We used a mixed methods design, which consisted of three parts: i) a questionnaire among family practices (FPs); ii) video recordings of hypertension-, cholesterol- and/or endocrine-related general practitioner visits; and iii) the database of Netherlands Information Network of General Practice. There was an overlap in the FPs used for the three different approaches.
Literature [12, 13] and interviews with general practitioners (n = 3) were used to determine cardiovascular primary preventive activities for patients without CVD. Six activities were selected for the questionnaire: 1) blood pressure measurement; 2) cardiovascular risk profiling; 3) a blood test related activity (i.e., explaining the need for a blood test to the patient, filling out a request for a blood test, and/or sharing the blood test results with the patient); 4) family history; 5) lifestyle history; and 6) lifestyle counselling. The questionnaire comprised questions on practice size and location, and about the frequency of the above mentioned cardiovascular primary preventive activities per week provided by general practitioners, health care assistants and practice nurses.
The questionnaire was emailed to 80 FPs from the 'Netherlands Information Network of General Practice' (LINH-DB) [14, 15] in 2009. Each year LINH contains 80 FPs and is a representative network of Dutch FPs as it takes into account the representativeness regarding practice type, urbanisation and the software system that is used . The sample of practices originates from the mid 1990's; eighty FPs are sufficient to make disease-related conclusions at the Dutch national level. Practices participate on a voluntary basis. The LINH database holds longitudinal data on morbidity, prescribing, and referrals, based on the routine electronic patient records that are kept by the participating practices. In order to enable longitudinal analyses, changes in the set of participating practices are kept to a minimum. There is a waiting list for practices to participate in the network. When a practice quits participating in the network, a new practice is invited. These FPs are spread throughout the Netherlands and are representative of all Dutch FPs. Reminders were sent to non-respondents two and four weeks later.
We used video recorded general practitioner consultations which were recorded as part of a larger study into doctor-patient communication in general practice in 2007-2008 . Neither general practitioners nor patients were aware of the topics of interest for the researchers. Forty Dutch general practitioners participated in the study. These general practitioners are representative of Dutch general practitioners regarding age, practice form and number of days worked . Eight hundred and eight consultations were video-recorded. These consultations were randomly recorded on week days, and are expected to represent Dutch general practice consultations. The study protocol adheres to the Dutch privacy legislation, approved by the Dutch Data Protection Authority. However, approval by a medical ethics committee was not required for this observational study, because the study did not interfere with a GPs usual work process and patients were not confronted with whatever project-related intervention. Our research complied with the Helsinki Declaration. All participating general practitioners and patients filled in an informed consent form before the recording of the consultation.
We included all recorded hypertension-, cholesterol- and endocrine-related visits for further analyses (we assumed that, for example, the time spent on a blood pressure measurement did not differ between primary and secondary prevention). The time spent on any of the six cardiovascular preventive activities (see previous paragraph) was measured.
Database of Netherlands Information Network of General Practice
Data on diagnosis and prescriptions were derived from the 'Netherlands Information Network of General Practice' (LINH-DB) [14, 15]. Focusing on the period 2005-2007, data from 161 FPs, spread throughout the Netherlands, were collected (data on the periods 2008 and 2009 were not yet available at the time of study). To investigate the prescription of primary preventive cardiovascular medication (PPCM) in FPs, we included FPs with complete data sets over the whole period 2005-2007. PPCM is defined as cardiovascular medication (i.e., all kind of beta blockers and statins) aimed at a determinant of cardiovascular disease for people without cardiovascular diseases, diabetes (types 1 and 2), or disorders of lipid metabolism (e.g. hypercholesterolemia). Cardiovascular treatment of patients who had consulted their general practitioner for cardiovascular diseases, diabetes, or lipid disorders in 2005-2007 was not regarded of primary preventive nature. Patients with these conditions were excluded, because the focus of the manuscript is on primary prevention. Primary prevention focuses on patients without cardiovascular diseases and/or diabetes and/or disorders of lipid metabolism, because in that case - according the Dutch guidelines - the use of measures belongs to regular care.
Based on the questionnaire, we determined the average weekly frequency a FP carried out the different cardiovascular primary preventive activities. The averages were determined separately for the general practitioners, health care assistants, and practice nurses in each practice. The video recorded consultations were used to determine the average amount of time required to carry out each of these activities. Based on these two figures we determined the average amount of time spent carrying out cardiovascular primary prevention activities per week. This was extrapolated to an annual figure. Finally, we estimated the total direct costs of cardiovascular primary preventive activities in FPs in the Netherlands in 2009 by taking into account the annual income and full time equivalents of general practitioners, health care assistants, and practice nurses as well as practice costs (i.e. housing, medical equipment, insurance and transportation) [18–20].
The volume of prescriptions of PPCM in FPs was determined using data of FPs with complete data sets from the period 2005 through 2007. Second, to get insight into which part of prescription of cardiovascular medication was intended as preventive, the patients were divided into four patient groups: 1) patients with cardiovascular and diabetes (types 1 and 2) and/or disorders of lipid metabolism (e.g. hypercholesterolemia); 2) patients with cardiovascular disease, but without diabetes or lipid disorders; 3) patients without cardiovascular disease, but with diabetes or lipid disorders; and 4) patients with neither cardiovascular disease nor diabetes or lipid disorders (primary prevention). The proportion of patients with i) prescription of cardiovascular medication, and ii) prescription of PPCM were calculated based on the existence of CVD, diabetes or lipid disordersas previously outlined. Multilevel analyses were conducted to investigate whether differences in prescription of cardiovascular medication for each patient group could be explained by family practice characteristics (urbanisation and practice type) and/or by patients' characteristics (age, gender, living in a disadvantaged neighbourhood (i.e. a geographically localised community within a larger city, town or suburb that contained a large proportion of people with a low social economic status), and insurance type). This was achieved using a mixed effects regression model, which estimates 'fixed' coefficients β for covariates at the patient level and at the family practice level i in FP j (Xij) and 'random' coefficients for the FPs j (θj). The parameter θj is assumed to be normally distributed with mean μ and variance τ2:
The variance (τ2) estimated in the model is a measure of the between family practice differences, and indicates the spreading of the prescribing behaviour of the individual FPs. To facilitate interpretation of the estimated between-FPs differences, we compared the FPs at the higher end of the outcome distribution (the 90th percentile) with the FPs at the lower end (the 10th percentile) of the outcome distribution. The relative difference in odds of prescribing behaviour in these two groups of FPs can be calculated from the parameter τ2: 80% OR range = exp (2.58*τ). The value 2.58 is the z-value corresponding to the width of the 80% (i.e. from the 10th percentile to the 90th percentile) confidence interval in a normal distribution (2*1.29).