Data were used from the Netherlands Information Network of General Practice (LINH). These data were retrieved from EMRs kept by a nationally representative sample of approximately 160 GPs working in 96 general practices with 360,000 registered patients in 2006. Data include information on consultations, morbidity, prescriptions and referrals to other healthcare professionals. Practices as well as patients are representative for the Dutch population [11, 12].
Diagnoses were recorded using the International Classification of Primary Care (ICPC-1) coding system . When issuing a prescription, a diagnostic code was recorded, and the selected drug was automatically linked to the Anatomical Therapeutic Chemical (ATC) coding system (http://www.whocc.no). In this study morbidity data and data on all prescriptions issued by the participating practices were used. Only individuals of 30 years and older were included in the statistical analyses, because individuals below the age of 30 years have a lower probability of having CVD , but also inflammatory arthritis, osteoarthritis, and/or diabetes mellitus. Practices that recorded data during less than six months in 2006 were also excluded from the analyses, because the selection of the patient groups are based on complaints and diseases presented during GP consultations. When more than half of the consultations in a year are missing, the prevalence rates of the studied diseases are underestimated.
Based on the ICPC coded morbidity and prescription data, it was determined whether the participating patients had any of the following diagnoses in 2006: 1) inflammatory arthritis (ICPC code L88; rheumatoid arthritis or ankylosing spondylitis), 2) diabetes mellitus (ICPC code T90), 3) osteoarthritis of knee and/or hip (ICPC code L89 and/or L90), 4) CVD (ICPC codes K75 (myocardial infarction), K89 (transient ischemic attack), and/or K90 (stroke / cerebrovascular accident), 5) hypertension (ICPC code K86 and/or K87) and 6) hypercholesterolemia (ICPC code T93). Individuals were classified within the above mentioned groups, when the diagnosis was recorded in 2006 or in previous years (up to 2004). By using a period of three years, we could determine the diagnoses more reliable, since not all patients are visiting their GP every year with complains related to the studied diseases.
It was not possible to discriminate between rheumatoid arthritis and ankylosing spondylitis within the group of inflammatory arthritis patients and between diabetes mellitus type 1 and 2. Per patient we also determined whether they used cardio protective medication (statins and/or anti-hypertensive agents) in 2006. All other patients listed in the participating practices without inflammatory arthritis, diabetes mellitus and/or osteoarthritis were used as controls.
First, CVD prevalence rates were calculated among individuals with inflammatory arthritis, diabetes mellitus, osteoarthritis and controls. Differences in age, gender, the prevalence rates of CVD, hypertension and hypercholesteremia and the use of statins and antihypertensives between the three patient groups and controls were tested with Student’s t-test and chi-square tests.
Second, the association between the presence of CVD and the presence of the three diseases was determined with multilevel logistic regression analyses with a random intercept using the second order PQL method  due to the hierarchical structure of the data (patients clustered in practices). Three models were calculated: 1) a model with the three studied independent variables (presence of inflammatory arthritis, diabetes mellitus and osteoarthritis), 2) model 1 plus age and gender, and 3) model 2 plus CVD risk factors (hypertension and hypercholesterolemia). Results are presented as odds ratios with 95%-confidence intervals. Interaction terms were introduced to test for selective amplification of the CVD risk by subjects with more than one of the investigated diseases and the possible modifying effect of age.
Multilevel logistic regression analyses were performed with MLwiN, a statistical program for multilevel analyses . Student’s t-test and chi-square tests were performed with Stata10.