Population preferences and choice of primary care models: A discrete choice experiment in Sweden

Transcription

1 Health Policy 83 (2007) Population preferences and choice of primary care models: A discrete choice experiment in Sweden Jonas Hjelmgren, Anders Anell The Swedish Institute for Health Economics (IHE), P.O. Box 2127, SE Lund, Sweden Abstract Objective: To examine which attributes are important when individuals choose between primary care models. In particular, we studied whether individuals that were given a choice preferred individual family physicians (GP) or a primary care team consisting of physicians and nurses (PCT). Method: A questionnaire survey, designed as a discrete choice experiment, was sent to 1600 individuals in Sweden. In the questionnaire, different primary care models were constructed on the basis of five attributes: GP versus primary care team, waiting time for non-emergency visits, user charges, ability to choose provider, and degree of influence over the care received. Individual preferences and willingness to pay for attributes were regressed against such characteristics as age, gender, education, and health status (as measured by EQ-5D). Results: The response rate was 58%. Waiting time, user charges, ability to choose provider, and degree of influence over the care received were each statistically significant discriminants (p < 0.001). Willingness to pay per visit was 224 SEK for great compared to limited degree of influence over the care received, 164 SEK for ability to choose a provider and 111 SEK for each 1 day reduction in waiting time for non-emergency visits. Subgroup analysis showed that older individuals and individuals in poor health preferred the option to register with a GP whereas working individuals and individuals living at a greater distance from a hospital preferred the option to register with a primary care team Elsevier Ireland Ltd. All rights reserved. Keywords: Primary care organisation; Sweden; Population preferences; Discrete choice 1. Introduction Organising primary health care is a difficult task, balancing such different objectives as efficiency, equity and responsiveness. An ambitious Canadian study, for example, evaluated the strengths and weaknesses of Corresponding author. Tel.: ; fax: address: jh@ihe.se (J. Hjelmgren). two competing models, health centres with geographical responsibility (such as traditionally practiced in Sweden and Finland) and the GP s with individual registration (as practised, for example, in the UK, Denmark and the Netherlands), but could not reach a definitive conclusionabout which primary care model is best [1]. While pros and cons could be identified, the choice between the two models involves a necessary trade-off between objectives that should be informed /$ see front matter 2007 Elsevier Ireland Ltd. All rights reserved. doi: /j.healthpol

2 J. Hjelmgren, A. Anell / Health Policy 83 (2007) by population preferences. As noted in other empirical studies [2], primary care services in several European countries are presently evolving towards group practices and more reliance on multidisciplinary teamwork. This will ultimately impact individual choice and offers an additional argument for researching population preferences about primary care organisation. Primary health care in Sweden has traditionally been organised around health centres with a broad professional competence that have responsibility for the delivery of health care within a geographical area. In the late 1980s, however, Swedish county councils responsible for the organisation of health care were criticised for the lack of consumer choice and influence. In 1993, inspired by the primary care models used in UK, Netherlands and Denmark, the then ruling conservative coalition government legislated a new national system of family physicians which featured consumer choice and funding following individual choices, a contrast from the previous system of health centres with geographical responsibility [3]. Following elections in 1994, a new Social-Democrat minority government came into office, and the new parliament voted to abolish the act. Decision-making related to the organisation of primary care was again decentralised to the 21 county councils, some of which continued to develop a system of family physicians with consumer choice, whereas others returned to health centres with a geographically defined responsibility. Currently, thus, there are several models of providing primary health care operating in the different county councils in Sweden. The debate of which model or models to select for the future has been intense, as different stakeholders favour different solutions [4]. Unfortunately, very little is known about preferences across the population and if the choice of different primary care models depends on characteristics, such as age, education, health status and residence in an urban or rural area. The purpose of this study is to evaluate population preferences for different primary care models in Sweden. In particular, we studied the importance of patient choice and if individuals that were given a choice preferred to choose an individual family physicians (GP) or a primary care team of physicians and nurses (PCT). To account for the tradeoffs between various organisational attributes related to different primary care models, we used discrete choice experiment. 2. Discrete choice modelling Discrete choice experiment (DCE) has previously been widely used to derive individual preferences in the fields of market research, transport economics and environmental economics [5 9]. This method has also been used in the health care context, e.g. to value patient benefits from health care services [10]. DCE methods are drawn upon Lancaster s economic theory of value, which states that the utility (U) an individual (i) obtains from consuming a good (q) is a function of the sum of its attributes ( X j ) [11] (Eq. (1)) U jq = α j X ijq (1) where α j should be interpreted as marginal utility weights of attribute j. If we assume that an individual faces a choice between two types of primary care models with different attributes (model A versus model B) he/she will choose, ceteris paribus, the alternative that provides most utility. The net welfare gain from choosing model A over model B would be (Eq. (2)): U = α j X ija α i X ijb = α i X ijq. (2) A DCE is conducted within a random utility theory (RUT) framework [12 15]. RUT proposes that respondents know their own utility U q for any good with certainty. The researcher is, however, unable to observe the individuals utility instead he/she observes the attributes of the goods that are selected and not selected. This means that the utility of the individuals only can be estimated by generalised prediction models (RUT models), which incorporate uncertainty. In order to take account of the randomness of individuals choices Eq. (2) can be expressed as Eq. (3), where the utility function of alternative (A) (U A )is separated into a systematic or measurable component (V) and a stochastic component (ε). According to RUT an individual will choose alternative A over another alternative B if (Eq. (3)): V ia ( αja X ija ) + ε A >V ib ( αjb X ijb ) + ε B (3) By assuming a linear utility function the relationship between the change in utility ( V = V ia V ib ) and the attributes of A and B ( α ia X ija α ib X ijb )ina

3 316 J. Hjelmgren, A. Anell / Health Policy 83 (2007) pairwise choice scenario can be expressed as Eq. (4): V = (α 0A α 0B ) + α j X ijq + e i (4) where α 0A α 0B is the constant term in the empirical model (e.g. a random effects logit model), α i represents the utility parameter of the attribute X j, and e i comprises all unobservable factors in the utility function. The estimated parameters in the empirical model can be used to estimate the trade off (marginal rate of substitution) of one attribute in terms of another. One important trade off is that of the price proxy (e.g. user charge) and one of the other attributes, which could be interpreted as the willingness to pay (WTP) for an attribute. The WTP for attribute j is obtained by dividing the parameter value α j with the price proxy parameter (i.e. α j / α price ). 3. Design of experiment and survey 3.1. Study outline The study was designed as an experiment using a postal survey, where respondents were presented with choices between alternative primary care models with different attributes. The questionnaire comprised four sections. The first section included questions about general background (sex, age, number of people in the household, present health status (EQ-5D), educational level, working situation). The second section included questions about experience with primary care services (travel distance to nearest health centre and hospital, number of visits to primary care physician latest 12 months, present registration with GP and general confidence in available primary care services). In the third and fourth sections, patient preferences for different primary care attributes were solicited independently (Section 3) and simultaneously based on individual choices of defined primary care models (Section 4). The questionnaire was tested for clarity and interpretation on 15 individuals of different age and background. After minor revisions the final questionnaire was distributed to 1600 randomly selected individuals in Sweden between the age of 18 and 85. The sample was stratified by residence: one stratum included the three largest cities (50% of the sample) and the other the rest of the country. Reminders were sent at 2 and 4 weeks following the TDM method [16] Preference assessment The study focused on two important policy issues related to the organisation of the Swedish primary care: (1) the value of being able to choose your own primary care provider compared to a policy with no choice and geographical responsibility for providers, and (2) the value of choosing (registering with) a GP versus the value of choosing (registering with) a primary care team comprising both physicians and nurses (see Table 1 for description of alternatives). Three additional attributes where included in the constructed alternatives: the degree of patient influence, user charges and waiting times for non-emergency visits. By including user charges as one of the attributes, the relative value of different attributes can be expressed in monetary terms. It should be noted that all individuals above 18 years in Sweden currently face co-payments of between SEK 100 and 150 per visit so the inclusion of user charges as an attribute is consistent with respondents expectations. Patient influence was included since this attribute has proven very important in previous and similar studies [17 20] and thus serves as a good benchmark measure for comparing the various attributes of interest in our study. Waiting time for Table 1 Definitions of primary care models given to respondents General practitioner You are registered with a general practitioner In case of non-acute conditions, you will always be treated by the same physician. Less severe conditions can, if possible, be handled by a district nurse In case, there is a need the general practitioner can refer you to a specialist or health care professional Primary care team You are registered with a primary care team consisting of several physicians and nurses Your condition or disease will determine which physician or nurse you will meet In case, there is a need the physician can refer you to a specialist or health care professional

4 J. Hjelmgren, A. Anell / Health Policy 83 (2007) non-emergency services was included since policies to shorten waiting times in Swedish primary care have been an important political priority. The first part of the preference assessment section included simple assessments related to the individual attributes one at a time, using a Likert scale with the endpoints 0 and 5 (0 = not important and 5 = very important). In addition, respondents were asked whether they would prefer to register with a GP or with a primary care team (with equally good as a third alternative). The purpose of this first part of the preference assessment section was twofold. First, respondents were being prepared for the more difficult choices by thinking of the value of each of the five attributes one by one. Second, it was possible to compare responses and thereby assess the practical value of the more demanding discrete choice methodology. The second section involved the DCE, where the different attributes were combined in levels to form a primary care model. The choice of levels for the five different attributes is illustrated in Table 2 and reflects a pragmatic trade off between giving respondents realistic and not too complicated choices and possibilities to assess parameter values. The complete choice set consists of 96 possible combinations (the full factorial design is ). To make this more manageable, we reduced this to 16 combinations (a fractional design) based on a catalogue [21]. We Table 2 Attributes and levels used to construct alternative primary care models Attributes Levels Primary care work model Patient influence Choice for individuals User charges (per visit) Waiting time for non-emergency visit Registration with GP Registration with primary care team Large influence Limited influence Individual choice of provider (GP or team) No choice 300 SEK 200 SEK 100 SEK 0 SEK 7 days 4 days 2 days then constructed 16 choice pairs by randomly matching the original 16 combinations with a fold over designed choice set. To minimise the cognitive overload of the respondents, the 16 choice combinations were randomly split into 4 choice sets and 4 different questionnaires were developed. Each respondent received one of the questionnaires, which included only one out of four choices (i.e. four choice scenarios), each involving an individual choice between two alternatives based on the five attributes and their related values. We estimated the DCE using a random effects logit regression model in Limdep 8.0. We analysed the data on two levels: (i) whole population analysis where only the attributes were included as explanatory variables, and (ii) subgroup analyses were we investigated if subgroups of the population differed in their preferences concerning the choice of primary care models. 4. Results 4.1. Survey population The response rate was 58%. There were no significant differences between the four questionnaire versions or between the larger cities and the rest of the country. Table 3 presents the attributes of the respondents in terms of demographic variables (age and sex), the size of the household, educational level, and work force participation. As can be seen the response rate was higher for men than for women. A majority of the respondents worked full-time or part-time. About 80 90% have reported no problem in the health dimensions Mobility, Self-care and Usual activities whereas a considerably smaller fraction, between 50 and 65%, respectively, have reported no problems regarding Pain/discomfort and Anxiety/discomfort. 5. Evaluation of individual attributes Short waiting time (mean score = 3.66) was, on average, valued as the most important characteristic followed by influence over the care received (3.42), possibility to choose a GP (3.05) and low user charge (2.75). Approximately the same percent of respondents preferred the two primary care alterna-

5 318 J. Hjelmgren, A. Anell / Health Policy 83 (2007) Table 3 Characteristics of the survey population No of responses (% of total) 924 (58) Women/men (%) 41/59 Age (S.D.) 48.9 (18.5) Household size (S.D.) 2.07 (1.39) Level of education Fraction (%) Compulsory school 26.6 Gymnasium 35.4 University 34.6 No answer 3.4 Work force participation Fraction (%) Working (full-time or part time) 60.9 Other (retired, on the sick list, parental leave) 35.7 No answer 3.4 The health status (EQ-5D health dimensions) Fraction (%) Health dimension Mobility No problems walking 80.3 Some problems walking about 13.3 Confined to bed 0.1 No answer 6.4 Self-care No problems with self-care 91.0 Some problems with washing or dressing self 1.9 Unable to wash or dress self 0.6 No answer 6.4 Usual activities (e.g. work, study, housework, family or leisure activities) No problems with usual activities 82.0 Some problems with usual activities 10.1 Unable to perform usual activities 1.3 No answer 6.6 Pain/discomfort No pain or discomfort 50.2 Moderate pain or discomfort 38.4 Extreme pain or discomfort 4.7 No answer 6.6 Anxiety/depression Not anxious or depressed 64.9 Moderately anxious or depressed 26.2 Extremely anxious or depressed 2.3 No answer 6.6 tives: 38.2% preferred registration with a GP and 36.6% preferred the primary team (mean difference of 1.6%; p = 0.048). The remaining 21.3% were indifferent and 3.9% did not answer The discrete choice experiment The whole population The results from the random effects logit regression equation for the survey population as a whole are shown in Table 4. All the regression coefficients except for GP/primary care (p = ) are significant at the level, indicating that user charges, the possibility to choose a GP/primary care team, the influence over the care received, and waiting times are important attributes in the choice of a primary care model. The direction of the effects are also as expected: increased user charges and waiting times decrease the probability for an primary care model to be chosen (negative coefficient) whereas the possibility to choose a GP/primary care team and influence over the care received have the opposite effect (positive coefficient). The WTP results imply that individuals are willing to pay more to have an influence over the care they receive (SEK 224 per visit) than to have an opportunity to choose their GP/primary care team (SEK 164 per visit). The WTP result for waiting times (SEK 111 per day) implies that individuals value a 2-day reduction in waiting time (SEK 222) equally as having an influence over the care they receive. According to the assessment using the Likert scale, short waiting time (mean score = 3.66) was on average considered as more important than influence over the care received (3.42) and possibility to choose a GP/primary care team (3.05). The ratings of the attributes in the DCE allow us to be more specific about the trade offs between the attributes; here, we are able to state that a shortening of the waiting time by 1 day is worth less than having influence by the care received whereas the opposite is true for a shortening of the waiting time by 3 days Subgroup analysis In order to investigate differences at subgroup level regarding the choice between GP/Primary care team we constructed interaction variables between the specific subgroup variable and the attribute variable GP/primary care team. Three subgroup analyses were performed: (i) a model including all interaction variables ( full subgroup model ), (ii) a model including only the significant variables (p < 0.05) from the full subgroup model ( reduced subgroup model ) and (iii)

6 J. Hjelmgren, A. Anell / Health Policy 83 (2007) Table 4 Random effect logit regression results and estimated WTP in SEK for the individual attributes Attributes Coefficient p-value WTP a Constant User charge <0.001 Possibility to choose (no possibility to choose = 0) < Influence over the care received (no influence = 0) < Waiting time (per day) < GP/primary care team (GP = 0) Log likelihood function, L 3237 Restricted log likelihood function, L McFadden s R 2 (1 L/L 0 ) a The marginal rate of substitution between user charge and characteristic i = β i / β price (=willingness to pay), where β i denotes the regression coefficient for the characteristic I and β price denotes the regression coefficient for user charge. a model including all the variables from the reduced subgroup model and interaction variables between user the charge attribute and variables correlated with the income level of the individual (age, occupational level, and health status) ( reduced model including user charge interactions ). The reduced subgroup model was estimated to assess the robustness of the significant variables in the full subgroup model whereas the reduced model including user charge interactions was estimated to investigate the price sensitivity and WTP for different subgroups of the population. The other attributes were assumed to be unaffected of the subgroup effects. The results of the subgroup analysis are illustrated in Table 5. The inclusion of the interaction variables have a modest effect on the coefficients for other attributes both in the full subgroup model and in the reduced subgroup model (compare Tables 4 and 5). Notable from the results of the full subgroup model is that individuals presently registered with GP and individuals who are extremely anxious and depressed prefer to be registered with a GP instead of the primary care team (negative coefficient). Extremely anxious or depressed individuals are also on average willing to pay the largest amount, SEK 431, to be registered with a GP, which imply that these individuals could either benefit the most (implement a system with GP registration) or lose the most (abolish a system with GP registration) from policy change. Registration with a primary care team is preferred by individuals who are working and live at a distance km from the nearest hospital (positive coefficient) (p < 0.05). Individuals who live at a distance km from the nearest hospital have the highest WTP to register with the primary care team, SEK 175. The result of the reduced subgroup model confirms the results from the full subgroup model although some of the results to a minor extent either are reinforced or weakened. In the reduced subgroup model, we are unable to conclude that individuals who are presently registered with a GP have preferences for the GP or the primary care team (p > 0.05). On the other hand, some of the other variables ( extremely anxious or depressed and Working full or part time ) are more significant than in the full subgroup model. These results indicate that the significant parameters in the full subgroup model are fairly stable although some of the included variables in the reduced subgroup model and the excluded variables in the full subgroup model are correlated. Age and user charge was the only interaction that was statistically significant in the model that included user charge interactions, which suggests that individuals in different age groups react differently to price changes. It can be seen that the parameter representing the age and user charge interaction are positive meaning that older people are less price sensitive than younger. On the other hand, the age square (age age) interaction works in the opposite direction implying that individuals are more sensitive to price chances. These effects are balanced at the age of 60, meaning that after the age of 60 individuals become more price sensitive and less willing (or able) to pay for primary care. An individual 59 years of age is willing to pay SEK 474 to have a large degree of influence over the care they

7 Table 5 Random effects logit regression results of the subgroup analysis and the estimated WTP in SEK for the individual attributes Attributes Full model Reduced model Reduced model including user charge interactions Coefficient p-value WTP a Coefficient p-value WTP a Coefficient p-value WTP b WTP c Constant User charge < < <0.001 Possibility to choose (no possibility to choose = 0) < < < b 167 Influence over the care received (no influence = 0) < < < b 211 Waiting time (per day) < < < b 117 Interaction variable between the specific variable below and primary care team (general practitioner = 0) Gender (Female = 0) Age (per increased age in years) < < Urban citizen (provincial = 0) Gymnasium (compulsory school = 0) University (compulsory school = 0) Working full or part time (other occupation = 0) < < Distance to health care centre 2 5 km (<2 km = 0) Distance to health care centre>5km(<2km=0) Distance to nearest hospital km (<10 km = 0) Distance to nearest hospital > 30 km (<10 km = 0) One visit at the health care centre (no visits = 0) visits at the health care centre (no visits = 0) >5 visits at the health care centre (no visits = 0) Present registration with GP (not registered = 0) Some problems walking about (no problems walking = 0) Confined to bed (no problems walking = 0) n/a n/a Some problems with washing or dressing self (no problems self care = 0) Unable to wash or dress self (no problems self care = 0) Some problems with usual activities (no problems usual activities = 0) Unable to perform usual activities (no problems usual activities = 0) Moderate pain or discomfort (no pain or discomfort = 0) Extreme pain or discomfort (no pain or discomfort = 0) Moderately anxious or depressed (not anxious or depressed = 0) Extremely anxious or depressed (not anxious or depressed = 0) Interaction variable between the specific variable below and the user charge Age <0.001 Age square <0.001 Working full or part time (other occupation = 0) Log likelihood function, L Restricted log likelihood function, L McFadden s R 2 (1 L/L 0 ) J. Hjelmgren, A. Anell / Health Policy 83 (2007) a The marginal rate of substitution between user charge and attribute i = βi / β price (=willingness to pay), where β i denotes the regression coefficient for attribute i and β price denotes the regression coefficient for user charge. b WTP when the interaction effect is not accounted for (i.e. when age is 0 and the individual is not working full time). c When the interaction effect is accounted for the user charge parameter is (WTP is evaluated for the in the mean of age (=48.9) and age square (=2731) and the share of working full time is resulting in = ).

8 J. Hjelmgren, A. Anell / Health Policy 83 (2007) receive whereas and individual 70 years of age is only willing to pay SEK 372. By comparing the four regression models in Tables 4 and 5 we notice that the best or most effective model was the reduced model including user charge interactions (McFadden s R 2 = 0.136). However, although the included variables in this model were highly significant the R 2 did not increase by more than 23.6% compared to the whole population model (0.136/0.110). 6. Discussion The idea behind the discrete choice methodology is that individuals generally make their choices by comparing the utility from a set of combined attributes associated with the alternatives. It is entirely logical that an individual responds that shorter waiting times for health care services are very important but at the same time choose an alternative with longer waiting times if this alternative is associated with other positive benefits, e.g. more influence over the care received. This point is well illustrated in our study. According to an individual assessment using a Likert scale, attributes seem to be equally important to respondents and it is not possible to assess their relative value. Significant attributes that influence individual choices between primary care models in our whole population regression analysis are degree of patient influence, choice of provider for individuals and waiting time for non-emergency visits (p < 0.001). The average marginal willingness-to-pay for a large instead of limited influence is 224 SEK per visit. This can be compared with a willingness-to-pay of 164 SEK for the possibility to choose ones primary care provider and 111 SEK per day for shorter waiting times. The importance of patient influence match results from other studies [17 20]. It should be noted that both our own and previous studies are limited by the fact that influence was not defined specifically. Thus, each respondent had to refer to his or her own interpretation of the attribute. As suggested by previous research in Swedish primary care more influence may imply more information for some respondents, whereas others also demand influence over the clinical decision-making [20]. In addition the two levels ( large or limited ) gives a very simplistic measurement. More detailed knowledge about the value of influence would have required a different design of the experiment. For our purposes, focusing on the value of choice and if individuals prefer a GP or a primary care team, a simplistic use of the attribute is preferred. In subgroup regression analysis, preferences for choice of GP or primary care team were found to vary. The elderly and individuals with poor health preferred registration with a GP, while working individuals and individuals living at a greater distance from hospitals preferred to register with a primary care team. Preference differences by age and health status seem logical in consideration of differences in need. Elderly people and individuals with poor health status generally have a need for more doctor visits. Previous studies in Sweden have shown that individuals in these situations prefer doctors that are familiar with their health history [22]. For individuals living at a greater distance from hospitals, preferences to register with a primary care team can be interpreted as a demand for a broader supply of local health care professionals. From a Swedish policy perspective, our study results indicate that improvement in waiting times for nonemergency visits need to be balanced against other important objectives, such as possibilities of a greater influence and choice of provider. A second lesson is that the population is divided about GP s or primary care teams. Offering individuals only one of these primary care models, as is the case presently in most Swedish county councils, will consequently not satisfy much of the population. To maximise value, individuals should have the possibility to chose between the two models, and be able to change their option when needed. Such a policy would not only give benefits in terms of increased satisfaction across the population, but would also provide options for primary care doctors and nurses to work in different ways. An important question within the context of a publicly provided health care system is whose preferences should be used. Our study has a broader perspective than most other DCE studies, which are primarily based on data from either patients or service users [23]. The relatively high response rate (58%) also indicates that individuals in the general population are willing to participate in a relatively complex discrete choice experimentation if they find the research questions relevant. One limitation though, as with many other DCE studies, was the relatively simple design. We included

9 322 J. Hjelmgren, A. Anell / Health Policy 83 (2007) only 5 attributes with 2 4 levels, resulting in a reduced choice set of 16 alternatives (primary care models), disregarding possible interaction effects between the attributes. Our choice set was also limited to a choice between two primary care models and their related attributes and we also intentionally disregarded the possibility for individuals to be non-demanders in order to avoid respondents picking alternatives that were least cognitively demanding. Previous studies have shown that an opt-out alternative could potentially provide for a more market-like choice context and more accurate estimations of the included model parameters [24]. The goodness of fit of the regression models we used in our analysis was at best for the model that included interactions between selected background subgroup variables and the attribute variable GP/primary care team as well as interactions between age and occupational status) and the user charge. One reason for the low predictive power of the models was the division of the design into four blocks resulting in separate choice sets that are not entirely orthogonal [15]. References [1] Lamarche PA, Beaulieu M, Pineault R, et al. Choices for change: the path for restructuring Primary Healthcare Services in Canada. Canadian Health Services Research Foundation; [2] Saltman RB, Rico A, Boerma WGW, editors. Primary care in the driverǐs seat? Organisational reform in European primary care. Berkshire: Open University Press; [3] Anell A. Swedish Health Care under pressure. Health Econ 2005;14(S1):S [4] Anell A. Primärvård i förändring. Studentlitteratur: Lund; [5] Cattin P, Wittink D. Commercial use of con-joint analysis: a survey. J Marketing 1982;46: [6] Kroes EP, Sheldon R. Stated preference methods: an introduction. J Transport Econ Policy 1988;22: [7] Hensher DA, Barnard PO, Truong TP. The role of states preference methods in studies of travel choice. J Transport Econ Policy 1988;22: [8] Opaluch J, Swallow S, Weaver T, et al. Evaluation impacts from noxious facilities: including public preferences in current siting mechanisms. J Environ Manage 1993;24: [9] Adamowicz W, Louviere J, Williams M. Combining revealed preference and stated preference methods for valuing environmental amenities. J Environ Manage 1994;6: [10] Ryan M, Gerard K. Using discrete choice experiments to value health care programmes: current practice and future research reflections. Appl Health Econ Health Policy 2003;2(1):1 10. [11] Lancaster KJ. A new approach to consumer theory. J Polit Econ 1966;74: [12] Manski C. The structure of random utility models. Theor Decis 1977;8: [13] Train KE. Discrete choice methods with simulation. Cambridge, UK/New York: Cambridge University Press; [14] Louvier JJ, Hensher DA, Swait JD. Stated Choice Methods: Analysis and Applications. Cambridge, UK/New York: Cambridge University Press; [15] Hensher DA, Rose JM, Green WH. Applied Choice Analysis: A Primer. Cambridge, UK/New York: Cambridge University Press; [16] Dillman DA. Mail and telephone surveys. In: The Total Design Method. New York: Wiley; [17] Anell A, Rosén P, Hjortsberg C. Choice and participation in the health services: a survey of preferences among Swedish residents. Health Policy 1997;40: [18] Vick S, Scott A. Agency in health care. Examining patients preferences for attributes of the doctor patient relationship. J Health Econ 1998;17: [19] Scott A. Economics of general practice. In: Culyer AJ, Newhouse JP, editors. Handbook of Health Economics, vol. 1. Elsevier Science; [20] Rosén P, Anell A, Hjortsberg C. Patient views on choice and participation in primary health care. Health Policy 2001;55: [21] The experimental plan was based on A Catalogue and Computer Program for the Design and Analysis of Orthogonal Symmetric and Asymmetric Fractional Factorial Experiments by GJ Hahn and Shapiro SS, Report 66-C-165. May General Electric Research and Development Centre, Schenectady, New York. [22] Rosen P, Anell A, Hjortsberg C, et al. Patientpreferenser i primärvården en empirisk undersökning kring patientinflytande. IHE Working Report 1998:3. Lund, [23] Bryan S, Dolan P. Discrete choice experiments in health economics for better or for worse? Eur J Health Econom 2004;5: [24] Ryan M, Skatun D. Modelling non-demanders in choice experiments. Health Econ 2004;13: