Applying a novel approach to scoping review incorporating artificial intelligence: mapping the natural history of gonorrhoea

BMC Medical Research Methodology

Jane Whelan¹  Ekkehard Beck²  Nicolas Médoc³  Mohammad Ghoniem³  Mike Apicella⁴ 

[1] GSK, Amsterdam, The Netherlands;GSK, Wavre, Belgium;Luxembourg Institute of Science and Technology, Esch-sur-Alzette, Luxembourg;University of Iowa, Iowa City, USA;

关键词: Scoping review;    Artificial intelligence;    Visual text mining;    Natural language processing;    Co-clustering;    Clinical outcomes;    Health problems;    Gonorrhoea;    N. gonorrhoeae;    Natural history;   

DOI  :  10.1186/s12874-021-01367-x

来源: Springer

【 摘 要 】

BackgroundSystematic and scoping literature searches are increasingly resource intensive. We present the results of a scoping review which combines the use of a novel artificial-intelligence-(AI)-assisted Medline search tool with two other ‘traditional’ literature search methods. We illustrate this novel approach with a case study to identify and map the range of conditions (clinical presentations, complications, coinfections and health problems) associated with gonorrhoea infection.MethodsTo fully characterize the range of health outcomes associated with gonorrhoea, we combined a high yield preliminary search with a traditional systematic search, then supplemented with the output of a novel AI-assisted Medline search tool based on natural language processing methods to identify eligible literature.ResultsWe identified 189 health conditions associated with gonorrhoea infection of which: 53 were identified through the initial ‘high yield’ search; 99 through the systematic search; and 124 through the AI-assisted search. These were extracted from 107 unique references and 21 International Statistical Classification of Diseases and Related Health Problems Ninth and Tenth Revision (ICD 9/10) or Read codes. Health conditions were mapped to the urogenital tract (n = 86), anorectal tract (n = 6) oropharyngeal tract (n = 5) and the eye (n = 14); and other conditions such as systemic (n = 61) and neonatal conditions (n = 7), psychosocial associations (n = 3), and co-infections (n = 7). The 107 unique references attained a Scottish Intercollegiate Guidelines Network (SIGN) score of ≥2++ (n = 2), 2+ (14 [13%]), 2- (30 [28%]) and 3 (45 [42%]), respectively. The remaining papers (n = 16) were reviews.ConclusionsThrough AI screening of Medline, we captured – titles, abstracts, case reports and case series related to rare but serious health conditions related to gonorrhoea infection. These outcomes might otherwise have been missed during a systematic search. The AI-assisted search provided a useful addition to traditional/manual literature searches especially when rapid results are required in an exploratory setting.

【 授权许可】

CC BY