Participating countries
The international multicentre study draws its name (Guliver) from the four centres involved: Gent University (Belgium), Utrecht University/NIVEL (the Netherlands), Liverpool University (United Kingdom) and the University of Verona (Italy).
Figure 1 shows some of the variables that describe the cultural background of the four countries from different perspectives, like Hofstede’s dimensions and geographic region.
Regarding the geographical location of the enrolled nations, although none of them is located in the Eastern Europe, the other three macro-regions (North, West and South) are present, assuring a wide variety of cultural backgrounds. As suggested by previous studies [11, 16, 17], health care public funding, quality of primary care and general practitioners gatekeeping role have been also reported in the table as possible moderator variables. These may effect modulate lay peoples’ preferences and expectations towards doctor-patient communication. In Italy and the UK, the National Health Service is based on Beveridge model [17] in which health care is provided and financed by the government through tax payments. The Netherlands and Belgium follow another model, named Bismarckmodel [17] that uses an insurance system with different health cost insurers who offer slightly different types of insurance packages.
Panel sample
A sample of 259 participants was recruited from the general population. This was balanced by a number of factors. Firstly, age, so that at least two persons were in the classes 18–30, 31–49 and >50 years of age, for a total of 6–8 participants, in order to guarantee a heterogeneous distribution in each group. Secondly, gender, with 117 males and 142 females. Finally there was a country balance with 64 in the Netherland, 72 in Italy, 75 in UK and 48 in Belgium.
The overall sample presented a satisfactory mixture of socio-demographic characteristics: marital status (45 % married, 44 % single, other 11 %); education (13 % primary, 40 % secondary, 47 % higher school); and, occupation (57 % employed, 20 % student, 5 % unemployed, 4 % unable to work through disability, 14 % housewife/retired). The frequency distribution of these variables within each country, shows statistically significant differences in the education level (X2 23.4 df = 6; higher school range: 36–60 % respectively for IT and UK) and occupational status (X2 58.24 df = 12; employed: 29–87 % respectively for NL and UK). More details of the participants sample clinical characteristics are reported elsewhere [18]. Recruitment took place in public areas, via calls in free local newspaper and by word of mouth. The protocol was approved by the Medical Education Research Ethics Committee of the University of Liverpool. The written informed consent of the participants was obtained in all four countries.
Study design and focus groups
Figure 2 illustrates the study design. A set of 35 focus group discussions (nine for each country, except Belgium with eight) were conducted following the same procedures, according to a detailed protocol [19]. Participants attended a 1-day-meeting where they watched four videotaped consultations and carried out different tasks [19]. The videotapes were standardized medical OSCE (Objective Structured Clinical Examination) consultations, in which eight different 4th year medical students from Liverpool Medical School -from now on called ‘doctors’- were assessed during their final examination. Consultations lasted on average 10 min. The maximum variation in the quality of doctors communication, was guaranteed by the combination of simulated patient ratings on a 10-point Likert scale (Global Simulated Patient Rating Scale, GSPRS) and examiners’ assessments on a checklist that included pre-established expert defined abilities defined by experts (Liverpool Communication Skills Assessment Scale, LCSAS). Two different scenarios were used, both about gynaecological problems associated with high levels of emotional distress. One was vaginal discharge related to unsafe sex- a Sexually Transmitted Disease (STD), the other was menstrual period pain (PP). As previously stated, participants were balanced by gender, anyway in order to encourage the free expression of opinions given the “gender sensitive” health problems shown in the videos, they attended gender specific focus groups.
As a prompt for the focus group discussion, participants watched four videos, based on the same scenario, in which the quality of doctors’ communication varied according to different scores of GPRS and LCSAS evaluations. Focus group discussions of 1 h followed, in which they were invited to explain their assessments from the first-round session, share their likes and dislikes regarding the doctors’ communication approach and provide underlying reasons.
Units of analysis and measures
In order to compare the qualitative data gathered through the focus group discussions, a content analysis was performed. This aimed at creating a coding system that would allow us to synthesize, and systematically organize, participants comments. The application of quantitative techniques to qualitative data, is one of the possible use of the Mixed-Method approach [20, 21].
Each centre adopted the same set of systematic and transparent procedures for arranging and processing the raw data in order to obtain valid and reliable inferences [18]. The researchers from each centre, two from the Netherlands (J.B., L.V), three from Italy (F.M., M.R., G.D.), and one from the UK (I.F), applied an inductive content analysis of a selected set of focus group discussions. These were previously videotaped, transcribed and translated into English by researchers, who are all fluent in the English language, and checked by a native speaker. This was in order to derive a common coding framework (“Guliver coding system”) with which to classify each participant’s statement. Details about the inter-rater reliability have been published elsewhere [18].
The resulting coding system, is divided into three levels, the area, category and sub-category, to which each statement has to refer. Specific examples of the Guliver coding system categories selected in the present paper are provided in an Appendix.
When a judgment was expressed in a participants’ comment, its value was coded as positive, negative or neutral. Figure 2 indicates the variables on which the analysis have been performed in order to answer each research question.
All the focus group transcripts have been coded in their original language and also translated into English in order to make them accessible to researchers in all four centres.
Statistical analyses
The analyses have been performed at the category level of the Guliver coding system in order to have a sufficient sample size in the comparisons between countries (Table 1 shows the consistency of the cells at the third level of classification – sub-category).
The exploration of the bivariate frequency distribution between content communication categories and participant nationality was performed using chi2 test and the adjusted residual analysis [22].
Two logistic regression models were estimated in order to investigate, both in terms of main effects and their interaction, the relationship between the outcome variable (positive and negative participant specific judgments) and the two independent ones which are, Guliver categories and participants’ country. Since the two independent variables are categorical, a reference category was needed for each of them; therefore Collecting information was the reference category for the coding system and the Netherlands for the country. The coefficients estimated by the models were expressed according to the odds ratio interpretative approach [23]. Therefore in the first model the main effects were expressed in terms of OR (that inform on the odds of positive statements of each category in relation to a specific reference category); while the interaction effects estimated in the second model were expressed in terms of odds (that indicate the increase of positive comments for each negative produced - calculated using Margins and Marginsplot STATA commands). This methodological choice was taken in order to facilitate the interpretation of the results.
To take into account the nested structure of the study design – repeated measures within participants – the cluster option of STATA commands, was adopted in the regression models.
A count of the frequency of positive and negative statements at the level of sub-category will be provided to describe better the results obtained.
All the analysis were performed using STATA13.0 [24].