The Weight of Surgical Knowledge

Abstract

THE CHALLENGE In the modern era of information overload, keeping up with literature is a Herculean task. Using evidence syntheses and clinical guidelines may relieve cognitive overload, but biases and suboptimal methodologies are not infrequent.1–3 Peer-reviewed literature is most surgeons’ dominant information source,4 but conventional methods we keep up with literature–reading journals, setting search engine alerts, and more recently, following peers on social media–may be inadequate for apprehending the growing weight of surgical knowledge. The surgeon who queries a clinical or research question (ie, on PubMed) will, most likely, review articles on the first results page.5 Accessing literature through conventional search engines may limit vast evidence to a handful of articles selected by search algorithms (usually based on text features such as title term matches rather than quality of evidence6). Moreover, fragmented lists of articles demand considerable effort for surgeons to understand evidence in context of existing knowledge and reappraise current clinical practice. Breadth of surgical practice, typically spanning across multiple pathologies, amplifies the challenge: how do we apprehend the comprehensive scope of vast surgical knowledge today? CHARACTERIZING THE WEIGHT OF SURGICAL KNOWLEDGE What is the weight of surgical knowledge? A research librarian (C.S.) designed and performed PubMed searches of observational studies, randomized controlled trials (RCTs), guidelines, reviews, and evidence syntheses (systematic reviews and meta-analyses) from all surgical specialties (as defined by the American College of Surgeons7) between January 1, 1970 to December 31, 2019. Figure 1A shows that surgical literature has grown exponentially. Between 2016 and 2019, an average of 133,952 articles were published annually, including 20,736 case reports, 7482 evidence syntheses, 5928 RCTs, and 902 guidelines (Fig. 1B). On average, 367 articles, including 20 evidence syntheses and 16 RCTs, are published daily. Although what constitutes surgical knowledge may be debatable, information overload opined nearly 50 years ago8 is a clear challenge today.FIGURE 1: Four-year rolling averages of articles published in surgical literature, 1970 to 2019 ([A] All articles; [B] Four study types. Observational studies, not shown for scale, comprise majority of surgical literature); Knowledge graphs of articles published on rib fractures between 1970 and 2020 ([C] Article citation network. Research subtopic clusters; [D] Article citation network. Articles delineated by publication year to evaluate research trends; [E] Article citation network. An influential article and its citing articles are highlighted; [F] Co-authorship network. The largest co-authorship network comprises many founding members of the Chest Wall Injury Society; [G] Institutional citation network. Institutions are highlighted by the average publication year of their articles on rib fractures; [H] Journal citation network. Journals are highlighted by the average publication year of their articles on rib fractures). Circles represent individual articles/authors/institutions/journals. Size of circles indicate linkage strength. Lines represent network relationships (eg, citations). Publication year scales are normalized based on knowledge graph.A POTENTIAL SOLUTION–VOSviewer VOSviewer is a scientometric software tool that can help navigate vast literature using knowledge graphs.9 Intuitive visualizations can accelerate acquiring knowledge, from understanding research landscapes and trends to tracing the flow of evidence and identifying collaboration networks. VOSviewer is being adopted by a growing number of disciplines, and it may help meet the critical need for surgeons, academic institutions, and societies to navigate the weight of modern surgical knowledge. VOSviewer offers several methods to examine published literature. Articles can be downloaded from online (eg, PubMed, Scopus, Web of Science) or through integrated application program interface engines (eg, Microsoft Academic, Semantic Scholar). Selected articles can be visualized as customizable knowledge graphs. Knowledge graphs cluster chosen units (eg, articles, authors, institutions) by similarity, display inter-unit relationships, and delineate influence within the field. For intuitive visualization, displayed units can be weighted and highlighted by several features, including publication year, number of documents, number of citations, and total link strength (a measure of influence). Knowledge graphs hyperlink to article texts, facilitating full text review while visualizing an article's place within the web of knowledge. Articles can be visualized by hovering the mouse over circles in the knowledge graph. EXAMPLE OF VOSviewer's POTENTIAL USING A SPECIFIC CLINICAL PROBLEM We used Microsoft Academic, a machine learning-based semantic search engine comprising 251 million articles from 49,000 journals, to identify articles on a specific clinical problem (rib fractures) published between January 1, 1970 and December 31, 2020. We generated VOSviewer knowledge graphs to answer several questions: 1) What are important subtopics in rib fracture research? 2) What are emerging trends in rib fracture research? 3) What are foundational articles from which evidence stem from? 4) Who are influential authors and institutions? 5) Which authors and institutions collaborate for research? 6) What are good target journals to publish rib fracture research? 7) How would a surgeon evaluate the evidence on surgical stabilization of rib fractures before adopting this management modality? VOSviewer showed potential to answer key questions from decades of rib fracture literature within minutes. VOSviewer identified 952 articles on rib fractures and clustered distinct research subtopics (Fig. 1C; select subtopics are labeled). The “article citation network” knowledge graph clustered articles by content relevance; dissimilar articles were clustered more distantly (ie, articles on rib fractures as radiotherapy complication vs the majority of articles on rib fractures as traumatic injury). Among subtopics, surgical stabilization of rib fractures and analgesic modalities attracted the most recent publications (Fig. 1D). The knowledge graph also highlighted the most influential articles in rib fracture literature (indicated by cluster centrality and circle size), such as the foundational study that revealed the association between older age and poor outcomes (Fig. 1E).10 A second knowledge graph revealed author collaboration networks (Fig. 1F). Author collaboration networks can be filtered by the minimum number of documents or citations per author. Among 141 authors who have published ≥3 articles on rib fractures, many key collaborators (highlighted) were founding members of the Chest Wall Injury Society. The collaboration network found the 3 most influential authors (delineated by total linkage strength) to be Drs. Fredric M. Pieracci, Thomas W. White, and Andrew R. Doben. Institutional citation network revealed collaborations among authors’ affiliated institutions and found the University of Colorado-Denver, Denver Health Medical Center, and Intermountain Medical Center to have the greatest influence within rib fracture literature (Fig. 1G). Figure 1G highlights institutions by average article publication year, displaying which institutions have greatest recent productivity within rib fracture literature. Using journal citation knowledge graphs, VOSviewer identified journals that have published the most number of rib fracture articles (eg, The Journal of Trauma and Acute Care Surgery–71), journals wherein articles accrued the most number of citations (eg, The Journal of Trauma and Acute Care Surgery–2894), and journals publishing more articles on rib fractures in recent years (eg, Trauma Surgery and Acute Care Open and Journal of Thoracic Disease; Fig. 1H). VOSviewer can facilitate understanding niche topics that a surgeon may not be an expert in. If a surgeon wanted to evaluate evidence on surgical stabilization of rib fractures, knowledge graphs can help visualize the quantity and span of literature, understand each article in context of supportive evidence, identify foundational articles for targeted evaluation, identify surgeons and institutions who may help provide expert advice, and understand emerging research trends for clinical consideration. VOSviewer BEYOND THE INDIVIDUAL SURGEON VOSviewer expands the potential to extract actionable knowledge from literature to parties beyond individual surgeons. Departments can quantify an individual's scholarly impact within a research niche rather than rely on general metrics such as the h-index. Surgical societies could strategize research agenda by identifying research questions that have not been investigated recently. Journals can identify expert reviewers or commentators, and mitigate peer-review bias by identifying reviewers who have not collaborated with an author. Additionally, surgical communities can trace the flow of evidence to articles that define modern practice and periodically re-examine foundational evidence. VOSviewer LIMITATIONS VOSviewer has several limitations. First, if articles are downloaded from online databases rather than integrated application program interface engines, imports must be batched according to database export size limitations (eg, PubMed–10,000 articles; Scopus–2000 articles). This may delay querying topics comprising large article volumes. Second, knowledge graphs may vary depending on article sources and search strategies. Authors using VOSviewer should consider detailing source database and search strategies for result replicability. Third, VOSviewer cannot appraise individual articles’ quality of evidence. For example, an influential article identified using article citation networks may reflect literature dissemination bias rather than the article's quality of evidence. However, VOSviewer knowledge graphs can help surgeons identify foundational articles within a research niche and appraise whether evidence stands on solid ground. NAVIGATING INFORMATION OVERLOAD: FUTURE DIRECTIONS Navigating modern surgical knowledge requires contemporary tools. VOSviewer knowledge graphs may be a powerful complement for the surgical community to critically evaluate and extract meaningful knowledge from literature. Other tools to navigate vast literature may be on the horizon. For example, the General Index,11 a recently-released corpus of more than 100 million journal articles and 355 billion words and phrases associated with these articles, may pioneer other literature navigation software using text mining and natural language processing. These literature navigation tools will require thoughtful development; how surgeons curate literature—which articles are read, gain attention, and cited—may prompt variable future surgical practices from the same corpus of surgical knowledge. Inaccuracies in published content12 and citations13 amplify the importance of ensuring surgeons have tools to appraise comprehensive literature and avoid knowledge silos. Information overload is here to stay. Efforts to understand and mitigate bias in published research14 must be matched by efforts to consider how we curate aggregated knowledge. How we navigate information overload may determine which knowledge is implemented at the bedside and dictate the future course of surgical science.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 10 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.