Understanding therapeutic traditions in a multilevel framework - new methodological approaches

Background: The observation of differences in the way apparently similar patients are treated from one health care setting to another, i.e., practice variation, have been recognized in numerous studies. The variations across the health care settings have been shown to exist for different outcomes. However, any measurement of any system will reveal some form of variation and this normal inevitable variation must be separated from special cause variation. Practice variation might, for example, be an expression of inefficient health care if a pharmacological agent is available in different brands at different prices and certain prescribers chose the more expensive one. In this situation, it is relevant to identify determinants of this variation in order to launch appropriate interventions. Inappropriate practice variation may have different origins. Since the process of drug prescription includes a number of phases it could be influenced at different levels (e.g. the patient, prescriber, health care practice). However, very few studies have so far tried to understand the relative importance of the different levels. Another aspect that rarely has been investigated when studying practice variation is what we denote as therapeutic traditions. This corresponds to the idea that cultural aspects at the practice level might exert a collective influence on prescribers working within the same practice. It can be expressed by the fact that the prescription behavior among physicians within the same practice may be more similar than the prescription behavior among physicians from different practices. This can be operationalized by investigating variance components in a multilevel framework. Many aspects concerning the analysis of variance components, i.e., quantifying overall practice variation, understanding the importance of different levels, monitoring and distinguishing appropriate from inappropriate practice variation, needs much more development. A further development of applied statistical and epidemiological methods may facilitate a more comprehensive understanding of the differences between health care units and thereby a more accurate measure of healthcare quality. Aims: Through the application of modern multilevel statistical techniques, we aim in this thesis to propose a model of analysis for investigating practice variation. We investigate therapeutic traditions in general and adherence to prescription guidelines and adoption of new drugs in particular. We also examine social and economic conditions at different levels of analysis and the role they play in this context. We focus on the combined analyses of measures of association and of variance and clustering as this can offer original and valuable information that could be of relevance for planning and evaluating interventions aimed to promote evidence-based prescription. We mainly study adherence to prescription guidelines for statin prescription as these lipid lowering drugs all have similar indications and efficacy. Material & Methods: The database LOMAS (Longitudinal Multilevel Analysis in Scania) is used and consists of unidentified information on all individuals living in Skåne region, Sweden during the period 1968 to 2008. One of the registers included is the Swedish Prescribed Drug Register that records information on sales of prescribed pharmaceutical agents by the Swedish Corporation of Pharmacies. We use multilevel regression models and generalized estimation equations, and provide an explanation for the application of these methods when focusing on measures of variance and clustering. Results: Adherence to guidelines for statin prescription and the early adoption of a new statin seemed to some extent to be conditioned by contextual factors particularly at the Health care practice level (HCP) level. The determinants of the individual behaviour are influenced directly by the contextual environment of the practice. Moreover, HCPs that follow guidelines for one drug type also appear to follow guidelines for other drug types, i.e., therapeutic traditions, acting at the HCP level, seems to influence the prescribing behaviour of individual physicians independently of specific drug type. Moreover, men with a lower income were prescribed the cheaper recommended statins to a higher degree than men with a high income. Conclusion: We present a model of analysis for investigating practice variation were we focus on the combined analysis of measures of association and measures of variance and clustering. We investigate therapeutic traditions in general and adherence to prescription guidelines and adoption of new drugs in particular. Since very few studies have tried to understand the relative importance of the different levels on the process of prescription, this thesis may eventually lead to a better understanding of the relationship between HCP- and individual level characteristics with respect to the prescription process. In turn, better understanding of the importance of different levels may facilitate for decision makers to focus interventions on the right factors at the right levels. By investigating the role of different health care levels on adherence to guidelines, researchers can more efficiently build and test models that capture factors influencing the prescription process