199 - Network Analysis of the 2016-2025 IR Residency Match: Algorithmic Identification of Geographic Association, Non-Geographic Clusters, and Influence
Assistant Professor Emory University School of Medicine, United States
Purpose: IR residency grew from 7 programs and 15 positions in 2016 to 96 programs and 196 positions in 2025. Analysis of prior years’ match data reveals the most well-connected schools and programs, geographic and non-geographic associations, and the network’s evolution over time.
Materials and Methods: 1389 of 1458 (95.3%) IR matches between 2016-2025 were compiled from publicly available sources: NRMP data, department websites, and medical school match lists. Network plots were generated using Gephi. Medical schools and residency programs (nodes) and matches (edges) were assessed for number of connections, clusters, and Eigenvector centrality (relative influence determined by how well-connected a node is and how well-connected are its connections). Data was evaluated for statistical significance using Chi-square or Fisher’s exact tests.
Results: The medical schools graduating the most students matched to IR are Rosalind Franklin (26), Johns Hopkins (23), Temple (23), and U Washington (23), while residency programs matching the greatest number of students into IR are Emory (37), Rush (36), U Pennsylvania (34), and U Michigan (30). A significant association between a student’s medical school region and their matched program region exists across the ten years of the IR match (p< .05), and this association persists when examining distinct time periods: 2016-2020, 2021-2022, and 2023-2025 defined by substantive match process changes. A force-directed network algorithm illustrates that even as the number of matched students invariably grows, non-geographic clusters of schools and programs appear and change temporally. Overall, the most well-connected nodes in the network of IR matches are Rush (52), U Pennsylvania (52), U Washington (51), and Duke (47). The most influential node in the 5-year network is Duke with an Eigenvector centrality of 1.0, followed by Rush (0.93) and U Penn (0.91).
Conclusion: IR is among the most competitive specialties with challenges for applicants and program directors alike - an issue exacerbated by the ease of virtual interviews and resultant increase in applications. Students may use network analysis to focus on programs within their own or other desirable clusters, or to identify an influential program for an away rotation to bridge clusters. Program directors may use network analysis to guide outreach and inform perceptions of a program’s relative standing within and amongst clusters or the broader IR training network. The ABR or NRMP may use network analysis to evaluate effectiveness of reporting signals and regional preference.
