Network Analysis of Academic Medical Center Websites in the United States

Shuhan He, David Chen, Kameron Collin Black, P. Andrew Chong, Sammer Marzouk, Byung-Jun Yoon, Kendrick Davis, Jarone LeeScientific 2023 · DOI

Abstract: Healthcare resources are published annually in repositories such as the AHA Annual Survey Database™. However, these data repositories are created via manual surveying techniques which are cumbersome in collection and not updated as frequently as website information of the respective hospital systems represented. Also, this resource is not widely available to patients in an easy-to-use format. Network analysis techniques have the potential to create topological maps which serve to aid in pathfinding for patients in their search for healthcare services. This study explores the topological structure of forty United States academic health center websites. Network analysis is utilized to analyze and visualize 48,686 webpages. Several elements of network structure are examined including basic network properties, and centrality measures distributions. The Louvain community detection algorithm is used to examine the extent to which these techniques allow identification of healthcare resources within networks. The results indicate that websites with related healthcare services tend to form observable clusters useful in mapping key resources within a hospital system.

Introduction

Finding specific healthcare information online can often feel like navigating a labyrinth. While the American Hospital Association (AHA) maintains comprehensive databases of hospital services, these repositories are typically compiled through manual surveys. This process is inherently slow to update and the resulting data is rarely presented in a format that is user-friendly for patients seeking immediate care.

In contrast, hospital websites are updated in real-time but often lack an intuitive, unified structure across different institutions. This research addresses the need for an automated, data-driven approach to understanding how healthcare information is organized digitally. By treating a website as a "network"—where individual pages are nodes and hyperlinks are edges—the study establishes a framework for optimizing patient navigation and automating the tracking of medical services across the United States.

Methodology

The researchers adopted a "Data Descriptor" approach to construct and analyze a massive graph representing the Academic Medical Center (AMC) web ecosystem. The workflow involved three primary stages: data crawling, network metric calculation, and community detection.

Methodological workflow for crawling, exporting, and analyzing AMC website network data.

Initially, the Screaming Frog SEO spider was used to crawl 73 AMC websites. Due to technical constraints such as processing limits and security firewalls, the final dataset was refined to focus on 40 centers. To evaluate the connectivity and importance of specific pages, the team utilized the Python library NetworkX to calculate several graph theory metrics:

  • Centrality Measures: Betweenness, Eigenvector, and Closeness centrality were employed to identify "gatekeeper" pages and highly influential hubs.
  • Importance Index: A novel composite score, calculated as the geometric mean of various centrality metrics, was used to rank the overall navigability and connectivity of different AMC sites.

Finally, the Louvain algorithm was applied for community detection, and visualization tools like Gephi were used to generate topological maps where node size and color reflect connectivity levels.

Data and Key Insights

The study resulted in a new, open-access repository hosted on Figshare, containing metadata for 48,686 individual webpages and a massive list of 1,165,520 internal and external hyperlinks. Analysis of this data revealed several critical insights into the digital structure of healthcare:

The "Scale-Free" Nature of Healthcare

Most AMC websites follow a "power-law" distribution. This indicates that a few "super-hub" pages—such as the homepage or main provider directories—hold the vast majority of links, while most niche pages have very few. This "rich-get-richer" structure is a hallmark of organic web growth.

Semantic Clustering via Topology

A significant finding was that the Louvain algorithm successfully grouped pages into functional medical departments (e.g., trauma centers, maternal health, or financial services) without any prior knowledge of the page text. This suggests that the internal linking structure created by web developers naturally mirrors the real-world organizational hierarchy of medical services.

Results and Comparative Analysis

The final network analysis provided a quantitative look at the digital footprint of these institutions. Despite the massive scale of nearly 50,000 nodes, the "average path length" across these sites was found to be only 2.619. This implies that most healthcare information is technically only two to three clicks away, provided the user is on the correct path.

Comparative analysis of network metrics across forty U.S. Academic Medical Centers.

By normalizing centrality measures into an Importance Index, the researchers were able to rank the navigability of various institutions. As shown in the comparative data, the University of Utah Health Sciences Center and Augusta University emerged as top performers in terms of site connectivity and ease of navigation.

Limitations and Future Directions

While the study proves the efficacy of network science in mapping healthcare resources, several hurdles remain. To maintain computational feasibility, many sites were limited to a crawl depth of three, potentially missing deeper, specialized content. Furthermore, 31 AMCs had to be excluded because their web structures were too large for current standard visualization tools to process efficiently.

Future work aims to incorporate edge directionality to better model the actual "flow" of a patient's journey through a site. Additionally, as websites are dynamic entities, the authors suggest that future systems should be fully automated to provide "live" surveillance of healthcare resource availability, ensuring that digital maps remain as current as the services they represent.

Summary updated

Abstract

Healthcare resources are published annually in repositories such as the AHA Annual Survey Database<sup>TM</sup>. However, these data repositories are created via manual surveying techniques which are cumbersome in collection and not updated as frequently as website information of the respective hospital systems represented. Also, this resource is not widely available to patients in an easy-to-use format. Network analysis techniques have the potential to create topological maps which serve to aid in pathfinding for patients in their search for healthcare services. This study explores the topological structure of forty United States academic health center websites. Network analysis is utilized to analyze and visualize 48,686 webpages. Several elements of network structure are examined including basic network properties, and centrality measures distributions. The Louvain community detection algorithm is used to examine the extent to which these techniques allow identification of healthcare resources within networks. The results indicate that websites with related healthcare services tend to form observable clusters useful in mapping key resources within a hospital system.
BibTeX
@inproceedings{he2023network,
  author = {Shuhan He and David Chen and Kameron Collin Black and P. Andrew Chong and Sammer Marzouk and Byung-Jun Yoon and Kendrick Davis and Jarone Lee},
  title = {{Network Analysis of Academic Medical Center Websites in the United States}},
  booktitle = {Scientific},
  year = {2023},
  doi = {10.1038/s41597-023-02104-3},
}
Figure