Objective: 
To investigate the utility of combining multiple publication-level metrics with author network analysis to identify influential external experts.

Method: 
A sample of 500 publications in 2019 relating to a selected drug was identified in July 2019. Author names were extracted and linked to publication metrics (weighted and scaled to form four factors - social media, scholarly, news and citations) and three key co-authorship network characteristics. 
Results: 
Of 500 publications, 477 had available altmetrics. From these publications 2080 authors were identified. The top 500 authors by number of network connections were selected, and 71 not connected to the main network then eliminated. The resulting network comprised 429 authors in 16 clusters. The most prolific author had 59 publications in this set, while 82 authors had only 1 publication. 
Cluster 4 (35 authors, 274 publications) had the highest total citation impact while cluster 1 (70 authors, 245 publications) had relatively low citation impact but the highest social media and news impact, and the highest total network influence centrality and importance. Cluster 3 (36 authors, 301 publications) had the highest scholarly impact, and also the highest total network betweenness centrality.
Correlations on an individual author level showed that influence range closeness centrality was the most strongly correlated with publication metrics (Spearman’s rank scores ranging from 0.484-0.687). The correlation between the number of publications attributed to an author was modestly correlated with the author’s impact in terms of either network centrality/importance (Spearman’s rank 0.451-0.786) or publication metrics (Spearman’s rank 0.503-0.701). Ranking authors on each of the 7 metrics allowed ‘fingerprinting’ to identify the most influential authors from each cluster, by overall metrics and by specific factors of interest. Visual inspection of the network map also allowed key linking authors to be identified.

Conclusions: 
Identifying influential external authors can be challenging to do in a quantitative way. Scientific publications are a useful indicator of an author’s engagement and influence in their field. However, publications vary greatly in their importance. While the impact factor (IF) of the journal has traditionally been used as a proxy for the importance and quality of the research, this can be highly unreliable.
Article-level metrics are not only directly relevant to the particular publication, but they also measure different types of interest and uptake. However, the wide variety of article-level metrics available can make them confusing to interpret, so simplifying them into four equal domains facilitates comparative assessments. 
In addition to the importance of their research output, the nature of an experts’ interactions with other experts in their field is an important guide to their influence. Co-authorship is a useful guide to this. Co-authorship network analysis allows authors to be clustered, enabling easy identification of informal ‘research groups’. Measures of network centrality allow different estimates of how central the author is to the overall network – authors with high centrality are the nodes through which other authors can be easily reached. Network importance gives an indication of the strength of an author’s connection to multiple other authors. 
By combining different measures of an author’s publication impact with an assessment of their position in the co-authorship network, deep qualitative and quantitative insights can be gained into the level of individual influence in a given field.

Key opinion leader identification through author publication altmetrics and network analysis