The study
An explanatory cluster randomised crossover trial was carried out which involved training GPs in the use of risk communication tools (information aids including numerical formats and charts), SDM competencies, or both. Patients with one of four chronic conditions (menorrhagia, atrial fibrillation, menopausal symptoms or prostatism) were invited by their GP to attend a ‘review consultation’ to discuss their continuing treatment. Decisions taken at these consultations related to previously diagnosed conditions with which the patients were familiar and may have already experienced treatment.
The study was in three phases.13 Twenty GPs from 20 different practices in Gwent, South Wales, UK, were recruited and consenting patients from their practices were invited to review consultations within one of three study phases (Figure 1). Each doctor would consult 48 patients: 12 in phase 1 (pre-training = control group), 24 in phase 2 where half the doctors are trained in risk communication and half trained in SDM, and 12 in phase 3 where all doctors receive the training they did not receive in phase 2 and were thus all now trained in both. The trial assessed the resulting effects on the doctor–patient interaction (level of SDM in consultations14), patient outcomes (SF-12)15 and satisfaction with the decision taken.12 Results of these aspects of the trial are available elsewhere.16,17 Briefly, training produced statistically and clinically significant changes in the process of consultations although patient-based outcomes showed no significant effects. All patients who attended the review consultation were eligible for the discrete choice experiment (Figure 2).
Figure 1 Randomised trial design with crossover.
Figure 2 Flow of patients.
The discrete choice experiment
Developed in the early 1980s, discrete choice experiments are increasingly being used by health economists to elicit patient preferences for different forms of healthcare delivery.18 Conventional economic theory asserts that consumers gain ‘utility’ (satisfaction) from consuming goods and services. One formulation of this theory asserts that goods and services can be described in terms of their utility bearing characteristics (attributes) and consumers demand those goods and services whose variations in attributes provide the highest utility.19 In the context of a GP consultation, the relevant (that is, utility bearing) attributes can include the length of the consultation, whether the doctor listens and so on.
How this fits in
Shared decision making (SDM) is widely advocated by health policy makers. This study shows that patients do value this approach, but not as much as other key attributes of consultations such as having a doctor who listens and being provided with easily understood information. In a randomised controlled comparison, the importance that patients attach to SDM increased among the patients who had experienced this approach. This suggests that SDM may gain greater acceptance among patients more widely once they have experienced it.
Discrete choice experiments involves asking patients to choose between pairs of scenarios that differ in terms of these consultation attributes. For example the attribute ‘who makes the treatment decision’ could be described in one scenario including the option ‘doctor decides’ and in the other the option ‘patient decides’. Such stated preference techniques are favoured in health economic analyses because they are grounded in utility theory (responders choose the alternative which give them the highest utility) and they simulate the types of decisions that individuals are accustomed to making in everyday life.20 The five stages of a discrete choice experiment are described below.
Identify attributes. Literature reviews on SDM and risk communication21,22 were used to identify key characteristics of a ‘successful’ consultation (Table 1). ‘Doctor listens’ implies patients want to transmit information to the doctor;23 ‘amount of information’ implies that information from the doctor is valued; ‘information easy to understand’ reflects the doctor's ability to explain health problems and alternative treatments; ‘who chooses your treatment’ reflects the patient's wish to be involved in the decision process;9 and ‘consultation length’ measures patients' preferences for more time with the doctor.
Table 1 Attributes and options included in the discrete choice experiments.
Choosing options for each attribute. The term ‘option’ is used here to reflect the ways that each attribute is described, that is, how the attribute varies. To retain the realistic nature of the comparison between scenarios it is important that the way that the options are described to reflect realistic choices (for example; doctor, patient, both for the attribute ‘who chooses your treatment’). These were informed by the literature reviews (Table 1).
Presenting scenarios. All possible combinations of attributes and options produces 72 scenarios (three attributes with two options plus two attributes with three options = 32 × 23 = 72, the full factorial design). However, evidence shows that responders usually manage between nine and 16 comparisons before losing interest.24 A sub-sample of 27 scenarios was deemed feasible while allowing enough degrees of freedom to perform the required analysis. This gave a relative theoretical D-efficiency of 0.32 compared to the full factorial design, and 1.43 compared to an orthogonal 1/6 fractional design.25
Within discrete choice experiments, one scenario is used in every choice (constant) and all others compared with it. Although the constant can be selected randomly,26 the scenario judged to most closely reflect current practice was chosen as the constant. A scenario assumed to be unambiguously of lower utility24 (‘worst’ scenario) was composed of ‘doctor does not listen’, ‘a small amount of information’, ‘information difficult to understand’, ‘doctor chooses’ and ‘less than 10 minutes’. This was used to assess if responders were making rational choices, such as choosing scenarios yielding more utility over those yielding less. Also, a scenario considered to hold highest utility (‘best’ scenario) was selected (mainly for statistical purposes). It was composed of ‘doctor listens’, ‘a large amount of information’, ‘information easy to understand’, ‘both choose’ and ‘more than 10 minutes’. This is a subjective judgement and responders can legitimately (that is, rationally) have different views of best and worst. In addition to these, 24 other scenarios were chosen to ensure balance between options as well as ‘utility balance’.27
To avoid overburdening patients, two questionnaires were developed. Each included comparison of the constant with ‘best’ and ‘worst’ scenarios with the remaining 24 comparisons randomly divided between the two. Thus, each patient was presented with 14 pair-wise choices; Table 2 shows one pair-wise choice. The scenario on the left represent the constant scenario (that is, current practice) as indicated by the GPs in the research group.
Table 2 Example of pairwise choice choice used for the discrete choice experiments.
The questionnaires explained the aim of the exercises and gave a brief description of the consultation attributes and options attached to them. A ranking exercise, where patients were asked to rank the attributes independently of the scenarios was included. In addition to the pair-wise choices, a series of questions was included to assess the patient's experience at the last consultation, their degree of difficulty in completing the questionnaire (likert scale) and to estimate the time taken to complete the questionnaire.
Administering and piloting the questionnaire. The questionnaire was sent by post to all study patients 6 months after attending the review consultation to allow time for reflection. Reminders were sent to non-responders 2 and 4 weeks after first mailing. The first 72 questionnaires were used for piloting.
Data analysis. Responders and non-responders were compared on the basis of demographic data, clinical condition and SF-12 collected at the review consultation. Patients who always chose the constant scenario were compared with the rest on demographic and clinical variables.
The discrete choice experiment was analysed with a multilevel logistic regression model28,29 using MlWin software. This takes into account correlations at practice level (level 3), individual (level 2) and the multiple responses from within each individual (level 1). The dependent variable was the probability that a patient included the attributes options including whether the doctor had been trained in risk communication, SDM or both, would choose the alternative scenario. The explanatory variables included the attributes and whether the doctor had been trained in risk communication, SDM or both. Dummy variables were used for attributes with three options to avoid assuming that the changes between attributes options were ordinal and that, for instance, little to moderate amount of information gives the same utility gain as moderate to large amounts of information. The main effects in the model described above show the importance that patients place on changing different attribute options of a consultation. To explore whether these remain stable as patients experience consultations with doctors who had been trained, interactions between the attributes and training were examined. It should be noted that patients were randomly allocated to consultations with doctors who had or had not been trained in risk communication, SDM or both and not to a consultation which followed any prescribed model. Nevertheless, within the trial, provision of each form of training significantly increased the level of patient involvement in treatment decisions,14 which suggests that the processes of these consultations were significantly different than in the baseline phase of the trial and specifically manifesting patient involvement competences.
