Surgical Innovation in the Era of Global Surgery

A Network Analysis

George Garas, MD, FRCS; Isabella Cingolani, PhD; Vanash Patel, MD, PhD, FRCS; Pietro Panzarasa, PhD; Derek Alderson, MD, FRCS; Ara Darzi, MD, KBE, FRCS; Thanos Athanasiou, MD, PhD, MBA, FRCS


Annals of Surgery. 2020;271(5):868-874. 

In This Article


The Dataset

This study draws on the Web of Science (WOS) platform (Clarivate Analytics, Philadelphia, PA). All articles on robotic surgery were extracted through the use of the MeSH terms: "robot OR robotic OR robot assisted OR robotic assisted OR robotically assisted OR robot-assisted OR robotic-assisted OR robotically-assisted." The Research Area was confined to "Surgery" and the Document Types to "Article."

The search was performed on January 17, 2017 and produced 3889 publications (peer-reviewed articles) published between July 1988 and January 2017 (Figure S1, These were all used for constructing the coauthorship network. Articles were generated from 1700 organizations nested within 62 countries, in turn nested within 6 geographical regions (Table S1,

The Surgical Collaboration Network

The collaboration network was constructed using VOSviewer (Leiden University, Leiden, The Netherlands), a software developed specifically for the study of scientific collaboration networks. In the network, each node represents an organization, and a link between 2 nodes represents collaboration between the corresponding organizations. As coauthorship has been shown to be a good proxy of collaboration,[11,12] links between nodes were based on coauthorship of articles. The resulting network is weighed: the value (or weight) of a collaborative link increases as a function of the intensity of collaboration (Figure S2,

Outcome Measures

For each organization, 2 outcome measures were computed: 1) the research impact; and 2) the innovation index. These measures were computed as follows:

  • Research impact: Research impact was measured as the sum of normalized citations received by all articles (co-) authored by scholars affiliated with a given organization in each year. To obtain normalized citations, the citation count for each publication in a given year was divided by the average number of citations obtained by all articles published in the same year. The greater the sum of normalized citations for a given organization, the greater the organization's research impact (ie, normalized citations are used to measure impact).

  • Innovation index: The innovation index represents a recently validated metric used to evaluate and rank surgical innovation (Table S2,[13] It captures the value of the innovative output produced by an organization as a function of the degree to which it reached an implementation stage (Figures S3–S4, Thus, the greater the innovation index of a given organization, the more innovative the organization's surgical research output.

For a detailed description of how each performance metric was calculated, please see sections S.3.1 and S.3.2 in the Supplementary Material,

Network Measures

Two established network measures were computed: 1) the clustering coefficient; and 2) closeness centrality. These measures were defined as follows:

  • Local clustering coefficient: The local clustering coefficient quantifies how closed an organization's ego-centered network is (ie, the network including connections between the organization and its partners as well as connections between these partners) enabling assessment of the extent to which an organization's collaborators also collaborate with each other or, alternatively, the extent to which an organization spans structural holes separating collaborators. The higher the local clustering coefficient of an organization, the more closed the organization's ego-centered network is. More specifically, the local clustering coefficient of an organization was defined as the ratio between the number of actual triangles containing the organization and its neighbors, and the maximum possible number of such triangles. A generalized weighted clustering coefficient was calculated to take into account the weights of links (see Supplementary Material, To facilitate interpretation, all values of the generalized weighted clustering coefficient were standardized.[14] Measuring the density of triangles in an organization's local network uncovers how open or closed the network is, and the extent to which the organization acts as the knowledge broker between otherwise disconnected organizations in the collaboration network.

  • Closeness centrality: The closeness centrality of an organization measures how close the organization is to all other organizations in the collaboration network. The higher an organization's closeness centrality, the greater the organization's access to the knowledge (or data, or any other resource) provided by other organizations in the collaboration network, and thus the greater the organization's influence on others as a result of its structural position.[15] The generalized weighted version of closeness centrality was used to account for the weights of links (see Supplementary Material,[16]

For a detailed description of how each network metric was calculated, please see sections S.3.3 and S.3.4 in the Supplementary Material,

Measuring the Geographical Dispersion of Collaborations

For each organization, the geographical entropy of collaborations was computed to capture the geographical dispersion of the organization's collaborators. An organization's geographical entropy increases as the organization collaborates with other organizations located in more countries and devotes an equal amount of collaborative effort towards each of these countries (Figure S5, For a detailed description of how geographical entropy was measured, please see section S.3.5 in the Supplementary Material,

Measuring Academic–Industry Collaborations

All organizations that were publicly registered as companies and classified as "corporate" entities in the WOS platform (through the InCites intelligence tool) were identified.[17,18] For each organization, the sum of the organization's collaborative efforts toward other industrial (corporate) partners during the study period was calculated. For a detailed description of how the strength of industrial collaborations was measured, please see section S.3.6 in the Supplementary Material,

Control Variables

Many other organizational characteristics may influence scientific performance. This study controlled for the following 2 additional variables: 1) each organization's institutional type (eg, academic, corporate, health, etc. as classified by the InCites intelligence tool in the WOS platform) and 2) a measure of volume, here referred to as "number of articles in WOS," given by the number of all articles, beyond robotic surgery, published by each organization that the InCites intelligence tool could retrieve in the WOS database (see section S.3.7 in Supplementary Material, Controlling for research volume enables the association between collaboration network and both citation count and innovation to be investigated by keeping the number of publications constant.

Statistical Analysis

Maximum-likelihood estimates of 2 hierarchical 3-level random-intercept models were computed in which organizations were nested within countries, in turn nested within geographical regions. A random-intercept structure was combined with heteroskedastic level-1 residuals by letting the variances of these residuals be a function of the organization's institutional type. All models were estimated using STATA 15 (StataCorp LP, College Station, TX). The significance threshold was set at P < 0.05.