Advertisement

Valuing the Extended Role of Prescribing Pharmacist in General Practice: Results from a Discrete Choice Experiment

Open ArchivePublished:May 25, 2012DOI:https://doi.org/10.1016/j.jval.2012.02.006

      Abstract

      Objectives

      : To quantify patients' preferences for new pharmacist independent prescribing services in general practice for managing common existing long-term conditions compared with usual medical prescribing.

      Methods

      A discrete choice experiment cross-sectional survey was conducted in five general practices in England (October–November 2009). Four service attributes reported on the length of consultation and aspects of patient-professional interaction. A choice between three alternatives—novel pharmacist independent prescribing service (“prescribing pharmacist”), “own (family) doctor” service, and “available (family) doctor” service—was presented. Alternative regression models were compared according to their goodness of fit, and the preferred one was used to inform policy analysis.

      Results

      A total of 451 patients completed questionnaires. Respondents preferred a “pharmacist” or “own doctor” compared with “available doctor,” with a larger value given to own doctor. All attributes on patient-professional interaction were important in choosing how to manage diagnosed hypertension, while the “length of consultation” (P = 0.42) did not have any impact. The impact of introducing a pharmacist prescribing service into a general practice setting was estimated from these findings. Patients' preferences suggested that about 16% of consultations with a patient's own doctor can be switched to a prescribing pharmacist instead. Although there is a stronger preference for seeing own doctor, alternative combinations of attribute levels can be used to compensate and reconfigure a more preferred prescribing pharmacist service.

      Conclusions

      The pharmacist service is valued by patients as an alternative to doctor prescribing in primary care and therefore represents an acceptable form of service delivery when informing policy.

      Keywords

      Introduction

      In the United Kingdom, appropriately qualified pharmacists and nurses prescribe independently, within their competence, any medicine for any condition: for pharmacists, this excludes controlled drugs. The introduction of independent prescribing across the whole formulary in 2006 built upon previous forms of nonmedical prescribing (NMP); “supplementary” prescribing was first introduced in the United Kingdom in 2003 to enable pharmacists and nurses to prescribe in partnership with an independent prescriber (doctor or dentist) and within a patient-specific clinical management plan [
      Department of Health
      Extending independent nurse prescribing within the NHS in England: a guide for implementation.
      ]1. A supplementary prescriber is able to prescribe any medicine, including controlled drugs, for any condition within his or her competence. The scope of supplementary prescribing is an issue that is agreed in the patient's clinical management plan and is for the medical judgment of the independent prescriber [
      • Latter S.
      • Blenkinsopp A.
      Non-medical prescribing: current and future contributions of pharmacists and nurses.
      ]. Health care professionals able to prescribe as supplementary prescribers undertake a training program in prescribing and learning practice with a medically qualified mentor. NMP policy aims to improve patient care, choice and access, safety, use of health professionals' skills, and more flexible team working across the National Health Service (NHS) [
      Department of Health
      The NHS plan: a plan for investment, a plan for reform.
      ,
      Department of Health
      Improving patients' access to medicines: a guide to implementing nurse and pharmacist independent prescribing within the NHS in England.
      ]. It can further provide important new opportunities to reconfigure the delivery of health care that has a greater patient focus and a better use of scarce resources. For example, the community pharmacist has been shown to provide valued additional support for medication use, adherence, and review [
      • Tinelli M.
      • Ryan M.
      • Bond C.
      Patient preferences for an increased pharmacist role in the management of drug therapy.
      ].
      In a patient-focused health service it follows that to take patients' preferences into consideration we need their views on what aspects of an extended professional role is valuable. We used the discrete choice experiment (DCE) approach as the best valuation tool for assessing patient preferences for the pharmacist independent prescribing (PIP) service in the current study.
      The DCE approach is a multiattribute-based survey method for valuing benefits on a latent utility index. The approach assumes individuals derived utility from the attributes of the service rather than the service per se and choose from a set of alternatives the one that gives highest utility. The DCE tool has been used to value a variety of patient experiences of health services [
      • de Bekker-Grob E.W.
      • Ryan M.
      • Gerard K.
      Discrete choice experiments in health economics: a review of the literature.
      ], for example, “quality of professional-patient relationship” and “continuity of care.” The output from a DCE is versatile; it allows examination of the important attributes of a service and discerning their impact on choosing between alternatives. Results can inform decision makers—for example, modeling the probability that particular service configurations are taken up.
      The DCE study reported forms part of a wider evaluation of independent NMP in England [
      • Latter S.
      • Blenkinsopp A.
      • Smith A.
      • et al.
      Evaluation of Nurse and Pharmacist Independent Prescribing Department of Health Policy Research Programme Project 016 0108.
      ]. The DCE identifies and quantifies patient preferences for new PIP services in general practice for managing common existing long-term conditions. The study set out to address the following question: what are patients' preferences for who prescribes their medication? It also aimed to quantify the extent to which choosing who to consult is considered important relative to other patient experience factors, explore preference differences for prescribing services among key patient subgroups, and use the findings to predict the uptake of “prescribing pharmacist” services and assess the impact of quality improvements to the service.

      Methods

      A DCE survey asks individuals to make hypothetical (yet realistic) choices about their most preferred option from a choice of service options uniquely described by combinations of attribute levels. The approach is based on the premise that all decisions involve choice and all choices involve sacrifice. Choosing forces the respondent to value attributes against each other. Typically, respondents are asked to complete a series of such choices. The underlying assumptions are that the individual can make informed choices by weighing up the differences in attribute levels and will consider all the information provided before selecting the alternative with the highest utility. A random utility model is used to estimate the impact of attributes and contextual variables on choice (preference) [
      • Ryan M.
      • Gerard K.
      • Amaya-Amaya M.
      Using Discrete Choice Experiments to Value Health and Health Care.
      ]. We used guidance on how to design a rigorous DCE to inform the development of this study [
      • Lancsar E.
      • Louviere J.
      Conducting discrete choice experiments to inform healthcare decision making: a user's guide.
      ].

      The DCE questionnaire

      Attributes and their levels

      Four key attributes were selected to characterize differences in consulting an NMP service (i.e., PIP) or a medical prescribing service. The attribute levels were combined to describe unique alternatives important for patients yet amenable to change by the service provider. Attribute selection was based on generic characteristics known to be important to patients wanting better access to family doctor services and a quality interpersonal relationship where the professional is committed to the patient's care [
      National Primary Care Research and Development Centre
      What Patients Want from Primary Care.
      ,
      National Primary Care ResearchDevelopment Centre
      What Do Patients Want from Their GP?.
      ]. These generic attributes were tailored to represent the current context: key differences in medical prescribing and NMP for a long-term condition managed in general practice. Table 1 presents the final list of attributes and levels used. Attribute levels were sufficiently varied to distinguish between alternatives by using data from the national survey of prescribing pharmacists and the pilot study.
      Table 1Attributes and levels.
      Attribute (short name)Levels
      Length of consultation (length)5, 10, 15, 20 (min)
      Professional's words and explanations about your medicines (words)Difficult to understand Easy to understand
      Attention paid by professional to your views about medicines (attention)Appears not to listen Appears to listen
      Health review covers (review)High blood pressure only High blood pressure and review of overall health
      The attribute reporting on access to prescribing services was the “length of consultation.” Studies have shown the importance of the length of a primary care consultation in patients' experience of satisfactory appointments [
      • Wensing K.
      • Jumg H.
      • Mainz I.
      • et al.
      A systematic review of the literature on patient preferences for general practice care, part 1: description of the research domain.
      ], although the type of health problem consulted for may mitigate this effect [
      National Primary Care ResearchDevelopment Centre
      What Do Patients Want from Their GP?.
      ]. Typical doctor appointments last between 5 and 10 minutes (although “double appointments” can be booked []). In contrast, a national survey of PIPs' working practices showed consultations for a long-term condition such as hypertension lasting, on average, 18 minutes (range 5–60 minutes) [
      • Latter S.
      • Blenkinsopp A.
      • Smith A.
      • et al.
      Evaluation of Nurse and Pharmacist Independent Prescribing Department of Health Policy Research Programme Project 016 0108.
      ]. The survey was distributed to all 358 pharmacists in England registered as an independent prescriber on May 1, 2008 [
      • Latter S.
      • Blenkinsopp A.
      • Smith A.
      • et al.
      Evaluation of Nurse and Pharmacist Independent Prescribing Department of Health Policy Research Programme Project 016 0108.
      ]. Longer PIP consultations, if they also enable high-quality patient-professional interaction, may be an alternative option to support the development of frontline primary care services.
      Attributes reporting on patient-professional interaction were “professional's words and explanations about medication,” “attention paid by the professional to the patient's views about medicines,” and “the extent of review undertaken.”
      Aside from the National Primary Care Research and Development Centre's own research [], a substantial body of evidence shows that the quality of patient-professional interaction matters to patients, for example, “whether the doctor seemed to listen” [
      • Scott A.
      • Watson M.
      • Ross S.
      Eliciting preferences of the community for out-of-hours care provided by general practitioners: a stated preference discrete choice experiment.
      ] and “doctor's manner” [
      • Morgan A.
      • Shackely P.
      • Pickin M.
      • Brazier J.
      Quantifying patient preferences for out-of-hours primary care.
      ]. In a survey of patient attitudes and satisfaction with pharmacist supplementary prescribers and doctors, value was attached to whether the professional knew the patient well and/or demonstrated an interest in getting to know them as “a whole person” rather than just their illness [
      • Stewart D.
      • George J.
      • Bond C.
      • et al.
      Exploring patients' perspectives of pharmacist supplementary prescribing in Scotland.
      ]. In a study of pharmacists' preferences for extending the community pharmacy role, undertaking regular medication review was a significant influence on choice [
      • Scott A.
      • Bond C.
      • Inch J.
      • Grant A.
      Preferences of community pharmacists for extended roles in primary care.
      ] and further research has shown that patients value pharmacist advice on all aspects of their medication, general health, and lifestyle [
      • Tinelli M.
      • Ryan M.
      • Bond C.
      Patient preferences for an increased pharmacist role in the management of drug therapy.
      ]. Pharmacists' preferences have also shown that they value providing patient-centered services such as medication therapy management and chronic diseases management [
      • Grindrod K.A.
      • Marra C.A.
      • Colley L.
      • et al.
      Pharmacists' preferences for providing patient-centered services: a discrete choice experiment to guide health policy.
      ]. The policymakers too believe that promoting patient-centered care will improve patient adherence to medication [
      National Institute for Health and Clinical Excellence
      Medicines adherence: involving patients in decisions about prescribed medicines and supporting adherence.
      ].

      The choice context and the services on offer

      We again drew on the PIP survey to ensure that respondents were presented with an appropriate choice context in as realistic and understandable way as possible. It showed that the most frequent setting for prescribing was general medical practice (55% of active pharmacists worked in this setting). Hypertension was the single most common condition pharmacists managed (25% of the PIPs reported this as the condition they prescribed most frequently for). Overall, 77% of the survey respondents reported that they worked from a diagnosis made by a doctor and 40% considered that in their most common treatment area their prescribing replaced medical prescribing.
      A “labeled”-choice experiment was selected for choosing between appointments for a prescribing pharmacist, “own (family) doctor,” or “available (family) doctor” service. Such labels expect to have an intrinsic value [
      • Louviere J.
      • Hensher D.
      • Swait J.
      Stated Choice Methods: Analysis and Application.
      ,
      • Hensher D.
      • Rose J.
      • Greene W.
      Applied Choice Analysis: A Primer.
      ], and this has been demonstrated, for example [
      • Baker R.
      • Freeman G.
      • Boulton M.
      • et al.
      Continuity of care: patients' and carers' views and choices in their use of primary care services
      2001. SDO/13b/2011/ Report for the National Co-ordinating Centre for NHS Service Delivery and Organisation R+D.
      ].
      By focusing on an existing long-term condition managed by either doctor or pharmacist (hypertension), we assumed prior diagnosis at the time of the current appointment and it followed that the subsequent consultation was for managing and reviewing medication, adherence, and providing a routine health checkup. We assumed that patients wanted to attend follow-up appointments with a prescribing pharmacist, “own doctor,” or an “available doctor.” In any such appointment, the prescriber (“own” doctor, “available” doctor, or prescribing pharmacist) would be expected to check the patient's blood pressure, review his or her response to treatment (including any side effects etc.), and make any necessary changes to treatment.
      When making appointments to see a family doctor, patients prefer to consult with someone they know and trust [
      • Turner D.
      • Tarrant C.
      • Windridge K.
      • et al.
      Do patients value continuity of care in general practice? An investigation using stated preference discrete choice experiments.
      ]. On this basis, we hypothesized that an appointment with any available doctor would be less preferred than with the patient's own doctor and used this as a fixed alternative least preferred choice.
      To help respondents choose, we used a vignette that described the purpose of the appointment (Fig. 1). This was based on the PIP survey, expertise within the evaluation team, and the literature [
      • Tinelli M.
      • Ryan M.
      • Bond C.
      Patient preferences for an increased pharmacist role in the management of drug therapy.
      ,
      • Hensher D.
      • Rose J.
      • Greene W.
      Applied Choice Analysis: A Primer.
      ,
      • Baker R.
      • Freeman G.
      • Boulton M.
      • et al.
      Continuity of care: patients' and carers' views and choices in their use of primary care services
      2001. SDO/13b/2011/ Report for the National Co-ordinating Centre for NHS Service Delivery and Organisation R+D.
      ,
      • Turner D.
      • Tarrant C.
      • Windridge K.
      • et al.
      Do patients value continuity of care in general practice? An investigation using stated preference discrete choice experiments.
      ,
      • Caldow J.
      • Bond C.
      • Ryan M.
      • et al.
      Treatment of minor illness in primary care: a national survey of patient satisfaction, attitudes and preferences regarding a wider nursing role.
      ,
      • Gerard K.
      • Salisbury C.
      • Street D.
      • et al.
      Is fast access to general practice all that should matter? A discrete choice experiment of patients' preferences.
      ,
      • Hole A.R.
      Modelling heterogeneity in patients' preferences for the attributes of a general practitioner appointment.
      ].
      Figure thumbnail gr1
      Fig. 1Vignette and example of a choice.

      Creating the choice set

      Current design practice was followed to design the choice set (see [
      • de Bekker-Grob E.W.
      • Ryan M.
      • Gerard K.
      Discrete choice experiments in health economics: a review of the literature.
      ]). An online design catalog was used to derive an orthogonal fractional factorial design (i.e., uncorrelated levels of attributes) with 16 profiles (www.research.att.com/∼njas/oadir/). The second choice was created by using a systematic-level change (a standard approach in which design codes assigned to the attribute increase by a constant factor to produce a uniquely different set of alternatives [
      • Louviere J.
      • Hensher D.
      • Swait J.
      Stated Choice Methods: Analysis and Application.
      ]). Presenting a third choice (available family doctor) meant that statistical properties and statistical efficiency of the final design was checked following piloting and the identification of attribute levels for this fixed option. The total number of choices to individuals was minimized by blocking the experimental design into four different questionnaire versions incorporating four choices each (details of the experimental design are available upon request).

      Developing the DCE questionnaire for the survey

      The pilot tested the feasibility of the DCE instrument with 12 patients attending a general practice (summer 2009) and collected information on respondent's own experience of primary care (to inform levels for the fixed alternative available doctor). Respondents evaluated the vignette and attributes as plausible and questionnaire length “about right” (on average 9 minutes to complete). Respondents described a typical visit to see the available doctor by using the attribute levels. This was seeing the doctor for a 10-minute consultation, the doctor used difficult to understand words and explanations about medication, the doctor appeared not to listen well to the patient's views about medicines, and the visit covered a limited health checkup.
      The final survey instrument contained four consultation choices per questionnaire version (and a fifth choice to check for “consistency”) and background questions about the individual— demographics (age, sex), socioeconomic status (income and whether they pay for NHS prescriptions), health status (chronic health problems, health today), use of prescriptions, expectations (of getting a prescription), and experience of NMP services. An example of a choice is given in Figure 1.

      The Survey, Sample Size, and Ethics Approval

      The survey was conducted during October and November 2009 in five general practices geographically spread across England providing NMP services with health professionals working within their premises. Three practices listed a total of two NMP personnel within their staff, while two practices employed one PIP. Respondents were current patients waiting to see health professionals in the doctor surgery. Overall, each practice was asked to hand out 150 questionnaires. However, because each practice could not keep close track of the 150, it was not possible to calculate a response rate. Each practice had 38 questionnaires of each version, with a minimum target response of 105 questionnaires (70%); overall, the target was 525. This sample size provided ample opportunity to explore subgroup analysis and respondent heterogeneity, because sample size calculation was based on a minimum 100 responses per subgroup [
      • Pearmain D.
      • Swanson J.
      • Kroes E.
      • Bradley M.
      Stated Preference Techniques: A Guide to Practice.
      ]. A priori two subgroups were of particular interest—the impact of previous experience of an NMP service and consistency of respondents. The importance of experience has previously been demonstrated [
      • Baker R.
      • Freeman G.
      • Boulton M.
      • et al.
      Continuity of care: patients' and carers' views and choices in their use of primary care services
      2001. SDO/13b/2011/ Report for the National Co-ordinating Centre for NHS Service Delivery and Organisation R+D.
      ]. DCEs rely on hypothetical choices, and much debate has grown up around the issue of whether to include all respondents or only those who have answered consistently [
      • de Bekker-Grob E.W.
      • Ryan M.
      • Gerard K.
      Discrete choice experiments in health economics: a review of the literature.
      ]. Because currently available tests of consistency are not conclusive, best practice is to explore the impact of consistency by considering utility models with/without inconsistent respondents but remain cautious of deleting responses because this may be inappropriate for policy-making purposes (see discussion on the matter reported by Lancsar and Louviere [
      • Lancsar E.
      • Louviere J.
      Conducting discrete choice experiments to inform healthcare decision making: a user's guide.
      ]).
      NHS Ethics approval was obtained from Dorset Research Ethics Committee in February 2009 (REC Ref No 08/H0201/163).

      Utility models, their comparison, and theoretical validity of responses

      When analyzing responses to multiple-choice DCE survey, typically the multinomial logit (MNL) model is used, although there is growing application of alternative models when relaxing its assumptions. One example is provided by the mixed logit (MXL) model, a framework already employed when evaluating health care intervention [
      Department of Health
      Improving patients' access to medicines: a guide to implementing nurse and pharmacist independent prescribing within the NHS in England.
      ]. Our strategy sought the most parsimonious model by starting with a basic MNL model to predict choice by using all responses (MNL1) followed by an MNL specification to allow for some predefined taste heterogeneity within respondents (MNL2) and an unrestricted MXL. Two subgroups were also examined separately by using the basic additive main effects MNL model: experienced in using NMP services (MNL3) and “consistent” respondents (MNL4) (further details of the MNL and MXL models used are available upon request). Empirical models were analyzed by using BIOGEME software (http://biogeme.epfl.ch/).
      Modeling comparison was undertaken and any improvement in model fit (i.e., MNL2 or MXL compared with MNL1) was assessed by the relative increase in Pseudo R2 value and log likelihood ratio test [
      • Hensher D.
      • Rose J.
      • Greene W.
      Applied Choice Analysis: A Primer.
      ]. Given the differences in the subsamples used in MNL3 and MNL 4 models, it is not appropriate to compare them.
      The theoretical validity of responses was checked by testing the sign of attribute coefficients. A priori we expected respondents to prefer longer consultations (positive sign), easy to understand words and explanations (positive sign), more attention paid to patient's views (positive sign), and more comprehensive review (positive sign).

      Policy analysis

      Results from the utility models can be used to estimate the probability of uptake for the introduction of a new (prescribing pharmacist) service. This kind of information may be a useful contribution to policy analysis. We use results from the preferred model to estimate the impact of introducing the prescribing pharmacist service into a general practice setting. In doing so, attribute levels are predefined and resultant utility scores converted to probabilities (for details, see Equation 1, Table 4, and [
      • Ryan M.
      • Gerard K.
      • Amaya-Amaya M.
      Using Discrete Choice Experiments to Value Health and Health Care.
      ]). We also used the same utility functions to derive utility scores for given levels of own doctor and new prescribing pharmacist services to assess the impact of quality improvements to the service. In this scenario, we considered that no additional resources were available to improve a current own doctor service but that by considering trade-off information between attributes how a better service could be configured by reducing one aspect of service (attribute level) for an increase in another attribute level and how, overall, this could increase utility. When using DCE findings to inform policy analysis, the overall representativeness of the sample needs to be considered and any limitation discussed.

      Results

      Patients' responses and their background characteristics

      Questionnaires were completed from 451 patients attending all practices. Response rates were not calculable. Table 2 shows the background characteristics of respondents and the distribution of choices across the alternatives. The age of the 451 patients sampled ranged from 35 to 62 years, and the average income was in the range of £21,000 to £40,000. Two hundred fifty patients (56.1%) came to the surgery expecting to get a prescription, and 50 patients (19.9%) were expecting to see a prescribing pharmacist. When asked whether they had previously seen a nonmedical prescriber, 166 (43.0%) indicated that they had. By using a standard test of consistency, we found that ninety-five respondents (21.1%) “failed” to choose the alternative depicting “more of a good thing” and could possibly be regarded as inconsistent.
      Table 2Characteristics of sample and choices.
      Characteristics (short name)n%
      All respondents (N = 451)
       Sex (female)21751.9
       Age (age), median (IQR)48(35–62)
       Lives with a chronic disease (chronic)18140.3
       Health today
        Very good5011.2
        Good13730.7
       Neither good nor poor11826.5
        Poor12127.1
        Very poor204.5
       Usually pays for NHS prescription (payNHS)28972.3
       Income status
        Up to £20,00011928.2
        £21,000–£40,00016539.1
        More than £40,00013832.7
       Expecting a prescription today25056.1
       Expecting to see pharmacist today5019.9
       Has experience consulting nonmedical prescriber (experienced)16643.0
       Passed consistency test
      Consistent responses were identified by building in a test of “consistency” into the questionnaire. A pseudo choice that contained one superior option in a set of choices was added, i.e., dominated on all the attribute levels—a “pass” was given for the dominant option being selected.
      (consistent)
      35621.1
      Observations (N = 1779)
       Choice
       Prescribing pharmacist76142.8
       Your own family doctor98455.3
       Available family doctor341.9
      IQR, interquartile range; NHS, National Health Service.
      low asterisk Consistent responses were identified by building in a test of “consistency” into the questionnaire. A pseudo choice that contained one superior option in a set of choices was added, i.e., dominated on all the attribute levels—a “pass” was given for the dominant option being selected.
      When considering responses to the key question of patients' preferred prescribing service, as predicted, the fixed alternative, “available family doctor,” was chosen on very few occasions (2%). Rather patients' choices were more evenly distributed between the more preferred alternatives prescribing pharmacist (43%) and “own family doctor” (55%).

      Utility models

      All respondents: MNL1, MNL2, and MXL

      Table 3 presents the regression results for all models run. The first series of results relates to the basic model, MNL1, and is shown in the third column of the table. Overall, this basic model, the most largely applied, retained decent model fit (pseudo R2 of 0.406) [
      • Hensher D.
      • Rose J.
      • Greene W.
      Applied Choice Analysis: A Primer.
      ]. The impact of attributes on choice shows. There is a general preference for managing preexisting hypertension by using a prescribing pharmacist (α1 = 2.05; P < 0.01) or own family doctor (α = 2.41; P < 0.01)—these health professionals were more preferred to the fixed alternative “available doctor.” This means that respondents preferred to move from a service provided by an available doctor to an alternative offering to see a prescribing pharmacist or own doctor; however, this move was not equally preferred. The larger value given to own doctor suggests that this alternative was the more preferred.
      Table 3The regression results.
      Utility modelsAll respondentsSubgroups
      All MNL1All with interactions MNL2All MXL
      Estimated with normal distributions assigned to variables β1 to β4 by using 100 Halton draws and convergence was achieved in 1000 iterations.
      Experienced in using NMP MNL3“Consistent” respondents MNL4
      PharmacistCoefficient (α1)2.05
      Statistically significant at the 1% level.
      1.13
      Statistically significant at the 1% level.
      Coefficient (α1)2.05
      Statistically significant at the 1% level.
      2.13
      Statistically significant at the 1% level.
      2.33
      Own doctorCoefficient (α2)2.41
      Statistically significant at the 1% level.
      1.73
      Statistically significant at the 1% level.
      Coefficient (α2)2.42
      Statistically significant at the 1% level.
      2.38
      Statistically significant at the 1% level.
      2.35
      Statistically significant at the 1% level.
      Length of consultationCoefficient (β1)−0.005−0.004Mean (β′1)−0.0050.006−0.004
      SD0.000
      Professional's wordsCoefficient (β2)0.686
      Statistically significant at the 1% level.
      0.695
      Statistically significant at the 1% level.
      Mean (β′2)0.693
      Statistically significant at the 1% level.
      0.668
      Statistically significant at the 1% level.
      0.824
      Statistically significant at the 1% level.
      SD0.005
      Attention paid to patient's viewsCoefficient (β3)0.889
      Statistically significant at the 1% level.
      0.906
      Statistically significant at the 1% level.
      Mean (β′3)0.906
      Statistically significant at the 1% level.
      0.844
      Statistically significant at the 1% level.
      1.020
      Statistically significant at the 1% level.
      SD0.262
      Extent of reviewCoefficient (β4)0.173
      Statistically significant at the 1% level.
      0.173
      Statistically significant at the 1% level.
      Mean (β′4)0.171
      Statistically significant at the 1% level.
      0.1150.135
      Statistically significant at the 1% level.
      SD0.000
      Female (β5)0.898
      Statistically significant at the 1% level.
      Female (β9)0.466
      Age (β6)0.000
      Age (β10)0.000
      Chronic (β7)0.930
      Statistically significant at the 1% level.
      Chronic (β11)0.931
      Statistically significant at the 1% level.
      PayNHS (β8)0.222
      PayNHS (β12)0.222
      Number of responses1767176717676531405
      Number of respondents451451451166356
      Log likelihood (0)−1940.437−1940.437−1954.43−716.583−1543.550
      Log likelihood (model)−1153.507−1131.297−1159.76−432.939−856.736
      Psuedo R20.4060.4100.4010.3870.441
      Log likelihood ratio test44.4212.506
      Note. The preferred model MNL1 used for policy analysis is reported in italics bold.
      MNL, multinomial logit; MXL, mixed logit; NHS, National Health Service; NMP, nonmedical prescribing.
      low asterisk Estimated with normal distributions assigned to variables β1 to β4 by using 100 Halton draws and convergence was achieved in 1000 iterations.
      Statistically significant at the 1% level.
      Furthermore, all service attributes considered in the experiment, with the exception of the length of consultation (β1 = −0.005; P = 0.42), were important in choosing how to manage diagnosed hypertension and had signs as expected. The attribute “attention paid by professional to the patient's views about medicines” was judged the most important (with greatest absolute value of 0.889; P = < 0.01), and respondents preferred professionals to appear to listen. Respondents were more likely to prefer a service offering professional's words and explanations about medication that were easy to understand (positive value), professional appearing to listen to their views about medicines (“attention paid by professional to the patient's views about medicines” with positive value), and provide a comprehensive health care review covering both issues of high blood pressure and overall health (“the extent of review undertaken” with positive value).
      Next, we turn to the impact of the alternative models MNL2 and MXL. MNL2 appears to have comparable goodness of fit to MNL1; the pseudo R2 value increased to 0.410 and the log likelihood ratio test comparing this model to MNL1 is statistically significant (χ2 = 44.42, 8 degrees of freedom, and P < 0.01). It also informs some differences when taking account of individual tastes. The fourth column of the table shows that in addition to the same attributes bearing similar relative importance in the decision to choose, female respondents and those who live with a chronic health condition have stronger preferences for who they consult. Female respondents more strongly prefer to choose a prescribing pharmacist service (β5 = 0.898; P < 0.01) and those who live with a chronic health condition hold stronger (and equal) preferences for seeing a prescribing pharmacist (β7 = 0.930; P < 0.01) and own family doctor (β11 = 0.931; P < 0.01) compared with the average respondent.
      The MXL model results are shown in column 6 of the table. In this case, the model does not appear to fit as well, the pseudo R2 value decreased to 0.401, and although overall there was a statistically significant improvement in the log likelihood ratio test compared with model MNL1 (χ2 = 12.506, 4 degrees of freedom, and P < 0.01), there was no added value in the information obtained. The same general picture is true as it is for MNL1: a general preference for managing preexisting hypertension by using a prescribing pharmacist (α1 = 2.05; P < 0.01) or own doctor (α2 = 2.42; P < 0.01) and the mean of the sample population random parameter estimates taking on similar importance. That is to say that the length of consultation was found not to be important (β′1 = −0.005; P = 0.42) but the remaining attributes were. Their mean estimates were statistically significant: “attention paid by professional to the patient's views about medicines” was again judged the most important (with greatest absolute value of 0.906; P = <0.01) followed by professionals appearing to listen (β′2 = 0.693; P < 0.01) and “health care review” (β′4 = 0.171; P < 0.01). None of the estimates for the spread around the mean values, however, was statistically significant, suggesting that no further heterogeneity was identified.

      Subgroups: Respondents experienced in using NMP and consistent respondents

      When considering the impact of experience of NMP on preferences (MNL3), the model appears to have an acceptable model fit [
      • Hensher D.
      • Rose J.
      • Greene W.
      Applied Choice Analysis: A Primer.
      ] (R2 = 0.387) even though two attributes “length of consultation” (β1= −0.004; P = 0.83) and “extent of review” (β4 = 0.173; P = 0.32) are not valued (see Table 3).
      The subgroup of respondents who were judged to have passed the test of consistency (MNL4) also had a good model fit (pseudo R2 = 0.441; see Table 3). Consistent respondents preferred, to similar extents, using a prescribing pharmacist service (α1 = 2.33; P < 0.01) or own family doctor (α = 2.35; P < 0.01) compared with the fixed alternative, “available doctor,” given the similar absolute values on these coefficients. As with MNL1, who was seen was the main influence on choice with the remaining significant attributes contributing to preference in a similar (relative) order of importance.

      Policy analysis

      Output from the basic MNL1 was used for policy analysis. The regression results can be used in a number of ways, including 1 predicting the impact of introducing a new service on the take up of services (or market share) and 2 assessing the impact of quality improvements to the service.

      Predicting the impact of introducing a new service on the take up of services

      This is demonstrated in Table 4 where we assume two hypothetical options currently available in a general practice for the ongoing management of hypertension and then explore the impact of introducing a new prescribing pharmacist service.
      Table 4Probability of take up (based on MNL1).
      Probability
      Probabilities calculated by using the following formula: Pc (option 1) = exp (V1n)/Σj exp Vjn (Equation 1), where individual n will choose option 1 within a choice set C of J (j = 1, ... , J) options. Note results based on the attribute-level assumptions given as follows: Option 1: α1 = 0, α2 = 0, β1= 10 min, β2 = difficult to understand words and explanations, β3 = doctor appears not to listen to patient's views about medicines, β4 = high blood pressure review; Option 2: α1 = 0, α2 = own doctor, β1= 10 min, β2 = easy to understand words and explanations, β3 = doctor appears not to listen to patient's views about medicines, β4 = high blood pressure review and review of overall health; Option 3: α1 = prescribing pharmacist, α2 = 0, β1= 10 min, β2 = easy to understand words and explanations, β3 = doctor appears to listen to patient's views about medicines, β4 = high blood pressure review and review of overall health.
      of choice
      Option 1: “Basic”Option 2: “Own doctor service”New option: “Prescribing pharmacist”
      Before new prescribing service3.6%96.3%
      After new prescribing service introduced3.0%79.5%17.3%
      MNL, multinomial logit.
      low asterisk Probabilities calculated by using the following formula: Pc (option 1) = exp (V1n)/Σj exp Vjn (Equation 1), where individual n will choose option 1 within a choice set C of J (j = 1, ... , J) options. Note results based on the attribute-level assumptions given as follows: Option 1: α1 = 0, α2 = 0, β1= 10 min, β2 = difficult to understand words and explanations, β3 = doctor appears not to listen to patient's views about medicines, β4 = high blood pressure review; Option 2: α1 = 0, α2 = own doctor, β1= 10 min, β2 = easy to understand words and explanations, β3 = doctor appears not to listen to patient's views about medicines, β4 = high blood pressure review and review of overall health; Option 3: α1 = prescribing pharmacist, α2 = 0, β1= 10 min, β2 = easy to understand words and explanations, β3 = doctor appears to listen to patient's views about medicines, β4 = high blood pressure review and review of overall health.
      The existing services are described as 1 “basic” (or available doctor), where patients can see the available doctor for a 10-minute consultation during which they do not find the doctor appearing to listen to their views about their medicines, or the doctor using easy to understand words and explanations about medication and performing a limited health review and 2 own doctor, where patients can see their doctor for a 10-minute consultation during which they also find the doctor not appearing to listen to their views about their medicines but the doctor uses easy to understand words and explanations about medication and performs a comprehensive health review. The new prescribing pharmacist service is similar to the own doctor option except that patients see a prescribing pharmacist for a 10-minute consultation.
      We use the MNL1 regression results to calculate the indirect utility for each option and from that estimate the probability of uptake (see Equation 1, as footnote to Table 4). Table 4 shows that before the introduction of the prescribing pharmacist service, the own doctor service makes up the predominant share of total consultations, 96.3% compared with 3.6% for the basic option. After the introduction of the new prescribing pharmacist service, the utility model suggests a significant change in who is consulted. About 17% of consultations are switched from family doctor services (either own doctor or available doctor) to prescribing pharmacist services, the large majority (16%) from own doctor. This can be seen as evidence of the acceptability of a prescribing pharmacist service to patients as a means of accessing their medicines and thus as a way of alleviating pressure on scarce doctor time.

      Assessing the impact of quality improvements to the service

      We further use the MNL1 regression results to estimate utility scores obtained from given combinations of attribute levels and overall preference for health professional consulted. From this we can learn about the relative ranking of alternative service options by overall utility and trade-offs between attributes that may inform how services can be reconfigured to yield greater patient utility. Table 5 shows the estimated utility scores of a current level of service (based on own doctor prescribing) and compares it with a possible new prescribing pharmacist service. Although there is a general stronger preference for seeing own doctor (α1 = 2.41), this is not the only factor contributing to utility. Thus, alternative combinations of attribute levels can be used to compensate for a given own doctor service. This is shown in the table by a prescribing pharmacist service that pays attention to the patients' views about medicines when the own doctor service does not, all else equal. This combination of attribute levels is more preferred (the utility model predicts a utility of 3.793 compared with 3.264 for the own doctor service described, a gain of 0.529 in overall utility).
      Table 5Utility scores for alternative service configurations (based on MNL1).
      FactorNew serviceCurrent service
      LevelCoefficientLevelCoefficient
      Alternative-specific constantPIP2.05Own doctor2.41
      Length of consultation10 min−0.00510 min−0.005
      Professional′s words and explanations about your medicinesEasy to understand0.686Easy to understand0.686
      Attention paid by professional to your views about medicinesAppears to listen0.889Appears not to listen0
      Health review coversComprehensive0.173Comprehensive0.173
      Total estimated utility3.7933.264
      Gain in utility0.529
      MNL, multinomial logit; PIP, prescribing independent pharmacist.

      Discussion

      The wider study set out to evaluate NMP in England to inform planning for current and future prescribers [
      • Latter S.
      • Blenkinsopp A.
      • Smith A.
      • et al.
      Evaluation of Nurse and Pharmacist Independent Prescribing Department of Health Policy Research Programme Project 016 0108.
      ]. Part of this evaluation included the DCE study reported, a study that investigated and provided new empirical evidence of the strength of patient preferences for using prescribing pharmacists in a primary care setting. In health economics and health policy, it is important to have such information on preferences and to use them in conjunction with efficient allocation of health care resources. By using the estimated parameters from a well-fitting regression model, we demonstrated not only patients' acceptability of using prescribing pharmacists for a certain condition but also explored how patients trade between who they prefer to consult and other important qualitative aspects of the service.
      At the commencement of the study, there was limited information of current working practices for the new prescribing pharmacist role(s) or how these might be developed as frontline services. Earlier evidence of supplementary prescribing pharmacist showed then prescribing practices to be predominantly based in primary care and for cardiovascular conditions [
      • Stewart D.
      • George J.
      • Bond C.
      • et al.
      Exploring patients' perspectives of pharmacist supplementary prescribing in Scotland.
      ]. Informed further by the seminal national survey of pharmacists working practices [
      • Latter S.
      • Blenkinsopp A.
      • Smith A.
      • et al.
      Evaluation of Nurse and Pharmacist Independent Prescribing Department of Health Policy Research Programme Project 016 0108.
      ], the decision was taken to elicit patient preferences within a context of offering choice of family doctor (either own doctor or available doctor) or prescribing pharmacist consultations. The study assessed the relative importance of service attributes for appointments to manage a preexisting condition, exemplified by the cardiovascular condition hypertension. This choice context is likely to remain relevant in the future development of frontline services of this kind.
      The study suggested that our sample found prescribing pharmacist services plausible alternatives for family doctor services. Depending on the model used to explain choices, all respondents demonstrated the greatest preference for being seen by their own doctor when compared with the available doctor but also demonstrated a strong preference for being seen by the prescribing pharmacist. This meant that there was evidence to support a move from a service provided by an available doctor to an alternative offering to see a prescribing pharmacist or own doctor, all else being equal. Furthermore, the attribute “length of consultation” was not on this occasion shown to be a statistically significant predictor of choice. One explanation is that follow-up consultations of long-term conditions such as hypertension may simply not require longer consultations because the patient is sufficiently knowledgeable about his or her condition and management; rather, it is the fine detail and quality of service that matters. Therefore, what appeared to subsequently matter to the level of patients' utility, and in order of relative importance, were attributes reporting on patient-professional interaction: “attention paid by the professional to the patient's views about medicines,” followed by “professional's words and explanations about medication,” and “the extent of review undertaken.”
      It was also observed that when key contextual variables were explored as further determinants of choice, statistically significant differences in tastes were limited to respondents' gender and health status. In our sample, it was female respondents who held stronger preferences for choosing a prescribing pharmacist service while those who live with a long-term health condition held equally stronger preferences for a prescribing pharmacist service and own doctor service when compared with the available doctor service. As a result, one policy recommendation may be to provide prescribing pharmacist services at locations and times at greatest convenience to women and individuals with existing long-term conditions. Findings from an RCT evaluating a community-pharmacy–based medicines management service for chronic patients with coronary heart disease showed females more satisfied with the new pharmacy service [
      • Tinelli M.
      • Blenkinsopp A.
      • Bond C.
      Community Pharmacy Medicines Management Evaluation Team
      Development, validation and application of a patient satisfaction score for a community pharmacy medicines management service.
      ].
      While others have argued the importance of exploring the extent to which individuals may have different preferences and that some of this may be unrelated to observable personal characteristics [
      • Hole A.R.
      Modelling heterogeneity in patients' preferences for the attributes of a general practitioner appointment.
      ], this study did not uncover any further preference heterogeneity of this kind, as evidenced by the lack of statistical significance in the estimates for the SD values associated with attributes in the MXL model that was estimated.
      An interesting finding was the difference found between relative preferences obtained by the complete sample and a more limited sample judged to have answered the DCE consistently. Here the main finding was that consistent respondents appeared to be more equally balanced in their preference for a prescribing pharmacist and family doctor than the whole sample. As previously discussed, there is much debate in the literature about what to do with such preferences. In the current study, the main impact of our subgroup findings served to strengthen the relative importance of prescribing pharmacists.
      When the results were used to explore the take up of new prescribing pharmacist services under different states of the world, our model showed that of the 17% of new prescribing pharmacist consultations, most (16%) could expect to arise from a switch from own family doctor services.
      We further explored how the DCE could be used to exploit “trade-off” information by considering whether patients might prefer to see a prescribing pharmacist if they were “compensated” by particular changes in the level of other service attributes. In the example given, we showed how an improvement in the attribute “attention paid by professional to patients' views about medication” yielded higher utility than the own family doctor service, all else being equal.
      Clearly, there are a number of strengths and weaknesses of an empirical study such as this one. A particular strength was using an evidence-based approach in planning the intervention, basing the research on key findings from the health care literature and previous DCE applications to pharmacy research as well as the national survey in the first phase of the wide study. The choice set for this study included an opt-out or status quo alternative with attribute levels defined during piloting work. This provided a further strength when limiting overestimation of responses [
      • Boyle K.L.
      • Holmes K.J.
      • Teisl T.P.
      • et al.
      A comparison of conjoint analysis response formats.
      ]. In the process of creating the experiment, contemporary issues about measuring design efficiency and choosing the most appropriate design available in terms of its statistical properties were considered. A further strength of the DCE method is that alternative DCE modeling can be considered and a strategy for choosing the preferred model developed [
      • Hensher D.
      • Rose J.
      • Greene W.
      Applied Choice Analysis: A Primer.
      ]. In this case, a reasonably extensive strategy was adopted although there is always scope for considering alternatives (e.g., different assumption in MXL models or latent class models; see [
      • Grindrod K.A.
      • Marra C.A.
      • Colley L.
      • et al.
      Pharmacists' preferences for providing patient-centered services: a discrete choice experiment to guide health policy.
      ] for latent class model with application to pharmacy). The impact of consistency of responses and patient experience of NMP on regression modeling was also considered. The implication of alternative models on policy decision making might also be a matter of discussion.
      A possible limitation of the study was the representativeness of the respondents and therefore generalizability of findings. Although the five sites involved in the study for data collection were spread across England and delivered comparable NMP services [
      • Latter S.
      • Blenkinsopp A.
      • Smith A.
      • et al.
      Evaluation of Nurse and Pharmacist Independent Prescribing Department of Health Policy Research Programme Project 016 0108.
      ], the overall representativeness of the sample remains unknown and no other evaluation of NMP services in the United Kingdom are available to compare findings. Poor patient responses from three specific general practice sites (i.e., fewer than 50 questionnaires per practice) could not allow testing for variability in patient preferences across sites. Any possible issue in the representativeness of the sample could have an impact on policy analysis and the use of findings to support any policy change.
      The DCE survey involved a separate sample of patients from the one replying to a wider patients' experience questionnaire that took place within the wider study [
      • Latter S.
      • Blenkinsopp A.
      • Smith A.
      • et al.
      Evaluation of Nurse and Pharmacist Independent Prescribing Department of Health Policy Research Programme Project 016 0108.
      ], but this two-phase survey strategy did not reduce the number of patients who were available to be given a DCE questionnaire. In both surveys we aimed at distributing a questionnaire to comparable samples from each participating site (200 for the patients' experience questionnaire and 150 for the DCE questionnaire). Unfortunately, data from the two surveys could not be linked and patients' views and experiences on their direct consultation with the independent prescribers could not be incorporated within the DCE modeling. A further limitation to the study is that response rates were not calculable because questionnaires handed out could not be tracked.
      The experimental design used for this application was created by adding a fixed opt-out option to a binary orthogonal choice set derived from a catalog of orthogonal arrays with application of foldover approach [
      • Louviere J.
      • Hensher D.
      • Swait J.
      Stated Choice Methods: Analysis and Application.
      ]. Alternative approaches are now available to researchers to create multiple-choice designs that are statistically efficient [

      Rose J. Issues in experimental design and sample size for discrete choice experiments. Presented at: The Third Conjoint Analysis in Health Conference. Newport Beach, CA, 2010.

      ]. Recent developments include using prior assumptions about parameters to improve statistical efficiency [
      • Bliemer M.C.J.
      • Rose J.
      Experimental design influences on stated choice outputs: an empirical study in air travel choice.
      ] and construct multiple-choice designs with customized reference alternatives for each respondent (http://www.choice-metrics.com).
      Other aspects, such as possible concerns about the appropriateness of the health care received, might be important to patients when choosing between different health care packages, although for our study the choice of these specific DCE attributes was supported by evidence from the literature, discussion with experts, and pilot work with patients. Unfortunately, no costing data on the delivery of the alternative services were considered. Future work should integrate costing and DCE output within a cost-effectiveness framework to investigate how preferences (and their heterogeneity) might influence cost-effective decisions.
      In conclusion our study demonstrates that patients have valid preferences for how primary care prescribing services in general practice for long-term conditions are delivered. On this occasion, the pharmacist prescribing service is valued by patients as an alternative to doctor prescribing and therefore represents an acceptable form of service delivery when informing policy and practice. In turn, it is important that policymakers and practices take note of these preferences.

      Acknowledgments

      We thank all the patients and practice staff who contributed to this study. We also thank Nick Thayer and Bernard Naughton, who cleaned, entered, and checked the survey data, and Andrew Sibley, who assisted with data collection.
      Source of financial support: This is an independent report commissioned and funded by the Policy Research Programme in the Department of Health. Karen Gerard is funded by a National Institute for Health Research Career Development Award and Michela Tinelli by an ESRC/MRC Interdisciplinary Postdoctoral Fellowship. The views expressed by the authors are their own and not those of their funders.

      References

        • Department of Health
        Extending independent nurse prescribing within the NHS in England: a guide for implementation.
        (Accessed February 27, 2012)
        • Latter S.
        • Blenkinsopp A.
        Non-medical prescribing: current and future contributions of pharmacists and nurses.
        Int J Pharm Pract. 2011; 196: 381-382
        • Department of Health
        The NHS plan: a plan for investment, a plan for reform.
        ([Accessed February 27, 2012])
        • Department of Health
        Improving patients' access to medicines: a guide to implementing nurse and pharmacist independent prescribing within the NHS in England.
        ([Accessed February 27, 2012])
        • Tinelli M.
        • Ryan M.
        • Bond C.
        Patient preferences for an increased pharmacist role in the management of drug therapy.
        In J Pharm Pract. 2009; 17: 275-282
        • de Bekker-Grob E.W.
        • Ryan M.
        • Gerard K.
        Discrete choice experiments in health economics: a review of the literature.
        Health Econ. 2012; 21: 145-172
        • Latter S.
        • Blenkinsopp A.
        • Smith A.
        • et al.
        Evaluation of Nurse and Pharmacist Independent Prescribing.
        University of Southampton, Southampton2010
        • Ryan M.
        • Gerard K.
        • Amaya-Amaya M.
        Using Discrete Choice Experiments to Value Health and Health Care.
        Springer, Dordrecht, The Netherlands2008
        • Lancsar E.
        • Louviere J.
        Conducting discrete choice experiments to inform healthcare decision making: a user's guide.
        Pharmacoeconomics. 2008; 26: 661-677
        • National Primary Care Research and Development Centre
        What Patients Want from Primary Care.
        University of Manchester, Manchester2005
        • National Primary Care ResearchDevelopment Centre
        What Do Patients Want from Their GP?.
        University of Manchester, Manchester2008
        • Wensing K.
        • Jumg H.
        • Mainz I.
        • et al.
        A systematic review of the literature on patient preferences for general practice care, part 1: description of the research domain.
        Soc Sci Med. 1998; 47: 1573-1588
        • Department of Health
        GP Workload Survey.
        ([Accessed October 15, 2011])
        • National Primary Care Research & Development Centre
        ([Accessed October 15, 2011])
        • Scott A.
        • Watson M.
        • Ross S.
        Eliciting preferences of the community for out-of-hours care provided by general practitioners: a stated preference discrete choice experiment.
        Soc Sci Med. 2003; 50: 803-884
        • Morgan A.
        • Shackely P.
        • Pickin M.
        • Brazier J.
        Quantifying patient preferences for out-of-hours primary care.
        J Health Serv Res Policy. 2000; 5: 214-218
        • Stewart D.
        • George J.
        • Bond C.
        • et al.
        Exploring patients' perspectives of pharmacist supplementary prescribing in Scotland.
        Pharm World Sci. 2008; 30: 892-897
        • Scott A.
        • Bond C.
        • Inch J.
        • Grant A.
        Preferences of community pharmacists for extended roles in primary care.
        Pharmacoeconomics. 2007; 25: 783-792
        • Grindrod K.A.
        • Marra C.A.
        • Colley L.
        • et al.
        Pharmacists' preferences for providing patient-centered services: a discrete choice experiment to guide health policy.
        Ann Pharmacother. 2010; 44: 1554-1564
        • National Institute for Health and Clinical Excellence
        Medicines adherence: involving patients in decisions about prescribed medicines and supporting adherence.
        ([Accessed February 27, 2012])
        • Louviere J.
        • Hensher D.
        • Swait J.
        Stated Choice Methods: Analysis and Application.
        Cambridge University Press, Cambridge2000
        • Hensher D.
        • Rose J.
        • Greene W.
        Applied Choice Analysis: A Primer.
        Cambridge University Press, Cambridge2005
        • Baker R.
        • Freeman G.
        • Boulton M.
        • et al.
        • Continuity of care: patients' and carers' views and choices in their use of primary care services
        2001. SDO/13b/2011/ Report for the National Co-ordinating Centre for NHS Service Delivery and Organisation R+D.
        ([Accessed February 27, 2012])
        • Turner D.
        • Tarrant C.
        • Windridge K.
        • et al.
        Do patients value continuity of care in general practice?.
        J Health Serv Res Policy. 2007; 12: 132-136
        • Caldow J.
        • Bond C.
        • Ryan M.
        • et al.
        Treatment of minor illness in primary care: a national survey of patient satisfaction, attitudes and preferences regarding a wider nursing role.
        Health Expect. 2007; 10: 30-45
        • Gerard K.
        • Salisbury C.
        • Street D.
        • et al.
        Is fast access to general practice all that should matter?.
        J Health Serv Res Policy. 2008; 13: 3-10
        • Hole A.R.
        Modelling heterogeneity in patients' preferences for the attributes of a general practitioner appointment.
        J Health Econ. 2008; 27: 1078-1094
        • Pearmain D.
        • Swanson J.
        • Kroes E.
        • Bradley M.
        Stated Preference Techniques: A Guide to Practice.
        Steer Davies & Gleave Ltd., Richmond, Surrey1991
        • Tinelli M.
        • Blenkinsopp A.
        • Bond C.
        • Community Pharmacy Medicines Management Evaluation Team
        Development, validation and application of a patient satisfaction score for a community pharmacy medicines management service.
        Int J Pharm Pract. 2011; 19: 144-155
        • Boyle K.L.
        • Holmes K.J.
        • Teisl T.P.
        • et al.
        A comparison of conjoint analysis response formats.
        Am J Agric Econ. 2001; 83: 441-454
      1. Rose J. Issues in experimental design and sample size for discrete choice experiments. Presented at: The Third Conjoint Analysis in Health Conference. Newport Beach, CA, 2010.

        • Bliemer M.C.J.
        • Rose J.
        Experimental design influences on stated choice outputs: an empirical study in air travel choice.
        Transp Res Part A Traffic Psycol Behav. 2011; 45: 63-79