Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Canadian Valuation of EQ-5D Health States: Preliminary Value Set and Considerations for Future Valuation Studies

  • Nick Bansback ,

    Affiliations School of Population and Public Health, University of British Columbia, Vancouver, British Columbia, Canada, Centre for Health Evaluation and Outcome Sciences, St Paul's Hospital, Vancouver, British Columbia, Canada, Health Economics and Decision Sciences, School of Health and Related Research, University of Sheffield, Sheffield, United Kingdom

  • Aki Tsuchiya,

    Affiliations Health Economics and Decision Sciences, School of Health and Related Research, University of Sheffield, Sheffield, United Kingdom, Department of Economics, University of Sheffield, Sheffield, United Kingdom

  • John Brazier,

    Affiliation Health Economics and Decision Sciences, School of Health and Related Research, University of Sheffield, Sheffield, United Kingdom

  • Aslam Anis

    Affiliations School of Population and Public Health, University of British Columbia, Vancouver, British Columbia, Canada, Centre for Health Evaluation and Outcome Sciences, St Paul's Hospital, Vancouver, British Columbia, Canada



The EQ-5D is a preference based instrument which provides a description of a respondent's health status, and an empirically derived value for that health state often from a representative sample of the general population. It is commonly used to derive Quality Adjusted Life Year calculations (QALY) in economic evaluations. However, values for health states have been found to differ between countries. The objective of this study was to develop a set of values for the EQ-5D health states for use in Canada.


Values for 48 different EQ-5D health states were elicited using the Time Trade Off (TTO) via a web survey in English. A random effect model was fitted to the data to estimate values for all 243 health states of the EQ-5D. Various model specifications were explored. Comparisons with EQ-5D values from the UK and US were made. Sensitivity analysis explored different transformations of values worse than dead, and exclusion criteria of subjects.


The final model was estimated from the values of 1145 subjects with socio-demographics broadly representative of Canadian general population with the exception of Quebec. This yielded a good fit with observed TTO values, with an overall R2 of 0.403 and a mean absolute error of 0.044.


A preference-weight algorithm for Canadian studies that include the EQ-5D is developed. The primary limitations regarded the representativeness of the final sample, given the language used (English only), the method of recruitment, and the difficulty in the task. Insights into potential issues for conducting valuation studies in countries as large and diverse as Canada are gained.


Many difficult decisions in healthcare require value judgments. It is important to understand how society values different attributes of health to inform some of these decisions. Preference based instruments provide a classification of a respondent's health status and an empirically derived value, or preference, for that health state often from representative samples of the general population [1]. The preference for that health state can then be combined with duration to calculate outcomes such as Quality Adjusted Life Years (QALYs) [2]. While several preference based instruments are available, the EuroQol group's EQ-5D [3], [4], which describes health status by a combination of 5 attributes each comprised of 3 levels, is currently the most commonly used [5].

The first set of values for the EQ-5D health states was obtained from a sample of the general population in the United Kingdom in the early 1990s [6]. Since then, findings that peoples' health related preferences vary between countries [7] have led to several other population-based values [8], enabling policy makers to make informed decisions based on values from the population they serve. However, to date, no such values have been generated for Canada and consequently many studies have used population values from either the UK or US [9][17].

Conducting face to face interviews – the conventional method for eliciting public preferences - in a representative sample of the general adult population in Canada presents a number of logistical and resource limitation challenges. This study uses a conventional time trade-off (TTO) [18] exercise via a web survey in a sample of Canadians recruited from a market research panel and predicts values for all 243 EQ-5D health states conditional on the observed valuation data. The objective of this study is two-fold: to derive the first set of Canadian values of EQ-5D health states, and to provide insights into research designs for future valuation studies in large diverse countries such as Canada.

Materials and Methods

Ethics Statement

Ethical approval was obtained from the University of British Columbia behavioral ethics board. After being given detailed information, participants had to give written consent to begin the study.

Survey design

The survey design was a quasi-replication of previous EQ-5D studies, using a protocol modified from the initial UK study [19]. The main differences from the original methodology include: a different selection of health states, a fewer number of health states valued by each participant, the use of a web survey instead of a face to face interview, no rank or visual analogue scale (VAS) exercise and lastly, recruitment via a market research panel. Reasons for these differences include multiple study objectives (the survey also included discrete choice experiment (DCE) questions based on the EQ-5D to study a methodological objective separate to the objective addressed in this paper) and resource limitations.

In total, 48 of the 243 possible EQ-5D health states were valued. This was based on a 36 item orthogonal array [20], supplemented with 12 further health states so that the 17 health states studied in nearly all previous EQ-5D surveys were included [21], [22]. With the constraints of the other tasks in the survey, pilot work suggested each respondent would be able to complete 5 different valuations in the time allocated. Consequently, the 48 health states were blocked into 12 sets using a computer algorithm so that each block was itself near orthogonal [20].

Valuation procedure

The TTO procedure required participants to first indicate whether the health state being valued was better or worse than dead (WTD). If the health state was considered better than dead, an iterative process was used where the respondent chose between living in the health state for 10-years or full health for x years. Changing x, the number of years in full health (beginning at 5 years and either increasing up to 10 years or decreasing to 0 years) to a point where the respondent was indifferent between the two choices, gave the value for the health state (x/10). A different procedure was used for states considered WTD whereby the choice was between immediate death, and spending a length of time (10−x) in the health state being valued followed by x years in full health. The value assigned to such health states was −x/(10−x). A visual prop (time board) was used to guide respondents (figure 1) [19]. Responses were measured in 3-month increments allowing the raw TTO scores v to range from 1 to -39. For consistency with most previous studies, values considered WTD (less than zero) were replaced by a monotonic transformation (where v′ = v/(1−v)) bounding values to −0.975 [23]. The alternative transformation for values WTD considered by Shaw (referred to as a linear transformation) was considered in a sensitivity analysis [24].

Figure 1. Example of the TTO task for better and worse than dead health states.

Sampling framework

Members of a market research panel were invited to participate in the survey via email. Quota sampling was used to obtain a sample roughly representative of the age and gender of the Canadian general population. No incentive was provided specifically for participating in this survey, but participants in the panel that regularly completed surveys were offered various monthly and annual rewards.

Using previous EQ-5D studies as a guide [8], we considered including 1–2,000 respondents would obtain 5–10,000 valuations generating 25–50 valuations for each health state, sufficient to assess possible heterogeneity in preferences.

Survey Structure

Individuals that accepted the email invitation to participate in the study were referred to a password-protected secure website that contained the survey. This presented information about the study, outlined the issues for consideration by completing the survey, and then gained consent. Respondents first described their own health using the EQ-5D descriptive system. After an introductory video, respondents were asked a series of questions including the 5 TTO tasks. The first TTO task included a logical test. Finally, respondents were asked to rate their difficulty in understanding and answering the TTO. Personal characteristics were not asked in the survey, but were obtained by the market research company for all invited individuals.

Derivation of analytic sample

Respondents that failed to understand or engage with the TTO elicitation process were excluded from the primary analysis as their responses are not considered to represent their preferences. We used a variety of candidate criteria including: (i) the failure of a logical test, (ii) all 5 health states valued identically, (iii) multiple health states valued equal to 0.5, (iv) multiple health states valued as WTD, and (v) multiple logical inconsistencies between health state values (further information in Table 1). Since each of these criteria are subjective, we employed a previously described technique to determine the precise rules for inclusion [25]. This begins with a sample determined to have no problems (e.g. did not fail any of the five criteria under the most restrictive rules) and then added respondents based on iterative changes in each criteria (e.g. include respondents with 1 logical inconsistency) and tested whether there is evidence of systematic differences in the values obtained between the groups (new and old sample). This was repeated for all possible combination of criteria until the largest sample with no systematic differences in values is determined. Tests for systematic differences included: mean absolute difference (MAD) between each health states mean value; whether each of the 48 health states mean difference is statistically significantly different (using paired t-tests); the maximum difference between each mean profile value; the additional number of pairwise logical inconsistencies between mean profile values; the mean difference in the number of values WTD for each profile; and the maximum difference in the number of values WTD for each profile. A sensitivity analysis was undertaken using the whole sample.

Table 1. Problems with understanding and engaging with task from the 2033 respondents completing all five TTO tasks.

Statistical analysis

Descriptive statistics of the sample's characteristics were calculated. Comparisons between subgroups were made using t tests for interval data and χ2 tests for nominal data. Visual comparisons were made with characteristics of the Canadian general population using the Canadian Community Health Survey [26] and a previous EQ-5D study from the Canadian population [27].

TTO values were subtracted from 1 so that the dependent variable represents a measure of disutility, with a value of 1 equal to ‘immediate death’ and a value of 0 equal to full health. A random effect model was fitted using an additive specification [23], [28]. Various strategies were tested to account for interactions in the main effects. The N3 model assigns a dummy variable equal to 1 if any of the attributes was at level 3, and 0 otherwise [23]. The D1 model comprises of 4 terms: D1 represents the number of attributes with problems beyond the first and replaces the constant term; I2 represents the number of attributes at level 2 beyond the first; I2-squared is the square of I2; I3 represents the number of attributes at level 3 beyond the first, and I3-squared is the square of I3 [24].

The goodness-of-fit of models was assessed using: the square of the Pearson product-moment correlation between the observed and predicted health state values for each individual (R2), the mean absolute error (MAE) for predicting the mean observed values by health state, and the number of health states with prediction errors greater than 0.05 or 0.10 in absolute magnitude.

Other analysis

Models were re-estimated using the linear transformation for values considered WTD, and using the whole sample instead of those defined to not have problematic TTO responses. Comparisons were also made with EQ-5D values obtained from the UK and US valuation surveys [23], [24].

All comparisons were explored using graphical means, the Pearson correlation, systematic differences identified by assessing the mean absolute difference (MAD), and number of states with a difference that was greater than 0.05 and 0.1.


Sample characteristics

Of the 7482 subjects invited to participate in the survey, 2394 responded and consented (32.0%) to participate. A total of 2326 respondents began the TTO tasks (97.2% of those that began the survey), of which 2033 completed all 5 TTO tasks (87.4% of those that began the tasks). Of the 293 that failed to complete all the TTO tasks, 197 subjects did not complete even the first task.

A total of 888 (43.7%) respondents that completed all five TTO tasks were identified to have potentially failed to understand or engage with the task (see Table 1 for breakdown). The final inclusion criteria used were: not all values the same, three or fewer values considered WTD, and one or no pairwise logical inconsistencies (further details available from author). In total, 1145 respondents, or 56.3% of the 2033 that completed the TTO were included in the primary analysis.

The socio-demographic characteristics of respondents and non-respondents are shown in Table 2, along with Canadian general population statistics. It can be seen that the invited sample (groups I–IV) generally matched the Canadian general population (group V) with the exception of education (subjects with less than secondary education), and geography (substantially fewer subjects in predominantly French speaking Quebec). Respondents (groups I–III) tended to be older than non-respondents (group IV, p<0.001), which plausibly explains differences in education, household income and marital status. Subjects failing to complete all five TTO tasks (group III) were typically older (p<0.001) than those that did complete the tasks (groups I and II). Interestingly, there were fewer differences in profiles between respondents completing all tasks but identified to have failed to understand or engage with the task (group II) to those that did (group I). Exceptions were gender (more females had no problems (p = 0.009)), geography (p = 0.002), and problems in usual activities (p = 0.006).

Table 2. Study sample characteristics and comparison with Canadian general population.

Of the individuals included in the final analysis, 88% (n = 1009) deemed the TTO task as not very or at all difficult to understand, while only 3 people found the task very difficult to understand. Some 50% (n = 571) found the task not very or at all difficult to answer, while 41% (n = 467) found it fairly difficult to answer. Interestingly, the difficulties in answering the task were not statistically different to the responses from the 888 individuals identified as potentially failing to understand or engage with the task, but difficulties in understanding the task were (with included individuals finding it easier as expected).


Amongst the main sample of 1145, on average there were over 97 values for each health state (range 74–185, with the exception of worst health state where values were obtained from all respondents). Mean values for each health state ranged from 0.885 (health state 21111) to −0.309 (33333) (Table 3).

The coefficients, model fit, and prediction statistics from the regression models based on 5725 observations are shown in Table 4. All the coefficients were statistically significant (p<0.01) and logically ordered with level 2 terms positive, and level 3 terms larger than level 2 for each attribute. When not including any interactions, the model fit resulted in an R2 of 0.403 and MAE of 0.044, similar to studies in the UK and US [19], [20]. Only three of the predicted 243 health state values differed to observed values by more than 0.1 (Table 3). While the addition of the N3 interaction term resulted in a statistically significant coefficient, it did not improve the model statistics. Only 2 of the D1 interaction terms were significant at the 5% level, and while there were minor improvements in MAE and R2, the number of health states with a difference in predicted versus observed value greater than 0.1 increased from three to five. We determined the final model to therefore not include any interactions, similar to previous studies in Japan [22], Denmark [29], and Zimbabwe [30]. Models were robust to the inclusion of socio-demographic variables, with the size of coefficients changing by less than 3 decimal points, and therefore no weighting was used to correct for non representativeness of the sample.

Table 4. Parameter estimates for random effects models including N3 and D1 terms (mean (SE)), and statistics relating to the comparison between observed and predicted values.

Sensitivity analysis

The inclusion of respondents deemed to not engage or understand the TTO task modified the coefficient estimates substantially. In particular, the constant increased from 0.111 to 0.487, which means the values for mild health states are dramatically different (e.g. for health state 21111 the value is 0.493 versus 0.843 in the main model). Figure 2 compares the 243 predicted health state values between the two sets of respondents demonstrating the systematic differences (MAD = 0.240, n>|0.05| = 236, n>|0.10| = 225).

Figure 2. Sensitivity analysis – comparison between predicted values from the primary model with (i) those using the full sample and (ii) using a linear WTD transformation.

Figure 2 also compares the values for only the main sample when health states WTD were transformed using the linear and monotonic methods. As with previous findings [24], we found the choice of transformation to impact results substantially (MAD = 0.085, n>|0.05| = 159, n>|0.10| = 79).

Comparison to values from other countries

Figure 3 is a scatter plot comparing the 243 predicted health states between the present study and from the previous EQ-5D studies in the US and the UK [23], [24]. While there is a high correlation between the predicted values (Pearson's rho = 0.964 and 0.963 for the US and UK respectively), it appears that Canadian values are systematically different to both the US and UK values, placing lower values on severe health states in comparison to the US (MAD = 0.057, n>|0.05| = 121, n>|0.10| = 37) and placing higher values on severe health states in comparison to the UK (MAD = 0.169, n>|0.05| = 178, n>|0.10| = 137). The figure also shows this pattern is also found in comparisons between the 20 common observed health states and so these differences do not appear to be an artifact of different model specifications. As explored above, the differences in US values could be attributable to the linear transformation of health states WTD.

Figure 3. Comparison of observed and predicted values from Canada with those in the UK and US.


This is the first study to provide a population-based set of values for EQ-5D health states in Canada. Coefficients were logically ordered and model fit was similar to other studies, but final values differed from previous value sets in the US and UK meaning previous economic evaluations using the EQ-5D may not provide accurate information to Canadian decision makers. Researchers can apply these values to studies collecting the EQ-5D to generate QALYs based on Canadian preferences (Tables S1 and S2).

While an objective of the study was to broadly follow previous country valuations, conducting such surveys in a country such as Canada provides unique challenges for both recruitment and administration which can influence the representativeness of values and the comparability of results with other country studies. Another important objective of this study was to identify issues for researchers conducting valuation studies in Canada and the potential influence of these issues on values and resources.

Our decision to use a web panel maintained by a market research company in a survey conducted in the English language may have affected the representativeness of the invited sample in comparison to the Canadian general population. Not all (only 80%) Canadians have access to the internet [31]. Moreover, subjects in market research panels may have a greater familiarity with understanding survey questions. Finally, the language this study was conducted in – English – is the preferred language for only 67% of Canadians [32].

These limitations however should be compared with alternative designs. Face-to-face interviews could overcome some of the highlighted problems, but conducting interviews in a country as large as Canada, in particular in rural areas, would require resources many times greater than those required for this study. Recruitment to such a study would be limited by the number of people who do not have publicly listed landline telephones [33]. Such a design would also not be able to compare socio-demographics of non-responders as we were able in this study. Whether such additional resources are worth the improvement in representativeness are debatable. Values for the Health Utilities Index [34], the predominant valuation study in Canada interviewed subjects from only one city, whereas the CLAMES survey only included 146 participants [35]. In contrast, this study included over 1000 respondents from all ten provinces including rural areas. Including a French version would improve representativeness. However, care should be given to ensure an accurate translation of the survey (in addition to the already existing Canadian French version of EQ-5D) and caution to the design and analysis to ensure values from the two versions can be combined appropriately.

Using a computer to elicit values instead of face-to-face interviews also may impact the comparability of the values to previous studies [36]. A computer based TTO has been used in previous EQ-5D studies [29], but often at an interview where assistance is available and not via the web. The advantages of a computer based TTO include the potential to reduce interviewer bias, errors in question routing and data input, and easy randomization of question ordering. Disadvantages of using a computer via the web are that engagement and understanding of the TTO task appears to suffer. This study found that 12.6% did not finish all the TTO tasks, and then excluded 43.7% of the respondents from the final models due to concerns over engagement or understanding. Previous EQ-5D studies, while using different exclusion criteria, have excluded between 7% [24] and 57% [25] of respondents from final models. It is reasonable to conjecture that the use of the web partially explains these differences. There is a strong argument to exclude values from respondents that appeared to fail to understand or engage with the TTO task since their responses do not represent their preferences, however since it is important to use representative samples in final models, including the elderly and low educated, any exclusion is in itself problematic. Looking for alternative elicitation methods such as rankings [37] or DCEs [38], [39] which may be simpler for subjects to understand should be explored.

The final issue regards how values for health states respondents considered WTD are interpreted. Our results find that the choice of method influences the values, similar to other findings. Unfortunately, the choice of method is arbitrary [40]. Our primary results employ the most commonly used method, enabling a more fair comparison with previous studies. The main values from this study should be cautiously compared to those derived in the US which used a different method for transforming values considered WTD. The consequence is that depending on which method is chosen, QALY gains of different size would be generated and policy-makers might be faced with opposing conclusions based on the choices. This highlights the importance of developing elicitation methods not requiring subjective transformations. Work on the ‘lead time’ TTO [41], [42] and the DCE [38], [39] appear to be promising alternatives in early development.

In conclusion, this study provides estimates for developing QALYs based on the EQ-5D using preferences from a broadly representative sample of the Canadian population with the exception of Quebec. With the resources available for this study, we conclude that the use of the internet and a market research panel is the preferred method for generating values to be used by policymakers in countries as large and diverse as Canada in comparison to alternative design. Limitations with the design remain, and we suggest a focus on cognitively easier methods that enable more respondents to engage and understand the tasks. Including a French version of the survey, and overcoming issues with the interpretation of health states considered worse than dead would also further improve future designs. Focus should be given to these limitations before the valuation of the 5-level EQ-5D commences [43]. Until these limitations can be addressed, the value set provided in this study offers substantial improvements over using preferences from the UK or US in the Canadian context for the EQ-5D.

Supporting Information

Table S2.

List of values for all health states.



We are grateful to Huiying Sun and Daphne Guh for their technical support, and survey respondents for their participation. We are also extremely grateful for comments received from the EuroQol valuation taskforce, and from the two anonymous reviewers of the manuscript. The usual disclaimer applies.

Author Contributions

Conceived and designed the experiments: NB JB AT AA. Performed the experiments: NB JB AT AA. Analyzed the data: NB. Contributed reagents/materials/analysis tools: NB JB AT AA. Wrote the paper: NB JB AT AA.


  1. 1. Neumann PJ, Goldie SJ, Weinstein MC (2000) Preference-based measures in economic evaluation in health care. Annu Rev Public Health 21: 587–611.
  2. 2. Weinstein MC, Torrance G, McGuire A (2009) QALYs: the basics. Value Health 12: S5–9.
  3. 3. Brooks R (1996) EuroQol: the current state of play. Health Policy 37: 53–72.
  4. 4. EuroQol G (1990) EuroQol–a new facility for the measurement of health-related quality of life. Health Policy 16: 199–208.
  5. 5. Brauer CA, Rosen AB, Greenberg D, Neumann PJ (2006) Trends in the Measurement of Health Utilities in Published Cost-Utility Analyses. Value Health 9: 213–218.
  6. 6. Williams A (1995) The measurement and valuation of health: a chronicle. Centre for Health Economics Discussion paper 136. Available: Accessed 2010 Jan 12.
  7. 7. Johnson JA, Luo N, Shaw JW, Kind P, Coons SJ (2005) Valuations of EQ-5D health states: are the United States and United Kingdom different? Med Care 43: 221–228.
  8. 8. Szende A, Oppe M, Devlin N (2007) EQ-5D Value Sets: Inventory, Comparative Review and User Guide. Series: EuroQol Group Monographs 2:
  9. 9. Drolet M, Brisson M, Schmader KE, Levin MJ, Johnson R, et al. (2010) The impact of herpes zoster and postherpetic neuralgia on health-related quality of life: a prospective study. CMAJ 182: 1731–1736.
  10. 10. Conner-Spady B, Sanmartin C, Johnston G, McGurran J, Kehler M, et al. (2008) Willingness of patients to change surgeons for a shorter waiting time for joint arthroplasty. CMAJ 179: 327–332.
  11. 11. Sumukadas D, Witham MD, Struthers AD, McMurdo ME (2007) Effect of perindopril on physical function in elderly people with functional impairment: a randomized controlled trial. CMAJ 177: 867–874.
  12. 12. Cameron C, Coyle D, Ur E, Klarenbach S (2010) Cost-effectiveness of self-monitoring of blood glucose in patients with type 2 diabetes mellitus managed without insulin. CMAJ 182: 28–34.
  13. 13. Cameron CG, Bennett HA (2009) Cost-effectiveness of insulin analogues for diabetes mellitus. CMAJ 180: 400–407.
  14. 14. Dendukuri N, Khetani K, McIsaac M, Brophy J (2007) Testing for HER2-positive breast cancer: a systematic review and cost-effectiveness analysis. CMAJ 176: 1429–1434.
  15. 15. Regier DA, Sunderji R, Lynd LD, Gin K, Marra CA (2006) Cost-effectiveness of self-managed versus physician-managed oral anticoagulation therapy. CMAJ 174: 1847–1852.
  16. 16. Shrive FM, Manns BJ, Galbraith PD, Knudtson ML, Ghali WA, et al. (2005) Economic evaluation of sirolimus-eluting stents. CMAJ 172: 345–351.
  17. 17. Ware MA, Wang T, Shapiro S, Robinson A, Ducruet T, et al. (2010) Smoked cannabis for chronic neuropathic pain: a randomized controlled trial. CMAJ 182: E694–701.
  18. 18. Torrance GW, Sackett D, Thomas W (1972) A utility maximization model for evaluation of health care programmes. Health Services Research 7: 118–133.
  19. 19. Gudex C (1994) Time Trade-off User Manual: Props and Self-completion Methods. Centre for Health Economics Discussion paper 20. Available: Accessed 2010 Jan 12.
  20. 20. Kuhfeld W (1997) Orthogonal arrays. SAS. Available: Accessed 2010 Jan 12.
  21. 21. Lamers LM, McDonnell J, Stalmeier PFM, Krabbe PFM, Busschbach JJV (2006) The Dutch tariff: results and arguments for an effective design for national EQ-5D valuation studies. Health Econ 15: 1121–1132.
  22. 22. Tsuchiya A, Ikeda S, Ikegami N, Nishimura S, Sakai I, et al. (2002) Estimating an EQ-5D population value set: the case of Japan. Health Econ 11: 341–353.
  23. 23. Dolan P (1997) Modeling valuations for EuroQol health states. Med Care 35: 1095–1108.
  24. 24. Shaw JW, Johnson JA, Coons SJ (2005) US Valuation of the EQ-5D Health States: Development and Testing of the D1 Valuation Model. Med Care 43: 203.
  25. 25. Devlin NJ, Hansen P, Kind P, Williams A (2003) Logical inconsistencies in survey respondents' health state valuations – a methodological challenge for estimating social tariffs. Health Econ 12: 529–544.
  26. 26. Canada Statistics (2006) Canadian Community Health Survey (CCHS) v 3.1. Available: Accessed 2010 Jan 12.
  27. 27. Johnson JA, Pickard AS (2000) Comparison of the EQ-5D and SF-12 health surveys in a general population survey in Alberta, Canada. Med Care 38: 115–121.
  28. 28. Brazier J, Ratcliffe J, Salomon J, Tsuchiya A (2007) Measuring and Valuing Health Benefits for Economic Evaluation. 1st ed. New York: Oxford University Press. 344 p.
  29. 29. Wittrup-Jensen KU, Lauridsen J, Gudex C, Pedersen KM (2009) Generation of a Danish TTO value set for EQ-5D health states. Scand J Public Health 37: 459–466.
  30. 30. Jelsma J, Hansen K, de Weerdt W, de Cock P, Kind P (2003) How do Zimbabweans value health states? Population Health Metrics 1: 11.
  31. 31. Canada Statistics (2009) Canadian Internet Use Survey (CIUS). Available: Accessed 2010 Jan 12.
  32. 32. Statistics Canada (2006) Census Profile of Federal Electoral Districts (2003 Representation Order): Language, Mobility and Migration and Immigration and Citizenship. Available: Accessed 2010 Jan 12.
  33. 33. Statistics Canada (2008) Residential Telephone Service Survey. Available: Accessed 2010 Jan 12.
  34. 34. Feeny D, Furlong W, Boyle M, Torrance GW (1995) Multi-attribute health status classification systems. Health Utilities Index. Pharmacoeconomics 7: 490–502.
  35. 35. McIntosh CN, Gorber SC, Bernier J, Berthelot JM (2007) Eliciting Canadian population preferences for health states using the Classification and Measurement System of Functional Health (CLAMES). Chronic Dis Can 28: 29–41.
  36. 36. Norman R, King MT, Clarke D, Viney R, Cronin P, et al. (2010) Does mode of administration matter? Comparison of online and face-to-face administration of a time trade-off task. Qual Life Res 19: 499–508.
  37. 37. Salomon J (2003) Reconsidering the use of rankings in the valuation of health states: a model for estimating cardinal values from ordinal data. Population Health Metrics 1: 12.
  38. 38. Stolk EA, Oppe M, Scalone L, Krabbe PF (2010) Discrete Choice Modeling for the Quantification of Health States: The Case of the EQ-5D. Value Health 13: 1005–1013.
  39. 39. Bansback N, Brazier J, Tsuchiya A, Anis A (2011) Using a discrete choice experiment to estimate health state utility values. Journal of Health Economics. In Press (PMID 22197308).
  40. 40. Lamers LM (2007) The transformation of utilities for health states worse than death - Consequences for the estimation of EQ-5D value sets. Med Care 45: 238–244.
  41. 41. Devlin NJ, Tsuchiya A, Buckingham K, Tilling C (2011) A uniform Time Trade Off method for states better and worse than dead: feasibility study of the ‘lead time’ approach. Health Econ 20: 348–361.
  42. 42. Attema AE, Versteegh MM, Oppe M, Brouwer WBF, Stolk EA (2011) Lead time TTO: Leading to better health state valuations? Erasmus University Rotterdam. Available: Accessed 2011 Jan 12.
  43. 43. Herdman M, Gudex C, Lloyd A, Janssen M, Kind P, et al. (2011) Development and preliminary testing of the new five-level version of EQ-5D (EQ-5D-5L). Qual Life Res 20(10): 1727–36.