Abstract
OBJECTIVES The main aim is to evaluate the variability of practitioners’ prescribing behaviour on statines in AUSL Bologna South from 2000 to 2003, as well as to quantify how much such a behaviour depends on culture, education and policy of practitioner. METHODS A multilevel model has been built to reflect the hierarchical data structure: patients with prescribed statines, nested in their prescribing general practitioner. The multilevelmodel advantage lies on the fact that one can insert, in the same analysis, independent variables related either to the general practitioners (level-2) or to the patients (level-1). Furthermore, one can quantify the variability at each level, which is, prescribing practitioners’ variability and patients’ variability. Particularly, a random intercept model has been built, where the response variable is the sum of the daily dosages given in the prescriptions for each patient. Actually, we refer to generalized linear model theory, because the dependent variable is Gamma-distributed. RESULTS The regressors referred to the patients (level-1) and inserted into the model are: age, sex, and use of other cardio-vascular drugs. The level-2 independent variables, hence referred to practitioner, are: age, sex, specialization (yes/no) on cardiology, ratio of statines’ prescriptions on total number of prescriptions of cardiovascular drugs and percentage of patients over 65 years old. The largest part of variability is obviously due to patients’ effect. Regarding the regressions of level 2, the age of the practitioner provides a negative and significant coefficient, indicating a tendency towards “prescriptive thrift” by older doctors. CONCLUSIONS It appears to be a clear evidence that the consumption of statines and, generally, of drugs cannot be simply reduced to individual characteristics. From a methodological point of view, it has been shown that multilevelapproach provides a coherent framework, in spite of the lack of applications to health sciences.