Modeling Unipartite Physician Network

• Activity Number: 229
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Sudeshna Paul*+ and James O'Malley
• Companies: Harvard Medical School and Harvard Medical School
• Keywords: bipartite ; network ; projection ; threshold ; unipartite

Abstract

Improving communication and coordination of care among physicians is important in improving quality of care. Physicians actively share clinical information among themselves through the sharing of patients. In this work, we study a physician-patient bipartite system and its projection to the unipartite physician network. The resulting ties in this network correspond to the number of shared patients between any two physicians. Using Medicare administrative data from 2006 on 51 HRRs (hospital referral region) across the US, a physician-physician network is identified for each HRR. In order to retain only the meaningful ties in the networks, a relative thresholding scheme was employed. Variation in network measures across the 51 HRRs was studied. We use negative binomial regression to model the valued ties in the full physician network assuming conditional independence. Predictors include patient and physician characteristics and fixed effects to account for HRR variation. Results indicate that physicians with similar individual and patient level attributes tend to share more patients.