utility is calculated from x by t (Ui = x/t) (48, 49).
Figure 2.4 Time Trade-off method (50)
Perfect health
Dead
Alternative 1
Alternative 2
p
1-p
Current state
Ui = x / t healthy
Dead
Full health
State i
x t
TIME
VALUE
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 13
Using conventional TTO to elicit worse than death health state is
problematic because the trade-offs task for the better than death health state was
different from worse than death health state. The method used to elicit preference of
state worse than death in Measurement and Valuation of Health (MVH) generates
extreme negative values (51). The lead-time TTO has been introduced to measure state
worse than death. The lead-time TTO technic can be applied equally for both better
than death and worse than death health state. This approach introduces ‘lead time’ in
perfect health prior to each of the alternatives (Figure 2.5). This approach can avoids
different valuation methods for better than death and worse than death health state.
To calculate utility from the lead time TTO, the lead time which is full
health state will be minus from both the numerator and the denominator in order to
give a result comparable with the usual TTO. If the lead time is 10 years, Ui = (x-10) /
(20-10), when x is number of year in full health of Life A at the point of indifference.
If x is greater than lead time in Life B, the utility will be positive. If x is less than lead
time in Life B, the utility will be negative.
Life A
Life B
Life A
Life B lead time
Figure 2.5 The lead time TTO (51)
Dark cell = full health, diagonal cell = health state i, white cell = death
Health state i = 10 yr
10 yr
20 yr
Health state i = 10 yr
Better than dead health state
Worse than dead health state
Juntana Pattanaphesaj Literature Review / 14
2) Indirect measurement
For indirect method, individual reports their health state’s preference
through multi-attribute instruments. Then, the answers are calculated by preference
function to obtain the individual’s utility. The preference function have been generated
through population survey study, which direct methods were used to elicit preference
of the possible health state (47). Health Utilities Index (HUI) and EQ-5D is the
example of indirect measures.
2.2 Measurement properties of health status questionnaires
2.2.1 Reliability
Reliability is the degree of consistency between two instruments of the
same thing (41). High reliability is needed for discriminative purpose (52). Reliability
coefficient ranges from 0 to 1. The weighted Cohen’s Kappa coefficient can be
employed to demonstrate reliability for ordinal measures. For continuous data,
intraclass correlation coefficient (ICC) will be derived from the variation in the
population (interindividual variation) divided by the total variation, which is the
interindividual variation plus the intraindividual variation (measurement error).
According to Fleiss’s standards for the strength of agreement for kappa values (53),
Cohen’s weighted kappa (k) was determined as follows: poor reproducibility (k < 0.4);
good reproducibility (0.4 < k < 0.75 ; excellent reproducibility (k > 0.75).
The methods used to test reliability differ in that they consider different
sources of error. For instance, test-retest reliability is used when we concern about
stability, and split-half method is used when we concern internal consistency (41).
The test-retest reliability testing is aimed to test the precision of measures
by administering the same test at a later time. The correlation coefficient between two
scores is used to demonstrate the stability of the individual’s scores (54). The duration
between the two tests should be long enough in order that the respondents can’t
remember the previous answers. However the time period should short enough in
order that the clinical symptoms are not changed. The appropriate period is 1-2 weeks,
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 15
however it is possible to be otherwise. At least 50 patients is required for the
assessment of reliability (52).
2.2.2 Validity
Validity is defined as the extent to which certain inferences can be made
from test scores or other measurement (41). The degree of validity is important as it is
useful oversimplification to think of validity as truthfulness. The magnitude and
direction of the relationship between two instruments is used to demonstrate the
validity (54). Pearson product-moment correlation coefficient (r) is a correlation
coefficient for continuous variables, while Spearman’s rank-order correlation (rs) is
employed for ordinal data. The correlation coefficient can range from -1 (perfect
negative relationship) through 0 (no systematic correlation), to +1 (perfect direct
relationship). The strength of correlation is determined as follows: absent (r < 0.20),
weak association (0.2≤ r< 0.35), moderate (0.35≤ r<0.50), and strong (r≥0.50) (55).
Content validity is related to the sufficient content of the test samples that covers the
domain about which inferences are to be made. The numerical expression is rarely
used for content validity. It is rather considered by a thorough inspection of the items.
Two persons may well make judgments about the match of the items to the domain.
Then, agreements of ratings could be calculated (41).
Criterion validity refers to the extent to which scores on a particular
instrument relate to a gold standard. It pertains to the empirical technique of studying
the relationship between the tests scores (predictors) and some independent external
measures (criteria). Some researchers make a distinction between two kinds of
criterion validity: concurrent validity and predictive validity. The only distinction
between these 2 types of validity is the ‘purpose of testing’ and ‘time period’ when the
criterion data are gathered. For concurrent validity, the data are collected at
approximately the same time as the test data. In addition, we are asking whether the
test score can be substitute for some less efficient way of gathering criterion data. For
predictive validity, the data are gathered at a later date and we are concerned with the
usefulness of the test score in predicting some future performance (41, 52). On the
other hand, construct validity refers to the extent to which scores on a particular
instrument relate to other measures which is consistent with theory concerning the
concepts that are being measured (52).
Juntana Pattanaphesaj Literature Review / 16
2.2.3 Discriminatory power
The Shannon index was initially developed from the information theory
(56). It normally is employed to measure diversity and richness of information in the
communications industry and the ecosystems. The Shannon index can be applied as a
quantifier of the information content of any classification system and can be used to
compare the informational richness of measures. Also, it could reflect discriminatory
power of the instrument. The Shannon index (H’) is defined as follows (56) :
Where;
H' = the absolute amount of informativity captured
C = the number of possible categories (levels)
pi = ni/N, the proportion of observations in the ith category (i = 1,...,C)
ni = the observed number of scores (responses) in category i
N = the total sample size
The Shannon index reflects the absolute information content and depends
on the number of categories. Whereas Shannon’s Evenness index (J') expresses the
relative informativity of a system or ‘evenness’ of a distribution, regardless the
number of categories. In case of an even distribution, the dimension is being most
efficiently used. This means that the discriminant ability of the level descriptors is
maximal.
Shannon’s Evenness index (J') is defined as follows (56) :
Where;
= log2C
C = the number of possible categories (levels)
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 17
The Shannon index can be calculated by dimension separately or whole
instrument. The following is the example of Shannon index calculation by dimension.
Suppose the EQ-5D-3L Mobility dimension is scored by 10 respondents:
no problems (n = 6), some problems (n = 3) and confined to bed (n = 1). Then, C = 3;
N = 10; Plevel1 = 6/10 = 0.6; Plevel2 = 3/10 = 0.3; Plevel3 = 1/10 = 0.1.
Shannon index for Mobility is calculated as
H’ = –((0.6 log2 0.6) + (0.3 log2 0.3) + (0.1 log20.1))
= 1.30
H’max = log23
= 1.58
J’ = 1.30/1.58
= 0.82.
Figure 2.6 illustrates the difference between H’ and J’. Figure 2.6a, which
is two different instruments and both has 3 levels, shows different distribution of
responses. The left figure is skewed distribution, while the right is a rectangular
distribution. This means that the right instrument is superior in discriminating between
patients and that Shannon index and Shannon’s evenness index reached their
maximum values.
Figure 2.6b illustrates the concept of relative informativity (J’). Both left
and right figure shows skewed distribution. However, the right instrument contains 5
levels of response and levels 2 and 4 are unused. As a result, compared to the left, H' is
equal and J' is lower reflecting the unutilized of the 2 levels added.
Figure 2.6c shows the increase in the value of absolute informativity (H').
Given both 3 and 5 level systems produce equally distributions, the value of J' will be
the same, however H' for the 5-level system is greater because it gives more details for
discriminating between respondents.
Juntana Pattanaphesaj Literature Review / 18
a 3-level system 3-level system H’ = 1.34 H’ = 1.58 J’ = 0.84 J’ = 1.00
b 3-level system 5-level system H’ = 1.34 H’ = 1.34 J’ = 0.84 J’ = 0.58
c 3-level system 5-level system H’ = 1.58 H’ = 2.32 J’ = 1.00 J’ = 1.00
Figure 2.6 The difference between absolute and relative informativity (H’ and J’) in a
3-level system and 5-level system
2.2.4 Coefficient of variation
Coefficient of variation (CV), is a measure used to show the extent of
variability in relation with the mean of the sample (57). It is obtained by the ratio of
the standard deviation to the mean. It is often presented as a percentage. The
advantage is that %CV allows comparison to other measure, while standard deviation
is often difficult to interpret or compared with other measure as its value based on the
sample data.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 19
2.3 SF-36 questionnaires
The Medical Outcomes Study 36-item Short-Form (SF-36) Health Survey
is a generic health survey which widely used both in Thailand and international. It has
been translated into more than 50 different languages (58). It contains 36 items,
divided into 8 dimensions, i.e. physical functioning (10 items), role limitation due to
physical problems (4 items), bodily pain (2 items), general health perceptions
(5 items), social functioning (2 items), vitality (4 items), role limitations due to
emotional problems (3 items), general mental health (5 items), and a single item
provided perceived change in health (47). The scoring system is a weighted Likert
scale. The items within the same dimension will be summed and then the summed
score is converted to a scale from 0 – 100 (100 indicating the best health level). The
SF-36 is suitable for the persons aged 14 or over. It can be managed by self-report,
computer, or face-to-face. Thus the SF-36 could be used among both general
population survey and clinic research.
Test-retest reliability study of the SF-36 was undertaken in the UK (59). In
that study, the SF-36 questionnaires were delivered to general populations by postal
survey. The second test among subsample of respondents were two weeks later. One
hundred and eighty seven responded the second test. The scores of the second test
were highly related with the first survey. The correlation coefficient ranged from 0.60
and 0.81. For discriminative validity, the study revealed that the SF-36 demonstrated
the best ability to discriminate between groups compared with other generic
instruments, i.e. EQ-5D, Nottingham Health Profile (NHP) (60).
The Thai version of SF-36 was retranslated using forward-backward
method in order to improve its reliability and validity of previous version (61). From
Thai population survey in 2005 (61), it took about 6 minute for self-complete by Thai
people. A psychometric property testing was undertaken by self-complete
administration in 448 respondents. Convergent validity was evaluated with multitrait
scaling. The correlation coefficient between items and its’ dimension which exceed
0.4 were accounted. The study found that the convergent validity were 96.3%.
Discriminant validity was considered from the correlation coefficient between items
and items of other dimension which over 2 times of standard error (2SE) or 0.092. It
was found that average correlation coefficient was 95%. Cronbach’s alpha coefficients
Juntana Pattanaphesaj Literature Review / 20
of all dimensions over 0.7, ranged from 0.72 and 0.86. The rate of missing data was
low (1.2%). Factor analysis yielded pattern for factor correlation of the new version
were comparable to that found in the previous version. It can be concluded that the
retranslated Thai version of SF-36 demonstrated good reliability and validity, except
vitality and role-emotional dimension. Thus, the results should be interpreted with
awareness.
2.4 EQ-5D questionnaires
The EQ-5D, a widely used generic instrument for describing and valuing
health outcome, has been developed since 1980s by the international multidisciplinary
researchers network named “EuroQol group” (10). Alan William was the important
economist who inspired and drove the EuroQol group's task. The group's founder
members came from various disciplines such as health economics, medicine,
sociology and psychology, academia, health care, public health and government. They
came from 4 countries: Finland, Netherlands, Sweden and the UK. However, at the
present, it includes members from North America, Australia, New Zealand, Africa,
and Asia, (62).
The current version of EQ-5D was published, following a 1993
moratorium on modification that has largely held until the present. It has now been
translated into more than 150 languages and is used worldwide including Thai. This is
because it was designed for self-complete so it is simple, and take only 2-3 minutes to
answer all items. The EQ-5D can be used in both clinical research, general population
survey, and face-to-face interviews. To use EQ-5D questionnaire, the researchers
register the study on EuroQol website. The licensing fees will be considered by the
EuroQol executive member based on the information described in the registration
form. The amount of fee is depending on the funding source, type of study, sample
size and number of requested version (63).
EQ-5D comprises 2 parts. The first part of questionnaire includes five
dimensions i.e. mobility, self-care, usual activities, pain/discomfort and
anxiety/depression. These 5 dimensions are basic “common core” of characteristics of
quality of life which a majority of people concern high value (10). Each dimension of
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 21
the EQ-5D-3L has three levels of responding namely no problems, some/moderate
problems, and extreme problems. Thus, it generates 243 (35) possible health states.
The second part values health conditions using the direct measurement namely VAS.
It is self-rated health on a 20 centimeters scale. The scale ranges from 0 to 100, where
0 means the worst health conditions that you can imagine and 100 mean the best
health conditions that you can imagine. However, utilities score of respondent is
calculated from the first part using country-specific value sets or tariffs.
As the EQ-5D-3L are limited to three levels of response categories, thus
substantial of ceiling effect was observed (13, 14). The respondents who are near
highest possible score can’t show any health improvement. As a result, it has
limitation in measuring small changes, especially in mild condition. In addition, EQ-
5D-3L appeared to be less sensitive when compared to SF-12 or SF-36 (13, 14).
In response to the problems previously mentioned, 5-level of EQ-5D (EQ-
5D-5L version) which comprises 5 levels of responding, has been developed by a task
force within EuroQol group since 2005 (19, 20). Several studies showed that EQ-5D-
5L had lower ceiling effect and greater power to distinguish the difference between
groups when compared with EQ-5D-3L (21, 22). In addition, it showed greater face
validity, and test-retest reliability also (21). At the present, 118 language versions of
the EQ-5D-5L are available (23).
2.4.1 The Thai version of EQ-5D-3L
In Thailand, the Thai version of EQ-5D-3L has been developed and widely
used. Thai national guideline of HTA has also recommended EQ-5D-3L as the
preferred methods for assessing the utility for health technology assessment (3, 4). The
Thai preference scores for the EQ-5D-3L health states were also established since
2009 by Tongsiri et al (11, 12). In estimating the preference-based score for EQ-5D-
3L, a sample of 1,409 Thai respondents from 17 provinces were interviewed. The
sample size calculation and the random selection were conducted by the National
Statistical Office (NSO), Thailand, using multistage stratified sampling method. The
study design followed standardized method namely Measurement and valuation in
Health (MVH) protocol. A total of 86 health states were arranged into 12 blocks. Each
block contains 11 health states which included two anchor states (11111 and 33333), 3
Juntana Pattanaphesaj Literature Review / 22
mild health conditions, 3 moderate health conditions and 3 severe health conditions.
The respondents were asked to rank and score all eleven health states according to
their preference. Then, the utility was directly elicited by time trade-off (TTO)
method. The algorithm was generated as follows:
Thai utility = 1 - 0.202 - (0.121 * mo) - (0.121 * sc) - (0.059 * ua) -
(0.072 * pd) - (0.032 * ad) - (0.190 * m2) - (0.065 * p2) -
(0.046 * a2) - (0.139 * N3)
where
mo = mobility
sc = self-care
ua = usual activities
pd = pain/discomfort
ad = anxiety/depression
The way to replace the variable is as follows:
If the answer is level 1, replace mo, sc, ua, pd, or ad with 0.
If the answer is level 2, replace mo, sc, ua, pd, or ad with 1.
If the answer is level 3, replace mo, sc, ua, pd, or ad with 2.
This study (11, 12) found that only mobility, pain/discomfort, and
anxiety/depression dimension, if the answer is 3, replace m2, p2, or a2 with 1. Replace
those variables with 0 if the answer is level 1 or 2. For all dimensions, if the level 3 is
responded at least once, replace N3 with 1.
In addition, the study found that the second best state is 11112 (score
0.766), and the worst state is 33333 (score -0.454). However, the current model still
suffers from floor and ceiling effects.
2.4.2 Measurement properties of the EQ-5D-3L
Test-retest reliability
The test-retest reliability of the EQ-5D-3L was undertaken in 20
rheumatoid arthritis patients who reporting no change in their arthritis (64). They were
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 23
asked to complete a second set of questionnaires after a 2 week interval. The intraclass
correlation coefficient (ICC) was used to demonstrate test-retest reliability. The study
found that the ICCs for EQ-5Dvas and EQ-5Dutility was 0.85 and 0.78, respectively.
These findings were similar to the study by Dorman et al (65). One-third of 2,253
patients with stroke was randomly sampled for secondly response within
approximately 3 weeks to the first responding. The ICC for overall dimension was
0.86. Thus, reproducibility of EQ-5D-3L was excellent.
The reproducibility of the EQ-5D-3L Thai version was also testing among
Thai people (66). After the first of face-to-face interview, one-fifth of 303 type 2
diabetic patients (n=64) was randomly selected to conduct 1-2 week of test-retest
reliability via telephone. The ICC which was calculated using preference weights from
the UK, US, and Japan was 0.74, 0.74, and 0.78, respectively. This study showed that
the reproducibility of the Thai version of EQ-5D-3L was good. However, if the second
test was administered by face-to-face interview, the test-retest reliability of the Thai
version of EQ-5D-3L may be better than the existing results.
Construct validity
Construct validity of the EQ-5D-3L were tested in general population in
Sheffield by postal survey (13). The people aged 16-74 years were randomly selected.
This study found that preference scores were distributed as expected among variables
(e.g. the sociodemographic variables, the diagnosis of health problems, and health
service received). The professional and managerial groups reported better health than
employee. The person who just received health services had worse health than the
person who did not use health services. The patients with chronic health problems had
poorer scores on the functioning dimension. However, the EQ-5D-3L generated
similar score for a matched sample with and without a diagnosis of chronic physical
problem, even though the differences was as expected.
Construct validity of the Thai version of EQ-5D-3L was also assessed
among Thai people (67). A large cohort (n=4,850) of occupational population was
studied. The respondents completed the Thai EQ-5D-3L and Short-Form 36 version 2
(SF-36v2), which was selected as the gold standard due to its widespread use in
clinical research and its validity among Thai people (68). The construct validity was
Juntana Pattanaphesaj Literature Review / 24
considered by testing the relationships between the EQ-5D-3L and SF-36v2. The
construct validity was agreed with expected relationships with SF-36v2 scale and total
scores. For example, the respondents who had problems (answer level 2 or 3) for any
of the EQ-5D dimensions reported lower SF-36v2 scores than the respondents who
reported no problems. Nevertheless, the EQ-5D-3L in Thai version showed a
considerable ceiling effect, with 48.7% of participants having an index score of 1.
According to the study, good construct validity of the EQ-5D-3L Thai version is
confirmed among the occupational population in Thailand.
The discriminative power of the EQ-5D-3L English version was
established by testing the relationship between the EQ-5D-3L and SF-36 (13). The
study was undertaken in general population aged 16-74 years in Sheffield by postal
survey. It was found that the distribution of the EQ-5D-3L responses is more skewed
than the UK SF-36 scores among comparable dimensions. The larger ceiling effect for
the EQ-5D-3L dimensions was confirmed. The percentage of the ceiling of the EQ-
5D-3L was higher than the SF-36 for the functional dimensions. A few categories of
the EQ-5D-3L leaded to skewness of the distributions.
Regarding measurement properties of the 3L, the findings were different
across type of patients. From systematic review (69), which assessing construct
validity and responsiveness of four generic measures in schizophrenia, the 3L gave
mixed evidence. The correlation between the 3L index score and specific disease
ranged from weak to strong. However, the correlation between the 3L dimensions
(anxiety/depression) and symptom or functioning measures was strong. For
responsiveness, the VAS score and health state dimensions (e.g. anxiety/depression)
were responsive to change in patients. While index scores did not respond to changes
in most symptom or functioning measures. Similar results were found in another
systematic review concerned about using generic measures among patients with visual
disorders (70). The performance of the 3L in visual disorders was also mixed. In
patients with age-related macular degeneration (AMD), the 3L was unable to
differentiate between severity levels and did not correlate well with other measures.
However, it performed well in patients with conjunctivitis. In patients with glaucoma,
the 3L distinguished between different levels of severity although it was not always
statistically significant.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 25
On the other hand, a review on the validity and comparative performance
of generic scales in rheumatoid arthritis (RA) (71) demonstrated that the 3L show good
validity and responsiveness for use in RA. Similar results were found in another review
concerned about using the 3L among patients with cardiovascular disease. It was found
that the validity and reliability of the 3L showed fairly strong convergent validity when
assessed by correlations with other HR-QoL measures. Also, the study demonstrated
good discriminatory power in detecting the changing of health status by a given clinical
degree. Nevertheless, the index values demonstrated significant ceiling effects.
Although the 3L demonstrated good reliability, validity, and discriminative
ability among general population. However, it may not appropriate for using in patients
with some specific diseases, for example, visual disorders, due to its performance.
2.4.3 Measurement properties of the EQ-5D-5L
Validity
The investigation of the face validity and content validity of the English
version of the 5L was undertaken in healthy participants and patients with chronic
diseases in the United Kingdom in 2008 (19). The study aimed to assess the ease of
use, comprehension, interpretation, and acceptability of two alternative 5-level
versions: alternative 1 ‘No problems-Minor problems-Moderate problems-Major
problems-Unable to’; alternative 2 ‘No problems-Slight problems-Moderate problems-
Severe problems-Unable to’. The study revealed that the health states based on new
labeling were relatively easy to understand. Regarding response scaling, the
participants preferred alternative 2, which used ‘slight’, ‘moderate’, and ‘severe’ for
the central levels.
The study regarding convergent validity of the 5L were conducted in
cancer patients (22). In this study, the participants completed the generic instruments
(English version of the 3L, the 5L, VAS) and the disease specific instruments
(Functional Assessment of Cancer Therapy - FACT). The participants were also
assessed function by health care personnel using Eastern Cancer Oncology Group
(ECOG). It was found that the 5L demonstrated slightly stronger correlations with
ECOG performance status compared with the 3L for all dimensions of health.
Juntana Pattanaphesaj Literature Review / 26
Similarly, there was a stronger correlation between the 5L and FACT. In addition, A
larger ceiling effect, defined as proportion of respondents reporting “no problems” for
all dimensions, was observed in the 3L (17%), as compared with the 5L (11%).
Reliability
Test-retest reliability of the 5L was evaluated among Asian breast cancer
patients (72). The time interval is 30 days after self-administering the baseline
questionnaire and states no change in performance status. The study found that the
ICC for the utility index was 0.81 (95% confident interval = 0.73 – 0.87) which
reflected excellent reproducibility.
2.4.4 Comparing the measurement property between the 3L and 5L
Ceiling effect
The ceiling effect is existed when the respondents who are near highest
possible score can’t show any health improvement. It is measured by the proportion of
respondents reporting “no problems” for all dimensions (11111) (22). The previous
studies consistently reported the decreasing of ceiling effect of the 5L compared with
the 3L. Among general population, the ceiling effect decreased from 44% (the 3L) to
35% (the 5L) in US population (25); and it decreased from 66% (the 3L) to 61% (the
5L) in Korean (29). With regards to patient group, the ceiling effect reduced from 17-
39% in the 3L to 10 – 36% in the 5L (22, 24, 26, 27, 30). It accounted for 3-17%
decreasing.
Considering the ceiling effect of the 5L by dimension, previous studies
found highest ceiling effect (80-97%) in self-care dimension (22, 24, 26, 27, 29, 30);
and the lowest ceiling effect (27-71%) was found in pain/discomfort dimension (24,
27, 29).
Test-retest reliability
The test-retest reliability of index scores was evaluated using the intraclass
correlation coefficient (ICC) and the reliability of each dimension was assessed with
Cohen’s weighted kappa coefficient. The duration of second test of previous studies
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 27
ranged from 7 to 21 days (21, 27, 30). It was found that the reliability of index score
was better for the 5L (0.52 for the 3L and 0.69 for the 5L) (21). With regard to
reliability by dimension, the 5L demonstrated similar or slightly better reliability
compared to the 3L (27, 30).
Discriminative power
All previous studies determined discriminative power using Shannon index
(21, 22, 24, 26, 27, 30), the results consistently showed that the 5L was better than the
3L in terms of discriminative power or informativity. However in terms of Shannon
evenness index which reflects the rectangularity of response distribution, the
evidences showed that the 5L was similar or slightly better than the 3L.
2.4.5 Interim method for mapping EQ-5D-5L to EQ-5D-3L value sets
EuroQol group proposed methodology for interim scoring for the 5L in
early 2012 (33). It based on the relationship between responses to the 3L and the 5L
descriptive system. In their model, transitional probabilities from 5L to 3L health state
are generated. Hence, a 3,125 x 243 matrix of transition probabilities was created. The
5L tariffs are obtained by multiplying the 3L tariffs with 243 transition probabilities,
and then subsequently summing of them.
According to EuroQol’s task, 3,691 respondents from 6 countries
(Denmark, England, Poland, Scotland, Italy, and the Netherlands,) completed both 3L
and 5L in order to generate transitional probabilities. Then, the 5L tariffs for the other
countries was calculated (i.e. France, Germany, UK, US, Spain, Japan, the
Netherlands, Thailand, and Zimbabwe) (32). Since transitional probabilities were
combined from many countries, which employed different translation method of the
5L, cultural difference may occur. Another limitation was an artificial floor effect on
5L scores. When mapping 5L to 3L value sets, crosswalk-based approach does not
allow the value of worst health state (55555) to be lower than that of 33333.
2.4.6 EQ-VT protocol and health state design for the 5L valuation
The EQ-VT (EuroQol Group’s Valuation Technology) protocol is
designed by the EuroQol group (73). It provides the methodology used for 5L
Juntana Pattanaphesaj Literature Review / 28
valuation study but no analysis procedures are recommended. The content in the EQ-VT
protocol includes interview preparations (for example, training of interviewer, pilot
test, recruiting a representative sample); guideline for EQ-VT software; interviewing;
and health states used for TTO and DCE valuation.
With regards to the respondents, the inclusion criteria were 1) 18+ years
old; 2) able to understand the tasks (as judged by the interviewer); and 3) able to give
informed consent. The exclusion criteria were presence of an acute illness or cognitive
impairment that in the opinion of the interviewer would interfere with the study
requirements.
The EQ-VT software was an offline application used for face-to-face
interviewing. Mobile device or laptop was recommended to be used for collecting the
data. Distinguish roles and download privileges for principal investigator (PI), Data PI,
technical staffs, and interviewer were specified in the protocol. This software needed
interviewer to upload the data to the central server via internet regularly or every day.
The data can be downloaded by PI to check quality of data in order to give feedback to
interviewers in the mean times.
Regarding health states, the EQ-VT protocol recommended 86 health
states used to elicit preference for TTO valuation. They were divided into 10 blocks,
and each block contained 10 health states. The EQ-VT protocol also recommends 196
health states used for DCE valuation. They were grouped into 28 blocks. Each block
contained 7 pairs of health states. After register the respondent, the software will
randomly select TTO block and DCE block for interviewing.
Regarding health state design for the 5L valuation, an experiment design
was used to create the content and structure of the set of EQ-5D health states that is
presented to the respondents (74). Efficiency design was minimizing the number of
states and the number of respondents needed to get significant parameters estimates.
Orthogonality (i.e. attribute levels are independent), minimum overlap (i.e. minimum
overlap of levels for each attribute) and level balance (i.e. levels of each attribute
appear the same number of times) are all design optimization criteria that are used to
minimize the number of respondents that are needed.
For lead-time TTO, the design aimed to maximize the information about
parameters. In order to create an optimal experimental design for lead-time TTO, a
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 29
Federov algorithm was used. The optimization criterion was D-optimality, which
seeks to minimize the determinant of the inverse matrix X'X of the design. This
criterion results in maximizing the differential Shannon information content of the
parameter estimates.
For DCE, the "efficient design" algorithm used Monte Carlo simulation to
derive a choice set based on a prespecified utility model. The advantage of this
approach is that it is less restrictive in the optimization process. The parameter
estimates of the main effects DCE model from the 3L pilot were used as priors for the
DCE design algorithm in this study. The efficient design algorithm used to optimize
the DCE was D-optimality criterion as the Federov algorithm.
A multi-national pilot study, namely the 4C study, was an experiment study
for the valuation of EQ-5D-5L which included both lead-time TTO and DCE. It was first
conducted in England, Italy, the Netherlands and Canada. The number of respondents of
was 400 per country, each respondent answered 10 DCE questions and 10 TTO questions.
The design needed 10 blocked as a total of 100 health states need to be valued. The results
of this study demonstrated how the algorithm works and this methodology can be used to
inform on the model specifications. The refined design algorithm and model can
subsequently be used in the official EQ-5D-5L valuation study.
Regarding logical inconsistency, it occurred when a more severe health
state get higher value (75-77). Excluding those logical inconsistencies responses may
lead to more data quality input into the model, while at the same time; it may reduce
the representativeness of the value sets estimated. No gold standard method used to
consider the degree of logical inconsistency, Dolan and Kind (76, 77) offered counting
the number of pairwise of inconsistency. For example, if the values of these states
were ranked: 11111 < 11112 < 11113, then there were 3 pairs inconsistency. However,
in order to compare which health state should be valued more or less than another
state, not all pairs of health states can be considered. An eligible pairs should have 1
compared dimension and the 4 remained dimensions should be the same. For example:
state A 13221 and state B 12221, then state A is logically worse than state B because
the level of state A is equal or worse than state B. Given state C is 22132, it can’t be
compared with state A or B because some levels in state C are worse and some are
better.
Juntana Pattanaphesaj Literature Review / 30
2.5 Multilevel analysis
Multilevel analysis is a type of regression model that is particularly
suitable for multilevel data, which is the data with complex pattern of variability, or
longitudinal data. The multilevel analysis may be known in literature under a variety
of names, i.e. hierarchical linear model (HLM), Longitudinal data, also called repeated
measurements in medicine or panel data in the social sciences, arise when the
measures were repeated to the same individuals over time (78, 79). If a multi-stage
sampling design has been employed, multilevel statistical model are always needed.
The use of panel data has increased dramatically since it is possibility to control
unobservable individual specific effect. If unobserved variables exist in the regression
model, OLS model will give bias and inconsistent coefficients (80). Multilevel
analysis differs from the usual multiple regression models in the fact that the equation
defining the multilevel model contains more than one error term as follows (81).
Where;
= the value observed for micro-unit I within macro-unit j
= the residual effect for micro-unit I within macro-unit j
= intercept (group-dependent)
= slope (group-dependent)
The group-dependent coefficients can be split into an average coefficient
and the group-dependent deviation:
Substitution leads to the model
and are level-two residuals, while is level-one residuals. All 3
residuals have means zero, given the values of the explanatory variable X. Thus, is
the average regression coefficient and is the average intercept. The first part
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 31
( ) is called fixed effect part of the model. The second part
( ) is called the random part. The term can be regarded as a
random interaction between group and X. This model implies that the groups are
characterized by two random effects: intercept and slope.
Multilevel analysis allows the investigation of heterogeneity across units
both in the overall level of the response and in the development over time (78). This
violates the typical assumptions of ordinary regression models and must be
accommodated to avoid invalid inference. The main statistical model for multilevel
analysis is the hierarchical linear model or random coefficient model which allows
intercepts as well as slopes to vary randomly (81). Another type of multilevel analysis
is random intercept model, which intercepts are allowed to vary and assumes that
slopes are fixed. Consequently, for each individual observation, the scores on the
dependent variable are predicted by the intercept that varies across groups.
2.5.1 Assumption of the multilevel analysis
The assumptions of the multilevel regression model are similar to the
conventional multiple regression: linearity, homoscedasticity, and normal distribution of
the error terms. The testing coefficients can be valid if these assumptions are satisfied.
To examine the linearity between dependent and independent variables, a
scatterplot between all variables of a regression model can be used (79). However,
scatterplot often only show the functional form of a relationship for small sample size.
To deal with larger sample sizes, more information is needed to improve the scatterplot.
The median trace is a tool to make a scatterplot smoother. To construct a median trace,
the values on x-axis were divided into strips and the median of y for each strip is
calculated. Then the medians are connected with straight lines (Figure 2.7).
-1-.
50
.51
Val
ue
5 10 15 20 25Severity
-1-.
50
.51
5 10 15 20 25Severity
Value Median bands
Figure 2.7 scatterplot without median trace (left), and with median trace (right)
Juntana Pattanaphesaj Literature Review / 32
Homoscedasticity, which is equal to variances of errors, is an important
assumption for OLS (81). This means that the residuals do not depend on the
explanatory variables. However, heteroscedasticity is a frequently occurring
phenomenon in panel data analysis. The techniques used in the multilevel analysis
allow to relax this assumption and replace it by the weaker assumption that variance
depend linearity or quadratically on explanatory variables. Therefore, the
heteroscedasticity will also be prevalent in many multilevel data analyses. This is
indeed the case. In fact, heteroscedasticity is an explicit part of most multilevel
models. The likelihood ratio test (LR test) can be used to inspect heteroscedasticity.
The null hypothesis is equal variance for errors (homoscedasticity). In case
heteroscedasticity was found, robust standard errors should be employed in the model
estimation.
Inspection of normality of residuals should be employed to find outlying
cases that may have high influence on the results of the statistical analysis (81).
Normality of the error terms (residuals) is needed for valid hypothesis testing. This
assumption assures that the p-values for the t-tests and F-test will be valid. Examining
residuals is a key part of all statistical modeling. Residuals are estimating by
subtracting the observed responses from the predicted responses. The results give
information about the reasonable of the assumptions and the appropriateness of the
model. There are various methods used to test for residual normality, i.e. Shapiro-Wilk
W test, normal probability plot, kernel density plot.
Model specification error or model misspecification generally refers to
errors of omission variables; including an irrelevant variables; and incorrect functional
form (82). For hierarchical linear model, specification test is to select relevant
explanatory variables in the fixed part, and relevant random slope in the random part.
The purpose of model specification is to arrive at a model that describes the observed
data to a satisfactory extent but without unnecessary complications. The consequence
of specification error is incorrect results of model estimation or misleading. The
Hausman specification test is used to differentiate between fixed effects model and
random effects model in panel data (83). The null hypothesis is that the errors are not
correlated with the regressors. If it fails to reject null hypothesis, random effects model
are preferred.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 33
2.5.2 Sampling weights
Surveys are often non-representative of the population of interest. This is
because of oversampling of certain groups and different non response rates (84).
Sampling weights should be applied with the data set. The idea is that applying these
sampling weights in the analysis corrects for the non-representativeness of the data by
giving underrepresented groups more weight and overrepresented groups less weight.
The results from unweighted data are sometimes inconsistent and bias.
2.5.3 Model selection criteria
Model selection is the task of selecting a statistical model from a set of
candidate models. Given candidate models of similar predictive or explanatory power,
the simplest model is likely to be the best choice. According to Hendry and Richard
(85), a model chosen for empirical analysis should satisfy the criteria as follows.
1) Data coherency – the residuals estimated from the regression model
should be random. Otherwise, specification error will be found.
2) Valid conditioning – the explanatory variables should be uncorrelated
with the error term.
3) Parameter constancy – the values of the parameters should be stable.
Otherwise, the prediction will not be reliable.
4) Data admissibility – the prediction made from the model should be
logically possible.
5) Theory consistency – the composition of the model, i.e. intercept,
coefficient, should make good sense.
6) Encompassing – the model should include important variables so that it
is capable to explain their results
Several statistics can be used as diagnostic measures. These criteria were
discussed: 1) cross validation; 2) R2 and adjusted R2; 3) Akaike information criterion
(AIC); 4) Concordance correlation coefficient (CCC); 5) Bland-Altman plot; and 6)
standardized response mean (SRM).
Cross validation is a technique used to validate the model in order to
assess the generalizability of tested data set. It is mainly used when the goal of
Juntana Pattanaphesaj Literature Review / 34
modelling is prediction, and the researcher needs to estimate the accuracy of a
predictive model in practice. The dataset is split into two independent halves, one half
being used for generating a model, and the other half for testing of effects (81). The
advantage is that the testing data and model specification are separated. So the testing
does not lose their validity because of capitalization on chance.
R2 criterion is a measure of goodness of fit of a regression model (82). It
ranges from 0 to 1. The better fitted model yields higher value or close to 1.
Nevertheless, there are some disadvantage of R2. Firstly, it measures goodness of fit in
the way of how close a predicted Y value to its actual value in the sample. Thus no
guarantee that it will predict precise again for another sample. Secondly, the dependent
variable should be the same when compared two or more R2. Thirdly, when more
variables are added to the model, an R2 always increase. Therefore, just adding more
variables into the model, R2 increases, but it may increase the variance of prediction
error also. Adjusted R2 is a penalty for adding independent variables to increase the R2
value. For comparative purposes, adjusted R2 is a better measure than R2. However,
dependent variables must be the same for the comparison to be valid.
The Akaike’s Information Criterion (AIC) is used for choosing best
predictor subsets in regression and often used for comparing models, which ordinary
statistical tests cannot do. The value of the AIC for a given data set has no meaning.
AIC can be useful when comparing the fit of several models to the same data. The
model with the lower AIC is the better model (model is considered to be closer to the
truth). However, the problem with AIC is that it is difficult to interpret as it does not
have well defined endpoints related to a perfect fit or a lack of fit. The AIC is
calculated from the deviance (which is -2 times the log-likelihood) as: AIC = d + 2p,
where d is deviance and p is the number of estimated parameters (86).
Concordance correlation coefficient (CCC), a goodness-of-fit statistic, is
used to determine whether the observed data significantly deviate from the line of
perfect concordance (87). This coefficient is not only measuring how far each
observation deviates from the line fit to the data (precision), but also how far this line
deviates from the 45° line through the origin (accuracy). The CCC value ranges from -
1 to 1. The CCC equation is
Where;
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 35
ρc = the concordance correlation coefficient
ρ = the Pearson correlation coefficient, which measures how far
each observation deviates from the best-fit line, and is a measure
of precision, and
Cb = bias correction factor that measures how far the best-fit line
deviates from the 45° line through the origin, and is a measure of
accuracy.
Bland-Altman plot is a simple graphical method used to show prediction
bias and precision (88). It plots the difference of predicted score and actual mean score
on y-axis against the average of those score for each subject on x-axis. The 95% limits
of agreement are calculated by 1.96 x standard deviation. It is expected that the 95%
of dot (the difference between predicted score and actual mean score) should lie within
the limits of agreement.
The standardized response mean (SRM) is one of several available effect
size indices used to gauge the responsiveness of scales to clinical change (89). There
is no consensus on the method used to determine the magnitude of the difference
between two different scores. The SRM is obtained by dividing the mean score change
by the standard deviation of the change as follows.
SRM = (Meanx1 – Meanx2) / SDchange scores
The SRM can be interpreted as: 0 - 0.19 trivial effect; 0.20 - 0.59 small
effect; 0.60 - 1.19 moderate effect; 1.20 - 1.99 large effect; 2.0 - 3.99 very large effect;
and 4.0 - nearly perfect.
2.6 Discrete choice experiment and logistic regression
A discrete choice experiment (DCE) is a task which respondents have to
choose among a set of alternatives. These alternatives consist of attributes and severity
level selected from a descriptive profile. In general, discrete choice models are usually
derived in a random utility model (RUM) framework in which respondents are
Juntana Pattanaphesaj Literature Review / 36
assumed to be utility maximizers. Differently from the classic TTO, DCE just needs
respondents to choose preferred health state from 2 options (90). The health state
chose by the respondents was assumed that it gave them higher utility.
The values obtained from DCE valuation were not directly observed and
have to be calculated from the binary responses, the conditional logit model (clogit)
can be employed to estimate health value (90, 91). This model can be used to analyze
the binary outcome data with one or more predictors, where observations are not
independent but are matched or grouped in some way. The values obtained from the
conditional logit model were on an arbitrary scale which is different from the utility
scale where 0 refers to dead and 1 refers to perfect health. Thus the coefficient
generated from the conditional logit model cannot be used directly to calculate QALY.
The transformation or rescale is needed.
Previous studies (91) have successfully anchored DCE results on the utility
scale. The worst health state (health state 55555) predicted by the lead-time TTO
model was taken to anchor health state 55555 of DCE valuation in order to rescale the
arbitrary scale of the conditional logistic model. Thus, both TTO and DCE model
produced the same index value for the worst health state. Rescale was undertaken by
dividing all coefficients obtained from a conditional logistic model by a scalar which
was calculated as follows (91).
(worst health statedce – 1) / (worst health statetto – 1)
After rescale calculation, the utility decrements for each coefficient of
DCE model were obtained.
The logistic regression allows predicting a discrete outcome from a set of
variables that may be dichotomous, categorical, or continuous (86). It has no
assumptions about normality, linearity, and homogeneity of variance for the
independent variables. Logistic regression can be used to fit and compare models. The
goodness-of-fit tests help to choose the model that does the best job of prediction with
the fewest predictors. The conditional logistic regression is a particular analysis used
to analyze the binary outcome data with one or more predictors, where observations
are not independent but are matched or grouped in some way (92). The criteria for
selection of the model are similar to the topic 2.6.3 mentioned above.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 37
CHAPTER III
METHODOLOGY
Comply with the objectives; there are 3 sessions of methodology as follows.
Session 3.1: Development of the Thai population-based preference scores
for the 5L Thai version
Session 3.2: Testing the measurement properties of the Thai version of the
5L compared to the 3L
Session 3.3: Comparison of economic evaluation results using preference
score derived from the 3L and the 5L
Session 3.1: Development of the Thai population-based preference
scores for the 5L Thai version
3.1.1 Study design
This study was a cross-sectional survey, using the EuroQol Group’s
Valuation Technology (EQ-VT) software generated by the Value Set Working Group
(VSWG) of the EuroQol Group.
3.1.2 Study location
This study was conducted in 12 provinces of Thailand: Bangkok, Sing
Buri, Trat, Suphan Buri, Chiang Mai, Chiang Rai, Sukhothai, Surin, Nong Bua Lam
Phu, Roi Et, Krabi, Nakhon Si Thammarat. The data collection was conducted
between 27 August 2013 and 26 January 2014.
Juntana Pattanaphesaj Methodology / 38
3.1.3 Study population
The study population was general Thai population. The sample size was
1,207 respondents. A representative sample was randomly selected by a stratified
three‐stage sampling method with the collaboration from the National Statistical
Office (NSO), Thailand. Firstly, all 77 provinces in Thailand were stratified into 4
regions and Bangkok: North, Northeast, Central, South and Bangkok. The primary
sampling unit was province. Eleven provinces and Bangkok (total=12) were randomly
selected using systematic sampling. The secondary sampling unit was enumeration
areas (EAs). Each EAs may vary considerably in size of household and population, so
probability proportional to size (PPS) was applied. In our study, 120 EAs was selected
from the total of approximately 120,000 EAs in Thailand. The third stage sampling
unit was individuals aged 18 years or over. Ten participants per EA were selected
using quota sampling by age and sex according to the Thai population structure.
Respondents were identified and contacted by area coordinator prior to interview date,
and all of them agreed to be interviewed.
Replacement of the respondent can be done only in specific situation.
Protocol for replacement is explained as follows; 1) If the eligible participant insistently
refuses the interview, individual who is at the same age and sex identified from the same
EAs will be chosen for replacement; 2) If EAs cannot be reached because of unexpected
bad weather or poor road, the closet area will be chosen for replacement.
3.1.4 Selection criteria
Comply with the EQ-VT protocol, the eligible participants was general
population aged more than 18 years old. They should be able to read Thai and able to
understand the tasks (determined by interviewer). The person who presents an acute
illness, cognitive impairment, drunk, or disabled that in the opinion of the interviewer
would interfere with the study requirement was excluded.
3.1.5 Data collection method
The participants were interviewed in a face-to-face setting. The data was
collected using a touch screen laptop which was installed with the software named
EQ-VT offline version. The version used in this study was translated to Thai language
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 39
by translator; and proofed by Thai researcher team (Appendix B). This digital method
could results in more consistent valuation interviews and more systematic responses
within valuation studies and between valuation studies. Six interviewers were
recruited and well-trained prior to work.
The brief processes of data collection using computer device and EQ-VT
offline version are followed.
1) The interviewer logins the program by entering the user name and
password.
2) Each respondent is registered. Only the subject’s identification, first
letter of a name and surname of the respondent are inserted.
3) The interviewer explains all process of the interview to the respondent.
At the same time, the introduction screen is showed (Appendix B).
4) The questionnaires used in the following interview consist of 9 parts as
follows.
i. The Thai version of the EQ-5D-5L (Appendix D)
ii. Visual Analog Scale (EQ VAS) (Appendix D)
iii. Background questions (Appendix E)
iv. Preference elicitation by TTO technique
Regarding health state for TTO method included in the EQ-VT
software, there were 10 blocks of health state (Appendix F). Each
block was contained 10 health states and included 1 anchor state
(55555). The blocks used for interview were randomly selected by
the EQ-VT software. The respondent was asked to imagine two
alternative health states which were described on the screen. For
better than dead health state, two alternative health states were life A
= x years in full health; and life B 10 years in the valued health state.
Then the respondent was asked to reduce the number of year in life
A, until they are indifferent between 2 alternatives (Figure 3.1). For
worse than dead health state, lead time approach was used. The lead
time was 10 years in this study, the two alternative health states was
life A 10+x years in full health; and life B 10 years in full health,
then 10 years in the valued health state. Then the respondent was
Juntana Pattanaphesaj Methodology / 40
asked to reduce the number of year in life A, until they were
indifferent between 2 alternatives (Figure 3.2).
Figure 3.1 TTO screen for state better than dead
Figure 3.2 Lead time TTO screen for state worse than dead
มปญหาในการเดนเลกนอย มปญหาในการอาบน า หหรออใเเออาดยยนนเอเเลกนอย มปญหาในการท ากจกรรมทท าเปนประจ าปานกลาเ มอาการเจบปยดหรอออาการไมเ บายนยอยเาเมาก ไมเร กยนกกเยลหรออซมเศรา
มปญหาในการเดนอยเาเมาก อาบน า หหรออใเเออาดยยนนเอเไมเได ไมเมปญหาในการท ากจกรรมทท าเปนประจ า มอาการเจบปยดหรอออาการไมเ บายนยอยเาเมาก ร กยนกกเยลหรออซมเศราอยเาเมากท ด
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 41
v. TTO feedback
After completed TTO valuation, the respondents were asked
how they agreed with 3 sentences: 1) It was easy to understand the
questions I was asked; 2) I found it easy to tell the difference
between the lives I was asked to think about; and 3) I found it
difficult to decide on the exact points where Life A and Life B were
about the same. The options were 5-point Likert scale where 1 =
agree and 5 = disagree (Appendix G).
vi. Preference elicitation by discrete choice experiment
There was 28 blocks of health state for DCE method
(Appendix H). Each block contained 7 pairs of health states; life A
and life B. Both life A and B consisted of 5 health state dimensions
with different level of severity. EQ-VT software randomly selected 1
block for each participant. The respondent was asked to imagine
both life A and B described on screen; and make a forced choice
from two alternative health states (Figure 3.3).
Figure 3.3 DCE screen
vii. DCE feedback
After completed DCE valuation, the respondents were asked
how they agreed with 3 sentences: 1) It was easy to understand the
มปญหาในการเดนอยเาเมาก
อาบน า หหรออใเเออาดยยนนเอเไมเได
ไมเมปญหาในการท ากจกรรมทท าเปนประจ า
มอาการเจบปยดหรอออาการไมเ บายนยอยเาเมาก
ร กยนกกเยลหรออซมเศราอยเาเมากท ด
มปญหาในการเดนเลกนอย
มปญหาในการอาบน า หหรออใเเออาดยยนนเอเเลกนอย
มปญหาในการท ากจกรรมทท าเปนประจ าปานกลาเ
มอาการเจบปยดหรอออาการไมเ บายนยอยเาเมาก
ไมเร กยนกกเยลหรออซมเศรา
อะไรดกยเากน หชยนแบบ หA หรออชยนแบบ หB
A B
Juntana Pattanaphesaj Methodology / 42
questions I was asked; 2) I found it easy to tell the difference
between the health states I was asked to think about; and 3) I found it
difficult to decide on my answers to the questions. The options were
5-point Likert scale where 1 = agree and 5 = disagree (Appendix I).
viii. Country-specific questionnaire
This paper-based questionnaire was created by Thai research
team. It contained 5 sentences regarding religious belief as follows.
1) Everything that happens in my life is a consequence of my actions
from my previous life.
2) When I am sick, there is someone looks after me.
3) Regardless of any serious sickness, I try to live as long as possible
to have something done.
4) According to my belief, to escape the problems by terminating
my own life is seriously wrong.
5) I hold the religious doctrine when I face the problems in my life.
The responses of country-specific questionnaire consisted of 5-
point Likert scales ranging from completely agree to completely
disagree (Appendix J).
ix. Qualitative questionnaire
This paper-based questionnaire was created by Thai research
team. It contained 3 questions (Appendix K); and was applied to
only some respondents who answered TTO questions with strange
reasons
5) At the end of interview, a screen expressed gratitude to the respondent
for the participation in the study. Respondents can give feedback by entering text in
the textbox of EQ-VT software.
6) The records were uploaded to the central data collection system of the
EuroQol group by the interviewers every day.
The respondents may control the device themselves. Nevertheless, Thai
people are not familiar with using computer device, some people may still unable to
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 43
use it after demonstration. In this case, the interviewer helped or controlled the device
instead.
Distinguish roles and download privileges for principal investigator (PI),
Data PI, and technical staffs were also specified in the protocol. Prior to data collection
begin, PI and co-investigators joined the 2-day EQ-VT training workshop offered by the
EuroQol group. Then, the interviewer was trained to fully acquaint with the operation
(installation of the EQ-VT offline, registration of respondents, data collection, and
uploading of data) and the central idea behind computer assisted application.
3.1.6 Study procedure
The procedures of the study were as follows.
1. Randomly selection of the area by the NSO
2. Coordination with study area’s coordinators.
3. Localized software development by EuroQol group and Thai team
4. Acquisition of the touch screen laptop
5. Translated interviewer instruction into Thai by EuroQol group and Thai
team
6. Recruiting and training 6 interviewers
7. Pilot testing (100 respondents in 8 primary care unit in Nonthaburi)
8. Data collection
9. Analysis and report
3.1.7 Data analysis for TTO valuation
3.1.7.1 Utility calculation
In order to calculate utility from TTO method, 2 different
methods were employed: method for better than dead (BTD) states and worse than
dead (WTD) states. Regarding BTD state, the utility (Ui) was calculated from x
divided by 10 (Ui = x/10) (48, 49), where x is number of year in full health of Life A
at the point of indifference. Regarding WTD state, the utility was calculated using lead
time approach which 10 years of full health were added to both alternative. The utility
was calculated from Ui = (x-10) / (10+t-10), when x is number of year in full health of
Life A at the point of indifference (51).
Juntana Pattanaphesaj Methodology / 44
Figure 3.4 shows the example of utility calculation. The upper
picture shows the BTD state which the indifference point is 4 years for life A. Thus
the utility was 4/10 = 0.4. The lower picture shows the WTD state which the
indifference point is 8 years for life A. Thus the utility was (8-10)/(10+10-10) = -0.2.
Life A 4
Life B
Life A 8
Life B 10
Figure 3.4 Example of utility calculation for TTO valuation (51)
Dark cell = full health, diagonal cell = health state i, white cell = death
3.1.7.2 Data management for TTO data
Prior to input data into the regression model, the 7 criteria were
applied to detect low quality of the data as follows.
1) logical inconsistency
2) positive slope
3) all 10 health states got the same value
4) too many health states (> 8 health states) were valued as
WTD
5) too many health states (>8 health states) were valued as zero
6) very mild states (severity level = 6 or 7) were given very
low value (value < 0)
7) magnitude of inconsistency
Health state i = 10 yr
Ui = 4/10 = 0.4
10 yr
Ui = (8-10)/(10+10-10) = -0.2 20 yr
Health state i = 10 yr
Better than dead health state
Worse than dead health state
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 45
In our study, severity level of health state was calculated as the
summation of 5 numbers representing each dimension. For example, the severity level
of health state ‘21232’= 2+1+2+3+2 = 10. Thus the lowest possible severity level was
5 (11111) and the highest possible severity level was 25 (55555).
The explanation of 7 criteria was described as follows. The
logical inconsistency exists when the TTO score of health state 55555 (severity level = 25)
is lower than that of the mildest health state within TTO block, as shown in Figure 3.5.
The positive slope means that the line (10 dots/respondent)
plotted between TTO score (y-axis) and severity level of health state (x-axis) has
positive slope. This means that the respondent gave higher TTO score when the
severity level of health states increased.
When the respondent gave the same values for all 10 health
states, this kind of responses was unreasonable as there was a big difference between
the mildest health state and the worst health state (severity level = 25). Then, such
records should be excluded from the analysis.
The data which many health states (> 8 health states) were
valued as zero or below (WTD) was excluded also. The reason was that each block (10
health states) contained 1-3 milder states (including only a level 2 on 1-2 dimensions);
1-6 moderate states; and 3-7 severe states (including at least one of level 5 and one of
level 4 or 5) (93). Thus it was possible that the respondents values all severe states as
zero or WTD. However, the records which 8 or more health states were valued zero or
less than zero were excluded.
Juntana Pattanaphesaj Methodology / 46
1. Logical inconsistency
2. Positive slope
3. All 10 states got the same value
4. Too many states were valued as WTD
5. Too many states were valued as zero
6. Very mild states got very low value
Figure 3.5 Six criteria used to detect low quality of the data for each respondent
After excluding respondents, whose at least 1 criteria were
met. The magnitude of inconsistency was calculated for each respondent. The
inconsistent response was defined as higher TTO score was assigned to a worse health
state (higher severity level). Table 3.1 demonstrates the calculation for the magnitude
of inconsistency of 1 respondent. The values in the table were calculated as ‘the TTO
score of milder state’ minus ‘the TTO score of more severe state’. The positive
number indicated consistent response while the negative number indicates the
inconsistency. As shown in Table 3.1, there were 3 inconsistency responses, i.e. 12513
Severity level
Severity level Severity level
Severity level Severity level
Severity level Severity level
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 47
vs 12344, 53221 vs 12344, and 14554 vs 44345, resulting in the magnitude of
inconsistency of 0.1+0.1+0.1 = 0.3.
Table 3.1 Magnitude of inconsistency responses of 1 respondent
Profile 11121 21112 12513 53221 12344 44125 54342 14554 44345 55555
Severity 6 7 12 13 14 16 18 19 20 25
Observed TTO score 1 1 0.9 0.9 1 0.8 0.8 0.7 0.8 0.4
11121 6 1
21112 7 1 0
12513 12 0.9 0.1 0.1
53221 13 0.9 0.1 0.1 0
12344 14 1 0 0 -0.1 -0.1 44125 16 0.8 0.2 0.2 0.1 0.1 0.2
54342 18 0.8 0.2 0.2 0.1 0.1 0.2 0
14554 19 0.7 0.3 0.3 0.2 0.2 0.3 0.1 0.1
44345 20 0.8 0.2 0.2 0.1 0.1 0.2 0 0 -0.1 55555 25 0.4 0.6 0.6 0.5 0.5 0.6 0.4 0.4 0.3 0.4
Since the appropriate cut-off point of magnitude of
inconsistence used to determine low quality data was unknown, this study divided
samples into subgroup according to the magnitude of inconsistent responses. Then
these subgroups were thoroughly compared using regression diagnostics. Finally, only
1 subgroup was chosen to estimate the Thai value sets.
3.1.7.3 Data diagnostic tests
Data diagnostic tests were used to check for potential problems
and evaluating the plausibility of key assumptions of multilevel analysis (81): 1)
linearity between dependent and independent variables; 2) homoscedasticity; and 3)
normal distributions of the residuals.
The linearity between dependent and independent variables
was demonstrated by the scatter plot between TTO value (Y-axis) and level of severity
(X- axis) of all respondents. However, the relation between y and x will not be seen
clearly when the sample size was large. The median trace was a crude way to show the
tendency in the relationship between y and x clearer (79). The median trace is a tool to
make a scatterplot smoother. To construct a median trace, the values on x-axis were
Juntana Pattanaphesaj Methodology / 48
divided into strips and the median of y for each strip is calculated. Then the medians
are connected with straight lines.
Heteroscedasticity was inspected using the likelihood ratio test
(LR test). The normality of residuals was tested using Shapiro-Wilk test, Kernel
density estimate, and probability-probability (P-P) plot.
The specification test was performed by the Hausman test,
which was used to determine whether fixed effect or random effect models should be
used. In addition, a spaghetti plot, which is a method of viewing pattern of response of
each individual data, was also considered. It was plotted between severity level
(x-axis) and predicted utility (y-axis). Since our data was large, only 10 respondents
were selected to demonstrate the pattern of relationship.
3.1.7.4 Regression modeling
Multilevel regression models (i.e. random coefficient model)
were undertaken to estimate preference scores using STATA 12 as the dataset was
longitudinal or panel data (81). The dependent variable was disutility (i.e. 1 – TTO
score). The independent variables were 5 dimensions of EQ-5D which each dimension
contained 5 options. So 20 dummy variables were produced (Table 3.2), and they were
consisted in the main effect model. Our data can be a representative of Thai population
as weighted values were attached to each individual in the regression model.
Table 3.2 Variables in the TTO model
Variable Definition
MO2 1 if mobility is at level 2, 0 otherwise MO3 1 if mobility is at level 3, 0 otherwise MO4 1 if mobility is at level 4, 0 otherwise MO5 1 if mobility is at level 5, 0 otherwise SC2 1 if self-care is at level 2, 0 otherwise SC3 1 if self-care is at level 3, 0 otherwise SC4 1 if self-care is at level 4, 0 otherwise SC5 1 if self-care is at level 5, 0 otherwise UA2 1 if usual activity is at level 2, 0 otherwise UA3 1 if usual activity is at level 3, 0 otherwise UA4 1 if usual activity is at level 4, 0 otherwise UA5 1 if usual activity is at level 5, 0 otherwise
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 49
Table 3.2 Variables in the TTO model (cont.)
Variable Definition
PD2 1 if pain/discomfort is at level 2, 0 otherwise PD3 1 if pain/discomfort is at level 3, 0 otherwise PD4 1 if pain/discomfort is at level 4, 0 otherwise PD5 1 if pain/discomfort is at level 5, 0 otherwise AD2 1 if pain/discomfort is at level 2, 0 otherwise AD3 1 if pain/discomfort is at level 3, 0 otherwise AD4 1 if pain/discomfort is at level 4, 0 otherwise AD5 1 if pain/discomfort is at level 5, 0 otherwise
The various functional forms were employed to the data sets as
shown in Table 3.3. The interaction terms tested were N45, N5, and D123^2. N5 and
N45 were similar to the N3 term purposed by Dolan (94). The term N5, which was a
dichotomous variable, was defined as whether extreme problems (level 5) in any
domain exist. N5 = 1 if there were at least 1 dimension with level 5. N5 = 0 if no level
5 in the profile. The term N45 was defined as whether severe or extreme problems
(level 4 or 5) in any domain exist. N45 = 1 if there were at least 1 dimension with level
4 or 5. N45 = 0 if no level 4 or 5 in the profile. The term D123^2 was defined as the
number of dimensions at level 4 or 5 minus one and then squared. For instance, the
health state 34435 has 3 dimensions with level 4 or 5. Thus D123^2 is calculated as
(3-1)^2 = 4. Also the interactions between levels and different dimensions were tested,
i.e. MO3*UA5.
Table 3.3 Functional forms
Model Functional form
Model 1 Constant + main effect
Model 2 Main effect
Model 3 Constant + main effect + N45
Model 4 Main effect + N45
Model 5 Constant + main effect + N5
Model 6 Main effect + N5
Model 7 Constant + main effect + D123^2
Model 8 Main effect + D123^2
Juntana Pattanaphesaj Methodology / 50
3.1.7.5 Evaluating the model performance
In order to select the best model for predicting the EQ-5D-5L
score for Thai people, four criteria were considered, i.e. consistency, predictive
performance, responsiveness and parsimony (85). Consistency means that more severe
problems associate with utility decrement (75, 95). Regarding consistency in the
regression model, the coefficient value of health dimension should associates with the
severity level of health dimension. Thus the inconsistence in the regression model
existed when the coefficients is lower when the severity level of health dimension
increases.
The predictive performance was evaluated using multiple
measures. Cross validation was employed by dividing the data set into 2 parts. Two-
third of data was used to generate a regression model, and one-third of data was kept as
testing data. The actual mean score from testing data was compared to the predicted
score by health profile. The number of health state was counted when the absolute
difference between the predicted score and mean actual score greater than 0.1 and 0.15.
The Akaike’s Information Criterion (AIC) was used for
comparing the fit of several models for the same data. The model with the lowest AIC
is likely to be the best model. This means that the model is considered to be closer to
the truth. Concordance correlation coefficient (CCC), was used to determine whether
the observed data significantly deviate from the line of perfect concordance (87). This
coefficient is not only measuring how far each observation deviates from the line fit to
the data (precision), but also how far this line deviates from the 45° line through the
origin (accuracy). The CCC value ranges from -1 to 1.
Bland-Altman plot was a graphical method used to show
prediction bias and precision (88). It plotted the difference of predicted score and
actual mean score on y-axis against the average of those score for each subject on x-axis.
The 95% limit of agreement was calculated as 1.96 x standard deviation. It is expected
that the 95% of dot (the difference between predicted score and actual mean score)
was within limits of agreement.
The standardized response mean (SRM) was used as an effect
size indices or reflected the responsiveness of health status change. The SRM was
obtained by dividing the mean score change by the standard deviation of the change.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 51
Lastly, if two or more multiple regression models performed similarly, the simplest
model (model parsimony) or the model with fewer explanation variables was
preferred.
3.1.7.6 Detecting logical inconsistency among 3,125 health
states
To be confident that the predicted scores for 3,125 health states
are logical consistent according to Dolan and Kind method (77), this study employed a
simple Excel program to investigate. All 3,125 health states with its preference score
were arranged into 2 columns and sorted by number in sheet1 (Figure 3.6). In sheet2,
all 3,125 health states were arranged into 5 columns (i.e.column 1-5) and 625 rows.
These 5 columns represented 5 levels of mobility dimension, i.e. column 1 consisted
health profiles which the mobility was level 1, column 2 consisted health profiles
which the mobility was level 2, and so on. Each column, the health profiles was sorted
by number. The command vlookup was employed to link the matched preference
score from sheet1 to sheet2 into the next 5 column (i.e.column 6-10). Then the next 25
columns (i.e.column 11-35) were the results of subtracting between preference score
of less severe profile and more severe profile for only eligible health state pairs
according to Dolan and Kind method (77). The negative figures indicated inconsistent
preference scores which were counted using the command countif.
Juntana Pattanaphesaj Methodology / 52
sheet1
Sheet2
Sheet2 (cont.)
Figure 3.6 Example of the method used to detect logical inconsistency
among 3,125 health states
Vlookup command was used to link preference score of matched profile from sheet1 to sheet2.
Cell M2 = G2-H2 Cell N2 = H2-I2 Cell O2 = I2-J2 …
Cell AI2 = G2-G3 Cell AJ2 = G3-G4 Cell AK2 = G4-G5 …
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 53
3.1.8 Data analysis for DCE valuation
Differently from TTO valuation, the values obtained from DCE valuation
were not directly observed and also have to be calculated from the binary responses.
The health state chosen by the respondents was assumed that it gave them higher
utility, so the conditional logit model can be employed to estimate health value (90,
91). The scale obtained from the conditional logit model is not the same as utility scale
where 0 refers to dead and 1 refers to perfect health. Thus the coefficient generated
from the conditional logit model cannot be used directly to calculate QALY. The
transformation or rescale is needed.
The dependent variable (Y) referred to the choice of respondents which
were rescaled to indicate that 0 stands for ‘not chosen health state’ and that 1 for ‘chosen
health state’. The independent variables consisted of 20 dummy variables. The value of
health state A of participant i (ViA) was explained by the following additive model (91).
Where XiAj are 20 dummy variables for participant i and health state A
(with 5 dimensions, each with 5 levels of severity). βj are the coefficients for each
independent variable j. Then the predictive model was consisted of ViA plus and error
term which was assumed to be random and show type 1 extreme value distribution.
Thus a conditional logistic model (clogit) can be applied (90, 91). The conditional
logistic model can be used to analyze the binary outcome data with one or more
predictors, where observations are not independent but are matched or grouped in
some way.
Consequently, the coefficients of the model were estimated. However the
values generated were on an arbitrary scale, they were needed to be rescaled. The
worst health state (health state 55555) predicted on the lead-time TTO model was
taken to anchor health state 55555 of DCE valuation in order to rescale the arbitrary
scale of the conditional logistic model. Thus, both TTO and DCE model of this study
produced the same index value for the worst health state. Rescale was undertaken by
dividing all coefficients obtained from a conditional logistic model by a scalar, which
was calculated as follows (91).
(worst health statedce – 1) / (worst health statetto – 1)
Juntana Pattanaphesaj Methodology / 54
After rescale calculation, the utility decrements for each coefficient of
DCE model were obtained.
With regard to data management of DCE data, the previous study (90)
proposed 4 criteria to consider low quality data as follows.
1) All responses (each respondent valued 7 pairs health state) were on the
same side e.g. AAAAAAA, BBBBBBB.
2) The pattern of responses had particular order e.g. ABABABA,
BABABAB.
3) The respondents spent too little time choosing preferred health state
(defined as < 10 second/pair)
4) More severe health state was preferred. This criteria was defined if the
preferred health state was more severe than rejected health (i.e level of severity of the
preferred health state was at least 8 level higher than rejected health state) and at least
4 dimensions of the preferred health state were more severe than the rejected health
state (Table 3.4)
Table 3.4 Example of more severe health state was preferred
Patient ID
Health state
Level of severity Explanation
1 A 35554 22 If the respondent preferred health state A, this record was not removed because it met only 1 criterion. Health state A was 8 level more severe than health state B, however, only 3 dimensions of health state A were more severe than health state B.
B 55211 14
1 A 12151 10 If the respondent preferred health state B, this record was removed because it met both criteria. Health state B was 10 levels more severe than health state A; and 4 dimensions of health state B were more severe than health state A.
B 35543 20
However, the model from both cleaned data set and original data set were
generated and compared.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 55
Session 3.2: Measurement properties of the Thai version of the 5L
compared to the 3L
3.2.1 Study design
This study was a cross-sectional survey.
3.2.2 Study location
The study was conducted at Ramathibodi Hospital, Bangkok between 7
January and 31 March 2013.
3.2.3 Study population
Participants in this study were those recruited in the study entitled
“Economic evaluation of self-monitoring of blood glucose (SMBG) intervention”. A
convenience sample of patients with type 1 and type 2 diabetes mellitus - who
received treatment at the outpatient department at Ramathibodi Hospital, Thailand,
was invited to participate. According to Terwee et al (52), the sample size was at least
50 respondents for subgroup.
The reasons for selection diabetic patient to test measurement properties of
the instrument were that the diabetic mellitus is a common chronic disease that
substantial affect quality of life of the patients. Additionally, diabetes was ranked as
third and eighth in terms of Disability-Adjusted Life Year (DALY) loss in Thai
women and men, respectively (96). The study by Holmes et al (97) demonstrated that
the quality of life of the people with diabetes was lower than general people in the
same age group. For these patients, their quality of life decreased with disease
progression and complications.
3.2.4 Selection criteria
For type 1 diabetic patient, eligibility criteria were 1) aged > 12 years old;
2) use insulin; 3) diagnosed with type 1 DM; and 4) consent to be interviewed.
For type 2 diabetic patient, eligibility crieteria were 1) aged > 18 years old;
2) use insulin; 3) diagnosed with type 2 DM; and 4) consent to be interviewed.
Pregnant women and disabled person were also excluded from the study.
Juntana Pattanaphesaj Methodology / 56
3.2.5 Data collection method
The questionnaires used in the interview consisted of 4 parts as follows:
1) Demographic information (Appendix L)
2) One page of Thai version of the 3L and 5L; Visual Analog Scale (EQ
VAS) (Appendix L)
3) Short-form 36 health survey version 2 (SF-36v2) in Thai (Appendix M)
4) Preference question (Appendix L)
The single page of the 3L and the 5L response scale contained the 5L
version on the left column and the 3L version on the right column, as shown in
Appendix L. Similar to previous studies (21, 24, 26), respondents were asked to
complete the 5L first, followed by the 3L in order to avoid the tendency to not choose
levels 2 and 4 - the “in-between” options - when the 3L was completed first.
Preference questions comprised 2 items, 1) Which response scale did you find easier
to use (3L or 5L or indifferent? and 2) Which response scale did you think better
expresses your health? (21) The interviewer explained the instruction and let the
respondents completed the questionnaire by themselves. After that the interviewers
gave the second set of questionnaires, which consisted of the Thai version of one page
of Thai version of the 3L and 5L and SF-36v2 in a pre-paid envelop to the
respondents. Each respondent was asked to complete the second set of questionnaire at
2 weeks later, and send it back to the researcher.
This study was approved by the Mahidol University Institutional Review
Board (MU-IRB), Thailand and the Institute for the Development of Human Research
Protections (IHRP), Ministry of Public Health, Thailand (Appendix A). All
participants provided written informed consent and all instruments were self-
administered. After completing the questionnaire, the respondents received 3.25 USD
for compensation (1 USD = 30.73 Baht).
3.2.6 Data analysis
The utility index of the 5L was obtained from an interim mapping
generated by the EuroQol group (34) as the valuation study of the 5L in Thailand has
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 57
not yet been completed. The 3L utility index was calculated using the Thai value sets
developed by Tongsiri et al (12).
3.2.6.1 Distribution pattern and ceiling effect
The distribution of the 3L and 5L responses was demonstrated
in terms of percentage of each level reported. The redistribution patterns of the
responses from the 3L to 5L for each dimension were also reported in terms of
percentage. Similar to previous studies (21, 27), the response inconsistency and size
were determined and are shown in Table 3.5 To determine the inconsistency, the
response of the 3L was converted into the 5L (the 3L5L) as follows: 1 = 1, 2 = 3, and 3
= 5. Then, the size of inconsistency was calculated as |3L5L-5L|-1. A size of
inconsistency of ≤ 0 indicated consistency, and thus only 7 pairs are considered as
consistent responses.
Table 3.5 Size of (in) consistent response
3L 5L
level 1 level 2 level 3 level 4 level 5
level 1 -1 0 1 2 3
level 2 1 0 -1 0 1
level 3 3 2 1 0 -1
Adapted from Janssen et al (21). The size of inconsistency of < 0 indicated consistency. The number in dark cells represents the size of inconsistency.
For the ceiling effect, the proportion of respondents reported
‘no problems’ for all five dimensions - the proportion of respondents scoring ‘11111’
(22) - was compared for the 3L and the 5L. The percentage reduction from the 3L to
the 5L was calculated as follows: (Ceiling 3L – Ceiling 5L)/ Ceiling 3L. We
hypothesized that the ceiling effect should be lower in the 5L compared with the 3L.
Feasibility was assessed by calculating the number of missing values for the 3L and 5L.
3.2.6.2 Convergent validity
The convergent validity of the 5L and 3L were evaluated by
comparing them with the SF-36 as it is a widely-used generic health survey in clinical
Juntana Pattanaphesaj Methodology / 58
research and has demonstrated validity among the Thai population (61, 67, 68). The
SF-36 contains 8 dimensions, i.e. physical functioning; role limitation due to physical
problems; bodily pain; general health perceptions; social functioning; vitality; role
limitations due to emotional problems; and general mental health (98). Since a
weighted Likert scale is used as the scoring system, the items for each dimension are
summed to provide a score which is then linearly transformed into a value from 0 – 100
(100 indicating the best health level).
Convergent validity was tested by assessing the relationship
between each dimension of the 5L and SF-36v2 using Spearman’s rank-order
correlation (Spearman’s rho). We hypothesized that each dimension in the 5L would
be more highly correlated to related subscales than to other subscales in the SF-36
compared to the 3L. The EQ-5D’s responses were recoded to signify that higher scores
presented better health statuses. The strength of correlation was determined as follows:
absent (r < 0.20), weak association (0.2≤ r< 0.35), moderate (0.35 ≤ r<0.50), and
strong (r≥0.50) (55).
3.2.6.3 Discriminative power
Discriminative power (or informativity) was determined by the
Shannon index ( ) and Shannon’s Evenness index ( ). These indices were initially
developed from the information theory (56) and are typically used to measure the
diversity and richness of information in ecosystems and the communications industry.
At present, and are often used to reflect the discriminatory power of health state
classification (21, 22, 24, 27, 56). High and values reflect discriminative power
and informativity of the instrument. The Shannon index ( ) is defined as (56):
Where;
= absolute amount of informativity captured
C = the number of possible categories (levels)
Pi = ni/N (the proportion of observations in the ith category (i = 1,...,C)
ni = the observed number of scores (responses) in category i
N = the total sample size
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 59
The Shannon index ( ) reflects the absolute information
content. The higher the , the more information is captured by the measure. The
also depends on the number of categories; if the number of levels (C) that are
actually needed increases, the index will also increase. However, if the newly added
levels are not used, the value will not increase. The index will reach its maximum
( max) which is equal to Log 2C (1.58 for 3L and 2.32 for 5L) when the optimal
amount of information is captured.
Shannon’s Evenness index ( ), on the other hand, expresses
the relative informativity of a system or the evenness of a distribution regardless of the
number of categories. is defined as:
In case of an even distribution, when all levels are filled with the
same frequency, is equal to 1. When comparing the 5L to the 3L, we expect the of
the 5L to be higher to reflect more discriminatory performance. On the other hand, the
of the 5L might slightly decrease as the extra level might not be used equally.
3.2.6.4 Test-retest reliability
The test-retest reliability of both EQ-5D index scores was
evaluated using the intraclass correlation coefficient (ICC) and the reliability of each
dimension was assessed with Cohen’s weighted kappa coefficient. According to
Fleiss’s standards for the strength of agreement for kappa values (53), Cohen’s
weighted kappa (k) was determined as follows: poor reproducibility (k < 0.4); good
reproducibility (0.4 < k < 0.75 ; excellent reproducibility (k > 0.75). Regarding intra-
rater reliability among each dimension at different times, the data set lacked variance
since most respondents responded with level 1 for self-care. The weighted kappa
coefficient could not be calculated, thus percentage agreement values was
demonstrated also (99, 100). It was calculated as: (a+d)/N, where the values of a and
d were obtained from a 2x2 table.
Juntana Pattanaphesaj Methodology / 60
3.2.6.5 Data variability
Coefficient of variation (CV) was used to compare data
variability between the 3L and the 5L. It is obtained by the ratio of the standard
deviation to the mean; and multiplied with 100 to show in a percentage. Higher value
of CV indicates greater variability of data. The advantage is that CV allows
comparison between measures as the unit was cancelled, while standard deviation is
often difficult to interpret or compared with other measures as its value based on the
sample data. CV of this study was calculated from all respondents except for the
respondents who reported perfect health.
All data in this session were analyzed using SPSS 19 and Microsoft Excel
2013. Statistical significance was set a priori as p < 0.05.
Session 3.3: Comparison of economic evaluation results using
preference score derived from the 3L and the 5L
In session 3.2, each participant answered both the 3L and the 5L. The 3L
utility was calculated using Thai value sets, which was estimated by classic TTO
technique (11, 12). The 5L utility was obtained from the value sets generated by
mapping method by EuroQol group (34); and also the Thai value sets obtained from
session 1. Utilities obtained from these 3 value sets were used to calculate QALY
(QALY = utility x life expectancy). Then, the QALY was inputting into the economic
evaluation model entitled “Economic evaluation of self-monitoring of blood glucose
(SMBG) intervention”. The incremental cost-effectiveness ratio (ICER) was used to
demonstrate the results of using 3 different value set. Given all parameters were
constant except for utility, ICER could be affected from QALY, which obtained from
different methods. The ICER was calculated as follows:
ICER = (C1-C2)/(Q1-Q2)
Where;
C1 = cost of reference intervention
C2 = cost of compared intervention
Q1 = QALY of reference intervention
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 61
Q2 = QALY of compared intervention
In order to cope with uncertainty of the parameters in the model, Monte
Carlo simulation was undertaken (101). It was done by input random parameters into
the model 1,500 times repeatedly. The results was plot on the cost-effectiveness plane
which the horizontal axis represented the incremental QALY and the vertical axis
represented incremental cost between the reference and compared interventions.
Uncertainty of the model was reflected by examining the distribution of those 1,500
dots. In addition, the cost-effectiveness acceptability curve (CEAC) was used to
summarize the information on uncertainty in cost-effectiveness analysis (35). It was
drawn by plotting the proportion of the costs and QALY pairs that were cost-effective
for the maximum acceptable ceiling ratio. The CEAC shows the probability that the
intervention is worth for a given value of the maximum acceptable ceiling ratio.
Head-to-head comparison of preference score of comparable health states
between the 3L and the 5L was also conducted. To identify comparable health states,
the level of 3L was converted to level of 5L as follows : level 23L to level 35L, level 33L
to level 55L and level13L to level 15L . For example, state 12323 of 3L was comparable
to state 13535 of the 5L.
Juntana Pattanaphesaj Results / 62
CHAPTER IV
RESULTS
Session 4.1: Development of the Thai population-based preference
scores for the 5L Thai version
4.1.1 Respondent’s characteristics
Table 4.1 presents the characteristic of 1,207 respondents interviewed. The
mean age of the respondents was 44 years old and about half of them (52%) were
female. A majority of respondent has been married (68%). Most of the respondent
graduated from primary school (44%), and high school (38%), respectively. Their
occupations were agriculture/fishery (35%), unskilled labor (19%), and business
owner (16%), respectively. The average household income of the respondents was
22,640 Baht per month. Table 4.2 shows that the percentage of the respondent from
the rural was slightly higher than those from urban (56% vs 44%).
Table 4.1 Demographic characteristic of respondents (n=1,207)
Demographic characteristic Mean (SD) Age (years) 43.55 (15.03) Household income (Baht / month) 22,640.36 (26,765.11) Number of child 1.75 (1.57)
n (%) Gender
Male 584 (48.38) Female 623 (51.62)
Marital status Married 817 (67.69) Single 230 (19.06) Widowed 88 (7.29) Divorced/Separated 72 (5.97)
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 63
Table 4.1 Demographic characteristic of respondents (n=1,207) (cont.)
Demographic characteristic n (%) Education
Primary school 526 (43.65) High school 458 (38.01) Bachelor’s degree 126 (10.46) Diploma 75 (6.22) Unlettered 16 (1.33) Master’s degree or higher 6 (0.50)
Occupation Agriculture/fishery 426 (35.29) Unskilled labor 227 (18.81) Business owner 193 (15.99) Housewife 128 (10.60) Student 65 (5.39) Government/state enterprise officer 45 (3.73) Employee 35 (2.90) Looking for a job 17 (1.41) Retired 16 (1.33) Unable to work due to sickness 7 (0.58) Other 48 (3.98)
Table 4.2 The number of respondents by age, gender, and residential area
Region / Province Age (SD) Gender (n (%)) Residential area (n (%))
Male Female Urban Rural
Bangkok 43.97 (15.49) 81 (48.21%) 87 (51.79%) 168 (100.00%) -
Central region Suphan Buri 43.90 (15.09) 109 (49.32%) 112 (50.68%) 70 (31.67%) 151 (68.33%)
Sing Buri 44.05 (15.57) 29 (47.54%) 32 (52.46%) 20 (32.79%) 41 (67.21%)
Sukhothai 43.38 (13.88) 13 (54.17%) 11 (45.83%) 9 (37.50%) 15 (62.50%)
Trat 42.80 (14.79) 28 (47.46%) 31 (52.54%) 20 (33.90%) 39 (66.10%)
North region Chiang Mai 42.80 (15.40) 60 (48.78%) 63 (51.22%) 80 (65.04%) 43 (34.96%)
Chiang Rai 43.25 (13.05) 32 (46.38%) 37 (53.62%) 29 (42.03%) 40 (57.97%)
Northeast region Nong Bua Lam Phu 43.48 (15.77) 23 (46.00%) 27 (54.00%) 20 (40.00%) 30 (60.00%)
Roi Et 43.43 (15.90) 72 (50.35%) 71 (49.65%) 50 (34.97%) 93 (65.03%)
Surin 43.87 (14.88) 66 (47.83%) 72 (52.17%) 20 (14.49%) 118 (85.51%)
South region Nakhon Si Thammarat 42.99 (14.48) 56 (46.28%) 65 (53.72%) 30 (24.79%) 91 (75.21%)
Krabi 44.40 (14.93) 15 (50.00%) 15 (50.00%) 10 (33.33%) 20 (66.67%)
Total 43.55 (15.03) 584 (48.38%) 623 (51.62%) 526 (43.58%) 681 (56.42%)
Juntana Pattanaphesaj Results / 64
Regarding health status of the respondents measured by the EQ-5D-5L, a
majority of them reported ‘no problem’ for each dimension; ranging from 47% for
pain/discomfort dimension to 96% for self-care dimension (Table 4.3). The mean VAS
score was 83. Out of a total of 1,207 respondents, 366 respondents (30%) reported
their own health as full health (11111). Only 1 respondent reported extreme problem
in dimension anxiety/depression. The mean time of computer-based interviewing was
37 minutes (SD = 12.65). This time was not include the process prior and after
computer-based interviewing, i.e. inform consent, interviewing with paper-based
country-specific question, and qualitative questions.
Table 4.3 Health status of respondents by level of severity
Dimensions No problems Slight problems
Moderate problems
Severe problems
Extreme
N % N % N % N % N % Mobility 873 72.33% 235 19.47% 84 6.96% 15 1.24% - Self-care 1,163 96.35% 32 2.65% 9 0.75% 3 0.25% - Usual activities 952 78.87% 188 15.58% 60 4.97% 7 0.58% - Pain/discomfort 571 47.31% 525 43.5% 98 8.12% 13 1.08% - Anxiety/depression 823 68.19% 313 25.93% 62 5.14% 8 0.66% 1 0.08
4.1.2 Valuation by TTO
The mean time used for 1 TTO task was 76.27 seconds (sd=56.08). It
should be noted that each respondent valued 10 TTO tasks. Table 4.4 presents the
mean observed value of 86 health state’s profiles, the highest mean TTO score was
0.94 for the following 4 health states: 11112, 11121, 12111, 21111. The lowest mean
TTO score was -0.31 for health state 55555. Eight out of 86 health states were valued
as worse than death, i.e. 35245, 55225, 44345, 55424, 44553, 52455, 43555, 55555.
With regard to TTO feedback questions, most of Thai respondents
understand the TTO questions, however it was not easy to differentiate between the
lives they were asked to imagine. They agreed that it was difficult to decide on the
exact points where Life A and Life B were about the same (Table 4.5).
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 65
Table 4.4 Observed mean values by health state’s profiles
Health state
Severity Mean SD Health state
Severity Mean SD Health state
Severity Mean SD
11112 6 0.94 0.01 23242 13 0.59 0.03 45133 16 0.33 0.03
11121 6 0.94 0.01 25222 13 0.68 0.02 51451 16 0.24 0.04
11211 6 0.93 0.01 32314 13 0.61 0.03 24443 17 0.19 0.04
12111 6 0.94 0.01 35311 13 0.59 0.02 34244 17 0.18 0.04
21111 6 0.94 0.01 42115 13 0.53 0.03 43514 17 0.25 0.04
11122 7 0.91 0.01 53221 13 0.54 0.03 45233 17 0.27 0.04
11212 7 0.88 0.01 12344 14 0.52 0.02 45413 17 0.20 0.04
11221 7 0.87 0.01 25331 14 0.59 0.02 53243 17 0.22 0.04
12112 7 0.88 0.01 31514 14 0.50 0.03 34155 18 0.06 0.04
12121 7 0.87 0.01 34232 14 0.56 0.02 34515 18 0.14 0.04
21112 7 0.88 0.01 51152 14 0.36 0.03 43542 18 0.13 0.04
11421 9 0.77 0.02 12543 15 0.50 0.03 45144 18 0.03 0.05
13122 9 0.79 0.02 21345 15 0.39 0.04 52335 18 0.24 0.04
14113 10 0.76 0.01 21444 15 0.39 0.03 53244 18 0.08 0.04
11414 11 0.67 0.02 22434 15 0.41 0.03 54153 18 0.06 0.04
13313 11 0.74 0.01 23514 15 0.42 0.03 54342 18 0.16 0.04
11235 12 0.61 0.03 24342 15 0.41 0.03 55233 18 0.19 0.04
12513 12 0.65 0.02 31524 15 0.41 0.03 14554 19 0.13 0.04
13224 12 0.68 0.02 52215 15 0.36 0.03 24445 19 0.02 0.05
21315 12 0.60 0.03 52431 15 0.40 0.04 24553 19 0.04 0.04
25122 12 0.63 0.03 53412 15 0.33 0.04 35245 19 -0.02 0.05
42321 12 0.61 0.03 54231 15 0.38 0.04 55225 19 -0.01 0.05
11425 13 0.61 0.02 31525 16 0.38 0.04 44345 20 -0.02 0.04
12244 13 0.51 0.03 32443 16 0.37 0.03 55424 20 -0.01 0.04
12334 13 0.57 0.03 33253 16 0.30 0.04 44553 21 -0.02 0.05
12514 13 0.48 0.03 35143 16 0.39 0.03 52455 21 -0.03 0.05
15151 13 0.50 0.03 35332 16 0.38 0.03 43555 22 -0.10 0.05
21334 13 0.57 0.02 43315 16 0.30 0.04 55555 25 -0.31 0.01
23152 13 0.58 0.03 44125 16 0.35 0.03
Juntana Pattanaphesaj Results / 66
Table 4.5 TTO feedback
Questions Completely agree Agree Neutral Disagree Completely
disagree Question 1 146 (12.10%) 606 (50.21%) 385 (31.90%) 63 (5.22%) 7 (0.58%) Question 2 39 (3.23%) 274 (22.7%) 346 (28.67%) 473 (39.19%) 75 (6.21%) Question 3 108 (8.95%) 639 (52.94%) 355 (29.41%) 97 (8.04%) 8 (0.66%) Question 1 : It was easy to understand the questions I was asked.
Question 2 : I found it easy to tell the difference between the lives I was asked to think about.
Question 3 : I found it difficult to decide on the exact points where Life A and Life B were about the same.
4.1.3 Data diagnostic tests
The models were checked for potential problems and evaluating the
plausibility of key assumptions of multilevel analysis (81): 1) linearity between
dependent and independent variables; 2) homoscedasticity; and 3) normal distributions
of the residuals.
The linearity between dependent and independent variables was
demonstrated by the scatter plot between TTO value (Y-axis) and level of severity (X- axis)
of all respondents with median band. Figure 4.1 reveals that the relationship between
TTO value and level of severity is linear. Thus linear regression could be applied to
our data set.
-1-.
50
.51
5 10 15 20 25Severity
Value Median bands
Figure 4.1 Scatter plot between TTO value (Y) and level of severity (X) of all
respondents
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 67
A spaghetti plot is a method of viewing individual data to visualize the
relationship between x and fitted y. Since our data was large, 10 respondents were
selected to show the pattern of relationship. Figure 4.2 demonstrates that both the
intercept and slope of the data are random. Thus random effect model could be used to
analyze the data of this study.
-2-1
01
2
Fitt
ed v
alue
s
5 10 15 20 25severity
Spaghetti plot of value severity id int_id1
Figure 4.2 spaghetti plot between fitted value (Y) and level of severity (X) of 10
selected respondents
Each model was tested for the mispecification using the Hausman test. The
Hausman test was used to determine whether fixed effect or random effect models
should be used. Heteroscedasticity of all models was inspected using the likelihood
ratio test (LR test). According to the test, heteroscedasticity was found in all models.
So, robust standard errors were employed in the model estimation.
The normality of residuals was tested using Shapiro-Wilk test, Kernel
density estimate, and probability-probability (P-P) plot. The Shapiro-Wilk test
indicated that the distribution of the data was not normality. However, the inspection
by visual graph (Figure 4.3 and 4.4) showed that the distribution of the data was near
normal distribution.
Juntana Pattanaphesaj Results / 68
0.5
11.
52
Den
sity
-1 -.5 0 .5 1 1.5Residuals
Kernel density estimateNormal density
kernel = epanechnikov, bandwidth = 0.0304
Kernel density estimate
Figure 4.3 Kernel density estimate of residuals
0.00
0.25
0.50
0.75
1.00
Nor
mal
F[(
resi
d_R
coef
-m)/
s]
0.00 0.25 0.50 0.75 1.00Empirical P[i] = i/(N+1)
Figure 4.4 Probability-probability (P-P) plot of residuals
4.1.4 Data selection to input in the regression model
According to the low quality data criteria (Table 4.6), twenty six out of
1,207 cases met the criteria. After excluding the 26 cases with low quality, magnitude
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 69
of inconsistency was calculate for each of the remaining cases (1,181 cases). Magnitude
of inconsistency per respondent ranged from 0 to 10, as shown in Figure 4.5.
Table 4.6 Number of respondents that met the criteria for low quality data for TTO
valuation
Criteria Number of
respondents*
1. logical inconsistency 2 (0.2%)
2. positive slope 1 (0.1%)
3. all 10 health states got the same value 0 (0.0%)
4. too many health states (> 8 health states) were valued as WTD 23 (1.9%)
5. too many health states (>8 health states) were valued as zero 1 (0.1%)
6. very mild states (severity level = 6 or 7) were given very low
value (value < 0)
5 (0.4%)
* The number of case can be double counted.
Figure 4.5 Magnitude of inconsistency among 1,181 respondents
Then, all 1,181 respondents were arbitrary divided into 3 groups (A, B, C)
according to the magnitude of inconsistency: group A consisted of 438 respondents
whose no inconsistency was found; group B consisted of 702 respondents whose
magnitude of inconsistency ranged between 0.01-2.99; and group C consisted of 41
respondents whose magnitude of inconsistency ranged between 3.0-10.0. By using
No. of respondent
Magnitude of inconsistency
Juntana Pattanaphesaj Results / 70
three as a cut-off point for magnitude of inconsistency, only 41 respondents (3.4%) of
sample will be excluded if group C showed poor fit statistics. It should be noted that if
cut-off point between group B and C was two, 72 respondents (6%) of sample will be
excluded.
By using both data quality criteria and magnitude of inconsistency, 4
subgroups were classified, as shown in Table 4.7. Subgroup 1 consisted of the
respondents whose data quality was satisfied and no inconsistency was found.
Subgroup 2 consisted of the respondents whose data quality was satisfied and
magnitude of inconsistency was lower than 3.0. Subgroup 3 consisted of the
respondents whose data quality was satisfied and their magnitude of inconsistency was
lower than 10.0. Lastly, subgroup 4 was the total respondents. These 4 subgroups were
compared in terms of mean TTO score, parameter estimates and the fit statistics in
order to find out the best subgroup to estimate Thai value sets. Table 4.8 and Figure
4.6 presents the mean TTO scores of 4 these subgroups.
Table 4.7 Subgroup classification by low quality data criteria and magnitude of
inconsistency
Subgroup Inclusion of
low quality data Magnitude of inconsistent value n
1 No 0 (group A) 438 2 No 0 – 2.99 (group A +B) 1,140 3 No 0 – 10 (group A+B+C) 1,181 4 Yes 0 – 10 (group A +B+C) 1,207
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 71
Table 4.8 Mean TTO scores by subgroup
Profile Subgroup 1 Subgroup 2 Subgroup 3 Subgroup 4
Mean SD n Mean SD n Mean SD n Mean SD n
11112 0.941 0.070 87 0.937 0.084 219 0.936 0.085 225 0.932 0.096 229
11121 0.947 0.068 86 0.935 0.086 236 0.934 0.086 246 0.931 0.094 255
11211 0.933 0.086 106 0.926 0.091 239 0.926 0.092 249 0.923 0.100 254
12111 0.954 0.066 94 0.941 0.082 227 0.938 0.086 233 0.928 0.152 240
21111 0.945 0.065 65 0.938 0.081 219 0.940 0.080 228 0.933 0.140 229
11122 0.920 0.091 49 0.907 0.100 116 0.908 0.099 117 0.883 0.216 120
11212 0.895 0.099 29 0.876 0.116 113 0.879 0.115 117 0.879 0.115 117
11221 0.892 0.103 45 0.881 0.116 111 0.872 0.127 116 0.870 0.128 120
12112 0.888 0.102 29 0.873 0.110 113 0.876 0.109 117 0.876 0.109 117
12121 0.877 0.126 76 0.858 0.138 134 0.859 0.138 135 0.854 0.148 139
21112 0.882 0.118 49 0.877 0.122 125 0.877 0.121 131 0.873 0.140 134
11421 0.817 0.107 36 0.780 0.151 106 0.770 0.167 111 0.757 0.217 112
13122 0.825 0.099 30 0.795 0.134 105 0.784 0.179 114 0.783 0.179 115
14113 0.796 0.125 39 0.760 0.139 106 0.754 0.141 112 0.752 0.141 114
11414 0.757 0.108 37 0.692 0.171 111 0.670 0.223 115 0.590 0.415 121
13313 0.756 0.120 36 0.734 0.153 106 0.736 0.156 111 0.722 0.213 112
11235 0.713 0.165 45 0.646 0.218 111 0.625 0.259 116 0.590 0.319 120
12513 0.681 0.133 49 0.634 0.222 125 0.616 0.277 131 0.580 0.362 134
13224 0.721 0.134 49 0.695 0.202 116 0.694 0.202 117 0.670 0.271 120
21315 0.690 0.146 39 0.622 0.217 106 0.615 0.232 112 0.593 0.288 114
25122 0.688 0.131 36 0.626 0.268 106 0.617 0.296 111 0.606 0.317 112
42321 0.679 0.158 49 0.634 0.222 116 0.621 0.263 117 0.598 0.313 120
11425 0.687 0.149 30 0.647 0.151 105 0.607 0.232 114 0.600 0.243 115
12244 0.621 0.143 36 0.516 0.261 106 0.491 0.302 111 0.480 0.324 112
12334 0.615 0.208 48 0.553 0.311 113 0.553 0.311 113 0.548 0.312 115
12514 0.594 0.183 45 0.532 0.264 111 0.499 0.326 116 0.462 0.382 120
15151 0.578 0.227 39 0.504 0.294 106 0.473 0.350 112 0.457 0.369 114
21334 0.600 0.195 48 0.549 0.285 113 0.549 0.285 113 0.535 0.304 115
23152 0.669 0.113 29 0.592 0.262 113 0.585 0.277 117 0.585 0.277 117
23242 0.661 0.162 48 0.569 0.321 113 0.569 0.321 113 0.566 0.324 115
25222 0.650 0.182 37 0.687 0.164 111 0.681 0.182 115 0.604 0.386 121
32314 0.663 0.200 48 0.590 0.317 113 0.590 0.317 113 0.580 0.344 115
35311 0.590 0.159 49 0.600 0.207 116 0.599 0.207 117 0.584 0.234 120
42115 0.595 0.157 30 0.561 0.218 105 0.515 0.321 114 0.507 0.331 115
53221 0.552 0.216 49 0.530 0.307 125 0.493 0.350 131 0.460 0.409 134
12344 0.468 0.229 49 0.480 0.298 125 0.474 0.308 131 0.442 0.373 134
25331 0.555 0.314 37 0.601 0.232 111 0.595 0.247 115 0.530 0.387 121
31514 0.489 0.313 37 0.519 0.255 111 0.503 0.279 115 0.432 0.414 121
34232 0.514 0.215 49 0.573 0.214 116 0.571 0.215 117 0.547 0.277 120
51152 0.502 0.224 30 0.408 0.306 105 0.343 0.385 114 0.338 0.388 115
12543 0.540 0.288 76 0.501 0.338 134 0.500 0.336 135 0.472 0.372 139
21345 0.553 0.130 29 0.384 0.407 113 0.383 0.403 117 0.383 0.403 117
21444 0.403 0.337 37 0.402 0.333 111 0.392 0.341 115 0.325 0.444 121
Juntana Pattanaphesaj Results / 72
Table 4.8 Mean TTO scores by subgroup (cont.)
Profile Subgroup 1 Subgroup 2 Subgroup 3 Subgroup 4
Mean SD n Mean SD n Mean SD n Mean SD n
22434 0.405 0.294 30 0.442 0.310 105 0.398 0.380 114 0.392 0.384 115
23514 0.419 0.329 76 0.433 0.321 134 0.433 0.320 135 0.405 0.358 139
24342 0.432 0.315 48 0.389 0.376 113 0.389 0.376 113 0.390 0.377 115
31524 0.440 0.261 39 0.418 0.315 106 0.399 0.351 112 0.382 0.373 114
52215 0.435 0.318 76 0.366 0.397 134 0.356 0.412 135 0.331 0.435 139
52431 0.473 0.268 39 0.417 0.338 106 0.416 0.336 112 0.397 0.364 114
53412 0.359 0.338 48 0.331 0.403 113 0.331 0.403 113 0.330 0.401 115
54231 0.451 0.297 45 0.424 0.359 111 0.381 0.420 116 0.354 0.441 120
31525 0.500 0.205 36 0.411 0.349 106 0.369 0.399 111 0.359 0.409 112
32443 0.290 0.377 76 0.392 0.335 134 0.393 0.334 135 0.364 0.371 139
33253 0.197 0.403 48 0.296 0.403 113 0.296 0.403 113 0.299 0.403 115
35143 0.332 0.356 37 0.384 0.345 111 0.385 0.348 115 0.319 0.447 121
35332 0.300 0.362 30 0.394 0.340 105 0.376 0.359 114 0.371 0.362 115
43315 0.333 0.258 39 0.306 0.368 106 0.304 0.386 112 0.287 0.405 114
44125 0.319 0.305 49 0.328 0.363 125 0.316 0.376 131 0.287 0.419 134
45133 0.242 0.406 76 0.354 0.366 134 0.353 0.365 135 0.324 0.402 139
51451 0.326 0.349 45 0.277 0.385 111 0.246 0.419 116 0.222 0.433 120
24443 0.208 0.331 39 0.178 0.439 106 0.190 0.432 112 0.175 0.444 114
34244 0.426 0.137 29 0.194 0.435 113 0.168 0.463 117 0.168 0.463 117
43514 0.483 0.174 29 0.266 0.437 113 0.250 0.449 117 0.250 0.449 117
45233 0.361 0.296 36 0.289 0.403 106 0.285 0.418 111 0.275 0.429 112
45413 0.208 0.378 30 0.224 0.423 105 0.188 0.454 114 0.184 0.454 115
53243 0.228 0.359 37 0.233 0.390 111 0.225 0.396 115 0.168 0.462 121
34155 -0.049 0.476 76 0.058 0.484 134 0.061 0.484 135 0.043 0.490 139
34515 0.179 0.415 45 0.171 0.431 111 0.157 0.442 116 0.130 0.461 120
43542 0.024 0.436 76 0.124 0.460 134 0.126 0.459 135 0.106 0.469 139
45144 0.061 0.440 45 0.058 0.472 111 0.045 0.481 116 0.022 0.492 120
52335 0.155 0.432 49 0.242 0.435 116 0.241 0.433 117 0.225 0.449 120
53244 -0.023 0.455 37 0.079 0.452 111 0.076 0.452 115 0.027 0.491 121
54153 0.071 0.384 39 0.054 0.462 106 0.058 0.463 112 0.046 0.471 114
54342 0.142 0.388 49 0.164 0.432 125 0.133 0.449 131 0.108 0.474 134
55233 0.258 0.357 36 0.200 0.432 106 0.201 0.431 111 0.194 0.436 112
14554 0.020 0.416 49 0.092 0.449 125 0.088 0.466 131 0.064 0.487 134
24445 -0.098 0.504 49 0.019 0.482 116 0.010 0.489 117 0.007 0.487 120
24553 -0.003 0.456 30 0.076 0.465 105 0.031 0.484 114 0.027 0.483 115
35245 -0.054 0.451 45 0.027 0.473 111 0.024 0.475 116 0.000 0.487 120
55225 -0.048 0.445 48 -0.032 0.515 113 -0.032 0.515 113 -0.035 0.512 115
44345 -0.111 0.436 49 -0.048 0.486 125 -0.050 0.483 131 -0.071 0.497 134
55424 0.307 0.158 29 -0.010 0.484 113 -0.013 0.479 117 -0.013 0.479 117
44553 0.253 0.160 29 -0.037 0.476 113 -0.031 0.485 117 -0.031 0.485 117
52455 0.122 0.400 36 -0.020 0.479 106 -0.036 0.494 111 -0.043 0.497 112
43555 -0.229 0.488 49 -0.088 0.515 116 -0.095 0.519 117 -0.108 0.526 120
55555 -0.271 0.454 438 -0.292 0.485 1140 -0.306 0.487 1181 -0.317 0.489 1207
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 73
Figure 4.6 Mean TTO scores by subgroup
Random effect model was employed to 4 subgroups to generate parameters
and fit statistics as shown in Table 4.9. Cross validation demonstrated that subgroup 2
performed the best as the absolute difference between actual mean scores and
predicted scores was the lowest. In terms of fit statistics (i.e. R square, RMSE, CCC,
and Cohen effect), compared to subgroup 2-4, subgroup 1 was the best, however it
produced too many number of states with absolute difference > 0.1. This may occur
because low number of sample size of subgroup 1 led to more variability between
individual although the inconsistency within individual was the lowest. The data from
subgroup 2 showed the best validation and acceptable fit statistics, so it was selected
to test the functional forms and estimate the Thai value sets.
Table 4.9 parameter estimates and the fit statistics by subgroup
Parameters Subgroup 1 Subgroup 2 Subgroup 3 Subgroup 4 (n=438) (n=1,140) (n=1,181) (n=1,207)
Cross validation - Number of inconsistent
coefficient in the regression model
0
0
0
0
- Number of states with absolute difference*> 0.1 (out of 86 health states)
30 11 14 22
Juntana Pattanaphesaj Results / 74
Table 4.9 parameter estimates and the fit statistics by subgroup (cont.)
Parameters Subgroup 1 Subgroup 2 Subgroup 3 Subgroup 4 (n=438) (n=1,140) (n=1,181) (n=1,207)
- Number of states with
absolute difference > 0.15 (out of 86 health states)
9 4 5 9
Maximum score
0.9817 (11211)
0.9607 (12111)
0.9555 (12111)
0.9458 (12111)
Minimum score (55555)
-0.2838 -0.3015 -0.3166 -0.3296
Range from the best to the worst score
1.2838 1.3015 1.3166 1.3296
R square 0.57 0.52 0.50 0.47 RMSE 0.24 0.27 0.28 0.29 CCC 0.73 0.68 0.67 0.64 Cohen effect 1.36 1.07 1.01 0.90 * absolute difference = the difference between actual mean scores and predicted scores
4.1.5 Testing for the functional form
The weighted data set of subgroup 2 was tested with various functional
forms, however only 8 functional forms were reported, i.e. with/without constant,
with/without N45 term, with/without N5 term, and with/without D123^2 term. Table
4.10 demonstrates coefficients and fit statistics generated from 8 functional forms. No
inconsistent coefficients was found among 20 dummy variables, however, the
coefficient of interaction terms (i.e. N45) were inconsistent because of negative value. No
inconsistent preference scores found was among 3,125 health states for all 8 models.
According to Figure 4.7, the predicted score from model 5 and 6 showed
unacceptable deviation from actual mean score. In addition, Bland-Altman plot
(Figure 4.8) confirmed that model 5 and 6 showed the most bias since only 73% and
87% of dot were covered with 95% confident interval (Table 4.10). Thus model 5 and
6 were excluded.
According to Figure 4.7, the predicted score from model 7 and 8 shows
more deviation from actual mean score when the predicted score (Y-axis) is lower
than 0.2 compared with model 1 and 2. The Bland-Altman plots also demonstrates that
model 7 and 8 has more prediction bias as 6% of dot are out of 95% confident interval
range (Table 4.10). Thus model 7 and 8 were excluded.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 75
In terms of AIC and responsiveness (measured by SRM), model 3 and 4 was
better than model 1 and 2. However, the coefficient of interaction term (N45) of model 3
and 4 was inconsistent because of negative value. Thus model 3 and 4 were excluded.
Bland-Altman plot showed that model 1 and 2 had lowest prediction bias as 96.5% of dot
were covered by 95% confident interval range. Model 1 and 2, which offered less
variables or more parsimony, were better.
Compared with model 1, model 2 was better in terms of higher
responsiveness and more parsimony. Thus the model 2 which was the model without
constant and interaction terms was selected to estimate the Thai value sets.
Table 4.10 Coefficients and fit statistics generated from data of subgroup 2 by functional form
Functional form
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Constant
+ main effect
Main effect Constant
+ main effect + N45a
Main effect
+ N45
Constant + main effect
+ N5b
Main effect
+ N5
Constant + main effect
+ D123^2c
Main effect + D123^2
Coef SE Coeff SE Coeff SE Coeff SE Coeff SE Coeff SE Coeff SE Coeff SE
MO2 0.052 0.009 0.056 0.009 0.072 0.010 0.075 0.009 0.052 0.010 0.056 0.009 0.058 0.010 0.062 0.009 MO3 0.112 0.010 0.114 0.010 0.137 0.010 0.139 0.010 0.112 0.011 0.114 0.011 0.122 0.010 0.125 0.010 MO4 0.229 0.010 0.231 0.010 0.256 0.011 0.259 0.010 0.228 0.010 0.231 0.010 0.227 0.010 0.230 0.010 MO5 0.306 0.011 0.307 0.011 0.324 0.011 0.325 0.011 0.306 0.011 0.307 0.011 0.300 0.011 0.301 0.011
SC2 0.028 0.008 0.033 0.007 0.045 0.008 0.049 0.007 0.028 0.008 0.033 0.007 0.032 0.008 0.036 0.007 SC3 0.107 0.009 0.108 0.009 0.121 0.009 0.122 0.009 0.107 0.009 0.108 0.009 0.109 0.009 0.111 0.009 SC4 0.223 0.012 0.225 0.012 0.242 0.012 0.243 0.012 0.223 0.012 0.225 0.012 0.216 0.012 0.217 0.012 SC5 0.252 0.009 0.254 0.009 0.269 0.009 0.271 0.009 0.252 0.009 0.253 0.009 0.237 0.011 0.237 0.011
UA2 0.038 0.008 0.043 0.008 0.056 0.009 0.061 0.008 0.038 0.008 0.043 0.008 0.041 0.009 0.045 0.008 UA3 0.070 0.011 0.075 0.011 0.092 0.011 0.097 0.011 0.070 0.011 0.075 0.011 0.073 0.011 0.077 0.011 UA4 0.161 0.009 0.165 0.009 0.191 0.011 0.194 0.010 0.161 0.009 0.165 0.009 0.155 0.010 0.158 0.009 UA5 0.205 0.009 0.207 0.009 0.224 0.010 0.226 0.009 0.205 0.009 0.207 0.009 0.192 0.010 0.194 0.010
PD2 0.034 0.008 0.040 0.007 0.049 0.008 0.053 0.007 0.034 0.008 0.040 0.007 0.038 0.008 0.042 0.007 PD3 0.068 0.010 0.068 0.010 0.086 0.011 0.086 0.011 0.068 0.010 0.068 0.010 0.072 0.010 0.073 0.010 PD4 0.234 0.010 0.233 0.010 0.247 0.010 0.246 0.010 0.234 0.010 0.234 0.010 0.226 0.010 0.225 0.010 PD5 0.262 0.011 0.266 0.011 0.283 0.012 0.287 0.011 0.262 0.011 0.266 0.011 0.250 0.011 0.253 0.011
AD2 0.025 0.008 0.032 0.007 0.035 0.008 0.041 0.007 0.025 0.008 0.032 0.007 0.028 0.008 0.034 0.007 AD3 0.091 0.011 0.097 0.010 0.109 0.011 0.113 0.011 0.091 0.011 0.096 0.011 0.098 0.011 0.103 0.011 AD4 0.198 0.009 0.202 0.009 0.227 0.010 0.232 0.009 0.198 0.009 0.202 0.009 0.195 0.010 0.198 0.009 AD5 0.245 0.010 0.249 0.010 0.265 0.011 0.268 0.011 0.245 0.010 0.249 0.010 0.236 0.011 0.239 0.011 Const 0.020 0.007 - - 0.017 0.007 - - 0.020 0.007 0.016 0.008 - - N45 - - - - -0.084 0.012 -0.085 0.012 - - - - - - N5 - - - - - - - - 0.000 0.009 0.001 0.009 - - - -
D123^2 - - - - - - - - - - - - 0.004 0.002 0.004 0.002
Juntana Pattanaphesaj R
esults / 76
Table 4.10 Coefficients and fit statistics generated from data of subgroup 2 by functional form (cont.)
Functional form
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Constant
+ main effect
Main effect Constant
+ main effect + N45a
Main effect
+ N45
Constant + main effect
+ N5b
Main effect
+ N5
Constant + main effect
+ D123^2c
Main effect + D123^2
Second best score (11112)
0.9553 0.9679 0.9480 0.9558 0.9554 0.9680 0.9522 0.9621
Minimum score (55555)
-0.2896 -0.2832 -0.2973 -0.2918 -0.2894 -0.2829 -0.2357 -0.2269
Range from the best to worst score
1.2896 1.2832 1.2973 1.2918 1.2894 1.2829 1.2357 1.2269
Inconsistent score among 3,125 health states
0 0 0 0 0 0 0 0
Number of negative value among 3,125 health state
135 126 151 139 135 126 94 86
AIC 4987 4987 4927 4927 4987 4989 4975 4974 CCC 0.68 0.68 0.68 0.68 0.68 0.68 0.68 0.68 SRM 1.68 1.75 1.82 1.86 1.78 1.85 1.82 1.85 % dots within limits of agreement from Bland-Altman plot
96.5 96.5 94.2 94.2 73.3 87.2 94.2 94.2
a The term N45 is defined as whether severe or extreme problems (level 4 or 5) exist in any domain. N45 = 1 if there are at least 1 dimension with level 4 or 5. N45 = 0 if no level 4 or 5 in the profile. b The term N5 is defined as whether extreme problems (level 5) exist in any domain. N5 = 1 if there are at least 1 dimension with level 5. N5 = 0 if no level 5 in the profile. c The term D123^2 is defined as the number of dimensions at level 4 or 5 minus one and then squared. For instance, the health state 34435 has 3 dimensions with level 4 or 5. Thus D123^2 is calculated as (3-1)^2 = 4.
Ph.D. (Pharm
acy Adm
inistration) / 77
Fac. of Grad. Studies, M
ahidol Univ.
Juntana Pattanaphesaj Results / 78
Figure 4.7 Comparison between actual mean score and predicted score of the model
1-8 using data from subgroup 2
Pre
dict
ed s
core
Health state
Health state
Pre
dict
ed s
core
Health state
Pre
dict
ed s
core
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 79
Figure 4.8 Bland-Altman plots of model 1-8
X-axis = mean of actual and predicted scores; Y-axis = difference of actual and predicted scores; The two dash lines are upper and lower limit of 95% confidence interval. The percent of dot that was out of 95%CI line of model 1-8 was 3, 3, 5, 5, 23, 11, 5 and 5%, respectively.
Juntana Pattanaphesaj Results / 80
4.1.6 The Thai algorithm and the preference scores
The Thai algorithm and the preference scores were estimated using the
weighted data from subgroup 2 and model 2 which had no constant and interaction
terms. Table 4.11 presents the coefficients of the Thai model. Thai preference scores
are calculated by 1 – disutility using the following algorithm.
Thai score for EQ-5D-5L = 1-(0.056*MO2)-(0.114*MO3)-(0.231*MO4)-
(0.307*MO5)-(0.033*SC2)-(0.108*SC3)-(0.225*SC4)-
(0.254*SC5)-(0.043*UA2)-(0.075*UA3)-(0.165*UA4)-
(0.207*UA5)-(0.040*PD2)-(0.068*PD3)-(0.233*PD4)-
(0.266*PD5)-(0.032*AD2)-(0.097*AD3)-(0.202*AD4)-
(0.249*AD5)
Where
MO2, SC2, UA2, PD2, AD2 = 1 if level 2 is responded, 0 otherwise
MO3, SC3, UA3, PD3, AD3 = 1 if level 3 is responded, 0 otherwise
MO4, SC4, UA4, PD4, AD4 = 1 if level 4 is responded, 0 otherwise
MO5, SC5, UA5, PD5, AD5 = 1 if level 5 is responded, 0 otherwise
Table 4.11 Coefficients for main effects of the Thai model
Variable Coefficients SE 95% CI MO2 0.056 0.009 0.039 - 0.074 MO3 0.114 0.010 0.095 - 0.134 MO4 0.231 0.010 0.212 - 0.251 MO5 0.307 0.011 0.287 - 0.328
SC2 0.033 0.007 0.019 - 0.048 SC3 0.108 0.009 0.090 - 0.126 SC4 0.225 0.012 0.202 - 0.248 SC5 0.254 0.009 0.236 - 0.271
UA2 0.043 0.008 0.028 - 0.058 UA3 0.075 0.011 0.054 - 0.095 UA4 0.165 0.009 0.147 - 0.182 UA5 0.207 0.009 0.190 - 0.225
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 81
Table 4.11 Coefficients for main effects of the Thai model (cont.)
Variable Coefficients SE 95% CI
PD2 0.040 0.007 0.025 - 0.054 PD3 0.068 0.010 0.048 - 0.088 PD4 0.233 0.010 0.214 - 0.253 PD5 0.266 0.011 0.246 - 0.287
AD2 0.032 0.007 0.017 - 0.047 AD3 0.097 0.010 0.076 - 0.117 AD4 0.202 0.009 0.185 - 0.220 AD5 0.249 0.010 0.229 - 0.269
Table 4.12 Examples for calculating the Thai preference score for the EQ-5D-5L
Variable Coefficients The preference score
for health state 12112
The preference score for health state
55555 MO2 0.056
MO3 0.114 MO4 0.231 MO5 0.307
-0.307
SC2 0.033 -0.033 SC3 0.108
SC4 0.225 SC5 0.254
-0.254
UA2 0.043 UA3 0.075 UA4 0.165 UA5 0.207
-0.207
PD2 0.040 PD3 0.068 PD4 0.233 PD5 0.266
-0.266
AD2 0.032 -0.032 AD3 0.097
AD4 0.202 AD5 0.249
-0.249 The preference score = 1-0.033-0.032 =1-0.307-0.254-
0.207-0.266-0.249 = 0.935 = -0.283
Juntana Pattanaphesaj Results / 82
According to the Thai algorithm, the level 5 of the mobility dimension
(unable to walk about) has the most impact to the utility decrement (coefficient = 0.307),
followed by extreme pain or discomfort (coefficient = 0.266). While the state that has
less impact to the utility decrement is slightly anxious or depressed and slight problems
washing or dressing myself (coefficient = 0.032 and 0.033). Since no constant in the
algorithm, the full health (11111) can be directly calculated from this algorithm without
assume to be equal to 1.00. The second best score is 0.968 for health state 11112 and the
worst score is -0.283 for health state 55555. The number of health state which the
preference score is lower than zero (worse than dead) is 126 health states (4.03%).
4.1.7 Comparing Thai preference score with the interim value sets
from mapping technique
The interim scoring for the Thai EQ-5D-5L was estimated by the EuroQol
group (33, 34) using mapping methodology while the 5L valuation study in Thailand
was on going. Figure 4.9 compares the preference score between those 2 value sets. The
Y-axis represents preference score and X-axis represents 3,125 health states sorted by
severity level and health profiles. It is clearly that the preference score from this study
was higher than the score from mapping. Paired T-test revealed that there was statistical
significant difference between those two sets (p-value < 0.001) and the mean difference
was 0.2. The interim scoring generated more states with negative scores than Thai value
sets (558 vs 126 states). No inconsistent score found among 3,125 states of both value
sets. The value sets generated from this study was better than mapping method in terms
of responsiveness (SRM = 1.75 vs 0.92).
Figure 4.9 Comparing EQ-5D-5L utility score obtaining from surveying and mapping
Health state
Preference score
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 83
Table 4.13 Comparing EQ-5D-5L value sets’ parameter between Thai’s and interim
scoring
Parameter Thai value sets Interim scoring Second best score (11112) 0.968 0.814 Worst score (55555) -0.283 -0.452 # negative score 126 (4.03%) 558 (17.86%) # inconsistent score among 3,125 states 0 0 SRM 1.75 0.92 Mean difference (paired t-test) 0.2 (p-value < 0.001)
4.1.8 Valuation by DCE
The mean time used by the respondent for valuing 1 pair of DCE task was
48 seconds (sd = 31.77). It is noted that each respondent valued 7 pairs of DCE tasks,
yielding 14 records per respondent: 7 preferred health states and 7 rejected health
states. Regarding data management of DCE data, 450 out of 16,898 records (2.7%)
were considered as low quality data, as shown in Table 4.14. The mean time of DCE
task among low quality data was 40 seconds (sd = 30), which was statistically
significantly shorter (p<0.001) than average time of all sample.
Table 4.14 Number of respondents that met the criteria for low DCE data quality
Criteria Number of records* 1. All responses were on the same side. 70
(5 respondents x 14 records) 2. The responses had particular order. 238
(17 respondents x 14 records) 3. The time spent was < 10 seconds/pair 88 4. More severe health state was preferred 86 * The number of records can be double counted.
Various models were tested with the data set, however only the important
results were shown in this report. Both conditional logit model (clogit) and random
effect logit model (xtlogit) were employed to both original and cleaned data set. Model
2 showed better fit statistics (in terms of pseudo R2 and AIC) than model 1 and 3
(Table 4.15). However, the specification test (linktest) revealed specification error
which usually means that there were omitted variables for model 1-3. Therefore the
interaction terms among 5 dimensions and 5 levels of severity (130 interaction terms)
were input in the model 2 and the results were showed in model 4. As a result, model 4
Juntana Pattanaphesaj Results / 84
showed the best fit model while no misspecification was found. Nevertheless, all 4
models showed several inconsistent coefficients.
Table 4.15 Coefficients and fit statistics of DCE model
Functional form
Model 1 Model 2 Model 3 Model 4 conditional logistic
model conditional logistic
model (cleaned data) random effect logit
model (cleaned data)
conditional logistic model with interaction terms*
(cleaned data) Coef SE Coeff SE Coeff SE Coeff SE
Const - - - - 0.312 0.071 - - MO2 -0.066 0.050 -0.063 0.051 -0.061 0.051 -0.175 0.148 MO3 0.092 0.053 0.088 0.054 0.079 0.055 -0.040 0.230 MO4 -0.106 0.052 -0.104 0.053 -0.105 0.053 -0.197 0.196 MO5 -0.291 0.054 -0.288 0.055 -0.288 0.056 -0.332 0.137
SC2 0.012 0.053 0.009 0.054 0.017 0.053 -0.100 0.160 SC3 0.010 0.053 0.006 0.054 0.004 0.054 -0.034 0.211 SC4 -0.104 0.053 -0.106 0.054 -0.104 0.055 -0.224 0.219 SC5 -0.141 0.051 -0.148 0.052 -0.144 0.053 -0.164 0.134
UA2 -0.046 0.052 -0.052 0.053 -0.056 0.053 -0.012 0.172 UA3 0.066 0.055 0.058 0.056 0.043 0.055 0.066 0.203 UA4 -0.061 0.052 -0.066 0.053 -0.067 0.054 -0.027 0.237 UA5 -0.157 0.053 -0.161 0.054 -0.163 0.054 -0.190 0.131
PD2 0.045 0.052 0.047 0.053 0.034 0.053 -0.124 0.167 PD3 0.059 0.053 0.062 0.054 0.046 0.054 0.067 0.207 PD4 -0.100 0.052 -0.099 0.053 -0.110 0.054 -0.293 0.229 PD5 -0.191 0.053 -0.189 0.054 -0.198 0.055 -0.243 0.135
AD2 0.003 0.053 0.000 0.054 -0.003 0.053 0.040 0.150 AD3 -0.101 0.052 -0.101 0.053 -0.104 0.054 -0.069 0.211 AD4 -0.198 0.052 -0.195 0.053 -0.191 0.054 -0.218 0.194 AD5 -0.260 0.052 -0.257 0.053 -0.264 0.054 -0.184 0.133
Pseudo R2 12.64 12.91 n/a 21.19 AIC 17363 16696 20295 15291
* The coefficients of interaction term were not showed due to too many significant terms (61 out of the total 130 interaction terms).
With regard to DCE feedback questions, most of Thai respondents
understand the DCE questions; however they reported that it was difficult to
differentiate between 2 health states which they were asked (Table 4.16).
Table 4.16 DCE feedback
Questions Completely Agree Agree Neutral Disagree Completely
disagree Question 1 308 (25.52%) 670 (55.51%) 205 (16.98%) 23 (1.91%) 1 (0.08%)
Question 2 29 (2.40%) 194 (16.07%) 285 (23.61%) 509 (42.17%) 190 (15.74%)
Question 3 215 (17.81%) 641 (53.11%) 267 (22.12%) 79 (6.55%) 5 (0.41%) Question 1 : It was easy to understand the questions I was asked.
Question 2 : I found it easy to tell the difference between the health state I was asked to think about.
Question 3 : I found it difficult to decide on my answers to the questions.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 85
4.1.9 Country-specific data
After completed TTO and DCE valuation, all respondents were asked with
5 questions designed by Thai researchers. These questions needed the respondents to
express their opinions about philosophy of life. They were asked whether they agreed
with the following sentences.
Sentence 1 - Everything that happens in my life is a consequence of my
actions from my previous life.
Sentence 2 - When I am sick, there is someone looks after me.
Sentence 3 - Regardless of any serious sickness, I try to live as long as
possible to have something done.
Sentence 4 - According to my belief, to escape the problems by
terminating my own life is seriously wrong.
Sentence 5 - I hold the religious doctrine when I face the problems in my
life.
Table 4.17 shows that a majority of the respondents (93%) hold the
religious doctrine when facing to the problems in their life; and 87% agreed that it was
seriously wrong to escape the problems by terminating their own life. Sixty three per
cent believed in a consequence of actions from the previous life. With regard to
sickness, 86% agreed that there was someone looking after them when they were sick;
and 75% expressed that they would try to live as long as possible to have something
done regardless of how serious the sickness.
Table 4.17 Opinions of the respondents towards philosophy of life (n=1,207)
Sentences Completely disagree Disagree Neutral Agree Completely
agree 1 - Everything that happens in my life is a consequence of my actions from my previous life.
49 (4.06%)
166 (13.75%)
233 (19.30%)
590 (48.88%)
169 (14.00%)
2 - When I am sick, there is someone looks after me.
12 (0.99%)
39 (3.23%)
115 (9.53%)
769 (63.71%)
272 (22.54%)
3 - Regardless of any serious sickness, I try to live as long as possible to have something done.
44 (3.65%)
148 (12.26%)
112 (9.28%)
595 (49.30%)
308 (25.52%)
Juntana Pattanaphesaj Results / 86
Table 4.17 Opinions of the respondents towards philosophy of life (n=1,207) (cont.)
Sentences Completely disagree Disagree Neutral Agree Completely
agree 4 - According to my belief, to escape the problems by terminating my own life is seriously wrong.
30 (2.49%)
69 (5.72%)
63 (5.22%)
436 (36.12%)
609 (50.46%)
5 - I hold the religious doctrine when I face the problems in my life.
2 (0.17%)
16 (1.33%)
63 (5.22%)
777 (64.37%)
350 (29.00%)
4.1.10 Qualitative data
After completion of TTO, DCE valuation, and country-specific questions,
105 out of 1,207 respondents, who had unique responses to the TTO questions, were
asked with the following 3 questions. The first question asked the respondents about
the particular reasons to their TTO responses. The second question required the
respondents to choose 2 dimensions (the most important and the least important
dimension) from the 5 dimensions of the EQ-5D. In the third question, specific
scenario was described then the respondents were required to choose 1 preferred life
from 3 options. The details of each questions were described below.
First question
According to the first question, the respondents were further classified into
3 groups according to their reasons as follows:
Group 1: Respondents who preferred longer life even that life had a severe
health.
Group 2: Respondents, whose TTO response were based on only 1-2
health dimensions.
Group 3: Respondents who preferred full health even it offered a shorter
life.
A majority of respondents was classified in group 1 (63%) or those who
preferred longer life even that life suffered from severe health. Four main reasons
explained by these groups were as follow: 1) they firmly believed that they would get
well care and support from their families; 2) they had obligations or life goals to be
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 87
completed; 3) they had good experience on severe illness; and 4) they had certain belief
about human life. These 4 reasons could be further explained in more details below.
1.1 Care and support of family
Most respondents, who preferred longer life even that life was suffered
from a severe health, argued that if they were ill, they would get well care and support
from their family. The warm care and well support from family were important for one
to endure and struggle with severe illness. While they were encouraged from their
family, they gain more spirit to fight against the illness. At the same time, their
attempts to fight with the sickness showed good example for their family members,
and also encouraged family members to be vigorous. On the other hand, if they felt
discouraged and gave up living, their family members were also discouraged.
1.2 Obligations or Life Goals
Respondents who were the head of family often argued that they must live
as long as possible to be a pillar of their family. They had some obligations to be
complete. For example, they had to earn for their family or children; they had to took
care of aging parents; and they had to took care a sibling with illness or disability.
Some respondents had life goal to be completed before death. For instance, they
wanted to look after their parents in return; they wanted to live longer with their
lovers; they wanted to do something that they had never done before. The attitude
towards life of the respondents from this group was that the life was still valuable
despite having a severe illness. They expressed that even they got illness; they could
still make positive contribution to society unlike death people who could never do
anything to anyone.
1.3 Good experience on illness
The good experience on illness could be either a direct or an indirect
experience. In terms of direct experience on illness, some respondents had experience
on a serious illness and almost died once before. However, they could fight and
survive ultimately. So they believed that if they face serious illness again, they could
fight, survive and live long again. In terms of indirect experience on illness, some
respondents had experiences in caring patients/parents/relatives who had severe
illness. They saw that those ill people still fight and can live with the illness. Those ill
people became their role model or inspiration for the respondents to struggle with
illness; and also caused the respondents realized that life is valuable.
Juntana Pattanaphesaj Results / 88
1.4 Belief about human life
Some respondents gave a reason on certain belief about human life. They
indicated very interesting reason that to be born as a human was very difficult. So it’s
worth to spend life as long as possible. They hoped to perform merit or contribute to
society as much as possible. They believed that it was uncertainty if they would have
the opportunity to be born as a human again in the next life. This belief reflected the
Buddhist lifestyle among Thai society.
There were 19% of respondents classified into group 2 (the respondents
whose TTO responses were based only on 1-2 health dimensions). The most important
health dimensions for them were anxiety/depression, mobility, and pain/discomfort,
respectively. The respondents who prioritized the anxiety/depression the first indicated
that the anxiety/depression made them desperate, disheartened, unhappy, and it could
become the root cause of other health problems. These symptoms were more difficult
to be treated than physical symptoms. In terms of mobility, the respondents expressed
that the problems in movement may trouble most activities in their life. They felt that
they were disabled and become a burden of their family. The respondents who mainly
make decision with pain/discomfort dimension thought that pain/discomfort was the
most suffering symptoms among all 5 dimensions; and it was too hard to bear.
There were 19% of respondents classified into group 3 (the respondents
preferred full health; even it offered a shorter life.) The main reasons were that the sick
people were burden of the family. The respondents aged between 18-29 years old
argued that they were still young, healthy and could do anything by themselves. If they
get sick for long time, they might be unable to accept it since sick people is recognized
as a burden of a family. Also the rest of their life would be filled with suffering and
unhappy. Whereas, the respondents aged over 60 years gave the meaning of sick
people that it was the person who was unable to take care of themselves and was a
burden of family and peers. If they were sick and had no one look after them; their life
would be difficult. They might be seen as disabled, no value, and could not afford their
family. The rest of life would be meaningless.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 89
Second question
The second question needed the respondents choose 2 dimensions (the
most important problem and the least important problem) from the 5 dimensions of the
EQ-5D. Table 4.18 showed the most important and least important dimension
specified by the respondents. Many respondents (40%) expressed that the most
important problem was unable to walk and the least important problem was unable to
do usual activities (23%) and extremely anxious or depressed (23%).
Table 4.18 Opinions of the respondents on the most and the least important dimension
(n=105)
Dimension Most important Least important
1. unable to walk about 42 (40%) 23 (22%)
2. unable to wash or dress myself 3 (3%) 20 (19%)
3. unable to do my usual activities 9 (9%) 24 (23%)
4. extreme pain or discomfort 24 (23%) 14 (13%)
5. extremely anxious or depressed 27 (26%) 24 (23%)
Third question
In the third question, the following scenario was given: "If you were in
severe health e.g. last stage cancer, you were suffered from severe pain, and that you
will die soon". Which option would you prefer?
Option 1: Being treated with medicine without charge. This medication
would extend the life for 5 years but in severe health condition and severe pain.
Option 2: Being treated with medicine without charge. This medication
would extend the life for 6 months with full health (no pain).
Option 3: Giving up living without any treatment.
Half of respondents (54%) chose option 2 which preferred full health even
their life was shorter. The key reason was that they had a bad experience with cancer.
They had ever seen their child, parents, relatives, or neighbors died from cancer. Prior
to death, those people suffered from the painfulness of chemotherapy as well as
symptoms of cancer. These produced bad impression and negative attitude towards
cancer. With a healthy life, even short time (6 months), they could do all activities,
Juntana Pattanaphesaj Results / 90
e.g. earning for family; giving child the parting instruction; calm their life down and
prepare to death.
Twenty nine percent chose option 1 which preferred long life with severe
pain. The main reasons were worrying about family; positive thinking; and good
experiences towards cancer, respectively. Some respondents argued that they wanted
to stay with their family as long as possible to see their children grow and succeed in
life. In addition, they believed that their family would take care of and encourage them
to fight against serious illness. Some respondents had positive thinking and optimism.
They argued that the illness was a simple matter of life which no one in this world
could escape from this truth. Some respondents had good experience in caring patient
with serious illness; so they understand well about suffering from serious illness. They
believed that if they had serious illness, they could fight and endure the pain to live as
long as possible.
Eleven percent chose option 3, which was giving up their lives. Some
respondents had negative attitude towards serious illness. They had ever seen a patient
suffered from serious illness, and finally died. Some respondents did not want to be a
burden to family. They expressed that if they had severe pain, their family would
suffer also. If they died, they would not be a burden to anyone anymore. Some
respondents mentioned about the truth of life that for all 3 options of life, finally
everybody dies. So it was no meaning to try to live longer with sufferings. Thus these
respondents chose to give up living without any treatment to eliminate all problems
and stop their own and family’s sufferings.
Session 4.2: Testing the measurement properties of the Thai version
of the 5L compared to the 3L
4.2.1 Characteristics of respondents
A total of 117 patients with diabetes mellitus who met the eligibility
criteria were included. The characteristics of the respondents are shown in Table 4.19.
The average age of the respondents was 45 years, with 62.4% being female. Sixty-four
(54.7%) respondents had type 2 diabetes. The average diabetes duration of the sample
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 91
was 9 years and the average BMI was 23.30. Of the 117 respondents who completed the
first survey, 101 respondents (86%) returned the second questionnaire set by postal mail.
Table 4.19 Demographic characteristic of respondents
Demographic characteristic n (%) Type of diabetes
Type 1 53 (45.3) Type 2 64 (54.7)
Gender Male 44 (37.6) Female 73 (62.4)
Marital status Single 58 (49.6) Married 46 (39.3) Widowed 9 (7.7) Divorced/Separated 4 (3.4)
Education High school 51 (43.6) Primary school 27 (23.1) Bachelor’s degree 25 (21.4) Diploma 10 (8.5) Master’s degree or higher 4 (3.4)
Occupation Student 50 (42.7) Government/state enterprise officer 20 (17.1) Housewife 14 (12.0) Business owner 11 (9.4) Unskilled labor 7 (6.0) Retired 6 (5.1) Employee 4 (3.4) Agriculture/fishery 2 (1.7) Other 3 (2.6)
Health insurance Civil Servants Medical Benefits Scheme 58 (49.6) Out of pocket 32 (27.4) Universal coverage 20 (17.1) Social security 7 (6.0)
Median (IQR) Age (years) 45.00 (40.0) Diabetes duration (yr) 9.00 (8.50) BMI (kg/m2) 23.30 (7.37) Household income per month (Baht) 30,000 (30,000)
Juntana Pattanaphesaj Results / 92
The health state ‘11111’ was observed in 29.1% in the 5L and 33.3% for
the 3L. The second-most frequent health state reported was ‘11121’ which was 14.5%
in the 5L and 15.4% in the 3L. There were no missing values from both the 5L and
the 3L, indicating good feasibility for both instruments.
4.2.2 Distribution and ceiling effect
For all dimensions, most respondents reported no problems (level 1) for
both the 3L (52-98%) and the 5L (44-97%), as shown in Figure 4.10. Among
responses with health problems, it was clear that the 5L demonstrated better severity
level distribution than the 3L except for self-care.
With regards to the ceiling effect, the 5L showed a slightly decreasing
trend for no problem responses compared with the 3L. The percentage of patients
reporting the health state ‘11111’ decreased from 33% in the 3L to 29% in the 5L.
Nevertheless, no statistically significant difference was found. Self-care reached the
highest ceiling effect (98% for the 3L, 97% for the 5L) and showed the smallest
reduction in ceiling effect (1%) with the 5L. In contrast, pain/discomfort showed the
smallest ceiling effect (52% for the 3L, 44% for the 5L) and also showed statistically
significant reduction in ceiling effect with the 5L. No statistically significant reduction
was found for the other dimensions.
Figure 4.10 Distribution across severity level of the 3L and 5L dimension
(MO = Mobility; SC = Self-care; UA = Usual activities; PD = Pain/discomfort; AD = Anxiety/depression; No level 3 reported for the 3L and no level 5 reported for the 5L)
% of respondent
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 93
4.2.3 Redistribution
Among the answers of no problem (level 1) on the 3L, most of them
(85-98%) remained the same (no problem) on the 5L while 2-15% redistributed to
slight problems (level 2) on the 5L, as shown in Table 4.20. The majority of the
respondents who reported moderate problems (level 2) on the 3L indicated slight
problems (level 2) on the 5L (69-100%), while 9-22% shifted to moderate problems
(level 3) on the 5L. As such, redistribution occurred the least in self-care. The mean
VAS score tended to be lower according to the severity level of the 5L. No
inconsistent response was found in this study.
Table 4.20 Redistribution pattern of response from the 3L to the 5L
Dimension 3L 5L n (%) Mean VAS
Size of inconsistent response *
Mobility 1
1 83 (98%) 81.02 -1
2 2 (2%) 85.00 0
2
2 22 (69%) 72.38 0
3 7 (22%) 71.43 -1
4 3 (9%) 72.67 0
Self-care 1
1 113 (98%) 79.19 -1
2 2 (2%) 70.00 0
2 2 2 (100%) 60.00 0
Usual activities 1
1 93 (98%) 80.82 -1
2 2 (2%) 80.00 0
2
2 20 (91%) 71.85 0
3 2 (9%) 50.00 -1
Pain/discomfort 1
1 52 (85%) 81.54 -1
2 9 (15%) 86.33 0
2
2 45 (80%) 77.77 0
3 10 (18%) 64.50 -1
4 1 (2%) 50.00 0
Anxiety/depression 1
1 84 (94%) 81.38 -1
2 5 (6%) 71.80 0
2
2 23 (82%) 73.48 0
3 4 (14%) 67.50 -1
4 1 (4%) 60.00 0
* The size of inconsistency of < 0 indicated consistency.
Juntana Pattanaphesaj Results / 94
4.2.4 Convergent validity
Table 4.21 demonstrates the Spearman’s correlation coefficients between
the EQ-5D and SF-36v2 dimensions. The relation among similar dimensions was
statistically significant. Both the 3L and 5L showed an acceptable degree of
association with the SF-36v2 except for self-care, where the degree of association was
relatively low. A strong association was found between mobility and physical
functioning for both the 3L and 5L (r = 0.54 for the 3L, r = 0.53 for the 5L). Mobility
was also moderately associated with bodily pain and general health perception. With
regards to self-care, the 5L had weak association with physical functioning and social
functioning while no association was found between the 3L and SF-36v2. The usual
activities of both the 3L and the 5L were weakly associated with various dimensions
of SF-36v2, i.e. bodily pain, physical functioning, and mental health. The association
between pain/discomfort and bodily pain improved in the 5L (r = 0.30 for the 3L, r =
0.35 for the 5L) and the association between anxiety/depression and mental health was
greater in the 5L as well (r = 0.45 for the 3L, r = 0.49 for the 5L). Additionally,
Pearson’s correlation coefficient between the VAS score and utility index was similar
between the 3L and 5L (0.36 for the 3L, 0.35 for the 5L with p-value < 0.001).
Table 4.21 Correlation coefficients between EQ-5D and SF-36v2 dimensions
Dimension PF RP BP GH VT SF RE MH
3L
Mobility .54** .28** .41** .42** .25** -0.07 0.11 0.14
Self-care 0.16 0.05 .19* 0.12 0.14 0.16 0.06 0.18
Usual activities .25** .21* .30** .19* .27** 0.18 0.13 .28**
Pain/discomfort .19* 0.17 .30** .24** .18* 0.11 .21* .22*
Anxiety/depression 0.05 0.09 .23* .22* .21* .32** .29** .45**
5L
Mobility .53** .29** .44** .44** .23* -0.08 0.09 0.11
Self-care .24** .20* .23* 0.18 0.16 .24** .21* .22*
Usual activities .30** .23* .29** .22* .24* 0.16 0.14 .24**
Pain/discomfort .24** .23* .35** .28** .22* 0.08 0.16 0.18
Anxiety/depression 0.08 0.12 .19* .21* .28** .35** .29** .49** PF (physical functioning), RP (role limitation due to physical problems), BP (bodily pain), GH (general health perceptions), SF (social functioning), VT (vitality), RE (role limitations due to emotional problems), MH (general mental health) * Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 95
4.2.5 Discriminative power
The absolute informativity ( ) of the 5L was higher than the 3L for all
dimensions as shown in Table 4.22. This reflected that the 5L generated more
informativity than the 3L. We also found that the 5L generated similar results
compared with the 3L when it came to relative informativity ( ).
Juntana Pattanaphesaj Results / 96
Table 4.22 Shannon index ( ) and Shannon’s Evenness index ( ) of the 3L and the 5L
Dimension
3L 5L
3L 5L Mobility 0.85 1.20
0.53 0.52
Self-care 0.12 0.21
0.08 0.09 Usual activities 0.70 0.78
0.44 0.34
Pain/discomfort 1.00 1.40
0.63 0.60 Anxiety/depression 0.79 1.06
0.50 0.46
4.2.6 Test-retest reliability
The time interval between the first and second test was approximately 3
weeks. Overall, the reliability coefficient and percentage agreement of the 5L were
slightly lower than the 3L (Table 4.23). The weighted kappa coefficient for the 3L
ranged between 0.39 and 0.70, and between 0.44 and 0.57 for the 5L; this indicated
that the 3L had better reproducibility than the 5L. The percentage agreement gave
higher values than the weighted kappa coefficient; it was between 0.78 and 0.98 for
the 3L and 0.67 and 0.97 for the 5L. The ICCs of the 3L and 5L indexes were 0.64
and 0.70, respectively, which indicated excellent reproducibility for both instruments.
Table 4.23 Test-retest reliability of the 3L and the 5L
Dimension Weighted kappa coefficient (95% CI) Percentage
agreement 3L 5L 3L 5L
Mobility 0.70 (0.53-0.86) 0.57 (0.40-0.74) 0.89 0.83 Self-care n/a* n/a* 0.98 0.97 Usual activities 0.39 (0.16-0.62) 0.45 (0.25-0.65) 0.82 0.81 Pain/discomfort 0.56 (0.39-0.72) 0.44 (0.29-0.58) 0.78 0.67 Anxiety/depression 0.50 (0.31-0.70) 0.49 (0.33-0.65) 0.82 0.77
Intraclass correlation coefficient ** EQ-5D index 0.64 (0.51-0.74) 0.70 (0.57-0.79)
* not enough information to calculate kappa coefficient for self-care dimension. ** ICC was 2-way random, single measures, and absolute agreement.
4.2.7 Coefficient of variation
The coefficient of variation (CV) was calculated from all respondents
(excluded the respondents who reported perfect health), since the majority of
respondents of this study had mild condition and the sample size was small, Figure 4.11
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 97
demonstrated that the preference score produced by 5L has lower variation than 3L
although standard deviations were similar.
Figure 4.11 Mean, standard deviation, and coefficient variation of preference score
4.2.8 Patient preferences
Thirty-six percent of respondents indicated that the 5L was easier to
answer than the 3L while 33% of respondents indicated that there was no difference
between the 5L and the 3L. In terms of reflecting health status, most respondents
(63%) agreed that the 5L was better in describing their health states while 29%
indicated that both versions were similar.
Session 4.3: Comparison of economic evaluation results using
preference score derived from the 3L and the 5L
This session showed the economic evaluation results generated from 3
different value sets (Thai population-based 3L value set, 5L value set generated from
mapping, and Thai population-based value set of the 5L generated in session 1). The
results of both DM type 1 and DM type 2 were similar. The mean preference scores
produced from Thai population-based 3L value set was similar to those of the 5L value
set generated from mapping, however both of them were lower than the population-
based preferences of the 5L generated from this study. Thus the QALY calculated
Juntana Pattanaphesaj Results / 98
from the 5L value set derived from this study was higher than those generated from the
Thai population-based 3L and the 5L value sets from mapping, as shown in Table 4.24.
The ICER, which was calculated by incremental cost divided by
incremental QALY, was derived using the probabilistic determination by fixing all
parameters accept for utility. The results showed that the scores generated from this
study yielded lower ICER than those generated from the Thai population-based 3L
value set and those generated from the 5L by mapping technique. Thus it implied that
the economic evaluation model which used utilities generated by Thai 5L population-
based value set was likely to be more cost-effective than those used the utilities
generated by the Thai population-based 3L value set and the 5L value set generated
from mapping.
Table 4.24 Economic evaluation results generated from 3 different value sets
Tariffs Mean utility
Standard error
∆ cost ∆ QALY ICER *
(Baht/QALY gained)
DM1 - Population-based 3L
value set 0.8323 0.0202 276,934 5.55 49,898
- 5L value set (mapping) 0.8335 0.0176 276,372 5.55 49,835 - Population-based 5L
value set 0.9597 0.0051 276,278 6.38 43,322
DM2 - Population-based 3L
value set 0.7308 0.0222 123,968 0.29 431,484
- 5L value set (mapping) 0.7366 0.0219 119,806 0.29 414,965 - Population-based 5L
value set 0.8991 0.0130 124,387 0.36 349,918
* ICER = ∆ cost / ∆ QALY
The sensitivity analysis of economic modeling was presented in cost-
effectiveness plane which was a scatter plot of the bootstrapped incremental costs and
effect (QALY) pairs. It was used to illustrate the uncertainty surrounding the expected
costs and expected effects. Since the utility produced from 5L value sets (mapping)
was similar to population-based 3L value set, only the results from population-based
3L value set and population-based 5L value set were demonstrated. The dots were
plotted by randomly input parameters into the model one thousand and five hundred
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 99
times repeatedly. In order to clearly demonstrate the effect of utility produced by 2
different value sets, utility values were varied to the possible range while all other
variables were fixed.
Figure 4.12 shows the cost-effectiveness plane which the solid line
represents the threshold for Thailand (160,000 Baht per QALY). The graph showed
that the population-based 5L value set produced less steep slope than population-based
3L value set. Thus using population-based 5L value set in economic model yielded
better value for money than population-based 3L value set.
Figure 4.12 Cost-effectiveness plane comparing SMBG group to no SMBG group for
DM type 1 & 2
3L
5L
3L 5L
Threshold =160,000 Baht/QALY
Threshold =160,000 Baht/QALY
Incremental cost
Incremental QALY
Incremental cost
Incremental QALY
DM1
DM2
Juntana Pattanaphesaj Results / 100
Figure 4.13 shows head-to-head comparison of preference score of
comparable health states between 3L and 5L. It is clearly that the preference score
produces by 5L is higher than 3L for all comparable health states. This evidence
confirm that using Thai 5L tariff in economic model is likely to yield better value for
money than Thai 3L tariff.
Figure 4.13 Preference score for comparable health states of the 3L and the 5L
The cost-effectiveness acceptability curve (CEAC) was used to summarize
the information on uncertainty in cost-effectiveness analysis. It was drawn by plotting
the proportion of the costs and QALY pairs that were cost-effective for the maximum
acceptable ceiling ratio. Figure 4.14 and 4.15 shows the probability that the intervention
group (SMBG) or reference group (no SMBG) was cost-effective. The CEAC generated
by population-based 5L value set showed steeper than population-based 3L value set,
especially among DM type 1. It implied that CEAC generated by population-based 5L
value set offered more certainty than population-based 3L value set. As a result, it was
easier for policy makers to make decision with the population-based 5L value set.
Preference score
Health state
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 101
Figure 4.14 Cost-effectiveness acceptability curve for DM type 1 group
3L
5L
Juntana Pattanaphesaj Results / 102
Figure 4.15 Cost-effectiveness acceptability curve for DM type 2 group
3L
5L
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 103
CHAPTER V
DISCUSSIONS
Session 5.1: Development of the Thai population-based preference
scores for the 5L
5.1.1 Sampling and characteristic of sample
To ensure representativeness for Thai population, various steps were
employed. The study areas were selected using probabilistic sampling by the NSO.
This technique reduced both systematic and sampling bias as every area in Thailand
had an equal opportunity for selection. Regarding individual selection, quota sampling
by age and sex was undertaken in order to obtain the sample that similar to Thai
population structure. Otherwise, most of respondents may be elderly because most of
young labors migrated to work in the big city. However, the sample in our study may
still not be a representative of Thai population due to non-response of some
respondents or oversampling of certain population. Thus, prior to analyze, the data was
weighted by the NSO to correct for non-representativeness. The sample of this study
was therefore representing the Thai population.
Regarding health status of the sample, most of them were healthy as mean
VAS score was 83, 30% reported perfect health, and less than 10% reported moderate
problem or severe problems. However, it was noticed that 44% of sample reported
having slight pain/discomfort (level 2). This may due to the fact that most of the
respondents (54%) worked in agriculture sector or unskilled labor which might cause
them fatigue. Compared with the previous valuation study for 3L in Thailand (11, 12),
63% of respondent also reported moderate pain/discomfort (level 2). This was also
consistent with the responses in session 2, which found that 46% of respondent
reported slight problems for pain in the 5L.
Juntana Pattanaphesaj Discussions / 104
5.1.2 Methodology
5.1.2.1 TTO valuation
The TTO valuation used to elicit population-based preferences
for the 5L was different from the 3L in many ways. Firstly, the 3L valuation in
Thailand (11, 12) used TTO board which was adapted from the props used in the
MVH protocol (94) to elicit health preference. One side of the TTO board was used
for valuation health states that the respondents considered as better than death while
the other side was used for health states viewed as worse than death. This tool was
difficult and inconvenience for use as interviewers had to change the year in full health
by themselves. In addition, this technique was more prone to human error. On the
other hand, the 5L valuation followed the EQ-VT protocol (73) which was a face-to-
face setting and used a laptop installed with the Thai version of ‘EQ-VT offline’ to
interview. The advantage of using digital technology in data collection was that it was
less burden to the interviewers. The missing values were also absent as the computer
program did not allowed to skip the questions. Nevertheless, a few missing records
occurred due to technical problems in software and/or internet connection during
upload the data to central server and update the software to the new version. Another
advantage was that it can reduce burden and error in data entry process. In addition,
using a touch screen laptop in data collecting could help the respondents enjoy
interview in the almost 1-hour interviewing per respondent. In the Thai context, the
respondents had various characteristics. Most respondents can use a touch screen
laptop skillfully. However, some respondents scared to touch it. In this case, the
interviewer demonstrated how to touch the screen at the earliest stage of interview and
then let the respondents touch the screen for the answers by themselves.
Secondly, in this study (5L valuation) the value of worse than
dead health state was elicited using lead-time TTO (51) which allowed both better than
death and worse than death health state to be valued with the same valuation technique
and concept. The purpose of adding lead time in full health to both two alternatives
(life A and life B) was to keep the utility calculation to be the same. It allowed
respondents to trade their lead time to avoid worse than death health state. Since
different lead time duration yielded different results, the appropriate length of lead
time was studies in many countries (51, 102, 103). A 10-year of lead time used in this
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 105
study would make possible comparison with the MVH value set elicited in the same
duration. On the other hand, the 3L valuation used the conventional TTO valuation
(MVH protocol) for worse than death state. In the conventional TTO valuation,
different trade-offs tasks and also different concept in utility calculation was used for
better than death and worse than death health state. As a results, MVH elicitation
produced an extreme negative values and needed transformation to bound the negative
value to -1 (104).
According to the EQ-VT protocol, the number of health state
used to elicit preference for 5L was 86 states out of the total 3,125 states, accounted
for 2.75%. While MVH protocol for the 3L valuation offered 42 health states (17%)
out of 243 states. The pilot study (74) demonstrated that these 86 selected health states
for 5L valuation were sufficient and efficient to predict 3,125 health states. However,
it was found that some selected health state were difficult to imagine for Thai, for
example state 51152 which the one was unable to walk but he/she had no problem in
self-care and usual activities. The selected health states which were difficult to
imagine may distort the TTO score and lead to inconsistent coefficients in the model.
The health states used to elicit preference should not just cover the full range of
severity; but also they should close to the reality.
Regarding the method used to determine the logical
inconsistency among 10 health states for each respondent in the TTO valuation, Devlin
et al. determined the logical inconsistency by counting the number of pair
inconsistency pairs (75) according to the methods proposed by Dolan and Kind (77).
In Devlin et al, the respondents were divided into 8 groups according to the degree of
inconsistency. It was found that the actual mean scores of each subgroup were not
statistically different from the subgroup with zero inconsistency. Thus Devlin
recommended estimating the tariff from the full sample. Differently, this study
determined the inconsistency by calculating the magnitude of inconsistency. The
inconsistency existed when a higher value was assigned to a worse health state (more
severity level). This method was different from Dolan and Kind (77) method.
According to Dolan and Kind (77), state A was logically worse than state B when all
level of state A was equal or worse than state B. If some levels of state A were worse
while some were better than state B, these two states can’t be compared. In addition,
Juntana Pattanaphesaj Discussions / 106
by using magnitude of inconsistency to exclude low quality data (41 respondents) we
found that the cross validation and fit statistics in the regression model were improved.
This study employed multilevel modeling for estimating Thai
value sets because the data set was panel data which 1 respondent were 10 times
repeated measured. Heteroscedasticity is common for panel data (81). The multilevel
modeling is more flexible with this kind of data as it can represent regression models
where the residual variance is not constant (81). The OLS was not appropriate for our
data because it might produce bias and misleading parameter estimates. This is
because one important assumption of OLS is homoscedasticity of residuals. Thus
heteroscedasticity of our data can be very problematic with OLS.
This study did not found the inconsistency of the coefficients
in regression model produced from TTO valuation. This may be due to the high
quality in data collection process. In our study, all 6 interviewers were well-trained.
The pilot study also allowed them to practice with the total of 100 respondents in the 8
settings. Importantly, regularly data uploading to a central server of the EuroQol
group and the quality control tool, which was an excel program developed by the
EuroQol group, allowed principal investigator (PI) to know all details of interviewing
real time, i.e. interviewing time, logical inconsistence responses. As the result, the PI
can notice the problems at the early step, which can help improving the quality of data
collection in due time. Nevertheless, the EQ-VT could not yet promptly identify the
logical inconsistency during interviewing.
5.1.2.2 DCE valuation
At the present, there is no standard protocol for calculating the
data from DCE valuation. Previous studies showed the feasible (105, 106) and
advantages of DCE over TTO valuation (90). The DCE allowed fewer data exclusions.
This is important when the representativeness of population was concerned. On the
other hand, TTO studies needs well data management as some respondents might not
understand the task. However, the main difficulty in analyzing DCE data was that the
values generated from the regression model were on arbitrary scale which needed
some methodology to anchor the values derived from DCE on the QALY scale (74,
91, 105). Two strategies for anchoring were employed (74, 91). The first method used
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 107
the worst value estimated from lead-time TTO model (55555) to anchor for the worst
state of DCE. The second method anchored on death state, which need more DCE
question, i.e. whether life A was WTD and whether life B was WTD. The DCE
questions from EQ-VT protocol had no such questions, so only first method can be
undertaken.
The inconsistent coefficients found in this study was similar to
those found in the Spanish (91) and English report (this conference material was not
allowed to cite). Although DCE valuation was claimed as easier than TTO valuation
(90, 106) which the respondents may not understand the task, difficulty of making
choices on prefer health state was found. Feedback questions reported that 81% of
Thai respondent understood the DCE task, however 71% of Thai respondents found
that it was difficult to choose preferred life on the DCE task. On the other hand, 62%
of Thai respondent indicated that they understood the TTO task, and 62% reported that
it was difficult to decide on the exact point where two lives were about the same. This
may implied that DCE task was more difficult on deciding the preferred life than TTO
task for Thai respondents. This finding was similar to Spanish report (91) which many
respondents did comment on the difficulty of making choices between health states.
This might because DCE task needed the respondents to imagine and compare 2 health
states with problems at the same time, while it might easier for Thai respondents to
imagine 1 health state with problems compared to perfect health in TTO task. During
interview, it was noticed that some respondents made decision on DCE task very
quick, i.e. they considered only whether the problems on a particular dimension
existed or not; and some of them did not consider the severity level of dimension.
5.1.3 Preference score
This study suggested the 5L algorithm without a constant or interaction
terms to estimate preference score for Thai population. It is interesting to compare the
5L value sets with those of other countries, unfortunately, the official publications on
the 5L valuation were not found. Comparing with the interim scoring generated by the
EuroQol group using mapping method (33, 34), the evidences demonstrated that the
predicted score of Thai population-based value sets for was statistically significant
different from the interim scoring. This might be due to the fact that the sample of
Juntana Pattanaphesaj Discussions / 108
interim scoring study was derived from 6 western countries although the Thai value set
of 3L was used to estimate interim value sets. Different context (i.e. culture, belief,
health behavior) between western and eastern could limit the validity of the value sets
estimated. One limitation of the mapping method was the floor effect as the worst state
of mapping method was anchored with the worst state of Thai value set for 3L (state
33333). As the result, it did not allow the value of state 55555 to be worse than state
33333.
The algorithm of this study was consistent with the Thai algorithm for the
3L (11, 12), and Brazilian study (107) which the smallest disutility was found in the
slightly (level 2) anxiety/depression dimension (0.032 in both the 3L and 5L, 0.062 for
Brazilian study). The largest disutility was found in the extreme problem (level 5) in
mobility dimension (0.432 for the 3L, 0.307 for the 5L, 0.40 for Brazilian study). The
greatest disutility was conformed to the opinion of the respondents that ‘unable to walk
about’ was the worst state. Since the 5L is more sensitive to mobility dimension, health
conditions which affect mobility could make great impact on the change of preference
score. However, considering health profile by each dimension is still needed (20).
With regard to the preference score of the 5L compared with 3L, the
second best state was 11112 (0.766 for the 3L, and 0.968 for the 5L). The score of the
worst health state of 3L and 5L was -0.454 and -0.283, respectively. The range of
value of the 5L was narrower than the 3L. It implied that using the 5L to measure
utility for the treatment of very severe conditions may resulting in lower QALY gains
than using the 3L. The differences in the range of score might mainly result from
different protocol used to elicit preference; especially the method used to elicit and
calculate WTD state. The current method used EQ-VT protocol which offered 10
years of full health as lead time for WTD state in order to keep the utility elicitation
and calculation to be the same as BTD state. However, the 3L valuation used MVH
protocol which produced an extreme negative values and needed transformation to
bound the negative value to -1 (104).
5.1.4 Limitations and future research
Sampling bias could occur when individuals were selected to be
interviewed. Although stratified-three stage sampling was undertaken, NSO employed
probability sampling for only first (provinces) and second stage (EAs). The third stage
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 109
(individuals) was selected by quota sampling which was non-probability sampling.
The quota sampling allowed the number of eligible participant with age and sex to
conform to the Thai population structure. The coordinators (mainly were the officer
from primary care unit) recruited participants according to the number of sex and age
range specified by the researcher. With this method, some members of the population
were less likely to be included than others resulting in a biased sample. Due to limitation
in budget and time, this study could not employ random sampling for individual.
The magnitude of inconsistency which was used as an indicator to
consider low quality data for 10 health states of TTO task may be crude and not
conform to Dolan and Kind method (76). These criteria determined the inconsistency
between health states from the level of severity. Nevertheless, the evidence
demonstrated that when the respondents with too high magnitude of inconsistency
(41 respondents) were excluded, the cross validation and fit statistics in the regression
model were improved.
The valuation study for the 5L using EQ-VT protocol consumed time and
budget as it needed a laptop for face-to-face interviewing. The research team also went
all over the country in order to get the data that could be a representative of Thai
population. The cost of travel, accommodation, and salary was substantial high.
Without financial support from funder, the valuation study was difficult to complete. It
also need well preparation, management, and good coordination with the community
leader or the officer at primary care unit.
The inconsistent coefficient found from DCE valuation needs the further
research to explore more about the cause of inconsistence and how to deal with these
problems.
Session 5.2: Testing the measurement properties of the Thai version
of the 5L compared to the 3L
This report is the first study in Thailand that assesses the measurement
properties of the 5L and compares it with the 3L. Similar to previous studies (22, 24,
26, 27, 29, 30), self-care showed the highest percentage of ceiling effect in both the 3L
and 5L. On the other hand, the lowest ceiling was found in pain/discomfort (44%) (24,
Juntana Pattanaphesaj Discussions / 110
27, 29). Similar to the previous studies (22, 24-27, 29, 30), the proportion of the
ceiling effect in our study was lower in the 5L (29%) compared with the 3L (33%).
However, in the previous studies that involved patients with a variety of severity
higher reduction in ceiling effect of the 5L (3-17%) was identified (22, 24, 27, 29).
The smaller reduction in ceiling effects found in our study may be due to the fact that
our respondents were likely to be healthy (median VAS score = 78).
In each dimension, more than half of the responses were in level 1 (no
problem) for both the 3L and 5L. Among the respondents with health problems, it was
clearly shown that the 5L can present more details of severity than the 3L. While we
found that the majority of level 1 in the 3L still remained at level 1 in the 5L (85-98%),
about 2% (self-care) to 15% (in pain/discomfort) were upgraded to level 2 in the 5L.
The redistribution from 3L-level 2 (some problems) to 5L-level 2 (slight problems)
was also high, ranging from 69% for mobility to 100% for self-care, and the
redistribution from 3L-level 2 to 5L-level 3, ranging from 9% for usual activities to
22% for mobility. This finding supports the inclusion of the slight problems (level 2)
in the 5L. However, no supportive evidence of the inclusion of severe problems (level
4) in the 5L was found in our study as no 3L-level 3 responses were reported. This
may also be due to the fact that our respondents were likely to be healthy.
No inconsistent responses were found in our study. This indicates that our
respondents were able to consistently answer both the 3L and 5L. This is similar to
previous studies (21, 24, 26, 27, 29, 30) which showed that inconsistency was quite
low, ranging from 0.5% to 3.5%. However, the consistent responses may be due to the
low number of the sample size and the characteristics of our sample - educated and
healthy diabetic patients. In addition, even when the respondents completed the
questionnaires themselves, they were well-advised by trained staff.
The measurement of reliability and agreement is important in health
classification as it reveals the amount of errors of the measurement. The concept of
‘reliability’ differs from ‘agreement’ in that reliability is a relative measure which is
the ratio of variability between subjects to the total variability of all measurement in
the sample (100). Thus, it reflects the ability of an instrument to differentiate between
subjects. In contrast, an agreement is an absolute measure which is the degree to which
responses are identical. Cohen’s weighted kappa is often used in assessing test-retest
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 111
reliability of ordinal instruments as it takes the chance agreement into account.
However, the lack of variance in the data set meant that the kappa could not be
calculated so it was necessary to rely on the percentage agreement values. However, it
should be cautioned that the percentage agreement may give higher reproducibility
figures than the kappa coefficient (99). Unlike previous studies (27, 29, 30), our results
of the test-retest reliability/agreement showed that the 5L was slightly less
reproducible than the 3L in all dimensions. This is probably due to the fact that the
average time interval between the two tests was too long (3 weeks), and therefore the
different environment of repeated measurements may have impacted the answers
(100). In addition, the 5L’s ability to better capture small changes in health status may
have produced slightly less reproducibility. It should also be noted that the respondents
completed the first set of questionnaires by themselves at the hospital and repeated it
again at their home for the second session.
Convergent validity was evaluated by correlations between the EQ-5D and
SF-36v2 dimensions. Both the 3L and 5L presented an acceptable degree of
association and similar correlation pattern with the SF-36v2 in some pairs of
dimension, i.e. mobility versus physical functioning; pain/discomfort versus bodily
pain; and anxiety/depression versus mental health. The findings were similar to the
study by Kimman et al (67) that assessed the relationship of the 3L with the SF-36v2
among the occupational population in Thailand.
Similar to previous studies (21, 22, 26), absolute informativity ( )
increased in all dimensions for the 5L while in terms of the evenness of distribution
evaluated by Shannon’s Evenness index ( ), the 5L was comparable to the 3L. While
the maximum value of for the 5L is 2.32, our values ranged from 0.21 to 1.40
which was lower than the findings from Pickard et al (22) (0.84-2.00) and Janssen et al
(21) (2.05-2.26). With the maximum value of set at 1.00, our values ranged from
0.09 to 0.60 which was also lower than Pickard et al (22) (0.36-0.86) and Janssen et al
(21) (0.88-0.97). The lower and values found in our study may have risen from
the mild characteristic of our sample since the extreme problems (3L-level 3 and 5L-
level 5) were not reported. As the result, the levels of responses of the EQ-5D were
used ineffectively, resulting in low and values.
Juntana Pattanaphesaj Discussions / 112
Although 5L offers 5 options to choose which seem to be more difficult to
make decision compared to 3L which offers only 3 options, the findings revealed that
Thai respondents preferred 5L than 3L as it was easier to use. Some of them expressed
that 3 options of 3L were crude, for example level 1 (no problem) was too good and
level 2 (moderate problems) was too bad to reflect their health. They would like to
choose in-between level (between level 1 and 2). Unfortunately no that option in 3L,
so it was hard to decide between level 1 and level 2 for 3L. Five options of 5L made
decision easier as the 5 choices offered were exactly reflect their health.
In our study, diabetic mellitus was chosen as it is a common chronic
disease that substantial affects quality of life (108, 109). Additionally, diabetes was
ranked as third and eighth in terms of Disability Adjusted Life Year (DALY) loss in
Thai women and men, respectively (96).We included patients with no complications in
our study to ensure that the health status will be stable enough in order to test the test-
retest reliability/agreement. However, given the mild condition of our sample, we
were unable to assess the redistribution of answers from the 3L-level 3 to the 5L.
Further studies should be conducted for patients with a variety of severe health
problems. In addition, it should be noted that the general findings of different groups
of patients should be made with caution as the pattern of responses may differ by
disease characteristics (14).
Regarding time to complete EQ-5D version, the methodology of this study did
not design to compare time to complete 3L and 5L. However the average time to complete 2
versions of EQ-5D (3L and 5L) was 4 minutes. Thus, the time used to complete 1 version of
EQ-5D by themselves for Thai respondents may less than 4 minutes.
One more limitation is that the 5L utility score was obtained from the interim
mapping generated by the EuroQol group since the valuation study for the 5L in Thailand
has not been completed yet. Although the calculation was based on the Thai 3L value sets,
the results of the mapping may deviate compared to the actual responses (66).
Session 5.3: Comparison of economic evaluation results using
preference score derived from the 3L and the 5L
This study showed the results from only one economic modeling which
used the preference score derived from the Thai 3L value set and the Thai 5L value
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 113
set. Thus generalizability of these results should be made with cautions. This study
found that among comparable health states between 3L and 5L, all preference scores
generated from the 5L value set was higher than those of the 3L value set resulting in
higher QALY. Thus ICER/QALY calculated using the Thai 5L value set yielded more
cost-effective than the Thai 3L value set. Nevertheless in terms of policy decision
making, only ICERs may not enough to decide which intervention is worth as ICERs
contained some degree of uncertainty from costs and effect variables (110). The
CEAC could help policy maker to better decide as it shows the information on
uncertainty of ICERs estimated. The CEAC from this study showed that the
uncertainty in the results was lower when the utilities obtained from the 5L value set
were employed. Thus using the Thai 5L value set in economic modelling could
produce more certainty of cost-effective evidence and could support better policy
decision making for Thailand context. Further studies comparing economic evaluation
results using preference score from 3L and 5L in different models, diseases, and
population are also needed.
Juntana Pattanaphesaj Conclusions / 114
CHAPTER VI
CONCLUSIONS
The preference score of the 5L for Thai population was estimated using the
lead-time TTO method. The random effect model with only main effect was selected.
The use of country-specific value sets is recommended since the evidence suggested
that the preference score of the Thai 5L value sets was significant different from the
interim scoring generated by the EuroQol group. The DCE valuation generated
inconsistent coefficient in the regression model, indicating the need to further examine
the cause of inconsistency and how to deal with these problems.
In term of measurement properties, this study suggests that the 5L was
better than the 3L in terms of greater distribution, less ceiling effect, more
informativity, more discriminatory power, more reliable for index score, and more
patient preferences. The 5L also showed reasonable convergent validity. In terms of
economic evaluation, the evidences indicated that using population-based 5L value set
in economic model yielded better value for money than population-based 3L value set.
In addition, we found that using population-based 5L value set produced less
uncertainty in cost-effective information compared to the Thai 3L value set.
Thus, the 5L should be recommended as a preferred health-related quality
of life questionnaire in Thailand.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 115
REFERENCES
1. Teerawattananon Y, Tantivess S, Yothasamut J, Kingkaew P, Chaisiri K.
Historical development of health technology assessment in Thailand. Int J
Technol Assess Health Care. 2009;25 Suppl 1:241-52.
2. Mohara A, Youngkong S, Velasco RP, Werayingyong P, Pachanee K,
Prakongsai P, et al. Using health technology assessment for informing
coverage decisions in Thailand. J compar eff resource. 2012;1(2):137-46.
3. อษา ฉายเกลดแกว, ยศ ตระวฒนานนท, สรพร คงพทยาชย, เนต สขสมบรณ. คมอการประเมนเทคโนโลยดานสขภาพส าหรบประเทศไทย. นนทบร: เดอะกราฟโกซสเตมส; 2552.
4. อษา ฉายเกลดแกว, ยศ ตระวฒนานนท. คมอการประเมนเทคโนโลยดานสขภาพส าหรบประเทศไทย ฉบบท 2 พ.ศ. 2556. นนทบร: โรงพมพวชรนทร พ.พ.; 2557.
5. Berger ML, Bingefors K, Hedblom EC, Pashos CL, Torrance GW. Health care
cost, quality, and outcomes: ISPOR book of terms. New Jersey:
International Society for Pharmacoeconomics and Outcomes Research
(ISPOR); 2003.
6. สพล ลมวฒนานนท. การวเคราะหความไวส าหรบความไมแนนอนของตวแปรทใชในรปแบบจ าลอง. ใน อษา ฉายเกลดแกว, ยศ ตระวฒนานนท บรรณาธการ. คมอการประเมนเทคโนโลยดานสขภาพส าหรบประเทศไทย ฉบบท 2 พศ 2556. นนทบร: วชรนทร พ.พ.; 2557.
7. Rawlins MD, Culyer AJ. National Institute for Clinical Excellence and its value
judgments. BMJ. 2004;329(7459):224-7.
8. Weinstein MC, Siegel JE, Gold MR, Kamlet MS, Russell LB. Recommendations
of the Panel on Cost-effectiveness in Health and Medicine. JAMA.
1996;276(15):1253-8.
9. Sakthong P. Measurement of clinical-effect: utility. J Med Assoc Thai. 2008;91
Suppl 2:S43-52.
Juntana Pattanaphesaj References / 116
10. Williams A. The EuroQol Instrument. In: Kind P, Brooks R, Rabin R, editors.
EQ-5D concepts and methods : a developmental history. Dordrecht:
Springer; 2005. p. 1-17.
11. Tongsiri S. The Thai population-based preference scores for EQ-5D health
states. Nonthaburi: Ministry of Public Health, 2009.
12. Tongsiri S, Cairns J. Estimating Population-Based Values for EQ-5D Health
States in Thailand. Value Health. 2011;14(8):1142-5.
13. Brazier J, Jones N, Kind P. Testing the validity of the Euroqol and comparing it
with the SF-36 health survey questionnaire. Qual Life Res. 1993;2(3):169-
80.
14. Johnson JA, Coons SJ. Comparison of the EQ-5D and SF-12 in an adult US
sample. Qual Life Res. 1998;7(2):155-66.
15. Sullivan PW, Lawrence WF, Ghushchyan V. A national catalog of preference-
based scores for chronic conditions in the United States. Med Care.
2005;43(7):736-49.
16. Badia X, Schiaffino A, Alonso J, Herdman M. Using the EuroQoI 5-D in the
Catalan general population: feasibility and construct validity. Qual Life
Res. 1998;7(4):311-22.
17. Kaarlola A, Pettila V, Kekki P. Performance of two measures of general health-
related quality of life, the EQ-5D and the RAND-36 among critically ill
patients. Intensive care medicine. 2004;30(12):2245-52.
18. Houle C, Berthelot J-M. A Head-to-Head Comparison of the Health Utilities
Mark 3 and the EQ-5D for the Population Living in Private Households in
Canada. Quality of Life Newsletter. 2000;24:5-6.
19. Herdman M, Gudex C, Lloyd A, Janssen M, Kind P, Parkin D, et al.
Development and preliminary testing of the new five-level version of EQ-
5D (EQ-5D-5L). Qual Life Res. 2011.
20. Oemar M, Janssen B. EQ-5D-5L User Guide. Rotterdam: EuroQol Group; 2013.
21. Janssen MF, Birnie E, Haagsma JA, Bonsel GJ. Comparing the standard EQ-5D
three-level system with a five-level version. Value Health.
2008;11(2):275-84.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 117
22. Pickard AS, De Leon MC, Kohlmann T, Cella D, Rosenbloom S. Psychometric
comparison of the standard EQ-5D to a 5 level version in cancer patients.
Med Care. 2007;45(3):259-63.
23. EuroQol group. EQ-5D-5L [cited 16 Apr 2014]. Available from:
http://www.euroqol.org/eq-5d-products/eq-5d-5l.html.
24. Janssen MF, Pickard AS, Golicki D, Gudex C, Niewada M, Scalone L, et al.
Measurement properties of the EQ-5D-5L compared to the EQ-5D-3L
across eight patient groups: a multi-country study. Qual Life Res.
2013;22(7):1717-27.
25. Craig BM, Pickard AS, Lubetkin EI. Health problems are more common, but
less severe when measured using newer EQ-5D versions. J Clin
Epidemiol. 2014;67(1):93-9.
26. Scalone L, Ciampichini R, Fagiuoli S, Gardini I, Fusco F, Gaeta L, et al.
Comparing the performance of the standard EQ-5D 3L with the new
version EQ-5D 5L in patients with chronic hepatic diseases. Qual Life
Res. 2013;22(7):1707-16.
27. Kim SH, Kim HJ, Lee SI, Jo MW. Comparing the psychometric properties of the
EQ-5D-3L and EQ-5D-5L in cancer patients in Korea. Qual Life Res.
2012;21(6):1065-73.
28. Tran BX, Ohinmaa A, Nguyen LT. Quality of life profile and psychometric
properties of the EQ-5D-5L in HIV/AIDS patients. Health Qual Life
Outcomes. 2012;10:132.
29. Kim TH, Jo MW, Lee SI, Kim SH, Chung SM. Psychometric properties of the
EQ-5D-5L in the general population of South Korea. Qual Life Res.
2013;22(8):2245-53.
30. Jia YX, Cui FQ, Li L, Zhang DL, Zhang GM, Wang FZ, et al. Comparison
between the EQ-5D-5L and the EQ-5D-3L in patients with hepatitis B.
Qual Life Res. 2014;March.
31. Cabasés JM, Errea M, Hernández-Arenaz I. Comparing the psychometric
properties of the EQ-5D-5L between mental and somatic chronic patients
populations. Spain: Department of Economics, Public University of
Navarra, 2013.
Juntana Pattanaphesaj References / 118
32. EuroQol group. EQ-5D-5L Value Sets2012 [cited 9 Oct 2012. Available from:
http://www.euroqol.org/fileadmin/user_upload/Documenten/Excel/Crossw
alk_5L/EQ-5D-5L_Crosswalk_Index_Value_Calculator.xls.
33. EuroQol group. Interim scoring for the EQ-5D-5L : mapping the EQ-5D-5L to
EQ-5D-3L value sets 2011 [11 Jan 2012]. Available from:
http://www.euroqol.org/eq-5d/valuation-of-eq-5d/eq-5d-5l-crosswalk-
value-sets.html.
34. van Hout B, Janssen MF, Feng YS, Kohlmann T, Busschbach J, Golicki D, et al.
Interim Scoring for the EQ-5D-5L: Mapping the EQ-5D-5L to EQ-5D-3L
Value Sets. Value Health. 2012;15(5):708-15.
35. Fenwick E, Byford S. A guide to cost-effectiveness acceptability curves. The
British journal of psychiatry : the journal of mental science. 2005;187:106-
8.
36. Kobelt G. Health economics: an introduction to economic evaluation. London:
BSC Print Ltd; 2002.
37. Feeny D. A utility approach to the assessment of health-related quality of life.
Med Care. 2000;38(9 Suppl):II151-4.
38. Sinnott PL, Joyce VR, Barnett PG. Guidebook preference measurement in
economic analysis. California: Health Economics Resource Center
(HERC); 2007.
39. Polygenis D, Frame D, Long KH, McCormick J. ISPOR taxonomy of patient
registries: classification, characteristics and terms. International Society
for Pharmacoeconomics and Outcomes Research (ISPOR): New Jersy;
2014.
40. Felce D, Perry J. Quality of life: its definition and measurement. Res Dev
Disabil. 1995;16(1):51-74.
41. Mehrens WA, Lehmann IJ. Measurement and Evaluation in Education and
Psychology. New York: Holt, Rinehart and Winston; 1984.
42. Warren C, McGraw AP, Boven LV. Values and preferences_defining preference
construction. WIREs Cognitive Science [Internet]. 2010 17 Apr 2010].
Available from: http://leeds-faculty.colorado.edu/mcgrawp/pdfwarren.
mcgraw.vanboven.inpress.pdf.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 119
43. Wolters Kluwer Health. Pharmacoeconomics guide chart. Pharmacoeconomics.
2005.
44. WHOQOL Group. The development of the World Health Organization Quality
of Life assessment instrument (the WHOQOL). In: J O, W K, editors.
Quality of life assessment : international perspectives. Berlin: Springer-
Verlag; 1994. p. 41-60.
45. Guyatt GH, Feeny DH, Patrick DL. Measuring health-related quality of life. Ann
Intern Med. 1993;118(8):622-9.
46. Wilson IB, Cleary PD. Linking clinical variables with health-related quality of
life. A conceptual model of patient outcomes. JAMA. 1995;273(1):59-65.
47. Coons SJ, Rao S, Keininger DL, Hays RD. A comparative review of generic
quality-of-life instruments. Pharmacoeconomics. 2000;17(1):13-35.
48. Green C, Brazier J, Deverill M. Valuing health-related quality of life. A review
of health state valuation techniques. Pharmacoeconomics. 2000;17(2):151-
65.
49. Kind P, Brooks R, Rabin R. EQ-5D concepts and methods : a developmental
history. Dordrecht: Springer; 2005.
50. Torrance GW, Thomas WH, Sackett DL. A utility maximization model for
evaluation of health care programs. Health Serv Res. 1972;7(2):118-33.
51. Devlin NJ, Tsuchiya A, Buckingham K, Tilling C. A uniform time trade off
method for states better and worse than dead: feasibility study of the 'lead
time' approach. Health Econ. 2011;20(3):348-61.
52. Terwee CB, Bot SD, de Boer MR, van der Windt DA, Knol DL, Dekker J, et al.
Quality criteria were proposed for measurement properties of health status
questionnaires. J Clin Epidemiol. 2007;60(1):34-42.
53. Fleiss JL, Levin B, Paik MC. The Measurement of Interrater Agreement.
Statistical Methods for Rates and Proportions. John Wiley & Sons, Inc.:
Hoboken, NJ, USA; 2004.
54. Hopkins KD, Stanley JC, Hopkins BR. Educational and psychological
measurement and evaluation. 6th ed. New Jersey: Prentice Hall; 1981.
55. Juniper EF, Guyatt GH, Jaeschke R. How to develop and validate a new quality
of life instrument. In: Spilker B, editor. Quality of Life and
Juntana Pattanaphesaj References / 120
Pharmacoeconomics in Clinical Trials. Philadelphia: Lippincott-Raven
Publishers; 1995. p. 49-56.
56. Bas Janssen MF, Birnie E, Bonsel GJ. Evaluating the discriminatory power of
EQ-5D, HUI2 and HUI3 in a US general population survey using
Shannon's indices. Qual Life Res. 2007;16(5):895-904.
57. Institute of Digital Research and Education (idre). What is the coefficient of
variation? [cited 1 July 2014]. Available from:
http://www.ats.ucla.edu/stat/mult_pkg/faq/general/coefficient_of_variation
.htm.
58. Medical Outcomes Trust. SF-36 [cited 14 Jan 2013]. Available from:
http://www.sf-36.org.
59. Brazier JE, Harper R, Jones NM, O'Cathain A, Thomas KJ, Usherwood T, et al.
Validating the SF-36 health survey questionnaire: new outcome measure
for primary care. BMJ. 1992;305(6846):160-4.
60. Essink-Bot ML, Krabbe PF, Bonsel GJ, Aaronson NK. An empirical comparison
of four generic health status measures. The Nottingham Health Profile, the
Medical Outcomes Study 36-item Short-Form Health Survey, the
COOP/WONCA charts, and the EuroQol instrument. Med Care.
1997;35(5):522-37.
61. Leurmarnkul W, Meetam P. Properties Testing of the Retranslated SF-36 (Thai
Version). Thai J Pharm Sci. 2005;29(1-2):69-88.
62. Cheung K, Oemar M, Oppe M, Rabin R. User guide : basic information on how
to use EQ-5D: EuroQol group; 2010.
63. EuroQol group. How to obtain EQ-5D 2012 [cited 16 Jan 2013]. Available from:
http://www.euroqol.org.
64. Hurst NP, Kind P, Ruta D, Hunter M, Stubbings A. Measuring health-related
quality of life in rheumatoid arthritis: validity, responsiveness and
reliability of EuroQol (EQ-5D). Br J Rheumatol. 1997;36(5):551-9.
65. Dorman P, Slattery J, Farrell B, Dennis M, Sandercock P. Qualitative
comparison of the reliability of health status assessments with the EuroQol
and SF-36 questionnaires after stroke. United Kingdom Collaborators in
the International Stroke Trial. Stroke. 1998;29(1):63-8.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 121
66. Sakthong P, Charoenvisuthiwongs R, Shabunthom R. A comparison of EQ-5D
index scores using the UK, US, and Japan preference weights in a Thai
sample with type 2 diabetes. Health Qual Life Outcomes. 2008;6:71.
67. Kimman M, Vathesatogkit P, Woodward M, Tai ES, Thumboo J, Yamwong S,
et al. Validity of the Thai EQ-5D in an occupational population in
Thailand. Qual Life Res. 2012.
68. Lim LL, Seubsman SA, Sleigh A. Thai SF-36 health survey: tests of data
quality, scaling assumptions, reliability and validity in healthy men and
women. Health Qual Life Outcomes. 2008;6:52.
69. Papaioannou D, Brazier J, Parry G. How valid and responsive are generic health
status measures, such as EQ-5D and SF-36, in schizophrenia? a systematic
review. Value Health. 2011;14:907-20.
70. Tosh J, Brazier J, Evans P, Longworth L. A review of generic preference-based
measures of health-related quality of life in visual disorders. Value Health.
2012;15(1):118-27.
71. Harrison MJ, Davies LM, Bansback NJ, Ingram M, Anis AH, Symmons DP.
The validity and responsiveness of generic utility measures in rheumatoid
arthritis: a review. J Rheumatol. 2008;35(4):592-602.
72. Lee CF, Ng R, Luo N, Wong NS, Yap YS, Lo SK, et al. The English and
Chinese versions of the five-level EuroQoL Group's five-dimension
questionnaire (EQ-5D) were valid and reliable and provided comparable
scores in Asian breast cancer patients. Support Care Cancer.
2013;21(1):201-9.
73. The EuroQol Group. EQ-VT operational instructions. 2013.
74. Oppe M, van Hout B. The optimal hybrid: Experimental design and modeling of
a combination of TTO and DCE. In: Yfantopoulos J, editor. 27th scientific
plenary meeting of the EuroQol group; Rotterdam2010.
75. Devlin NJ, Hansen P, Kind P, Williams A. Logical inconsistencies in survey
respondents' health state valuations -- a methodological challenge for
estimating social tariffs. Health Econ. 2003;12(7):529-44.
76. Dolan P, Kind P. Inconsistency and health state valuations. In: Kind P, Brooks
R, Rabin R, editors. EQ-5D concepts and methods : a developmental
history. Dordrecht: Springer; 2005.
Juntana Pattanaphesaj References / 122
77. Dolan P, Kind P. Inconsistency and health state valuations. Soc Sci Med.
1996;42(4):609-15.
78. Skrondal A, Rabe-Hesketh S. Multilevel and related models for longitudinal
data. In: Leeuw Jd, Meijer E, editors. Handbook of multilevel analysis.
New York: Springer Science+Business Media; 2008.
79. Kohler U, Kreuter F. Data analysis using Stata. Texas: A Stata press publication;
2005.
80. Metcalf GE. Specification testing in panel data with instrumental variables.
National Bureau of Economic Research, 1992.
81. Snijders T, Bosker R. Multilevel analysis: an introduction to basic and advanced
multilevel modeling. London: Sage publications; 1999.
82. Gujarati DN. Basic Econometrics. 4th edition ed. Singapore: McGraw-Hill
Bookk; 1995.
83. Hausman J. Specification tests in econometrics. Econometrica. 1978;46:1251-71.
84. Leeuw Jd, Meijer E. Introduction to multilevel analysis. In: Leeuw Jd, Meijer E,
editors. Handbook of multilevel analysis. New York: Springer
Science+Business Media; 2008.
85. D.F.Hendry, J.F.Richard. The econometric analysis of economic time series.
International Statistical Review. 1983;51:3-33.
86. Tabachnick BG, Fidell LS. Using multivariate statistics. Massachusetts: Pearson
Education; 2007.
87. McCarthy WF, Guo N. The assessment of the degree of concordance between
the observed values and the predicted values of a mixed-effect model
using “method of comparison” Techniques. Maryland medical research
institute, 2007.
88. Myles PS, Cui J. Using the Bland-Altman method to measure agreement with
repeated measures. British journal of anaesthesia. 2007;99(3):309-11.
89. Middel B, van Sonderen E. Statistical significant change versus relevant or
important change in (quasi) experimental design: some conceptual and
methodological problems in estimating magnitude of intervention-related
change in health services research. International journal of integrated care.
2002;2:e15.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 123
90. Bansback N, Brazier J, Tsuchiya A, Anis A. Using a discrete choice experiment
to estimate health state utility values. Journal of health economics.
2012;31(1):306-18.
91. Ramos-Goni JM, Rivero-Arias O, Errea M, Stolk EA, Herdman M, Cabases JM.
Dealing with the health state 'dead' when using discrete choice
experiments to obtain values for EQ-5D-5L heath states. Eur J Health
Econ. 2013;14 Suppl 1:S33-42.
92. M J Campbell. Statistics at square two: understanding modern statistical
applications in medicine. London: BMJ Publishing; 2001.
93. Janssen BM, Oppe M, Versteegh MM, Stolk EA. Introducing the composite time
trade-off: a test of feasibility and face validity. Eur J Health Econ. 2013;14
Suppl 1:S5-13.
94. Dolan P. Modeling valuations for EuroQol health states. Med Care.
1997;35(11):1095-108.
95. Devlin N, Hout Bv, Shah K, Mulhern B, Feng Y, editors. Developing an EQ-5D-
5L value set for England. Patient Reported Outcomes Measures (PROMs)
research conference; 2013 3rd December 2013; London.
96. Bundhamcharoen K, Odton P, Phulkerd S, Tangcharoensathien V. Burden of
disease in Thailand: changes in health gap between 1999 and 2004. BMC
Public Health. 2011;11:53.
97. Holmes J, McGill S, Kind P, Bottomley J, Gillam S, Murphy M. Health-related
quality of life in type 2 diabetes (TARDIS-2). Value Health. 2000;3 Suppl
1:47-51.
98. Ware JE, Jr., Sherbourne CD. The MOS 36-item short-form health survey (SF-
36). I. Conceptual framework and item selection. Med Care.
1992;30(6):473-83.
99. Laver-Fawcett A. Principles of Assessment and Outcome Measurement for
Occupational Therapists and Physiotherapists: Theory, Skills and
Application. London: John Wiley and Sons Ltd.; 2007.
100. Kottner J, Audige L, Brorson S, Donner A, Gajewski BJ, Hrobjartsson A, et al.
Guidelines for Reporting Reliability and Agreement Studies (GRRAS)
were proposed. J Clin Epidemiol. 2011;64(1):96-106.
Juntana Pattanaphesaj References / 124
101. Limwattananon S. Handling uncertainty of the economic evaluation result:
sensitivity analysis. J Med Assoc Thai. 2008;91 Suppl 2:S59-65.
102. Luo N, Li M, Stolk EA, Devlin NJ. The effects of lead time and visual aids in
TTO valuation: a study of the EQ-VT framework. Eur J Health Econ.
2013;14 Suppl 1:S15-24.
103. Versteegh MM, Attema AE, Oppe M, Devlin NJ, Stolk EA. Time to tweak the
TTO: results from a comparison of alternative specifications of the TTO.
Eur J Health Econ. 2013;14 Suppl 1:S43-51.
104. Lamers LM. The transformation of utilities for health states worse than death:
consequences for the estimation of EQ-5D value sets. Med Care.
2007;45(3):238-44.
105. Stolk EA, Oppe M, Scalone L, Krabbe PF. Discrete choice modeling for the
quantification of health states: the case of the EQ-5D. Value Health.
2010;13(8):1005-13.
106. Norman R, Cronin P, Viney R. A pilot discrete choice experiment to explore
preferences for EQ-5D-5L health states. Applied health economics and
health policy. 2013;11(3):287-98.
107. Andrade MV, Noranha KVMdS, Maia AC, Kind P. What matters most?
Evidence-based findings of health dimensions affecting societal
preferences of EQ-5D health states. Cad Saúde Pública. 2013;29(Sup:s59-72).
108. Rubin RR, Peyrot M. Quality of life and diabetes. Diabetes/metabolism research
and reviews. 1999;15(3):205-18.
109. Jacobson AM, Groot MD, Samson JA. The evaluation of two measures of
quality of life in patients with type I and type II diabetes. Diabetes care.
1994;17(4).
110. Fenwick E, Marshall DA, Levy AR, Nichol G. Using and interpreting cost-
effectiveness acceptability curves: an example using data from a trial of
management strategies for atrial fibrillation. BMC health services research.
2006;6:52.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 125
APPENDICES
Juntana Pattanaphesaj Appendices / 126
APPENDIX A
CERTIFICATE OF ETHICAL CONSIDERATION
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 127
APPENDIX B
THE EXAMPLE OF EQ-VT SCREEN
1. On-screen introduction
2. TTO wheelchair example, Life A = 10 years
มปญหาในการเดนเลกนอย มปญหาในการอาบน า หรอใสเสอผาดวยตนเองเลกนอย มปญหาในการท ากจกรรมทท าเปนประจ าปานกลาง มอาการเจบปวดหรออาการไมสบายตวอยางมาก ไมรสกวตกกงวลหรอซมเศรา
Juntana Pattanaphesaj Appendices / 128
3. TTO wheelchair example, Life A = 5 years
4. Lead time TTO example, Life A = 10 years, state worse than dead
มปญหาในการเดนอยางมาก อาบน า หรอใสเสอผาดวยตนเองไมได ไมมปญหาในการท ากจกรรมทท าเปนประจ า มอาการเจบปวดหรออาการไมสบายตวอยางมาก รสกวตกกงวลหรอซมเศราอยางมากทสด
มปญหาในการเดนเลกนอย มปญหาในการอาบน า หรอใสเสอผาดวยตนเองเลกนอย มปญหาในการท ากจกรรมทท าเปนประจ าปานกลาง มอาการเจบปวดหรออาการไมสบายตวอยางมาก ไมรสกวตกกงวลหรอซมเศรา
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 129
5. On-screen DCE introduction
มปญหาในการเดนอยางมาก
อาบน า หรอใสเสอผาดวยตนเองไมได
ไมมปญหาในการท ากจกรรมทท าเปนประจ า
มอาการเจบปวดหรออาการไมสบายตวอยางมาก
รสกวตกกงวลหรอซมเศราอยางมากทสด
มปญหาในการเดนเลกนอย
มปญหาในการอาบน า หรอใสเสอผาดวยตนเองเลกนอย
มปญหาในการท ากจกรรมทท าเปนประจ าปานกลาง
มอาการเจบปวดหรออาการไมสบายตวอยางมาก
ไมรสกวตกกงวลหรอซมเศรา
อะไรดกวากน ชวตแบบ A หรอชวตแบบ B
A B
Juntana Pattanaphesaj Appendices / 130
APPENDIX C
EQ-5D-3L THAI VERSION
กรณาท าเครองหมาย x ลงในชองสเหลยมของค าถามแตละขอทตรงกบภาวะสขภาพของทานในวนนมากทสด การเคลอนไหว ขาพเจาไมมปญหาในการเดน ขาพเจามปญหาในการเดนบาง ขาพเจาไมสามารถไปไหนได และจ าเปนตองนอนอยบนเตยง การดแลตนเอง ขาพเจาไมมปญหาในการดแลตนเอง ขาพเจามปญหาในการอาบน าหรอแตงตวบาง ขาพเจาไมสามารถอาบน าหรอแตงตวดวยตนเองได กจกรรมทท าเปนประจ า (เชน การท างาน การเรยนหนงสอ การท างานบาน การท ากจกรรมในครอบครว หรอการท ากจกรรมยามวาง) ขาพเจาไมมปญหาในการท ากจกรรมทท าเปนประจ า ขาพเจามปญหาในการท ากจกรรมทท าเปนประจ าอยบาง ขาพเจาไมสามารถท ากจกรรมทท าเปนประจ าได ความเจบปวด/ไมสขสบาย ขาพเจาไมมอาการเจบปวดหรออาการไมสขสบาย ขาพเจามอาการเจบปวดหรออาการไมสขสบายปานกลาง ขาพเจามอาการเจบปวดหรออาการไมสขสบายมากทสด ความวตกกงวล/ซมเศรา ขาพเจาไมรสกวตกกงวลหรอซมเศรา ขาพเจารสกวตกกงวลหรอซมเศราปานกลาง ขาพเจารสกวตกกงวลหรอซมเศรามากทสด Thailand (Thai) © 2002 EuroQol Group. EQ-5D™ is a trade mark of the EuroQol Group
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 131
เพอชวยในการประเมนภาวะสขภาพของทาน, ทางเราไดจดท าสเกลวดระดบสขภาพขน เรมตงแตระดบ 0 ถง 100 โดยท 100 หมายถงภาวะสขภาพทดทสด และ 0 หมายถง ภาวะสขภาพทแยทสด ตามความคดของทาน
กรณาประเมนภาวะสขภาพของทานในวนนวาดหรอไมดเพยงไร โดยการลากเสน จากชองสเหลยมขางลางนไปยงจดบนสเกลวดระดบสขภาพทตรงกบภาวะสขภาพ ของทานในวนน
ภาวะสขภาพของทาน ในวนน
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
100
ภาวะสขภาพททาน รสกวาแยทสด
0
ภาวะสขภาพททาน รสกวาดทสด
Thailand (Thai) © 2002 EuroQol Group. EQ-5D™ is a trade mark of the EuroQol Group
Juntana Pattanaphesaj Appendices / 132
APPENDIX D
EQ-5D-5L THAI VERSION
แบบสอบถามเรองสขภาพ
ฉบบภาษาไทยส าหรบใชในประเทศไทย
(Thai version for Thailand)
Thailand (Thai) © 2012 EuroQol Group. EQ-5D™ is a trade mark of the EuroQol Group
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 133
ในแตละหวขอ กรณาท าเครองหมาย ลงในชองสเหลยม เพยงชองเดยว ทตรงกบสขภาพของทานในวนน มากทสด การเคลอนไหว ขาพเจาไมมปญหาในการเดน ขาพเจามปญหาในการเดนเลกนอย ขาพเจามปญหาในการเดนปานกลาง ขาพเจามปญหาในการเดนอยางมาก ขาพเจาเดนไมได การดแลตนเอง ขาพเจาไมมปญหาในการอาบน า หรอใสเสอผาดวยตนเอง ขาพเจามปญหาในการอาบน า หรอใสเสอผาดวยตนเองเลกนอย ขาพเจามปญหาในการอาบน า หรอใสเสอผาดวยตนเองปานกลาง ขาพเจามปญหาในการอาบน า หรอใสเสอผาดวยตนเองอยางมาก ขาพเจาอาบน า หรอใสเสอผาดวยตนเองไมได กจกรรมทท าเปนประจ า (เชน ท ำงำน, เรยนหนงสอ, ท ำงำนบำน, กจกรรมในครอบครว หรอกจกรรมยำมวำง) ขาพเจาไมมปญหาในการท ากจกรรมทท าเปนประจ า ขาพเจามปญหาในการท ากจกรรมทท าเปนประจ าเลกนอย ขาพเจามปญหาในการท ากจกรรมทท าเปนประจ าปานกลาง ขาพเจามปญหาในการท ากจกรรมทท าเปนประจ าอยางมาก ขาพเจาท ากจกรรมทท าเปนประจ าไมได อาการเจบปวด/อาการไมสบายตว ขาพเจาไมมอาการเจบปวดหรออาการไมสบายตว ขาพเจามอาการเจบปวดหรออาการไมสบายตวเลกนอย ขาพเจามอาการเจบปวดหรออาการไมสบายตวปานกลาง ขาพเจามอาการเจบปวดหรออาการไมสบายตวอยางมาก ขาพเจามอาการเจบปวดหรออาการไมสบายตวอยางมากทสด ความวตกกงวล/ความซมเศรา ขาพเจาไมรสกวตกกงวลหรอซมเศรา ขาพเจารสกวตกกงวลหรอซมเศราเลกนอย ขาพเจารสกวตกกงวลหรอซมเศราปานกลาง ขาพเจารสกวตกกงวลหรอซมเศราอยางมาก ขาพเจารสกวตกกงวลหรอซมเศราอยางมากทสด
Thailand (Thai) © 2012 EuroQol Group. EQ-5D™ is a trade mark of the EuroQol Group
Juntana Pattanaphesaj Appendices / 134
เราอยากทราบวาสขภาพของทานเปนอยางไรในวนน
สเกลวดสขภาพนมตวเลขตงแต 0 ถง 100
100 หมายถง สขภาพดทสด ตามความคดของทาน
0 หมายถง สขภาพแยทสด ตามความคดของทาน
ท าเครองหมาย X บนสเกลเพอระบวาสขภาพของทานเปนอยางไรในวนน
ตอนน กรณาใสตวเลขทคณไดท าเครองหมายไวบนสเกลในชองสเหลยมดานลางน
สขภาพของทานในวนน =
Thailand (Thai) ©2012 EuroQol Group. EQ-5D™ is a trade mark of the EuroQol Group
10
0
20
30
40
50
60
80
70
90
100
5
15
25
35
45
55
75
65
85
95
สขภาพดทสด
ตามความคดของทาน
สขภาพแยทสด
ตามความคดของทาน
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 135
APPENDIX E
BACKGROUND QUESTIONS
1. Have you experienced serious illness?
In your self Yes No
In your family Yes No
In caring for others Yes No
2. How old are you? years
3. Are you male or female? Male Female
4. What is your marital status?
Single Married Widowed Divorced/Separated
5. How many children do you have?
6. What is the highest level of education that you have completed?
Unlettered Primary school
High school Diploma
Bachelor’s degree Master’s degree or higher
7. What is your main occupation?
Agriculture/fishery Business owner
Unskilled labor Employee
Government/state enterprise officer Housewife
Student Retired
Looking for a job Unable to work due to sickness
Other.......................................................
8. How much is your household income per month? Baht
Juntana Pattanaphesaj Appendices / 136
APPENDIX F
TTO HEALTH STATES INCLUDED IN THE EQ-VT
Block State # State Block State # State Block State # State 1 1 11221 5 33 43315 9 65 11414 2 11235 34 54153 66 25331 3 54231 35 52431 67 25222 4 51451 36 24443 68 21444 5 34515 37 14113 69 31514 6 35245 38 31524 70 53243 7 12514 39 15151 71 53244 8 45144 40 21315 72 35143 82 12111 85 11112 84 11121 86 55555 86 55555 86 55555
2 9 12543 6 41 12112 10 73 11122 10 12121 42 11212 74 52335 11 43542 43 44553 75 35311 12 34155 44 21345 76 43555 13 52215 45 34244 77 24445 14 45133 46 23152 78 13224 15 32443 47 43514 79 34232 16 23514 48 55424 80 42321 83 11211 81 21111 82 12111 86 55555 86 55555 86 55555
3 17 45233 7 49 13122 18 55233 50 24553 19 31525 51 51152 20 52455 52 11425 21 12244 53 22434 22 13313 54 42115 23 25122 55 35332 24 11421 56 45413 81 21111 83 11211 86 55555 86 55555
4 25 21112 8 57 33253 26 14554 58 23242 27 12513 59 24342 28 44345 60 32314 29 12344 61 12334 30 53221 62 21334 31 54342 63 55225 32 44125 64 53412 84 11121 85 11112 86 55555 86 55555
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 137
APPENDIX G
TTO FEEDBACK QUESTIONS
Please tell us what you thought about the questions you just answered, where you were
comparing two different ‘lives’
Agr
ee
Dis
agre
e
It was easy to understand the questions I was
asked.
I found it easy to tell the difference between the
lives I was asked to think about.
I found it difficult to decide on the exact points
where Life A and Life B were about the same.
Juntana Pattanaphesaj Appendices / 138
APPENDIX H
DCE PAIRS INCLUDED IN THE EQ-VT
Block Option 1 Option 2 Block Option 1 Option 2 Block Option 1 Option 2 1 35554 55211 6 13432 13245 11 15244 44241 43141 25554 24314 43222 44151 53242 31135 11444 51354 41335 22413 22331 25515 22251 43244 25522 41424 35533 42441 21415 12253 12551 42452 23144 22411 43133 23513 52254 53422 42525 33225 53314 54121 44322 23122 12415
2 52132 21534 7 23551 43135 12 43534 32125 31331 35124 51255 31343 24155 32534 42255 55524 11352 31413 22433 12443 23235 11141 25212 32443 52422 55254 34412 54253 43412 13342 55244 53531 35312 14422 54424 15321 22222 25514 13553 31234 34134 45325 12111 21121
3 51311 32154 8 14552 55325 13 44134 22352 34355 43342 51114 41253 42243 35433 14333 24424 25235 13413 22512 55313 22453 13442 25145 52244 15534 43454 41552 22422 45533 14444 55153 22521 45115 54225 51552 35513 21235 12243 12112 22211 21111 12121 12521 41115
4 44115 21455 9 25312 41532 14 11214 45312 23443 25113 41315 15121 25342 51152 31451 45431 44351 24415 34442 15214 45552 32413 24145 32253 21114 52432 25332 51544 51424 35525 35252 32254 41114 24142 23552 32244 42512 23544 11212 22112 13222 31131 54344 15411
5 12145 15344 10 35321 53215 15 55335 53442 33424 41542 24453 41331 45542 42133 35521 43355 21423 13114 12151 35543 44323 21525 51331 22421 43245 34324 52155 45231 35235 42325 13515 11324 33443 54133 22544 35452 41312 24253 11211 22111 51131 35353 11121 21211
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 139
Block Option 1 Option 2 Block Option 1 Option 2 Block Option 1 Option 2 16 23442 25414 21 13111 11215 26 44521 41153 52544 34222 13251 53313 54455 55234 52211 11325 44234 33441 44231 25533 33224 42113 21522 25324 15555 53455 51123 43451 45515 34433 22343 34513 15241 12352 43525 23444 11112 12221 14344 52454 42153 53151 21112 12211
17 52523 54142 22 22341 45145 27 14533 21542 23451 34354 32334 22254 23134 14314 53551 21224 41545 33531 53431 52255 53125 31415 55235 22533 51522 45244 15113 14434 15424 33322 14224 32322 13334 45441 32241 51525 44145 45432 11122 23111 32211 14211 11221 22122
18 33223 21232 23 11512 22241 28 42323 55223 31521 43152 34345 51325 41325 13445 44123 51232 42122 31325 34333 33142 14455 15514 45531 14334 23231 25323 25545 35225 51214 45153 31444 11353 33111 32545 15351 14312 15335 43532 41431 24212 21335 44551 35431 51323
19 35231 53554 24 32442 54441 42421 54255 11545 14113 54423 32314 52223 54132 23233 12411 11234 21532 22123 11155 35322 41535 21445 55141 13131 23113 54454 24511 55534 33355
20 35211 42551 25 33432 15551 34132 24445 14122 54231 24523 45125 51324 34543 52111 11431 33243 11115 21354 41321 34234 13533 54555 35535 23531 53133 11445 32115 53543 41215
Juntana Pattanaphesaj Appendices / 140
APPENDIX I
DCE FEEDBACK QUESTIONS
Please tell us what you thought about the questions you just answered, where you were
comparing two different states of health.
Agr
ee
Dis
agre
e
It was easy to understand the questions I was
asked.
I found it easy to tell the difference between the
health states I was asked to think about.
I found it difficult to decide on my answers to the
questions.
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 141
APPENDIX J
COUNTRY-SPECIFIC QUESTIONS
แบบสอบถามเฉพาะประเทศ – paper and pencil version วนทสมภาษณ……………………………………ชอพนกงานสมภาษณ……………………………………… Identifier ………………………………..………อาย............ป เพศ 1. ชาย 2. หญง ทานเหนดวยกบขอความนมากนอยเพยงใด ใหท าเครองหมายวงกลมทตวเลข เพยง 1 ค าตอบ 1. ทกสงทเกดขนกบฉนในชวตนเปนผลมาจากการกระท าของฉนเมอชาตกอนๆ
1 ไมเหนดวย มากทสด
2 ไมเหนดวย
3 เฉย ๆ /
ไมมความคดเหน
4 เหนดวย
5 เหนดวยมากทสด
2. ฉนมคนทจะชวยดแลฉนในยามทเจบปวย 1
ไมเหนดวย มากทสด
2 ไมเหนดวย
3 เฉย ๆ /
ไมมความคดเหน
4 เหนดวย
5 เหนดวยมากทสด
3. ไมวาฉนจะเจบปวยรนแรงเพยงใดฉนจะพยายามมชวตอยตอไปใหนานทสดเพอท าบางสงบางอยาง 1
ไมเหนดวย มากทสด
2 ไมเหนดวย
3 เฉย ๆ /
ไมมความคดเหน
4 เหนดวย
5 เหนดวยมากทสด
4. การจบชวตเพอหนปญหาเปนสงทผดมากตามความเชอของฉน
1 ไมเหนดวย มากทสด
2 ไมเหนดวย
3 เฉย ๆ /
ไมมความคดเหน
4 เหนดวย
5 เหนดวยมากทสด
5. ฉนใชหลกค าสอนทางศาสนาในการเผชญกบปญหาในชวตของฉน 1
ไมเหนดวย มากทสด
2 ไมเหนดวย
3 เฉย ๆ /
ไมมความคดเหน
4 เหนดวย
5 เหนดวยมากทสด
Juntana Pattanaphesaj Appendices / 142
APPENDIX K
QUALITATIVE QUESTIONS
ค าถามเชงคณภาพ วนทสมภาษณ……………………………………ชอพนกงานสมภาษณ……………………………………… Identifier ………………………………..………อาย............ป เพศ 1. ชาย 2. หญง
เกณฑการคดเลอกผตอบแบบสอบถาม 1. มลกษณะเฉพาะในการตอบ TTO
1.1 เลอก “ชวตแบบ A & B เหมอนกน” ในจดทชวตแบบ A สนมาก (0-4 ป) เปนสวนใหญ 1.2 เลอกทจะมชวตอยนานๆ (เลอกชวตแบบ B เปนสวนใหญ) แมวาชวตแบบ B จะมอาการรนแรง (มสถานะรนแรงมากทสด/ท าไมได 2 ขอขนไป)
1.3 เลอกทจะเสยชวตเปนสวนใหญ แมชวตแบบ B จะมสถานะสขภาพทไมรนแรง 2. มกจะตดสนใจดวยมตสขภาพเพยง 1-2 มต (เชน การเคลอนไหว การดแลตนเอง) 3. ........................................................................................................................................
1. เหตผลทเลอกค าตอบเชนนน
........................................................................................................................................................................................
........................................................................................................................................................................................ ........................................................................................................................................................................................
2. สถานะสขภาพทง 5 ขอน ขอใดททานคดวาแยทสด และขอใดททานรบไดมากทสด โดยใหท าเครองหมายถก () ลงในชองสเหลยมทตรงกบความคดเหนของทานมากทสด
แยทสด สถานะสขภาพ รบไดมากทสด
1. เดนไมได
2. อาบน า หรอใสเสอผาดวยตนเองไมได
3. ท ากจกรรมทท าเปนประจ าไมได
4. เจบปวดหรออาการไมสบายตวอยางมากทสด
5. วตกกงวลหรอซมเศราอยางมากทสด
กจกรรมทท าเปนประจ า เชน ท างาน, เรยนหนงสอ, ท างานบาน, กจกรรมในครอบครว หรอกจกรรมยามวาง
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 143
3. ในเหตการณสมมตน ถาหากมชวตใหทานเลอก 3 แบบ ทานจะเลอกแบบใด เพราะอะไร หากทานมความเจบปวยรนแรงมาก เชน เปนมะเรงระยะสดทาย ซงเจบปวดทรมาน และก าลงจะตองเสยชวตในอกไมชา
รกษาดวยยา โดยไมเสยคาใชจาย ซงชวยใหมชวตตอไปไดอก 5 ป แตยงมอาการเจบปวด ทรมานเหมอนเดม
รกษาดวยยา โดยไมเสยคาใชจาย ซงชวยใหมชวตตอไปไดอก 6 เดอน โดยมสขภาพด ไมมอาการเจบปวด ทรมาน
ยอมเสยชวต โดยไมรกษา
เหตผลทเลอกค าตอบเชนนน
....................................................................................................................................................................................................................................................................................................................................................................
..................................................................................................................................................................................
ขอขอบคณทกทานทใหความรวมมอในการตอบแบบสอบถาม
Juntana Pattanaphesaj Appendices / 144
APPENDIX L
QUESTIONNAIRE FOR TESTING MEASUREMENT PROPERTY
โครงการประเมนเทคโนโลยและนโยบายดานสขภาพ
(Health Intervention and Technology Assessment Program)
โครงการวจย “การวดอรรถประโยชนของผปวยเบาหวานชนดท 1 และชนดท 2 ทพงอนซลน” และ การเปรยบเทยบคณสมบตทางจตวทยาของแบบสอบถาม EQ-5D-3L และ EQ-5D-5L ฉบบภาษาไทย”
สวนประกอบของแบบสมภาษณ
สวนท 1 ขอมลทวไปของผถกสมภาษณ
สวนท 2 แบบสอบถามเรองสขภาพ EQ-5D และความคดเหนตอแบบสอบถาม
สวนท 3 แบบส ารวจสขภาพ SF-36 ฉบบภาษาไทย
แบบสอบถามนมจดประสงคเพอ
1. ประเมนคณภาพชวตผปวยเบาหวานชนดท 1 และชนดท 2 ทพงอนซลน
2. ทดสอบคณสมบตของแบบสอบถาม EQ-5D-5L ฉบบภาษาไทย
ขอมลนไมมการระบชอผถกสมภาษณ และขอมลจะถกเกบไวเปนความลบ
ส าหรบใชในงานวจยนเทานน เพอประโยชนแกการพฒนาระบบบรการสขภาพในอนาคต
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 145
สวนท 1 ขอมลทวไปของผถกสมภาษณ หมายเลขแบบสอบถาม
ชอ-นามสกล ผสมภาษณ ................................................................. เรมสมภาษณเวลา ............ : ............
วน เดอน ป (พ.ศ.) ทสมภาษณ //
ผสมภาษณอธบายวตถประสงคของการศกษาวจยแกผถกสมภาษณ
สวนท 1: ขอมลทวไปของผถกสมภาษณ Variable code
1. เพศ
1. ชาย 2. หญง
Sex
2. อาย ............... ปเตม Age
3. สถานภาพ
1. โสด 2. ค 3. หมาย 4. หยา/แยกกนอย
Married
4. ระดบการศกษา
1. ไมไดเรยนหนงสอ 2. ประถมศกษา
3. มธยมศกษา/ปวช. 4. อนปรญญา/ปวส.หรอเทยบเทา
5. ปรญญาตรหรอเทยบเทา 6. สงกวาปรญญาตร
Education
5. อาชพหลก
1. เกษตรกรรม/ประมง 2. คาขาย/เจาของกจการ
3. ผใชแรงงาน/รบจางทวไป 4. พนกงานบรษทเอกชน
5. ขาราชการ/รฐวสาหกจ 6. พอบาน/แมบาน
7. นกเรยน/นกศกษา 8. เกษยณ
9. อยระหวางหางาน 10. ไมสามารถท างานไดเพราะปวย
Occupation
6. รายไดเฉลยของครอบครวตอเดอน....................................................บาท Income
Juntana Pattanaphesaj Appendices / 146
7. สทธการรกษา
1. ประกนสขภาพถวนหนา 2. ประกนสงคม
3. ขาราชการ/รฐวสาหกจ 4. ช าระเอง/ไมมสทธใดๆ
Health_Insure
8. ทานปวยเปนเบาหวานชนด
1. ชนดท 1 2. ชนดท 2
Type
9. ทานปวยเปนเบาหวานมานาน_____ป Duration
10. น าหนก______________กโลกรม Weight
11. สวนสง_________เซนตเมตร height
เสรจเวลา ............ : ........... Time_1
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 147
สวนท 2 แบบสอบถามเรองสขภาพ EQ-5D กรณาท าเครองหมาย ลงในชองสเหลยมของค าถามแตละขอทตรงกบภาวะสขภาพของทานในวนนมากทสด ในแตละขอ ทานจะตองตอบ 2 ครง โดยดานซายมค าตอบใหเลอก 5 ระดบ ดานขวามค าตอบใหเลอก 3 ระดบ เรมเวลา ...........:............
12. การเคลอนไหว ขาพเจาไมมปญหาในการเดน ขาพเจามปญหาในการเดนเลกนอย ขาพเจามปญหาในการเดนปานกลาง ขาพเจามปญหาในการเดนอยางมาก ขาพเจาเดนไมได
ขาพเจาไมมปญหาในการเดน ขาพเจามปญหาในการเดนบาง ขาพเจาไมสามารถไปไหนได และจ าเปนตองนอนอยบนเตยง
13. การดแลตนเอง ขาพเจาไมมปญหาในการอาบน าหรอใสเสอผาดวยตนเอง ขาพเจามปญหาในการอาบน า หรอใสเสอผาดวยตนเองเลกนอย ขาพเจามปญหาในการอาบน า หรอใสเสอผาดวยตนเองปานกลาง ขาพเจามปญหาในการอาบน าหรอใสเสอผาดวยตนเองอยางมาก ขาพเจาอาบน าหรอใสเสอผาดวยตนเองไมได
ขาพเจาไมมปญหาในการดแลตนเอง ขาพเจามปญหาในการอาบน าหรอแตงตวบาง ขาพเจาไมสามารถอาบน าหรอแตงตวดวยตนเองได
14. กจกรรมทท าเปนประจ า(เชน ท ำงำน, เรยนหนงสอ, ท ำงำนบำน,กจกรรมในครอบครว หรอกจกรรมยำมวำง) ขาพเจาไมมปญหาในการท ากจกรรมทท าเปนประจ า ขาพเจามปญหาในการท ากจกรรมทท าเปนประจ าเลกนอย ขาพเจามปญหาในการท ากจกรรมทท าเปนประจ าปานกลาง ขาพเจามปญหาในการท ากจกรรมทท าเปนประจ าอยางมาก ขาพเจาท ากจกรรมทท าเปนประจ าไมได
ขาพเจาไมมปญหาในการท ากจกรรมทท าเปนประจ า ขาพเจามปญหาในการท ากจกรรมทท าเปนประจ าอยบาง ขาพเจาไมสามารถท ากจกรรมทท าเปนประจ าได
15. อาการเจบปวด/อาการไมสบายตว ขาพเจาไมมอาการเจบปวดหรออาการไมสบายตว ขาพเจามอาการเจบปวดหรออาการไมสบายตวเลกนอย ขาพเจามอาการเจบปวดหรออาการไมสบายตวปานกลาง ขาพเจามอาการเจบปวดหรออาการไมสบายตวอยางมาก ขาพเจามอาการเจบปวดหรออาการไมสบายตวอยางมากทสด
ขาพเจาไมมอาการเจบปวดหรออาการไมสขสบาย ขาพเจามอาการเจบปวดหรออาการไมสขสบายปานกลาง ขาพเจามอาการเจบปวดหรออาการไมสขสบายมากทสด
16. ความวตกกงวล/ความซมเศรา ขาพเจาไมรสกวตกกงวลหรอซมเศรา ขาพเจารสกวตกกงวลหรอซมเศราเลกนอย ขาพเจารสกวตกกงวลหรอซมเศราปานกลาง ขาพเจารสกวตกกงวลหรอซมเศราอยางมาก ขาพเจารสกวตกกงวลหรอซมเศราอยางมากทสด
ขาพเจาไมรสกวตกกงวลหรอซมเศรา ขาพเจารสกวตกกงวลหรอซมเศราปานกลาง ขาพเจารสกวตกกงวลหรอซมเศรามากทสด
Juntana Pattanaphesaj Appendices / 148
เราอยากทราบวาสขภาพของทานเปนอยางไรในวนน
สเกลวดสขภาพนมตวเลขตงแต 0 ถง 100
100 หมายถง สขภาพดทสด ตามความคดของทาน
0 หมายถง สขภาพแยทสด ตามความคดของทาน
ท าเครองหมาย X บนสเกลเพอระบวาสขภาพของทานเปนอยางไรในวนน
ตอนน กรณาใสตวเลขทคณไดท าเครองหมายไวบนสเกลในชองสเหลยม
ดานลางน
สขภาพของทานในวนน =
10
0
20
30
40
50
60
80
70
90
100
5
15
25
35
45
55
75
65
85
95
สขภาพดทสด
ตามความคดของทาน
สขภาพแยทสด
ตามความคดของทาน
Thailand (Thai) © 2012 EuroQol Group. EQ-5D™ is a trade mark of the EuroQol Group
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 149
17. จากแบบสอบถามคณภาพชวตดานสขภาพแบบทมค าตอบใหเลอก (ขอ 12-16) ลกษณะค าตอบแบบใดทตอบงายทสด ตามความคดเหนของทาน ค าตอบ 3 ระดบ ค าตอบ 5 ระดบ ไมตางกน
18. จากแบบสอบถามคณภาพชวตดานสขภาพแบบทมค าตอบใหเลอก (ขอ 12-16) ลกษณะค าตอบแบบใดท
สะทอนสขภาพของทานไดดทสด ตามความคดเหนของทาน
ค าตอบ 3 ระดบ ค าตอบ 5 ระดบ ไมตางกน
Juntana Pattanaphesaj Appendices / 150
APPENDIX M
SF-36V2 THAI VERSION
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 151
Juntana Pattanaphesaj Appendices / 152
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 153
Juntana Pattanaphesaj Appendices / 154
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 155
APPENDIX N
THAI PREFERENCE SCORE FOR EQ-5D-5L HEALTH STATES
Profile utility Profile utility Profile utility 11111 1.000 11243 0.627 11425 0.547
11112 0.968 11244 0.521 11431 0.767
11113 0.903 11245 0.475 11432 0.735
11114 0.798 11251 0.691 11433 0.671
11115 0.751 11252 0.659 11434 0.565
11121 0.960 11253 0.594 11435 0.519
11122 0.928 11254 0.488 11441 0.602
11123 0.864 11255 0.442 11442 0.570
11124 0.758 11311 0.925 11443 0.505
11125 0.712 11312 0.893 11444 0.400
11131 0.932 11313 0.829 11445 0.353
11132 0.900 11314 0.723 11451 0.569
11133 0.835 11315 0.677 11452 0.537
11134 0.730 11321 0.886 11453 0.473
11135 0.683 11322 0.854 11454 0.367
11141 0.767 11323 0.789 11455 0.320
11142 0.734 11324 0.683 11511 0.793
11143 0.670 11325 0.637 11512 0.761
11144 0.564 11331 0.857 11513 0.696
11145 0.518 11332 0.825 11514 0.590
11151 0.734 11333 0.761 11515 0.544
11152 0.702 11334 0.655 11521 0.753
11153 0.637 11335 0.609 11522 0.721
11154 0.531 11341 0.692 11523 0.656
11155 0.485 11342 0.660 11524 0.551
11211 0.957 11343 0.595 11525 0.504
11212 0.925 11344 0.490 11531 0.725
11213 0.860 11345 0.443 11532 0.693
11214 0.755 11351 0.659 11533 0.628
11215 0.708 11352 0.627 11534 0.522
11221 0.917 11353 0.563 11535 0.476
11222 0.885 11354 0.457 11541 0.559
11223 0.821 11355 0.410 11542 0.527
11224 0.715 11411 0.835 11543 0.463
11225 0.669 11412 0.803 11544 0.357
11231 0.889 11413 0.739 11545 0.310
11232 0.857 11414 0.633 11551 0.526
11233 0.792 11415 0.587 11552 0.494
11234 0.687 11421 0.796 11553 0.430
11235 0.640 11422 0.764 11554 0.324
11241 0.724 11423 0.699 11555 0.278
11242 0.691 11424 0.593
Juntana Pattanaphesaj Appendices / 156
Profile utility Profile utility Profile utility 12111 0.967 12243 0.594 12425 0.514
12112 0.935 12244 0.488 12431 0.734
12113 0.870 12245 0.442 12432 0.702
12114 0.765 12251 0.658 12433 0.638
12115 0.718 12252 0.626 12434 0.532
12121 0.927 12253 0.561 12435 0.485
12122 0.895 12254 0.455 12441 0.569
12123 0.831 12255 0.409 12442 0.537
12124 0.725 12311 0.892 12443 0.472
12125 0.678 12312 0.860 12444 0.366
12131 0.899 12313 0.796 12445 0.320
12132 0.867 12314 0.690 12451 0.536
12133 0.802 12315 0.643 12452 0.504
12134 0.696 12321 0.853 12453 0.439
12135 0.650 12322 0.820 12454 0.334
12141 0.733 12323 0.756 12455 0.287
12142 0.701 12324 0.650 12511 0.760
12143 0.637 12325 0.604 12512 0.727
12144 0.531 12331 0.824 12513 0.663
12145 0.485 12332 0.792 12514 0.557
12151 0.701 12333 0.728 12515 0.511
12152 0.669 12334 0.622 12521 0.720
12153 0.604 12335 0.575 12522 0.688
12154 0.498 12341 0.659 12523 0.623
12155 0.452 12342 0.627 12524 0.518
12211 0.924 12343 0.562 12525 0.471
12212 0.892 12344 0.456 12531 0.691
12213 0.827 12345 0.410 12532 0.659
12214 0.722 12351 0.626 12533 0.595
12215 0.675 12352 0.594 12534 0.489
12221 0.884 12353 0.529 12535 0.443
12222 0.852 12354 0.424 12541 0.526
12223 0.788 12355 0.377 12542 0.494
12224 0.682 12411 0.802 12543 0.430
12225 0.635 12412 0.770 12544 0.324
12231 0.856 12413 0.706 12545 0.277
12232 0.824 12414 0.600 12551 0.493
12233 0.759 12415 0.553 12552 0.461
12234 0.654 12421 0.763 12553 0.397
12235 0.607 12422 0.731 12554 0.291
12241 0.690 12423 0.666 12555 0.244
12242 0.658 12424 0.560
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 157
Profile utility Profile utility Profile utility 13111 0.892 13243 0.519 13425 0.439
13112 0.860 13244 0.413 13431 0.659
13113 0.795 13245 0.367 13432 0.627
13114 0.690 13251 0.583 13433 0.563
13115 0.643 13252 0.551 13434 0.457
13121 0.852 13253 0.486 13435 0.411
13122 0.820 13254 0.380 13441 0.494
13123 0.756 13255 0.334 13442 0.462
13124 0.650 13311 0.817 13443 0.397
13125 0.604 13312 0.785 13444 0.292
13131 0.824 13313 0.721 13445 0.245
13132 0.792 13314 0.615 13451 0.461
13133 0.727 13315 0.569 13452 0.429
13134 0.622 13321 0.778 13453 0.365
13135 0.575 13322 0.746 13454 0.259
13141 0.659 13323 0.681 13455 0.212
13142 0.626 13324 0.575 13511 0.685
13143 0.562 13325 0.529 13512 0.653
13144 0.456 13331 0.749 13513 0.588
13145 0.410 13332 0.717 13514 0.482
13151 0.626 13333 0.653 13515 0.436
13152 0.594 13334 0.547 13521 0.645
13153 0.529 13335 0.501 13522 0.613
13154 0.423 13341 0.584 13523 0.549
13155 0.377 13342 0.552 13524 0.443
13211 0.849 13343 0.487 13525 0.396
13212 0.817 13344 0.382 13531 0.617
13213 0.752 13345 0.335 13532 0.585
13214 0.647 13351 0.551 13533 0.520
13215 0.600 13352 0.519 13534 0.414
13221 0.809 13353 0.455 13535 0.368
13222 0.777 13354 0.349 13541 0.451
13223 0.713 13355 0.302 13542 0.419
13224 0.607 13411 0.727 13543 0.355
13225 0.561 13412 0.695 13544 0.249
13231 0.781 13413 0.631 13545 0.202
13232 0.749 13414 0.525 13551 0.418
13233 0.684 13415 0.479 13552 0.386
13234 0.579 13421 0.688 13553 0.322
13235 0.532 13422 0.656 13554 0.216
13241 0.616 13423 0.591 13555 0.170
13242 0.584 13424 0.485
Juntana Pattanaphesaj Appendices / 158
Profile utility Profile utility Profile utility 14111 0.775 14243 0.402 14425 0.322
14112 0.743 14244 0.297 14431 0.543
14113 0.679 14245 0.250 14432 0.510
14114 0.573 14251 0.466 14433 0.446
14115 0.526 14252 0.434 14434 0.340
14121 0.736 14253 0.369 14435 0.294
14122 0.704 14254 0.264 14441 0.377
14123 0.639 14255 0.217 14442 0.345
14124 0.533 14311 0.701 14443 0.281
14125 0.487 14312 0.669 14444 0.175
14131 0.707 14313 0.604 14445 0.128
14132 0.675 14314 0.498 14451 0.344
14133 0.611 14315 0.452 14452 0.312
14134 0.505 14321 0.661 14453 0.248
14135 0.458 14322 0.629 14454 0.142
14141 0.542 14323 0.564 14455 0.096
14142 0.510 14324 0.459 14511 0.568
14143 0.445 14325 0.412 14512 0.536
14144 0.339 14331 0.633 14513 0.471
14145 0.293 14332 0.600 14514 0.366
14151 0.509 14333 0.536 14515 0.319
14152 0.477 14334 0.430 14521 0.528
14153 0.412 14335 0.384 14522 0.496
14154 0.307 14341 0.467 14523 0.432
14155 0.260 14342 0.435 14524 0.326
14211 0.732 14343 0.371 14525 0.279
14212 0.700 14344 0.265 14531 0.500
14213 0.636 14345 0.218 14532 0.468
14214 0.530 14351 0.434 14533 0.403
14215 0.483 14352 0.402 14534 0.298
14221 0.693 14353 0.338 14535 0.251
14222 0.661 14354 0.232 14541 0.334
14223 0.596 14355 0.186 14542 0.302
14224 0.490 14411 0.611 14543 0.238
14225 0.444 14412 0.579 14544 0.132
14231 0.664 14413 0.514 14545 0.086
14232 0.632 14414 0.408 14551 0.302
14233 0.568 14415 0.362 14552 0.270
14234 0.462 14421 0.571 14553 0.205
14235 0.415 14422 0.539 14554 0.099
14241 0.499 14423 0.474 14555 0.053
14242 0.467 14424 0.369
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 159
Profile utility Profile utility Profile utility 15111 0.746 15243 0.373 15425 0.293
15112 0.714 15244 0.268 15431 0.514
15113 0.650 15245 0.221 15432 0.482
15114 0.544 15251 0.437 15433 0.417
15115 0.498 15252 0.405 15434 0.311
15121 0.707 15253 0.341 15435 0.265
15122 0.675 15254 0.235 15441 0.348
15123 0.610 15255 0.188 15442 0.316
15124 0.504 15311 0.672 15443 0.252
15125 0.458 15312 0.640 15444 0.146
15131 0.678 15313 0.575 15445 0.100
15132 0.646 15314 0.469 15451 0.316
15133 0.582 15315 0.423 15452 0.283
15134 0.476 15321 0.632 15453 0.219
15135 0.430 15322 0.600 15454 0.113
15141 0.513 15323 0.536 15455 0.067
15142 0.481 15324 0.430 15511 0.539
15143 0.416 15325 0.383 15512 0.507
15144 0.311 15331 0.604 15513 0.443
15145 0.264 15332 0.572 15514 0.337
15151 0.480 15333 0.507 15515 0.290
15152 0.448 15334 0.401 15521 0.499
15153 0.384 15335 0.355 15522 0.467
15154 0.278 15341 0.438 15523 0.403
15155 0.231 15342 0.406 15524 0.297
15211 0.703 15343 0.342 15525 0.251
15212 0.671 15344 0.236 15531 0.471
15213 0.607 15345 0.190 15532 0.439
15214 0.501 15351 0.406 15533 0.374
15215 0.455 15352 0.373 15534 0.269
15221 0.664 15353 0.309 15535 0.222
15222 0.632 15354 0.203 15541 0.306
15223 0.567 15355 0.157 15542 0.274
15224 0.461 15411 0.582 15543 0.209
15225 0.415 15412 0.550 15544 0.103
15231 0.635 15413 0.485 15545 0.057
15232 0.603 15414 0.379 15551 0.273
15233 0.539 15415 0.333 15552 0.241
15234 0.433 15421 0.542 15553 0.176
15235 0.387 15422 0.510 15554 0.071
15241 0.470 15423 0.446 15555 0.024
15242 0.438 15424 0.340
Juntana Pattanaphesaj Appendices / 160
Profile utility Profile utility Profile utility 21111 0.944 21243 0.571 21425 0.491
21112 0.912 21244 0.465 21431 0.711
21113 0.847 21245 0.418 21432 0.679
21114 0.741 21251 0.634 21433 0.614
21115 0.695 21252 0.602 21434 0.509
21121 0.904 21253 0.538 21435 0.462
21122 0.872 21254 0.432 21441 0.546
21123 0.807 21255 0.386 21442 0.514
21124 0.702 21311 0.869 21443 0.449
21125 0.655 21312 0.837 21444 0.343
21131 0.876 21313 0.772 21445 0.297
21132 0.844 21314 0.667 21451 0.513
21133 0.779 21315 0.620 21452 0.481
21134 0.673 21321 0.829 21453 0.416
21135 0.627 21322 0.797 21454 0.311
21141 0.710 21323 0.733 21455 0.264
21142 0.678 21324 0.627 21511 0.736
21143 0.614 21325 0.581 21512 0.704
21144 0.508 21331 0.801 21513 0.640
21145 0.461 21332 0.769 21514 0.534
21151 0.677 21333 0.704 21515 0.488
21152 0.645 21334 0.599 21521 0.697
21153 0.581 21335 0.552 21522 0.665
21154 0.475 21341 0.636 21523 0.600
21155 0.429 21342 0.604 21524 0.494
21211 0.901 21343 0.539 21525 0.448
21212 0.869 21344 0.433 21531 0.668
21213 0.804 21345 0.387 21532 0.636
21214 0.698 21351 0.603 21533 0.572
21215 0.652 21352 0.571 21534 0.466
21221 0.861 21353 0.506 21535 0.420
21222 0.829 21354 0.401 21541 0.503
21223 0.764 21355 0.354 21542 0.471
21224 0.659 21411 0.779 21543 0.406
21225 0.612 21412 0.747 21544 0.301
21231 0.833 21413 0.683 21545 0.254
21232 0.801 21414 0.577 21551 0.470
21233 0.736 21415 0.530 21552 0.438
21234 0.630 21421 0.739 21553 0.374
21235 0.584 21422 0.707 21554 0.268
21241 0.667 21423 0.643 21555 0.221
21242 0.635 21424 0.537
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 161
Profile utility Profile utility Profile utility 22111 0.911 22243 0.538 22425 0.457
22112 0.878 22244 0.432 22431 0.678
22113 0.814 22245 0.385 22432 0.646
22114 0.708 22251 0.601 22433 0.581
22115 0.662 22252 0.569 22434 0.476
22121 0.871 22253 0.505 22435 0.429
22122 0.839 22254 0.399 22441 0.512
22123 0.774 22255 0.353 22442 0.480
22124 0.669 22311 0.836 22443 0.416
22125 0.622 22312 0.804 22444 0.310
22131 0.842 22313 0.739 22445 0.264
22132 0.810 22314 0.634 22451 0.480
22133 0.746 22315 0.587 22452 0.448
22134 0.640 22321 0.796 22453 0.383
22135 0.594 22322 0.764 22454 0.277
22141 0.677 22323 0.700 22455 0.231
22142 0.645 22324 0.594 22511 0.703
22143 0.581 22325 0.547 22512 0.671
22144 0.475 22331 0.768 22513 0.607
22145 0.428 22332 0.736 22514 0.501
22151 0.644 22333 0.671 22515 0.454
22152 0.612 22334 0.566 22521 0.664
22153 0.548 22335 0.519 22522 0.632
22154 0.442 22341 0.602 22523 0.567
22155 0.395 22342 0.570 22524 0.461
22211 0.868 22343 0.506 22525 0.415
22212 0.835 22344 0.400 22531 0.635
22213 0.771 22345 0.354 22532 0.603
22214 0.665 22351 0.570 22533 0.539
22215 0.619 22352 0.538 22534 0.433
22221 0.828 22353 0.473 22535 0.386
22222 0.796 22354 0.367 22541 0.470
22223 0.731 22355 0.321 22542 0.438
22224 0.626 22411 0.746 22543 0.373
22225 0.579 22412 0.714 22544 0.267
22231 0.800 22413 0.649 22545 0.221
22232 0.767 22414 0.544 22551 0.437
22233 0.703 22415 0.497 22552 0.405
22234 0.597 22421 0.706 22553 0.340
22235 0.551 22422 0.674 22554 0.235
22241 0.634 22423 0.610 22555 0.188
22242 0.602 22424 0.504
Juntana Pattanaphesaj Appendices / 162
Profile utility Profile utility Profile utility 23111 0.836 23243 0.463 23425 0.383
23112 0.804 23244 0.357 23431 0.603
23113 0.739 23245 0.310 23432 0.571
23114 0.633 23251 0.527 23433 0.506
23115 0.587 23252 0.494 23434 0.401
23121 0.796 23253 0.430 23435 0.354
23122 0.764 23254 0.324 23441 0.438
23123 0.699 23255 0.278 23442 0.406
23124 0.594 23311 0.761 23443 0.341
23125 0.547 23312 0.729 23444 0.235
23131 0.768 23313 0.665 23445 0.189
23132 0.736 23314 0.559 23451 0.405
23133 0.671 23315 0.512 23452 0.373
23134 0.565 23321 0.721 23453 0.308
23135 0.519 23322 0.689 23454 0.203
23141 0.602 23323 0.625 23455 0.156
23142 0.570 23324 0.519 23511 0.628
23143 0.506 23325 0.473 23512 0.596
23144 0.400 23331 0.693 23513 0.532
23145 0.353 23332 0.661 23514 0.426
23151 0.569 23333 0.596 23515 0.380
23152 0.537 23334 0.491 23521 0.589
23153 0.473 23335 0.444 23522 0.557
23154 0.367 23341 0.528 23523 0.492
23155 0.321 23342 0.496 23524 0.386
23211 0.793 23343 0.431 23525 0.340
23212 0.761 23344 0.325 23531 0.560
23213 0.696 23345 0.279 23532 0.528
23214 0.590 23351 0.495 23533 0.464
23215 0.544 23352 0.463 23534 0.358
23221 0.753 23353 0.398 23535 0.312
23222 0.721 23354 0.293 23541 0.395
23223 0.657 23355 0.246 23542 0.363
23224 0.551 23411 0.671 23543 0.298
23225 0.504 23412 0.639 23544 0.193
23231 0.725 23413 0.575 23545 0.146
23232 0.693 23414 0.469 23551 0.362
23233 0.628 23415 0.422 23552 0.330
23234 0.522 23421 0.631 23553 0.266
23235 0.476 23422 0.599 23554 0.160
23241 0.559 23423 0.535 23555 0.113
23242 0.527 23424 0.429
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 163
Profile utility Profile utility Profile utility 24111 0.719 24243 0.346 24425 0.266
24112 0.687 24244 0.240 24431 0.486
24113 0.622 24245 0.194 24432 0.454
24114 0.517 24251 0.410 24433 0.390
24115 0.470 24252 0.378 24434 0.284
24121 0.679 24253 0.313 24435 0.237
24122 0.647 24254 0.207 24441 0.321
24123 0.583 24255 0.161 24442 0.289
24124 0.477 24311 0.644 24443 0.224
24125 0.430 24312 0.612 24444 0.119
24131 0.651 24313 0.548 24445 0.072
24132 0.619 24314 0.442 24451 0.288
24133 0.554 24315 0.395 24452 0.256
24134 0.449 24321 0.605 24453 0.191
24135 0.402 24322 0.573 24454 0.086
24141 0.485 24323 0.508 24455 0.039
24142 0.453 24324 0.402 24511 0.512
24143 0.389 24325 0.356 24512 0.480
24144 0.283 24331 0.576 24513 0.415
24145 0.237 24332 0.544 24514 0.309
24151 0.453 24333 0.480 24515 0.263
24152 0.421 24334 0.374 24521 0.472
24153 0.356 24335 0.327 24522 0.440
24154 0.250 24341 0.411 24523 0.375
24155 0.204 24342 0.379 24524 0.270
24211 0.676 24343 0.314 24525 0.223
24212 0.644 24344 0.209 24531 0.444
24213 0.579 24345 0.162 24532 0.411
24214 0.474 24351 0.378 24533 0.347
24215 0.427 24352 0.346 24534 0.241
24221 0.636 24353 0.281 24535 0.195
24222 0.604 24354 0.176 24541 0.278
24223 0.540 24355 0.129 24542 0.246
24224 0.434 24411 0.554 24543 0.182
24225 0.387 24412 0.522 24544 0.076
24231 0.608 24413 0.458 24545 0.029
24232 0.576 24414 0.352 24551 0.245
24233 0.511 24415 0.306 24552 0.213
24234 0.406 24421 0.515 24553 0.149
24235 0.359 24422 0.483 24554 0.043
24241 0.443 24423 0.418 24555 -0.003
24242 0.410 24424 0.312
Juntana Pattanaphesaj Appendices / 164
Profile utility Profile utility Profile utility 25111 0.690 25243 0.317 25425 0.237
25112 0.658 25244 0.211 25431 0.457
25113 0.593 25245 0.165 25432 0.425
25114 0.488 25251 0.381 25433 0.361
25115 0.441 25252 0.349 25434 0.255
25121 0.650 25253 0.284 25435 0.209
25122 0.618 25254 0.179 25441 0.292
25123 0.554 25255 0.132 25442 0.260
25124 0.448 25311 0.615 25443 0.195
25125 0.402 25312 0.583 25444 0.090
25131 0.622 25313 0.519 25445 0.043
25132 0.590 25314 0.413 25451 0.259
25133 0.525 25315 0.367 25452 0.227
25134 0.420 25321 0.576 25453 0.163
25135 0.373 25322 0.544 25454 0.057
25141 0.457 25323 0.479 25455 0.010
25142 0.425 25324 0.374 25511 0.483
25143 0.360 25325 0.327 25512 0.451
25144 0.254 25331 0.547 25513 0.386
25145 0.208 25332 0.515 25514 0.280
25151 0.424 25333 0.451 25515 0.234
25152 0.392 25334 0.345 25521 0.443
25153 0.327 25335 0.299 25522 0.411
25154 0.222 25341 0.382 25523 0.347
25155 0.175 25342 0.350 25524 0.241
25211 0.647 25343 0.285 25525 0.194
25212 0.615 25344 0.180 25531 0.415
25213 0.551 25345 0.133 25532 0.383
25214 0.445 25351 0.349 25533 0.318
25215 0.398 25352 0.317 25534 0.212
25221 0.607 25353 0.253 25535 0.166
25222 0.575 25354 0.147 25541 0.249
25223 0.511 25355 0.100 25542 0.217
25224 0.405 25411 0.525 25543 0.153
25225 0.359 25412 0.493 25544 0.047
25231 0.579 25413 0.429 25545 0.001
25232 0.547 25414 0.323 25551 0.217
25233 0.482 25415 0.277 25552 0.184
25234 0.377 25421 0.486 25553 0.120
25235 0.330 25422 0.454 25554 0.014
25241 0.414 25423 0.389 25555 -0.032
25242 0.382 25424 0.284
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 165
Profile utility Profile utility Profile utility 31111 0.886 31243 0.513 31425 0.432
31112 0.853 31244 0.407 31431 0.653
31113 0.789 31245 0.360 31432 0.621
31114 0.683 31251 0.576 31433 0.556
31115 0.637 31252 0.544 31434 0.451
31121 0.846 31253 0.480 31435 0.404
31122 0.814 31254 0.374 31441 0.487
31123 0.749 31255 0.327 31442 0.455
31124 0.644 31311 0.811 31443 0.391
31125 0.597 31312 0.779 31444 0.285
31131 0.817 31313 0.714 31445 0.239
31132 0.785 31314 0.609 31451 0.455
31133 0.721 31315 0.562 31452 0.423
31134 0.615 31321 0.771 31453 0.358
31135 0.569 31322 0.739 31454 0.252
31141 0.652 31323 0.675 31455 0.206
31142 0.620 31324 0.569 31511 0.678
31143 0.555 31325 0.522 31512 0.646
31144 0.450 31331 0.743 31513 0.582
31145 0.403 31332 0.711 31514 0.476
31151 0.619 31333 0.646 31515 0.429
31152 0.587 31334 0.541 31521 0.639
31153 0.523 31335 0.494 31522 0.606
31154 0.417 31341 0.577 31523 0.542
31155 0.370 31342 0.545 31524 0.436
31211 0.843 31343 0.481 31525 0.390
31212 0.810 31344 0.375 31531 0.610
31213 0.746 31345 0.329 31532 0.578
31214 0.640 31351 0.545 31533 0.514
31215 0.594 31352 0.513 31534 0.408
31221 0.803 31353 0.448 31535 0.361
31222 0.771 31354 0.342 31541 0.445
31223 0.706 31355 0.296 31542 0.413
31224 0.601 31411 0.721 31543 0.348
31225 0.554 31412 0.689 31544 0.242
31231 0.774 31413 0.624 31545 0.196
31232 0.742 31414 0.519 31551 0.412
31233 0.678 31415 0.472 31552 0.380
31234 0.572 31421 0.681 31553 0.315
31235 0.526 31422 0.649 31554 0.210
31241 0.609 31423 0.585 31555 0.163
31242 0.577 31424 0.479
Juntana Pattanaphesaj Appendices / 166
Profile utility Profile utility Profile utility 32111 0.852 32243 0.479 32425 0.399
32112 0.820 32244 0.374 32431 0.620
32113 0.756 32245 0.327 32432 0.588
32114 0.650 32251 0.543 32433 0.523
32115 0.604 32252 0.511 32434 0.417
32121 0.813 32253 0.447 32435 0.371
32122 0.781 32254 0.341 32441 0.454
32123 0.716 32255 0.294 32442 0.422
32124 0.610 32311 0.778 32443 0.358
32125 0.564 32312 0.746 32444 0.252
32131 0.784 32313 0.681 32445 0.205
32132 0.752 32314 0.575 32451 0.422
32133 0.688 32315 0.529 32452 0.389
32134 0.582 32321 0.738 32453 0.325
32135 0.535 32322 0.706 32454 0.219
32141 0.619 32323 0.642 32455 0.173
32142 0.587 32324 0.536 32511 0.645
32143 0.522 32325 0.489 32512 0.613
32144 0.417 32331 0.710 32513 0.548
32145 0.370 32332 0.678 32514 0.443
32151 0.586 32333 0.613 32515 0.396
32152 0.554 32334 0.507 32521 0.605
32153 0.490 32335 0.461 32522 0.573
32154 0.384 32341 0.544 32523 0.509
32155 0.337 32342 0.512 32524 0.403
32211 0.809 32343 0.448 32525 0.357
32212 0.777 32344 0.342 32531 0.577
32213 0.713 32345 0.295 32532 0.545
32214 0.607 32351 0.512 32533 0.480
32215 0.561 32352 0.479 32534 0.375
32221 0.770 32353 0.415 32535 0.328
32222 0.738 32354 0.309 32541 0.412
32223 0.673 32355 0.263 32542 0.380
32224 0.567 32411 0.688 32543 0.315
32225 0.521 32412 0.656 32544 0.209
32231 0.741 32413 0.591 32545 0.163
32232 0.709 32414 0.485 32551 0.379
32233 0.645 32415 0.439 32552 0.347
32234 0.539 32421 0.648 32553 0.282
32235 0.493 32422 0.616 32554 0.177
32241 0.576 32423 0.552 32555 0.130
32242 0.544 32424 0.446
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 167
Profile utility Profile utility Profile utility 33111 0.778 33243 0.405 33425 0.324
33112 0.745 33244 0.299 33431 0.545
33113 0.681 33245 0.252 33432 0.513
33114 0.575 33251 0.468 33433 0.448
33115 0.529 33252 0.436 33434 0.343
33121 0.738 33253 0.372 33435 0.296
33122 0.706 33254 0.266 33441 0.380
33123 0.641 33255 0.220 33442 0.347
33124 0.536 33311 0.703 33443 0.283
33125 0.489 33312 0.671 33444 0.177
33131 0.709 33313 0.606 33445 0.131
33132 0.677 33314 0.501 33451 0.347
33133 0.613 33315 0.454 33452 0.315
33134 0.507 33321 0.663 33453 0.250
33135 0.461 33322 0.631 33454 0.144
33141 0.544 33323 0.567 33455 0.098
33142 0.512 33324 0.461 33511 0.570
33143 0.448 33325 0.414 33512 0.538
33144 0.342 33331 0.635 33513 0.474
33145 0.295 33332 0.603 33514 0.368
33151 0.511 33333 0.538 33515 0.321
33152 0.479 33334 0.433 33521 0.531
33153 0.415 33335 0.386 33522 0.499
33154 0.309 33341 0.469 33523 0.434
33155 0.262 33342 0.437 33524 0.328
33211 0.735 33343 0.373 33525 0.282
33212 0.702 33344 0.267 33531 0.502
33213 0.638 33345 0.221 33532 0.470
33214 0.532 33351 0.437 33533 0.406
33215 0.486 33352 0.405 33534 0.300
33221 0.695 33353 0.340 33535 0.253
33222 0.663 33354 0.234 33541 0.337
33223 0.598 33355 0.188 33542 0.305
33224 0.493 33411 0.613 33543 0.240
33225 0.446 33412 0.581 33544 0.134
33231 0.667 33413 0.516 33545 0.088
33232 0.634 33414 0.411 33551 0.304
33233 0.570 33415 0.364 33552 0.272
33234 0.464 33421 0.573 33553 0.207
33235 0.418 33422 0.541 33554 0.102
33241 0.501 33423 0.477 33555 0.055
33242 0.469 33424 0.371
Juntana Pattanaphesaj Appendices / 168
Profile utility Profile utility Profile utility 34111 0.661 34243 0.288 34425 0.208
34112 0.629 34244 0.182 34431 0.428
34113 0.564 34245 0.136 34432 0.396
34114 0.458 34251 0.352 34433 0.332
34115 0.412 34252 0.319 34434 0.226
34121 0.621 34253 0.255 34435 0.179
34122 0.589 34254 0.149 34441 0.263
34123 0.525 34255 0.103 34442 0.231
34124 0.419 34311 0.586 34443 0.166
34125 0.372 34312 0.554 34444 0.060
34131 0.593 34313 0.490 34445 0.014
34132 0.561 34314 0.384 34451 0.230
34133 0.496 34315 0.337 34452 0.198
34134 0.390 34321 0.546 34453 0.133
34135 0.344 34322 0.514 34454 0.028
34141 0.427 34323 0.450 34455 -0.019
34142 0.395 34324 0.344 34511 0.453
34143 0.331 34325 0.298 34512 0.421
34144 0.225 34331 0.518 34513 0.357
34145 0.178 34332 0.486 34514 0.251
34151 0.395 34333 0.421 34515 0.205
34152 0.362 34334 0.316 34521 0.414
34153 0.298 34335 0.269 34522 0.382
34154 0.192 34341 0.353 34523 0.317
34155 0.146 34342 0.321 34524 0.211
34211 0.618 34343 0.256 34525 0.165
34212 0.586 34344 0.150 34531 0.385
34213 0.521 34345 0.104 34532 0.353
34214 0.415 34351 0.320 34533 0.289
34215 0.369 34352 0.288 34534 0.183
34221 0.578 34353 0.223 34535 0.137
34222 0.546 34354 0.118 34541 0.220
34223 0.482 34355 0.071 34542 0.188
34224 0.376 34411 0.496 34543 0.123
34225 0.329 34412 0.464 34544 0.018
34231 0.550 34413 0.400 34545 -0.029
34232 0.518 34414 0.294 34551 0.187
34233 0.453 34415 0.247 34552 0.155
34234 0.347 34421 0.457 34553 0.091
34235 0.301 34422 0.424 34554 -0.015
34241 0.384 34423 0.360 34555 -0.062
34242 0.352 34424 0.254
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 169
Profile utility Profile utility Profile utility 35111 0.632 35243 0.259 35425 0.179
35112 0.600 35244 0.153 35431 0.399
35113 0.535 35245 0.107 35432 0.367
35114 0.430 35251 0.323 35433 0.303
35115 0.383 35252 0.291 35434 0.197
35121 0.592 35253 0.226 35435 0.150
35122 0.560 35254 0.120 35441 0.234
35123 0.496 35255 0.074 35442 0.202
35124 0.390 35311 0.557 35443 0.137
35125 0.343 35312 0.525 35444 0.032
35131 0.564 35313 0.461 35445 -0.015
35132 0.532 35314 0.355 35451 0.201
35133 0.467 35315 0.308 35452 0.169
35134 0.362 35321 0.518 35453 0.104
35135 0.315 35322 0.486 35454 -0.001
35141 0.398 35323 0.421 35455 -0.048
35142 0.366 35324 0.315 35511 0.425
35143 0.302 35325 0.269 35512 0.393
35144 0.196 35331 0.489 35513 0.328
35145 0.150 35332 0.457 35514 0.222
35151 0.366 35333 0.393 35515 0.176
35152 0.334 35334 0.287 35521 0.385
35153 0.269 35335 0.240 35522 0.353
35154 0.163 35341 0.324 35523 0.288
35155 0.117 35342 0.292 35524 0.183
35211 0.589 35343 0.227 35525 0.136
35212 0.557 35344 0.122 35531 0.357
35213 0.492 35345 0.075 35532 0.324
35214 0.387 35351 0.291 35533 0.260
35215 0.340 35352 0.259 35534 0.154
35221 0.549 35353 0.194 35535 0.108
35222 0.517 35354 0.089 35541 0.191
35223 0.453 35355 0.042 35542 0.159
35224 0.347 35411 0.467 35543 0.095
35225 0.300 35412 0.435 35544 -0.011
35231 0.521 35413 0.371 35545 -0.058
35232 0.489 35414 0.265 35551 0.158
35233 0.424 35415 0.218 35552 0.126
35234 0.319 35421 0.428 35553 0.062
35235 0.272 35422 0.396 35554 -0.044
35241 0.355 35423 0.331 35555 -0.090
35242 0.323 35424 0.225
Juntana Pattanaphesaj Appendices / 170
Profile utility Profile utility Profile utility 41111 0.769 41243 0.396 41425 0.315
41112 0.736 41244 0.290 41431 0.536
41113 0.672 41245 0.243 41432 0.504
41114 0.566 41251 0.459 41433 0.439
41115 0.520 41252 0.427 41434 0.334
41121 0.729 41253 0.363 41435 0.287
41122 0.697 41254 0.257 41441 0.370
41123 0.632 41255 0.210 41442 0.338
41124 0.527 41311 0.694 41443 0.274
41125 0.480 41312 0.662 41444 0.168
41131 0.700 41313 0.597 41445 0.122
41132 0.668 41314 0.492 41451 0.338
41133 0.604 41315 0.445 41452 0.306
41134 0.498 41321 0.654 41453 0.241
41135 0.452 41322 0.622 41454 0.135
41141 0.535 41323 0.558 41455 0.089
41142 0.503 41324 0.452 41511 0.561
41143 0.438 41325 0.405 41512 0.529
41144 0.333 41331 0.626 41513 0.465
41145 0.286 41332 0.594 41514 0.359
41151 0.502 41333 0.529 41515 0.312
41152 0.470 41334 0.424 41521 0.522
41153 0.406 41335 0.377 41522 0.489
41154 0.300 41341 0.460 41523 0.425
41155 0.253 41342 0.428 41524 0.319
41211 0.726 41343 0.364 41525 0.273
41212 0.693 41344 0.258 41531 0.493
41213 0.629 41345 0.212 41532 0.461
41214 0.523 41351 0.428 41533 0.397
41215 0.477 41352 0.396 41534 0.291
41221 0.686 41353 0.331 41535 0.244
41222 0.654 41354 0.225 41541 0.328
41223 0.589 41355 0.179 41542 0.296
41224 0.484 41411 0.604 41543 0.231
41225 0.437 41412 0.572 41544 0.125
41231 0.657 41413 0.507 41545 0.079
41232 0.625 41414 0.402 41551 0.295
41233 0.561 41415 0.355 41552 0.263
41234 0.455 41421 0.564 41553 0.198
41235 0.409 41422 0.532 41554 0.093
41241 0.492 41423 0.468 41555 0.046
41242 0.460 41424 0.362
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 171
Profile utility Profile utility Profile utility 42111 0.735 42243 0.362 42425 0.282
42112 0.703 42244 0.257 42431 0.503
42113 0.639 42245 0.210 42432 0.471
42114 0.533 42251 0.426 42433 0.406
42115 0.487 42252 0.394 42434 0.300
42121 0.696 42253 0.330 42435 0.254
42122 0.664 42254 0.224 42441 0.337
42123 0.599 42255 0.177 42442 0.305
42124 0.493 42311 0.661 42443 0.241
42125 0.447 42312 0.629 42444 0.135
42131 0.667 42313 0.564 42445 0.089
42132 0.635 42314 0.458 42451 0.305
42133 0.571 42315 0.412 42452 0.272
42134 0.465 42321 0.621 42453 0.208
42135 0.418 42322 0.589 42454 0.102
42141 0.502 42323 0.525 42455 0.056
42142 0.470 42324 0.419 42511 0.528
42143 0.405 42325 0.372 42512 0.496
42144 0.300 42331 0.593 42513 0.431
42145 0.253 42332 0.561 42514 0.326
42151 0.469 42333 0.496 42515 0.279
42152 0.437 42334 0.390 42521 0.488
42153 0.373 42335 0.344 42522 0.456
42154 0.267 42341 0.427 42523 0.392
42155 0.220 42342 0.395 42524 0.286
42211 0.692 42343 0.331 42525 0.240
42212 0.660 42344 0.225 42531 0.460
42213 0.596 42345 0.178 42532 0.428
42214 0.490 42351 0.395 42533 0.363
42215 0.444 42352 0.362 42534 0.258
42221 0.653 42353 0.298 42535 0.211
42222 0.621 42354 0.192 42541 0.295
42223 0.556 42355 0.146 42542 0.263
42224 0.450 42411 0.571 42543 0.198
42225 0.404 42412 0.539 42544 0.092
42231 0.624 42413 0.474 42545 0.046
42232 0.592 42414 0.368 42551 0.262
42233 0.528 42415 0.322 42552 0.230
42234 0.422 42421 0.531 42553 0.165
42235 0.376 42422 0.499 42554 0.060
42241 0.459 42423 0.435 42555 0.013
42242 0.427 42424 0.329
Juntana Pattanaphesaj Appendices / 172
Profile utility Profile utility Profile utility 43111 0.661 43243 0.288 43425 0.207
43112 0.628 43244 0.182 43431 0.428
43113 0.564 43245 0.135 43432 0.396
43114 0.458 43251 0.351 43433 0.331
43115 0.412 43252 0.319 43434 0.226
43121 0.621 43253 0.255 43435 0.179
43122 0.589 43254 0.149 43441 0.263
43123 0.524 43255 0.103 43442 0.230
43124 0.419 43311 0.586 43443 0.166
43125 0.372 43312 0.554 43444 0.060
43131 0.593 43313 0.489 43445 0.014
43132 0.560 43314 0.384 43451 0.230
43133 0.496 43315 0.337 43452 0.198
43134 0.390 43321 0.546 43453 0.133
43135 0.344 43322 0.514 43454 0.027
43141 0.427 43323 0.450 43455 -0.019
43142 0.395 43324 0.344 43511 0.453
43143 0.331 43325 0.297 43512 0.421
43144 0.225 43331 0.518 43513 0.357
43145 0.178 43332 0.486 43514 0.251
43151 0.394 43333 0.421 43515 0.204
43152 0.362 43334 0.316 43521 0.414
43153 0.298 43335 0.269 43522 0.382
43154 0.192 43341 0.352 43523 0.317
43155 0.145 43342 0.320 43524 0.211
43211 0.618 43343 0.256 43525 0.165
43212 0.585 43344 0.150 43531 0.385
43213 0.521 43345 0.104 43532 0.353
43214 0.415 43351 0.320 43533 0.289
43215 0.369 43352 0.288 43534 0.183
43221 0.578 43353 0.223 43535 0.136
43222 0.546 43354 0.117 43541 0.220
43223 0.481 43355 0.071 43542 0.188
43224 0.376 43411 0.496 43543 0.123
43225 0.329 43412 0.464 43544 0.017
43231 0.550 43413 0.399 43545 -0.029
43232 0.517 43414 0.294 43551 0.187
43233 0.453 43415 0.247 43552 0.155
43234 0.347 43421 0.456 43553 0.090
43235 0.301 43422 0.424 43554 -0.015
43241 0.384 43423 0.360 43555 -0.062
43242 0.352 43424 0.254
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 173
Profile utility Profile utility Profile utility 44111 0.544 44243 0.171 44425 0.091
44112 0.512 44244 0.065 44431 0.311
44113 0.447 44245 0.019 44432 0.279
44114 0.341 44251 0.235 44433 0.215
44115 0.295 44252 0.202 44434 0.109
44121 0.504 44253 0.138 44435 0.062
44122 0.472 44254 0.032 44441 0.146
44123 0.408 44255 -0.014 44442 0.114
44124 0.302 44311 0.469 44443 0.049
44125 0.255 44312 0.437 44444 -0.057
44131 0.476 44313 0.373 44445 -0.103
44132 0.444 44314 0.267 44451 0.113
44133 0.379 44315 0.220 44452 0.081
44134 0.273 44321 0.430 44453 0.016
44135 0.227 44322 0.397 44454 -0.089
44141 0.310 44323 0.333 44455 -0.136
44142 0.278 44324 0.227 44511 0.336
44143 0.214 44325 0.181 44512 0.304
44144 0.108 44331 0.401 44513 0.240
44145 0.061 44332 0.369 44514 0.134
44151 0.278 44333 0.305 44515 0.088
44152 0.245 44334 0.199 44521 0.297
44153 0.181 44335 0.152 44522 0.265
44154 0.075 44341 0.236 44523 0.200
44155 0.029 44342 0.204 44524 0.095
44211 0.501 44343 0.139 44525 0.048
44212 0.469 44344 0.033 44531 0.268
44213 0.404 44345 -0.013 44532 0.236
44214 0.298 44351 0.203 44533 0.172
44215 0.252 44352 0.171 44534 0.066
44221 0.461 44353 0.106 44535 0.020
44222 0.429 44354 0.001 44541 0.103
44223 0.365 44355 -0.046 44542 0.071
44224 0.259 44411 0.379 44543 0.006
44225 0.212 44412 0.347 44544 -0.099
44231 0.433 44413 0.283 44545 -0.146
44232 0.401 44414 0.177 44551 0.070
44233 0.336 44415 0.130 44552 0.038
44234 0.230 44421 0.340 44553 -0.026
44235 0.184 44422 0.307 44554 -0.132
44241 0.267 44423 0.243 44555 -0.179
44242 0.235 44424 0.137
Juntana Pattanaphesaj Appendices / 174
Profile utility Profile utility Profile utility 45111 0.515 45243 0.142 45425 0.062
45112 0.483 45244 0.036 45431 0.282
45113 0.418 45245 -0.010 45432 0.250
45114 0.313 45251 0.206 45433 0.186
45115 0.266 45252 0.174 45434 0.080
45121 0.475 45253 0.109 45435 0.033
45122 0.443 45254 0.003 45441 0.117
45123 0.379 45255 -0.043 45442 0.085
45124 0.273 45311 0.440 45443 0.020
45125 0.226 45312 0.408 45444 -0.085
45131 0.447 45313 0.344 45445 -0.132
45132 0.415 45314 0.238 45451 0.084
45133 0.350 45315 0.191 45452 0.052
45134 0.245 45321 0.401 45453 -0.013
45135 0.198 45322 0.369 45454 -0.118
45141 0.281 45323 0.304 45455 -0.165
45142 0.249 45324 0.198 45511 0.308
45143 0.185 45325 0.152 45512 0.276
45144 0.079 45331 0.372 45513 0.211
45145 0.033 45332 0.340 45514 0.105
45151 0.249 45333 0.276 45515 0.059
45152 0.217 45334 0.170 45521 0.268
45153 0.152 45335 0.123 45522 0.236
45154 0.046 45341 0.207 45523 0.171
45155 0.000 45342 0.175 45524 0.066
45211 0.472 45343 0.110 45525 0.019
45212 0.440 45344 0.005 45531 0.240
45213 0.375 45345 -0.042 45532 0.207
45214 0.270 45351 0.174 45533 0.143
45215 0.223 45352 0.142 45534 0.037
45221 0.432 45353 0.077 45535 -0.009
45222 0.400 45354 -0.028 45541 0.074
45223 0.336 45355 -0.075 45542 0.042
45224 0.230 45411 0.350 45543 -0.022
45225 0.183 45412 0.318 45544 -0.128
45231 0.404 45413 0.254 45545 -0.175
45232 0.372 45414 0.148 45551 0.041
45233 0.307 45415 0.101 45552 0.009
45234 0.202 45421 0.311 45553 -0.055
45235 0.155 45422 0.279 45554 -0.161
45241 0.238 45423 0.214 45555 -0.207
45242 0.206 45424 0.108
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 175
Profile utility Profile utility Profile utility 51111 0.693 51243 0.320 51425 0.240
51112 0.661 51244 0.214 51431 0.460
51113 0.596 51245 0.168 51432 0.428
51114 0.491 51251 0.384 51433 0.364
51115 0.444 51252 0.352 51434 0.258
51121 0.653 51253 0.287 51435 0.211
51122 0.621 51254 0.181 51441 0.295
51123 0.557 51255 0.135 51442 0.263
51124 0.451 51311 0.618 51443 0.198
51125 0.404 51312 0.586 51444 0.092
51131 0.625 51313 0.522 51445 0.046
51132 0.593 51314 0.416 51451 0.262
51133 0.528 51315 0.369 51452 0.230
51134 0.422 51321 0.579 51453 0.165
51135 0.376 51322 0.546 51454 0.060
51141 0.459 51323 0.482 51455 0.013
51142 0.427 51324 0.376 51511 0.486
51143 0.363 51325 0.330 51512 0.453
51144 0.257 51331 0.550 51513 0.389
51145 0.211 51332 0.518 51514 0.283
51151 0.427 51333 0.454 51515 0.237
51152 0.394 51334 0.348 51521 0.446
51153 0.330 51335 0.301 51522 0.414
51154 0.224 51341 0.385 51523 0.349
51155 0.178 51342 0.353 51524 0.244
51211 0.650 51343 0.288 51525 0.197
51212 0.618 51344 0.182 51531 0.417
51213 0.553 51345 0.136 51532 0.385
51214 0.448 51351 0.352 51533 0.321
51215 0.401 51352 0.320 51534 0.215
51221 0.610 51353 0.255 51535 0.169
51222 0.578 51354 0.150 51541 0.252
51223 0.514 51355 0.103 51542 0.220
51224 0.408 51411 0.528 51543 0.155
51225 0.361 51412 0.496 51544 0.050
51231 0.582 51413 0.432 51545 0.003
51232 0.550 51414 0.326 51551 0.219
51233 0.485 51415 0.279 51552 0.187
51234 0.379 51421 0.489 51553 0.123
51235 0.333 51422 0.456 51554 0.017
51241 0.416 51423 0.392 51555 -0.030
51242 0.384 51424 0.286
Juntana Pattanaphesaj Appendices / 176
Profile utility Profile utility Profile utility 52111 0.660 52243 0.287 52425 0.207
52112 0.628 52244 0.181 52431 0.427
52113 0.563 52245 0.134 52432 0.395
52114 0.457 52251 0.350 52433 0.330
52115 0.411 52252 0.318 52434 0.225
52121 0.620 52253 0.254 52435 0.178
52122 0.588 52254 0.148 52441 0.262
52123 0.523 52255 0.102 52442 0.230
52124 0.418 52311 0.585 52443 0.165
52125 0.371 52312 0.553 52444 0.059
52131 0.592 52313 0.488 52445 0.013
52132 0.560 52314 0.383 52451 0.229
52133 0.495 52315 0.336 52452 0.197
52134 0.389 52321 0.545 52453 0.132
52135 0.343 52322 0.513 52454 0.027
52141 0.426 52323 0.449 52455 -0.020
52142 0.394 52324 0.343 52511 0.452
52143 0.330 52325 0.297 52512 0.420
52144 0.224 52331 0.517 52513 0.356
52145 0.177 52332 0.485 52514 0.250
52151 0.393 52333 0.420 52515 0.204
52152 0.361 52334 0.315 52521 0.413
52153 0.297 52335 0.268 52522 0.381
52154 0.191 52341 0.352 52523 0.316
52155 0.145 52342 0.320 52524 0.210
52211 0.617 52343 0.255 52525 0.164
52212 0.585 52344 0.149 52531 0.384
52213 0.520 52345 0.103 52532 0.352
52214 0.414 52351 0.319 52533 0.288
52215 0.368 52352 0.287 52534 0.182
52221 0.577 52353 0.222 52535 0.135
52222 0.545 52354 0.116 52541 0.219
52223 0.480 52355 0.070 52542 0.187
52224 0.375 52411 0.495 52543 0.122
52225 0.328 52412 0.463 52544 0.017
52231 0.549 52413 0.398 52545 -0.030
52232 0.517 52414 0.293 52551 0.186
52233 0.452 52415 0.246 52552 0.154
52234 0.346 52421 0.455 52553 0.090
52235 0.300 52422 0.423 52554 -0.016
52241 0.383 52423 0.359 52555 -0.063
52242 0.351 52424 0.253
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 177
Profile utility Profile utility Profile utility 53111 0.585 53243 0.212 53425 0.132
53112 0.553 53244 0.106 53431 0.352
53113 0.488 53245 0.060 53432 0.320
53114 0.383 53251 0.276 53433 0.256
53115 0.336 53252 0.244 53434 0.150
53121 0.545 53253 0.179 53435 0.103
53122 0.513 53254 0.073 53441 0.187
53123 0.449 53255 0.027 53442 0.155
53124 0.343 53311 0.510 53443 0.090
53125 0.296 53312 0.478 53444 -0.016
53131 0.517 53313 0.414 53445 -0.062
53132 0.485 53314 0.308 53451 0.154
53133 0.420 53315 0.261 53452 0.122
53134 0.314 53321 0.471 53453 0.057
53135 0.268 53322 0.439 53454 -0.048
53141 0.351 53323 0.374 53455 -0.095
53142 0.319 53324 0.268 53511 0.378
53143 0.255 53325 0.222 53512 0.345
53144 0.149 53331 0.442 53513 0.281
53145 0.103 53332 0.410 53514 0.175
53151 0.319 53333 0.346 53515 0.129
53152 0.287 53334 0.240 53521 0.338
53153 0.222 53335 0.193 53522 0.306
53154 0.116 53341 0.277 53523 0.241
53155 0.070 53342 0.245 53524 0.136
53211 0.542 53343 0.180 53525 0.089
53212 0.510 53344 0.074 53531 0.309
53213 0.445 53345 0.028 53532 0.277
53214 0.340 53351 0.244 53533 0.213
53215 0.293 53352 0.212 53534 0.107
53221 0.502 53353 0.147 53535 0.061
53222 0.470 53354 0.042 53541 0.144
53223 0.406 53355 -0.005 53542 0.112
53224 0.300 53411 0.420 53543 0.048
53225 0.253 53412 0.388 53544 -0.058
53231 0.474 53413 0.324 53545 -0.105
53232 0.442 53414 0.218 53551 0.111
53233 0.377 53415 0.171 53552 0.079
53234 0.272 53421 0.381 53553 0.015
53235 0.225 53422 0.349 53554 -0.091
53241 0.308 53423 0.284 53555 -0.138
53242 0.276 53424 0.178
Juntana Pattanaphesaj Appendices / 178
Profile utility Profile utility Profile utility 54111 0.468 54243 0.095 54425 0.015
54112 0.436 54244 -0.011 54431 0.235
54113 0.371 54245 -0.057 54432 0.203
54114 0.266 54251 0.159 54433 0.139
54115 0.219 54252 0.127 54434 0.033
54121 0.428 54253 0.062 54435 -0.013
54122 0.396 54254 -0.043 54441 0.070
54123 0.332 54255 -0.090 54442 0.038
54124 0.226 54311 0.393 54443 -0.027
54125 0.180 54312 0.361 54444 -0.132
54131 0.400 54313 0.297 54445 -0.179
54132 0.368 54314 0.191 54451 0.037
54133 0.303 54315 0.145 54452 0.005
54134 0.198 54321 0.354 54453 -0.059
54135 0.151 54322 0.322 54454 -0.165
54141 0.235 54323 0.257 54455 -0.212
54142 0.203 54324 0.151 54511 0.261
54143 0.138 54325 0.105 54512 0.229
54144 0.032 54331 0.325 54513 0.164
54145 -0.014 54332 0.293 54514 0.058
54151 0.202 54333 0.229 54515 0.012
54152 0.170 54334 0.123 54521 0.221
54153 0.105 54335 0.077 54522 0.189
54154 -0.001 54341 0.160 54523 0.125
54155 -0.047 54342 0.128 54524 0.019
54211 0.425 54343 0.063 54525 -0.028
54212 0.393 54344 -0.042 54531 0.193
54213 0.328 54345 -0.089 54532 0.161
54214 0.223 54351 0.127 54533 0.096
54215 0.176 54352 0.095 54534 -0.010
54221 0.385 54353 0.031 54535 -0.056
54222 0.353 54354 -0.075 54541 0.027
54223 0.289 54355 -0.122 54542 -0.005
54224 0.183 54411 0.303 54543 -0.069
54225 0.137 54412 0.271 54544 -0.175
54231 0.357 54413 0.207 54545 -0.222
54232 0.325 54414 0.101 54551 -0.005
54233 0.260 54415 0.055 54552 -0.038
54234 0.155 54421 0.264 54553 -0.102
54235 0.108 54422 0.232 54554 -0.208
54241 0.192 54423 0.167 54555 -0.254
54242 0.160 54424 0.061
Fac. of Grad. Studies, Mahidol Univ. Ph.D. (Pharmacy Administration) / 179
Profile utility Profile utility Profile utility 55111 0.439 55243 0.066 55425 -0.014
55112 0.407 55244 -0.040 55431 0.207
55113 0.343 55245 -0.086 55432 0.174
55114 0.237 55251 0.130 55433 0.110
55115 0.190 55252 0.098 55434 0.004
55121 0.400 55253 0.033 55435 -0.042
55122 0.367 55254 -0.072 55441 0.041
55123 0.303 55255 -0.119 55442 0.009
55124 0.197 55311 0.365 55443 -0.055
55125 0.151 55312 0.333 55444 -0.161
55131 0.371 55313 0.268 55445 -0.208
55132 0.339 55314 0.162 55451 0.008
55133 0.275 55315 0.116 55452 -0.024
55134 0.169 55321 0.325 55453 -0.088
55135 0.122 55322 0.293 55454 -0.194
55141 0.206 55323 0.228 55455 -0.240
55142 0.174 55324 0.123 55511 0.232
55143 0.109 55325 0.076 55512 0.200
55144 0.003 55331 0.297 55513 0.135
55145 -0.043 55332 0.264 55514 0.030
55151 0.173 55333 0.200 55515 -0.017
55152 0.141 55334 0.094 55521 0.192
55153 0.076 55335 0.048 55522 0.160
55154 -0.029 55341 0.131 55523 0.096
55155 -0.076 55342 0.099 55524 -0.010
55211 0.396 55343 0.035 55525 -0.057
55212 0.364 55344 -0.071 55531 0.164
55213 0.300 55345 -0.118 55532 0.132
55214 0.194 55351 0.098 55533 0.067
55215 0.147 55352 0.066 55534 -0.038
55221 0.357 55353 0.002 55535 -0.085
55222 0.325 55354 -0.104 55541 -0.002
55223 0.260 55355 -0.150 55542 -0.034
55224 0.154 55411 0.275 55543 -0.098
55225 0.108 55412 0.243 55544 -0.204
55231 0.328 55413 0.178 55545 -0.250
55232 0.296 55414 0.072 55551 -0.034
55233 0.232 55415 0.026 55552 -0.066
55234 0.126 55421 0.235 55553 -0.131
55235 0.079 55422 0.203 55554 -0.237
55241 0.163 55423 0.138 55555 -0.283
55242 0.131 55424 0.033
Juntana Pattanaphesaj Biography / 180
BIOGRAPHY
NAME Miss Juntana Pattanaphesaj
DATE OF BIRTH July 19, 1974
PLACE OF BIRTH Kampangphet, Thailand
INSTITUTIONS ATTENDED Chiang Mai University, 1992-1997:
Bachelor of Science in Pharmacy
Mahidol University, 2005-2007:
Master of Science in Pharmacy
(Pharmacy Administration)
Mahidol University, 2010-2014:
Doctor of Philosophy
(Pharmacy Administration)
RESEARCH GRANTS 1) Burden of Diseases Project, Thailand;
2) EuroQol Foundation, The Netherlands
HOME ADDRESS 119/153 The Terrace Village, Soi Tiwanon 3,
Tiwanon Rd., Taladkwan, Muang,
Nonthaburi, Thailand
EMPLOYMENT ADDRESS Health Intervention and Technology
Assessment Program (HITAP), Ministry of
Public Health, Nonthaburi, Thailand
Position : researcher
Tel. 0-2590-4549
E-mail : [email protected]