Network Analysis of Past Presidents of the North American Skull Base Society

Abstract

Background  Little is known about the relative contributions and interactions of the past presidents of the North American Skull Base Society (NASBS) and skull base centers.

Objectives  (1) Measure academic contributions of past presidents; (2) identify influential nodes of academic collaboration; (3) identify opportunities for future collaboration.

Methods  Peer-reviewed publications of past presidents of NASBS from 1964 to July 2019 were identified using Scopus author name search. Network structures were constructed and analyzed using the graph-tool python library to produce a weighted co-authorship network base and compute centrality measures. Girvan–Newman clustering was applied to identify community structure. Network maps were then produced using Gephi network visualization software with force-directed layout algorithms.

Results  The coauthor network of 29 presidents was fully connected, with a maximum shortest-path distance between presidents of 5. The mean number of connections from each node without respect to weighting was 5.31 (standard deviation [SD]: 3.53), and the mean number of connections with weighting was 8.40 (SD: 7.28). The number of unweighted connections ranged from 1 to 14 and weighted connections ranged from 0.25 to 24.7. Girvan–Newman clustering identified three communities with two that covered 93% of the network. The largest communities contained 14 and 13 presidents. The number of connections was correlated with h-index, both unweighted ( r ²  = 0.34) and weighted ( r ²  = 0.26).

Conclusion  Network mapping of past presidents of the NASBS helps to capture the history of the NASBS and reveals areas of concentration and influence within the specialty.

Introduction

Past presidents of the North American Skull Base Society (NASBS) include pioneers and leaders across multiple disciplines in skull base surgery. Little is known about the relative contributions and interactions of past presidents and skull base centers. Compared with other specialties, skull base surgery is a relatively small network. The NASBS represents skull base physicians from multiple disciplines across North America. From 1990 through 2022, there have been 31 presidents representing multiple specialties. ¹

Social network analysis is the application of graph theory to gain insight into the structure of relationships between social entities. ² These entities may be organizations, events, or publications in addition to individuals. With social network analysis, one can (1) better understand the flow of communication and spread of information within the community; (2) identify hotspots of collaboration and participation; (3) identify opportunities for networking and new collaboration; and (4) create visual representations of these interactions.

The objectives of this study were to (1) measure the academic contributions of past presidents of the NASBS; (2) identify influential nodes of academic collaboration; and (3) provide a resource to the membership of the NASBS for exploration and building of new collaborations. These data were presented at the NASBS annual meeting in 2020 and include data for 29 past presidents through 2019. ¹

Methods

Scopus search engine was utilized for review of publications due to the quality and accuracy of citation data, widespread adoption, and ease of use. Peer-reviewed publications of past presidents of NASBS from 1964 to July 2019 were identified using Scopus author name search. Author and abstract records were collected in an SQL database for offline refinement. Ambiguous author profiles and alternate spellings were merged via automated n-gram distance similarity matching ³ followed by manual review using institution affiliation and publication history. Most presidents had multiple significantly sized author records in Scopus, requiring manual intervention for merging. h-Index was calculated from citation data provided by Scopus after merging author records. The search included years prior to the foundation of the NASBS to more accurately capture all citations for earlier presidents.

Network structures were constructed and analyzed using the graph-tool ⁴ python library. A weighted co-authorship network graph was generated, and centrality measures calculated. Network weights were calculated with normalization per the method described by Newman, ⁵ under the simplified assumption that as the number of coauthors on a publication increases the collaboration between any pair of individuals is weaker. Girvan–Newman clustering ⁶ was applied to identify isolated communities of collaborators and the respective bridging collaborations. Visual network maps were then produced using Gephi ⁷ network visualization software with force-directed layout algorithms. To quantify and visualize regions of strongest influence and information spreading in the network, measures of closeness centrality and betweenness centrality ⁵ were calculated for every node (member) in the network and used for visualization. Both are based on determining the shortest paths between members, with the shortest path between two members being the minimum number of hops connecting them in the co-authorship network. Closeness centrality is one useful numeric measure for determining the relative importance or influence of each member in the context of the network. It allows highlighting members that are efficiently connected to the entire network, quantifying how directly a member collaborates with all others in the network. To calculate closeness centrality of a member, the length of the shortest path from that member to every other member is determined. The average of these lengths represents how far or relatively disconnected a member of the network is, so the closeness is defined as the inverse of this average.

Betweenness centrality is another useful measure of network member influence. First, for every pair of members in the network, the shortest path connecting them is determined. The betweenness centrality of a member is the number of times a shortest path between other pairs in the network pass through them. It therefore allows identifying the members that are the strongest bridges of collaboration, influence, and information spreading throughout the network. ⁵ While closeness centrality identifies members with a large quantity of connections with few intermediaries, in networks with highly internally connected and isolated clusters it may overestimate the importance of some of these connections. The betweenness centrality allows identifying key influencers and critical information bridges in the network, especially networks with distinct clusters.

For the unweighted network, the length of a path between members is simply the number of connections traversed. For the weighted network, the length is the sum of the weights of the connections.

Results

The coauthor network of 29 presidents was fully connected, with a maximum shortest-path distance between presidents of 5 ( Figs. 1 2 3 ). The mean number of connections from each node without respect to weighting was 5.31 (standard deviation [SD]: 3.53), and the mean number of connections with weighting was 8.40 (SD: 7.28). The number of unweighted connections ranged from 1 to 14 and weighted connections ranged from 0.25 to 24.7. A total of 12 members (41.4%) had weighted connections above the mean. Girvan–Newman clustering identified three communities with two that covered 93% of the network. The largest communities contained 14 and 13 presidents and had mean unweighted connections of 5.79 and 5.31 and mean weighted connections of 10.10 and 7.45, respectively. The number of connections was correlated with h-index, ⁸ both unweighted ( r ²  = 0.34) and weighted ( r ²  = 0.26).

Fig. 1.

Node size: weighted degree connections. Node color: community.

Fig. 2.

Node size: closeness centrality. Node color: specialty.

Fig. 3.

Node size: betweenness centrality. Node color: specialty.

Review of Fig. 1 demonstrates that the network is dominated by two groups. The blue group consists of mostly neurosurgeons and closely affiliated neurotologists across multiple institutions. The red group consists of mostly otolaryngologists across multiple institutions. Primary connections across the two groups are surgical partners (otolaryngology and neurosurgery) at University of Pittsburgh Medical Center (UPMC), MD Anderson, and Loyola. The strongest individual connections are between pairs of surgeons at Memorial Sloan Kettering, UPMC, Loyola, and MD Anderson.

When grouped by specialty, closeness centrality ( Fig. 2 ) is widely distributed among neurosurgery, whereas otolaryngology is dominated by two individuals with minimal overlap at specific institutions for both groups. Betweenness centrality ( Fig. 3 ) yields similar results, showing that the same individuals dominate both networks.

Discussion

This network analysis of NASBS past presidents provides historical context for the contributions of individual surgeons and academic programs to the development of skull base surgery and medical literature. Historically, the network of NASBS past presidents has been dominated by two major communities ( Fig. 1 ), as measured by shared peer-reviewed publications. These collaborations cross specialties and institutions and encompass overlapping eras of transcranial and endoscopic endonasal surgery. It is notable that the strongest collaborations between specialties are seen between Dr. Sekhar and Dr. Janecka at UPMC and Dr. DeMonte and Dr. Hanna at MD Anderson. A very strong collaboration is also observed between an otolaryngologist (Dr. Gullane) and plastic surgeon (Dr. Neligan) in Toronto. One limitation of such an analysis is that it reflects the network over a long time period and does not show how it has evolved over time.

In addition to providing a historical context for past academic collaboration, network connections can be exploited to disseminate information and foster collaboration more efficiently. As mentioned previously, closeness centrality is useful in finding the individuals who are best placed to influence the entire network most quickly. Betweenness centrality indicates which nodes act as “bridges” between communities in a network and is useful in finding individuals who influence the flow around a system. An individual may have few connections to the entire group (low closeness centrality) but be the only connection between two separate groups (high betweenness centrality). This person can provide information about the activities of both groups and facilitate connections between them. Conversely, an individual may be centrally located in the network with lots of connections to others (high closeness centrality) but low betweenness centrality if these groups are already connected through other nodes. This individual is a good pathway for rapid dissemination of information across the entire network.

In this network, there is minimal difference between closeness centrality and betweenness centrality ( Figs. 2 and 3 ), indicating that the same individuals are influential in both regards. This may reflect the small size of the network. Nevertheless, the network maps identify those past presidents who are likely to be most influential. Additional network mapping of other societies in otolaryngology and neurosurgery may identify important connections of past presidents with other academic communities. Nodes of influence can be exploited to promulgate new initiatives more efficiently and coordinate activities across multiple networks.

Conclusion

Network mapping of past presidents of the NASBS helps to capture the history of the NASBS and reveals areas of concentration and influence within the specialty. These findings, represented visually, can be used to identify nodes of influence than can be utilized by the society to promote rapid dissemination of new initiatives and collaboration. Ongoing network analysis can display trends in evolution of collaboration within and across specialties and societies.