Modeling the Association Between Physician Risky-Prescribing and the Complex Network Structure of Physician Shared-Patient Relationships

Homophily is the social network phenomenon whereby sim-ilar individuals have a greater propensity to form ties. Motivated by concerns of risky-prescribing among older patients in the United States, we developed exponential random graph models to estimate the effect of homophily of risky prescribing net of other physician characteristics and network features in a complex network. We also developed novel network measures and associated non-parametric statistical tests that allow for greater homophily in specific triadic confgurations (“super-homophily"). Using a shared-patient network of all physicians who treated patients residing in the US state of Ohio in 2014, we found statistical evidence of physician homophily (both in level and heterogeneity across regions) and triadic homophily on risky prescribing. Our findings may explain the emergence of prescriber communities, motivate group-level prescriber interventions to directly reduce risky-prescribing, and motivate interventions that reshape physician shared-patient networks to indirectly reduce risky-prescribing.