The Effect on Physician's Prescribing from Prescribing Tendencies of Physicians in their Patient-Sharing Network

Health services researchers often discuss concepts of “local” physician practice styles and treatment preferences within physician networks.  For example, geographic variation in treatment rates is often attributed to the coalescing of physician preferences in local areas.  Based on this notion, measures of local area treatment rates have often been used as the basis for instrumental variables in treatment choice.  Despite widespread acceptance that local-area treatment preferences exist, the question of whether physician treatment preferences are sensitive to the practices of other physicians in their local-area network has not been studied. This research examines whether physician high-dose statin prescribing rates are affected by the prescribing tendencies of the physicians with whom they are connected in a patient-sharing network.  Analysis used Medicare claims data for all fee-for-service Medicare beneficiaries hospitalized with acute myocardial infarction (AMI) in 2008 or 2009 and discharged home.  During this period, high-dose statins were available but guidelines were not clear and their use was discretionary.  Medicare Part D claims were used to measure physician-specific high-dose statin prescribing and physician patient-sharing networks.  Physician-specific high-dose statin ratios (HDSR) were defined as the ratio of high-dose statins to all statins prescribed by each physician in the 30-days after AMI discharge across all of their patients.  Networks for each physician were measured by identifying all patients who were prescribed any medication by the physician and then connecting the physician to all other physicians who prescribed medications to these patients.  The high-dose statin prescribing rate of each physician network was measured as the weighted average HDSR across all physicians in the network.  Weights were defined using the number of patients shared with each physician in the network.  We regressed the change in physician-specific HDSR from 2008 to 2009 on the difference between physician and network HDSR in 2008 and a squared-term of this difference.  We found that the difference between physician and network HDSR in 2008 was significantly associated with the change in physician-specific HDSR from 2008 to 2009.  This suggests that physicians react to the prescribing tendencies of the physicians in their local network and, on average, assimilate to the tendencies of their network.  Moreover, this result suggests that the rate at which physician HDSR converges towards their network HDSR is positively associated with the magnitude of difference between physician and network HDSR.  These inferences are robust to alternative model specifications and physician inclusion criteria.  This is, to our knowledge, the first study to show that physicians prescribing tendencies are sensitive to the prescribing tendencies of physicians in their network over time using observational data.

JEL Classification Codes:  D850, I100, I110