303 Social Network Analysis of Patient Sharing Among Providers: Implications for Analyzing Disparities in Cancer Screening

Abstract

OBJECTIVES/GOALS: Many providers share patients resulting in networks where clinical information is exchanged, and which can impact the quality and efficiency of care. Here we analyzed the network properties of a primary care service area (PCSA) in Harris County TX, motivating our ongoing analysis of how they are associated with disparities in cancer screening. METHODS/STUDY POPULATION: Data.All providers (n=731, Medicare 2018) from the PCSA with the most providers in Harris County TX, with gender, specialty, and the number of shared patients. Method. Modeled the data as a network consisting of provider nodes, connected in pairs by edges if they shared >11 patients (an empirically-determined threshold). Analyzed the network structure using (1) modularity maximization and its significance to identify densely-connected communities; (2) degree centralization to measure whether a few providers shared many patients, and betweenness centralization to measure whether a few providers connected densely-connected communities; and (3) chi-squared to measure if pairs of connected providers tended to be of the same gender compared to disconnected provider pairs. RESULTS/ANTICIPATED RESULTS: The results (Fig. 1, http://www.skbhavnani.com/DIVA/Images/Fig-1-SNA-Network.jpg [http://www.skbhavnani.com/DIVA/Images/Fig-1-SNA-Network.jpg]) revealed a fragmented network with 120 small components (connected subnetworks, not part of any larger connected subnetwork), and 1 large component. The large component (n=244) had strong and significant modularity (Q=0.73, z=53.13, P<.001) with communities of providers that shared more patients than expected by chance; low degree centralization (dc=0.11) suggesting that no provider dominated patient sharing, in addition to high and significant betweenness centralization (bc=0.5, P<.01) suggesting that a few providers were responsible for connecting the densely-connected communities; and a significant gender bias (X2=10.05, df=1, P< .01) among those that shared patients, versus those that did not. DISCUSSION/SIGNIFICANCE: The analysis revealed a specific type of vulnerability (betweenness) for network fragmentation, and a gender bias in how patients were shared. These results motivated our ongoing analysis on how the network properties are associated with disparity in cancer screening within PCSAs across Texas.